1
The following changes since commit 03a3a62fbd0aa5227e978eef3c67d3978aec9e5f:
1
The following changes since commit 56f9e46b841c7be478ca038d8d4085d776ab4b0d:
2
2
3
Merge tag 'for-upstream' of https://gitlab.com/bonzini/qemu into staging (2023-09-07 10:29:06 -0400)
3
Merge remote-tracking branch 'remotes/armbru/tags/pull-qapi-2017-02-20' into staging (2017-02-20 17:42:47 +0000)
4
4
5
are available in the Git repository at:
5
are available in the git repository at:
6
6
7
https://gitlab.com/stefanha/qemu.git tags/block-pull-request
7
git://github.com/stefanha/qemu.git tags/block-pull-request
8
8
9
for you to fetch changes up to 631c872614aca91eaf947c1748f0f27f99635d92:
9
for you to fetch changes up to a7b91d35bab97a2d3e779d0c64c9b837b52a6cf7:
10
10
11
tests/qtest: Introduce tests for UFS (2023-09-07 14:01:29 -0400)
11
coroutine-lock: make CoRwlock thread-safe and fair (2017-02-21 11:39:40 +0000)
12
12
13
----------------------------------------------------------------
13
----------------------------------------------------------------
14
Pull request
14
Pull request
15
15
16
- Jeuk Kim's emulated UFS device
16
v2:
17
- Fabiano Rosas' IOThread GSource "name" debugging aid
17
* Rebased to resolve scsi conflicts
18
18
19
----------------------------------------------------------------
19
----------------------------------------------------------------
20
20
21
Fabiano Rosas (1):
21
Paolo Bonzini (24):
22
iothread: Set the GSource "name" field
22
block: move AioContext, QEMUTimer, main-loop to libqemuutil
23
aio: introduce aio_co_schedule and aio_co_wake
24
block-backend: allow blk_prw from coroutine context
25
test-thread-pool: use generic AioContext infrastructure
26
io: add methods to set I/O handlers on AioContext
27
io: make qio_channel_yield aware of AioContexts
28
nbd: convert to use qio_channel_yield
29
coroutine-lock: reschedule coroutine on the AioContext it was running
30
on
31
blkdebug: reschedule coroutine on the AioContext it is running on
32
qed: introduce qed_aio_start_io and qed_aio_next_io_cb
33
aio: push aio_context_acquire/release down to dispatching
34
block: explicitly acquire aiocontext in timers that need it
35
block: explicitly acquire aiocontext in callbacks that need it
36
block: explicitly acquire aiocontext in bottom halves that need it
37
block: explicitly acquire aiocontext in aio callbacks that need it
38
aio-posix: partially inline aio_dispatch into aio_poll
39
async: remove unnecessary inc/dec pairs
40
block: document fields protected by AioContext lock
41
coroutine-lock: make CoMutex thread-safe
42
coroutine-lock: add limited spinning to CoMutex
43
test-aio-multithread: add performance comparison with thread-based
44
mutexes
45
coroutine-lock: place CoMutex before CoQueue in header
46
coroutine-lock: add mutex argument to CoQueue APIs
47
coroutine-lock: make CoRwlock thread-safe and fair
23
48
24
Jeuk Kim (4):
49
Makefile.objs | 4 -
25
hw/ufs: Initial commit for emulated Universal-Flash-Storage
50
stubs/Makefile.objs | 1 +
26
hw/ufs: Support for Query Transfer Requests
51
tests/Makefile.include | 19 +-
27
hw/ufs: Support for UFS logical unit
52
util/Makefile.objs | 6 +-
28
tests/qtest: Introduce tests for UFS
53
block/nbd-client.h | 2 +-
29
54
block/qed.h | 3 +
30
MAINTAINERS | 7 +
55
include/block/aio.h | 38 ++-
31
docs/specs/pci-ids.rst | 2 +
56
include/block/block_int.h | 64 +++--
32
meson.build | 1 +
57
include/io/channel.h | 72 +++++-
33
hw/ufs/trace.h | 1 +
58
include/qemu/coroutine.h | 84 ++++---
34
hw/ufs/ufs.h | 131 ++++
59
include/qemu/coroutine_int.h | 11 +-
35
include/block/ufs.h | 1090 +++++++++++++++++++++++++++
60
include/sysemu/block-backend.h | 14 +-
36
include/hw/pci/pci.h | 1 +
61
tests/iothread.h | 25 ++
37
include/hw/pci/pci_ids.h | 1 +
62
block/backup.c | 2 +-
38
include/scsi/constants.h | 1 +
63
block/blkdebug.c | 9 +-
39
hw/ufs/lu.c | 1445 ++++++++++++++++++++++++++++++++++++
64
block/blkreplay.c | 2 +-
40
hw/ufs/ufs.c | 1502 ++++++++++++++++++++++++++++++++++++++
65
block/block-backend.c | 13 +-
41
iothread.c | 14 +-
66
block/curl.c | 44 +++-
42
tests/qtest/ufs-test.c | 587 +++++++++++++++
67
block/gluster.c | 9 +-
43
hw/Kconfig | 1 +
68
block/io.c | 42 +---
44
hw/meson.build | 1 +
69
block/iscsi.c | 15 +-
45
hw/ufs/Kconfig | 4 +
70
block/linux-aio.c | 10 +-
46
hw/ufs/meson.build | 1 +
71
block/mirror.c | 12 +-
47
hw/ufs/trace-events | 58 ++
72
block/nbd-client.c | 119 +++++----
48
tests/qtest/meson.build | 1 +
73
block/nfs.c | 9 +-
49
19 files changed, 4843 insertions(+), 6 deletions(-)
74
block/qcow2-cluster.c | 4 +-
50
create mode 100644 hw/ufs/trace.h
75
block/qed-cluster.c | 2 +
51
create mode 100644 hw/ufs/ufs.h
76
block/qed-table.c | 12 +-
52
create mode 100644 include/block/ufs.h
77
block/qed.c | 58 +++--
53
create mode 100644 hw/ufs/lu.c
78
block/sheepdog.c | 31 +--
54
create mode 100644 hw/ufs/ufs.c
79
block/ssh.c | 29 +--
55
create mode 100644 tests/qtest/ufs-test.c
80
block/throttle-groups.c | 4 +-
56
create mode 100644 hw/ufs/Kconfig
81
block/win32-aio.c | 9 +-
57
create mode 100644 hw/ufs/meson.build
82
dma-helpers.c | 2 +
58
create mode 100644 hw/ufs/trace-events
83
hw/9pfs/9p.c | 2 +-
84
hw/block/virtio-blk.c | 19 +-
85
hw/scsi/scsi-bus.c | 2 +
86
hw/scsi/scsi-disk.c | 15 ++
87
hw/scsi/scsi-generic.c | 20 +-
88
hw/scsi/virtio-scsi.c | 7 +
89
io/channel-command.c | 13 +
90
io/channel-file.c | 11 +
91
io/channel-socket.c | 16 +-
92
io/channel-tls.c | 12 +
93
io/channel-watch.c | 6 +
94
io/channel.c | 97 ++++++--
95
nbd/client.c | 2 +-
96
nbd/common.c | 9 +-
97
nbd/server.c | 94 +++-----
98
stubs/linux-aio.c | 32 +++
99
stubs/set-fd-handler.c | 11 -
100
tests/iothread.c | 91 +++++++
101
tests/test-aio-multithread.c | 463 ++++++++++++++++++++++++++++++++++++
102
tests/test-thread-pool.c | 12 +-
103
aio-posix.c => util/aio-posix.c | 62 ++---
104
aio-win32.c => util/aio-win32.c | 30 +--
105
util/aiocb.c | 55 +++++
106
async.c => util/async.c | 84 ++++++-
107
iohandler.c => util/iohandler.c | 0
108
main-loop.c => util/main-loop.c | 0
109
util/qemu-coroutine-lock.c | 254 ++++++++++++++++++--
110
util/qemu-coroutine-sleep.c | 2 +-
111
util/qemu-coroutine.c | 8 +
112
qemu-timer.c => util/qemu-timer.c | 0
113
thread-pool.c => util/thread-pool.c | 8 +-
114
trace-events | 11 -
115
util/trace-events | 17 +-
116
67 files changed, 1712 insertions(+), 533 deletions(-)
117
create mode 100644 tests/iothread.h
118
create mode 100644 stubs/linux-aio.c
119
create mode 100644 tests/iothread.c
120
create mode 100644 tests/test-aio-multithread.c
121
rename aio-posix.c => util/aio-posix.c (94%)
122
rename aio-win32.c => util/aio-win32.c (95%)
123
create mode 100644 util/aiocb.c
124
rename async.c => util/async.c (82%)
125
rename iohandler.c => util/iohandler.c (100%)
126
rename main-loop.c => util/main-loop.c (100%)
127
rename qemu-timer.c => util/qemu-timer.c (100%)
128
rename thread-pool.c => util/thread-pool.c (97%)
59
129
60
--
130
--
61
2.41.0
131
2.9.3
132
133
diff view generated by jsdifflib
1
From: Jeuk Kim <jeuk20.kim@samsung.com>
1
From: Paolo Bonzini <pbonzini@redhat.com>
2
2
3
This patch includes the following tests
3
AioContext is fairly self contained, the only dependency is QEMUTimer but
4
Test mmio read
4
that in turn doesn't need anything else. So move them out of block-obj-y
5
Test ufs device initialization and ufs-lu recognition
5
to avoid introducing a dependency from io/ to block-obj-y.
6
Test I/O (Performs a write followed by a read to verify)
6
7
7
main-loop and its dependency iohandler also need to be moved, because
8
Signed-off-by: Jeuk Kim <jeuk20.kim@samsung.com>
8
later in this series io/ will call iohandler_get_aio_context.
9
Acked-by: Thomas Huth <thuth@redhat.com>
9
10
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
10
[Changed copyright "the QEMU team" to "other QEMU contributors" as
11
Message-id: 9e9207f54505e9ba30931849f949ff6f474ac333.1693980783.git.jeuk20.kim@gmail.com
11
suggested by Daniel Berrange and agreed by Paolo.
12
--Stefan]
13
14
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
15
Reviewed-by: Fam Zheng <famz@redhat.com>
16
Message-id: 20170213135235.12274-2-pbonzini@redhat.com
12
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
17
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
13
---
18
---
14
MAINTAINERS | 1 +
19
Makefile.objs | 4 ---
15
tests/qtest/ufs-test.c | 587 ++++++++++++++++++++++++++++++++++++++++
20
stubs/Makefile.objs | 1 +
16
tests/qtest/meson.build | 1 +
21
tests/Makefile.include | 11 ++++----
17
3 files changed, 589 insertions(+)
22
util/Makefile.objs | 6 +++-
18
create mode 100644 tests/qtest/ufs-test.c
23
block/io.c | 29 -------------------
19
24
stubs/linux-aio.c | 32 +++++++++++++++++++++
20
diff --git a/MAINTAINERS b/MAINTAINERS
25
stubs/set-fd-handler.c | 11 --------
21
index XXXXXXX..XXXXXXX 100644
26
aio-posix.c => util/aio-posix.c | 2 +-
22
--- a/MAINTAINERS
27
aio-win32.c => util/aio-win32.c | 0
23
+++ b/MAINTAINERS
28
util/aiocb.c | 55 +++++++++++++++++++++++++++++++++++++
24
@@ -XXX,XX +XXX,XX @@ M: Jeuk Kim <jeuk20.kim@samsung.com>
29
async.c => util/async.c | 3 +-
25
S: Supported
30
iohandler.c => util/iohandler.c | 0
26
F: hw/ufs/*
31
main-loop.c => util/main-loop.c | 0
27
F: include/block/ufs.h
32
qemu-timer.c => util/qemu-timer.c | 0
28
+F: tests/qtest/ufs-test.c
33
thread-pool.c => util/thread-pool.c | 2 +-
29
34
trace-events | 11 --------
30
megasas
35
util/trace-events | 11 ++++++++
31
M: Hannes Reinecke <hare@suse.com>
36
17 files changed, 114 insertions(+), 64 deletions(-)
32
diff --git a/tests/qtest/ufs-test.c b/tests/qtest/ufs-test.c
37
create mode 100644 stubs/linux-aio.c
38
rename aio-posix.c => util/aio-posix.c (99%)
39
rename aio-win32.c => util/aio-win32.c (100%)
40
create mode 100644 util/aiocb.c
41
rename async.c => util/async.c (99%)
42
rename iohandler.c => util/iohandler.c (100%)
43
rename main-loop.c => util/main-loop.c (100%)
44
rename qemu-timer.c => util/qemu-timer.c (100%)
45
rename thread-pool.c => util/thread-pool.c (99%)
46
47
diff --git a/Makefile.objs b/Makefile.objs
48
index XXXXXXX..XXXXXXX 100644
49
--- a/Makefile.objs
50
+++ b/Makefile.objs
51
@@ -XXX,XX +XXX,XX @@ chardev-obj-y = chardev/
52
#######################################################################
53
# block-obj-y is code used by both qemu system emulation and qemu-img
54
55
-block-obj-y = async.o thread-pool.o
56
block-obj-y += nbd/
57
block-obj-y += block.o blockjob.o
58
-block-obj-y += main-loop.o iohandler.o qemu-timer.o
59
-block-obj-$(CONFIG_POSIX) += aio-posix.o
60
-block-obj-$(CONFIG_WIN32) += aio-win32.o
61
block-obj-y += block/
62
block-obj-y += qemu-io-cmds.o
63
block-obj-$(CONFIG_REPLICATION) += replication.o
64
diff --git a/stubs/Makefile.objs b/stubs/Makefile.objs
65
index XXXXXXX..XXXXXXX 100644
66
--- a/stubs/Makefile.objs
67
+++ b/stubs/Makefile.objs
68
@@ -XXX,XX +XXX,XX @@ stub-obj-y += get-vm-name.o
69
stub-obj-y += iothread.o
70
stub-obj-y += iothread-lock.o
71
stub-obj-y += is-daemonized.o
72
+stub-obj-$(CONFIG_LINUX_AIO) += linux-aio.o
73
stub-obj-y += machine-init-done.o
74
stub-obj-y += migr-blocker.o
75
stub-obj-y += monitor.o
76
diff --git a/tests/Makefile.include b/tests/Makefile.include
77
index XXXXXXX..XXXXXXX 100644
78
--- a/tests/Makefile.include
79
+++ b/tests/Makefile.include
80
@@ -XXX,XX +XXX,XX @@ check-unit-y += tests/test-visitor-serialization$(EXESUF)
81
check-unit-y += tests/test-iov$(EXESUF)
82
gcov-files-test-iov-y = util/iov.c
83
check-unit-y += tests/test-aio$(EXESUF)
84
+gcov-files-test-aio-y = util/async.c util/qemu-timer.o
85
+gcov-files-test-aio-$(CONFIG_WIN32) += util/aio-win32.c
86
+gcov-files-test-aio-$(CONFIG_POSIX) += util/aio-posix.c
87
check-unit-y += tests/test-throttle$(EXESUF)
88
gcov-files-test-aio-$(CONFIG_WIN32) = aio-win32.c
89
gcov-files-test-aio-$(CONFIG_POSIX) = aio-posix.c
90
@@ -XXX,XX +XXX,XX @@ tests/check-qjson$(EXESUF): tests/check-qjson.o $(test-util-obj-y)
91
tests/check-qom-interface$(EXESUF): tests/check-qom-interface.o $(test-qom-obj-y)
92
tests/check-qom-proplist$(EXESUF): tests/check-qom-proplist.o $(test-qom-obj-y)
93
94
-tests/test-char$(EXESUF): tests/test-char.o qemu-timer.o \
95
-    $(test-util-obj-y) $(qtest-obj-y) $(test-block-obj-y) $(chardev-obj-y)
96
+tests/test-char$(EXESUF): tests/test-char.o $(test-util-obj-y) $(qtest-obj-y) $(test-io-obj-y) $(chardev-obj-y)
97
tests/test-coroutine$(EXESUF): tests/test-coroutine.o $(test-block-obj-y)
98
tests/test-aio$(EXESUF): tests/test-aio.o $(test-block-obj-y)
99
tests/test-throttle$(EXESUF): tests/test-throttle.o $(test-block-obj-y)
100
@@ -XXX,XX +XXX,XX @@ tests/test-vmstate$(EXESUF): tests/test-vmstate.o \
101
    migration/vmstate.o migration/qemu-file.o \
102
migration/qemu-file-channel.o migration/qjson.o \
103
    $(test-io-obj-y)
104
-tests/test-timed-average$(EXESUF): tests/test-timed-average.o qemu-timer.o \
105
-    $(test-util-obj-y)
106
+tests/test-timed-average$(EXESUF): tests/test-timed-average.o $(test-util-obj-y)
107
tests/test-base64$(EXESUF): tests/test-base64.o \
108
    libqemuutil.a libqemustub.a
109
tests/ptimer-test$(EXESUF): tests/ptimer-test.o tests/ptimer-test-stubs.o hw/core/ptimer.o libqemustub.a
110
@@ -XXX,XX +XXX,XX @@ tests/usb-hcd-ehci-test$(EXESUF): tests/usb-hcd-ehci-test.o $(libqos-usb-obj-y)
111
tests/usb-hcd-xhci-test$(EXESUF): tests/usb-hcd-xhci-test.o $(libqos-usb-obj-y)
112
tests/pc-cpu-test$(EXESUF): tests/pc-cpu-test.o
113
tests/postcopy-test$(EXESUF): tests/postcopy-test.o
114
-tests/vhost-user-test$(EXESUF): tests/vhost-user-test.o qemu-timer.o \
115
+tests/vhost-user-test$(EXESUF): tests/vhost-user-test.o $(test-util-obj-y) \
116
    $(qtest-obj-y) $(test-io-obj-y) $(libqos-virtio-obj-y) $(libqos-pc-obj-y) \
117
    $(chardev-obj-y)
118
tests/qemu-iotests/socket_scm_helper$(EXESUF): tests/qemu-iotests/socket_scm_helper.o
119
diff --git a/util/Makefile.objs b/util/Makefile.objs
120
index XXXXXXX..XXXXXXX 100644
121
--- a/util/Makefile.objs
122
+++ b/util/Makefile.objs
123
@@ -XXX,XX +XXX,XX @@
124
util-obj-y = osdep.o cutils.o unicode.o qemu-timer-common.o
125
util-obj-y += bufferiszero.o
126
util-obj-y += lockcnt.o
127
+util-obj-y += aiocb.o async.o thread-pool.o qemu-timer.o
128
+util-obj-y += main-loop.o iohandler.o
129
+util-obj-$(CONFIG_POSIX) += aio-posix.o
130
util-obj-$(CONFIG_POSIX) += compatfd.o
131
util-obj-$(CONFIG_POSIX) += event_notifier-posix.o
132
util-obj-$(CONFIG_POSIX) += mmap-alloc.o
133
util-obj-$(CONFIG_POSIX) += oslib-posix.o
134
util-obj-$(CONFIG_POSIX) += qemu-openpty.o
135
util-obj-$(CONFIG_POSIX) += qemu-thread-posix.o
136
-util-obj-$(CONFIG_WIN32) += event_notifier-win32.o
137
util-obj-$(CONFIG_POSIX) += memfd.o
138
+util-obj-$(CONFIG_WIN32) += aio-win32.o
139
+util-obj-$(CONFIG_WIN32) += event_notifier-win32.o
140
util-obj-$(CONFIG_WIN32) += oslib-win32.o
141
util-obj-$(CONFIG_WIN32) += qemu-thread-win32.o
142
util-obj-y += envlist.o path.o module.o
143
diff --git a/block/io.c b/block/io.c
144
index XXXXXXX..XXXXXXX 100644
145
--- a/block/io.c
146
+++ b/block/io.c
147
@@ -XXX,XX +XXX,XX @@ BlockAIOCB *bdrv_aio_flush(BlockDriverState *bs,
148
return &acb->common;
149
}
150
151
-void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
152
- BlockCompletionFunc *cb, void *opaque)
153
-{
154
- BlockAIOCB *acb;
155
-
156
- acb = g_malloc(aiocb_info->aiocb_size);
157
- acb->aiocb_info = aiocb_info;
158
- acb->bs = bs;
159
- acb->cb = cb;
160
- acb->opaque = opaque;
161
- acb->refcnt = 1;
162
- return acb;
163
-}
164
-
165
-void qemu_aio_ref(void *p)
166
-{
167
- BlockAIOCB *acb = p;
168
- acb->refcnt++;
169
-}
170
-
171
-void qemu_aio_unref(void *p)
172
-{
173
- BlockAIOCB *acb = p;
174
- assert(acb->refcnt > 0);
175
- if (--acb->refcnt == 0) {
176
- g_free(acb);
177
- }
178
-}
179
-
180
/**************************************************************/
181
/* Coroutine block device emulation */
182
183
diff --git a/stubs/linux-aio.c b/stubs/linux-aio.c
33
new file mode 100644
184
new file mode 100644
34
index XXXXXXX..XXXXXXX
185
index XXXXXXX..XXXXXXX
35
--- /dev/null
186
--- /dev/null
36
+++ b/tests/qtest/ufs-test.c
187
+++ b/stubs/linux-aio.c
37
@@ -XXX,XX +XXX,XX @@
188
@@ -XXX,XX +XXX,XX @@
38
+/*
189
+/*
39
+ * QTest testcase for UFS
190
+ * Linux native AIO support.
40
+ *
191
+ *
41
+ * Copyright (c) 2023 Samsung Electronics Co., Ltd. All rights reserved.
192
+ * Copyright (C) 2009 IBM, Corp.
42
+ *
193
+ * Copyright (C) 2009 Red Hat, Inc.
43
+ * SPDX-License-Identifier: GPL-2.0-or-later
194
+ *
195
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
196
+ * See the COPYING file in the top-level directory.
44
+ */
197
+ */
45
+
46
+#include "qemu/osdep.h"
198
+#include "qemu/osdep.h"
47
+#include "qemu/module.h"
199
+#include "block/aio.h"
48
+#include "qemu/units.h"
200
+#include "block/raw-aio.h"
49
+#include "libqtest.h"
201
+
50
+#include "libqos/qgraph.h"
202
+void laio_detach_aio_context(LinuxAioState *s, AioContext *old_context)
51
+#include "libqos/pci.h"
203
+{
52
+#include "scsi/constants.h"
204
+ abort();
53
+#include "include/block/ufs.h"
205
+}
54
+
206
+
55
+/* Test images sizes in Bytes */
207
+void laio_attach_aio_context(LinuxAioState *s, AioContext *new_context)
56
+#define TEST_IMAGE_SIZE (64 * 1024 * 1024)
208
+{
57
+/* Timeout for various operations, in seconds. */
209
+ abort();
58
+#define TIMEOUT_SECONDS 10
210
+}
59
+/* Maximum PRD entry count */
211
+
60
+#define MAX_PRD_ENTRY_COUNT 10
212
+LinuxAioState *laio_init(void)
61
+#define PRD_ENTRY_DATA_SIZE 4096
213
+{
62
+/* Constants to build upiu */
214
+ abort();
63
+#define UTP_COMMAND_DESCRIPTOR_SIZE 4096
215
+}
64
+#define UTP_RESPONSE_UPIU_OFFSET 1024
216
+
65
+#define UTP_PRDT_UPIU_OFFSET 2048
217
+void laio_cleanup(LinuxAioState *s)
66
+
218
+{
67
+typedef struct QUfs QUfs;
219
+ abort();
68
+
220
+}
69
+struct QUfs {
221
diff --git a/stubs/set-fd-handler.c b/stubs/set-fd-handler.c
70
+ QOSGraphObject obj;
222
index XXXXXXX..XXXXXXX 100644
71
+ QPCIDevice dev;
223
--- a/stubs/set-fd-handler.c
72
+ QPCIBar bar;
224
+++ b/stubs/set-fd-handler.c
73
+
225
@@ -XXX,XX +XXX,XX @@ void qemu_set_fd_handler(int fd,
74
+ uint64_t utrlba;
226
{
75
+ uint64_t utmrlba;
227
abort();
76
+ uint64_t cmd_desc_addr;
228
}
77
+ uint64_t data_buffer_addr;
229
-
78
+
230
-void aio_set_fd_handler(AioContext *ctx,
79
+ bool enabled;
231
- int fd,
80
+};
232
- bool is_external,
81
+
233
- IOHandler *io_read,
82
+static inline uint32_t ufs_rreg(QUfs *ufs, size_t offset)
234
- IOHandler *io_write,
83
+{
235
- AioPollFn *io_poll,
84
+ return qpci_io_readl(&ufs->dev, ufs->bar, offset);
236
- void *opaque)
85
+}
237
-{
86
+
238
- abort();
87
+static inline void ufs_wreg(QUfs *ufs, size_t offset, uint32_t value)
239
-}
88
+{
240
diff --git a/aio-posix.c b/util/aio-posix.c
89
+ qpci_io_writel(&ufs->dev, ufs->bar, offset, value);
241
similarity index 99%
90
+}
242
rename from aio-posix.c
91
+
243
rename to util/aio-posix.c
92
+static void ufs_wait_for_irq(QUfs *ufs)
244
index XXXXXXX..XXXXXXX 100644
93
+{
245
--- a/aio-posix.c
94
+ uint64_t end_time;
246
+++ b/util/aio-posix.c
95
+ uint32_t is;
247
@@ -XXX,XX +XXX,XX @@
96
+ /* Wait for device to reset as the linux driver does. */
248
#include "qemu/rcu_queue.h"
97
+ end_time = g_get_monotonic_time() + TIMEOUT_SECONDS * G_TIME_SPAN_SECOND;
249
#include "qemu/sockets.h"
98
+ do {
250
#include "qemu/cutils.h"
99
+ qtest_clock_step(ufs->dev.bus->qts, 100);
251
-#include "trace-root.h"
100
+ is = ufs_rreg(ufs, A_IS);
252
+#include "trace.h"
101
+ } while (is == 0 && g_get_monotonic_time() < end_time);
253
#ifdef CONFIG_EPOLL_CREATE1
102
+}
254
#include <sys/epoll.h>
103
+
255
#endif
104
+static UtpTransferReqDesc ufs_build_req_utrd(uint64_t cmd_desc_addr,
256
diff --git a/aio-win32.c b/util/aio-win32.c
105
+ uint8_t slot,
257
similarity index 100%
106
+ uint32_t data_direction,
258
rename from aio-win32.c
107
+ uint16_t prd_table_length)
259
rename to util/aio-win32.c
108
+{
260
diff --git a/util/aiocb.c b/util/aiocb.c
109
+ UtpTransferReqDesc req = { 0 };
261
new file mode 100644
110
+ uint64_t command_desc_base_addr =
262
index XXXXXXX..XXXXXXX
111
+ cmd_desc_addr + slot * UTP_COMMAND_DESCRIPTOR_SIZE;
263
--- /dev/null
112
+
264
+++ b/util/aiocb.c
113
+ req.header.dword_0 =
265
@@ -XXX,XX +XXX,XX @@
114
+ cpu_to_le32(1 << 28 | data_direction | UFS_UTP_REQ_DESC_INT_CMD);
266
+/*
115
+ req.header.dword_2 = cpu_to_le32(UFS_OCS_INVALID_COMMAND_STATUS);
267
+ * BlockAIOCB allocation
116
+
268
+ *
117
+ req.command_desc_base_addr_hi = cpu_to_le32(command_desc_base_addr >> 32);
269
+ * Copyright (c) 2003-2017 Fabrice Bellard and other QEMU contributors
118
+ req.command_desc_base_addr_lo =
270
+ *
119
+ cpu_to_le32(command_desc_base_addr & 0xffffffff);
271
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
120
+ req.response_upiu_offset =
272
+ * of this software and associated documentation files (the "Software"), to deal
121
+ cpu_to_le16(UTP_RESPONSE_UPIU_OFFSET / sizeof(uint32_t));
273
+ * in the Software without restriction, including without limitation the rights
122
+ req.response_upiu_length = cpu_to_le16(sizeof(UtpUpiuRsp));
274
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
123
+ req.prd_table_offset = cpu_to_le16(UTP_PRDT_UPIU_OFFSET / sizeof(uint32_t));
275
+ * copies of the Software, and to permit persons to whom the Software is
124
+ req.prd_table_length = cpu_to_le16(prd_table_length);
276
+ * furnished to do so, subject to the following conditions:
125
+ return req;
277
+ *
126
+}
278
+ * The above copyright notice and this permission notice shall be included in
127
+
279
+ * all copies or substantial portions of the Software.
128
+static void ufs_send_nop_out(QUfs *ufs, uint8_t slot,
280
+ *
129
+ UtpTransferReqDesc *utrd_out, UtpUpiuRsp *rsp_out)
281
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
130
+{
282
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
131
+ /* Build up utp transfer request descriptor */
283
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
132
+ UtpTransferReqDesc utrd = ufs_build_req_utrd(ufs->cmd_desc_addr, slot,
284
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
133
+ UFS_UTP_NO_DATA_TRANSFER, 0);
285
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
134
+ uint64_t utrd_addr = ufs->utrlba + slot * sizeof(UtpTransferReqDesc);
286
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
135
+ uint64_t req_upiu_addr =
287
+ * THE SOFTWARE.
136
+ ufs->cmd_desc_addr + slot * UTP_COMMAND_DESCRIPTOR_SIZE;
288
+ */
137
+ uint64_t rsp_upiu_addr = req_upiu_addr + UTP_RESPONSE_UPIU_OFFSET;
289
+
138
+ qtest_memwrite(ufs->dev.bus->qts, utrd_addr, &utrd, sizeof(utrd));
290
+#include "qemu/osdep.h"
139
+
291
+#include "block/aio.h"
140
+ /* Build up request upiu */
292
+
141
+ UtpUpiuReq req_upiu = { 0 };
293
+void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
142
+ req_upiu.header.trans_type = UFS_UPIU_TRANSACTION_NOP_OUT;
294
+ BlockCompletionFunc *cb, void *opaque)
143
+ req_upiu.header.task_tag = slot;
295
+{
144
+ qtest_memwrite(ufs->dev.bus->qts, req_upiu_addr, &req_upiu,
296
+ BlockAIOCB *acb;
145
+ sizeof(req_upiu));
297
+
146
+
298
+ acb = g_malloc(aiocb_info->aiocb_size);
147
+ /* Ring Doorbell */
299
+ acb->aiocb_info = aiocb_info;
148
+ ufs_wreg(ufs, A_UTRLDBR, 1);
300
+ acb->bs = bs;
149
+ ufs_wait_for_irq(ufs);
301
+ acb->cb = cb;
150
+ g_assert_true(FIELD_EX32(ufs_rreg(ufs, A_IS), IS, UTRCS));
302
+ acb->opaque = opaque;
151
+ ufs_wreg(ufs, A_IS, FIELD_DP32(0, IS, UTRCS, 1));
303
+ acb->refcnt = 1;
152
+
304
+ return acb;
153
+ qtest_memread(ufs->dev.bus->qts, utrd_addr, utrd_out, sizeof(*utrd_out));
305
+}
154
+ qtest_memread(ufs->dev.bus->qts, rsp_upiu_addr, rsp_out, sizeof(*rsp_out));
306
+
155
+}
307
+void qemu_aio_ref(void *p)
156
+
308
+{
157
+static void ufs_send_query(QUfs *ufs, uint8_t slot, uint8_t query_function,
309
+ BlockAIOCB *acb = p;
158
+ uint8_t query_opcode, uint8_t idn, uint8_t index,
310
+ acb->refcnt++;
159
+ UtpTransferReqDesc *utrd_out, UtpUpiuRsp *rsp_out)
311
+}
160
+{
312
+
161
+ /* Build up utp transfer request descriptor */
313
+void qemu_aio_unref(void *p)
162
+ UtpTransferReqDesc utrd = ufs_build_req_utrd(ufs->cmd_desc_addr, slot,
314
+{
163
+ UFS_UTP_NO_DATA_TRANSFER, 0);
315
+ BlockAIOCB *acb = p;
164
+ uint64_t utrd_addr = ufs->utrlba + slot * sizeof(UtpTransferReqDesc);
316
+ assert(acb->refcnt > 0);
165
+ uint64_t req_upiu_addr =
317
+ if (--acb->refcnt == 0) {
166
+ ufs->cmd_desc_addr + slot * UTP_COMMAND_DESCRIPTOR_SIZE;
318
+ g_free(acb);
167
+ uint64_t rsp_upiu_addr = req_upiu_addr + UTP_RESPONSE_UPIU_OFFSET;
168
+ qtest_memwrite(ufs->dev.bus->qts, utrd_addr, &utrd, sizeof(utrd));
169
+
170
+ /* Build up request upiu */
171
+ UtpUpiuReq req_upiu = { 0 };
172
+ req_upiu.header.trans_type = UFS_UPIU_TRANSACTION_QUERY_REQ;
173
+ req_upiu.header.query_func = query_function;
174
+ req_upiu.header.task_tag = slot;
175
+ /*
176
+ * QEMU UFS does not currently support Write descriptor and Write attribute,
177
+ * so the value of data_segment_length is always 0.
178
+ */
179
+ req_upiu.header.data_segment_length = 0;
180
+ req_upiu.qr.opcode = query_opcode;
181
+ req_upiu.qr.idn = idn;
182
+ req_upiu.qr.index = index;
183
+ qtest_memwrite(ufs->dev.bus->qts, req_upiu_addr, &req_upiu,
184
+ sizeof(req_upiu));
185
+
186
+ /* Ring Doorbell */
187
+ ufs_wreg(ufs, A_UTRLDBR, 1);
188
+ ufs_wait_for_irq(ufs);
189
+ g_assert_true(FIELD_EX32(ufs_rreg(ufs, A_IS), IS, UTRCS));
190
+ ufs_wreg(ufs, A_IS, FIELD_DP32(0, IS, UTRCS, 1));
191
+
192
+ qtest_memread(ufs->dev.bus->qts, utrd_addr, utrd_out, sizeof(*utrd_out));
193
+ qtest_memread(ufs->dev.bus->qts, rsp_upiu_addr, rsp_out, sizeof(*rsp_out));
194
+}
195
+
196
+static void ufs_send_scsi_command(QUfs *ufs, uint8_t slot, uint8_t lun,
197
+ const uint8_t *cdb, const uint8_t *data_in,
198
+ size_t data_in_len, uint8_t *data_out,
199
+ size_t data_out_len,
200
+ UtpTransferReqDesc *utrd_out,
201
+ UtpUpiuRsp *rsp_out)
202
+
203
+{
204
+ /* Build up PRDT */
205
+ UfshcdSgEntry entries[MAX_PRD_ENTRY_COUNT] = {
206
+ 0,
207
+ };
208
+ uint8_t flags;
209
+ uint16_t prd_table_length, i;
210
+ uint32_t data_direction, data_len;
211
+ uint64_t req_upiu_addr =
212
+ ufs->cmd_desc_addr + slot * UTP_COMMAND_DESCRIPTOR_SIZE;
213
+ uint64_t prdt_addr = req_upiu_addr + UTP_PRDT_UPIU_OFFSET;
214
+
215
+ g_assert_true(data_in_len < MAX_PRD_ENTRY_COUNT * PRD_ENTRY_DATA_SIZE);
216
+ g_assert_true(data_out_len < MAX_PRD_ENTRY_COUNT * PRD_ENTRY_DATA_SIZE);
217
+ if (data_in_len > 0) {
218
+ g_assert_nonnull(data_in);
219
+ data_direction = UFS_UTP_HOST_TO_DEVICE;
220
+ data_len = data_in_len;
221
+ flags = UFS_UPIU_CMD_FLAGS_WRITE;
222
+ } else if (data_out_len > 0) {
223
+ g_assert_nonnull(data_out);
224
+ data_direction = UFS_UTP_DEVICE_TO_HOST;
225
+ data_len = data_out_len;
226
+ flags = UFS_UPIU_CMD_FLAGS_READ;
227
+ } else {
228
+ data_direction = UFS_UTP_NO_DATA_TRANSFER;
229
+ data_len = 0;
230
+ flags = UFS_UPIU_CMD_FLAGS_NONE;
231
+ }
319
+ }
232
+ prd_table_length = DIV_ROUND_UP(data_len, PRD_ENTRY_DATA_SIZE);
320
+}
233
+
321
diff --git a/async.c b/util/async.c
234
+ qtest_memset(ufs->dev.bus->qts, ufs->data_buffer_addr, 0,
322
similarity index 99%
235
+ MAX_PRD_ENTRY_COUNT * PRD_ENTRY_DATA_SIZE);
323
rename from async.c
236
+ if (data_in_len) {
324
rename to util/async.c
237
+ qtest_memwrite(ufs->dev.bus->qts, ufs->data_buffer_addr, data_in,
325
index XXXXXXX..XXXXXXX 100644
238
+ data_in_len);
326
--- a/async.c
239
+ }
327
+++ b/util/async.c
240
+
328
@@ -XXX,XX +XXX,XX @@
241
+ for (i = 0; i < prd_table_length; i++) {
329
/*
242
+ entries[i].addr =
330
- * QEMU System Emulator
243
+ cpu_to_le64(ufs->data_buffer_addr + i * sizeof(UfshcdSgEntry));
331
+ * Data plane event loop
244
+ if (i + 1 != prd_table_length) {
332
*
245
+ entries[i].size = cpu_to_le32(PRD_ENTRY_DATA_SIZE - 1);
333
* Copyright (c) 2003-2008 Fabrice Bellard
246
+ } else {
334
+ * Copyright (c) 2009-2017 QEMU contributors
247
+ entries[i].size = cpu_to_le32(
335
*
248
+ data_len - (PRD_ENTRY_DATA_SIZE * (prd_table_length - 1)) - 1);
336
* Permission is hereby granted, free of charge, to any person obtaining a copy
249
+ }
337
* of this software and associated documentation files (the "Software"), to deal
250
+ }
338
diff --git a/iohandler.c b/util/iohandler.c
251
+ qtest_memwrite(ufs->dev.bus->qts, prdt_addr, entries,
339
similarity index 100%
252
+ prd_table_length * sizeof(UfshcdSgEntry));
340
rename from iohandler.c
253
+
341
rename to util/iohandler.c
254
+ /* Build up utp transfer request descriptor */
342
diff --git a/main-loop.c b/util/main-loop.c
255
+ UtpTransferReqDesc utrd = ufs_build_req_utrd(
343
similarity index 100%
256
+ ufs->cmd_desc_addr, slot, data_direction, prd_table_length);
344
rename from main-loop.c
257
+ uint64_t utrd_addr = ufs->utrlba + slot * sizeof(UtpTransferReqDesc);
345
rename to util/main-loop.c
258
+ uint64_t rsp_upiu_addr = req_upiu_addr + UTP_RESPONSE_UPIU_OFFSET;
346
diff --git a/qemu-timer.c b/util/qemu-timer.c
259
+ qtest_memwrite(ufs->dev.bus->qts, utrd_addr, &utrd, sizeof(utrd));
347
similarity index 100%
260
+
348
rename from qemu-timer.c
261
+ /* Build up request upiu */
349
rename to util/qemu-timer.c
262
+ UtpUpiuReq req_upiu = { 0 };
350
diff --git a/thread-pool.c b/util/thread-pool.c
263
+ req_upiu.header.trans_type = UFS_UPIU_TRANSACTION_COMMAND;
351
similarity index 99%
264
+ req_upiu.header.flags = flags;
352
rename from thread-pool.c
265
+ req_upiu.header.lun = lun;
353
rename to util/thread-pool.c
266
+ req_upiu.header.task_tag = slot;
354
index XXXXXXX..XXXXXXX 100644
267
+ req_upiu.sc.exp_data_transfer_len = cpu_to_be32(data_len);
355
--- a/thread-pool.c
268
+ memcpy(req_upiu.sc.cdb, cdb, UFS_CDB_SIZE);
356
+++ b/util/thread-pool.c
269
+ qtest_memwrite(ufs->dev.bus->qts, req_upiu_addr, &req_upiu,
357
@@ -XXX,XX +XXX,XX @@
270
+ sizeof(req_upiu));
358
#include "qemu/queue.h"
271
+
359
#include "qemu/thread.h"
272
+ /* Ring Doorbell */
360
#include "qemu/coroutine.h"
273
+ ufs_wreg(ufs, A_UTRLDBR, 1);
361
-#include "trace-root.h"
274
+ ufs_wait_for_irq(ufs);
362
+#include "trace.h"
275
+ g_assert_true(FIELD_EX32(ufs_rreg(ufs, A_IS), IS, UTRCS));
363
#include "block/thread-pool.h"
276
+ ufs_wreg(ufs, A_IS, FIELD_DP32(0, IS, UTRCS, 1));
364
#include "qemu/main-loop.h"
277
+
365
278
+ qtest_memread(ufs->dev.bus->qts, utrd_addr, utrd_out, sizeof(*utrd_out));
366
diff --git a/trace-events b/trace-events
279
+ qtest_memread(ufs->dev.bus->qts, rsp_upiu_addr, rsp_out, sizeof(*rsp_out));
367
index XXXXXXX..XXXXXXX 100644
280
+ if (data_out_len) {
368
--- a/trace-events
281
+ qtest_memread(ufs->dev.bus->qts, ufs->data_buffer_addr, data_out,
369
+++ b/trace-events
282
+ data_out_len);
370
@@ -XXX,XX +XXX,XX @@
283
+ }
371
#
284
+}
372
# The <format-string> should be a sprintf()-compatible format string.
285
+
373
286
+/**
374
-# aio-posix.c
287
+ * Initialize Ufs host controller and logical unit.
375
-run_poll_handlers_begin(void *ctx, int64_t max_ns) "ctx %p max_ns %"PRId64
288
+ * After running this function, you can make a transfer request to the UFS.
376
-run_poll_handlers_end(void *ctx, bool progress) "ctx %p progress %d"
289
+ */
377
-poll_shrink(void *ctx, int64_t old, int64_t new) "ctx %p old %"PRId64" new %"PRId64
290
+static void ufs_init(QUfs *ufs, QGuestAllocator *alloc)
378
-poll_grow(void *ctx, int64_t old, int64_t new) "ctx %p old %"PRId64" new %"PRId64
291
+{
379
-
292
+ uint64_t end_time;
380
-# thread-pool.c
293
+ uint32_t nutrs, nutmrs;
381
-thread_pool_submit(void *pool, void *req, void *opaque) "pool %p req %p opaque %p"
294
+ uint32_t hcs, is, ucmdarg2, cap;
382
-thread_pool_complete(void *pool, void *req, void *opaque, int ret) "pool %p req %p opaque %p ret %d"
295
+ uint32_t hce = 0, ie = 0;
383
-thread_pool_cancel(void *req, void *opaque) "req %p opaque %p"
296
+ UtpTransferReqDesc utrd;
384
-
297
+ UtpUpiuRsp rsp_upiu;
385
# ioport.c
298
+
386
cpu_in(unsigned int addr, char size, unsigned int val) "addr %#x(%c) value %u"
299
+ ufs->bar = qpci_iomap(&ufs->dev, 0, NULL);
387
cpu_out(unsigned int addr, char size, unsigned int val) "addr %#x(%c) value %u"
300
+ qpci_device_enable(&ufs->dev);
388
diff --git a/util/trace-events b/util/trace-events
301
+
389
index XXXXXXX..XXXXXXX 100644
302
+ /* Start host controller initialization */
390
--- a/util/trace-events
303
+ hce = FIELD_DP32(hce, HCE, HCE, 1);
391
+++ b/util/trace-events
304
+ ufs_wreg(ufs, A_HCE, hce);
392
@@ -XXX,XX +XXX,XX @@
305
+
393
# See docs/tracing.txt for syntax documentation.
306
+ /* Wait for device to reset */
394
307
+ end_time = g_get_monotonic_time() + TIMEOUT_SECONDS * G_TIME_SPAN_SECOND;
395
+# util/aio-posix.c
308
+ do {
396
+run_poll_handlers_begin(void *ctx, int64_t max_ns) "ctx %p max_ns %"PRId64
309
+ qtest_clock_step(ufs->dev.bus->qts, 100);
397
+run_poll_handlers_end(void *ctx, bool progress) "ctx %p progress %d"
310
+ hce = FIELD_EX32(ufs_rreg(ufs, A_HCE), HCE, HCE);
398
+poll_shrink(void *ctx, int64_t old, int64_t new) "ctx %p old %"PRId64" new %"PRId64
311
+ } while (hce == 0 && g_get_monotonic_time() < end_time);
399
+poll_grow(void *ctx, int64_t old, int64_t new) "ctx %p old %"PRId64" new %"PRId64
312
+ g_assert_cmpuint(hce, ==, 1);
400
+
313
+
401
+# util/thread-pool.c
314
+ /* Enable interrupt */
402
+thread_pool_submit(void *pool, void *req, void *opaque) "pool %p req %p opaque %p"
315
+ ie = FIELD_DP32(ie, IE, UCCE, 1);
403
+thread_pool_complete(void *pool, void *req, void *opaque, int ret) "pool %p req %p opaque %p ret %d"
316
+ ie = FIELD_DP32(ie, IE, UHESE, 1);
404
+thread_pool_cancel(void *req, void *opaque) "req %p opaque %p"
317
+ ie = FIELD_DP32(ie, IE, UHXSE, 1);
405
+
318
+ ie = FIELD_DP32(ie, IE, UPMSE, 1);
406
# util/buffer.c
319
+ ufs_wreg(ufs, A_IE, ie);
407
buffer_resize(const char *buf, size_t olen, size_t len) "%s: old %zd, new %zd"
320
+
408
buffer_move_empty(const char *buf, size_t len, const char *from) "%s: %zd bytes from %s"
321
+ /* Send DME_LINK_STARTUP uic command */
322
+ hcs = ufs_rreg(ufs, A_HCS);
323
+ g_assert_true(FIELD_EX32(hcs, HCS, UCRDY));
324
+
325
+ ufs_wreg(ufs, A_UCMDARG1, 0);
326
+ ufs_wreg(ufs, A_UCMDARG2, 0);
327
+ ufs_wreg(ufs, A_UCMDARG3, 0);
328
+ ufs_wreg(ufs, A_UICCMD, UFS_UIC_CMD_DME_LINK_STARTUP);
329
+
330
+ is = ufs_rreg(ufs, A_IS);
331
+ g_assert_true(FIELD_EX32(is, IS, UCCS));
332
+ ufs_wreg(ufs, A_IS, FIELD_DP32(0, IS, UCCS, 1));
333
+
334
+ ucmdarg2 = ufs_rreg(ufs, A_UCMDARG2);
335
+ g_assert_cmpuint(ucmdarg2, ==, 0);
336
+ is = ufs_rreg(ufs, A_IS);
337
+ g_assert_cmpuint(is, ==, 0);
338
+ hcs = ufs_rreg(ufs, A_HCS);
339
+ g_assert_true(FIELD_EX32(hcs, HCS, DP));
340
+ g_assert_true(FIELD_EX32(hcs, HCS, UTRLRDY));
341
+ g_assert_true(FIELD_EX32(hcs, HCS, UTMRLRDY));
342
+ g_assert_true(FIELD_EX32(hcs, HCS, UCRDY));
343
+
344
+ /* Enable all interrupt functions */
345
+ ie = FIELD_DP32(ie, IE, UTRCE, 1);
346
+ ie = FIELD_DP32(ie, IE, UEE, 1);
347
+ ie = FIELD_DP32(ie, IE, UPMSE, 1);
348
+ ie = FIELD_DP32(ie, IE, UHXSE, 1);
349
+ ie = FIELD_DP32(ie, IE, UHESE, 1);
350
+ ie = FIELD_DP32(ie, IE, UTMRCE, 1);
351
+ ie = FIELD_DP32(ie, IE, UCCE, 1);
352
+ ie = FIELD_DP32(ie, IE, DFEE, 1);
353
+ ie = FIELD_DP32(ie, IE, HCFEE, 1);
354
+ ie = FIELD_DP32(ie, IE, SBFEE, 1);
355
+ ie = FIELD_DP32(ie, IE, CEFEE, 1);
356
+ ufs_wreg(ufs, A_IE, ie);
357
+ ufs_wreg(ufs, A_UTRIACR, 0);
358
+
359
+ /* Enable tranfer request and task management request */
360
+ cap = ufs_rreg(ufs, A_CAP);
361
+ nutrs = FIELD_EX32(cap, CAP, NUTRS) + 1;
362
+ nutmrs = FIELD_EX32(cap, CAP, NUTMRS) + 1;
363
+ ufs->cmd_desc_addr =
364
+ guest_alloc(alloc, nutrs * UTP_COMMAND_DESCRIPTOR_SIZE);
365
+ ufs->data_buffer_addr =
366
+ guest_alloc(alloc, MAX_PRD_ENTRY_COUNT * PRD_ENTRY_DATA_SIZE);
367
+ ufs->utrlba = guest_alloc(alloc, nutrs * sizeof(UtpTransferReqDesc));
368
+ ufs->utmrlba = guest_alloc(alloc, nutmrs * sizeof(UtpTaskReqDesc));
369
+
370
+ ufs_wreg(ufs, A_UTRLBA, ufs->utrlba & 0xffffffff);
371
+ ufs_wreg(ufs, A_UTRLBAU, ufs->utrlba >> 32);
372
+ ufs_wreg(ufs, A_UTMRLBA, ufs->utmrlba & 0xffffffff);
373
+ ufs_wreg(ufs, A_UTMRLBAU, ufs->utmrlba >> 32);
374
+ ufs_wreg(ufs, A_UTRLRSR, 1);
375
+ ufs_wreg(ufs, A_UTMRLRSR, 1);
376
+
377
+ /* Send nop out to test transfer request */
378
+ ufs_send_nop_out(ufs, 0, &utrd, &rsp_upiu);
379
+ g_assert_cmpuint(le32_to_cpu(utrd.header.dword_2), ==, UFS_OCS_SUCCESS);
380
+
381
+ /* Set fDeviceInit flag via query request */
382
+ ufs_send_query(ufs, 0, UFS_UPIU_QUERY_FUNC_STANDARD_WRITE_REQUEST,
383
+ UFS_UPIU_QUERY_OPCODE_SET_FLAG,
384
+ UFS_QUERY_FLAG_IDN_FDEVICEINIT, 0, &utrd, &rsp_upiu);
385
+ g_assert_cmpuint(le32_to_cpu(utrd.header.dword_2), ==, UFS_OCS_SUCCESS);
386
+
387
+ /* Wait for device to reset */
388
+ end_time = g_get_monotonic_time() + TIMEOUT_SECONDS * G_TIME_SPAN_SECOND;
389
+ do {
390
+ qtest_clock_step(ufs->dev.bus->qts, 100);
391
+ ufs_send_query(ufs, 0, UFS_UPIU_QUERY_FUNC_STANDARD_READ_REQUEST,
392
+ UFS_UPIU_QUERY_OPCODE_READ_FLAG,
393
+ UFS_QUERY_FLAG_IDN_FDEVICEINIT, 0, &utrd, &rsp_upiu);
394
+ } while (be32_to_cpu(rsp_upiu.qr.value) != 0 &&
395
+ g_get_monotonic_time() < end_time);
396
+ g_assert_cmpuint(be32_to_cpu(rsp_upiu.qr.value), ==, 0);
397
+
398
+ ufs->enabled = true;
399
+}
400
+
401
+static void ufs_exit(QUfs *ufs, QGuestAllocator *alloc)
402
+{
403
+ if (ufs->enabled) {
404
+ guest_free(alloc, ufs->utrlba);
405
+ guest_free(alloc, ufs->utmrlba);
406
+ guest_free(alloc, ufs->cmd_desc_addr);
407
+ guest_free(alloc, ufs->data_buffer_addr);
408
+ }
409
+
410
+ qpci_iounmap(&ufs->dev, ufs->bar);
411
+}
412
+
413
+static void *ufs_get_driver(void *obj, const char *interface)
414
+{
415
+ QUfs *ufs = obj;
416
+
417
+ if (!g_strcmp0(interface, "pci-device")) {
418
+ return &ufs->dev;
419
+ }
420
+
421
+ fprintf(stderr, "%s not present in ufs\n", interface);
422
+ g_assert_not_reached();
423
+}
424
+
425
+static void *ufs_create(void *pci_bus, QGuestAllocator *alloc, void *addr)
426
+{
427
+ QUfs *ufs = g_new0(QUfs, 1);
428
+ QPCIBus *bus = pci_bus;
429
+
430
+ qpci_device_init(&ufs->dev, bus, addr);
431
+ ufs->obj.get_driver = ufs_get_driver;
432
+
433
+ return &ufs->obj;
434
+}
435
+
436
+static void ufstest_reg_read(void *obj, void *data, QGuestAllocator *alloc)
437
+{
438
+ QUfs *ufs = obj;
439
+ uint32_t cap;
440
+
441
+ ufs->bar = qpci_iomap(&ufs->dev, 0, NULL);
442
+ qpci_device_enable(&ufs->dev);
443
+
444
+ cap = ufs_rreg(ufs, A_CAP);
445
+ g_assert_cmpuint(FIELD_EX32(cap, CAP, NUTRS), ==, 31);
446
+ g_assert_cmpuint(FIELD_EX32(cap, CAP, NUTMRS), ==, 7);
447
+ g_assert_cmpuint(FIELD_EX32(cap, CAP, 64AS), ==, 1);
448
+
449
+ qpci_iounmap(&ufs->dev, ufs->bar);
450
+}
451
+
452
+static void ufstest_init(void *obj, void *data, QGuestAllocator *alloc)
453
+{
454
+ QUfs *ufs = obj;
455
+
456
+ uint8_t buf[4096] = { 0 };
457
+ const uint8_t report_luns_cdb[UFS_CDB_SIZE] = {
458
+ /* allocation length 4096 */
459
+ REPORT_LUNS, 0x00, 0x00, 0x00, 0x00, 0x00,
460
+ 0x00, 0x00, 0x10, 0x00, 0x00, 0x00
461
+ };
462
+ const uint8_t test_unit_ready_cdb[UFS_CDB_SIZE] = {
463
+ TEST_UNIT_READY,
464
+ };
465
+ UtpTransferReqDesc utrd;
466
+ UtpUpiuRsp rsp_upiu;
467
+
468
+ ufs_init(ufs, alloc);
469
+
470
+ /* Check REPORT_LUNS */
471
+ ufs_send_scsi_command(ufs, 0, 0, report_luns_cdb, NULL, 0, buf, sizeof(buf),
472
+ &utrd, &rsp_upiu);
473
+ g_assert_cmpuint(le32_to_cpu(utrd.header.dword_2), ==, UFS_OCS_SUCCESS);
474
+ g_assert_cmpuint(rsp_upiu.header.scsi_status, ==, GOOD);
475
+ /* LUN LIST LENGTH should be 8, in big endian */
476
+ g_assert_cmpuint(buf[3], ==, 8);
477
+ /* There is one logical unit whose lun is 0 */
478
+ g_assert_cmpuint(buf[9], ==, 0);
479
+
480
+ /* Check TEST_UNIT_READY */
481
+ ufs_send_scsi_command(ufs, 0, 0, test_unit_ready_cdb, NULL, 0, NULL, 0,
482
+ &utrd, &rsp_upiu);
483
+ g_assert_cmpuint(le32_to_cpu(utrd.header.dword_2), ==, UFS_OCS_SUCCESS);
484
+ g_assert_cmpuint(rsp_upiu.header.scsi_status, ==, GOOD);
485
+
486
+ ufs_exit(ufs, alloc);
487
+}
488
+
489
+static void ufstest_read_write(void *obj, void *data, QGuestAllocator *alloc)
490
+{
491
+ QUfs *ufs = obj;
492
+ uint8_t read_buf[4096] = { 0 };
493
+ uint8_t write_buf[4096] = { 0 };
494
+ const uint8_t read_capacity_cdb[UFS_CDB_SIZE] = {
495
+ /* allocation length 4096 */
496
+ SERVICE_ACTION_IN_16,
497
+ SAI_READ_CAPACITY_16,
498
+ 0x00,
499
+ 0x00,
500
+ 0x00,
501
+ 0x00,
502
+ 0x00,
503
+ 0x00,
504
+ 0x00,
505
+ 0x00,
506
+ 0x00,
507
+ 0x00,
508
+ 0x10,
509
+ 0x00,
510
+ 0x00,
511
+ 0x00
512
+ };
513
+ const uint8_t read_cdb[UFS_CDB_SIZE] = {
514
+ /* READ(10) to LBA 0, transfer length 1 */
515
+ READ_10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00
516
+ };
517
+ const uint8_t write_cdb[UFS_CDB_SIZE] = {
518
+ /* WRITE(10) to LBA 0, transfer length 1 */
519
+ WRITE_10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00
520
+ };
521
+ uint32_t block_size;
522
+ UtpTransferReqDesc utrd;
523
+ UtpUpiuRsp rsp_upiu;
524
+
525
+ ufs_init(ufs, alloc);
526
+
527
+ /* Read capacity */
528
+ ufs_send_scsi_command(ufs, 0, 1, read_capacity_cdb, NULL, 0, read_buf,
529
+ sizeof(read_buf), &utrd, &rsp_upiu);
530
+ g_assert_cmpuint(le32_to_cpu(utrd.header.dword_2), ==, UFS_OCS_SUCCESS);
531
+ g_assert_cmpuint(rsp_upiu.header.scsi_status, ==,
532
+ UFS_COMMAND_RESULT_SUCESS);
533
+ block_size = ldl_be_p(&read_buf[8]);
534
+ g_assert_cmpuint(block_size, ==, 4096);
535
+
536
+ /* Write data */
537
+ memset(write_buf, rand() % 255 + 1, block_size);
538
+ ufs_send_scsi_command(ufs, 0, 1, write_cdb, write_buf, block_size, NULL, 0,
539
+ &utrd, &rsp_upiu);
540
+ g_assert_cmpuint(le32_to_cpu(utrd.header.dword_2), ==, UFS_OCS_SUCCESS);
541
+ g_assert_cmpuint(rsp_upiu.header.scsi_status, ==,
542
+ UFS_COMMAND_RESULT_SUCESS);
543
+
544
+ /* Read data and verify */
545
+ ufs_send_scsi_command(ufs, 0, 1, read_cdb, NULL, 0, read_buf, block_size,
546
+ &utrd, &rsp_upiu);
547
+ g_assert_cmpuint(le32_to_cpu(utrd.header.dword_2), ==, UFS_OCS_SUCCESS);
548
+ g_assert_cmpuint(rsp_upiu.header.scsi_status, ==,
549
+ UFS_COMMAND_RESULT_SUCESS);
550
+ g_assert_cmpint(memcmp(read_buf, write_buf, block_size), ==, 0);
551
+
552
+ ufs_exit(ufs, alloc);
553
+}
554
+
555
+static void drive_destroy(void *path)
556
+{
557
+ unlink(path);
558
+ g_free(path);
559
+ qos_invalidate_command_line();
560
+}
561
+
562
+static char *drive_create(void)
563
+{
564
+ int fd, ret;
565
+ char *t_path;
566
+
567
+ /* Create a temporary raw image */
568
+ fd = g_file_open_tmp("qtest-ufs.XXXXXX", &t_path, NULL);
569
+ g_assert_cmpint(fd, >=, 0);
570
+ ret = ftruncate(fd, TEST_IMAGE_SIZE);
571
+ g_assert_cmpint(ret, ==, 0);
572
+ close(fd);
573
+
574
+ g_test_queue_destroy(drive_destroy, t_path);
575
+ return t_path;
576
+}
577
+
578
+static void *ufs_blk_test_setup(GString *cmd_line, void *arg)
579
+{
580
+ char *tmp_path = drive_create();
581
+
582
+ g_string_append_printf(cmd_line,
583
+ " -blockdev file,filename=%s,node-name=drv1 "
584
+ "-device ufs-lu,bus=ufs0,drive=drv1,lun=1 ",
585
+ tmp_path);
586
+
587
+ return arg;
588
+}
589
+
590
+static void ufs_register_nodes(void)
591
+{
592
+ const char *arch;
593
+ QOSGraphEdgeOptions edge_opts = {
594
+ .before_cmd_line = "-blockdev null-co,node-name=drv0,read-zeroes=on",
595
+ .after_cmd_line = "-device ufs-lu,bus=ufs0,drive=drv0,lun=0",
596
+ .extra_device_opts = "addr=04.0,id=ufs0,nutrs=32,nutmrs=8"
597
+ };
598
+
599
+ QOSGraphTestOptions io_test_opts = {
600
+ .before = ufs_blk_test_setup,
601
+ };
602
+
603
+ add_qpci_address(&edge_opts, &(QPCIAddress){ .devfn = QPCI_DEVFN(4, 0) });
604
+
605
+ qos_node_create_driver("ufs", ufs_create);
606
+ qos_node_consumes("ufs", "pci-bus", &edge_opts);
607
+ qos_node_produces("ufs", "pci-device");
608
+
609
+ qos_add_test("reg-read", "ufs", ufstest_reg_read, NULL);
610
+
611
+ /*
612
+ * Check architecture
613
+ * TODO: Enable ufs io tests for ppc64
614
+ */
615
+ arch = qtest_get_arch();
616
+ if (!strcmp(arch, "ppc64")) {
617
+ g_test_message("Skipping ufs io tests for ppc64");
618
+ return;
619
+ }
620
+ qos_add_test("init", "ufs", ufstest_init, NULL);
621
+ qos_add_test("read-write", "ufs", ufstest_read_write, &io_test_opts);
622
+}
623
+
624
+libqos_init(ufs_register_nodes);
625
diff --git a/tests/qtest/meson.build b/tests/qtest/meson.build
626
index XXXXXXX..XXXXXXX 100644
627
--- a/tests/qtest/meson.build
628
+++ b/tests/qtest/meson.build
629
@@ -XXX,XX +XXX,XX @@ qos_test_ss.add(
630
'virtio-iommu-test.c',
631
'vmxnet3-test.c',
632
'igb-test.c',
633
+ 'ufs-test.c',
634
)
635
636
if config_all_devices.has_key('CONFIG_VIRTIO_SERIAL')
637
--
409
--
638
2.41.0
410
2.9.3
411
412
diff view generated by jsdifflib
1
From: Jeuk Kim <jeuk20.kim@samsung.com>
1
From: Paolo Bonzini <pbonzini@redhat.com>
2
2
3
Universal Flash Storage (UFS) is a high-performance mass storage device
3
aio_co_wake provides the infrastructure to start a coroutine on a "home"
4
with a serial interface. It is primarily used as a high-performance
4
AioContext. It will be used by CoMutex and CoQueue, so that coroutines
5
data storage device for embedded applications.
5
don't jump from one context to another when they go to sleep on a
6
6
mutex or waitqueue. However, it can also be used as a more efficient
7
This commit contains code for UFS device to be recognized
7
alternative to one-shot bottom halves, and saves the effort of tracking
8
as a UFS PCI device.
8
which AioContext a coroutine is running on.
9
Patches to handle UFS logical unit and Transfer Request will follow.
9
10
10
aio_co_schedule is the part of aio_co_wake that starts a coroutine
11
Signed-off-by: Jeuk Kim <jeuk20.kim@samsung.com>
11
on a remove AioContext, but it is also useful to implement e.g.
12
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
12
bdrv_set_aio_context callbacks.
13
Message-id: 10232660d462ee5cd10cf673f1a9a1205fc8276c.1693980783.git.jeuk20.kim@gmail.com
13
14
The implementation of aio_co_schedule is based on a lock-free
15
multiple-producer, single-consumer queue. The multiple producers use
16
cmpxchg to add to a LIFO stack. The consumer (a per-AioContext bottom
17
half) grabs all items added so far, inverts the list to make it FIFO,
18
and goes through it one item at a time until it's empty. The data
19
structure was inspired by OSv, which uses it in the very code we'll
20
"port" to QEMU for the thread-safe CoMutex.
21
22
Most of the new code is really tests.
23
24
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
25
Reviewed-by: Fam Zheng <famz@redhat.com>
26
Message-id: 20170213135235.12274-3-pbonzini@redhat.com
14
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
27
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
15
---
28
---
16
MAINTAINERS | 6 +
29
tests/Makefile.include | 8 +-
17
docs/specs/pci-ids.rst | 2 +
30
include/block/aio.h | 32 +++++++
18
meson.build | 1 +
31
include/qemu/coroutine_int.h | 11 ++-
19
hw/ufs/trace.h | 1 +
32
tests/iothread.h | 25 +++++
20
hw/ufs/ufs.h | 42 ++
33
tests/iothread.c | 91 ++++++++++++++++++
21
include/block/ufs.h | 1090 ++++++++++++++++++++++++++++++++++++++
34
tests/test-aio-multithread.c | 213 +++++++++++++++++++++++++++++++++++++++++++
22
include/hw/pci/pci.h | 1 +
35
util/async.c | 65 +++++++++++++
23
include/hw/pci/pci_ids.h | 1 +
36
util/qemu-coroutine.c | 8 ++
24
hw/ufs/ufs.c | 278 ++++++++++
37
util/trace-events | 4 +
25
hw/Kconfig | 1 +
38
9 files changed, 453 insertions(+), 4 deletions(-)
26
hw/meson.build | 1 +
39
create mode 100644 tests/iothread.h
27
hw/ufs/Kconfig | 4 +
40
create mode 100644 tests/iothread.c
28
hw/ufs/meson.build | 1 +
41
create mode 100644 tests/test-aio-multithread.c
29
hw/ufs/trace-events | 32 ++
42
30
14 files changed, 1461 insertions(+)
43
diff --git a/tests/Makefile.include b/tests/Makefile.include
31
create mode 100644 hw/ufs/trace.h
32
create mode 100644 hw/ufs/ufs.h
33
create mode 100644 include/block/ufs.h
34
create mode 100644 hw/ufs/ufs.c
35
create mode 100644 hw/ufs/Kconfig
36
create mode 100644 hw/ufs/meson.build
37
create mode 100644 hw/ufs/trace-events
38
39
diff --git a/MAINTAINERS b/MAINTAINERS
40
index XXXXXXX..XXXXXXX 100644
44
index XXXXXXX..XXXXXXX 100644
41
--- a/MAINTAINERS
45
--- a/tests/Makefile.include
42
+++ b/MAINTAINERS
46
+++ b/tests/Makefile.include
43
@@ -XXX,XX +XXX,XX @@ F: tests/qtest/nvme-test.c
47
@@ -XXX,XX +XXX,XX @@ check-unit-y += tests/test-aio$(EXESUF)
44
F: docs/system/devices/nvme.rst
48
gcov-files-test-aio-y = util/async.c util/qemu-timer.o
45
T: git git://git.infradead.org/qemu-nvme.git nvme-next
49
gcov-files-test-aio-$(CONFIG_WIN32) += util/aio-win32.c
46
50
gcov-files-test-aio-$(CONFIG_POSIX) += util/aio-posix.c
47
+ufs
51
+check-unit-y += tests/test-aio-multithread$(EXESUF)
48
+M: Jeuk Kim <jeuk20.kim@samsung.com>
52
+gcov-files-test-aio-multithread-y = $(gcov-files-test-aio-y)
49
+S: Supported
53
+gcov-files-test-aio-multithread-y += util/qemu-coroutine.c tests/iothread.c
50
+F: hw/ufs/*
54
check-unit-y += tests/test-throttle$(EXESUF)
51
+F: include/block/ufs.h
55
-gcov-files-test-aio-$(CONFIG_WIN32) = aio-win32.c
52
+
56
-gcov-files-test-aio-$(CONFIG_POSIX) = aio-posix.c
53
megasas
57
check-unit-y += tests/test-thread-pool$(EXESUF)
54
M: Hannes Reinecke <hare@suse.com>
58
gcov-files-test-thread-pool-y = thread-pool.c
55
L: qemu-block@nongnu.org
59
gcov-files-test-hbitmap-y = util/hbitmap.c
56
diff --git a/docs/specs/pci-ids.rst b/docs/specs/pci-ids.rst
60
@@ -XXX,XX +XXX,XX @@ test-qapi-obj-y = tests/test-qapi-visit.o tests/test-qapi-types.o \
61
    $(test-qom-obj-y)
62
test-crypto-obj-y = $(crypto-obj-y) $(test-qom-obj-y)
63
test-io-obj-y = $(io-obj-y) $(test-crypto-obj-y)
64
-test-block-obj-y = $(block-obj-y) $(test-io-obj-y)
65
+test-block-obj-y = $(block-obj-y) $(test-io-obj-y) tests/iothread.o
66
67
tests/check-qint$(EXESUF): tests/check-qint.o $(test-util-obj-y)
68
tests/check-qstring$(EXESUF): tests/check-qstring.o $(test-util-obj-y)
69
@@ -XXX,XX +XXX,XX @@ tests/check-qom-proplist$(EXESUF): tests/check-qom-proplist.o $(test-qom-obj-y)
70
tests/test-char$(EXESUF): tests/test-char.o $(test-util-obj-y) $(qtest-obj-y) $(test-io-obj-y) $(chardev-obj-y)
71
tests/test-coroutine$(EXESUF): tests/test-coroutine.o $(test-block-obj-y)
72
tests/test-aio$(EXESUF): tests/test-aio.o $(test-block-obj-y)
73
+tests/test-aio-multithread$(EXESUF): tests/test-aio-multithread.o $(test-block-obj-y)
74
tests/test-throttle$(EXESUF): tests/test-throttle.o $(test-block-obj-y)
75
tests/test-blockjob$(EXESUF): tests/test-blockjob.o $(test-block-obj-y) $(test-util-obj-y)
76
tests/test-blockjob-txn$(EXESUF): tests/test-blockjob-txn.o $(test-block-obj-y) $(test-util-obj-y)
77
diff --git a/include/block/aio.h b/include/block/aio.h
57
index XXXXXXX..XXXXXXX 100644
78
index XXXXXXX..XXXXXXX 100644
58
--- a/docs/specs/pci-ids.rst
79
--- a/include/block/aio.h
59
+++ b/docs/specs/pci-ids.rst
80
+++ b/include/block/aio.h
60
@@ -XXX,XX +XXX,XX @@ PCI devices (other than virtio):
81
@@ -XXX,XX +XXX,XX @@ typedef void QEMUBHFunc(void *opaque);
61
PCI PVPanic device (``-device pvpanic-pci``)
82
typedef bool AioPollFn(void *opaque);
62
1b36:0012
83
typedef void IOHandler(void *opaque);
63
PCI ACPI ERST device (``-device acpi-erst``)
84
64
+1b36:0013
85
+struct Coroutine;
65
+ PCI UFS device (``-device ufs``)
86
struct ThreadPool;
66
87
struct LinuxAioState;
67
All these devices are documented in :doc:`index`.
88
68
89
@@ -XXX,XX +XXX,XX @@ struct AioContext {
69
diff --git a/meson.build b/meson.build
90
bool notified;
91
EventNotifier notifier;
92
93
+ QSLIST_HEAD(, Coroutine) scheduled_coroutines;
94
+ QEMUBH *co_schedule_bh;
95
+
96
/* Thread pool for performing work and receiving completion callbacks.
97
* Has its own locking.
98
*/
99
@@ -XXX,XX +XXX,XX @@ static inline bool aio_node_check(AioContext *ctx, bool is_external)
100
}
101
102
/**
103
+ * aio_co_schedule:
104
+ * @ctx: the aio context
105
+ * @co: the coroutine
106
+ *
107
+ * Start a coroutine on a remote AioContext.
108
+ *
109
+ * The coroutine must not be entered by anyone else while aio_co_schedule()
110
+ * is active. In addition the coroutine must have yielded unless ctx
111
+ * is the context in which the coroutine is running (i.e. the value of
112
+ * qemu_get_current_aio_context() from the coroutine itself).
113
+ */
114
+void aio_co_schedule(AioContext *ctx, struct Coroutine *co);
115
+
116
+/**
117
+ * aio_co_wake:
118
+ * @co: the coroutine
119
+ *
120
+ * Restart a coroutine on the AioContext where it was running last, thus
121
+ * preventing coroutines from jumping from one context to another when they
122
+ * go to sleep.
123
+ *
124
+ * aio_co_wake may be executed either in coroutine or non-coroutine
125
+ * context. The coroutine must not be entered by anyone else while
126
+ * aio_co_wake() is active.
127
+ */
128
+void aio_co_wake(struct Coroutine *co);
129
+
130
+/**
131
* Return the AioContext whose event loop runs in the current thread.
132
*
133
* If called from an IOThread this will be the IOThread's AioContext. If
134
diff --git a/include/qemu/coroutine_int.h b/include/qemu/coroutine_int.h
70
index XXXXXXX..XXXXXXX 100644
135
index XXXXXXX..XXXXXXX 100644
71
--- a/meson.build
136
--- a/include/qemu/coroutine_int.h
72
+++ b/meson.build
137
+++ b/include/qemu/coroutine_int.h
73
@@ -XXX,XX +XXX,XX @@ if have_system
138
@@ -XXX,XX +XXX,XX @@ struct Coroutine {
74
'hw/ssi',
139
CoroutineEntry *entry;
75
'hw/timer',
140
void *entry_arg;
76
'hw/tpm',
141
Coroutine *caller;
77
+ 'hw/ufs',
142
+
78
'hw/usb',
143
+ /* Only used when the coroutine has terminated. */
79
'hw/vfio',
144
QSLIST_ENTRY(Coroutine) pool_next;
80
'hw/virtio',
145
+
81
diff --git a/hw/ufs/trace.h b/hw/ufs/trace.h
146
size_t locks_held;
147
148
- /* Coroutines that should be woken up when we yield or terminate */
149
+ /* Coroutines that should be woken up when we yield or terminate.
150
+ * Only used when the coroutine is running.
151
+ */
152
QSIMPLEQ_HEAD(, Coroutine) co_queue_wakeup;
153
+
154
+ /* Only used when the coroutine has yielded. */
155
+ AioContext *ctx;
156
QSIMPLEQ_ENTRY(Coroutine) co_queue_next;
157
+ QSLIST_ENTRY(Coroutine) co_scheduled_next;
158
};
159
160
Coroutine *qemu_coroutine_new(void);
161
diff --git a/tests/iothread.h b/tests/iothread.h
82
new file mode 100644
162
new file mode 100644
83
index XXXXXXX..XXXXXXX
163
index XXXXXXX..XXXXXXX
84
--- /dev/null
164
--- /dev/null
85
+++ b/hw/ufs/trace.h
165
+++ b/tests/iothread.h
86
@@ -0,0 +1 @@
166
@@ -XXX,XX +XXX,XX @@
87
+#include "trace/trace-hw_ufs.h"
167
+/*
88
diff --git a/hw/ufs/ufs.h b/hw/ufs/ufs.h
168
+ * Event loop thread implementation for unit tests
169
+ *
170
+ * Copyright Red Hat Inc., 2013, 2016
171
+ *
172
+ * Authors:
173
+ * Stefan Hajnoczi <stefanha@redhat.com>
174
+ * Paolo Bonzini <pbonzini@redhat.com>
175
+ *
176
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
177
+ * See the COPYING file in the top-level directory.
178
+ */
179
+#ifndef TEST_IOTHREAD_H
180
+#define TEST_IOTHREAD_H
181
+
182
+#include "block/aio.h"
183
+#include "qemu/thread.h"
184
+
185
+typedef struct IOThread IOThread;
186
+
187
+IOThread *iothread_new(void);
188
+void iothread_join(IOThread *iothread);
189
+AioContext *iothread_get_aio_context(IOThread *iothread);
190
+
191
+#endif
192
diff --git a/tests/iothread.c b/tests/iothread.c
89
new file mode 100644
193
new file mode 100644
90
index XXXXXXX..XXXXXXX
194
index XXXXXXX..XXXXXXX
91
--- /dev/null
195
--- /dev/null
92
+++ b/hw/ufs/ufs.h
196
+++ b/tests/iothread.c
93
@@ -XXX,XX +XXX,XX @@
197
@@ -XXX,XX +XXX,XX @@
94
+/*
198
+/*
95
+ * QEMU UFS
199
+ * Event loop thread implementation for unit tests
96
+ *
200
+ *
97
+ * Copyright (c) 2023 Samsung Electronics Co., Ltd. All rights reserved.
201
+ * Copyright Red Hat Inc., 2013, 2016
98
+ *
202
+ *
99
+ * Written by Jeuk Kim <jeuk20.kim@samsung.com>
203
+ * Authors:
100
+ *
204
+ * Stefan Hajnoczi <stefanha@redhat.com>
101
+ * SPDX-License-Identifier: GPL-2.0-or-later
205
+ * Paolo Bonzini <pbonzini@redhat.com>
206
+ *
207
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
208
+ * See the COPYING file in the top-level directory.
209
+ *
102
+ */
210
+ */
103
+
211
+
104
+#ifndef HW_UFS_UFS_H
212
+#include "qemu/osdep.h"
105
+#define HW_UFS_UFS_H
213
+#include "qapi/error.h"
106
+
214
+#include "block/aio.h"
107
+#include "hw/pci/pci_device.h"
215
+#include "qemu/main-loop.h"
108
+#include "hw/scsi/scsi.h"
216
+#include "qemu/rcu.h"
109
+#include "block/ufs.h"
217
+#include "iothread.h"
110
+
218
+
111
+#define UFS_MAX_LUS 32
219
+struct IOThread {
112
+#define UFS_BLOCK_SIZE 4096
220
+ AioContext *ctx;
113
+
221
+
114
+typedef struct UfsParams {
222
+ QemuThread thread;
115
+ char *serial;
223
+ QemuMutex init_done_lock;
116
+ uint8_t nutrs; /* Number of UTP Transfer Request Slots */
224
+ QemuCond init_done_cond; /* is thread initialization done? */
117
+ uint8_t nutmrs; /* Number of UTP Task Management Request Slots */
225
+ bool stopping;
118
+} UfsParams;
226
+};
119
+
227
+
120
+typedef struct UfsHc {
228
+static __thread IOThread *my_iothread;
121
+ PCIDevice parent_obj;
229
+
122
+ MemoryRegion iomem;
230
+AioContext *qemu_get_current_aio_context(void)
123
+ UfsReg reg;
231
+{
124
+ UfsParams params;
232
+ return my_iothread ? my_iothread->ctx : qemu_get_aio_context();
125
+ uint32_t reg_size;
233
+}
126
+
234
+
127
+ qemu_irq irq;
235
+static void *iothread_run(void *opaque)
128
+ QEMUBH *doorbell_bh;
236
+{
129
+ QEMUBH *complete_bh;
237
+ IOThread *iothread = opaque;
130
+} UfsHc;
238
+
131
+
239
+ rcu_register_thread();
132
+#define TYPE_UFS "ufs"
240
+
133
+#define UFS(obj) OBJECT_CHECK(UfsHc, (obj), TYPE_UFS)
241
+ my_iothread = iothread;
134
+
242
+ qemu_mutex_lock(&iothread->init_done_lock);
135
+#endif /* HW_UFS_UFS_H */
243
+ iothread->ctx = aio_context_new(&error_abort);
136
diff --git a/include/block/ufs.h b/include/block/ufs.h
244
+ qemu_cond_signal(&iothread->init_done_cond);
245
+ qemu_mutex_unlock(&iothread->init_done_lock);
246
+
247
+ while (!atomic_read(&iothread->stopping)) {
248
+ aio_poll(iothread->ctx, true);
249
+ }
250
+
251
+ rcu_unregister_thread();
252
+ return NULL;
253
+}
254
+
255
+void iothread_join(IOThread *iothread)
256
+{
257
+ iothread->stopping = true;
258
+ aio_notify(iothread->ctx);
259
+ qemu_thread_join(&iothread->thread);
260
+ qemu_cond_destroy(&iothread->init_done_cond);
261
+ qemu_mutex_destroy(&iothread->init_done_lock);
262
+ aio_context_unref(iothread->ctx);
263
+ g_free(iothread);
264
+}
265
+
266
+IOThread *iothread_new(void)
267
+{
268
+ IOThread *iothread = g_new0(IOThread, 1);
269
+
270
+ qemu_mutex_init(&iothread->init_done_lock);
271
+ qemu_cond_init(&iothread->init_done_cond);
272
+ qemu_thread_create(&iothread->thread, NULL, iothread_run,
273
+ iothread, QEMU_THREAD_JOINABLE);
274
+
275
+ /* Wait for initialization to complete */
276
+ qemu_mutex_lock(&iothread->init_done_lock);
277
+ while (iothread->ctx == NULL) {
278
+ qemu_cond_wait(&iothread->init_done_cond,
279
+ &iothread->init_done_lock);
280
+ }
281
+ qemu_mutex_unlock(&iothread->init_done_lock);
282
+ return iothread;
283
+}
284
+
285
+AioContext *iothread_get_aio_context(IOThread *iothread)
286
+{
287
+ return iothread->ctx;
288
+}
289
diff --git a/tests/test-aio-multithread.c b/tests/test-aio-multithread.c
137
new file mode 100644
290
new file mode 100644
138
index XXXXXXX..XXXXXXX
291
index XXXXXXX..XXXXXXX
139
--- /dev/null
292
--- /dev/null
140
+++ b/include/block/ufs.h
293
+++ b/tests/test-aio-multithread.c
141
@@ -XXX,XX +XXX,XX @@
142
+/* SPDX-License-Identifier: GPL-2.0-or-later */
143
+
144
+#ifndef BLOCK_UFS_H
145
+#define BLOCK_UFS_H
146
+
147
+#include "hw/registerfields.h"
148
+
149
+typedef struct QEMU_PACKED UfsReg {
150
+ uint32_t cap;
151
+ uint32_t rsvd0;
152
+ uint32_t ver;
153
+ uint32_t rsvd1;
154
+ uint32_t hcpid;
155
+ uint32_t hcmid;
156
+ uint32_t ahit;
157
+ uint32_t rsvd2;
158
+ uint32_t is;
159
+ uint32_t ie;
160
+ uint32_t rsvd3[2];
161
+ uint32_t hcs;
162
+ uint32_t hce;
163
+ uint32_t uecpa;
164
+ uint32_t uecdl;
165
+ uint32_t uecn;
166
+ uint32_t uect;
167
+ uint32_t uecdme;
168
+ uint32_t utriacr;
169
+ uint32_t utrlba;
170
+ uint32_t utrlbau;
171
+ uint32_t utrldbr;
172
+ uint32_t utrlclr;
173
+ uint32_t utrlrsr;
174
+ uint32_t utrlcnr;
175
+ uint32_t rsvd4[2];
176
+ uint32_t utmrlba;
177
+ uint32_t utmrlbau;
178
+ uint32_t utmrldbr;
179
+ uint32_t utmrlclr;
180
+ uint32_t utmrlrsr;
181
+ uint32_t rsvd5[3];
182
+ uint32_t uiccmd;
183
+ uint32_t ucmdarg1;
184
+ uint32_t ucmdarg2;
185
+ uint32_t ucmdarg3;
186
+ uint32_t rsvd6[4];
187
+ uint32_t rsvd7[4];
188
+ uint32_t rsvd8[16];
189
+ uint32_t ccap;
190
+} UfsReg;
191
+
192
+REG32(CAP, offsetof(UfsReg, cap))
193
+ FIELD(CAP, NUTRS, 0, 5)
194
+ FIELD(CAP, RTT, 8, 8)
195
+ FIELD(CAP, NUTMRS, 16, 3)
196
+ FIELD(CAP, AUTOH8, 23, 1)
197
+ FIELD(CAP, 64AS, 24, 1)
198
+ FIELD(CAP, OODDS, 25, 1)
199
+ FIELD(CAP, UICDMETMS, 26, 1)
200
+ FIELD(CAP, CS, 28, 1)
201
+REG32(VER, offsetof(UfsReg, ver))
202
+REG32(HCPID, offsetof(UfsReg, hcpid))
203
+REG32(HCMID, offsetof(UfsReg, hcmid))
204
+REG32(AHIT, offsetof(UfsReg, ahit))
205
+REG32(IS, offsetof(UfsReg, is))
206
+ FIELD(IS, UTRCS, 0, 1)
207
+ FIELD(IS, UDEPRI, 1, 1)
208
+ FIELD(IS, UE, 2, 1)
209
+ FIELD(IS, UTMS, 3, 1)
210
+ FIELD(IS, UPMS, 4, 1)
211
+ FIELD(IS, UHXS, 5, 1)
212
+ FIELD(IS, UHES, 6, 1)
213
+ FIELD(IS, ULLS, 7, 1)
214
+ FIELD(IS, ULSS, 8, 1)
215
+ FIELD(IS, UTMRCS, 9, 1)
216
+ FIELD(IS, UCCS, 10, 1)
217
+ FIELD(IS, DFES, 11, 1)
218
+ FIELD(IS, UTPES, 12, 1)
219
+ FIELD(IS, HCFES, 16, 1)
220
+ FIELD(IS, SBFES, 17, 1)
221
+ FIELD(IS, CEFES, 18, 1)
222
+REG32(IE, offsetof(UfsReg, ie))
223
+ FIELD(IE, UTRCE, 0, 1)
224
+ FIELD(IE, UDEPRIE, 1, 1)
225
+ FIELD(IE, UEE, 2, 1)
226
+ FIELD(IE, UTMSE, 3, 1)
227
+ FIELD(IE, UPMSE, 4, 1)
228
+ FIELD(IE, UHXSE, 5, 1)
229
+ FIELD(IE, UHESE, 6, 1)
230
+ FIELD(IE, ULLSE, 7, 1)
231
+ FIELD(IE, ULSSE, 8, 1)
232
+ FIELD(IE, UTMRCE, 9, 1)
233
+ FIELD(IE, UCCE, 10, 1)
234
+ FIELD(IE, DFEE, 11, 1)
235
+ FIELD(IE, UTPEE, 12, 1)
236
+ FIELD(IE, HCFEE, 16, 1)
237
+ FIELD(IE, SBFEE, 17, 1)
238
+ FIELD(IE, CEFEE, 18, 1)
239
+REG32(HCS, offsetof(UfsReg, hcs))
240
+ FIELD(HCS, DP, 0, 1)
241
+ FIELD(HCS, UTRLRDY, 1, 1)
242
+ FIELD(HCS, UTMRLRDY, 2, 1)
243
+ FIELD(HCS, UCRDY, 3, 1)
244
+ FIELD(HCS, UPMCRS, 8, 3)
245
+REG32(HCE, offsetof(UfsReg, hce))
246
+ FIELD(HCE, HCE, 0, 1)
247
+ FIELD(HCE, CGE, 1, 1)
248
+REG32(UECPA, offsetof(UfsReg, uecpa))
249
+REG32(UECDL, offsetof(UfsReg, uecdl))
250
+REG32(UECN, offsetof(UfsReg, uecn))
251
+REG32(UECT, offsetof(UfsReg, uect))
252
+REG32(UECDME, offsetof(UfsReg, uecdme))
253
+REG32(UTRIACR, offsetof(UfsReg, utriacr))
254
+REG32(UTRLBA, offsetof(UfsReg, utrlba))
255
+ FIELD(UTRLBA, UTRLBA, 9, 22)
256
+REG32(UTRLBAU, offsetof(UfsReg, utrlbau))
257
+REG32(UTRLDBR, offsetof(UfsReg, utrldbr))
258
+REG32(UTRLCLR, offsetof(UfsReg, utrlclr))
259
+REG32(UTRLRSR, offsetof(UfsReg, utrlrsr))
260
+REG32(UTRLCNR, offsetof(UfsReg, utrlcnr))
261
+REG32(UTMRLBA, offsetof(UfsReg, utmrlba))
262
+ FIELD(UTMRLBA, UTMRLBA, 9, 22)
263
+REG32(UTMRLBAU, offsetof(UfsReg, utmrlbau))
264
+REG32(UTMRLDBR, offsetof(UfsReg, utmrldbr))
265
+REG32(UTMRLCLR, offsetof(UfsReg, utmrlclr))
266
+REG32(UTMRLRSR, offsetof(UfsReg, utmrlrsr))
267
+REG32(UICCMD, offsetof(UfsReg, uiccmd))
268
+REG32(UCMDARG1, offsetof(UfsReg, ucmdarg1))
269
+REG32(UCMDARG2, offsetof(UfsReg, ucmdarg2))
270
+REG32(UCMDARG3, offsetof(UfsReg, ucmdarg3))
271
+REG32(CCAP, offsetof(UfsReg, ccap))
272
+
273
+#define UFS_INTR_MASK \
274
+ ((1 << R_IS_CEFES_SHIFT) | (1 << R_IS_SBFES_SHIFT) | \
275
+ (1 << R_IS_HCFES_SHIFT) | (1 << R_IS_UTPES_SHIFT) | \
276
+ (1 << R_IS_DFES_SHIFT) | (1 << R_IS_UCCS_SHIFT) | \
277
+ (1 << R_IS_UTMRCS_SHIFT) | (1 << R_IS_ULSS_SHIFT) | \
278
+ (1 << R_IS_ULLS_SHIFT) | (1 << R_IS_UHES_SHIFT) | \
279
+ (1 << R_IS_UHXS_SHIFT) | (1 << R_IS_UPMS_SHIFT) | \
280
+ (1 << R_IS_UTMS_SHIFT) | (1 << R_IS_UE_SHIFT) | \
281
+ (1 << R_IS_UDEPRI_SHIFT) | (1 << R_IS_UTRCS_SHIFT))
282
+
283
+#define UFS_UPIU_HEADER_TRANSACTION_TYPE_SHIFT 24
284
+#define UFS_UPIU_HEADER_TRANSACTION_TYPE_MASK 0xff
285
+#define UFS_UPIU_HEADER_TRANSACTION_TYPE(dword0) \
286
+ ((be32_to_cpu(dword0) >> UFS_UPIU_HEADER_TRANSACTION_TYPE_SHIFT) & \
287
+ UFS_UPIU_HEADER_TRANSACTION_TYPE_MASK)
288
+
289
+#define UFS_UPIU_HEADER_QUERY_FUNC_SHIFT 16
290
+#define UFS_UPIU_HEADER_QUERY_FUNC_MASK 0xff
291
+#define UFS_UPIU_HEADER_QUERY_FUNC(dword1) \
292
+ ((be32_to_cpu(dword1) >> UFS_UPIU_HEADER_QUERY_FUNC_SHIFT) & \
293
+ UFS_UPIU_HEADER_QUERY_FUNC_MASK)
294
+
295
+#define UFS_UPIU_HEADER_DATA_SEGMENT_LENGTH_SHIFT 0
296
+#define UFS_UPIU_HEADER_DATA_SEGMENT_LENGTH_MASK 0xffff
297
+#define UFS_UPIU_HEADER_DATA_SEGMENT_LENGTH(dword2) \
298
+ ((be32_to_cpu(dword2) >> UFS_UPIU_HEADER_DATA_SEGMENT_LENGTH_SHIFT) & \
299
+ UFS_UPIU_HEADER_DATA_SEGMENT_LENGTH_MASK)
300
+
301
+typedef struct QEMU_PACKED DeviceDescriptor {
302
+ uint8_t length;
303
+ uint8_t descriptor_idn;
304
+ uint8_t device;
305
+ uint8_t device_class;
306
+ uint8_t device_sub_class;
307
+ uint8_t protocol;
308
+ uint8_t number_lu;
309
+ uint8_t number_wlu;
310
+ uint8_t boot_enable;
311
+ uint8_t descr_access_en;
312
+ uint8_t init_power_mode;
313
+ uint8_t high_priority_lun;
314
+ uint8_t secure_removal_type;
315
+ uint8_t security_lu;
316
+ uint8_t background_ops_term_lat;
317
+ uint8_t init_active_icc_level;
318
+ uint16_t spec_version;
319
+ uint16_t manufacture_date;
320
+ uint8_t manufacturer_name;
321
+ uint8_t product_name;
322
+ uint8_t serial_number;
323
+ uint8_t oem_id;
324
+ uint16_t manufacturer_id;
325
+ uint8_t ud_0_base_offset;
326
+ uint8_t ud_config_p_length;
327
+ uint8_t device_rtt_cap;
328
+ uint16_t periodic_rtc_update;
329
+ uint8_t ufs_features_support;
330
+ uint8_t ffu_timeout;
331
+ uint8_t queue_depth;
332
+ uint16_t device_version;
333
+ uint8_t num_secure_wp_area;
334
+ uint32_t psa_max_data_size;
335
+ uint8_t psa_state_timeout;
336
+ uint8_t product_revision_level;
337
+ uint8_t reserved[36];
338
+ uint32_t extended_ufs_features_support;
339
+ uint8_t write_booster_buffer_preserve_user_space_en;
340
+ uint8_t write_booster_buffer_type;
341
+ uint32_t num_shared_write_booster_buffer_alloc_units;
342
+} DeviceDescriptor;
343
+
344
+typedef struct QEMU_PACKED GeometryDescriptor {
345
+ uint8_t length;
346
+ uint8_t descriptor_idn;
347
+ uint8_t media_technology;
348
+ uint8_t reserved;
349
+ uint64_t total_raw_device_capacity;
350
+ uint8_t max_number_lu;
351
+ uint32_t segment_size;
352
+ uint8_t allocation_unit_size;
353
+ uint8_t min_addr_block_size;
354
+ uint8_t optimal_read_block_size;
355
+ uint8_t optimal_write_block_size;
356
+ uint8_t max_in_buffer_size;
357
+ uint8_t max_out_buffer_size;
358
+ uint8_t rpmb_read_write_size;
359
+ uint8_t dynamic_capacity_resource_policy;
360
+ uint8_t data_ordering;
361
+ uint8_t max_context_id_number;
362
+ uint8_t sys_data_tag_unit_size;
363
+ uint8_t sys_data_tag_res_size;
364
+ uint8_t supported_sec_r_types;
365
+ uint16_t supported_memory_types;
366
+ uint32_t system_code_max_n_alloc_u;
367
+ uint16_t system_code_cap_adj_fac;
368
+ uint32_t non_persist_max_n_alloc_u;
369
+ uint16_t non_persist_cap_adj_fac;
370
+ uint32_t enhanced_1_max_n_alloc_u;
371
+ uint16_t enhanced_1_cap_adj_fac;
372
+ uint32_t enhanced_2_max_n_alloc_u;
373
+ uint16_t enhanced_2_cap_adj_fac;
374
+ uint32_t enhanced_3_max_n_alloc_u;
375
+ uint16_t enhanced_3_cap_adj_fac;
376
+ uint32_t enhanced_4_max_n_alloc_u;
377
+ uint16_t enhanced_4_cap_adj_fac;
378
+ uint32_t optimal_logical_block_size;
379
+ uint8_t reserved2[7];
380
+ uint32_t write_booster_buffer_max_n_alloc_units;
381
+ uint8_t device_max_write_booster_l_us;
382
+ uint8_t write_booster_buffer_cap_adj_fac;
383
+ uint8_t supported_write_booster_buffer_user_space_reduction_types;
384
+ uint8_t supported_write_booster_buffer_types;
385
+} GeometryDescriptor;
386
+
387
+#define UFS_GEOMETRY_CAPACITY_SHIFT 9
388
+
389
+typedef struct QEMU_PACKED UnitDescriptor {
390
+ uint8_t length;
391
+ uint8_t descriptor_idn;
392
+ uint8_t unit_index;
393
+ uint8_t lu_enable;
394
+ uint8_t boot_lun_id;
395
+ uint8_t lu_write_protect;
396
+ uint8_t lu_queue_depth;
397
+ uint8_t psa_sensitive;
398
+ uint8_t memory_type;
399
+ uint8_t data_reliability;
400
+ uint8_t logical_block_size;
401
+ uint64_t logical_block_count;
402
+ uint32_t erase_block_size;
403
+ uint8_t provisioning_type;
404
+ uint64_t phy_mem_resource_count;
405
+ uint16_t context_capabilities;
406
+ uint8_t large_unit_granularity_m1;
407
+ uint8_t reserved[6];
408
+ uint32_t lu_num_write_booster_buffer_alloc_units;
409
+} UnitDescriptor;
410
+
411
+typedef struct QEMU_PACKED RpmbUnitDescriptor {
412
+ uint8_t length;
413
+ uint8_t descriptor_idn;
414
+ uint8_t unit_index;
415
+ uint8_t lu_enable;
416
+ uint8_t boot_lun_id;
417
+ uint8_t lu_write_protect;
418
+ uint8_t lu_queue_depth;
419
+ uint8_t psa_sensitive;
420
+ uint8_t memory_type;
421
+ uint8_t reserved;
422
+ uint8_t logical_block_size;
423
+ uint64_t logical_block_count;
424
+ uint32_t erase_block_size;
425
+ uint8_t provisioning_type;
426
+ uint64_t phy_mem_resource_count;
427
+ uint8_t reserved2[3];
428
+} RpmbUnitDescriptor;
429
+
430
+typedef struct QEMU_PACKED PowerParametersDescriptor {
431
+ uint8_t length;
432
+ uint8_t descriptor_idn;
433
+ uint16_t active_icc_levels_vcc[16];
434
+ uint16_t active_icc_levels_vccq[16];
435
+ uint16_t active_icc_levels_vccq_2[16];
436
+} PowerParametersDescriptor;
437
+
438
+typedef struct QEMU_PACKED InterconnectDescriptor {
439
+ uint8_t length;
440
+ uint8_t descriptor_idn;
441
+ uint16_t bcd_unipro_version;
442
+ uint16_t bcd_mphy_version;
443
+} InterconnectDescriptor;
444
+
445
+typedef struct QEMU_PACKED StringDescriptor {
446
+ uint8_t length;
447
+ uint8_t descriptor_idn;
448
+ uint16_t UC[126];
449
+} StringDescriptor;
450
+
451
+typedef struct QEMU_PACKED DeviceHealthDescriptor {
452
+ uint8_t length;
453
+ uint8_t descriptor_idn;
454
+ uint8_t pre_eol_info;
455
+ uint8_t device_life_time_est_a;
456
+ uint8_t device_life_time_est_b;
457
+ uint8_t vendor_prop_info[32];
458
+ uint32_t refresh_total_count;
459
+ uint32_t refresh_progress;
460
+} DeviceHealthDescriptor;
461
+
462
+typedef struct QEMU_PACKED Flags {
463
+ uint8_t reserved;
464
+ uint8_t device_init;
465
+ uint8_t permanent_wp_en;
466
+ uint8_t power_on_wp_en;
467
+ uint8_t background_ops_en;
468
+ uint8_t device_life_span_mode_en;
469
+ uint8_t purge_enable;
470
+ uint8_t refresh_enable;
471
+ uint8_t phy_resource_removal;
472
+ uint8_t busy_rtc;
473
+ uint8_t reserved2;
474
+ uint8_t permanently_disable_fw_update;
475
+ uint8_t reserved3[2];
476
+ uint8_t wb_en;
477
+ uint8_t wb_buffer_flush_en;
478
+ uint8_t wb_buffer_flush_during_hibernate;
479
+ uint8_t reserved4[2];
480
+} Flags;
481
+
482
+typedef struct Attributes {
483
+ uint8_t boot_lun_en;
484
+ uint8_t reserved;
485
+ uint8_t current_power_mode;
486
+ uint8_t active_icc_level;
487
+ uint8_t out_of_order_data_en;
488
+ uint8_t background_op_status;
489
+ uint8_t purge_status;
490
+ uint8_t max_data_in_size;
491
+ uint8_t max_data_out_size;
492
+ uint32_t dyn_cap_needed;
493
+ uint8_t ref_clk_freq;
494
+ uint8_t config_descr_lock;
495
+ uint8_t max_num_of_rtt;
496
+ uint16_t exception_event_control;
497
+ uint16_t exception_event_status;
498
+ uint32_t seconds_passed;
499
+ uint16_t context_conf;
500
+ uint8_t device_ffu_status;
501
+ uint8_t psa_state;
502
+ uint32_t psa_data_size;
503
+ uint8_t ref_clk_gating_wait_time;
504
+ uint8_t device_case_rough_temperaure;
505
+ uint8_t device_too_high_temp_boundary;
506
+ uint8_t device_too_low_temp_boundary;
507
+ uint8_t throttling_status;
508
+ uint8_t wb_buffer_flush_status;
509
+ uint8_t available_wb_buffer_size;
510
+ uint8_t wb_buffer_life_time_est;
511
+ uint32_t current_wb_buffer_size;
512
+ uint8_t refresh_status;
513
+ uint8_t refresh_freq;
514
+ uint8_t refresh_unit;
515
+ uint8_t refresh_method;
516
+} Attributes;
517
+
518
+#define UFS_TRANSACTION_SPECIFIC_FIELD_SIZE 20
519
+#define UFS_MAX_QUERY_DATA_SIZE 256
520
+
521
+/* Command response result code */
522
+typedef enum CommandRespCode {
523
+ UFS_COMMAND_RESULT_SUCESS = 0x00,
524
+ UFS_COMMAND_RESULT_FAIL = 0x01,
525
+} CommandRespCode;
526
+
527
+enum {
528
+ UFS_UPIU_FLAG_UNDERFLOW = 0x20,
529
+ UFS_UPIU_FLAG_OVERFLOW = 0x40,
530
+};
531
+
532
+typedef struct QEMU_PACKED UtpUpiuHeader {
533
+ uint8_t trans_type;
534
+ uint8_t flags;
535
+ uint8_t lun;
536
+ uint8_t task_tag;
537
+ uint8_t iid_cmd_set_type;
538
+ uint8_t query_func;
539
+ uint8_t response;
540
+ uint8_t scsi_status;
541
+ uint8_t ehs_len;
542
+ uint8_t device_inf;
543
+ uint16_t data_segment_length;
544
+} UtpUpiuHeader;
545
+
546
+/*
547
+ * The code below is copied from the linux kernel
548
+ * ("include/uapi/scsi/scsi_bsg_ufs.h") and modified to fit the qemu style.
549
+ */
550
+
551
+typedef struct QEMU_PACKED UtpUpiuQuery {
552
+ uint8_t opcode;
553
+ uint8_t idn;
554
+ uint8_t index;
555
+ uint8_t selector;
556
+ uint16_t reserved_osf;
557
+ uint16_t length;
558
+ uint32_t value;
559
+ uint32_t reserved[2];
560
+ /* EHS length should be 0. We don't have to worry about EHS area. */
561
+ uint8_t data[UFS_MAX_QUERY_DATA_SIZE];
562
+} UtpUpiuQuery;
563
+
564
+#define UFS_CDB_SIZE 16
565
+
566
+/*
567
+ * struct UtpUpiuCmd - Command UPIU structure
568
+ * @data_transfer_len: Data Transfer Length DW-3
569
+ * @cdb: Command Descriptor Block CDB DW-4 to DW-7
570
+ */
571
+typedef struct QEMU_PACKED UtpUpiuCmd {
572
+ uint32_t exp_data_transfer_len;
573
+ uint8_t cdb[UFS_CDB_SIZE];
574
+} UtpUpiuCmd;
575
+
576
+/*
577
+ * struct UtpUpiuReq - general upiu request structure
578
+ * @header:UPIU header structure DW-0 to DW-2
579
+ * @sc: fields structure for scsi command DW-3 to DW-7
580
+ * @qr: fields structure for query request DW-3 to DW-7
581
+ * @uc: use utp_upiu_query to host the 4 dwords of uic command
582
+ */
583
+typedef struct QEMU_PACKED UtpUpiuReq {
584
+ UtpUpiuHeader header;
585
+ union {
586
+ UtpUpiuCmd sc;
587
+ UtpUpiuQuery qr;
588
+ };
589
+} UtpUpiuReq;
590
+
591
+/*
592
+ * The code below is copied from the linux kernel ("include/ufs/ufshci.h") and
593
+ * modified to fit the qemu style.
594
+ */
595
+
596
+enum {
597
+ UFS_PWR_OK = 0x0,
598
+ UFS_PWR_LOCAL = 0x01,
599
+ UFS_PWR_REMOTE = 0x02,
600
+ UFS_PWR_BUSY = 0x03,
601
+ UFS_PWR_ERROR_CAP = 0x04,
602
+ UFS_PWR_FATAL_ERROR = 0x05,
603
+};
604
+
605
+/* UIC Commands */
606
+enum uic_cmd_dme {
607
+ UFS_UIC_CMD_DME_GET = 0x01,
608
+ UFS_UIC_CMD_DME_SET = 0x02,
609
+ UFS_UIC_CMD_DME_PEER_GET = 0x03,
610
+ UFS_UIC_CMD_DME_PEER_SET = 0x04,
611
+ UFS_UIC_CMD_DME_POWERON = 0x10,
612
+ UFS_UIC_CMD_DME_POWEROFF = 0x11,
613
+ UFS_UIC_CMD_DME_ENABLE = 0x12,
614
+ UFS_UIC_CMD_DME_RESET = 0x14,
615
+ UFS_UIC_CMD_DME_END_PT_RST = 0x15,
616
+ UFS_UIC_CMD_DME_LINK_STARTUP = 0x16,
617
+ UFS_UIC_CMD_DME_HIBER_ENTER = 0x17,
618
+ UFS_UIC_CMD_DME_HIBER_EXIT = 0x18,
619
+ UFS_UIC_CMD_DME_TEST_MODE = 0x1A,
620
+};
621
+
622
+/* UIC Config result code / Generic error code */
623
+enum {
624
+ UFS_UIC_CMD_RESULT_SUCCESS = 0x00,
625
+ UFS_UIC_CMD_RESULT_INVALID_ATTR = 0x01,
626
+ UFS_UIC_CMD_RESULT_FAILURE = 0x01,
627
+ UFS_UIC_CMD_RESULT_INVALID_ATTR_VALUE = 0x02,
628
+ UFS_UIC_CMD_RESULT_READ_ONLY_ATTR = 0x03,
629
+ UFS_UIC_CMD_RESULT_WRITE_ONLY_ATTR = 0x04,
630
+ UFS_UIC_CMD_RESULT_BAD_INDEX = 0x05,
631
+ UFS_UIC_CMD_RESULT_LOCKED_ATTR = 0x06,
632
+ UFS_UIC_CMD_RESULT_BAD_TEST_FEATURE_INDEX = 0x07,
633
+ UFS_UIC_CMD_RESULT_PEER_COMM_FAILURE = 0x08,
634
+ UFS_UIC_CMD_RESULT_BUSY = 0x09,
635
+ UFS_UIC_CMD_RESULT_DME_FAILURE = 0x0A,
636
+};
637
+
638
+#define UFS_MASK_UIC_COMMAND_RESULT 0xFF
639
+
640
+/*
641
+ * Request Descriptor Definitions
642
+ */
643
+
644
+/* Transfer request command type */
645
+enum {
646
+ UFS_UTP_CMD_TYPE_SCSI = 0x0,
647
+ UFS_UTP_CMD_TYPE_UFS = 0x1,
648
+ UFS_UTP_CMD_TYPE_DEV_MANAGE = 0x2,
649
+};
650
+
651
+/* To accommodate UFS2.0 required Command type */
652
+enum {
653
+ UFS_UTP_CMD_TYPE_UFS_STORAGE = 0x1,
654
+};
655
+
656
+enum {
657
+ UFS_UTP_SCSI_COMMAND = 0x00000000,
658
+ UFS_UTP_NATIVE_UFS_COMMAND = 0x10000000,
659
+ UFS_UTP_DEVICE_MANAGEMENT_FUNCTION = 0x20000000,
660
+ UFS_UTP_REQ_DESC_INT_CMD = 0x01000000,
661
+ UFS_UTP_REQ_DESC_CRYPTO_ENABLE_CMD = 0x00800000,
662
+};
663
+
664
+/* UTP Transfer Request Data Direction (DD) */
665
+enum {
666
+ UFS_UTP_NO_DATA_TRANSFER = 0x00000000,
667
+ UFS_UTP_HOST_TO_DEVICE = 0x02000000,
668
+ UFS_UTP_DEVICE_TO_HOST = 0x04000000,
669
+};
670
+
671
+/* Overall command status values */
672
+enum UtpOcsCodes {
673
+ UFS_OCS_SUCCESS = 0x0,
674
+ UFS_OCS_INVALID_CMD_TABLE_ATTR = 0x1,
675
+ UFS_OCS_INVALID_PRDT_ATTR = 0x2,
676
+ UFS_OCS_MISMATCH_DATA_BUF_SIZE = 0x3,
677
+ UFS_OCS_MISMATCH_RESP_UPIU_SIZE = 0x4,
678
+ UFS_OCS_PEER_COMM_FAILURE = 0x5,
679
+ UFS_OCS_ABORTED = 0x6,
680
+ UFS_OCS_FATAL_ERROR = 0x7,
681
+ UFS_OCS_DEVICE_FATAL_ERROR = 0x8,
682
+ UFS_OCS_INVALID_CRYPTO_CONFIG = 0x9,
683
+ UFS_OCS_GENERAL_CRYPTO_ERROR = 0xa,
684
+ UFS_OCS_INVALID_COMMAND_STATUS = 0xf,
685
+};
686
+
687
+enum {
688
+ UFS_MASK_OCS = 0x0F,
689
+};
690
+
691
+/*
692
+ * struct UfshcdSgEntry - UFSHCI PRD Entry
693
+ * @addr: Physical address; DW-0 and DW-1.
694
+ * @reserved: Reserved for future use DW-2
695
+ * @size: size of physical segment DW-3
696
+ */
697
+typedef struct QEMU_PACKED UfshcdSgEntry {
698
+ uint64_t addr;
699
+ uint32_t reserved;
700
+ uint32_t size;
701
+ /*
702
+ * followed by variant-specific fields if
703
+ * CONFIG_SCSI_UFS_VARIABLE_SG_ENTRY_SIZE has been defined.
704
+ */
705
+} UfshcdSgEntry;
706
+
707
+/*
708
+ * struct RequestDescHeader - Descriptor Header common to both UTRD and UTMRD
709
+ * @dword0: Descriptor Header DW0
710
+ * @dword1: Descriptor Header DW1
711
+ * @dword2: Descriptor Header DW2
712
+ * @dword3: Descriptor Header DW3
713
+ */
714
+typedef struct QEMU_PACKED RequestDescHeader {
715
+ uint32_t dword_0;
716
+ uint32_t dword_1;
717
+ uint32_t dword_2;
718
+ uint32_t dword_3;
719
+} RequestDescHeader;
720
+
721
+/*
722
+ * struct UtpTransferReqDesc - UTP Transfer Request Descriptor (UTRD)
723
+ * @header: UTRD header DW-0 to DW-3
724
+ * @command_desc_base_addr_lo: UCD base address low DW-4
725
+ * @command_desc_base_addr_hi: UCD base address high DW-5
726
+ * @response_upiu_length: response UPIU length DW-6
727
+ * @response_upiu_offset: response UPIU offset DW-6
728
+ * @prd_table_length: Physical region descriptor length DW-7
729
+ * @prd_table_offset: Physical region descriptor offset DW-7
730
+ */
731
+typedef struct QEMU_PACKED UtpTransferReqDesc {
732
+ /* DW 0-3 */
733
+ RequestDescHeader header;
734
+
735
+ /* DW 4-5*/
736
+ uint32_t command_desc_base_addr_lo;
737
+ uint32_t command_desc_base_addr_hi;
738
+
739
+ /* DW 6 */
740
+ uint16_t response_upiu_length;
741
+ uint16_t response_upiu_offset;
742
+
743
+ /* DW 7 */
744
+ uint16_t prd_table_length;
745
+ uint16_t prd_table_offset;
746
+} UtpTransferReqDesc;
747
+
748
+/*
749
+ * UTMRD structure.
750
+ */
751
+typedef struct QEMU_PACKED UtpTaskReqDesc {
752
+ /* DW 0-3 */
753
+ RequestDescHeader header;
754
+
755
+ /* DW 4-11 - Task request UPIU structure */
756
+ struct {
757
+ UtpUpiuHeader req_header;
758
+ uint32_t input_param1;
759
+ uint32_t input_param2;
760
+ uint32_t input_param3;
761
+ uint32_t reserved1[2];
762
+ } upiu_req;
763
+
764
+ /* DW 12-19 - Task Management Response UPIU structure */
765
+ struct {
766
+ UtpUpiuHeader rsp_header;
767
+ uint32_t output_param1;
768
+ uint32_t output_param2;
769
+ uint32_t reserved2[3];
770
+ } upiu_rsp;
771
+} UtpTaskReqDesc;
772
+
773
+/*
774
+ * The code below is copied from the linux kernel ("include/ufs/ufs.h") and
775
+ * modified to fit the qemu style.
776
+ */
777
+
778
+#define UFS_GENERAL_UPIU_REQUEST_SIZE (sizeof(UtpUpiuReq))
779
+#define UFS_QUERY_DESC_MAX_SIZE 255
780
+#define UFS_QUERY_DESC_MIN_SIZE 2
781
+#define UFS_QUERY_DESC_HDR_SIZE 2
782
+#define UFS_QUERY_OSF_SIZE (GENERAL_UPIU_REQUEST_SIZE - (sizeof(UtpUpiuHeader)))
783
+#define UFS_SENSE_SIZE 18
784
+
785
+/*
786
+ * UFS device may have standard LUs and LUN id could be from 0x00 to
787
+ * 0x7F. Standard LUs use "Peripheral Device Addressing Format".
788
+ * UFS device may also have the Well Known LUs (also referred as W-LU)
789
+ * which again could be from 0x00 to 0x7F. For W-LUs, device only use
790
+ * the "Extended Addressing Format" which means the W-LUNs would be
791
+ * from 0xc100 (SCSI_W_LUN_BASE) onwards.
792
+ * This means max. LUN number reported from UFS device could be 0xC17F.
793
+ */
794
+#define UFS_UPIU_MAX_UNIT_NUM_ID 0x7F
795
+#define UFS_UPIU_WLUN_ID (1 << 7)
796
+
797
+/* WriteBooster buffer is available only for the logical unit from 0 to 7 */
798
+#define UFS_UPIU_MAX_WB_LUN_ID 8
799
+
800
+/*
801
+ * WriteBooster buffer lifetime has a limit setted by vendor.
802
+ * If it is over the limit, WriteBooster feature will be disabled.
803
+ */
804
+#define UFS_WB_EXCEED_LIFETIME 0x0B
805
+
806
+/*
807
+ * In UFS Spec, the Extra Header Segment (EHS) starts from byte 32 in UPIU
808
+ * request/response packet
809
+ */
810
+#define UFS_EHS_OFFSET_IN_RESPONSE 32
811
+
812
+/* Well known logical unit id in LUN field of UPIU */
813
+enum {
814
+ UFS_UPIU_REPORT_LUNS_WLUN = 0x81,
815
+ UFS_UPIU_UFS_DEVICE_WLUN = 0xD0,
816
+ UFS_UPIU_BOOT_WLUN = 0xB0,
817
+ UFS_UPIU_RPMB_WLUN = 0xC4,
818
+};
819
+
820
+/*
821
+ * UFS Protocol Information Unit related definitions
822
+ */
823
+
824
+/* Task management functions */
825
+enum {
826
+ UFS_ABORT_TASK = 0x01,
827
+ UFS_ABORT_TASK_SET = 0x02,
828
+ UFS_CLEAR_TASK_SET = 0x04,
829
+ UFS_LOGICAL_RESET = 0x08,
830
+ UFS_QUERY_TASK = 0x80,
831
+ UFS_QUERY_TASK_SET = 0x81,
832
+};
833
+
834
+/* UTP UPIU Transaction Codes Initiator to Target */
835
+enum {
836
+ UFS_UPIU_TRANSACTION_NOP_OUT = 0x00,
837
+ UFS_UPIU_TRANSACTION_COMMAND = 0x01,
838
+ UFS_UPIU_TRANSACTION_DATA_OUT = 0x02,
839
+ UFS_UPIU_TRANSACTION_TASK_REQ = 0x04,
840
+ UFS_UPIU_TRANSACTION_QUERY_REQ = 0x16,
841
+};
842
+
843
+/* UTP UPIU Transaction Codes Target to Initiator */
844
+enum {
845
+ UFS_UPIU_TRANSACTION_NOP_IN = 0x20,
846
+ UFS_UPIU_TRANSACTION_RESPONSE = 0x21,
847
+ UFS_UPIU_TRANSACTION_DATA_IN = 0x22,
848
+ UFS_UPIU_TRANSACTION_TASK_RSP = 0x24,
849
+ UFS_UPIU_TRANSACTION_READY_XFER = 0x31,
850
+ UFS_UPIU_TRANSACTION_QUERY_RSP = 0x36,
851
+ UFS_UPIU_TRANSACTION_REJECT_UPIU = 0x3F,
852
+};
853
+
854
+/* UPIU Read/Write flags */
855
+enum {
856
+ UFS_UPIU_CMD_FLAGS_NONE = 0x00,
857
+ UFS_UPIU_CMD_FLAGS_WRITE = 0x20,
858
+ UFS_UPIU_CMD_FLAGS_READ = 0x40,
859
+};
860
+
861
+/* UPIU Task Attributes */
862
+enum {
863
+ UFS_UPIU_TASK_ATTR_SIMPLE = 0x00,
864
+ UFS_UPIU_TASK_ATTR_ORDERED = 0x01,
865
+ UFS_UPIU_TASK_ATTR_HEADQ = 0x02,
866
+ UFS_UPIU_TASK_ATTR_ACA = 0x03,
867
+};
868
+
869
+/* UPIU Query request function */
870
+enum {
871
+ UFS_UPIU_QUERY_FUNC_STANDARD_READ_REQUEST = 0x01,
872
+ UFS_UPIU_QUERY_FUNC_STANDARD_WRITE_REQUEST = 0x81,
873
+};
874
+
875
+/* Flag idn for Query Requests*/
876
+enum flag_idn {
877
+ UFS_QUERY_FLAG_IDN_FDEVICEINIT = 0x01,
878
+ UFS_QUERY_FLAG_IDN_PERMANENT_WPE = 0x02,
879
+ UFS_QUERY_FLAG_IDN_PWR_ON_WPE = 0x03,
880
+ UFS_QUERY_FLAG_IDN_BKOPS_EN = 0x04,
881
+ UFS_QUERY_FLAG_IDN_LIFE_SPAN_MODE_ENABLE = 0x05,
882
+ UFS_QUERY_FLAG_IDN_PURGE_ENABLE = 0x06,
883
+ UFS_QUERY_FLAG_IDN_REFRESH_ENABLE = 0x07,
884
+ UFS_QUERY_FLAG_IDN_FPHYRESOURCEREMOVAL = 0x08,
885
+ UFS_QUERY_FLAG_IDN_BUSY_RTC = 0x09,
886
+ UFS_QUERY_FLAG_IDN_RESERVED3 = 0x0A,
887
+ UFS_QUERY_FLAG_IDN_PERMANENTLY_DISABLE_FW_UPDATE = 0x0B,
888
+ UFS_QUERY_FLAG_IDN_WB_EN = 0x0E,
889
+ UFS_QUERY_FLAG_IDN_WB_BUFF_FLUSH_EN = 0x0F,
890
+ UFS_QUERY_FLAG_IDN_WB_BUFF_FLUSH_DURING_HIBERN8 = 0x10,
891
+ UFS_QUERY_FLAG_IDN_HPB_RESET = 0x11,
892
+ UFS_QUERY_FLAG_IDN_HPB_EN = 0x12,
893
+ UFS_QUERY_FLAG_IDN_COUNT,
894
+};
895
+
896
+/* Attribute idn for Query requests */
897
+enum attr_idn {
898
+ UFS_QUERY_ATTR_IDN_BOOT_LU_EN = 0x00,
899
+ UFS_QUERY_ATTR_IDN_MAX_HPB_SINGLE_CMD = 0x01,
900
+ UFS_QUERY_ATTR_IDN_POWER_MODE = 0x02,
901
+ UFS_QUERY_ATTR_IDN_ACTIVE_ICC_LVL = 0x03,
902
+ UFS_QUERY_ATTR_IDN_OOO_DATA_EN = 0x04,
903
+ UFS_QUERY_ATTR_IDN_BKOPS_STATUS = 0x05,
904
+ UFS_QUERY_ATTR_IDN_PURGE_STATUS = 0x06,
905
+ UFS_QUERY_ATTR_IDN_MAX_DATA_IN = 0x07,
906
+ UFS_QUERY_ATTR_IDN_MAX_DATA_OUT = 0x08,
907
+ UFS_QUERY_ATTR_IDN_DYN_CAP_NEEDED = 0x09,
908
+ UFS_QUERY_ATTR_IDN_REF_CLK_FREQ = 0x0A,
909
+ UFS_QUERY_ATTR_IDN_CONF_DESC_LOCK = 0x0B,
910
+ UFS_QUERY_ATTR_IDN_MAX_NUM_OF_RTT = 0x0C,
911
+ UFS_QUERY_ATTR_IDN_EE_CONTROL = 0x0D,
912
+ UFS_QUERY_ATTR_IDN_EE_STATUS = 0x0E,
913
+ UFS_QUERY_ATTR_IDN_SECONDS_PASSED = 0x0F,
914
+ UFS_QUERY_ATTR_IDN_CNTX_CONF = 0x10,
915
+ UFS_QUERY_ATTR_IDN_CORR_PRG_BLK_NUM = 0x11,
916
+ UFS_QUERY_ATTR_IDN_RESERVED2 = 0x12,
917
+ UFS_QUERY_ATTR_IDN_RESERVED3 = 0x13,
918
+ UFS_QUERY_ATTR_IDN_FFU_STATUS = 0x14,
919
+ UFS_QUERY_ATTR_IDN_PSA_STATE = 0x15,
920
+ UFS_QUERY_ATTR_IDN_PSA_DATA_SIZE = 0x16,
921
+ UFS_QUERY_ATTR_IDN_REF_CLK_GATING_WAIT_TIME = 0x17,
922
+ UFS_QUERY_ATTR_IDN_CASE_ROUGH_TEMP = 0x18,
923
+ UFS_QUERY_ATTR_IDN_HIGH_TEMP_BOUND = 0x19,
924
+ UFS_QUERY_ATTR_IDN_LOW_TEMP_BOUND = 0x1A,
925
+ UFS_QUERY_ATTR_IDN_THROTTLING_STATUS = 0x1B,
926
+ UFS_QUERY_ATTR_IDN_WB_FLUSH_STATUS = 0x1C,
927
+ UFS_QUERY_ATTR_IDN_AVAIL_WB_BUFF_SIZE = 0x1D,
928
+ UFS_QUERY_ATTR_IDN_WB_BUFF_LIFE_TIME_EST = 0x1E,
929
+ UFS_QUERY_ATTR_IDN_CURR_WB_BUFF_SIZE = 0x1F,
930
+ UFS_QUERY_ATTR_IDN_REFRESH_STATUS = 0x2C,
931
+ UFS_QUERY_ATTR_IDN_REFRESH_FREQ = 0x2D,
932
+ UFS_QUERY_ATTR_IDN_REFRESH_UNIT = 0x2E,
933
+ UFS_QUERY_ATTR_IDN_COUNT,
934
+};
935
+
936
+/* Descriptor idn for Query requests */
937
+enum desc_idn {
938
+ UFS_QUERY_DESC_IDN_DEVICE = 0x0,
939
+ UFS_QUERY_DESC_IDN_CONFIGURATION = 0x1,
940
+ UFS_QUERY_DESC_IDN_UNIT = 0x2,
941
+ UFS_QUERY_DESC_IDN_RFU_0 = 0x3,
942
+ UFS_QUERY_DESC_IDN_INTERCONNECT = 0x4,
943
+ UFS_QUERY_DESC_IDN_STRING = 0x5,
944
+ UFS_QUERY_DESC_IDN_RFU_1 = 0x6,
945
+ UFS_QUERY_DESC_IDN_GEOMETRY = 0x7,
946
+ UFS_QUERY_DESC_IDN_POWER = 0x8,
947
+ UFS_QUERY_DESC_IDN_HEALTH = 0x9,
948
+ UFS_QUERY_DESC_IDN_MAX,
949
+};
950
+
951
+enum desc_header_offset {
952
+ UFS_QUERY_DESC_LENGTH_OFFSET = 0x00,
953
+ UFS_QUERY_DESC_DESC_TYPE_OFFSET = 0x01,
954
+};
955
+
956
+/* Unit descriptor parameters offsets in bytes*/
957
+enum unit_desc_param {
958
+ UFS_UNIT_DESC_PARAM_LEN = 0x0,
959
+ UFS_UNIT_DESC_PARAM_TYPE = 0x1,
960
+ UFS_UNIT_DESC_PARAM_UNIT_INDEX = 0x2,
961
+ UFS_UNIT_DESC_PARAM_LU_ENABLE = 0x3,
962
+ UFS_UNIT_DESC_PARAM_BOOT_LUN_ID = 0x4,
963
+ UFS_UNIT_DESC_PARAM_LU_WR_PROTECT = 0x5,
964
+ UFS_UNIT_DESC_PARAM_LU_Q_DEPTH = 0x6,
965
+ UFS_UNIT_DESC_PARAM_PSA_SENSITIVE = 0x7,
966
+ UFS_UNIT_DESC_PARAM_MEM_TYPE = 0x8,
967
+ UFS_UNIT_DESC_PARAM_DATA_RELIABILITY = 0x9,
968
+ UFS_UNIT_DESC_PARAM_LOGICAL_BLK_SIZE = 0xA,
969
+ UFS_UNIT_DESC_PARAM_LOGICAL_BLK_COUNT = 0xB,
970
+ UFS_UNIT_DESC_PARAM_ERASE_BLK_SIZE = 0x13,
971
+ UFS_UNIT_DESC_PARAM_PROVISIONING_TYPE = 0x17,
972
+ UFS_UNIT_DESC_PARAM_PHY_MEM_RSRC_CNT = 0x18,
973
+ UFS_UNIT_DESC_PARAM_CTX_CAPABILITIES = 0x20,
974
+ UFS_UNIT_DESC_PARAM_LARGE_UNIT_SIZE_M1 = 0x22,
975
+ UFS_UNIT_DESC_PARAM_HPB_LU_MAX_ACTIVE_RGNS = 0x23,
976
+ UFS_UNIT_DESC_PARAM_HPB_PIN_RGN_START_OFF = 0x25,
977
+ UFS_UNIT_DESC_PARAM_HPB_NUM_PIN_RGNS = 0x27,
978
+ UFS_UNIT_DESC_PARAM_WB_BUF_ALLOC_UNITS = 0x29,
979
+};
980
+
981
+/* RPMB Unit descriptor parameters offsets in bytes*/
982
+enum rpmb_unit_desc_param {
983
+ UFS_RPMB_UNIT_DESC_PARAM_LEN = 0x0,
984
+ UFS_RPMB_UNIT_DESC_PARAM_TYPE = 0x1,
985
+ UFS_RPMB_UNIT_DESC_PARAM_UNIT_INDEX = 0x2,
986
+ UFS_RPMB_UNIT_DESC_PARAM_LU_ENABLE = 0x3,
987
+ UFS_RPMB_UNIT_DESC_PARAM_BOOT_LUN_ID = 0x4,
988
+ UFS_RPMB_UNIT_DESC_PARAM_LU_WR_PROTECT = 0x5,
989
+ UFS_RPMB_UNIT_DESC_PARAM_LU_Q_DEPTH = 0x6,
990
+ UFS_RPMB_UNIT_DESC_PARAM_PSA_SENSITIVE = 0x7,
991
+ UFS_RPMB_UNIT_DESC_PARAM_MEM_TYPE = 0x8,
992
+ UFS_RPMB_UNIT_DESC_PARAM_REGION_EN = 0x9,
993
+ UFS_RPMB_UNIT_DESC_PARAM_LOGICAL_BLK_SIZE = 0xA,
994
+ UFS_RPMB_UNIT_DESC_PARAM_LOGICAL_BLK_COUNT = 0xB,
995
+ UFS_RPMB_UNIT_DESC_PARAM_REGION0_SIZE = 0x13,
996
+ UFS_RPMB_UNIT_DESC_PARAM_REGION1_SIZE = 0x14,
997
+ UFS_RPMB_UNIT_DESC_PARAM_REGION2_SIZE = 0x15,
998
+ UFS_RPMB_UNIT_DESC_PARAM_REGION3_SIZE = 0x16,
999
+ UFS_RPMB_UNIT_DESC_PARAM_PROVISIONING_TYPE = 0x17,
1000
+ UFS_RPMB_UNIT_DESC_PARAM_PHY_MEM_RSRC_CNT = 0x18,
1001
+};
1002
+
1003
+/* Device descriptor parameters offsets in bytes*/
1004
+enum device_desc_param {
1005
+ UFS_DEVICE_DESC_PARAM_LEN = 0x0,
1006
+ UFS_DEVICE_DESC_PARAM_TYPE = 0x1,
1007
+ UFS_DEVICE_DESC_PARAM_DEVICE_TYPE = 0x2,
1008
+ UFS_DEVICE_DESC_PARAM_DEVICE_CLASS = 0x3,
1009
+ UFS_DEVICE_DESC_PARAM_DEVICE_SUB_CLASS = 0x4,
1010
+ UFS_DEVICE_DESC_PARAM_PRTCL = 0x5,
1011
+ UFS_DEVICE_DESC_PARAM_NUM_LU = 0x6,
1012
+ UFS_DEVICE_DESC_PARAM_NUM_WLU = 0x7,
1013
+ UFS_DEVICE_DESC_PARAM_BOOT_ENBL = 0x8,
1014
+ UFS_DEVICE_DESC_PARAM_DESC_ACCSS_ENBL = 0x9,
1015
+ UFS_DEVICE_DESC_PARAM_INIT_PWR_MODE = 0xA,
1016
+ UFS_DEVICE_DESC_PARAM_HIGH_PR_LUN = 0xB,
1017
+ UFS_DEVICE_DESC_PARAM_SEC_RMV_TYPE = 0xC,
1018
+ UFS_DEVICE_DESC_PARAM_SEC_LU = 0xD,
1019
+ UFS_DEVICE_DESC_PARAM_BKOP_TERM_LT = 0xE,
1020
+ UFS_DEVICE_DESC_PARAM_ACTVE_ICC_LVL = 0xF,
1021
+ UFS_DEVICE_DESC_PARAM_SPEC_VER = 0x10,
1022
+ UFS_DEVICE_DESC_PARAM_MANF_DATE = 0x12,
1023
+ UFS_DEVICE_DESC_PARAM_MANF_NAME = 0x14,
1024
+ UFS_DEVICE_DESC_PARAM_PRDCT_NAME = 0x15,
1025
+ UFS_DEVICE_DESC_PARAM_SN = 0x16,
1026
+ UFS_DEVICE_DESC_PARAM_OEM_ID = 0x17,
1027
+ UFS_DEVICE_DESC_PARAM_MANF_ID = 0x18,
1028
+ UFS_DEVICE_DESC_PARAM_UD_OFFSET = 0x1A,
1029
+ UFS_DEVICE_DESC_PARAM_UD_LEN = 0x1B,
1030
+ UFS_DEVICE_DESC_PARAM_RTT_CAP = 0x1C,
1031
+ UFS_DEVICE_DESC_PARAM_FRQ_RTC = 0x1D,
1032
+ UFS_DEVICE_DESC_PARAM_UFS_FEAT = 0x1F,
1033
+ UFS_DEVICE_DESC_PARAM_FFU_TMT = 0x20,
1034
+ UFS_DEVICE_DESC_PARAM_Q_DPTH = 0x21,
1035
+ UFS_DEVICE_DESC_PARAM_DEV_VER = 0x22,
1036
+ UFS_DEVICE_DESC_PARAM_NUM_SEC_WPA = 0x24,
1037
+ UFS_DEVICE_DESC_PARAM_PSA_MAX_DATA = 0x25,
1038
+ UFS_DEVICE_DESC_PARAM_PSA_TMT = 0x29,
1039
+ UFS_DEVICE_DESC_PARAM_PRDCT_REV = 0x2A,
1040
+ UFS_DEVICE_DESC_PARAM_HPB_VER = 0x40,
1041
+ UFS_DEVICE_DESC_PARAM_HPB_CONTROL = 0x42,
1042
+ UFS_DEVICE_DESC_PARAM_EXT_UFS_FEATURE_SUP = 0x4F,
1043
+ UFS_DEVICE_DESC_PARAM_WB_PRESRV_USRSPC_EN = 0x53,
1044
+ UFS_DEVICE_DESC_PARAM_WB_TYPE = 0x54,
1045
+ UFS_DEVICE_DESC_PARAM_WB_SHARED_ALLOC_UNITS = 0x55,
1046
+};
1047
+
1048
+/* Interconnect descriptor parameters offsets in bytes*/
1049
+enum interconnect_desc_param {
1050
+ UFS_INTERCONNECT_DESC_PARAM_LEN = 0x0,
1051
+ UFS_INTERCONNECT_DESC_PARAM_TYPE = 0x1,
1052
+ UFS_INTERCONNECT_DESC_PARAM_UNIPRO_VER = 0x2,
1053
+ UFS_INTERCONNECT_DESC_PARAM_MPHY_VER = 0x4,
1054
+};
1055
+
1056
+/* Geometry descriptor parameters offsets in bytes*/
1057
+enum geometry_desc_param {
1058
+ UFS_GEOMETRY_DESC_PARAM_LEN = 0x0,
1059
+ UFS_GEOMETRY_DESC_PARAM_TYPE = 0x1,
1060
+ UFS_GEOMETRY_DESC_PARAM_DEV_CAP = 0x4,
1061
+ UFS_GEOMETRY_DESC_PARAM_MAX_NUM_LUN = 0xC,
1062
+ UFS_GEOMETRY_DESC_PARAM_SEG_SIZE = 0xD,
1063
+ UFS_GEOMETRY_DESC_PARAM_ALLOC_UNIT_SIZE = 0x11,
1064
+ UFS_GEOMETRY_DESC_PARAM_MIN_BLK_SIZE = 0x12,
1065
+ UFS_GEOMETRY_DESC_PARAM_OPT_RD_BLK_SIZE = 0x13,
1066
+ UFS_GEOMETRY_DESC_PARAM_OPT_WR_BLK_SIZE = 0x14,
1067
+ UFS_GEOMETRY_DESC_PARAM_MAX_IN_BUF_SIZE = 0x15,
1068
+ UFS_GEOMETRY_DESC_PARAM_MAX_OUT_BUF_SIZE = 0x16,
1069
+ UFS_GEOMETRY_DESC_PARAM_RPMB_RW_SIZE = 0x17,
1070
+ UFS_GEOMETRY_DESC_PARAM_DYN_CAP_RSRC_PLC = 0x18,
1071
+ UFS_GEOMETRY_DESC_PARAM_DATA_ORDER = 0x19,
1072
+ UFS_GEOMETRY_DESC_PARAM_MAX_NUM_CTX = 0x1A,
1073
+ UFS_GEOMETRY_DESC_PARAM_TAG_UNIT_SIZE = 0x1B,
1074
+ UFS_GEOMETRY_DESC_PARAM_TAG_RSRC_SIZE = 0x1C,
1075
+ UFS_GEOMETRY_DESC_PARAM_SEC_RM_TYPES = 0x1D,
1076
+ UFS_GEOMETRY_DESC_PARAM_MEM_TYPES = 0x1E,
1077
+ UFS_GEOMETRY_DESC_PARAM_SCM_MAX_NUM_UNITS = 0x20,
1078
+ UFS_GEOMETRY_DESC_PARAM_SCM_CAP_ADJ_FCTR = 0x24,
1079
+ UFS_GEOMETRY_DESC_PARAM_NPM_MAX_NUM_UNITS = 0x26,
1080
+ UFS_GEOMETRY_DESC_PARAM_NPM_CAP_ADJ_FCTR = 0x2A,
1081
+ UFS_GEOMETRY_DESC_PARAM_ENM1_MAX_NUM_UNITS = 0x2C,
1082
+ UFS_GEOMETRY_DESC_PARAM_ENM1_CAP_ADJ_FCTR = 0x30,
1083
+ UFS_GEOMETRY_DESC_PARAM_ENM2_MAX_NUM_UNITS = 0x32,
1084
+ UFS_GEOMETRY_DESC_PARAM_ENM2_CAP_ADJ_FCTR = 0x36,
1085
+ UFS_GEOMETRY_DESC_PARAM_ENM3_MAX_NUM_UNITS = 0x38,
1086
+ UFS_GEOMETRY_DESC_PARAM_ENM3_CAP_ADJ_FCTR = 0x3C,
1087
+ UFS_GEOMETRY_DESC_PARAM_ENM4_MAX_NUM_UNITS = 0x3E,
1088
+ UFS_GEOMETRY_DESC_PARAM_ENM4_CAP_ADJ_FCTR = 0x42,
1089
+ UFS_GEOMETRY_DESC_PARAM_OPT_LOG_BLK_SIZE = 0x44,
1090
+ UFS_GEOMETRY_DESC_PARAM_HPB_REGION_SIZE = 0x48,
1091
+ UFS_GEOMETRY_DESC_PARAM_HPB_NUMBER_LU = 0x49,
1092
+ UFS_GEOMETRY_DESC_PARAM_HPB_SUBREGION_SIZE = 0x4A,
1093
+ UFS_GEOMETRY_DESC_PARAM_HPB_MAX_ACTIVE_REGS = 0x4B,
1094
+ UFS_GEOMETRY_DESC_PARAM_WB_MAX_ALLOC_UNITS = 0x4F,
1095
+ UFS_GEOMETRY_DESC_PARAM_WB_MAX_WB_LUNS = 0x53,
1096
+ UFS_GEOMETRY_DESC_PARAM_WB_BUFF_CAP_ADJ = 0x54,
1097
+ UFS_GEOMETRY_DESC_PARAM_WB_SUP_RED_TYPE = 0x55,
1098
+ UFS_GEOMETRY_DESC_PARAM_WB_SUP_WB_TYPE = 0x56,
1099
+};
1100
+
1101
+/* Health descriptor parameters offsets in bytes*/
1102
+enum health_desc_param {
1103
+ UFS_HEALTH_DESC_PARAM_LEN = 0x0,
1104
+ UFS_HEALTH_DESC_PARAM_TYPE = 0x1,
1105
+ UFS_HEALTH_DESC_PARAM_EOL_INFO = 0x2,
1106
+ UFS_HEALTH_DESC_PARAM_LIFE_TIME_EST_A = 0x3,
1107
+ UFS_HEALTH_DESC_PARAM_LIFE_TIME_EST_B = 0x4,
1108
+};
1109
+
1110
+/* WriteBooster buffer mode */
1111
+enum {
1112
+ UFS_WB_BUF_MODE_LU_DEDICATED = 0x0,
1113
+ UFS_WB_BUF_MODE_SHARED = 0x1,
1114
+};
1115
+
1116
+/*
1117
+ * Logical Unit Write Protect
1118
+ * 00h: LU not write protected
1119
+ * 01h: LU write protected when fPowerOnWPEn =1
1120
+ * 02h: LU permanently write protected when fPermanentWPEn =1
1121
+ */
1122
+enum ufs_lu_wp_type {
1123
+ UFS_LU_NO_WP = 0x00,
1124
+ UFS_LU_POWER_ON_WP = 0x01,
1125
+ UFS_LU_PERM_WP = 0x02,
1126
+};
1127
+
1128
+/* UTP QUERY Transaction Specific Fields OpCode */
1129
+enum query_opcode {
1130
+ UFS_UPIU_QUERY_OPCODE_NOP = 0x0,
1131
+ UFS_UPIU_QUERY_OPCODE_READ_DESC = 0x1,
1132
+ UFS_UPIU_QUERY_OPCODE_WRITE_DESC = 0x2,
1133
+ UFS_UPIU_QUERY_OPCODE_READ_ATTR = 0x3,
1134
+ UFS_UPIU_QUERY_OPCODE_WRITE_ATTR = 0x4,
1135
+ UFS_UPIU_QUERY_OPCODE_READ_FLAG = 0x5,
1136
+ UFS_UPIU_QUERY_OPCODE_SET_FLAG = 0x6,
1137
+ UFS_UPIU_QUERY_OPCODE_CLEAR_FLAG = 0x7,
1138
+ UFS_UPIU_QUERY_OPCODE_TOGGLE_FLAG = 0x8,
1139
+};
1140
+
1141
+/* Query response result code */
1142
+typedef enum QueryRespCode {
1143
+ UFS_QUERY_RESULT_SUCCESS = 0x00,
1144
+ UFS_QUERY_RESULT_NOT_READABLE = 0xF6,
1145
+ UFS_QUERY_RESULT_NOT_WRITEABLE = 0xF7,
1146
+ UFS_QUERY_RESULT_ALREADY_WRITTEN = 0xF8,
1147
+ UFS_QUERY_RESULT_INVALID_LENGTH = 0xF9,
1148
+ UFS_QUERY_RESULT_INVALID_VALUE = 0xFA,
1149
+ UFS_QUERY_RESULT_INVALID_SELECTOR = 0xFB,
1150
+ UFS_QUERY_RESULT_INVALID_INDEX = 0xFC,
1151
+ UFS_QUERY_RESULT_INVALID_IDN = 0xFD,
1152
+ UFS_QUERY_RESULT_INVALID_OPCODE = 0xFE,
1153
+ UFS_QUERY_RESULT_GENERAL_FAILURE = 0xFF,
1154
+} QueryRespCode;
1155
+
1156
+/* UTP Transfer Request Command Type (CT) */
1157
+enum {
1158
+ UFS_UPIU_COMMAND_SET_TYPE_SCSI = 0x0,
1159
+ UFS_UPIU_COMMAND_SET_TYPE_UFS = 0x1,
1160
+ UFS_UPIU_COMMAND_SET_TYPE_QUERY = 0x2,
1161
+};
1162
+
1163
+/* Task management service response */
1164
+enum {
1165
+ UFS_UPIU_TASK_MANAGEMENT_FUNC_COMPL = 0x00,
1166
+ UFS_UPIU_TASK_MANAGEMENT_FUNC_NOT_SUPPORTED = 0x04,
1167
+ UFS_UPIU_TASK_MANAGEMENT_FUNC_SUCCEEDED = 0x08,
1168
+ UFS_UPIU_TASK_MANAGEMENT_FUNC_FAILED = 0x05,
1169
+ UFS_UPIU_INCORRECT_LOGICAL_UNIT_NO = 0x09,
1170
+};
1171
+
1172
+/* UFS device power modes */
1173
+enum ufs_dev_pwr_mode {
1174
+ UFS_ACTIVE_PWR_MODE = 1,
1175
+ UFS_SLEEP_PWR_MODE = 2,
1176
+ UFS_POWERDOWN_PWR_MODE = 3,
1177
+ UFS_DEEPSLEEP_PWR_MODE = 4,
1178
+};
1179
+
1180
+/*
1181
+ * struct UtpCmdRsp - Response UPIU structure
1182
+ * @residual_transfer_count: Residual transfer count DW-3
1183
+ * @reserved: Reserved double words DW-4 to DW-7
1184
+ * @sense_data_len: Sense data length DW-8 U16
1185
+ * @sense_data: Sense data field DW-8 to DW-12
1186
+ */
1187
+typedef struct QEMU_PACKED UtpCmdRsp {
1188
+ uint32_t residual_transfer_count;
1189
+ uint32_t reserved[4];
1190
+ uint16_t sense_data_len;
1191
+ uint8_t sense_data[UFS_SENSE_SIZE];
1192
+} UtpCmdRsp;
1193
+
1194
+/*
1195
+ * struct UtpUpiuRsp - general upiu response structure
1196
+ * @header: UPIU header structure DW-0 to DW-2
1197
+ * @sr: fields structure for scsi command DW-3 to DW-12
1198
+ * @qr: fields structure for query request DW-3 to DW-7
1199
+ */
1200
+typedef struct QEMU_PACKED UtpUpiuRsp {
1201
+ UtpUpiuHeader header;
1202
+ union {
1203
+ UtpCmdRsp sr;
1204
+ UtpUpiuQuery qr;
1205
+ };
1206
+} UtpUpiuRsp;
1207
+
1208
+static inline void _ufs_check_size(void)
1209
+{
1210
+ QEMU_BUILD_BUG_ON(sizeof(UfsReg) != 0x104);
1211
+ QEMU_BUILD_BUG_ON(sizeof(DeviceDescriptor) != 89);
1212
+ QEMU_BUILD_BUG_ON(sizeof(GeometryDescriptor) != 87);
1213
+ QEMU_BUILD_BUG_ON(sizeof(UnitDescriptor) != 45);
1214
+ QEMU_BUILD_BUG_ON(sizeof(RpmbUnitDescriptor) != 35);
1215
+ QEMU_BUILD_BUG_ON(sizeof(PowerParametersDescriptor) != 98);
1216
+ QEMU_BUILD_BUG_ON(sizeof(InterconnectDescriptor) != 6);
1217
+ QEMU_BUILD_BUG_ON(sizeof(StringDescriptor) != 254);
1218
+ QEMU_BUILD_BUG_ON(sizeof(DeviceHealthDescriptor) != 45);
1219
+ QEMU_BUILD_BUG_ON(sizeof(Flags) != 0x13);
1220
+ QEMU_BUILD_BUG_ON(sizeof(UtpUpiuHeader) != 12);
1221
+ QEMU_BUILD_BUG_ON(sizeof(UtpUpiuQuery) != 276);
1222
+ QEMU_BUILD_BUG_ON(sizeof(UtpUpiuCmd) != 20);
1223
+ QEMU_BUILD_BUG_ON(sizeof(UtpUpiuReq) != 288);
1224
+ QEMU_BUILD_BUG_ON(sizeof(UfshcdSgEntry) != 16);
1225
+ QEMU_BUILD_BUG_ON(sizeof(RequestDescHeader) != 16);
1226
+ QEMU_BUILD_BUG_ON(sizeof(UtpTransferReqDesc) != 32);
1227
+ QEMU_BUILD_BUG_ON(sizeof(UtpTaskReqDesc) != 80);
1228
+ QEMU_BUILD_BUG_ON(sizeof(UtpCmdRsp) != 40);
1229
+ QEMU_BUILD_BUG_ON(sizeof(UtpUpiuRsp) != 288);
1230
+}
1231
+#endif
1232
diff --git a/include/hw/pci/pci.h b/include/hw/pci/pci.h
1233
index XXXXXXX..XXXXXXX 100644
1234
--- a/include/hw/pci/pci.h
1235
+++ b/include/hw/pci/pci.h
1236
@@ -XXX,XX +XXX,XX @@ extern bool pci_available;
1237
#define PCI_DEVICE_ID_REDHAT_NVME 0x0010
1238
#define PCI_DEVICE_ID_REDHAT_PVPANIC 0x0011
1239
#define PCI_DEVICE_ID_REDHAT_ACPI_ERST 0x0012
1240
+#define PCI_DEVICE_ID_REDHAT_UFS 0x0013
1241
#define PCI_DEVICE_ID_REDHAT_QXL 0x0100
1242
1243
#define FMT_PCIBUS PRIx64
1244
diff --git a/include/hw/pci/pci_ids.h b/include/hw/pci/pci_ids.h
1245
index XXXXXXX..XXXXXXX 100644
1246
--- a/include/hw/pci/pci_ids.h
1247
+++ b/include/hw/pci/pci_ids.h
1248
@@ -XXX,XX +XXX,XX @@
1249
#define PCI_CLASS_STORAGE_SATA 0x0106
1250
#define PCI_CLASS_STORAGE_SAS 0x0107
1251
#define PCI_CLASS_STORAGE_EXPRESS 0x0108
1252
+#define PCI_CLASS_STORAGE_UFS 0x0109
1253
#define PCI_CLASS_STORAGE_OTHER 0x0180
1254
1255
#define PCI_BASE_CLASS_NETWORK 0x02
1256
diff --git a/hw/ufs/ufs.c b/hw/ufs/ufs.c
1257
new file mode 100644
1258
index XXXXXXX..XXXXXXX
1259
--- /dev/null
1260
+++ b/hw/ufs/ufs.c
1261
@@ -XXX,XX +XXX,XX @@
294
@@ -XXX,XX +XXX,XX @@
1262
+/*
295
+/*
1263
+ * QEMU Universal Flash Storage (UFS) Controller
296
+ * AioContext multithreading tests
1264
+ *
297
+ *
1265
+ * Copyright (c) 2023 Samsung Electronics Co., Ltd. All rights reserved.
298
+ * Copyright Red Hat, Inc. 2016
1266
+ *
299
+ *
1267
+ * Written by Jeuk Kim <jeuk20.kim@samsung.com>
300
+ * Authors:
1268
+ *
301
+ * Paolo Bonzini <pbonzini@redhat.com>
1269
+ * SPDX-License-Identifier: GPL-2.0-or-later
302
+ *
303
+ * This work is licensed under the terms of the GNU LGPL, version 2 or later.
304
+ * See the COPYING.LIB file in the top-level directory.
1270
+ */
305
+ */
1271
+
306
+
1272
+#include "qemu/osdep.h"
307
+#include "qemu/osdep.h"
308
+#include <glib.h>
309
+#include "block/aio.h"
1273
+#include "qapi/error.h"
310
+#include "qapi/error.h"
1274
+#include "migration/vmstate.h"
311
+#include "qemu/coroutine.h"
312
+#include "qemu/thread.h"
313
+#include "qemu/error-report.h"
314
+#include "iothread.h"
315
+
316
+/* AioContext management */
317
+
318
+#define NUM_CONTEXTS 5
319
+
320
+static IOThread *threads[NUM_CONTEXTS];
321
+static AioContext *ctx[NUM_CONTEXTS];
322
+static __thread int id = -1;
323
+
324
+static QemuEvent done_event;
325
+
326
+/* Run a function synchronously on a remote iothread. */
327
+
328
+typedef struct CtxRunData {
329
+ QEMUBHFunc *cb;
330
+ void *arg;
331
+} CtxRunData;
332
+
333
+static void ctx_run_bh_cb(void *opaque)
334
+{
335
+ CtxRunData *data = opaque;
336
+
337
+ data->cb(data->arg);
338
+ qemu_event_set(&done_event);
339
+}
340
+
341
+static void ctx_run(int i, QEMUBHFunc *cb, void *opaque)
342
+{
343
+ CtxRunData data = {
344
+ .cb = cb,
345
+ .arg = opaque
346
+ };
347
+
348
+ qemu_event_reset(&done_event);
349
+ aio_bh_schedule_oneshot(ctx[i], ctx_run_bh_cb, &data);
350
+ qemu_event_wait(&done_event);
351
+}
352
+
353
+/* Starting the iothreads. */
354
+
355
+static void set_id_cb(void *opaque)
356
+{
357
+ int *i = opaque;
358
+
359
+ id = *i;
360
+}
361
+
362
+static void create_aio_contexts(void)
363
+{
364
+ int i;
365
+
366
+ for (i = 0; i < NUM_CONTEXTS; i++) {
367
+ threads[i] = iothread_new();
368
+ ctx[i] = iothread_get_aio_context(threads[i]);
369
+ }
370
+
371
+ qemu_event_init(&done_event, false);
372
+ for (i = 0; i < NUM_CONTEXTS; i++) {
373
+ ctx_run(i, set_id_cb, &i);
374
+ }
375
+}
376
+
377
+/* Stopping the iothreads. */
378
+
379
+static void join_aio_contexts(void)
380
+{
381
+ int i;
382
+
383
+ for (i = 0; i < NUM_CONTEXTS; i++) {
384
+ aio_context_ref(ctx[i]);
385
+ }
386
+ for (i = 0; i < NUM_CONTEXTS; i++) {
387
+ iothread_join(threads[i]);
388
+ }
389
+ for (i = 0; i < NUM_CONTEXTS; i++) {
390
+ aio_context_unref(ctx[i]);
391
+ }
392
+ qemu_event_destroy(&done_event);
393
+}
394
+
395
+/* Basic test for the stuff above. */
396
+
397
+static void test_lifecycle(void)
398
+{
399
+ create_aio_contexts();
400
+ join_aio_contexts();
401
+}
402
+
403
+/* aio_co_schedule test. */
404
+
405
+static Coroutine *to_schedule[NUM_CONTEXTS];
406
+
407
+static bool now_stopping;
408
+
409
+static int count_retry;
410
+static int count_here;
411
+static int count_other;
412
+
413
+static bool schedule_next(int n)
414
+{
415
+ Coroutine *co;
416
+
417
+ co = atomic_xchg(&to_schedule[n], NULL);
418
+ if (!co) {
419
+ atomic_inc(&count_retry);
420
+ return false;
421
+ }
422
+
423
+ if (n == id) {
424
+ atomic_inc(&count_here);
425
+ } else {
426
+ atomic_inc(&count_other);
427
+ }
428
+
429
+ aio_co_schedule(ctx[n], co);
430
+ return true;
431
+}
432
+
433
+static void finish_cb(void *opaque)
434
+{
435
+ schedule_next(id);
436
+}
437
+
438
+static coroutine_fn void test_multi_co_schedule_entry(void *opaque)
439
+{
440
+ g_assert(to_schedule[id] == NULL);
441
+ atomic_mb_set(&to_schedule[id], qemu_coroutine_self());
442
+
443
+ while (!atomic_mb_read(&now_stopping)) {
444
+ int n;
445
+
446
+ n = g_test_rand_int_range(0, NUM_CONTEXTS);
447
+ schedule_next(n);
448
+ qemu_coroutine_yield();
449
+
450
+ g_assert(to_schedule[id] == NULL);
451
+ atomic_mb_set(&to_schedule[id], qemu_coroutine_self());
452
+ }
453
+}
454
+
455
+
456
+static void test_multi_co_schedule(int seconds)
457
+{
458
+ int i;
459
+
460
+ count_here = count_other = count_retry = 0;
461
+ now_stopping = false;
462
+
463
+ create_aio_contexts();
464
+ for (i = 0; i < NUM_CONTEXTS; i++) {
465
+ Coroutine *co1 = qemu_coroutine_create(test_multi_co_schedule_entry, NULL);
466
+ aio_co_schedule(ctx[i], co1);
467
+ }
468
+
469
+ g_usleep(seconds * 1000000);
470
+
471
+ atomic_mb_set(&now_stopping, true);
472
+ for (i = 0; i < NUM_CONTEXTS; i++) {
473
+ ctx_run(i, finish_cb, NULL);
474
+ to_schedule[i] = NULL;
475
+ }
476
+
477
+ join_aio_contexts();
478
+ g_test_message("scheduled %d, queued %d, retry %d, total %d\n",
479
+ count_other, count_here, count_retry,
480
+ count_here + count_other + count_retry);
481
+}
482
+
483
+static void test_multi_co_schedule_1(void)
484
+{
485
+ test_multi_co_schedule(1);
486
+}
487
+
488
+static void test_multi_co_schedule_10(void)
489
+{
490
+ test_multi_co_schedule(10);
491
+}
492
+
493
+/* End of tests. */
494
+
495
+int main(int argc, char **argv)
496
+{
497
+ init_clocks();
498
+
499
+ g_test_init(&argc, &argv, NULL);
500
+ g_test_add_func("/aio/multi/lifecycle", test_lifecycle);
501
+ if (g_test_quick()) {
502
+ g_test_add_func("/aio/multi/schedule", test_multi_co_schedule_1);
503
+ } else {
504
+ g_test_add_func("/aio/multi/schedule", test_multi_co_schedule_10);
505
+ }
506
+ return g_test_run();
507
+}
508
diff --git a/util/async.c b/util/async.c
509
index XXXXXXX..XXXXXXX 100644
510
--- a/util/async.c
511
+++ b/util/async.c
512
@@ -XXX,XX +XXX,XX @@
513
#include "qemu/main-loop.h"
514
#include "qemu/atomic.h"
515
#include "block/raw-aio.h"
516
+#include "qemu/coroutine_int.h"
1275
+#include "trace.h"
517
+#include "trace.h"
1276
+#include "ufs.h"
518
1277
+
519
/***********************************************************/
1278
+/* The QEMU-UFS device follows spec version 3.1 */
520
/* bottom halves (can be seen as timers which expire ASAP) */
1279
+#define UFS_SPEC_VER 0x00000310
521
@@ -XXX,XX +XXX,XX @@ aio_ctx_finalize(GSource *source)
1280
+#define UFS_MAX_NUTRS 32
522
}
1281
+#define UFS_MAX_NUTMRS 8
523
#endif
1282
+
524
1283
+static void ufs_irq_check(UfsHc *u)
525
+ assert(QSLIST_EMPTY(&ctx->scheduled_coroutines));
1284
+{
526
+ qemu_bh_delete(ctx->co_schedule_bh);
1285
+ PCIDevice *pci = PCI_DEVICE(u);
527
+
1286
+
528
qemu_lockcnt_lock(&ctx->list_lock);
1287
+ if ((u->reg.is & UFS_INTR_MASK) & u->reg.ie) {
529
assert(!qemu_lockcnt_count(&ctx->list_lock));
1288
+ trace_ufs_irq_raise();
530
while (ctx->first_bh) {
1289
+ pci_irq_assert(pci);
531
@@ -XXX,XX +XXX,XX @@ static bool event_notifier_poll(void *opaque)
532
return atomic_read(&ctx->notified);
533
}
534
535
+static void co_schedule_bh_cb(void *opaque)
536
+{
537
+ AioContext *ctx = opaque;
538
+ QSLIST_HEAD(, Coroutine) straight, reversed;
539
+
540
+ QSLIST_MOVE_ATOMIC(&reversed, &ctx->scheduled_coroutines);
541
+ QSLIST_INIT(&straight);
542
+
543
+ while (!QSLIST_EMPTY(&reversed)) {
544
+ Coroutine *co = QSLIST_FIRST(&reversed);
545
+ QSLIST_REMOVE_HEAD(&reversed, co_scheduled_next);
546
+ QSLIST_INSERT_HEAD(&straight, co, co_scheduled_next);
547
+ }
548
+
549
+ while (!QSLIST_EMPTY(&straight)) {
550
+ Coroutine *co = QSLIST_FIRST(&straight);
551
+ QSLIST_REMOVE_HEAD(&straight, co_scheduled_next);
552
+ trace_aio_co_schedule_bh_cb(ctx, co);
553
+ qemu_coroutine_enter(co);
554
+ }
555
+}
556
+
557
AioContext *aio_context_new(Error **errp)
558
{
559
int ret;
560
@@ -XXX,XX +XXX,XX @@ AioContext *aio_context_new(Error **errp)
561
}
562
g_source_set_can_recurse(&ctx->source, true);
563
qemu_lockcnt_init(&ctx->list_lock);
564
+
565
+ ctx->co_schedule_bh = aio_bh_new(ctx, co_schedule_bh_cb, ctx);
566
+ QSLIST_INIT(&ctx->scheduled_coroutines);
567
+
568
aio_set_event_notifier(ctx, &ctx->notifier,
569
false,
570
(EventNotifierHandler *)
571
@@ -XXX,XX +XXX,XX @@ fail:
572
return NULL;
573
}
574
575
+void aio_co_schedule(AioContext *ctx, Coroutine *co)
576
+{
577
+ trace_aio_co_schedule(ctx, co);
578
+ QSLIST_INSERT_HEAD_ATOMIC(&ctx->scheduled_coroutines,
579
+ co, co_scheduled_next);
580
+ qemu_bh_schedule(ctx->co_schedule_bh);
581
+}
582
+
583
+void aio_co_wake(struct Coroutine *co)
584
+{
585
+ AioContext *ctx;
586
+
587
+ /* Read coroutine before co->ctx. Matches smp_wmb in
588
+ * qemu_coroutine_enter.
589
+ */
590
+ smp_read_barrier_depends();
591
+ ctx = atomic_read(&co->ctx);
592
+
593
+ if (ctx != qemu_get_current_aio_context()) {
594
+ aio_co_schedule(ctx, co);
595
+ return;
596
+ }
597
+
598
+ if (qemu_in_coroutine()) {
599
+ Coroutine *self = qemu_coroutine_self();
600
+ assert(self != co);
601
+ QSIMPLEQ_INSERT_TAIL(&self->co_queue_wakeup, co, co_queue_next);
1290
+ } else {
602
+ } else {
1291
+ trace_ufs_irq_lower();
603
+ aio_context_acquire(ctx);
1292
+ pci_irq_deassert(pci);
604
+ qemu_coroutine_enter(co);
1293
+ }
605
+ aio_context_release(ctx);
1294
+}
606
+ }
1295
+
607
+}
1296
+static void ufs_process_uiccmd(UfsHc *u, uint32_t val)
608
+
1297
+{
609
void aio_context_ref(AioContext *ctx)
1298
+ trace_ufs_process_uiccmd(val, u->reg.ucmdarg1, u->reg.ucmdarg2,
610
{
1299
+ u->reg.ucmdarg3);
611
g_source_ref(&ctx->source);
1300
+ /*
612
diff --git a/util/qemu-coroutine.c b/util/qemu-coroutine.c
1301
+ * Only the essential uic commands for running drivers on Linux and Windows
613
index XXXXXXX..XXXXXXX 100644
1302
+ * are implemented.
614
--- a/util/qemu-coroutine.c
615
+++ b/util/qemu-coroutine.c
616
@@ -XXX,XX +XXX,XX @@
617
#include "qemu/atomic.h"
618
#include "qemu/coroutine.h"
619
#include "qemu/coroutine_int.h"
620
+#include "block/aio.h"
621
622
enum {
623
POOL_BATCH_SIZE = 64,
624
@@ -XXX,XX +XXX,XX @@ void qemu_coroutine_enter(Coroutine *co)
625
}
626
627
co->caller = self;
628
+ co->ctx = qemu_get_current_aio_context();
629
+
630
+ /* Store co->ctx before anything that stores co. Matches
631
+ * barrier in aio_co_wake.
1303
+ */
632
+ */
1304
+ switch (val) {
633
+ smp_wmb();
1305
+ case UFS_UIC_CMD_DME_LINK_STARTUP:
634
+
1306
+ u->reg.hcs = FIELD_DP32(u->reg.hcs, HCS, DP, 1);
635
ret = qemu_coroutine_switch(self, co, COROUTINE_ENTER);
1307
+ u->reg.hcs = FIELD_DP32(u->reg.hcs, HCS, UTRLRDY, 1);
636
1308
+ u->reg.hcs = FIELD_DP32(u->reg.hcs, HCS, UTMRLRDY, 1);
637
qemu_co_queue_run_restart(co);
1309
+ u->reg.ucmdarg2 = UFS_UIC_CMD_RESULT_SUCCESS;
638
diff --git a/util/trace-events b/util/trace-events
1310
+ break;
1311
+ /* TODO: Revisit it when Power Management is implemented */
1312
+ case UFS_UIC_CMD_DME_HIBER_ENTER:
1313
+ u->reg.is = FIELD_DP32(u->reg.is, IS, UHES, 1);
1314
+ u->reg.hcs = FIELD_DP32(u->reg.hcs, HCS, UPMCRS, UFS_PWR_LOCAL);
1315
+ u->reg.ucmdarg2 = UFS_UIC_CMD_RESULT_SUCCESS;
1316
+ break;
1317
+ case UFS_UIC_CMD_DME_HIBER_EXIT:
1318
+ u->reg.is = FIELD_DP32(u->reg.is, IS, UHXS, 1);
1319
+ u->reg.hcs = FIELD_DP32(u->reg.hcs, HCS, UPMCRS, UFS_PWR_LOCAL);
1320
+ u->reg.ucmdarg2 = UFS_UIC_CMD_RESULT_SUCCESS;
1321
+ break;
1322
+ default:
1323
+ u->reg.ucmdarg2 = UFS_UIC_CMD_RESULT_FAILURE;
1324
+ }
1325
+
1326
+ u->reg.is = FIELD_DP32(u->reg.is, IS, UCCS, 1);
1327
+
1328
+ ufs_irq_check(u);
1329
+}
1330
+
1331
+static void ufs_write_reg(UfsHc *u, hwaddr offset, uint32_t data, unsigned size)
1332
+{
1333
+ switch (offset) {
1334
+ case A_IS:
1335
+ u->reg.is &= ~data;
1336
+ ufs_irq_check(u);
1337
+ break;
1338
+ case A_IE:
1339
+ u->reg.ie = data;
1340
+ ufs_irq_check(u);
1341
+ break;
1342
+ case A_HCE:
1343
+ if (!FIELD_EX32(u->reg.hce, HCE, HCE) && FIELD_EX32(data, HCE, HCE)) {
1344
+ u->reg.hcs = FIELD_DP32(u->reg.hcs, HCS, UCRDY, 1);
1345
+ u->reg.hce = FIELD_DP32(u->reg.hce, HCE, HCE, 1);
1346
+ } else if (FIELD_EX32(u->reg.hce, HCE, HCE) &&
1347
+ !FIELD_EX32(data, HCE, HCE)) {
1348
+ u->reg.hcs = 0;
1349
+ u->reg.hce = FIELD_DP32(u->reg.hce, HCE, HCE, 0);
1350
+ }
1351
+ break;
1352
+ case A_UTRLBA:
1353
+ u->reg.utrlba = data & R_UTRLBA_UTRLBA_MASK;
1354
+ break;
1355
+ case A_UTRLBAU:
1356
+ u->reg.utrlbau = data;
1357
+ break;
1358
+ case A_UTRLDBR:
1359
+ /* Not yet supported */
1360
+ break;
1361
+ case A_UTRLRSR:
1362
+ u->reg.utrlrsr = data;
1363
+ break;
1364
+ case A_UTRLCNR:
1365
+ u->reg.utrlcnr &= ~data;
1366
+ break;
1367
+ case A_UTMRLBA:
1368
+ u->reg.utmrlba = data & R_UTMRLBA_UTMRLBA_MASK;
1369
+ break;
1370
+ case A_UTMRLBAU:
1371
+ u->reg.utmrlbau = data;
1372
+ break;
1373
+ case A_UICCMD:
1374
+ ufs_process_uiccmd(u, data);
1375
+ break;
1376
+ case A_UCMDARG1:
1377
+ u->reg.ucmdarg1 = data;
1378
+ break;
1379
+ case A_UCMDARG2:
1380
+ u->reg.ucmdarg2 = data;
1381
+ break;
1382
+ case A_UCMDARG3:
1383
+ u->reg.ucmdarg3 = data;
1384
+ break;
1385
+ case A_UTRLCLR:
1386
+ case A_UTMRLDBR:
1387
+ case A_UTMRLCLR:
1388
+ case A_UTMRLRSR:
1389
+ trace_ufs_err_unsupport_register_offset(offset);
1390
+ break;
1391
+ default:
1392
+ trace_ufs_err_invalid_register_offset(offset);
1393
+ break;
1394
+ }
1395
+}
1396
+
1397
+static uint64_t ufs_mmio_read(void *opaque, hwaddr addr, unsigned size)
1398
+{
1399
+ UfsHc *u = (UfsHc *)opaque;
1400
+ uint8_t *ptr = (uint8_t *)&u->reg;
1401
+ uint64_t value;
1402
+
1403
+ if (addr > sizeof(u->reg) - size) {
1404
+ trace_ufs_err_invalid_register_offset(addr);
1405
+ return 0;
1406
+ }
1407
+
1408
+ value = *(uint32_t *)(ptr + addr);
1409
+ trace_ufs_mmio_read(addr, value, size);
1410
+ return value;
1411
+}
1412
+
1413
+static void ufs_mmio_write(void *opaque, hwaddr addr, uint64_t data,
1414
+ unsigned size)
1415
+{
1416
+ UfsHc *u = (UfsHc *)opaque;
1417
+
1418
+ if (addr > sizeof(u->reg) - size) {
1419
+ trace_ufs_err_invalid_register_offset(addr);
1420
+ return;
1421
+ }
1422
+
1423
+ trace_ufs_mmio_write(addr, data, size);
1424
+ ufs_write_reg(u, addr, data, size);
1425
+}
1426
+
1427
+static const MemoryRegionOps ufs_mmio_ops = {
1428
+ .read = ufs_mmio_read,
1429
+ .write = ufs_mmio_write,
1430
+ .endianness = DEVICE_LITTLE_ENDIAN,
1431
+ .impl = {
1432
+ .min_access_size = 4,
1433
+ .max_access_size = 4,
1434
+ },
1435
+};
1436
+
1437
+static bool ufs_check_constraints(UfsHc *u, Error **errp)
1438
+{
1439
+ if (u->params.nutrs > UFS_MAX_NUTRS) {
1440
+ error_setg(errp, "nutrs must be less than or equal to %d",
1441
+ UFS_MAX_NUTRS);
1442
+ return false;
1443
+ }
1444
+
1445
+ if (u->params.nutmrs > UFS_MAX_NUTMRS) {
1446
+ error_setg(errp, "nutmrs must be less than or equal to %d",
1447
+ UFS_MAX_NUTMRS);
1448
+ return false;
1449
+ }
1450
+
1451
+ return true;
1452
+}
1453
+
1454
+static void ufs_init_pci(UfsHc *u, PCIDevice *pci_dev)
1455
+{
1456
+ uint8_t *pci_conf = pci_dev->config;
1457
+
1458
+ pci_conf[PCI_INTERRUPT_PIN] = 1;
1459
+ pci_config_set_prog_interface(pci_conf, 0x1);
1460
+
1461
+ memory_region_init_io(&u->iomem, OBJECT(u), &ufs_mmio_ops, u, "ufs",
1462
+ u->reg_size);
1463
+ pci_register_bar(pci_dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY, &u->iomem);
1464
+ u->irq = pci_allocate_irq(pci_dev);
1465
+}
1466
+
1467
+static void ufs_init_hc(UfsHc *u)
1468
+{
1469
+ uint32_t cap = 0;
1470
+
1471
+ u->reg_size = pow2ceil(sizeof(UfsReg));
1472
+
1473
+ memset(&u->reg, 0, sizeof(u->reg));
1474
+ cap = FIELD_DP32(cap, CAP, NUTRS, (u->params.nutrs - 1));
1475
+ cap = FIELD_DP32(cap, CAP, RTT, 2);
1476
+ cap = FIELD_DP32(cap, CAP, NUTMRS, (u->params.nutmrs - 1));
1477
+ cap = FIELD_DP32(cap, CAP, AUTOH8, 0);
1478
+ cap = FIELD_DP32(cap, CAP, 64AS, 1);
1479
+ cap = FIELD_DP32(cap, CAP, OODDS, 0);
1480
+ cap = FIELD_DP32(cap, CAP, UICDMETMS, 0);
1481
+ cap = FIELD_DP32(cap, CAP, CS, 0);
1482
+ u->reg.cap = cap;
1483
+ u->reg.ver = UFS_SPEC_VER;
1484
+}
1485
+
1486
+static void ufs_realize(PCIDevice *pci_dev, Error **errp)
1487
+{
1488
+ UfsHc *u = UFS(pci_dev);
1489
+
1490
+ if (!ufs_check_constraints(u, errp)) {
1491
+ return;
1492
+ }
1493
+
1494
+ ufs_init_hc(u);
1495
+ ufs_init_pci(u, pci_dev);
1496
+}
1497
+
1498
+static Property ufs_props[] = {
1499
+ DEFINE_PROP_STRING("serial", UfsHc, params.serial),
1500
+ DEFINE_PROP_UINT8("nutrs", UfsHc, params.nutrs, 32),
1501
+ DEFINE_PROP_UINT8("nutmrs", UfsHc, params.nutmrs, 8),
1502
+ DEFINE_PROP_END_OF_LIST(),
1503
+};
1504
+
1505
+static const VMStateDescription ufs_vmstate = {
1506
+ .name = "ufs",
1507
+ .unmigratable = 1,
1508
+};
1509
+
1510
+static void ufs_class_init(ObjectClass *oc, void *data)
1511
+{
1512
+ DeviceClass *dc = DEVICE_CLASS(oc);
1513
+ PCIDeviceClass *pc = PCI_DEVICE_CLASS(oc);
1514
+
1515
+ pc->realize = ufs_realize;
1516
+ pc->vendor_id = PCI_VENDOR_ID_REDHAT;
1517
+ pc->device_id = PCI_DEVICE_ID_REDHAT_UFS;
1518
+ pc->class_id = PCI_CLASS_STORAGE_UFS;
1519
+
1520
+ set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
1521
+ dc->desc = "Universal Flash Storage";
1522
+ device_class_set_props(dc, ufs_props);
1523
+ dc->vmsd = &ufs_vmstate;
1524
+}
1525
+
1526
+static const TypeInfo ufs_info = {
1527
+ .name = TYPE_UFS,
1528
+ .parent = TYPE_PCI_DEVICE,
1529
+ .class_init = ufs_class_init,
1530
+ .instance_size = sizeof(UfsHc),
1531
+ .interfaces = (InterfaceInfo[]){ { INTERFACE_PCIE_DEVICE }, {} },
1532
+};
1533
+
1534
+static void ufs_register_types(void)
1535
+{
1536
+ type_register_static(&ufs_info);
1537
+}
1538
+
1539
+type_init(ufs_register_types)
1540
diff --git a/hw/Kconfig b/hw/Kconfig
1541
index XXXXXXX..XXXXXXX 100644
639
index XXXXXXX..XXXXXXX 100644
1542
--- a/hw/Kconfig
640
--- a/util/trace-events
1543
+++ b/hw/Kconfig
641
+++ b/util/trace-events
1544
@@ -XXX,XX +XXX,XX @@ source smbios/Kconfig
642
@@ -XXX,XX +XXX,XX @@ run_poll_handlers_end(void *ctx, bool progress) "ctx %p progress %d"
1545
source ssi/Kconfig
643
poll_shrink(void *ctx, int64_t old, int64_t new) "ctx %p old %"PRId64" new %"PRId64
1546
source timer/Kconfig
644
poll_grow(void *ctx, int64_t old, int64_t new) "ctx %p old %"PRId64" new %"PRId64
1547
source tpm/Kconfig
645
1548
+source ufs/Kconfig
646
+# util/async.c
1549
source usb/Kconfig
647
+aio_co_schedule(void *ctx, void *co) "ctx %p co %p"
1550
source virtio/Kconfig
648
+aio_co_schedule_bh_cb(void *ctx, void *co) "ctx %p co %p"
1551
source vfio/Kconfig
649
+
1552
diff --git a/hw/meson.build b/hw/meson.build
650
# util/thread-pool.c
1553
index XXXXXXX..XXXXXXX 100644
651
thread_pool_submit(void *pool, void *req, void *opaque) "pool %p req %p opaque %p"
1554
--- a/hw/meson.build
652
thread_pool_complete(void *pool, void *req, void *opaque, int ret) "pool %p req %p opaque %p ret %d"
1555
+++ b/hw/meson.build
1556
@@ -XXX,XX +XXX,XX @@ subdir('smbios')
1557
subdir('ssi')
1558
subdir('timer')
1559
subdir('tpm')
1560
+subdir('ufs')
1561
subdir('usb')
1562
subdir('vfio')
1563
subdir('virtio')
1564
diff --git a/hw/ufs/Kconfig b/hw/ufs/Kconfig
1565
new file mode 100644
1566
index XXXXXXX..XXXXXXX
1567
--- /dev/null
1568
+++ b/hw/ufs/Kconfig
1569
@@ -XXX,XX +XXX,XX @@
1570
+config UFS_PCI
1571
+ bool
1572
+ default y if PCI_DEVICES
1573
+ depends on PCI
1574
diff --git a/hw/ufs/meson.build b/hw/ufs/meson.build
1575
new file mode 100644
1576
index XXXXXXX..XXXXXXX
1577
--- /dev/null
1578
+++ b/hw/ufs/meson.build
1579
@@ -0,0 +1 @@
1580
+system_ss.add(when: 'CONFIG_UFS_PCI', if_true: files('ufs.c'))
1581
diff --git a/hw/ufs/trace-events b/hw/ufs/trace-events
1582
new file mode 100644
1583
index XXXXXXX..XXXXXXX
1584
--- /dev/null
1585
+++ b/hw/ufs/trace-events
1586
@@ -XXX,XX +XXX,XX @@
1587
+# ufs.c
1588
+ufs_irq_raise(void) "INTx"
1589
+ufs_irq_lower(void) "INTx"
1590
+ufs_mmio_read(uint64_t addr, uint64_t data, unsigned size) "addr 0x%"PRIx64" data 0x%"PRIx64" size %d"
1591
+ufs_mmio_write(uint64_t addr, uint64_t data, unsigned size) "addr 0x%"PRIx64" data 0x%"PRIx64" size %d"
1592
+ufs_process_db(uint32_t slot) "UTRLDBR slot %"PRIu32""
1593
+ufs_process_req(uint32_t slot) "UTRLDBR slot %"PRIu32""
1594
+ufs_complete_req(uint32_t slot) "UTRLDBR slot %"PRIu32""
1595
+ufs_sendback_req(uint32_t slot) "UTRLDBR slot %"PRIu32""
1596
+ufs_exec_nop_cmd(uint32_t slot) "UTRLDBR slot %"PRIu32""
1597
+ufs_exec_scsi_cmd(uint32_t slot, uint8_t lun, uint8_t opcode) "slot %"PRIu32", lun 0x%"PRIx8", opcode 0x%"PRIx8""
1598
+ufs_exec_query_cmd(uint32_t slot, uint8_t opcode) "slot %"PRIu32", opcode 0x%"PRIx8""
1599
+ufs_process_uiccmd(uint32_t uiccmd, uint32_t ucmdarg1, uint32_t ucmdarg2, uint32_t ucmdarg3) "uiccmd 0x%"PRIx32", ucmdarg1 0x%"PRIx32", ucmdarg2 0x%"PRIx32", ucmdarg3 0x%"PRIx32""
1600
+
1601
+# error condition
1602
+ufs_err_dma_read_utrd(uint32_t slot, uint64_t addr) "failed to read utrd. UTRLDBR slot %"PRIu32", UTRD dma addr %"PRIu64""
1603
+ufs_err_dma_read_req_upiu(uint32_t slot, uint64_t addr) "failed to read req upiu. UTRLDBR slot %"PRIu32", request upiu addr %"PRIu64""
1604
+ufs_err_dma_read_prdt(uint32_t slot, uint64_t addr) "failed to read prdt. UTRLDBR slot %"PRIu32", prdt addr %"PRIu64""
1605
+ufs_err_dma_write_utrd(uint32_t slot, uint64_t addr) "failed to write utrd. UTRLDBR slot %"PRIu32", UTRD dma addr %"PRIu64""
1606
+ufs_err_dma_write_rsp_upiu(uint32_t slot, uint64_t addr) "failed to write rsp upiu. UTRLDBR slot %"PRIu32", response upiu addr %"PRIu64""
1607
+ufs_err_utrl_slot_busy(uint32_t slot) "UTRLDBR slot %"PRIu32" is busy"
1608
+ufs_err_unsupport_register_offset(uint32_t offset) "Register offset 0x%"PRIx32" is not yet supported"
1609
+ufs_err_invalid_register_offset(uint32_t offset) "Register offset 0x%"PRIx32" is invalid"
1610
+ufs_err_scsi_cmd_invalid_lun(uint8_t lun) "scsi command has invalid lun: 0x%"PRIx8""
1611
+ufs_err_query_flag_not_readable(uint8_t idn) "query flag idn 0x%"PRIx8" is denied to read"
1612
+ufs_err_query_flag_not_writable(uint8_t idn) "query flag idn 0x%"PRIx8" is denied to write"
1613
+ufs_err_query_attr_not_readable(uint8_t idn) "query attribute idn 0x%"PRIx8" is denied to read"
1614
+ufs_err_query_attr_not_writable(uint8_t idn) "query attribute idn 0x%"PRIx8" is denied to write"
1615
+ufs_err_query_invalid_opcode(uint8_t opcode) "query request has invalid opcode. opcode: 0x%"PRIx8""
1616
+ufs_err_query_invalid_idn(uint8_t opcode, uint8_t idn) "query request has invalid idn. opcode: 0x%"PRIx8", idn 0x%"PRIx8""
1617
+ufs_err_query_invalid_index(uint8_t opcode, uint8_t index) "query request has invalid index. opcode: 0x%"PRIx8", index 0x%"PRIx8""
1618
+ufs_err_invalid_trans_code(uint32_t slot, uint8_t trans_code) "request upiu has invalid transaction code. slot: %"PRIu32", trans_code: 0x%"PRIx8""
1619
--
653
--
1620
2.41.0
654
2.9.3
655
656
diff view generated by jsdifflib
New patch
1
From: Paolo Bonzini <pbonzini@redhat.com>
1
2
3
qcow2_create2 calls this. Do not run a nested event loop, as that
4
breaks when aio_co_wake tries to queue the coroutine on the co_queue_wakeup
5
list of the currently running one.
6
7
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
8
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9
Reviewed-by: Fam Zheng <famz@redhat.com>
10
Message-id: 20170213135235.12274-4-pbonzini@redhat.com
11
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
12
---
13
block/block-backend.c | 12 ++++++++----
14
1 file changed, 8 insertions(+), 4 deletions(-)
15
16
diff --git a/block/block-backend.c b/block/block-backend.c
17
index XXXXXXX..XXXXXXX 100644
18
--- a/block/block-backend.c
19
+++ b/block/block-backend.c
20
@@ -XXX,XX +XXX,XX @@ static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf,
21
{
22
QEMUIOVector qiov;
23
struct iovec iov;
24
- Coroutine *co;
25
BlkRwCo rwco;
26
27
iov = (struct iovec) {
28
@@ -XXX,XX +XXX,XX @@ static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf,
29
.ret = NOT_DONE,
30
};
31
32
- co = qemu_coroutine_create(co_entry, &rwco);
33
- qemu_coroutine_enter(co);
34
- BDRV_POLL_WHILE(blk_bs(blk), rwco.ret == NOT_DONE);
35
+ if (qemu_in_coroutine()) {
36
+ /* Fast-path if already in coroutine context */
37
+ co_entry(&rwco);
38
+ } else {
39
+ Coroutine *co = qemu_coroutine_create(co_entry, &rwco);
40
+ qemu_coroutine_enter(co);
41
+ BDRV_POLL_WHILE(blk_bs(blk), rwco.ret == NOT_DONE);
42
+ }
43
44
return rwco.ret;
45
}
46
--
47
2.9.3
48
49
diff view generated by jsdifflib
New patch
1
From: Paolo Bonzini <pbonzini@redhat.com>
1
2
3
Once the thread pool starts using aio_co_wake, it will also need
4
qemu_get_current_aio_context(). Make test-thread-pool create
5
an AioContext with qemu_init_main_loop, so that stubs/iothread.c
6
and tests/iothread.c can provide the rest.
7
8
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
9
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
10
Reviewed-by: Fam Zheng <famz@redhat.com>
11
Message-id: 20170213135235.12274-5-pbonzini@redhat.com
12
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
13
---
14
tests/test-thread-pool.c | 12 +++---------
15
1 file changed, 3 insertions(+), 9 deletions(-)
16
17
diff --git a/tests/test-thread-pool.c b/tests/test-thread-pool.c
18
index XXXXXXX..XXXXXXX 100644
19
--- a/tests/test-thread-pool.c
20
+++ b/tests/test-thread-pool.c
21
@@ -XXX,XX +XXX,XX @@
22
#include "qapi/error.h"
23
#include "qemu/timer.h"
24
#include "qemu/error-report.h"
25
+#include "qemu/main-loop.h"
26
27
static AioContext *ctx;
28
static ThreadPool *pool;
29
@@ -XXX,XX +XXX,XX @@ static void test_cancel_async(void)
30
int main(int argc, char **argv)
31
{
32
int ret;
33
- Error *local_error = NULL;
34
35
- init_clocks();
36
-
37
- ctx = aio_context_new(&local_error);
38
- if (!ctx) {
39
- error_reportf_err(local_error, "Failed to create AIO Context: ");
40
- exit(1);
41
- }
42
+ qemu_init_main_loop(&error_abort);
43
+ ctx = qemu_get_current_aio_context();
44
pool = aio_get_thread_pool(ctx);
45
46
g_test_init(&argc, &argv, NULL);
47
@@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv)
48
49
ret = g_test_run();
50
51
- aio_context_unref(ctx);
52
return ret;
53
}
54
--
55
2.9.3
56
57
diff view generated by jsdifflib
New patch
1
1
From: Paolo Bonzini <pbonzini@redhat.com>
2
3
This is in preparation for making qio_channel_yield work on
4
AioContexts other than the main one.
5
6
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
7
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
8
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9
Reviewed-by: Fam Zheng <famz@redhat.com>
10
Message-id: 20170213135235.12274-6-pbonzini@redhat.com
11
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
12
---
13
include/io/channel.h | 25 +++++++++++++++++++++++++
14
io/channel-command.c | 13 +++++++++++++
15
io/channel-file.c | 11 +++++++++++
16
io/channel-socket.c | 16 +++++++++++-----
17
io/channel-tls.c | 12 ++++++++++++
18
io/channel-watch.c | 6 ++++++
19
io/channel.c | 11 +++++++++++
20
7 files changed, 89 insertions(+), 5 deletions(-)
21
22
diff --git a/include/io/channel.h b/include/io/channel.h
23
index XXXXXXX..XXXXXXX 100644
24
--- a/include/io/channel.h
25
+++ b/include/io/channel.h
26
@@ -XXX,XX +XXX,XX @@
27
28
#include "qemu-common.h"
29
#include "qom/object.h"
30
+#include "block/aio.h"
31
32
#define TYPE_QIO_CHANNEL "qio-channel"
33
#define QIO_CHANNEL(obj) \
34
@@ -XXX,XX +XXX,XX @@ struct QIOChannelClass {
35
off_t offset,
36
int whence,
37
Error **errp);
38
+ void (*io_set_aio_fd_handler)(QIOChannel *ioc,
39
+ AioContext *ctx,
40
+ IOHandler *io_read,
41
+ IOHandler *io_write,
42
+ void *opaque);
43
};
44
45
/* General I/O handling functions */
46
@@ -XXX,XX +XXX,XX @@ void qio_channel_yield(QIOChannel *ioc,
47
void qio_channel_wait(QIOChannel *ioc,
48
GIOCondition condition);
49
50
+/**
51
+ * qio_channel_set_aio_fd_handler:
52
+ * @ioc: the channel object
53
+ * @ctx: the AioContext to set the handlers on
54
+ * @io_read: the read handler
55
+ * @io_write: the write handler
56
+ * @opaque: the opaque value passed to the handler
57
+ *
58
+ * This is used internally by qio_channel_yield(). It can
59
+ * be used by channel implementations to forward the handlers
60
+ * to another channel (e.g. from #QIOChannelTLS to the
61
+ * underlying socket).
62
+ */
63
+void qio_channel_set_aio_fd_handler(QIOChannel *ioc,
64
+ AioContext *ctx,
65
+ IOHandler *io_read,
66
+ IOHandler *io_write,
67
+ void *opaque);
68
+
69
#endif /* QIO_CHANNEL_H */
70
diff --git a/io/channel-command.c b/io/channel-command.c
71
index XXXXXXX..XXXXXXX 100644
72
--- a/io/channel-command.c
73
+++ b/io/channel-command.c
74
@@ -XXX,XX +XXX,XX @@ static int qio_channel_command_close(QIOChannel *ioc,
75
}
76
77
78
+static void qio_channel_command_set_aio_fd_handler(QIOChannel *ioc,
79
+ AioContext *ctx,
80
+ IOHandler *io_read,
81
+ IOHandler *io_write,
82
+ void *opaque)
83
+{
84
+ QIOChannelCommand *cioc = QIO_CHANNEL_COMMAND(ioc);
85
+ aio_set_fd_handler(ctx, cioc->readfd, false, io_read, NULL, NULL, opaque);
86
+ aio_set_fd_handler(ctx, cioc->writefd, false, NULL, io_write, NULL, opaque);
87
+}
88
+
89
+
90
static GSource *qio_channel_command_create_watch(QIOChannel *ioc,
91
GIOCondition condition)
92
{
93
@@ -XXX,XX +XXX,XX @@ static void qio_channel_command_class_init(ObjectClass *klass,
94
ioc_klass->io_set_blocking = qio_channel_command_set_blocking;
95
ioc_klass->io_close = qio_channel_command_close;
96
ioc_klass->io_create_watch = qio_channel_command_create_watch;
97
+ ioc_klass->io_set_aio_fd_handler = qio_channel_command_set_aio_fd_handler;
98
}
99
100
static const TypeInfo qio_channel_command_info = {
101
diff --git a/io/channel-file.c b/io/channel-file.c
102
index XXXXXXX..XXXXXXX 100644
103
--- a/io/channel-file.c
104
+++ b/io/channel-file.c
105
@@ -XXX,XX +XXX,XX @@ static int qio_channel_file_close(QIOChannel *ioc,
106
}
107
108
109
+static void qio_channel_file_set_aio_fd_handler(QIOChannel *ioc,
110
+ AioContext *ctx,
111
+ IOHandler *io_read,
112
+ IOHandler *io_write,
113
+ void *opaque)
114
+{
115
+ QIOChannelFile *fioc = QIO_CHANNEL_FILE(ioc);
116
+ aio_set_fd_handler(ctx, fioc->fd, false, io_read, io_write, NULL, opaque);
117
+}
118
+
119
static GSource *qio_channel_file_create_watch(QIOChannel *ioc,
120
GIOCondition condition)
121
{
122
@@ -XXX,XX +XXX,XX @@ static void qio_channel_file_class_init(ObjectClass *klass,
123
ioc_klass->io_seek = qio_channel_file_seek;
124
ioc_klass->io_close = qio_channel_file_close;
125
ioc_klass->io_create_watch = qio_channel_file_create_watch;
126
+ ioc_klass->io_set_aio_fd_handler = qio_channel_file_set_aio_fd_handler;
127
}
128
129
static const TypeInfo qio_channel_file_info = {
130
diff --git a/io/channel-socket.c b/io/channel-socket.c
131
index XXXXXXX..XXXXXXX 100644
132
--- a/io/channel-socket.c
133
+++ b/io/channel-socket.c
134
@@ -XXX,XX +XXX,XX @@ qio_channel_socket_set_blocking(QIOChannel *ioc,
135
qemu_set_block(sioc->fd);
136
} else {
137
qemu_set_nonblock(sioc->fd);
138
-#ifdef WIN32
139
- WSAEventSelect(sioc->fd, ioc->event,
140
- FD_READ | FD_ACCEPT | FD_CLOSE |
141
- FD_CONNECT | FD_WRITE | FD_OOB);
142
-#endif
143
}
144
return 0;
145
}
146
@@ -XXX,XX +XXX,XX @@ qio_channel_socket_shutdown(QIOChannel *ioc,
147
return 0;
148
}
149
150
+static void qio_channel_socket_set_aio_fd_handler(QIOChannel *ioc,
151
+ AioContext *ctx,
152
+ IOHandler *io_read,
153
+ IOHandler *io_write,
154
+ void *opaque)
155
+{
156
+ QIOChannelSocket *sioc = QIO_CHANNEL_SOCKET(ioc);
157
+ aio_set_fd_handler(ctx, sioc->fd, false, io_read, io_write, NULL, opaque);
158
+}
159
+
160
static GSource *qio_channel_socket_create_watch(QIOChannel *ioc,
161
GIOCondition condition)
162
{
163
@@ -XXX,XX +XXX,XX @@ static void qio_channel_socket_class_init(ObjectClass *klass,
164
ioc_klass->io_set_cork = qio_channel_socket_set_cork;
165
ioc_klass->io_set_delay = qio_channel_socket_set_delay;
166
ioc_klass->io_create_watch = qio_channel_socket_create_watch;
167
+ ioc_klass->io_set_aio_fd_handler = qio_channel_socket_set_aio_fd_handler;
168
}
169
170
static const TypeInfo qio_channel_socket_info = {
171
diff --git a/io/channel-tls.c b/io/channel-tls.c
172
index XXXXXXX..XXXXXXX 100644
173
--- a/io/channel-tls.c
174
+++ b/io/channel-tls.c
175
@@ -XXX,XX +XXX,XX @@ static int qio_channel_tls_close(QIOChannel *ioc,
176
return qio_channel_close(tioc->master, errp);
177
}
178
179
+static void qio_channel_tls_set_aio_fd_handler(QIOChannel *ioc,
180
+ AioContext *ctx,
181
+ IOHandler *io_read,
182
+ IOHandler *io_write,
183
+ void *opaque)
184
+{
185
+ QIOChannelTLS *tioc = QIO_CHANNEL_TLS(ioc);
186
+
187
+ qio_channel_set_aio_fd_handler(tioc->master, ctx, io_read, io_write, opaque);
188
+}
189
+
190
static GSource *qio_channel_tls_create_watch(QIOChannel *ioc,
191
GIOCondition condition)
192
{
193
@@ -XXX,XX +XXX,XX @@ static void qio_channel_tls_class_init(ObjectClass *klass,
194
ioc_klass->io_close = qio_channel_tls_close;
195
ioc_klass->io_shutdown = qio_channel_tls_shutdown;
196
ioc_klass->io_create_watch = qio_channel_tls_create_watch;
197
+ ioc_klass->io_set_aio_fd_handler = qio_channel_tls_set_aio_fd_handler;
198
}
199
200
static const TypeInfo qio_channel_tls_info = {
201
diff --git a/io/channel-watch.c b/io/channel-watch.c
202
index XXXXXXX..XXXXXXX 100644
203
--- a/io/channel-watch.c
204
+++ b/io/channel-watch.c
205
@@ -XXX,XX +XXX,XX @@ GSource *qio_channel_create_socket_watch(QIOChannel *ioc,
206
GSource *source;
207
QIOChannelSocketSource *ssource;
208
209
+#ifdef WIN32
210
+ WSAEventSelect(socket, ioc->event,
211
+ FD_READ | FD_ACCEPT | FD_CLOSE |
212
+ FD_CONNECT | FD_WRITE | FD_OOB);
213
+#endif
214
+
215
source = g_source_new(&qio_channel_socket_source_funcs,
216
sizeof(QIOChannelSocketSource));
217
ssource = (QIOChannelSocketSource *)source;
218
diff --git a/io/channel.c b/io/channel.c
219
index XXXXXXX..XXXXXXX 100644
220
--- a/io/channel.c
221
+++ b/io/channel.c
222
@@ -XXX,XX +XXX,XX @@ GSource *qio_channel_create_watch(QIOChannel *ioc,
223
}
224
225
226
+void qio_channel_set_aio_fd_handler(QIOChannel *ioc,
227
+ AioContext *ctx,
228
+ IOHandler *io_read,
229
+ IOHandler *io_write,
230
+ void *opaque)
231
+{
232
+ QIOChannelClass *klass = QIO_CHANNEL_GET_CLASS(ioc);
233
+
234
+ klass->io_set_aio_fd_handler(ioc, ctx, io_read, io_write, opaque);
235
+}
236
+
237
guint qio_channel_add_watch(QIOChannel *ioc,
238
GIOCondition condition,
239
QIOChannelFunc func,
240
--
241
2.9.3
242
243
diff view generated by jsdifflib
New patch
1
1
From: Paolo Bonzini <pbonzini@redhat.com>
2
3
Support separate coroutines for reading and writing, and place the
4
read/write handlers on the AioContext that the QIOChannel is registered
5
with.
6
7
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
8
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
9
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
10
Reviewed-by: Fam Zheng <famz@redhat.com>
11
Message-id: 20170213135235.12274-7-pbonzini@redhat.com
12
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
13
---
14
include/io/channel.h | 47 ++++++++++++++++++++++++++--
15
io/channel.c | 86 +++++++++++++++++++++++++++++++++++++++-------------
16
2 files changed, 109 insertions(+), 24 deletions(-)
17
18
diff --git a/include/io/channel.h b/include/io/channel.h
19
index XXXXXXX..XXXXXXX 100644
20
--- a/include/io/channel.h
21
+++ b/include/io/channel.h
22
@@ -XXX,XX +XXX,XX @@
23
24
#include "qemu-common.h"
25
#include "qom/object.h"
26
+#include "qemu/coroutine.h"
27
#include "block/aio.h"
28
29
#define TYPE_QIO_CHANNEL "qio-channel"
30
@@ -XXX,XX +XXX,XX @@ struct QIOChannel {
31
Object parent;
32
unsigned int features; /* bitmask of QIOChannelFeatures */
33
char *name;
34
+ AioContext *ctx;
35
+ Coroutine *read_coroutine;
36
+ Coroutine *write_coroutine;
37
#ifdef _WIN32
38
HANDLE event; /* For use with GSource on Win32 */
39
#endif
40
@@ -XXX,XX +XXX,XX @@ guint qio_channel_add_watch(QIOChannel *ioc,
41
42
43
/**
44
+ * qio_channel_attach_aio_context:
45
+ * @ioc: the channel object
46
+ * @ctx: the #AioContext to set the handlers on
47
+ *
48
+ * Request that qio_channel_yield() sets I/O handlers on
49
+ * the given #AioContext. If @ctx is %NULL, qio_channel_yield()
50
+ * uses QEMU's main thread event loop.
51
+ *
52
+ * You can move a #QIOChannel from one #AioContext to another even if
53
+ * I/O handlers are set for a coroutine. However, #QIOChannel provides
54
+ * no synchronization between the calls to qio_channel_yield() and
55
+ * qio_channel_attach_aio_context().
56
+ *
57
+ * Therefore you should first call qio_channel_detach_aio_context()
58
+ * to ensure that the coroutine is not entered concurrently. Then,
59
+ * while the coroutine has yielded, call qio_channel_attach_aio_context(),
60
+ * and then aio_co_schedule() to place the coroutine on the new
61
+ * #AioContext. The calls to qio_channel_detach_aio_context()
62
+ * and qio_channel_attach_aio_context() should be protected with
63
+ * aio_context_acquire() and aio_context_release().
64
+ */
65
+void qio_channel_attach_aio_context(QIOChannel *ioc,
66
+ AioContext *ctx);
67
+
68
+/**
69
+ * qio_channel_detach_aio_context:
70
+ * @ioc: the channel object
71
+ *
72
+ * Disable any I/O handlers set by qio_channel_yield(). With the
73
+ * help of aio_co_schedule(), this allows moving a coroutine that was
74
+ * paused by qio_channel_yield() to another context.
75
+ */
76
+void qio_channel_detach_aio_context(QIOChannel *ioc);
77
+
78
+/**
79
* qio_channel_yield:
80
* @ioc: the channel object
81
* @condition: the I/O condition to wait for
82
*
83
- * Yields execution from the current coroutine until
84
- * the condition indicated by @condition becomes
85
- * available.
86
+ * Yields execution from the current coroutine until the condition
87
+ * indicated by @condition becomes available. @condition must
88
+ * be either %G_IO_IN or %G_IO_OUT; it cannot contain both. In
89
+ * addition, no two coroutine can be waiting on the same condition
90
+ * and channel at the same time.
91
*
92
* This must only be called from coroutine context
93
*/
94
diff --git a/io/channel.c b/io/channel.c
95
index XXXXXXX..XXXXXXX 100644
96
--- a/io/channel.c
97
+++ b/io/channel.c
98
@@ -XXX,XX +XXX,XX @@
99
#include "qemu/osdep.h"
100
#include "io/channel.h"
101
#include "qapi/error.h"
102
-#include "qemu/coroutine.h"
103
+#include "qemu/main-loop.h"
104
105
bool qio_channel_has_feature(QIOChannel *ioc,
106
QIOChannelFeature feature)
107
@@ -XXX,XX +XXX,XX @@ off_t qio_channel_io_seek(QIOChannel *ioc,
108
}
109
110
111
-typedef struct QIOChannelYieldData QIOChannelYieldData;
112
-struct QIOChannelYieldData {
113
- QIOChannel *ioc;
114
- Coroutine *co;
115
-};
116
+static void qio_channel_set_aio_fd_handlers(QIOChannel *ioc);
117
118
+static void qio_channel_restart_read(void *opaque)
119
+{
120
+ QIOChannel *ioc = opaque;
121
+ Coroutine *co = ioc->read_coroutine;
122
+
123
+ ioc->read_coroutine = NULL;
124
+ qio_channel_set_aio_fd_handlers(ioc);
125
+ aio_co_wake(co);
126
+}
127
128
-static gboolean qio_channel_yield_enter(QIOChannel *ioc,
129
- GIOCondition condition,
130
- gpointer opaque)
131
+static void qio_channel_restart_write(void *opaque)
132
{
133
- QIOChannelYieldData *data = opaque;
134
- qemu_coroutine_enter(data->co);
135
- return FALSE;
136
+ QIOChannel *ioc = opaque;
137
+ Coroutine *co = ioc->write_coroutine;
138
+
139
+ ioc->write_coroutine = NULL;
140
+ qio_channel_set_aio_fd_handlers(ioc);
141
+ aio_co_wake(co);
142
}
143
144
+static void qio_channel_set_aio_fd_handlers(QIOChannel *ioc)
145
+{
146
+ IOHandler *rd_handler = NULL, *wr_handler = NULL;
147
+ AioContext *ctx;
148
+
149
+ if (ioc->read_coroutine) {
150
+ rd_handler = qio_channel_restart_read;
151
+ }
152
+ if (ioc->write_coroutine) {
153
+ wr_handler = qio_channel_restart_write;
154
+ }
155
+
156
+ ctx = ioc->ctx ? ioc->ctx : iohandler_get_aio_context();
157
+ qio_channel_set_aio_fd_handler(ioc, ctx, rd_handler, wr_handler, ioc);
158
+}
159
+
160
+void qio_channel_attach_aio_context(QIOChannel *ioc,
161
+ AioContext *ctx)
162
+{
163
+ AioContext *old_ctx;
164
+ if (ioc->ctx == ctx) {
165
+ return;
166
+ }
167
+
168
+ old_ctx = ioc->ctx ? ioc->ctx : iohandler_get_aio_context();
169
+ qio_channel_set_aio_fd_handler(ioc, old_ctx, NULL, NULL, NULL);
170
+ ioc->ctx = ctx;
171
+ qio_channel_set_aio_fd_handlers(ioc);
172
+}
173
+
174
+void qio_channel_detach_aio_context(QIOChannel *ioc)
175
+{
176
+ ioc->read_coroutine = NULL;
177
+ ioc->write_coroutine = NULL;
178
+ qio_channel_set_aio_fd_handlers(ioc);
179
+ ioc->ctx = NULL;
180
+}
181
182
void coroutine_fn qio_channel_yield(QIOChannel *ioc,
183
GIOCondition condition)
184
{
185
- QIOChannelYieldData data;
186
-
187
assert(qemu_in_coroutine());
188
- data.ioc = ioc;
189
- data.co = qemu_coroutine_self();
190
- qio_channel_add_watch(ioc,
191
- condition,
192
- qio_channel_yield_enter,
193
- &data,
194
- NULL);
195
+ if (condition == G_IO_IN) {
196
+ assert(!ioc->read_coroutine);
197
+ ioc->read_coroutine = qemu_coroutine_self();
198
+ } else if (condition == G_IO_OUT) {
199
+ assert(!ioc->write_coroutine);
200
+ ioc->write_coroutine = qemu_coroutine_self();
201
+ } else {
202
+ abort();
203
+ }
204
+ qio_channel_set_aio_fd_handlers(ioc);
205
qemu_coroutine_yield();
206
}
207
208
--
209
2.9.3
210
211
diff view generated by jsdifflib
New patch
1
From: Paolo Bonzini <pbonzini@redhat.com>
1
2
3
In the client, read the reply headers from a coroutine, switching the
4
read side between the "read header" coroutine and the I/O coroutine that
5
reads the body of the reply.
6
7
In the server, if the server can read more requests it will create a new
8
"read request" coroutine as soon as a request has been read. Otherwise,
9
the new coroutine is created in nbd_request_put.
10
11
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
12
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
13
Reviewed-by: Fam Zheng <famz@redhat.com>
14
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
15
Message-id: 20170213135235.12274-8-pbonzini@redhat.com
16
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
17
---
18
block/nbd-client.h | 2 +-
19
block/nbd-client.c | 117 ++++++++++++++++++++++++-----------------------------
20
nbd/client.c | 2 +-
21
nbd/common.c | 9 +----
22
nbd/server.c | 94 +++++++++++++-----------------------------
23
5 files changed, 83 insertions(+), 141 deletions(-)
24
25
diff --git a/block/nbd-client.h b/block/nbd-client.h
26
index XXXXXXX..XXXXXXX 100644
27
--- a/block/nbd-client.h
28
+++ b/block/nbd-client.h
29
@@ -XXX,XX +XXX,XX @@ typedef struct NBDClientSession {
30
31
CoMutex send_mutex;
32
CoQueue free_sema;
33
- Coroutine *send_coroutine;
34
+ Coroutine *read_reply_co;
35
int in_flight;
36
37
Coroutine *recv_coroutine[MAX_NBD_REQUESTS];
38
diff --git a/block/nbd-client.c b/block/nbd-client.c
39
index XXXXXXX..XXXXXXX 100644
40
--- a/block/nbd-client.c
41
+++ b/block/nbd-client.c
42
@@ -XXX,XX +XXX,XX @@
43
#define HANDLE_TO_INDEX(bs, handle) ((handle) ^ ((uint64_t)(intptr_t)bs))
44
#define INDEX_TO_HANDLE(bs, index) ((index) ^ ((uint64_t)(intptr_t)bs))
45
46
-static void nbd_recv_coroutines_enter_all(NBDClientSession *s)
47
+static void nbd_recv_coroutines_enter_all(BlockDriverState *bs)
48
{
49
+ NBDClientSession *s = nbd_get_client_session(bs);
50
int i;
51
52
for (i = 0; i < MAX_NBD_REQUESTS; i++) {
53
@@ -XXX,XX +XXX,XX @@ static void nbd_recv_coroutines_enter_all(NBDClientSession *s)
54
qemu_coroutine_enter(s->recv_coroutine[i]);
55
}
56
}
57
+ BDRV_POLL_WHILE(bs, s->read_reply_co);
58
}
59
60
static void nbd_teardown_connection(BlockDriverState *bs)
61
@@ -XXX,XX +XXX,XX @@ static void nbd_teardown_connection(BlockDriverState *bs)
62
qio_channel_shutdown(client->ioc,
63
QIO_CHANNEL_SHUTDOWN_BOTH,
64
NULL);
65
- nbd_recv_coroutines_enter_all(client);
66
+ nbd_recv_coroutines_enter_all(bs);
67
68
nbd_client_detach_aio_context(bs);
69
object_unref(OBJECT(client->sioc));
70
@@ -XXX,XX +XXX,XX @@ static void nbd_teardown_connection(BlockDriverState *bs)
71
client->ioc = NULL;
72
}
73
74
-static void nbd_reply_ready(void *opaque)
75
+static coroutine_fn void nbd_read_reply_entry(void *opaque)
76
{
77
- BlockDriverState *bs = opaque;
78
- NBDClientSession *s = nbd_get_client_session(bs);
79
+ NBDClientSession *s = opaque;
80
uint64_t i;
81
int ret;
82
83
- if (!s->ioc) { /* Already closed */
84
- return;
85
- }
86
-
87
- if (s->reply.handle == 0) {
88
- /* No reply already in flight. Fetch a header. It is possible
89
- * that another thread has done the same thing in parallel, so
90
- * the socket is not readable anymore.
91
- */
92
+ for (;;) {
93
+ assert(s->reply.handle == 0);
94
ret = nbd_receive_reply(s->ioc, &s->reply);
95
- if (ret == -EAGAIN) {
96
- return;
97
- }
98
if (ret < 0) {
99
- s->reply.handle = 0;
100
- goto fail;
101
+ break;
102
}
103
- }
104
105
- /* There's no need for a mutex on the receive side, because the
106
- * handler acts as a synchronization point and ensures that only
107
- * one coroutine is called until the reply finishes. */
108
- i = HANDLE_TO_INDEX(s, s->reply.handle);
109
- if (i >= MAX_NBD_REQUESTS) {
110
- goto fail;
111
- }
112
+ /* There's no need for a mutex on the receive side, because the
113
+ * handler acts as a synchronization point and ensures that only
114
+ * one coroutine is called until the reply finishes.
115
+ */
116
+ i = HANDLE_TO_INDEX(s, s->reply.handle);
117
+ if (i >= MAX_NBD_REQUESTS || !s->recv_coroutine[i]) {
118
+ break;
119
+ }
120
121
- if (s->recv_coroutine[i]) {
122
- qemu_coroutine_enter(s->recv_coroutine[i]);
123
- return;
124
+ /* We're woken up by the recv_coroutine itself. Note that there
125
+ * is no race between yielding and reentering read_reply_co. This
126
+ * is because:
127
+ *
128
+ * - if recv_coroutine[i] runs on the same AioContext, it is only
129
+ * entered after we yield
130
+ *
131
+ * - if recv_coroutine[i] runs on a different AioContext, reentering
132
+ * read_reply_co happens through a bottom half, which can only
133
+ * run after we yield.
134
+ */
135
+ aio_co_wake(s->recv_coroutine[i]);
136
+ qemu_coroutine_yield();
137
}
138
-
139
-fail:
140
- nbd_teardown_connection(bs);
141
-}
142
-
143
-static void nbd_restart_write(void *opaque)
144
-{
145
- BlockDriverState *bs = opaque;
146
-
147
- qemu_coroutine_enter(nbd_get_client_session(bs)->send_coroutine);
148
+ s->read_reply_co = NULL;
149
}
150
151
static int nbd_co_send_request(BlockDriverState *bs,
152
@@ -XXX,XX +XXX,XX @@ static int nbd_co_send_request(BlockDriverState *bs,
153
QEMUIOVector *qiov)
154
{
155
NBDClientSession *s = nbd_get_client_session(bs);
156
- AioContext *aio_context;
157
int rc, ret, i;
158
159
qemu_co_mutex_lock(&s->send_mutex);
160
@@ -XXX,XX +XXX,XX @@ static int nbd_co_send_request(BlockDriverState *bs,
161
return -EPIPE;
162
}
163
164
- s->send_coroutine = qemu_coroutine_self();
165
- aio_context = bdrv_get_aio_context(bs);
166
-
167
- aio_set_fd_handler(aio_context, s->sioc->fd, false,
168
- nbd_reply_ready, nbd_restart_write, NULL, bs);
169
if (qiov) {
170
qio_channel_set_cork(s->ioc, true);
171
rc = nbd_send_request(s->ioc, request);
172
@@ -XXX,XX +XXX,XX @@ static int nbd_co_send_request(BlockDriverState *bs,
173
} else {
174
rc = nbd_send_request(s->ioc, request);
175
}
176
- aio_set_fd_handler(aio_context, s->sioc->fd, false,
177
- nbd_reply_ready, NULL, NULL, bs);
178
- s->send_coroutine = NULL;
179
qemu_co_mutex_unlock(&s->send_mutex);
180
return rc;
181
}
182
@@ -XXX,XX +XXX,XX @@ static void nbd_co_receive_reply(NBDClientSession *s,
183
{
184
int ret;
185
186
- /* Wait until we're woken up by the read handler. TODO: perhaps
187
- * peek at the next reply and avoid yielding if it's ours? */
188
+ /* Wait until we're woken up by nbd_read_reply_entry. */
189
qemu_coroutine_yield();
190
*reply = s->reply;
191
if (reply->handle != request->handle ||
192
@@ -XXX,XX +XXX,XX @@ static void nbd_coroutine_start(NBDClientSession *s,
193
/* s->recv_coroutine[i] is set as soon as we get the send_lock. */
194
}
195
196
-static void nbd_coroutine_end(NBDClientSession *s,
197
+static void nbd_coroutine_end(BlockDriverState *bs,
198
NBDRequest *request)
199
{
200
+ NBDClientSession *s = nbd_get_client_session(bs);
201
int i = HANDLE_TO_INDEX(s, request->handle);
202
+
203
s->recv_coroutine[i] = NULL;
204
- if (s->in_flight-- == MAX_NBD_REQUESTS) {
205
- qemu_co_queue_next(&s->free_sema);
206
+ s->in_flight--;
207
+ qemu_co_queue_next(&s->free_sema);
208
+
209
+ /* Kick the read_reply_co to get the next reply. */
210
+ if (s->read_reply_co) {
211
+ aio_co_wake(s->read_reply_co);
212
}
213
}
214
215
@@ -XXX,XX +XXX,XX @@ int nbd_client_co_preadv(BlockDriverState *bs, uint64_t offset,
216
} else {
217
nbd_co_receive_reply(client, &request, &reply, qiov);
218
}
219
- nbd_coroutine_end(client, &request);
220
+ nbd_coroutine_end(bs, &request);
221
return -reply.error;
222
}
223
224
@@ -XXX,XX +XXX,XX @@ int nbd_client_co_pwritev(BlockDriverState *bs, uint64_t offset,
225
} else {
226
nbd_co_receive_reply(client, &request, &reply, NULL);
227
}
228
- nbd_coroutine_end(client, &request);
229
+ nbd_coroutine_end(bs, &request);
230
return -reply.error;
231
}
232
233
@@ -XXX,XX +XXX,XX @@ int nbd_client_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
234
} else {
235
nbd_co_receive_reply(client, &request, &reply, NULL);
236
}
237
- nbd_coroutine_end(client, &request);
238
+ nbd_coroutine_end(bs, &request);
239
return -reply.error;
240
}
241
242
@@ -XXX,XX +XXX,XX @@ int nbd_client_co_flush(BlockDriverState *bs)
243
} else {
244
nbd_co_receive_reply(client, &request, &reply, NULL);
245
}
246
- nbd_coroutine_end(client, &request);
247
+ nbd_coroutine_end(bs, &request);
248
return -reply.error;
249
}
250
251
@@ -XXX,XX +XXX,XX @@ int nbd_client_co_pdiscard(BlockDriverState *bs, int64_t offset, int count)
252
} else {
253
nbd_co_receive_reply(client, &request, &reply, NULL);
254
}
255
- nbd_coroutine_end(client, &request);
256
+ nbd_coroutine_end(bs, &request);
257
return -reply.error;
258
259
}
260
261
void nbd_client_detach_aio_context(BlockDriverState *bs)
262
{
263
- aio_set_fd_handler(bdrv_get_aio_context(bs),
264
- nbd_get_client_session(bs)->sioc->fd,
265
- false, NULL, NULL, NULL, NULL);
266
+ NBDClientSession *client = nbd_get_client_session(bs);
267
+ qio_channel_detach_aio_context(QIO_CHANNEL(client->sioc));
268
}
269
270
void nbd_client_attach_aio_context(BlockDriverState *bs,
271
AioContext *new_context)
272
{
273
- aio_set_fd_handler(new_context, nbd_get_client_session(bs)->sioc->fd,
274
- false, nbd_reply_ready, NULL, NULL, bs);
275
+ NBDClientSession *client = nbd_get_client_session(bs);
276
+ qio_channel_attach_aio_context(QIO_CHANNEL(client->sioc), new_context);
277
+ aio_co_schedule(new_context, client->read_reply_co);
278
}
279
280
void nbd_client_close(BlockDriverState *bs)
281
@@ -XXX,XX +XXX,XX @@ int nbd_client_init(BlockDriverState *bs,
282
/* Now that we're connected, set the socket to be non-blocking and
283
* kick the reply mechanism. */
284
qio_channel_set_blocking(QIO_CHANNEL(sioc), false, NULL);
285
-
286
+ client->read_reply_co = qemu_coroutine_create(nbd_read_reply_entry, client);
287
nbd_client_attach_aio_context(bs, bdrv_get_aio_context(bs));
288
289
logout("Established connection with NBD server\n");
290
diff --git a/nbd/client.c b/nbd/client.c
291
index XXXXXXX..XXXXXXX 100644
292
--- a/nbd/client.c
293
+++ b/nbd/client.c
294
@@ -XXX,XX +XXX,XX @@ ssize_t nbd_receive_reply(QIOChannel *ioc, NBDReply *reply)
295
ssize_t ret;
296
297
ret = read_sync(ioc, buf, sizeof(buf));
298
- if (ret < 0) {
299
+ if (ret <= 0) {
300
return ret;
301
}
302
303
diff --git a/nbd/common.c b/nbd/common.c
304
index XXXXXXX..XXXXXXX 100644
305
--- a/nbd/common.c
306
+++ b/nbd/common.c
307
@@ -XXX,XX +XXX,XX @@ ssize_t nbd_wr_syncv(QIOChannel *ioc,
308
}
309
if (len == QIO_CHANNEL_ERR_BLOCK) {
310
if (qemu_in_coroutine()) {
311
- /* XXX figure out if we can create a variant on
312
- * qio_channel_yield() that works with AIO contexts
313
- * and consider using that in this branch */
314
- qemu_coroutine_yield();
315
- } else if (done) {
316
- /* XXX this is needed by nbd_reply_ready. */
317
- qio_channel_wait(ioc,
318
- do_read ? G_IO_IN : G_IO_OUT);
319
+ qio_channel_yield(ioc, do_read ? G_IO_IN : G_IO_OUT);
320
} else {
321
return -EAGAIN;
322
}
323
diff --git a/nbd/server.c b/nbd/server.c
324
index XXXXXXX..XXXXXXX 100644
325
--- a/nbd/server.c
326
+++ b/nbd/server.c
327
@@ -XXX,XX +XXX,XX @@ struct NBDClient {
328
CoMutex send_lock;
329
Coroutine *send_coroutine;
330
331
- bool can_read;
332
-
333
QTAILQ_ENTRY(NBDClient) next;
334
int nb_requests;
335
bool closing;
336
@@ -XXX,XX +XXX,XX @@ struct NBDClient {
337
338
/* That's all folks */
339
340
-static void nbd_set_handlers(NBDClient *client);
341
-static void nbd_unset_handlers(NBDClient *client);
342
-static void nbd_update_can_read(NBDClient *client);
343
+static void nbd_client_receive_next_request(NBDClient *client);
344
345
static gboolean nbd_negotiate_continue(QIOChannel *ioc,
346
GIOCondition condition,
347
@@ -XXX,XX +XXX,XX @@ void nbd_client_put(NBDClient *client)
348
*/
349
assert(client->closing);
350
351
- nbd_unset_handlers(client);
352
+ qio_channel_detach_aio_context(client->ioc);
353
object_unref(OBJECT(client->sioc));
354
object_unref(OBJECT(client->ioc));
355
if (client->tlscreds) {
356
@@ -XXX,XX +XXX,XX @@ static NBDRequestData *nbd_request_get(NBDClient *client)
357
358
assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
359
client->nb_requests++;
360
- nbd_update_can_read(client);
361
362
req = g_new0(NBDRequestData, 1);
363
nbd_client_get(client);
364
@@ -XXX,XX +XXX,XX @@ static void nbd_request_put(NBDRequestData *req)
365
g_free(req);
366
367
client->nb_requests--;
368
- nbd_update_can_read(client);
369
+ nbd_client_receive_next_request(client);
370
+
371
nbd_client_put(client);
372
}
373
374
@@ -XXX,XX +XXX,XX @@ static void blk_aio_attached(AioContext *ctx, void *opaque)
375
exp->ctx = ctx;
376
377
QTAILQ_FOREACH(client, &exp->clients, next) {
378
- nbd_set_handlers(client);
379
+ qio_channel_attach_aio_context(client->ioc, ctx);
380
+ if (client->recv_coroutine) {
381
+ aio_co_schedule(ctx, client->recv_coroutine);
382
+ }
383
+ if (client->send_coroutine) {
384
+ aio_co_schedule(ctx, client->send_coroutine);
385
+ }
386
}
387
}
388
389
@@ -XXX,XX +XXX,XX @@ static void blk_aio_detach(void *opaque)
390
TRACE("Export %s: Detaching clients from AIO context %p\n", exp->name, exp->ctx);
391
392
QTAILQ_FOREACH(client, &exp->clients, next) {
393
- nbd_unset_handlers(client);
394
+ qio_channel_detach_aio_context(client->ioc);
395
}
396
397
exp->ctx = NULL;
398
@@ -XXX,XX +XXX,XX @@ static ssize_t nbd_co_send_reply(NBDRequestData *req, NBDReply *reply,
399
g_assert(qemu_in_coroutine());
400
qemu_co_mutex_lock(&client->send_lock);
401
client->send_coroutine = qemu_coroutine_self();
402
- nbd_set_handlers(client);
403
404
if (!len) {
405
rc = nbd_send_reply(client->ioc, reply);
406
@@ -XXX,XX +XXX,XX @@ static ssize_t nbd_co_send_reply(NBDRequestData *req, NBDReply *reply,
407
}
408
409
client->send_coroutine = NULL;
410
- nbd_set_handlers(client);
411
qemu_co_mutex_unlock(&client->send_lock);
412
return rc;
413
}
414
@@ -XXX,XX +XXX,XX @@ static ssize_t nbd_co_receive_request(NBDRequestData *req,
415
ssize_t rc;
416
417
g_assert(qemu_in_coroutine());
418
- client->recv_coroutine = qemu_coroutine_self();
419
- nbd_update_can_read(client);
420
-
421
+ assert(client->recv_coroutine == qemu_coroutine_self());
422
rc = nbd_receive_request(client->ioc, request);
423
if (rc < 0) {
424
if (rc != -EAGAIN) {
425
@@ -XXX,XX +XXX,XX @@ static ssize_t nbd_co_receive_request(NBDRequestData *req,
426
427
out:
428
client->recv_coroutine = NULL;
429
- nbd_update_can_read(client);
430
+ nbd_client_receive_next_request(client);
431
432
return rc;
433
}
434
435
-static void nbd_trip(void *opaque)
436
+/* Owns a reference to the NBDClient passed as opaque. */
437
+static coroutine_fn void nbd_trip(void *opaque)
438
{
439
NBDClient *client = opaque;
440
NBDExport *exp = client->exp;
441
NBDRequestData *req;
442
- NBDRequest request;
443
+ NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */
444
NBDReply reply;
445
ssize_t ret;
446
int flags;
447
448
TRACE("Reading request.");
449
if (client->closing) {
450
+ nbd_client_put(client);
451
return;
452
}
453
454
@@ -XXX,XX +XXX,XX @@ static void nbd_trip(void *opaque)
455
456
done:
457
nbd_request_put(req);
458
+ nbd_client_put(client);
459
return;
460
461
out:
462
nbd_request_put(req);
463
client_close(client);
464
+ nbd_client_put(client);
465
}
466
467
-static void nbd_read(void *opaque)
468
+static void nbd_client_receive_next_request(NBDClient *client)
469
{
470
- NBDClient *client = opaque;
471
-
472
- if (client->recv_coroutine) {
473
- qemu_coroutine_enter(client->recv_coroutine);
474
- } else {
475
- qemu_coroutine_enter(qemu_coroutine_create(nbd_trip, client));
476
- }
477
-}
478
-
479
-static void nbd_restart_write(void *opaque)
480
-{
481
- NBDClient *client = opaque;
482
-
483
- qemu_coroutine_enter(client->send_coroutine);
484
-}
485
-
486
-static void nbd_set_handlers(NBDClient *client)
487
-{
488
- if (client->exp && client->exp->ctx) {
489
- aio_set_fd_handler(client->exp->ctx, client->sioc->fd, true,
490
- client->can_read ? nbd_read : NULL,
491
- client->send_coroutine ? nbd_restart_write : NULL,
492
- NULL, client);
493
- }
494
-}
495
-
496
-static void nbd_unset_handlers(NBDClient *client)
497
-{
498
- if (client->exp && client->exp->ctx) {
499
- aio_set_fd_handler(client->exp->ctx, client->sioc->fd, true, NULL,
500
- NULL, NULL, NULL);
501
- }
502
-}
503
-
504
-static void nbd_update_can_read(NBDClient *client)
505
-{
506
- bool can_read = client->recv_coroutine ||
507
- client->nb_requests < MAX_NBD_REQUESTS;
508
-
509
- if (can_read != client->can_read) {
510
- client->can_read = can_read;
511
- nbd_set_handlers(client);
512
-
513
- /* There is no need to invoke aio_notify(), since aio_set_fd_handler()
514
- * in nbd_set_handlers() will have taken care of that */
515
+ if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) {
516
+ nbd_client_get(client);
517
+ client->recv_coroutine = qemu_coroutine_create(nbd_trip, client);
518
+ aio_co_schedule(client->exp->ctx, client->recv_coroutine);
519
}
520
}
521
522
@@ -XXX,XX +XXX,XX @@ static coroutine_fn void nbd_co_client_start(void *opaque)
523
goto out;
524
}
525
qemu_co_mutex_init(&client->send_lock);
526
- nbd_set_handlers(client);
527
528
if (exp) {
529
QTAILQ_INSERT_TAIL(&exp->clients, client, next);
530
}
531
+
532
+ nbd_client_receive_next_request(client);
533
+
534
out:
535
g_free(data);
536
}
537
@@ -XXX,XX +XXX,XX @@ void nbd_client_new(NBDExport *exp,
538
object_ref(OBJECT(client->sioc));
539
client->ioc = QIO_CHANNEL(sioc);
540
object_ref(OBJECT(client->ioc));
541
- client->can_read = true;
542
client->close = close_fn;
543
544
data->client = client;
545
--
546
2.9.3
547
548
diff view generated by jsdifflib
New patch
1
From: Paolo Bonzini <pbonzini@redhat.com>
1
2
3
As a small step towards the introduction of multiqueue, we want
4
coroutines to remain on the same AioContext that started them,
5
unless they are moved explicitly with e.g. aio_co_schedule. This patch
6
avoids that coroutines switch AioContext when they use a CoMutex.
7
For now it does not make much of a difference, because the CoMutex
8
is not thread-safe and the AioContext itself is used to protect the
9
CoMutex from concurrent access. However, this is going to change.
10
11
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
12
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
13
Reviewed-by: Fam Zheng <famz@redhat.com>
14
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
15
Message-id: 20170213135235.12274-9-pbonzini@redhat.com
16
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
17
---
18
util/qemu-coroutine-lock.c | 5 ++---
19
util/trace-events | 1 -
20
2 files changed, 2 insertions(+), 4 deletions(-)
21
22
diff --git a/util/qemu-coroutine-lock.c b/util/qemu-coroutine-lock.c
23
index XXXXXXX..XXXXXXX 100644
24
--- a/util/qemu-coroutine-lock.c
25
+++ b/util/qemu-coroutine-lock.c
26
@@ -XXX,XX +XXX,XX @@
27
#include "qemu/coroutine.h"
28
#include "qemu/coroutine_int.h"
29
#include "qemu/queue.h"
30
+#include "block/aio.h"
31
#include "trace.h"
32
33
void qemu_co_queue_init(CoQueue *queue)
34
@@ -XXX,XX +XXX,XX @@ void qemu_co_queue_run_restart(Coroutine *co)
35
36
static bool qemu_co_queue_do_restart(CoQueue *queue, bool single)
37
{
38
- Coroutine *self = qemu_coroutine_self();
39
Coroutine *next;
40
41
if (QSIMPLEQ_EMPTY(&queue->entries)) {
42
@@ -XXX,XX +XXX,XX @@ static bool qemu_co_queue_do_restart(CoQueue *queue, bool single)
43
44
while ((next = QSIMPLEQ_FIRST(&queue->entries)) != NULL) {
45
QSIMPLEQ_REMOVE_HEAD(&queue->entries, co_queue_next);
46
- QSIMPLEQ_INSERT_TAIL(&self->co_queue_wakeup, next, co_queue_next);
47
- trace_qemu_co_queue_next(next);
48
+ aio_co_wake(next);
49
if (single) {
50
break;
51
}
52
diff --git a/util/trace-events b/util/trace-events
53
index XXXXXXX..XXXXXXX 100644
54
--- a/util/trace-events
55
+++ b/util/trace-events
56
@@ -XXX,XX +XXX,XX @@ qemu_coroutine_terminate(void *co) "self %p"
57
58
# util/qemu-coroutine-lock.c
59
qemu_co_queue_run_restart(void *co) "co %p"
60
-qemu_co_queue_next(void *nxt) "next %p"
61
qemu_co_mutex_lock_entry(void *mutex, void *self) "mutex %p self %p"
62
qemu_co_mutex_lock_return(void *mutex, void *self) "mutex %p self %p"
63
qemu_co_mutex_unlock_entry(void *mutex, void *self) "mutex %p self %p"
64
--
65
2.9.3
66
67
diff view generated by jsdifflib
New patch
1
From: Paolo Bonzini <pbonzini@redhat.com>
1
2
3
Keep the coroutine on the same AioContext. Without this change,
4
there would be a race between yielding the coroutine and reentering it.
5
While the race cannot happen now, because the code only runs from a single
6
AioContext, this will change with multiqueue support in the block layer.
7
8
While doing the change, replace custom bottom half with aio_co_schedule.
9
10
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
11
Reviewed-by: Fam Zheng <famz@redhat.com>
12
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
13
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
14
Message-id: 20170213135235.12274-10-pbonzini@redhat.com
15
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
16
---
17
block/blkdebug.c | 9 +--------
18
1 file changed, 1 insertion(+), 8 deletions(-)
19
20
diff --git a/block/blkdebug.c b/block/blkdebug.c
21
index XXXXXXX..XXXXXXX 100644
22
--- a/block/blkdebug.c
23
+++ b/block/blkdebug.c
24
@@ -XXX,XX +XXX,XX @@ out:
25
return ret;
26
}
27
28
-static void error_callback_bh(void *opaque)
29
-{
30
- Coroutine *co = opaque;
31
- qemu_coroutine_enter(co);
32
-}
33
-
34
static int inject_error(BlockDriverState *bs, BlkdebugRule *rule)
35
{
36
BDRVBlkdebugState *s = bs->opaque;
37
@@ -XXX,XX +XXX,XX @@ static int inject_error(BlockDriverState *bs, BlkdebugRule *rule)
38
}
39
40
if (!immediately) {
41
- aio_bh_schedule_oneshot(bdrv_get_aio_context(bs), error_callback_bh,
42
- qemu_coroutine_self());
43
+ aio_co_schedule(qemu_get_current_aio_context(), qemu_coroutine_self());
44
qemu_coroutine_yield();
45
}
46
47
--
48
2.9.3
49
50
diff view generated by jsdifflib
1
From: Jeuk Kim <jeuk20.kim@samsung.com>
1
From: Paolo Bonzini <pbonzini@redhat.com>
2
2
3
This commit makes the UFS device support query
3
qed_aio_start_io and qed_aio_next_io will not have to acquire/release
4
and nop out transfer requests.
4
the AioContext, while qed_aio_next_io_cb will. Split the functionality
5
and gain a little type-safety in the process.
5
6
6
The next patch would be support for UFS logical
7
unit and scsi command transfer request.
8
9
Signed-off-by: Jeuk Kim <jeuk20.kim@samsung.com>
10
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
7
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
11
Message-id: ff7a5f0fd26761936a553ffb89d3df0ba62844e9.1693980783.git.jeuk20.kim@gmail.com
8
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9
Reviewed-by: Fam Zheng <famz@redhat.com>
10
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
11
Message-id: 20170213135235.12274-11-pbonzini@redhat.com
12
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
12
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
13
---
13
---
14
hw/ufs/ufs.h | 46 +++
14
block/qed.c | 39 +++++++++++++++++++++++++--------------
15
hw/ufs/ufs.c | 988 +++++++++++++++++++++++++++++++++++++++++++-
15
1 file changed, 25 insertions(+), 14 deletions(-)
16
hw/ufs/trace-events | 1 +
17
3 files changed, 1033 insertions(+), 2 deletions(-)
18
16
19
diff --git a/hw/ufs/ufs.h b/hw/ufs/ufs.h
17
diff --git a/block/qed.c b/block/qed.c
20
index XXXXXXX..XXXXXXX 100644
18
index XXXXXXX..XXXXXXX 100644
21
--- a/hw/ufs/ufs.h
19
--- a/block/qed.c
22
+++ b/hw/ufs/ufs.h
20
+++ b/block/qed.c
23
@@ -XXX,XX +XXX,XX @@
21
@@ -XXX,XX +XXX,XX @@ static CachedL2Table *qed_new_l2_table(BDRVQEDState *s)
24
#define UFS_MAX_LUS 32
22
return l2_table;
25
#define UFS_BLOCK_SIZE 4096
23
}
26
24
27
+typedef enum UfsRequestState {
25
-static void qed_aio_next_io(void *opaque, int ret);
28
+ UFS_REQUEST_IDLE = 0,
26
+static void qed_aio_next_io(QEDAIOCB *acb, int ret);
29
+ UFS_REQUEST_READY = 1,
30
+ UFS_REQUEST_RUNNING = 2,
31
+ UFS_REQUEST_COMPLETE = 3,
32
+ UFS_REQUEST_ERROR = 4,
33
+} UfsRequestState;
34
+
27
+
35
+typedef enum UfsReqResult {
28
+static void qed_aio_start_io(QEDAIOCB *acb)
36
+ UFS_REQUEST_SUCCESS = 0,
37
+ UFS_REQUEST_FAIL = 1,
38
+} UfsReqResult;
39
+
40
+typedef struct UfsRequest {
41
+ struct UfsHc *hc;
42
+ UfsRequestState state;
43
+ int slot;
44
+
45
+ UtpTransferReqDesc utrd;
46
+ UtpUpiuReq req_upiu;
47
+ UtpUpiuRsp rsp_upiu;
48
+
49
+ /* for scsi command */
50
+ QEMUSGList *sg;
51
+} UfsRequest;
52
+
53
typedef struct UfsParams {
54
char *serial;
55
uint8_t nutrs; /* Number of UTP Transfer Request Slots */
56
@@ -XXX,XX +XXX,XX @@ typedef struct UfsHc {
57
UfsReg reg;
58
UfsParams params;
59
uint32_t reg_size;
60
+ UfsRequest *req_list;
61
+
62
+ DeviceDescriptor device_desc;
63
+ GeometryDescriptor geometry_desc;
64
+ Attributes attributes;
65
+ Flags flags;
66
67
qemu_irq irq;
68
QEMUBH *doorbell_bh;
69
@@ -XXX,XX +XXX,XX @@ typedef struct UfsHc {
70
#define TYPE_UFS "ufs"
71
#define UFS(obj) OBJECT_CHECK(UfsHc, (obj), TYPE_UFS)
72
73
+typedef enum UfsQueryFlagPerm {
74
+ UFS_QUERY_FLAG_NONE = 0x0,
75
+ UFS_QUERY_FLAG_READ = 0x1,
76
+ UFS_QUERY_FLAG_SET = 0x2,
77
+ UFS_QUERY_FLAG_CLEAR = 0x4,
78
+ UFS_QUERY_FLAG_TOGGLE = 0x8,
79
+} UfsQueryFlagPerm;
80
+
81
+typedef enum UfsQueryAttrPerm {
82
+ UFS_QUERY_ATTR_NONE = 0x0,
83
+ UFS_QUERY_ATTR_READ = 0x1,
84
+ UFS_QUERY_ATTR_WRITE = 0x2,
85
+} UfsQueryAttrPerm;
86
+
87
#endif /* HW_UFS_UFS_H */
88
diff --git a/hw/ufs/ufs.c b/hw/ufs/ufs.c
89
index XXXXXXX..XXXXXXX 100644
90
--- a/hw/ufs/ufs.c
91
+++ b/hw/ufs/ufs.c
92
@@ -XXX,XX +XXX,XX @@
93
#include "ufs.h"
94
95
/* The QEMU-UFS device follows spec version 3.1 */
96
-#define UFS_SPEC_VER 0x00000310
97
+#define UFS_SPEC_VER 0x0310
98
#define UFS_MAX_NUTRS 32
99
#define UFS_MAX_NUTMRS 8
100
101
+static MemTxResult ufs_addr_read(UfsHc *u, hwaddr addr, void *buf, int size)
102
+{
29
+{
103
+ hwaddr hi = addr + size - 1;
30
+ qed_aio_next_io(acb, 0);
104
+
105
+ if (hi < addr) {
106
+ return MEMTX_DECODE_ERROR;
107
+ }
108
+
109
+ if (!FIELD_EX32(u->reg.cap, CAP, 64AS) && (hi >> 32)) {
110
+ return MEMTX_DECODE_ERROR;
111
+ }
112
+
113
+ return pci_dma_read(PCI_DEVICE(u), addr, buf, size);
114
+}
31
+}
115
+
32
+
116
+static MemTxResult ufs_addr_write(UfsHc *u, hwaddr addr, const void *buf,
33
+static void qed_aio_next_io_cb(void *opaque, int ret)
117
+ int size)
118
+{
34
+{
119
+ hwaddr hi = addr + size - 1;
35
+ QEDAIOCB *acb = opaque;
120
+ if (hi < addr) {
121
+ return MEMTX_DECODE_ERROR;
122
+ }
123
+
36
+
124
+ if (!FIELD_EX32(u->reg.cap, CAP, 64AS) && (hi >> 32)) {
37
+ qed_aio_next_io(acb, ret);
125
+ return MEMTX_DECODE_ERROR;
126
+ }
127
+
128
+ return pci_dma_write(PCI_DEVICE(u), addr, buf, size);
129
+}
38
+}
130
+
39
131
+static void ufs_complete_req(UfsRequest *req, UfsReqResult req_result);
40
static void qed_plug_allocating_write_reqs(BDRVQEDState *s)
132
+
133
+static inline hwaddr ufs_get_utrd_addr(UfsHc *u, uint32_t slot)
134
+{
135
+ hwaddr utrl_base_addr = (((hwaddr)u->reg.utrlbau) << 32) + u->reg.utrlba;
136
+ hwaddr utrd_addr = utrl_base_addr + slot * sizeof(UtpTransferReqDesc);
137
+
138
+ return utrd_addr;
139
+}
140
+
141
+static inline hwaddr ufs_get_req_upiu_base_addr(const UtpTransferReqDesc *utrd)
142
+{
143
+ uint32_t cmd_desc_base_addr_lo =
144
+ le32_to_cpu(utrd->command_desc_base_addr_lo);
145
+ uint32_t cmd_desc_base_addr_hi =
146
+ le32_to_cpu(utrd->command_desc_base_addr_hi);
147
+
148
+ return (((hwaddr)cmd_desc_base_addr_hi) << 32) + cmd_desc_base_addr_lo;
149
+}
150
+
151
+static inline hwaddr ufs_get_rsp_upiu_base_addr(const UtpTransferReqDesc *utrd)
152
+{
153
+ hwaddr req_upiu_base_addr = ufs_get_req_upiu_base_addr(utrd);
154
+ uint32_t rsp_upiu_byte_off =
155
+ le16_to_cpu(utrd->response_upiu_offset) * sizeof(uint32_t);
156
+ return req_upiu_base_addr + rsp_upiu_byte_off;
157
+}
158
+
159
+static MemTxResult ufs_dma_read_utrd(UfsRequest *req)
160
+{
161
+ UfsHc *u = req->hc;
162
+ hwaddr utrd_addr = ufs_get_utrd_addr(u, req->slot);
163
+ MemTxResult ret;
164
+
165
+ ret = ufs_addr_read(u, utrd_addr, &req->utrd, sizeof(req->utrd));
166
+ if (ret) {
167
+ trace_ufs_err_dma_read_utrd(req->slot, utrd_addr);
168
+ }
169
+ return ret;
170
+}
171
+
172
+static MemTxResult ufs_dma_read_req_upiu(UfsRequest *req)
173
+{
174
+ UfsHc *u = req->hc;
175
+ hwaddr req_upiu_base_addr = ufs_get_req_upiu_base_addr(&req->utrd);
176
+ UtpUpiuReq *req_upiu = &req->req_upiu;
177
+ uint32_t copy_size;
178
+ uint16_t data_segment_length;
179
+ MemTxResult ret;
180
+
181
+ /*
182
+ * To know the size of the req_upiu, we need to read the
183
+ * data_segment_length in the header first.
184
+ */
185
+ ret = ufs_addr_read(u, req_upiu_base_addr, &req_upiu->header,
186
+ sizeof(UtpUpiuHeader));
187
+ if (ret) {
188
+ trace_ufs_err_dma_read_req_upiu(req->slot, req_upiu_base_addr);
189
+ return ret;
190
+ }
191
+ data_segment_length = be16_to_cpu(req_upiu->header.data_segment_length);
192
+
193
+ copy_size = sizeof(UtpUpiuHeader) + UFS_TRANSACTION_SPECIFIC_FIELD_SIZE +
194
+ data_segment_length;
195
+
196
+ ret = ufs_addr_read(u, req_upiu_base_addr, &req->req_upiu, copy_size);
197
+ if (ret) {
198
+ trace_ufs_err_dma_read_req_upiu(req->slot, req_upiu_base_addr);
199
+ }
200
+ return ret;
201
+}
202
+
203
+static MemTxResult ufs_dma_read_prdt(UfsRequest *req)
204
+{
205
+ UfsHc *u = req->hc;
206
+ uint16_t prdt_len = le16_to_cpu(req->utrd.prd_table_length);
207
+ uint16_t prdt_byte_off =
208
+ le16_to_cpu(req->utrd.prd_table_offset) * sizeof(uint32_t);
209
+ uint32_t prdt_size = prdt_len * sizeof(UfshcdSgEntry);
210
+ g_autofree UfshcdSgEntry *prd_entries = NULL;
211
+ hwaddr req_upiu_base_addr, prdt_base_addr;
212
+ int err;
213
+
214
+ assert(!req->sg);
215
+
216
+ if (prdt_size == 0) {
217
+ return MEMTX_OK;
218
+ }
219
+ prd_entries = g_new(UfshcdSgEntry, prdt_size);
220
+
221
+ req_upiu_base_addr = ufs_get_req_upiu_base_addr(&req->utrd);
222
+ prdt_base_addr = req_upiu_base_addr + prdt_byte_off;
223
+
224
+ err = ufs_addr_read(u, prdt_base_addr, prd_entries, prdt_size);
225
+ if (err) {
226
+ trace_ufs_err_dma_read_prdt(req->slot, prdt_base_addr);
227
+ return err;
228
+ }
229
+
230
+ req->sg = g_malloc0(sizeof(QEMUSGList));
231
+ pci_dma_sglist_init(req->sg, PCI_DEVICE(u), prdt_len);
232
+
233
+ for (uint16_t i = 0; i < prdt_len; ++i) {
234
+ hwaddr data_dma_addr = le64_to_cpu(prd_entries[i].addr);
235
+ uint32_t data_byte_count = le32_to_cpu(prd_entries[i].size) + 1;
236
+ qemu_sglist_add(req->sg, data_dma_addr, data_byte_count);
237
+ }
238
+ return MEMTX_OK;
239
+}
240
+
241
+static MemTxResult ufs_dma_read_upiu(UfsRequest *req)
242
+{
243
+ MemTxResult ret;
244
+
245
+ ret = ufs_dma_read_utrd(req);
246
+ if (ret) {
247
+ return ret;
248
+ }
249
+
250
+ ret = ufs_dma_read_req_upiu(req);
251
+ if (ret) {
252
+ return ret;
253
+ }
254
+
255
+ ret = ufs_dma_read_prdt(req);
256
+ if (ret) {
257
+ return ret;
258
+ }
259
+
260
+ return 0;
261
+}
262
+
263
+static MemTxResult ufs_dma_write_utrd(UfsRequest *req)
264
+{
265
+ UfsHc *u = req->hc;
266
+ hwaddr utrd_addr = ufs_get_utrd_addr(u, req->slot);
267
+ MemTxResult ret;
268
+
269
+ ret = ufs_addr_write(u, utrd_addr, &req->utrd, sizeof(req->utrd));
270
+ if (ret) {
271
+ trace_ufs_err_dma_write_utrd(req->slot, utrd_addr);
272
+ }
273
+ return ret;
274
+}
275
+
276
+static MemTxResult ufs_dma_write_rsp_upiu(UfsRequest *req)
277
+{
278
+ UfsHc *u = req->hc;
279
+ hwaddr rsp_upiu_base_addr = ufs_get_rsp_upiu_base_addr(&req->utrd);
280
+ uint32_t rsp_upiu_byte_len =
281
+ le16_to_cpu(req->utrd.response_upiu_length) * sizeof(uint32_t);
282
+ uint16_t data_segment_length =
283
+ be16_to_cpu(req->rsp_upiu.header.data_segment_length);
284
+ uint32_t copy_size = sizeof(UtpUpiuHeader) +
285
+ UFS_TRANSACTION_SPECIFIC_FIELD_SIZE +
286
+ data_segment_length;
287
+ MemTxResult ret;
288
+
289
+ if (copy_size > rsp_upiu_byte_len) {
290
+ copy_size = rsp_upiu_byte_len;
291
+ }
292
+
293
+ ret = ufs_addr_write(u, rsp_upiu_base_addr, &req->rsp_upiu, copy_size);
294
+ if (ret) {
295
+ trace_ufs_err_dma_write_rsp_upiu(req->slot, rsp_upiu_base_addr);
296
+ }
297
+ return ret;
298
+}
299
+
300
+static MemTxResult ufs_dma_write_upiu(UfsRequest *req)
301
+{
302
+ MemTxResult ret;
303
+
304
+ ret = ufs_dma_write_rsp_upiu(req);
305
+ if (ret) {
306
+ return ret;
307
+ }
308
+
309
+ return ufs_dma_write_utrd(req);
310
+}
311
+
312
static void ufs_irq_check(UfsHc *u)
313
{
41
{
314
PCIDevice *pci = PCI_DEVICE(u);
42
@@ -XXX,XX +XXX,XX @@ static void qed_unplug_allocating_write_reqs(BDRVQEDState *s)
315
@@ -XXX,XX +XXX,XX @@ static void ufs_irq_check(UfsHc *u)
43
44
acb = QSIMPLEQ_FIRST(&s->allocating_write_reqs);
45
if (acb) {
46
- qed_aio_next_io(acb, 0);
47
+ qed_aio_start_io(acb);
316
}
48
}
317
}
49
}
318
50
319
+static void ufs_process_db(UfsHc *u, uint32_t val)
51
@@ -XXX,XX +XXX,XX @@ static void qed_aio_complete(QEDAIOCB *acb, int ret)
320
+{
52
QSIMPLEQ_REMOVE_HEAD(&s->allocating_write_reqs, next);
321
+ unsigned long doorbell;
53
acb = QSIMPLEQ_FIRST(&s->allocating_write_reqs);
322
+ uint32_t slot;
54
if (acb) {
323
+ uint32_t nutrs = u->params.nutrs;
55
- qed_aio_next_io(acb, 0);
324
+ UfsRequest *req;
56
+ qed_aio_start_io(acb);
325
+
57
} else if (s->header.features & QED_F_NEED_CHECK) {
326
+ val &= ~u->reg.utrldbr;
58
qed_start_need_check_timer(s);
327
+ if (!val) {
59
}
328
+ return;
60
@@ -XXX,XX +XXX,XX @@ static void qed_commit_l2_update(void *opaque, int ret)
329
+ }
61
acb->request.l2_table = qed_find_l2_cache_entry(&s->l2_cache, l2_offset);
330
+
62
assert(acb->request.l2_table != NULL);
331
+ doorbell = val;
63
332
+ slot = find_first_bit(&doorbell, nutrs);
64
- qed_aio_next_io(opaque, ret);
333
+
65
+ qed_aio_next_io(acb, ret);
334
+ while (slot < nutrs) {
335
+ req = &u->req_list[slot];
336
+ if (req->state == UFS_REQUEST_ERROR) {
337
+ trace_ufs_err_utrl_slot_error(req->slot);
338
+ return;
339
+ }
340
+
341
+ if (req->state != UFS_REQUEST_IDLE) {
342
+ trace_ufs_err_utrl_slot_busy(req->slot);
343
+ return;
344
+ }
345
+
346
+ trace_ufs_process_db(slot);
347
+ req->state = UFS_REQUEST_READY;
348
+ slot = find_next_bit(&doorbell, nutrs, slot + 1);
349
+ }
350
+
351
+ qemu_bh_schedule(u->doorbell_bh);
352
+}
353
+
354
static void ufs_process_uiccmd(UfsHc *u, uint32_t val)
355
{
356
trace_ufs_process_uiccmd(val, u->reg.ucmdarg1, u->reg.ucmdarg2,
357
@@ -XXX,XX +XXX,XX @@ static void ufs_write_reg(UfsHc *u, hwaddr offset, uint32_t data, unsigned size)
358
u->reg.utrlbau = data;
359
break;
360
case A_UTRLDBR:
361
- /* Not yet supported */
362
+ ufs_process_db(u, data);
363
+ u->reg.utrldbr |= data;
364
break;
365
case A_UTRLRSR:
366
u->reg.utrlrsr = data;
367
@@ -XXX,XX +XXX,XX @@ static const MemoryRegionOps ufs_mmio_ops = {
368
},
369
};
370
371
+static void ufs_build_upiu_header(UfsRequest *req, uint8_t trans_type,
372
+ uint8_t flags, uint8_t response,
373
+ uint8_t scsi_status,
374
+ uint16_t data_segment_length)
375
+{
376
+ memcpy(&req->rsp_upiu.header, &req->req_upiu.header, sizeof(UtpUpiuHeader));
377
+ req->rsp_upiu.header.trans_type = trans_type;
378
+ req->rsp_upiu.header.flags = flags;
379
+ req->rsp_upiu.header.response = response;
380
+ req->rsp_upiu.header.scsi_status = scsi_status;
381
+ req->rsp_upiu.header.data_segment_length = cpu_to_be16(data_segment_length);
382
+}
383
+
384
+static UfsReqResult ufs_exec_nop_cmd(UfsRequest *req)
385
+{
386
+ trace_ufs_exec_nop_cmd(req->slot);
387
+ ufs_build_upiu_header(req, UFS_UPIU_TRANSACTION_NOP_IN, 0, 0, 0, 0);
388
+ return UFS_REQUEST_SUCCESS;
389
+}
390
+
391
+/*
392
+ * This defines the permission of flags based on their IDN. There are some
393
+ * things that are declared read-only, which is inconsistent with the ufs spec,
394
+ * because we want to return an error for features that are not yet supported.
395
+ */
396
+static const int flag_permission[UFS_QUERY_FLAG_IDN_COUNT] = {
397
+ [UFS_QUERY_FLAG_IDN_FDEVICEINIT] = UFS_QUERY_FLAG_READ | UFS_QUERY_FLAG_SET,
398
+ /* Write protection is not supported */
399
+ [UFS_QUERY_FLAG_IDN_PERMANENT_WPE] = UFS_QUERY_FLAG_READ,
400
+ [UFS_QUERY_FLAG_IDN_PWR_ON_WPE] = UFS_QUERY_FLAG_READ,
401
+ [UFS_QUERY_FLAG_IDN_BKOPS_EN] = UFS_QUERY_FLAG_READ | UFS_QUERY_FLAG_SET |
402
+ UFS_QUERY_FLAG_CLEAR |
403
+ UFS_QUERY_FLAG_TOGGLE,
404
+ [UFS_QUERY_FLAG_IDN_LIFE_SPAN_MODE_ENABLE] =
405
+ UFS_QUERY_FLAG_READ | UFS_QUERY_FLAG_SET | UFS_QUERY_FLAG_CLEAR |
406
+ UFS_QUERY_FLAG_TOGGLE,
407
+ /* Purge Operation is not supported */
408
+ [UFS_QUERY_FLAG_IDN_PURGE_ENABLE] = UFS_QUERY_FLAG_NONE,
409
+ /* Refresh Operation is not supported */
410
+ [UFS_QUERY_FLAG_IDN_REFRESH_ENABLE] = UFS_QUERY_FLAG_NONE,
411
+ /* Physical Resource Removal is not supported */
412
+ [UFS_QUERY_FLAG_IDN_FPHYRESOURCEREMOVAL] = UFS_QUERY_FLAG_READ,
413
+ [UFS_QUERY_FLAG_IDN_BUSY_RTC] = UFS_QUERY_FLAG_READ,
414
+ [UFS_QUERY_FLAG_IDN_PERMANENTLY_DISABLE_FW_UPDATE] = UFS_QUERY_FLAG_READ,
415
+ /* Write Booster is not supported */
416
+ [UFS_QUERY_FLAG_IDN_WB_EN] = UFS_QUERY_FLAG_READ,
417
+ [UFS_QUERY_FLAG_IDN_WB_BUFF_FLUSH_EN] = UFS_QUERY_FLAG_READ,
418
+ [UFS_QUERY_FLAG_IDN_WB_BUFF_FLUSH_DURING_HIBERN8] = UFS_QUERY_FLAG_READ,
419
+};
420
+
421
+static inline QueryRespCode ufs_flag_check_idn_valid(uint8_t idn, int op)
422
+{
423
+ if (idn >= UFS_QUERY_FLAG_IDN_COUNT) {
424
+ return UFS_QUERY_RESULT_INVALID_IDN;
425
+ }
426
+
427
+ if (!(flag_permission[idn] & op)) {
428
+ if (op == UFS_QUERY_FLAG_READ) {
429
+ trace_ufs_err_query_flag_not_readable(idn);
430
+ return UFS_QUERY_RESULT_NOT_READABLE;
431
+ }
432
+ trace_ufs_err_query_flag_not_writable(idn);
433
+ return UFS_QUERY_RESULT_NOT_WRITEABLE;
434
+ }
435
+
436
+ return UFS_QUERY_RESULT_SUCCESS;
437
+}
438
+
439
+static const int attr_permission[UFS_QUERY_ATTR_IDN_COUNT] = {
440
+ /* booting is not supported */
441
+ [UFS_QUERY_ATTR_IDN_BOOT_LU_EN] = UFS_QUERY_ATTR_READ,
442
+ [UFS_QUERY_ATTR_IDN_POWER_MODE] = UFS_QUERY_ATTR_READ,
443
+ [UFS_QUERY_ATTR_IDN_ACTIVE_ICC_LVL] =
444
+ UFS_QUERY_ATTR_READ | UFS_QUERY_ATTR_WRITE,
445
+ [UFS_QUERY_ATTR_IDN_OOO_DATA_EN] = UFS_QUERY_ATTR_READ,
446
+ [UFS_QUERY_ATTR_IDN_BKOPS_STATUS] = UFS_QUERY_ATTR_READ,
447
+ [UFS_QUERY_ATTR_IDN_PURGE_STATUS] = UFS_QUERY_ATTR_READ,
448
+ [UFS_QUERY_ATTR_IDN_MAX_DATA_IN] =
449
+ UFS_QUERY_ATTR_READ | UFS_QUERY_ATTR_WRITE,
450
+ [UFS_QUERY_ATTR_IDN_MAX_DATA_OUT] =
451
+ UFS_QUERY_ATTR_READ | UFS_QUERY_ATTR_WRITE,
452
+ [UFS_QUERY_ATTR_IDN_DYN_CAP_NEEDED] = UFS_QUERY_ATTR_READ,
453
+ [UFS_QUERY_ATTR_IDN_REF_CLK_FREQ] =
454
+ UFS_QUERY_ATTR_READ | UFS_QUERY_ATTR_WRITE,
455
+ [UFS_QUERY_ATTR_IDN_CONF_DESC_LOCK] = UFS_QUERY_ATTR_READ,
456
+ [UFS_QUERY_ATTR_IDN_MAX_NUM_OF_RTT] =
457
+ UFS_QUERY_ATTR_READ | UFS_QUERY_ATTR_WRITE,
458
+ [UFS_QUERY_ATTR_IDN_EE_CONTROL] =
459
+ UFS_QUERY_ATTR_READ | UFS_QUERY_ATTR_WRITE,
460
+ [UFS_QUERY_ATTR_IDN_EE_STATUS] = UFS_QUERY_ATTR_READ,
461
+ [UFS_QUERY_ATTR_IDN_SECONDS_PASSED] = UFS_QUERY_ATTR_WRITE,
462
+ [UFS_QUERY_ATTR_IDN_CNTX_CONF] = UFS_QUERY_ATTR_READ,
463
+ [UFS_QUERY_ATTR_IDN_FFU_STATUS] = UFS_QUERY_ATTR_READ,
464
+ [UFS_QUERY_ATTR_IDN_PSA_STATE] = UFS_QUERY_ATTR_READ | UFS_QUERY_ATTR_WRITE,
465
+ [UFS_QUERY_ATTR_IDN_PSA_DATA_SIZE] =
466
+ UFS_QUERY_ATTR_READ | UFS_QUERY_ATTR_WRITE,
467
+ [UFS_QUERY_ATTR_IDN_REF_CLK_GATING_WAIT_TIME] = UFS_QUERY_ATTR_READ,
468
+ [UFS_QUERY_ATTR_IDN_CASE_ROUGH_TEMP] = UFS_QUERY_ATTR_READ,
469
+ [UFS_QUERY_ATTR_IDN_HIGH_TEMP_BOUND] = UFS_QUERY_ATTR_READ,
470
+ [UFS_QUERY_ATTR_IDN_LOW_TEMP_BOUND] = UFS_QUERY_ATTR_READ,
471
+ [UFS_QUERY_ATTR_IDN_THROTTLING_STATUS] = UFS_QUERY_ATTR_READ,
472
+ [UFS_QUERY_ATTR_IDN_WB_FLUSH_STATUS] = UFS_QUERY_ATTR_READ,
473
+ [UFS_QUERY_ATTR_IDN_AVAIL_WB_BUFF_SIZE] = UFS_QUERY_ATTR_READ,
474
+ [UFS_QUERY_ATTR_IDN_WB_BUFF_LIFE_TIME_EST] = UFS_QUERY_ATTR_READ,
475
+ [UFS_QUERY_ATTR_IDN_CURR_WB_BUFF_SIZE] = UFS_QUERY_ATTR_READ,
476
+ /* refresh operation is not supported */
477
+ [UFS_QUERY_ATTR_IDN_REFRESH_STATUS] = UFS_QUERY_ATTR_READ,
478
+ [UFS_QUERY_ATTR_IDN_REFRESH_FREQ] = UFS_QUERY_ATTR_READ,
479
+ [UFS_QUERY_ATTR_IDN_REFRESH_UNIT] = UFS_QUERY_ATTR_READ,
480
+};
481
+
482
+static inline QueryRespCode ufs_attr_check_idn_valid(uint8_t idn, int op)
483
+{
484
+ if (idn >= UFS_QUERY_ATTR_IDN_COUNT) {
485
+ return UFS_QUERY_RESULT_INVALID_IDN;
486
+ }
487
+
488
+ if (!(attr_permission[idn] & op)) {
489
+ if (op == UFS_QUERY_ATTR_READ) {
490
+ trace_ufs_err_query_attr_not_readable(idn);
491
+ return UFS_QUERY_RESULT_NOT_READABLE;
492
+ }
493
+ trace_ufs_err_query_attr_not_writable(idn);
494
+ return UFS_QUERY_RESULT_NOT_WRITEABLE;
495
+ }
496
+
497
+ return UFS_QUERY_RESULT_SUCCESS;
498
+}
499
+
500
+static QueryRespCode ufs_exec_query_flag(UfsRequest *req, int op)
501
+{
502
+ UfsHc *u = req->hc;
503
+ uint8_t idn = req->req_upiu.qr.idn;
504
+ uint32_t value;
505
+ QueryRespCode ret;
506
+
507
+ ret = ufs_flag_check_idn_valid(idn, op);
508
+ if (ret) {
509
+ return ret;
510
+ }
511
+
512
+ if (idn == UFS_QUERY_FLAG_IDN_FDEVICEINIT) {
513
+ value = 0;
514
+ } else if (op == UFS_QUERY_FLAG_READ) {
515
+ value = *(((uint8_t *)&u->flags) + idn);
516
+ } else if (op == UFS_QUERY_FLAG_SET) {
517
+ value = 1;
518
+ } else if (op == UFS_QUERY_FLAG_CLEAR) {
519
+ value = 0;
520
+ } else if (op == UFS_QUERY_FLAG_TOGGLE) {
521
+ value = *(((uint8_t *)&u->flags) + idn);
522
+ value = !value;
523
+ } else {
524
+ trace_ufs_err_query_invalid_opcode(op);
525
+ return UFS_QUERY_RESULT_INVALID_OPCODE;
526
+ }
527
+
528
+ *(((uint8_t *)&u->flags) + idn) = value;
529
+ req->rsp_upiu.qr.value = cpu_to_be32(value);
530
+ return UFS_QUERY_RESULT_SUCCESS;
531
+}
532
+
533
+static uint32_t ufs_read_attr_value(UfsHc *u, uint8_t idn)
534
+{
535
+ switch (idn) {
536
+ case UFS_QUERY_ATTR_IDN_BOOT_LU_EN:
537
+ return u->attributes.boot_lun_en;
538
+ case UFS_QUERY_ATTR_IDN_POWER_MODE:
539
+ return u->attributes.current_power_mode;
540
+ case UFS_QUERY_ATTR_IDN_ACTIVE_ICC_LVL:
541
+ return u->attributes.active_icc_level;
542
+ case UFS_QUERY_ATTR_IDN_OOO_DATA_EN:
543
+ return u->attributes.out_of_order_data_en;
544
+ case UFS_QUERY_ATTR_IDN_BKOPS_STATUS:
545
+ return u->attributes.background_op_status;
546
+ case UFS_QUERY_ATTR_IDN_PURGE_STATUS:
547
+ return u->attributes.purge_status;
548
+ case UFS_QUERY_ATTR_IDN_MAX_DATA_IN:
549
+ return u->attributes.max_data_in_size;
550
+ case UFS_QUERY_ATTR_IDN_MAX_DATA_OUT:
551
+ return u->attributes.max_data_out_size;
552
+ case UFS_QUERY_ATTR_IDN_DYN_CAP_NEEDED:
553
+ return be32_to_cpu(u->attributes.dyn_cap_needed);
554
+ case UFS_QUERY_ATTR_IDN_REF_CLK_FREQ:
555
+ return u->attributes.ref_clk_freq;
556
+ case UFS_QUERY_ATTR_IDN_CONF_DESC_LOCK:
557
+ return u->attributes.config_descr_lock;
558
+ case UFS_QUERY_ATTR_IDN_MAX_NUM_OF_RTT:
559
+ return u->attributes.max_num_of_rtt;
560
+ case UFS_QUERY_ATTR_IDN_EE_CONTROL:
561
+ return be16_to_cpu(u->attributes.exception_event_control);
562
+ case UFS_QUERY_ATTR_IDN_EE_STATUS:
563
+ return be16_to_cpu(u->attributes.exception_event_status);
564
+ case UFS_QUERY_ATTR_IDN_SECONDS_PASSED:
565
+ return be32_to_cpu(u->attributes.seconds_passed);
566
+ case UFS_QUERY_ATTR_IDN_CNTX_CONF:
567
+ return be16_to_cpu(u->attributes.context_conf);
568
+ case UFS_QUERY_ATTR_IDN_FFU_STATUS:
569
+ return u->attributes.device_ffu_status;
570
+ case UFS_QUERY_ATTR_IDN_PSA_STATE:
571
+ return be32_to_cpu(u->attributes.psa_state);
572
+ case UFS_QUERY_ATTR_IDN_PSA_DATA_SIZE:
573
+ return be32_to_cpu(u->attributes.psa_data_size);
574
+ case UFS_QUERY_ATTR_IDN_REF_CLK_GATING_WAIT_TIME:
575
+ return u->attributes.ref_clk_gating_wait_time;
576
+ case UFS_QUERY_ATTR_IDN_CASE_ROUGH_TEMP:
577
+ return u->attributes.device_case_rough_temperaure;
578
+ case UFS_QUERY_ATTR_IDN_HIGH_TEMP_BOUND:
579
+ return u->attributes.device_too_high_temp_boundary;
580
+ case UFS_QUERY_ATTR_IDN_LOW_TEMP_BOUND:
581
+ return u->attributes.device_too_low_temp_boundary;
582
+ case UFS_QUERY_ATTR_IDN_THROTTLING_STATUS:
583
+ return u->attributes.throttling_status;
584
+ case UFS_QUERY_ATTR_IDN_WB_FLUSH_STATUS:
585
+ return u->attributes.wb_buffer_flush_status;
586
+ case UFS_QUERY_ATTR_IDN_AVAIL_WB_BUFF_SIZE:
587
+ return u->attributes.available_wb_buffer_size;
588
+ case UFS_QUERY_ATTR_IDN_WB_BUFF_LIFE_TIME_EST:
589
+ return u->attributes.wb_buffer_life_time_est;
590
+ case UFS_QUERY_ATTR_IDN_CURR_WB_BUFF_SIZE:
591
+ return be32_to_cpu(u->attributes.current_wb_buffer_size);
592
+ case UFS_QUERY_ATTR_IDN_REFRESH_STATUS:
593
+ return u->attributes.refresh_status;
594
+ case UFS_QUERY_ATTR_IDN_REFRESH_FREQ:
595
+ return u->attributes.refresh_freq;
596
+ case UFS_QUERY_ATTR_IDN_REFRESH_UNIT:
597
+ return u->attributes.refresh_unit;
598
+ }
599
+ return 0;
600
+}
601
+
602
+static void ufs_write_attr_value(UfsHc *u, uint8_t idn, uint32_t value)
603
+{
604
+ switch (idn) {
605
+ case UFS_QUERY_ATTR_IDN_ACTIVE_ICC_LVL:
606
+ u->attributes.active_icc_level = value;
607
+ break;
608
+ case UFS_QUERY_ATTR_IDN_MAX_DATA_IN:
609
+ u->attributes.max_data_in_size = value;
610
+ break;
611
+ case UFS_QUERY_ATTR_IDN_MAX_DATA_OUT:
612
+ u->attributes.max_data_out_size = value;
613
+ break;
614
+ case UFS_QUERY_ATTR_IDN_REF_CLK_FREQ:
615
+ u->attributes.ref_clk_freq = value;
616
+ break;
617
+ case UFS_QUERY_ATTR_IDN_MAX_NUM_OF_RTT:
618
+ u->attributes.max_num_of_rtt = value;
619
+ break;
620
+ case UFS_QUERY_ATTR_IDN_EE_CONTROL:
621
+ u->attributes.exception_event_control = cpu_to_be16(value);
622
+ break;
623
+ case UFS_QUERY_ATTR_IDN_SECONDS_PASSED:
624
+ u->attributes.seconds_passed = cpu_to_be32(value);
625
+ break;
626
+ case UFS_QUERY_ATTR_IDN_PSA_STATE:
627
+ u->attributes.psa_state = value;
628
+ break;
629
+ case UFS_QUERY_ATTR_IDN_PSA_DATA_SIZE:
630
+ u->attributes.psa_data_size = cpu_to_be32(value);
631
+ break;
632
+ }
633
+}
634
+
635
+static QueryRespCode ufs_exec_query_attr(UfsRequest *req, int op)
636
+{
637
+ UfsHc *u = req->hc;
638
+ uint8_t idn = req->req_upiu.qr.idn;
639
+ uint32_t value;
640
+ QueryRespCode ret;
641
+
642
+ ret = ufs_attr_check_idn_valid(idn, op);
643
+ if (ret) {
644
+ return ret;
645
+ }
646
+
647
+ if (op == UFS_QUERY_ATTR_READ) {
648
+ value = ufs_read_attr_value(u, idn);
649
+ } else {
650
+ value = be32_to_cpu(req->req_upiu.qr.value);
651
+ ufs_write_attr_value(u, idn, value);
652
+ }
653
+
654
+ req->rsp_upiu.qr.value = cpu_to_be32(value);
655
+ return UFS_QUERY_RESULT_SUCCESS;
656
+}
657
+
658
+static const RpmbUnitDescriptor rpmb_unit_desc = {
659
+ .length = sizeof(RpmbUnitDescriptor),
660
+ .descriptor_idn = 2,
661
+ .unit_index = UFS_UPIU_RPMB_WLUN,
662
+ .lu_enable = 0,
663
+};
664
+
665
+static QueryRespCode ufs_read_unit_desc(UfsRequest *req)
666
+{
667
+ uint8_t lun = req->req_upiu.qr.index;
668
+
669
+ if (lun != UFS_UPIU_RPMB_WLUN && lun > UFS_MAX_LUS) {
670
+ trace_ufs_err_query_invalid_index(req->req_upiu.qr.opcode, lun);
671
+ return UFS_QUERY_RESULT_INVALID_INDEX;
672
+ }
673
+
674
+ if (lun == UFS_UPIU_RPMB_WLUN) {
675
+ memcpy(&req->rsp_upiu.qr.data, &rpmb_unit_desc, rpmb_unit_desc.length);
676
+ } else {
677
+ /* unit descriptor is not yet supported */
678
+ return UFS_QUERY_RESULT_INVALID_INDEX;
679
+ }
680
+
681
+ return UFS_QUERY_RESULT_SUCCESS;
682
+}
683
+
684
+static inline StringDescriptor manufacturer_str_desc(void)
685
+{
686
+ StringDescriptor desc = {
687
+ .length = 0x12,
688
+ .descriptor_idn = UFS_QUERY_DESC_IDN_STRING,
689
+ };
690
+ desc.UC[0] = cpu_to_be16('R');
691
+ desc.UC[1] = cpu_to_be16('E');
692
+ desc.UC[2] = cpu_to_be16('D');
693
+ desc.UC[3] = cpu_to_be16('H');
694
+ desc.UC[4] = cpu_to_be16('A');
695
+ desc.UC[5] = cpu_to_be16('T');
696
+ return desc;
697
+}
698
+
699
+static inline StringDescriptor product_name_str_desc(void)
700
+{
701
+ StringDescriptor desc = {
702
+ .length = 0x22,
703
+ .descriptor_idn = UFS_QUERY_DESC_IDN_STRING,
704
+ };
705
+ desc.UC[0] = cpu_to_be16('Q');
706
+ desc.UC[1] = cpu_to_be16('E');
707
+ desc.UC[2] = cpu_to_be16('M');
708
+ desc.UC[3] = cpu_to_be16('U');
709
+ desc.UC[4] = cpu_to_be16(' ');
710
+ desc.UC[5] = cpu_to_be16('U');
711
+ desc.UC[6] = cpu_to_be16('F');
712
+ desc.UC[7] = cpu_to_be16('S');
713
+ return desc;
714
+}
715
+
716
+static inline StringDescriptor product_rev_level_str_desc(void)
717
+{
718
+ StringDescriptor desc = {
719
+ .length = 0x0a,
720
+ .descriptor_idn = UFS_QUERY_DESC_IDN_STRING,
721
+ };
722
+ desc.UC[0] = cpu_to_be16('0');
723
+ desc.UC[1] = cpu_to_be16('0');
724
+ desc.UC[2] = cpu_to_be16('0');
725
+ desc.UC[3] = cpu_to_be16('1');
726
+ return desc;
727
+}
728
+
729
+static const StringDescriptor null_str_desc = {
730
+ .length = 0x02,
731
+ .descriptor_idn = UFS_QUERY_DESC_IDN_STRING,
732
+};
733
+
734
+static QueryRespCode ufs_read_string_desc(UfsRequest *req)
735
+{
736
+ UfsHc *u = req->hc;
737
+ uint8_t index = req->req_upiu.qr.index;
738
+ StringDescriptor desc;
739
+
740
+ if (index == u->device_desc.manufacturer_name) {
741
+ desc = manufacturer_str_desc();
742
+ memcpy(&req->rsp_upiu.qr.data, &desc, desc.length);
743
+ } else if (index == u->device_desc.product_name) {
744
+ desc = product_name_str_desc();
745
+ memcpy(&req->rsp_upiu.qr.data, &desc, desc.length);
746
+ } else if (index == u->device_desc.serial_number) {
747
+ memcpy(&req->rsp_upiu.qr.data, &null_str_desc, null_str_desc.length);
748
+ } else if (index == u->device_desc.oem_id) {
749
+ memcpy(&req->rsp_upiu.qr.data, &null_str_desc, null_str_desc.length);
750
+ } else if (index == u->device_desc.product_revision_level) {
751
+ desc = product_rev_level_str_desc();
752
+ memcpy(&req->rsp_upiu.qr.data, &desc, desc.length);
753
+ } else {
754
+ trace_ufs_err_query_invalid_index(req->req_upiu.qr.opcode, index);
755
+ return UFS_QUERY_RESULT_INVALID_INDEX;
756
+ }
757
+ return UFS_QUERY_RESULT_SUCCESS;
758
+}
759
+
760
+static inline InterconnectDescriptor interconnect_desc(void)
761
+{
762
+ InterconnectDescriptor desc = {
763
+ .length = sizeof(InterconnectDescriptor),
764
+ .descriptor_idn = UFS_QUERY_DESC_IDN_INTERCONNECT,
765
+ };
766
+ desc.bcd_unipro_version = cpu_to_be16(0x180);
767
+ desc.bcd_mphy_version = cpu_to_be16(0x410);
768
+ return desc;
769
+}
770
+
771
+static QueryRespCode ufs_read_desc(UfsRequest *req)
772
+{
773
+ UfsHc *u = req->hc;
774
+ QueryRespCode status;
775
+ uint8_t idn = req->req_upiu.qr.idn;
776
+ uint16_t length = be16_to_cpu(req->req_upiu.qr.length);
777
+ InterconnectDescriptor desc;
778
+
779
+ switch (idn) {
780
+ case UFS_QUERY_DESC_IDN_DEVICE:
781
+ memcpy(&req->rsp_upiu.qr.data, &u->device_desc, sizeof(u->device_desc));
782
+ status = UFS_QUERY_RESULT_SUCCESS;
783
+ break;
784
+ case UFS_QUERY_DESC_IDN_UNIT:
785
+ status = ufs_read_unit_desc(req);
786
+ break;
787
+ case UFS_QUERY_DESC_IDN_GEOMETRY:
788
+ memcpy(&req->rsp_upiu.qr.data, &u->geometry_desc,
789
+ sizeof(u->geometry_desc));
790
+ status = UFS_QUERY_RESULT_SUCCESS;
791
+ break;
792
+ case UFS_QUERY_DESC_IDN_INTERCONNECT: {
793
+ desc = interconnect_desc();
794
+ memcpy(&req->rsp_upiu.qr.data, &desc, sizeof(InterconnectDescriptor));
795
+ status = UFS_QUERY_RESULT_SUCCESS;
796
+ break;
797
+ }
798
+ case UFS_QUERY_DESC_IDN_STRING:
799
+ status = ufs_read_string_desc(req);
800
+ break;
801
+ case UFS_QUERY_DESC_IDN_POWER:
802
+ /* mocking of power descriptor is not supported */
803
+ memset(&req->rsp_upiu.qr.data, 0, sizeof(PowerParametersDescriptor));
804
+ req->rsp_upiu.qr.data[0] = sizeof(PowerParametersDescriptor);
805
+ req->rsp_upiu.qr.data[1] = UFS_QUERY_DESC_IDN_POWER;
806
+ status = UFS_QUERY_RESULT_SUCCESS;
807
+ break;
808
+ case UFS_QUERY_DESC_IDN_HEALTH:
809
+ /* mocking of health descriptor is not supported */
810
+ memset(&req->rsp_upiu.qr.data, 0, sizeof(DeviceHealthDescriptor));
811
+ req->rsp_upiu.qr.data[0] = sizeof(DeviceHealthDescriptor);
812
+ req->rsp_upiu.qr.data[1] = UFS_QUERY_DESC_IDN_HEALTH;
813
+ status = UFS_QUERY_RESULT_SUCCESS;
814
+ break;
815
+ default:
816
+ length = 0;
817
+ trace_ufs_err_query_invalid_idn(req->req_upiu.qr.opcode, idn);
818
+ status = UFS_QUERY_RESULT_INVALID_IDN;
819
+ }
820
+
821
+ if (length > req->rsp_upiu.qr.data[0]) {
822
+ length = req->rsp_upiu.qr.data[0];
823
+ }
824
+ req->rsp_upiu.qr.opcode = req->req_upiu.qr.opcode;
825
+ req->rsp_upiu.qr.idn = req->req_upiu.qr.idn;
826
+ req->rsp_upiu.qr.index = req->req_upiu.qr.index;
827
+ req->rsp_upiu.qr.selector = req->req_upiu.qr.selector;
828
+ req->rsp_upiu.qr.length = cpu_to_be16(length);
829
+
830
+ return status;
831
+}
832
+
833
+static QueryRespCode ufs_exec_query_read(UfsRequest *req)
834
+{
835
+ QueryRespCode status;
836
+ switch (req->req_upiu.qr.opcode) {
837
+ case UFS_UPIU_QUERY_OPCODE_NOP:
838
+ status = UFS_QUERY_RESULT_SUCCESS;
839
+ break;
840
+ case UFS_UPIU_QUERY_OPCODE_READ_DESC:
841
+ status = ufs_read_desc(req);
842
+ break;
843
+ case UFS_UPIU_QUERY_OPCODE_READ_ATTR:
844
+ status = ufs_exec_query_attr(req, UFS_QUERY_ATTR_READ);
845
+ break;
846
+ case UFS_UPIU_QUERY_OPCODE_READ_FLAG:
847
+ status = ufs_exec_query_flag(req, UFS_QUERY_FLAG_READ);
848
+ break;
849
+ default:
850
+ trace_ufs_err_query_invalid_opcode(req->req_upiu.qr.opcode);
851
+ status = UFS_QUERY_RESULT_INVALID_OPCODE;
852
+ break;
853
+ }
854
+
855
+ return status;
856
+}
857
+
858
+static QueryRespCode ufs_exec_query_write(UfsRequest *req)
859
+{
860
+ QueryRespCode status;
861
+ switch (req->req_upiu.qr.opcode) {
862
+ case UFS_UPIU_QUERY_OPCODE_NOP:
863
+ status = UFS_QUERY_RESULT_SUCCESS;
864
+ break;
865
+ case UFS_UPIU_QUERY_OPCODE_WRITE_DESC:
866
+ /* write descriptor is not supported */
867
+ status = UFS_QUERY_RESULT_NOT_WRITEABLE;
868
+ break;
869
+ case UFS_UPIU_QUERY_OPCODE_WRITE_ATTR:
870
+ status = ufs_exec_query_attr(req, UFS_QUERY_ATTR_WRITE);
871
+ break;
872
+ case UFS_UPIU_QUERY_OPCODE_SET_FLAG:
873
+ status = ufs_exec_query_flag(req, UFS_QUERY_FLAG_SET);
874
+ break;
875
+ case UFS_UPIU_QUERY_OPCODE_CLEAR_FLAG:
876
+ status = ufs_exec_query_flag(req, UFS_QUERY_FLAG_CLEAR);
877
+ break;
878
+ case UFS_UPIU_QUERY_OPCODE_TOGGLE_FLAG:
879
+ status = ufs_exec_query_flag(req, UFS_QUERY_FLAG_TOGGLE);
880
+ break;
881
+ default:
882
+ trace_ufs_err_query_invalid_opcode(req->req_upiu.qr.opcode);
883
+ status = UFS_QUERY_RESULT_INVALID_OPCODE;
884
+ break;
885
+ }
886
+
887
+ return status;
888
+}
889
+
890
+static UfsReqResult ufs_exec_query_cmd(UfsRequest *req)
891
+{
892
+ uint8_t query_func = req->req_upiu.header.query_func;
893
+ uint16_t data_segment_length;
894
+ QueryRespCode status;
895
+
896
+ trace_ufs_exec_query_cmd(req->slot, req->req_upiu.qr.opcode);
897
+ if (query_func == UFS_UPIU_QUERY_FUNC_STANDARD_READ_REQUEST) {
898
+ status = ufs_exec_query_read(req);
899
+ } else if (query_func == UFS_UPIU_QUERY_FUNC_STANDARD_WRITE_REQUEST) {
900
+ status = ufs_exec_query_write(req);
901
+ } else {
902
+ status = UFS_QUERY_RESULT_GENERAL_FAILURE;
903
+ }
904
+
905
+ data_segment_length = be16_to_cpu(req->rsp_upiu.qr.length);
906
+ ufs_build_upiu_header(req, UFS_UPIU_TRANSACTION_QUERY_RSP, 0, status, 0,
907
+ data_segment_length);
908
+
909
+ if (status != UFS_QUERY_RESULT_SUCCESS) {
910
+ return UFS_REQUEST_FAIL;
911
+ }
912
+ return UFS_REQUEST_SUCCESS;
913
+}
914
+
915
+static void ufs_exec_req(UfsRequest *req)
916
+{
917
+ UfsReqResult req_result;
918
+
919
+ if (ufs_dma_read_upiu(req)) {
920
+ return;
921
+ }
922
+
923
+ switch (req->req_upiu.header.trans_type) {
924
+ case UFS_UPIU_TRANSACTION_NOP_OUT:
925
+ req_result = ufs_exec_nop_cmd(req);
926
+ break;
927
+ case UFS_UPIU_TRANSACTION_COMMAND:
928
+ /* Not yet implemented */
929
+ req_result = UFS_REQUEST_FAIL;
930
+ break;
931
+ case UFS_UPIU_TRANSACTION_QUERY_REQ:
932
+ req_result = ufs_exec_query_cmd(req);
933
+ break;
934
+ default:
935
+ trace_ufs_err_invalid_trans_code(req->slot,
936
+ req->req_upiu.header.trans_type);
937
+ req_result = UFS_REQUEST_FAIL;
938
+ }
939
+
940
+ ufs_complete_req(req, req_result);
941
+}
942
+
943
+static void ufs_process_req(void *opaque)
944
+{
945
+ UfsHc *u = opaque;
946
+ UfsRequest *req;
947
+ int slot;
948
+
949
+ for (slot = 0; slot < u->params.nutrs; slot++) {
950
+ req = &u->req_list[slot];
951
+
952
+ if (req->state != UFS_REQUEST_READY) {
953
+ continue;
954
+ }
955
+ trace_ufs_process_req(slot);
956
+ req->state = UFS_REQUEST_RUNNING;
957
+
958
+ ufs_exec_req(req);
959
+ }
960
+}
961
+
962
+static void ufs_complete_req(UfsRequest *req, UfsReqResult req_result)
963
+{
964
+ UfsHc *u = req->hc;
965
+ assert(req->state == UFS_REQUEST_RUNNING);
966
+
967
+ if (req_result == UFS_REQUEST_SUCCESS) {
968
+ req->utrd.header.dword_2 = cpu_to_le32(UFS_OCS_SUCCESS);
969
+ } else {
970
+ req->utrd.header.dword_2 = cpu_to_le32(UFS_OCS_INVALID_CMD_TABLE_ATTR);
971
+ }
972
+
973
+ trace_ufs_complete_req(req->slot);
974
+ req->state = UFS_REQUEST_COMPLETE;
975
+ qemu_bh_schedule(u->complete_bh);
976
+}
977
+
978
+static void ufs_clear_req(UfsRequest *req)
979
+{
980
+ if (req->sg != NULL) {
981
+ qemu_sglist_destroy(req->sg);
982
+ g_free(req->sg);
983
+ req->sg = NULL;
984
+ }
985
+
986
+ memset(&req->utrd, 0, sizeof(req->utrd));
987
+ memset(&req->req_upiu, 0, sizeof(req->req_upiu));
988
+ memset(&req->rsp_upiu, 0, sizeof(req->rsp_upiu));
989
+}
990
+
991
+static void ufs_sendback_req(void *opaque)
992
+{
993
+ UfsHc *u = opaque;
994
+ UfsRequest *req;
995
+ int slot;
996
+
997
+ for (slot = 0; slot < u->params.nutrs; slot++) {
998
+ req = &u->req_list[slot];
999
+
1000
+ if (req->state != UFS_REQUEST_COMPLETE) {
1001
+ continue;
1002
+ }
1003
+
1004
+ if (ufs_dma_write_upiu(req)) {
1005
+ req->state = UFS_REQUEST_ERROR;
1006
+ continue;
1007
+ }
1008
+
1009
+ /*
1010
+ * TODO: UTP Transfer Request Interrupt Aggregation Control is not yet
1011
+ * supported
1012
+ */
1013
+ if (le32_to_cpu(req->utrd.header.dword_2) != UFS_OCS_SUCCESS ||
1014
+ le32_to_cpu(req->utrd.header.dword_0) & UFS_UTP_REQ_DESC_INT_CMD) {
1015
+ u->reg.is = FIELD_DP32(u->reg.is, IS, UTRCS, 1);
1016
+ }
1017
+
1018
+ u->reg.utrldbr &= ~(1 << slot);
1019
+ u->reg.utrlcnr |= (1 << slot);
1020
+
1021
+ trace_ufs_sendback_req(req->slot);
1022
+
1023
+ ufs_clear_req(req);
1024
+ req->state = UFS_REQUEST_IDLE;
1025
+ }
1026
+
1027
+ ufs_irq_check(u);
1028
+}
1029
+
1030
static bool ufs_check_constraints(UfsHc *u, Error **errp)
1031
{
1032
if (u->params.nutrs > UFS_MAX_NUTRS) {
1033
@@ -XXX,XX +XXX,XX @@ static void ufs_init_pci(UfsHc *u, PCIDevice *pci_dev)
1034
u->irq = pci_allocate_irq(pci_dev);
1035
}
66
}
1036
67
1037
+static void ufs_init_state(UfsHc *u)
68
/**
1038
+{
69
@@ -XXX,XX +XXX,XX @@ static void qed_aio_write_l2_update(QEDAIOCB *acb, int ret, uint64_t offset)
1039
+ u->req_list = g_new0(UfsRequest, u->params.nutrs);
70
if (need_alloc) {
1040
+
71
/* Write out the whole new L2 table */
1041
+ for (int i = 0; i < u->params.nutrs; i++) {
72
qed_write_l2_table(s, &acb->request, 0, s->table_nelems, true,
1042
+ u->req_list[i].hc = u;
73
- qed_aio_write_l1_update, acb);
1043
+ u->req_list[i].slot = i;
74
+ qed_aio_write_l1_update, acb);
1044
+ u->req_list[i].sg = NULL;
75
} else {
1045
+ u->req_list[i].state = UFS_REQUEST_IDLE;
76
/* Write out only the updated part of the L2 table */
1046
+ }
77
qed_write_l2_table(s, &acb->request, index, acb->cur_nclusters, false,
1047
+
78
- qed_aio_next_io, acb);
1048
+ u->doorbell_bh = qemu_bh_new_guarded(ufs_process_req, u,
79
+ qed_aio_next_io_cb, acb);
1049
+ &DEVICE(u)->mem_reentrancy_guard);
80
}
1050
+ u->complete_bh = qemu_bh_new_guarded(ufs_sendback_req, u,
81
return;
1051
+ &DEVICE(u)->mem_reentrancy_guard);
82
1052
+}
83
@@ -XXX,XX +XXX,XX @@ static void qed_aio_write_main(void *opaque, int ret)
1053
+
84
}
1054
static void ufs_init_hc(UfsHc *u)
85
1055
{
86
if (acb->find_cluster_ret == QED_CLUSTER_FOUND) {
1056
uint32_t cap = 0;
87
- next_fn = qed_aio_next_io;
1057
@@ -XXX,XX +XXX,XX @@ static void ufs_init_hc(UfsHc *u)
88
+ next_fn = qed_aio_next_io_cb;
1058
cap = FIELD_DP32(cap, CAP, CS, 0);
89
} else {
1059
u->reg.cap = cap;
90
if (s->bs->backing) {
1060
u->reg.ver = UFS_SPEC_VER;
91
next_fn = qed_aio_write_flush_before_l2_update;
1061
+
92
@@ -XXX,XX +XXX,XX @@ static void qed_aio_write_alloc(QEDAIOCB *acb, size_t len)
1062
+ memset(&u->device_desc, 0, sizeof(DeviceDescriptor));
93
if (acb->flags & QED_AIOCB_ZERO) {
1063
+ u->device_desc.length = sizeof(DeviceDescriptor);
94
/* Skip ahead if the clusters are already zero */
1064
+ u->device_desc.descriptor_idn = UFS_QUERY_DESC_IDN_DEVICE;
95
if (acb->find_cluster_ret == QED_CLUSTER_ZERO) {
1065
+ u->device_desc.device_sub_class = 0x01;
96
- qed_aio_next_io(acb, 0);
1066
+ u->device_desc.number_lu = 0x00;
97
+ qed_aio_start_io(acb);
1067
+ u->device_desc.number_wlu = 0x04;
98
return;
1068
+ /* TODO: Revisit it when Power Management is implemented */
99
}
1069
+ u->device_desc.init_power_mode = 0x01; /* Active Mode */
100
1070
+ u->device_desc.high_priority_lun = 0x7F; /* Same Priority */
101
@@ -XXX,XX +XXX,XX @@ static void qed_aio_read_data(void *opaque, int ret,
1071
+ u->device_desc.spec_version = cpu_to_be16(UFS_SPEC_VER);
102
/* Handle zero cluster and backing file reads */
1072
+ u->device_desc.manufacturer_name = 0x00;
103
if (ret == QED_CLUSTER_ZERO) {
1073
+ u->device_desc.product_name = 0x01;
104
qemu_iovec_memset(&acb->cur_qiov, 0, 0, acb->cur_qiov.size);
1074
+ u->device_desc.serial_number = 0x02;
105
- qed_aio_next_io(acb, 0);
1075
+ u->device_desc.oem_id = 0x03;
106
+ qed_aio_start_io(acb);
1076
+ u->device_desc.ud_0_base_offset = 0x16;
107
return;
1077
+ u->device_desc.ud_config_p_length = 0x1A;
108
} else if (ret != QED_CLUSTER_FOUND) {
1078
+ u->device_desc.device_rtt_cap = 0x02;
109
qed_read_backing_file(s, acb->cur_pos, &acb->cur_qiov,
1079
+ u->device_desc.queue_depth = u->params.nutrs;
110
- &acb->backing_qiov, qed_aio_next_io, acb);
1080
+ u->device_desc.product_revision_level = 0x04;
111
+ &acb->backing_qiov, qed_aio_next_io_cb, acb);
1081
+
1082
+ memset(&u->geometry_desc, 0, sizeof(GeometryDescriptor));
1083
+ u->geometry_desc.length = sizeof(GeometryDescriptor);
1084
+ u->geometry_desc.descriptor_idn = UFS_QUERY_DESC_IDN_GEOMETRY;
1085
+ u->geometry_desc.max_number_lu = (UFS_MAX_LUS == 32) ? 0x1 : 0x0;
1086
+ u->geometry_desc.segment_size = cpu_to_be32(0x2000); /* 4KB */
1087
+ u->geometry_desc.allocation_unit_size = 0x1; /* 4KB */
1088
+ u->geometry_desc.min_addr_block_size = 0x8; /* 4KB */
1089
+ u->geometry_desc.max_in_buffer_size = 0x8;
1090
+ u->geometry_desc.max_out_buffer_size = 0x8;
1091
+ u->geometry_desc.rpmb_read_write_size = 0x40;
1092
+ u->geometry_desc.data_ordering =
1093
+ 0x0; /* out-of-order data transfer is not supported */
1094
+ u->geometry_desc.max_context_id_number = 0x5;
1095
+ u->geometry_desc.supported_memory_types = cpu_to_be16(0x8001);
1096
+
1097
+ memset(&u->attributes, 0, sizeof(u->attributes));
1098
+ u->attributes.max_data_in_size = 0x08;
1099
+ u->attributes.max_data_out_size = 0x08;
1100
+ u->attributes.ref_clk_freq = 0x01; /* 26 MHz */
1101
+ /* configure descriptor is not supported */
1102
+ u->attributes.config_descr_lock = 0x01;
1103
+ u->attributes.max_num_of_rtt = 0x02;
1104
+
1105
+ memset(&u->flags, 0, sizeof(u->flags));
1106
+ u->flags.permanently_disable_fw_update = 1;
1107
}
1108
1109
static void ufs_realize(PCIDevice *pci_dev, Error **errp)
1110
@@ -XXX,XX +XXX,XX @@ static void ufs_realize(PCIDevice *pci_dev, Error **errp)
1111
return;
112
return;
1112
}
113
}
1113
114
1114
+ ufs_init_state(u);
115
BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO);
1115
ufs_init_hc(u);
116
bdrv_aio_readv(bs->file, offset / BDRV_SECTOR_SIZE,
1116
ufs_init_pci(u, pci_dev);
117
&acb->cur_qiov, acb->cur_qiov.size / BDRV_SECTOR_SIZE,
118
- qed_aio_next_io, acb);
119
+ qed_aio_next_io_cb, acb);
120
return;
121
122
err:
123
@@ -XXX,XX +XXX,XX @@ err:
124
/**
125
* Begin next I/O or complete the request
126
*/
127
-static void qed_aio_next_io(void *opaque, int ret)
128
+static void qed_aio_next_io(QEDAIOCB *acb, int ret)
129
{
130
- QEDAIOCB *acb = opaque;
131
BDRVQEDState *s = acb_to_s(acb);
132
QEDFindClusterFunc *io_fn = (acb->flags & QED_AIOCB_WRITE) ?
133
qed_aio_write_data : qed_aio_read_data;
134
@@ -XXX,XX +XXX,XX @@ static BlockAIOCB *qed_aio_setup(BlockDriverState *bs,
135
qemu_iovec_init(&acb->cur_qiov, qiov->niov);
136
137
/* Start request */
138
- qed_aio_next_io(acb, 0);
139
+ qed_aio_start_io(acb);
140
return &acb->common;
1117
}
141
}
1118
142
1119
+static void ufs_exit(PCIDevice *pci_dev)
1120
+{
1121
+ UfsHc *u = UFS(pci_dev);
1122
+
1123
+ qemu_bh_delete(u->doorbell_bh);
1124
+ qemu_bh_delete(u->complete_bh);
1125
+
1126
+ for (int i = 0; i < u->params.nutrs; i++) {
1127
+ ufs_clear_req(&u->req_list[i]);
1128
+ }
1129
+ g_free(u->req_list);
1130
+}
1131
+
1132
static Property ufs_props[] = {
1133
DEFINE_PROP_STRING("serial", UfsHc, params.serial),
1134
DEFINE_PROP_UINT8("nutrs", UfsHc, params.nutrs, 32),
1135
@@ -XXX,XX +XXX,XX @@ static void ufs_class_init(ObjectClass *oc, void *data)
1136
PCIDeviceClass *pc = PCI_DEVICE_CLASS(oc);
1137
1138
pc->realize = ufs_realize;
1139
+ pc->exit = ufs_exit;
1140
pc->vendor_id = PCI_VENDOR_ID_REDHAT;
1141
pc->device_id = PCI_DEVICE_ID_REDHAT_UFS;
1142
pc->class_id = PCI_CLASS_STORAGE_UFS;
1143
diff --git a/hw/ufs/trace-events b/hw/ufs/trace-events
1144
index XXXXXXX..XXXXXXX 100644
1145
--- a/hw/ufs/trace-events
1146
+++ b/hw/ufs/trace-events
1147
@@ -XXX,XX +XXX,XX @@ ufs_err_dma_read_req_upiu(uint32_t slot, uint64_t addr) "failed to read req upiu
1148
ufs_err_dma_read_prdt(uint32_t slot, uint64_t addr) "failed to read prdt. UTRLDBR slot %"PRIu32", prdt addr %"PRIu64""
1149
ufs_err_dma_write_utrd(uint32_t slot, uint64_t addr) "failed to write utrd. UTRLDBR slot %"PRIu32", UTRD dma addr %"PRIu64""
1150
ufs_err_dma_write_rsp_upiu(uint32_t slot, uint64_t addr) "failed to write rsp upiu. UTRLDBR slot %"PRIu32", response upiu addr %"PRIu64""
1151
+ufs_err_utrl_slot_error(uint32_t slot) "UTRLDBR slot %"PRIu32" is in error"
1152
ufs_err_utrl_slot_busy(uint32_t slot) "UTRLDBR slot %"PRIu32" is busy"
1153
ufs_err_unsupport_register_offset(uint32_t offset) "Register offset 0x%"PRIx32" is not yet supported"
1154
ufs_err_invalid_register_offset(uint32_t offset) "Register offset 0x%"PRIx32" is invalid"
1155
--
143
--
1156
2.41.0
144
2.9.3
145
146
diff view generated by jsdifflib
New patch
1
From: Paolo Bonzini <pbonzini@redhat.com>
1
2
3
The AioContext data structures are now protected by list_lock and/or
4
they are walked with FOREACH_RCU primitives. There is no need anymore
5
to acquire the AioContext for the entire duration of aio_dispatch.
6
Instead, just acquire it before and after invoking the callbacks.
7
The next step is then to push it further down.
8
9
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
10
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
11
Reviewed-by: Fam Zheng <famz@redhat.com>
12
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
13
Message-id: 20170213135235.12274-12-pbonzini@redhat.com
14
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
15
---
16
util/aio-posix.c | 25 +++++++++++--------------
17
util/aio-win32.c | 15 +++++++--------
18
util/async.c | 2 ++
19
3 files changed, 20 insertions(+), 22 deletions(-)
20
21
diff --git a/util/aio-posix.c b/util/aio-posix.c
22
index XXXXXXX..XXXXXXX 100644
23
--- a/util/aio-posix.c
24
+++ b/util/aio-posix.c
25
@@ -XXX,XX +XXX,XX @@ static bool aio_dispatch_handlers(AioContext *ctx)
26
(revents & (G_IO_IN | G_IO_HUP | G_IO_ERR)) &&
27
aio_node_check(ctx, node->is_external) &&
28
node->io_read) {
29
+ aio_context_acquire(ctx);
30
node->io_read(node->opaque);
31
+ aio_context_release(ctx);
32
33
/* aio_notify() does not count as progress */
34
if (node->opaque != &ctx->notifier) {
35
@@ -XXX,XX +XXX,XX @@ static bool aio_dispatch_handlers(AioContext *ctx)
36
(revents & (G_IO_OUT | G_IO_ERR)) &&
37
aio_node_check(ctx, node->is_external) &&
38
node->io_write) {
39
+ aio_context_acquire(ctx);
40
node->io_write(node->opaque);
41
+ aio_context_release(ctx);
42
progress = true;
43
}
44
45
@@ -XXX,XX +XXX,XX @@ bool aio_dispatch(AioContext *ctx, bool dispatch_fds)
46
}
47
48
/* Run our timers */
49
+ aio_context_acquire(ctx);
50
progress |= timerlistgroup_run_timers(&ctx->tlg);
51
+ aio_context_release(ctx);
52
53
return progress;
54
}
55
@@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking)
56
int64_t timeout;
57
int64_t start = 0;
58
59
- aio_context_acquire(ctx);
60
- progress = false;
61
-
62
/* aio_notify can avoid the expensive event_notifier_set if
63
* everything (file descriptors, bottom halves, timers) will
64
* be re-evaluated before the next blocking poll(). This is
65
@@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking)
66
start = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
67
}
68
69
- if (try_poll_mode(ctx, blocking)) {
70
- progress = true;
71
- } else {
72
+ aio_context_acquire(ctx);
73
+ progress = try_poll_mode(ctx, blocking);
74
+ aio_context_release(ctx);
75
+
76
+ if (!progress) {
77
assert(npfd == 0);
78
79
/* fill pollfds */
80
@@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking)
81
timeout = blocking ? aio_compute_timeout(ctx) : 0;
82
83
/* wait until next event */
84
- if (timeout) {
85
- aio_context_release(ctx);
86
- }
87
if (aio_epoll_check_poll(ctx, pollfds, npfd, timeout)) {
88
AioHandler epoll_handler;
89
90
@@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking)
91
} else {
92
ret = qemu_poll_ns(pollfds, npfd, timeout);
93
}
94
- if (timeout) {
95
- aio_context_acquire(ctx);
96
- }
97
}
98
99
if (blocking) {
100
@@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking)
101
progress = true;
102
}
103
104
- aio_context_release(ctx);
105
-
106
return progress;
107
}
108
109
diff --git a/util/aio-win32.c b/util/aio-win32.c
110
index XXXXXXX..XXXXXXX 100644
111
--- a/util/aio-win32.c
112
+++ b/util/aio-win32.c
113
@@ -XXX,XX +XXX,XX @@ static bool aio_dispatch_handlers(AioContext *ctx, HANDLE event)
114
(revents || event_notifier_get_handle(node->e) == event) &&
115
node->io_notify) {
116
node->pfd.revents = 0;
117
+ aio_context_acquire(ctx);
118
node->io_notify(node->e);
119
+ aio_context_release(ctx);
120
121
/* aio_notify() does not count as progress */
122
if (node->e != &ctx->notifier) {
123
@@ -XXX,XX +XXX,XX @@ static bool aio_dispatch_handlers(AioContext *ctx, HANDLE event)
124
(node->io_read || node->io_write)) {
125
node->pfd.revents = 0;
126
if ((revents & G_IO_IN) && node->io_read) {
127
+ aio_context_acquire(ctx);
128
node->io_read(node->opaque);
129
+ aio_context_release(ctx);
130
progress = true;
131
}
132
if ((revents & G_IO_OUT) && node->io_write) {
133
+ aio_context_acquire(ctx);
134
node->io_write(node->opaque);
135
+ aio_context_release(ctx);
136
progress = true;
137
}
138
139
@@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking)
140
int count;
141
int timeout;
142
143
- aio_context_acquire(ctx);
144
progress = false;
145
146
/* aio_notify can avoid the expensive event_notifier_set if
147
@@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking)
148
149
timeout = blocking && !have_select_revents
150
? qemu_timeout_ns_to_ms(aio_compute_timeout(ctx)) : 0;
151
- if (timeout) {
152
- aio_context_release(ctx);
153
- }
154
ret = WaitForMultipleObjects(count, events, FALSE, timeout);
155
if (blocking) {
156
assert(first);
157
atomic_sub(&ctx->notify_me, 2);
158
}
159
- if (timeout) {
160
- aio_context_acquire(ctx);
161
- }
162
163
if (first) {
164
aio_notify_accept(ctx);
165
@@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking)
166
progress |= aio_dispatch_handlers(ctx, event);
167
} while (count > 0);
168
169
+ aio_context_acquire(ctx);
170
progress |= timerlistgroup_run_timers(&ctx->tlg);
171
-
172
aio_context_release(ctx);
173
return progress;
174
}
175
diff --git a/util/async.c b/util/async.c
176
index XXXXXXX..XXXXXXX 100644
177
--- a/util/async.c
178
+++ b/util/async.c
179
@@ -XXX,XX +XXX,XX @@ int aio_bh_poll(AioContext *ctx)
180
ret = 1;
181
}
182
bh->idle = 0;
183
+ aio_context_acquire(ctx);
184
aio_bh_call(bh);
185
+ aio_context_release(ctx);
186
}
187
if (bh->deleted) {
188
deleted = true;
189
--
190
2.9.3
191
192
diff view generated by jsdifflib
New patch
1
1
From: Paolo Bonzini <pbonzini@redhat.com>
2
3
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
4
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5
Reviewed-by: Fam Zheng <famz@redhat.com>
6
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
7
Message-id: 20170213135235.12274-13-pbonzini@redhat.com
8
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
9
---
10
block/qed.h | 3 +++
11
block/curl.c | 2 ++
12
block/io.c | 5 +++++
13
block/iscsi.c | 8 ++++++--
14
block/null.c | 4 ++++
15
block/qed.c | 12 ++++++++++++
16
block/throttle-groups.c | 2 ++
17
util/aio-posix.c | 2 --
18
util/aio-win32.c | 2 --
19
util/qemu-coroutine-sleep.c | 2 +-
20
10 files changed, 35 insertions(+), 7 deletions(-)
21
22
diff --git a/block/qed.h b/block/qed.h
23
index XXXXXXX..XXXXXXX 100644
24
--- a/block/qed.h
25
+++ b/block/qed.h
26
@@ -XXX,XX +XXX,XX @@ enum {
27
*/
28
typedef void QEDFindClusterFunc(void *opaque, int ret, uint64_t offset, size_t len);
29
30
+void qed_acquire(BDRVQEDState *s);
31
+void qed_release(BDRVQEDState *s);
32
+
33
/**
34
* Generic callback for chaining async callbacks
35
*/
36
diff --git a/block/curl.c b/block/curl.c
37
index XXXXXXX..XXXXXXX 100644
38
--- a/block/curl.c
39
+++ b/block/curl.c
40
@@ -XXX,XX +XXX,XX @@ static void curl_multi_timeout_do(void *arg)
41
return;
42
}
43
44
+ aio_context_acquire(s->aio_context);
45
curl_multi_socket_action(s->multi, CURL_SOCKET_TIMEOUT, 0, &running);
46
47
curl_multi_check_completion(s);
48
+ aio_context_release(s->aio_context);
49
#else
50
abort();
51
#endif
52
diff --git a/block/io.c b/block/io.c
53
index XXXXXXX..XXXXXXX 100644
54
--- a/block/io.c
55
+++ b/block/io.c
56
@@ -XXX,XX +XXX,XX @@ void bdrv_aio_cancel(BlockAIOCB *acb)
57
if (acb->aiocb_info->get_aio_context) {
58
aio_poll(acb->aiocb_info->get_aio_context(acb), true);
59
} else if (acb->bs) {
60
+ /* qemu_aio_ref and qemu_aio_unref are not thread-safe, so
61
+ * assert that we're not using an I/O thread. Thread-safe
62
+ * code should use bdrv_aio_cancel_async exclusively.
63
+ */
64
+ assert(bdrv_get_aio_context(acb->bs) == qemu_get_aio_context());
65
aio_poll(bdrv_get_aio_context(acb->bs), true);
66
} else {
67
abort();
68
diff --git a/block/iscsi.c b/block/iscsi.c
69
index XXXXXXX..XXXXXXX 100644
70
--- a/block/iscsi.c
71
+++ b/block/iscsi.c
72
@@ -XXX,XX +XXX,XX @@ static void iscsi_retry_timer_expired(void *opaque)
73
struct IscsiTask *iTask = opaque;
74
iTask->complete = 1;
75
if (iTask->co) {
76
- qemu_coroutine_enter(iTask->co);
77
+ aio_co_wake(iTask->co);
78
}
79
}
80
81
@@ -XXX,XX +XXX,XX @@ static void iscsi_nop_timed_event(void *opaque)
82
{
83
IscsiLun *iscsilun = opaque;
84
85
+ aio_context_acquire(iscsilun->aio_context);
86
if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) {
87
error_report("iSCSI: NOP timeout. Reconnecting...");
88
iscsilun->request_timed_out = true;
89
} else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
90
error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
91
- return;
92
+ goto out;
93
}
94
95
timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
96
iscsi_set_events(iscsilun);
97
+
98
+out:
99
+ aio_context_release(iscsilun->aio_context);
100
}
101
102
static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
103
diff --git a/block/null.c b/block/null.c
104
index XXXXXXX..XXXXXXX 100644
105
--- a/block/null.c
106
+++ b/block/null.c
107
@@ -XXX,XX +XXX,XX @@ static void null_bh_cb(void *opaque)
108
static void null_timer_cb(void *opaque)
109
{
110
NullAIOCB *acb = opaque;
111
+ AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
112
+
113
+ aio_context_acquire(ctx);
114
acb->common.cb(acb->common.opaque, 0);
115
+ aio_context_release(ctx);
116
timer_deinit(&acb->timer);
117
qemu_aio_unref(acb);
118
}
119
diff --git a/block/qed.c b/block/qed.c
120
index XXXXXXX..XXXXXXX 100644
121
--- a/block/qed.c
122
+++ b/block/qed.c
123
@@ -XXX,XX +XXX,XX @@ static void qed_need_check_timer_cb(void *opaque)
124
125
trace_qed_need_check_timer_cb(s);
126
127
+ qed_acquire(s);
128
qed_plug_allocating_write_reqs(s);
129
130
/* Ensure writes are on disk before clearing flag */
131
bdrv_aio_flush(s->bs->file->bs, qed_clear_need_check, s);
132
+ qed_release(s);
133
+}
134
+
135
+void qed_acquire(BDRVQEDState *s)
136
+{
137
+ aio_context_acquire(bdrv_get_aio_context(s->bs));
138
+}
139
+
140
+void qed_release(BDRVQEDState *s)
141
+{
142
+ aio_context_release(bdrv_get_aio_context(s->bs));
143
}
144
145
static void qed_start_need_check_timer(BDRVQEDState *s)
146
diff --git a/block/throttle-groups.c b/block/throttle-groups.c
147
index XXXXXXX..XXXXXXX 100644
148
--- a/block/throttle-groups.c
149
+++ b/block/throttle-groups.c
150
@@ -XXX,XX +XXX,XX @@ static void timer_cb(BlockBackend *blk, bool is_write)
151
qemu_mutex_unlock(&tg->lock);
152
153
/* Run the request that was waiting for this timer */
154
+ aio_context_acquire(blk_get_aio_context(blk));
155
empty_queue = !qemu_co_enter_next(&blkp->throttled_reqs[is_write]);
156
+ aio_context_release(blk_get_aio_context(blk));
157
158
/* If the request queue was empty then we have to take care of
159
* scheduling the next one */
160
diff --git a/util/aio-posix.c b/util/aio-posix.c
161
index XXXXXXX..XXXXXXX 100644
162
--- a/util/aio-posix.c
163
+++ b/util/aio-posix.c
164
@@ -XXX,XX +XXX,XX @@ bool aio_dispatch(AioContext *ctx, bool dispatch_fds)
165
}
166
167
/* Run our timers */
168
- aio_context_acquire(ctx);
169
progress |= timerlistgroup_run_timers(&ctx->tlg);
170
- aio_context_release(ctx);
171
172
return progress;
173
}
174
diff --git a/util/aio-win32.c b/util/aio-win32.c
175
index XXXXXXX..XXXXXXX 100644
176
--- a/util/aio-win32.c
177
+++ b/util/aio-win32.c
178
@@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking)
179
progress |= aio_dispatch_handlers(ctx, event);
180
} while (count > 0);
181
182
- aio_context_acquire(ctx);
183
progress |= timerlistgroup_run_timers(&ctx->tlg);
184
- aio_context_release(ctx);
185
return progress;
186
}
187
188
diff --git a/util/qemu-coroutine-sleep.c b/util/qemu-coroutine-sleep.c
189
index XXXXXXX..XXXXXXX 100644
190
--- a/util/qemu-coroutine-sleep.c
191
+++ b/util/qemu-coroutine-sleep.c
192
@@ -XXX,XX +XXX,XX @@ static void co_sleep_cb(void *opaque)
193
{
194
CoSleepCB *sleep_cb = opaque;
195
196
- qemu_coroutine_enter(sleep_cb->co);
197
+ aio_co_wake(sleep_cb->co);
198
}
199
200
void coroutine_fn co_aio_sleep_ns(AioContext *ctx, QEMUClockType type,
201
--
202
2.9.3
203
204
diff view generated by jsdifflib
New patch
1
From: Paolo Bonzini <pbonzini@redhat.com>
1
2
3
This covers both file descriptor callbacks and polling callbacks,
4
since they execute related code.
5
6
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
7
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8
Reviewed-by: Fam Zheng <famz@redhat.com>
9
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
10
Message-id: 20170213135235.12274-14-pbonzini@redhat.com
11
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
12
---
13
block/curl.c | 16 +++++++++++++---
14
block/iscsi.c | 4 ++++
15
block/linux-aio.c | 4 ++++
16
block/nfs.c | 6 ++++++
17
block/sheepdog.c | 29 +++++++++++++++--------------
18
block/ssh.c | 29 +++++++++--------------------
19
block/win32-aio.c | 10 ++++++----
20
hw/block/virtio-blk.c | 5 ++++-
21
hw/scsi/virtio-scsi.c | 7 +++++++
22
util/aio-posix.c | 7 -------
23
util/aio-win32.c | 6 ------
24
11 files changed, 68 insertions(+), 55 deletions(-)
25
26
diff --git a/block/curl.c b/block/curl.c
27
index XXXXXXX..XXXXXXX 100644
28
--- a/block/curl.c
29
+++ b/block/curl.c
30
@@ -XXX,XX +XXX,XX @@ static void curl_multi_check_completion(BDRVCURLState *s)
31
}
32
}
33
34
-static void curl_multi_do(void *arg)
35
+static void curl_multi_do_locked(CURLState *s)
36
{
37
- CURLState *s = (CURLState *)arg;
38
CURLSocket *socket, *next_socket;
39
int running;
40
int r;
41
@@ -XXX,XX +XXX,XX @@ static void curl_multi_do(void *arg)
42
}
43
}
44
45
+static void curl_multi_do(void *arg)
46
+{
47
+ CURLState *s = (CURLState *)arg;
48
+
49
+ aio_context_acquire(s->s->aio_context);
50
+ curl_multi_do_locked(s);
51
+ aio_context_release(s->s->aio_context);
52
+}
53
+
54
static void curl_multi_read(void *arg)
55
{
56
CURLState *s = (CURLState *)arg;
57
58
- curl_multi_do(arg);
59
+ aio_context_acquire(s->s->aio_context);
60
+ curl_multi_do_locked(s);
61
curl_multi_check_completion(s->s);
62
+ aio_context_release(s->s->aio_context);
63
}
64
65
static void curl_multi_timeout_do(void *arg)
66
diff --git a/block/iscsi.c b/block/iscsi.c
67
index XXXXXXX..XXXXXXX 100644
68
--- a/block/iscsi.c
69
+++ b/block/iscsi.c
70
@@ -XXX,XX +XXX,XX @@ iscsi_process_read(void *arg)
71
IscsiLun *iscsilun = arg;
72
struct iscsi_context *iscsi = iscsilun->iscsi;
73
74
+ aio_context_acquire(iscsilun->aio_context);
75
iscsi_service(iscsi, POLLIN);
76
iscsi_set_events(iscsilun);
77
+ aio_context_release(iscsilun->aio_context);
78
}
79
80
static void
81
@@ -XXX,XX +XXX,XX @@ iscsi_process_write(void *arg)
82
IscsiLun *iscsilun = arg;
83
struct iscsi_context *iscsi = iscsilun->iscsi;
84
85
+ aio_context_acquire(iscsilun->aio_context);
86
iscsi_service(iscsi, POLLOUT);
87
iscsi_set_events(iscsilun);
88
+ aio_context_release(iscsilun->aio_context);
89
}
90
91
static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
92
diff --git a/block/linux-aio.c b/block/linux-aio.c
93
index XXXXXXX..XXXXXXX 100644
94
--- a/block/linux-aio.c
95
+++ b/block/linux-aio.c
96
@@ -XXX,XX +XXX,XX @@ static void qemu_laio_completion_cb(EventNotifier *e)
97
LinuxAioState *s = container_of(e, LinuxAioState, e);
98
99
if (event_notifier_test_and_clear(&s->e)) {
100
+ aio_context_acquire(s->aio_context);
101
qemu_laio_process_completions_and_submit(s);
102
+ aio_context_release(s->aio_context);
103
}
104
}
105
106
@@ -XXX,XX +XXX,XX @@ static bool qemu_laio_poll_cb(void *opaque)
107
return false;
108
}
109
110
+ aio_context_acquire(s->aio_context);
111
qemu_laio_process_completions_and_submit(s);
112
+ aio_context_release(s->aio_context);
113
return true;
114
}
115
116
diff --git a/block/nfs.c b/block/nfs.c
117
index XXXXXXX..XXXXXXX 100644
118
--- a/block/nfs.c
119
+++ b/block/nfs.c
120
@@ -XXX,XX +XXX,XX @@ static void nfs_set_events(NFSClient *client)
121
static void nfs_process_read(void *arg)
122
{
123
NFSClient *client = arg;
124
+
125
+ aio_context_acquire(client->aio_context);
126
nfs_service(client->context, POLLIN);
127
nfs_set_events(client);
128
+ aio_context_release(client->aio_context);
129
}
130
131
static void nfs_process_write(void *arg)
132
{
133
NFSClient *client = arg;
134
+
135
+ aio_context_acquire(client->aio_context);
136
nfs_service(client->context, POLLOUT);
137
nfs_set_events(client);
138
+ aio_context_release(client->aio_context);
139
}
140
141
static void nfs_co_init_task(BlockDriverState *bs, NFSRPC *task)
142
diff --git a/block/sheepdog.c b/block/sheepdog.c
143
index XXXXXXX..XXXXXXX 100644
144
--- a/block/sheepdog.c
145
+++ b/block/sheepdog.c
146
@@ -XXX,XX +XXX,XX @@ static coroutine_fn int send_co_req(int sockfd, SheepdogReq *hdr, void *data,
147
return ret;
148
}
149
150
-static void restart_co_req(void *opaque)
151
-{
152
- Coroutine *co = opaque;
153
-
154
- qemu_coroutine_enter(co);
155
-}
156
-
157
typedef struct SheepdogReqCo {
158
int sockfd;
159
BlockDriverState *bs;
160
@@ -XXX,XX +XXX,XX @@ typedef struct SheepdogReqCo {
161
unsigned int *rlen;
162
int ret;
163
bool finished;
164
+ Coroutine *co;
165
} SheepdogReqCo;
166
167
+static void restart_co_req(void *opaque)
168
+{
169
+ SheepdogReqCo *srco = opaque;
170
+
171
+ aio_co_wake(srco->co);
172
+}
173
+
174
static coroutine_fn void do_co_req(void *opaque)
175
{
176
int ret;
177
- Coroutine *co;
178
SheepdogReqCo *srco = opaque;
179
int sockfd = srco->sockfd;
180
SheepdogReq *hdr = srco->hdr;
181
@@ -XXX,XX +XXX,XX @@ static coroutine_fn void do_co_req(void *opaque)
182
unsigned int *wlen = srco->wlen;
183
unsigned int *rlen = srco->rlen;
184
185
- co = qemu_coroutine_self();
186
+ srco->co = qemu_coroutine_self();
187
aio_set_fd_handler(srco->aio_context, sockfd, false,
188
- NULL, restart_co_req, NULL, co);
189
+ NULL, restart_co_req, NULL, srco);
190
191
ret = send_co_req(sockfd, hdr, data, wlen);
192
if (ret < 0) {
193
@@ -XXX,XX +XXX,XX @@ static coroutine_fn void do_co_req(void *opaque)
194
}
195
196
aio_set_fd_handler(srco->aio_context, sockfd, false,
197
- restart_co_req, NULL, NULL, co);
198
+ restart_co_req, NULL, NULL, srco);
199
200
ret = qemu_co_recv(sockfd, hdr, sizeof(*hdr));
201
if (ret != sizeof(*hdr)) {
202
@@ -XXX,XX +XXX,XX @@ out:
203
aio_set_fd_handler(srco->aio_context, sockfd, false,
204
NULL, NULL, NULL, NULL);
205
206
+ srco->co = NULL;
207
srco->ret = ret;
208
srco->finished = true;
209
if (srco->bs) {
210
@@ -XXX,XX +XXX,XX @@ static void coroutine_fn aio_read_response(void *opaque)
211
* We've finished all requests which belong to the AIOCB, so
212
* we can switch back to sd_co_readv/writev now.
213
*/
214
- qemu_coroutine_enter(acb->coroutine);
215
+ aio_co_wake(acb->coroutine);
216
}
217
218
return;
219
@@ -XXX,XX +XXX,XX @@ static void co_read_response(void *opaque)
220
s->co_recv = qemu_coroutine_create(aio_read_response, opaque);
221
}
222
223
- qemu_coroutine_enter(s->co_recv);
224
+ aio_co_wake(s->co_recv);
225
}
226
227
static void co_write_request(void *opaque)
228
{
229
BDRVSheepdogState *s = opaque;
230
231
- qemu_coroutine_enter(s->co_send);
232
+ aio_co_wake(s->co_send);
233
}
234
235
/*
236
diff --git a/block/ssh.c b/block/ssh.c
237
index XXXXXXX..XXXXXXX 100644
238
--- a/block/ssh.c
239
+++ b/block/ssh.c
240
@@ -XXX,XX +XXX,XX @@ static void restart_coroutine(void *opaque)
241
242
DPRINTF("co=%p", co);
243
244
- qemu_coroutine_enter(co);
245
+ aio_co_wake(co);
246
}
247
248
-static coroutine_fn void set_fd_handler(BDRVSSHState *s, BlockDriverState *bs)
249
+/* A non-blocking call returned EAGAIN, so yield, ensuring the
250
+ * handlers are set up so that we'll be rescheduled when there is an
251
+ * interesting event on the socket.
252
+ */
253
+static coroutine_fn void co_yield(BDRVSSHState *s, BlockDriverState *bs)
254
{
255
int r;
256
IOHandler *rd_handler = NULL, *wr_handler = NULL;
257
@@ -XXX,XX +XXX,XX @@ static coroutine_fn void set_fd_handler(BDRVSSHState *s, BlockDriverState *bs)
258
259
aio_set_fd_handler(bdrv_get_aio_context(bs), s->sock,
260
false, rd_handler, wr_handler, NULL, co);
261
-}
262
-
263
-static coroutine_fn void clear_fd_handler(BDRVSSHState *s,
264
- BlockDriverState *bs)
265
-{
266
- DPRINTF("s->sock=%d", s->sock);
267
- aio_set_fd_handler(bdrv_get_aio_context(bs), s->sock,
268
- false, NULL, NULL, NULL, NULL);
269
-}
270
-
271
-/* A non-blocking call returned EAGAIN, so yield, ensuring the
272
- * handlers are set up so that we'll be rescheduled when there is an
273
- * interesting event on the socket.
274
- */
275
-static coroutine_fn void co_yield(BDRVSSHState *s, BlockDriverState *bs)
276
-{
277
- set_fd_handler(s, bs);
278
qemu_coroutine_yield();
279
- clear_fd_handler(s, bs);
280
+ DPRINTF("s->sock=%d - back", s->sock);
281
+ aio_set_fd_handler(bdrv_get_aio_context(bs), s->sock, false,
282
+ NULL, NULL, NULL, NULL);
283
}
284
285
/* SFTP has a function `libssh2_sftp_seek64' which seeks to a position
286
diff --git a/block/win32-aio.c b/block/win32-aio.c
287
index XXXXXXX..XXXXXXX 100644
288
--- a/block/win32-aio.c
289
+++ b/block/win32-aio.c
290
@@ -XXX,XX +XXX,XX @@ struct QEMUWin32AIOState {
291
HANDLE hIOCP;
292
EventNotifier e;
293
int count;
294
- bool is_aio_context_attached;
295
+ AioContext *aio_ctx;
296
};
297
298
typedef struct QEMUWin32AIOCB {
299
@@ -XXX,XX +XXX,XX @@ static void win32_aio_process_completion(QEMUWin32AIOState *s,
300
}
301
302
303
+ aio_context_acquire(s->aio_ctx);
304
waiocb->common.cb(waiocb->common.opaque, ret);
305
+ aio_context_release(s->aio_ctx);
306
qemu_aio_unref(waiocb);
307
}
308
309
@@ -XXX,XX +XXX,XX @@ void win32_aio_detach_aio_context(QEMUWin32AIOState *aio,
310
AioContext *old_context)
311
{
312
aio_set_event_notifier(old_context, &aio->e, false, NULL, NULL);
313
- aio->is_aio_context_attached = false;
314
+ aio->aio_ctx = NULL;
315
}
316
317
void win32_aio_attach_aio_context(QEMUWin32AIOState *aio,
318
AioContext *new_context)
319
{
320
- aio->is_aio_context_attached = true;
321
+ aio->aio_ctx = new_context;
322
aio_set_event_notifier(new_context, &aio->e, false,
323
win32_aio_completion_cb, NULL);
324
}
325
@@ -XXX,XX +XXX,XX @@ out_free_state:
326
327
void win32_aio_cleanup(QEMUWin32AIOState *aio)
328
{
329
- assert(!aio->is_aio_context_attached);
330
+ assert(!aio->aio_ctx);
331
CloseHandle(aio->hIOCP);
332
event_notifier_cleanup(&aio->e);
333
g_free(aio);
334
diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c
335
index XXXXXXX..XXXXXXX 100644
336
--- a/hw/block/virtio-blk.c
337
+++ b/hw/block/virtio-blk.c
338
@@ -XXX,XX +XXX,XX @@ static void virtio_blk_ioctl_complete(void *opaque, int status)
339
{
340
VirtIOBlockIoctlReq *ioctl_req = opaque;
341
VirtIOBlockReq *req = ioctl_req->req;
342
- VirtIODevice *vdev = VIRTIO_DEVICE(req->dev);
343
+ VirtIOBlock *s = req->dev;
344
+ VirtIODevice *vdev = VIRTIO_DEVICE(s);
345
struct virtio_scsi_inhdr *scsi;
346
struct sg_io_hdr *hdr;
347
348
@@ -XXX,XX +XXX,XX @@ bool virtio_blk_handle_vq(VirtIOBlock *s, VirtQueue *vq)
349
MultiReqBuffer mrb = {};
350
bool progress = false;
351
352
+ aio_context_acquire(blk_get_aio_context(s->blk));
353
blk_io_plug(s->blk);
354
355
do {
356
@@ -XXX,XX +XXX,XX @@ bool virtio_blk_handle_vq(VirtIOBlock *s, VirtQueue *vq)
357
}
358
359
blk_io_unplug(s->blk);
360
+ aio_context_release(blk_get_aio_context(s->blk));
361
return progress;
362
}
363
364
diff --git a/hw/scsi/virtio-scsi.c b/hw/scsi/virtio-scsi.c
365
index XXXXXXX..XXXXXXX 100644
366
--- a/hw/scsi/virtio-scsi.c
367
+++ b/hw/scsi/virtio-scsi.c
368
@@ -XXX,XX +XXX,XX @@ bool virtio_scsi_handle_ctrl_vq(VirtIOSCSI *s, VirtQueue *vq)
369
VirtIOSCSIReq *req;
370
bool progress = false;
371
372
+ virtio_scsi_acquire(s);
373
while ((req = virtio_scsi_pop_req(s, vq))) {
374
progress = true;
375
virtio_scsi_handle_ctrl_req(s, req);
376
}
377
+ virtio_scsi_release(s);
378
return progress;
379
}
380
381
@@ -XXX,XX +XXX,XX @@ bool virtio_scsi_handle_cmd_vq(VirtIOSCSI *s, VirtQueue *vq)
382
383
QTAILQ_HEAD(, VirtIOSCSIReq) reqs = QTAILQ_HEAD_INITIALIZER(reqs);
384
385
+ virtio_scsi_acquire(s);
386
do {
387
virtio_queue_set_notification(vq, 0);
388
389
@@ -XXX,XX +XXX,XX @@ bool virtio_scsi_handle_cmd_vq(VirtIOSCSI *s, VirtQueue *vq)
390
QTAILQ_FOREACH_SAFE(req, &reqs, next, next) {
391
virtio_scsi_handle_cmd_req_submit(s, req);
392
}
393
+ virtio_scsi_release(s);
394
return progress;
395
}
396
397
@@ -XXX,XX +XXX,XX @@ out:
398
399
bool virtio_scsi_handle_event_vq(VirtIOSCSI *s, VirtQueue *vq)
400
{
401
+ virtio_scsi_acquire(s);
402
if (s->events_dropped) {
403
virtio_scsi_push_event(s, NULL, VIRTIO_SCSI_T_NO_EVENT, 0);
404
+ virtio_scsi_release(s);
405
return true;
406
}
407
+ virtio_scsi_release(s);
408
return false;
409
}
410
411
diff --git a/util/aio-posix.c b/util/aio-posix.c
412
index XXXXXXX..XXXXXXX 100644
413
--- a/util/aio-posix.c
414
+++ b/util/aio-posix.c
415
@@ -XXX,XX +XXX,XX @@ static bool aio_dispatch_handlers(AioContext *ctx)
416
(revents & (G_IO_IN | G_IO_HUP | G_IO_ERR)) &&
417
aio_node_check(ctx, node->is_external) &&
418
node->io_read) {
419
- aio_context_acquire(ctx);
420
node->io_read(node->opaque);
421
- aio_context_release(ctx);
422
423
/* aio_notify() does not count as progress */
424
if (node->opaque != &ctx->notifier) {
425
@@ -XXX,XX +XXX,XX @@ static bool aio_dispatch_handlers(AioContext *ctx)
426
(revents & (G_IO_OUT | G_IO_ERR)) &&
427
aio_node_check(ctx, node->is_external) &&
428
node->io_write) {
429
- aio_context_acquire(ctx);
430
node->io_write(node->opaque);
431
- aio_context_release(ctx);
432
progress = true;
433
}
434
435
@@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking)
436
start = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
437
}
438
439
- aio_context_acquire(ctx);
440
progress = try_poll_mode(ctx, blocking);
441
- aio_context_release(ctx);
442
-
443
if (!progress) {
444
assert(npfd == 0);
445
446
diff --git a/util/aio-win32.c b/util/aio-win32.c
447
index XXXXXXX..XXXXXXX 100644
448
--- a/util/aio-win32.c
449
+++ b/util/aio-win32.c
450
@@ -XXX,XX +XXX,XX @@ static bool aio_dispatch_handlers(AioContext *ctx, HANDLE event)
451
(revents || event_notifier_get_handle(node->e) == event) &&
452
node->io_notify) {
453
node->pfd.revents = 0;
454
- aio_context_acquire(ctx);
455
node->io_notify(node->e);
456
- aio_context_release(ctx);
457
458
/* aio_notify() does not count as progress */
459
if (node->e != &ctx->notifier) {
460
@@ -XXX,XX +XXX,XX @@ static bool aio_dispatch_handlers(AioContext *ctx, HANDLE event)
461
(node->io_read || node->io_write)) {
462
node->pfd.revents = 0;
463
if ((revents & G_IO_IN) && node->io_read) {
464
- aio_context_acquire(ctx);
465
node->io_read(node->opaque);
466
- aio_context_release(ctx);
467
progress = true;
468
}
469
if ((revents & G_IO_OUT) && node->io_write) {
470
- aio_context_acquire(ctx);
471
node->io_write(node->opaque);
472
- aio_context_release(ctx);
473
progress = true;
474
}
475
476
--
477
2.9.3
478
479
diff view generated by jsdifflib
New patch
1
From: Paolo Bonzini <pbonzini@redhat.com>
1
2
3
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
4
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5
Reviewed-by: Fam Zheng <famz@redhat.com>
6
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
7
Message-id: 20170213135235.12274-15-pbonzini@redhat.com
8
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
9
---
10
block/archipelago.c | 3 +++
11
block/blkreplay.c | 2 +-
12
block/block-backend.c | 6 ++++++
13
block/curl.c | 26 ++++++++++++++++++--------
14
block/gluster.c | 9 +--------
15
block/io.c | 6 +++++-
16
block/iscsi.c | 6 +++++-
17
block/linux-aio.c | 15 +++++++++------
18
block/nfs.c | 3 ++-
19
block/null.c | 4 ++++
20
block/qed.c | 3 +++
21
block/rbd.c | 4 ++++
22
dma-helpers.c | 2 ++
23
hw/block/virtio-blk.c | 2 ++
24
hw/scsi/scsi-bus.c | 2 ++
25
util/async.c | 4 ++--
26
util/thread-pool.c | 2 ++
27
17 files changed, 71 insertions(+), 28 deletions(-)
28
29
diff --git a/block/archipelago.c b/block/archipelago.c
30
index XXXXXXX..XXXXXXX 100644
31
--- a/block/archipelago.c
32
+++ b/block/archipelago.c
33
@@ -XXX,XX +XXX,XX @@ static void qemu_archipelago_complete_aio(void *opaque)
34
{
35
AIORequestData *reqdata = (AIORequestData *) opaque;
36
ArchipelagoAIOCB *aio_cb = (ArchipelagoAIOCB *) reqdata->aio_cb;
37
+ AioContext *ctx = bdrv_get_aio_context(aio_cb->common.bs);
38
39
+ aio_context_acquire(ctx);
40
aio_cb->common.cb(aio_cb->common.opaque, aio_cb->ret);
41
+ aio_context_release(ctx);
42
aio_cb->status = 0;
43
44
qemu_aio_unref(aio_cb);
45
diff --git a/block/blkreplay.c b/block/blkreplay.c
46
index XXXXXXX..XXXXXXX 100755
47
--- a/block/blkreplay.c
48
+++ b/block/blkreplay.c
49
@@ -XXX,XX +XXX,XX @@ static int64_t blkreplay_getlength(BlockDriverState *bs)
50
static void blkreplay_bh_cb(void *opaque)
51
{
52
Request *req = opaque;
53
- qemu_coroutine_enter(req->co);
54
+ aio_co_wake(req->co);
55
qemu_bh_delete(req->bh);
56
g_free(req);
57
}
58
diff --git a/block/block-backend.c b/block/block-backend.c
59
index XXXXXXX..XXXXXXX 100644
60
--- a/block/block-backend.c
61
+++ b/block/block-backend.c
62
@@ -XXX,XX +XXX,XX @@ int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags)
63
static void error_callback_bh(void *opaque)
64
{
65
struct BlockBackendAIOCB *acb = opaque;
66
+ AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
67
68
bdrv_dec_in_flight(acb->common.bs);
69
+ aio_context_acquire(ctx);
70
acb->common.cb(acb->common.opaque, acb->ret);
71
+ aio_context_release(ctx);
72
qemu_aio_unref(acb);
73
}
74
75
@@ -XXX,XX +XXX,XX @@ static void blk_aio_complete(BlkAioEmAIOCB *acb)
76
static void blk_aio_complete_bh(void *opaque)
77
{
78
BlkAioEmAIOCB *acb = opaque;
79
+ AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
80
81
assert(acb->has_returned);
82
+ aio_context_acquire(ctx);
83
blk_aio_complete(acb);
84
+ aio_context_release(ctx);
85
}
86
87
static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes,
88
diff --git a/block/curl.c b/block/curl.c
89
index XXXXXXX..XXXXXXX 100644
90
--- a/block/curl.c
91
+++ b/block/curl.c
92
@@ -XXX,XX +XXX,XX @@ static void curl_readv_bh_cb(void *p)
93
{
94
CURLState *state;
95
int running;
96
+ int ret = -EINPROGRESS;
97
98
CURLAIOCB *acb = p;
99
- BDRVCURLState *s = acb->common.bs->opaque;
100
+ BlockDriverState *bs = acb->common.bs;
101
+ BDRVCURLState *s = bs->opaque;
102
+ AioContext *ctx = bdrv_get_aio_context(bs);
103
104
size_t start = acb->sector_num * BDRV_SECTOR_SIZE;
105
size_t end;
106
107
+ aio_context_acquire(ctx);
108
+
109
// In case we have the requested data already (e.g. read-ahead),
110
// we can just call the callback and be done.
111
switch (curl_find_buf(s, start, acb->nb_sectors * BDRV_SECTOR_SIZE, acb)) {
112
@@ -XXX,XX +XXX,XX @@ static void curl_readv_bh_cb(void *p)
113
qemu_aio_unref(acb);
114
// fall through
115
case FIND_RET_WAIT:
116
- return;
117
+ goto out;
118
default:
119
break;
120
}
121
@@ -XXX,XX +XXX,XX @@ static void curl_readv_bh_cb(void *p)
122
// No cache found, so let's start a new request
123
state = curl_init_state(acb->common.bs, s);
124
if (!state) {
125
- acb->common.cb(acb->common.opaque, -EIO);
126
- qemu_aio_unref(acb);
127
- return;
128
+ ret = -EIO;
129
+ goto out;
130
}
131
132
acb->start = 0;
133
@@ -XXX,XX +XXX,XX @@ static void curl_readv_bh_cb(void *p)
134
state->orig_buf = g_try_malloc(state->buf_len);
135
if (state->buf_len && state->orig_buf == NULL) {
136
curl_clean_state(state);
137
- acb->common.cb(acb->common.opaque, -ENOMEM);
138
- qemu_aio_unref(acb);
139
- return;
140
+ ret = -ENOMEM;
141
+ goto out;
142
}
143
state->acb[0] = acb;
144
145
@@ -XXX,XX +XXX,XX @@ static void curl_readv_bh_cb(void *p)
146
147
/* Tell curl it needs to kick things off */
148
curl_multi_socket_action(s->multi, CURL_SOCKET_TIMEOUT, 0, &running);
149
+
150
+out:
151
+ if (ret != -EINPROGRESS) {
152
+ acb->common.cb(acb->common.opaque, ret);
153
+ qemu_aio_unref(acb);
154
+ }
155
+ aio_context_release(ctx);
156
}
157
158
static BlockAIOCB *curl_aio_readv(BlockDriverState *bs,
159
diff --git a/block/gluster.c b/block/gluster.c
160
index XXXXXXX..XXXXXXX 100644
161
--- a/block/gluster.c
162
+++ b/block/gluster.c
163
@@ -XXX,XX +XXX,XX @@ static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf,
164
return qemu_gluster_glfs_init(gconf, errp);
165
}
166
167
-static void qemu_gluster_complete_aio(void *opaque)
168
-{
169
- GlusterAIOCB *acb = (GlusterAIOCB *)opaque;
170
-
171
- qemu_coroutine_enter(acb->coroutine);
172
-}
173
-
174
/*
175
* AIO callback routine called from GlusterFS thread.
176
*/
177
@@ -XXX,XX +XXX,XX @@ static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, void *arg)
178
acb->ret = -EIO; /* Partial read/write - fail it */
179
}
180
181
- aio_bh_schedule_oneshot(acb->aio_context, qemu_gluster_complete_aio, acb);
182
+ aio_co_schedule(acb->aio_context, acb->coroutine);
183
}
184
185
static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags)
186
diff --git a/block/io.c b/block/io.c
187
index XXXXXXX..XXXXXXX 100644
188
--- a/block/io.c
189
+++ b/block/io.c
190
@@ -XXX,XX +XXX,XX @@ static void bdrv_co_drain_bh_cb(void *opaque)
191
bdrv_dec_in_flight(bs);
192
bdrv_drained_begin(bs);
193
data->done = true;
194
- qemu_coroutine_enter(co);
195
+ aio_co_wake(co);
196
}
197
198
static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs)
199
@@ -XXX,XX +XXX,XX @@ static void bdrv_co_complete(BlockAIOCBCoroutine *acb)
200
static void bdrv_co_em_bh(void *opaque)
201
{
202
BlockAIOCBCoroutine *acb = opaque;
203
+ BlockDriverState *bs = acb->common.bs;
204
+ AioContext *ctx = bdrv_get_aio_context(bs);
205
206
assert(!acb->need_bh);
207
+ aio_context_acquire(ctx);
208
bdrv_co_complete(acb);
209
+ aio_context_release(ctx);
210
}
211
212
static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine *acb)
213
diff --git a/block/iscsi.c b/block/iscsi.c
214
index XXXXXXX..XXXXXXX 100644
215
--- a/block/iscsi.c
216
+++ b/block/iscsi.c
217
@@ -XXX,XX +XXX,XX @@ static void
218
iscsi_bh_cb(void *p)
219
{
220
IscsiAIOCB *acb = p;
221
+ AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
222
223
qemu_bh_delete(acb->bh);
224
225
g_free(acb->buf);
226
acb->buf = NULL;
227
228
+ aio_context_acquire(ctx);
229
acb->common.cb(acb->common.opaque, acb->status);
230
+ aio_context_release(ctx);
231
232
if (acb->task != NULL) {
233
scsi_free_scsi_task(acb->task);
234
@@ -XXX,XX +XXX,XX @@ iscsi_schedule_bh(IscsiAIOCB *acb)
235
static void iscsi_co_generic_bh_cb(void *opaque)
236
{
237
struct IscsiTask *iTask = opaque;
238
+
239
iTask->complete = 1;
240
- qemu_coroutine_enter(iTask->co);
241
+ aio_co_wake(iTask->co);
242
}
243
244
static void iscsi_retry_timer_expired(void *opaque)
245
diff --git a/block/linux-aio.c b/block/linux-aio.c
246
index XXXXXXX..XXXXXXX 100644
247
--- a/block/linux-aio.c
248
+++ b/block/linux-aio.c
249
@@ -XXX,XX +XXX,XX @@ struct LinuxAioState {
250
io_context_t ctx;
251
EventNotifier e;
252
253
- /* io queue for submit at batch */
254
+ /* io queue for submit at batch. Protected by AioContext lock. */
255
LaioQueue io_q;
256
257
- /* I/O completion processing */
258
+ /* I/O completion processing. Only runs in I/O thread. */
259
QEMUBH *completion_bh;
260
int event_idx;
261
int event_max;
262
@@ -XXX,XX +XXX,XX @@ static inline ssize_t io_event_ret(struct io_event *ev)
263
*/
264
static void qemu_laio_process_completion(struct qemu_laiocb *laiocb)
265
{
266
+ LinuxAioState *s = laiocb->ctx;
267
int ret;
268
269
ret = laiocb->ret;
270
@@ -XXX,XX +XXX,XX @@ static void qemu_laio_process_completion(struct qemu_laiocb *laiocb)
271
}
272
273
laiocb->ret = ret;
274
+ aio_context_acquire(s->aio_context);
275
if (laiocb->co) {
276
/* If the coroutine is already entered it must be in ioq_submit() and
277
* will notice laio->ret has been filled in when it eventually runs
278
@@ -XXX,XX +XXX,XX @@ static void qemu_laio_process_completion(struct qemu_laiocb *laiocb)
279
laiocb->common.cb(laiocb->common.opaque, ret);
280
qemu_aio_unref(laiocb);
281
}
282
+ aio_context_release(s->aio_context);
283
}
284
285
/**
286
@@ -XXX,XX +XXX,XX @@ static void qemu_laio_process_completions(LinuxAioState *s)
287
static void qemu_laio_process_completions_and_submit(LinuxAioState *s)
288
{
289
qemu_laio_process_completions(s);
290
+
291
+ aio_context_acquire(s->aio_context);
292
if (!s->io_q.plugged && !QSIMPLEQ_EMPTY(&s->io_q.pending)) {
293
ioq_submit(s);
294
}
295
+ aio_context_release(s->aio_context);
296
}
297
298
static void qemu_laio_completion_bh(void *opaque)
299
@@ -XXX,XX +XXX,XX @@ static void qemu_laio_completion_cb(EventNotifier *e)
300
LinuxAioState *s = container_of(e, LinuxAioState, e);
301
302
if (event_notifier_test_and_clear(&s->e)) {
303
- aio_context_acquire(s->aio_context);
304
qemu_laio_process_completions_and_submit(s);
305
- aio_context_release(s->aio_context);
306
}
307
}
308
309
@@ -XXX,XX +XXX,XX @@ static bool qemu_laio_poll_cb(void *opaque)
310
return false;
311
}
312
313
- aio_context_acquire(s->aio_context);
314
qemu_laio_process_completions_and_submit(s);
315
- aio_context_release(s->aio_context);
316
return true;
317
}
318
319
@@ -XXX,XX +XXX,XX @@ void laio_detach_aio_context(LinuxAioState *s, AioContext *old_context)
320
{
321
aio_set_event_notifier(old_context, &s->e, false, NULL, NULL);
322
qemu_bh_delete(s->completion_bh);
323
+ s->aio_context = NULL;
324
}
325
326
void laio_attach_aio_context(LinuxAioState *s, AioContext *new_context)
327
diff --git a/block/nfs.c b/block/nfs.c
328
index XXXXXXX..XXXXXXX 100644
329
--- a/block/nfs.c
330
+++ b/block/nfs.c
331
@@ -XXX,XX +XXX,XX @@ static void nfs_co_init_task(BlockDriverState *bs, NFSRPC *task)
332
static void nfs_co_generic_bh_cb(void *opaque)
333
{
334
NFSRPC *task = opaque;
335
+
336
task->complete = 1;
337
- qemu_coroutine_enter(task->co);
338
+ aio_co_wake(task->co);
339
}
340
341
static void
342
diff --git a/block/null.c b/block/null.c
343
index XXXXXXX..XXXXXXX 100644
344
--- a/block/null.c
345
+++ b/block/null.c
346
@@ -XXX,XX +XXX,XX @@ static const AIOCBInfo null_aiocb_info = {
347
static void null_bh_cb(void *opaque)
348
{
349
NullAIOCB *acb = opaque;
350
+ AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
351
+
352
+ aio_context_acquire(ctx);
353
acb->common.cb(acb->common.opaque, 0);
354
+ aio_context_release(ctx);
355
qemu_aio_unref(acb);
356
}
357
358
diff --git a/block/qed.c b/block/qed.c
359
index XXXXXXX..XXXXXXX 100644
360
--- a/block/qed.c
361
+++ b/block/qed.c
362
@@ -XXX,XX +XXX,XX @@ static void qed_update_l2_table(BDRVQEDState *s, QEDTable *table, int index,
363
static void qed_aio_complete_bh(void *opaque)
364
{
365
QEDAIOCB *acb = opaque;
366
+ BDRVQEDState *s = acb_to_s(acb);
367
BlockCompletionFunc *cb = acb->common.cb;
368
void *user_opaque = acb->common.opaque;
369
int ret = acb->bh_ret;
370
@@ -XXX,XX +XXX,XX @@ static void qed_aio_complete_bh(void *opaque)
371
qemu_aio_unref(acb);
372
373
/* Invoke callback */
374
+ qed_acquire(s);
375
cb(user_opaque, ret);
376
+ qed_release(s);
377
}
378
379
static void qed_aio_complete(QEDAIOCB *acb, int ret)
380
diff --git a/block/rbd.c b/block/rbd.c
381
index XXXXXXX..XXXXXXX 100644
382
--- a/block/rbd.c
383
+++ b/block/rbd.c
384
@@ -XXX,XX +XXX,XX @@ shutdown:
385
static void qemu_rbd_complete_aio(RADOSCB *rcb)
386
{
387
RBDAIOCB *acb = rcb->acb;
388
+ AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
389
int64_t r;
390
391
r = rcb->ret;
392
@@ -XXX,XX +XXX,XX @@ static void qemu_rbd_complete_aio(RADOSCB *rcb)
393
qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
394
}
395
qemu_vfree(acb->bounce);
396
+
397
+ aio_context_acquire(ctx);
398
acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
399
+ aio_context_release(ctx);
400
401
qemu_aio_unref(acb);
402
}
403
diff --git a/dma-helpers.c b/dma-helpers.c
404
index XXXXXXX..XXXXXXX 100644
405
--- a/dma-helpers.c
406
+++ b/dma-helpers.c
407
@@ -XXX,XX +XXX,XX @@ static void dma_blk_cb(void *opaque, int ret)
408
QEMU_ALIGN_DOWN(dbs->iov.size, dbs->align));
409
}
410
411
+ aio_context_acquire(dbs->ctx);
412
dbs->acb = dbs->io_func(dbs->offset, &dbs->iov,
413
dma_blk_cb, dbs, dbs->io_func_opaque);
414
+ aio_context_release(dbs->ctx);
415
assert(dbs->acb);
416
}
417
418
diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c
419
index XXXXXXX..XXXXXXX 100644
420
--- a/hw/block/virtio-blk.c
421
+++ b/hw/block/virtio-blk.c
422
@@ -XXX,XX +XXX,XX @@ static void virtio_blk_dma_restart_bh(void *opaque)
423
424
s->rq = NULL;
425
426
+ aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
427
while (req) {
428
VirtIOBlockReq *next = req->next;
429
if (virtio_blk_handle_request(req, &mrb)) {
430
@@ -XXX,XX +XXX,XX @@ static void virtio_blk_dma_restart_bh(void *opaque)
431
if (mrb.num_reqs) {
432
virtio_blk_submit_multireq(s->blk, &mrb);
433
}
434
+ aio_context_release(blk_get_aio_context(s->conf.conf.blk));
435
}
436
437
static void virtio_blk_dma_restart_cb(void *opaque, int running,
438
diff --git a/hw/scsi/scsi-bus.c b/hw/scsi/scsi-bus.c
439
index XXXXXXX..XXXXXXX 100644
440
--- a/hw/scsi/scsi-bus.c
441
+++ b/hw/scsi/scsi-bus.c
442
@@ -XXX,XX +XXX,XX @@ static void scsi_dma_restart_bh(void *opaque)
443
qemu_bh_delete(s->bh);
444
s->bh = NULL;
445
446
+ aio_context_acquire(blk_get_aio_context(s->conf.blk));
447
QTAILQ_FOREACH_SAFE(req, &s->requests, next, next) {
448
scsi_req_ref(req);
449
if (req->retry) {
450
@@ -XXX,XX +XXX,XX @@ static void scsi_dma_restart_bh(void *opaque)
451
}
452
scsi_req_unref(req);
453
}
454
+ aio_context_release(blk_get_aio_context(s->conf.blk));
455
}
456
457
void scsi_req_retry(SCSIRequest *req)
458
diff --git a/util/async.c b/util/async.c
459
index XXXXXXX..XXXXXXX 100644
460
--- a/util/async.c
461
+++ b/util/async.c
462
@@ -XXX,XX +XXX,XX @@ int aio_bh_poll(AioContext *ctx)
463
ret = 1;
464
}
465
bh->idle = 0;
466
- aio_context_acquire(ctx);
467
aio_bh_call(bh);
468
- aio_context_release(ctx);
469
}
470
if (bh->deleted) {
471
deleted = true;
472
@@ -XXX,XX +XXX,XX @@ static void co_schedule_bh_cb(void *opaque)
473
Coroutine *co = QSLIST_FIRST(&straight);
474
QSLIST_REMOVE_HEAD(&straight, co_scheduled_next);
475
trace_aio_co_schedule_bh_cb(ctx, co);
476
+ aio_context_acquire(ctx);
477
qemu_coroutine_enter(co);
478
+ aio_context_release(ctx);
479
}
480
}
481
482
diff --git a/util/thread-pool.c b/util/thread-pool.c
483
index XXXXXXX..XXXXXXX 100644
484
--- a/util/thread-pool.c
485
+++ b/util/thread-pool.c
486
@@ -XXX,XX +XXX,XX @@ static void thread_pool_completion_bh(void *opaque)
487
ThreadPool *pool = opaque;
488
ThreadPoolElement *elem, *next;
489
490
+ aio_context_acquire(pool->ctx);
491
restart:
492
QLIST_FOREACH_SAFE(elem, &pool->head, all, next) {
493
if (elem->state != THREAD_DONE) {
494
@@ -XXX,XX +XXX,XX @@ restart:
495
qemu_aio_unref(elem);
496
}
497
}
498
+ aio_context_release(pool->ctx);
499
}
500
501
static void thread_pool_cancel(BlockAIOCB *acb)
502
--
503
2.9.3
504
505
diff view generated by jsdifflib
New patch
1
From: Paolo Bonzini <pbonzini@redhat.com>
1
2
3
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
4
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5
Reviewed-by: Fam Zheng <famz@redhat.com>
6
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
7
Message-id: 20170213135235.12274-16-pbonzini@redhat.com
8
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
9
---
10
block/archipelago.c | 3 ---
11
block/block-backend.c | 7 -------
12
block/curl.c | 2 +-
13
block/io.c | 6 +-----
14
block/iscsi.c | 3 ---
15
block/linux-aio.c | 5 +----
16
block/mirror.c | 12 +++++++++---
17
block/null.c | 8 --------
18
block/qed-cluster.c | 2 ++
19
block/qed-table.c | 12 ++++++++++--
20
block/qed.c | 4 ++--
21
block/rbd.c | 4 ----
22
block/win32-aio.c | 3 ---
23
hw/block/virtio-blk.c | 12 +++++++++++-
24
hw/scsi/scsi-disk.c | 15 +++++++++++++++
25
hw/scsi/scsi-generic.c | 20 +++++++++++++++++---
26
util/thread-pool.c | 4 +++-
27
17 files changed, 72 insertions(+), 50 deletions(-)
28
29
diff --git a/block/archipelago.c b/block/archipelago.c
30
index XXXXXXX..XXXXXXX 100644
31
--- a/block/archipelago.c
32
+++ b/block/archipelago.c
33
@@ -XXX,XX +XXX,XX @@ static void qemu_archipelago_complete_aio(void *opaque)
34
{
35
AIORequestData *reqdata = (AIORequestData *) opaque;
36
ArchipelagoAIOCB *aio_cb = (ArchipelagoAIOCB *) reqdata->aio_cb;
37
- AioContext *ctx = bdrv_get_aio_context(aio_cb->common.bs);
38
39
- aio_context_acquire(ctx);
40
aio_cb->common.cb(aio_cb->common.opaque, aio_cb->ret);
41
- aio_context_release(ctx);
42
aio_cb->status = 0;
43
44
qemu_aio_unref(aio_cb);
45
diff --git a/block/block-backend.c b/block/block-backend.c
46
index XXXXXXX..XXXXXXX 100644
47
--- a/block/block-backend.c
48
+++ b/block/block-backend.c
49
@@ -XXX,XX +XXX,XX @@ int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags)
50
static void error_callback_bh(void *opaque)
51
{
52
struct BlockBackendAIOCB *acb = opaque;
53
- AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
54
55
bdrv_dec_in_flight(acb->common.bs);
56
- aio_context_acquire(ctx);
57
acb->common.cb(acb->common.opaque, acb->ret);
58
- aio_context_release(ctx);
59
qemu_aio_unref(acb);
60
}
61
62
@@ -XXX,XX +XXX,XX @@ static void blk_aio_complete(BlkAioEmAIOCB *acb)
63
static void blk_aio_complete_bh(void *opaque)
64
{
65
BlkAioEmAIOCB *acb = opaque;
66
- AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
67
-
68
assert(acb->has_returned);
69
- aio_context_acquire(ctx);
70
blk_aio_complete(acb);
71
- aio_context_release(ctx);
72
}
73
74
static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes,
75
diff --git a/block/curl.c b/block/curl.c
76
index XXXXXXX..XXXXXXX 100644
77
--- a/block/curl.c
78
+++ b/block/curl.c
79
@@ -XXX,XX +XXX,XX @@ static void curl_readv_bh_cb(void *p)
80
curl_multi_socket_action(s->multi, CURL_SOCKET_TIMEOUT, 0, &running);
81
82
out:
83
+ aio_context_release(ctx);
84
if (ret != -EINPROGRESS) {
85
acb->common.cb(acb->common.opaque, ret);
86
qemu_aio_unref(acb);
87
}
88
- aio_context_release(ctx);
89
}
90
91
static BlockAIOCB *curl_aio_readv(BlockDriverState *bs,
92
diff --git a/block/io.c b/block/io.c
93
index XXXXXXX..XXXXXXX 100644
94
--- a/block/io.c
95
+++ b/block/io.c
96
@@ -XXX,XX +XXX,XX @@ static void bdrv_co_io_em_complete(void *opaque, int ret)
97
CoroutineIOCompletion *co = opaque;
98
99
co->ret = ret;
100
- qemu_coroutine_enter(co->coroutine);
101
+ aio_co_wake(co->coroutine);
102
}
103
104
static int coroutine_fn bdrv_driver_preadv(BlockDriverState *bs,
105
@@ -XXX,XX +XXX,XX @@ static void bdrv_co_complete(BlockAIOCBCoroutine *acb)
106
static void bdrv_co_em_bh(void *opaque)
107
{
108
BlockAIOCBCoroutine *acb = opaque;
109
- BlockDriverState *bs = acb->common.bs;
110
- AioContext *ctx = bdrv_get_aio_context(bs);
111
112
assert(!acb->need_bh);
113
- aio_context_acquire(ctx);
114
bdrv_co_complete(acb);
115
- aio_context_release(ctx);
116
}
117
118
static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine *acb)
119
diff --git a/block/iscsi.c b/block/iscsi.c
120
index XXXXXXX..XXXXXXX 100644
121
--- a/block/iscsi.c
122
+++ b/block/iscsi.c
123
@@ -XXX,XX +XXX,XX @@ static void
124
iscsi_bh_cb(void *p)
125
{
126
IscsiAIOCB *acb = p;
127
- AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
128
129
qemu_bh_delete(acb->bh);
130
131
g_free(acb->buf);
132
acb->buf = NULL;
133
134
- aio_context_acquire(ctx);
135
acb->common.cb(acb->common.opaque, acb->status);
136
- aio_context_release(ctx);
137
138
if (acb->task != NULL) {
139
scsi_free_scsi_task(acb->task);
140
diff --git a/block/linux-aio.c b/block/linux-aio.c
141
index XXXXXXX..XXXXXXX 100644
142
--- a/block/linux-aio.c
143
+++ b/block/linux-aio.c
144
@@ -XXX,XX +XXX,XX @@ static inline ssize_t io_event_ret(struct io_event *ev)
145
*/
146
static void qemu_laio_process_completion(struct qemu_laiocb *laiocb)
147
{
148
- LinuxAioState *s = laiocb->ctx;
149
int ret;
150
151
ret = laiocb->ret;
152
@@ -XXX,XX +XXX,XX @@ static void qemu_laio_process_completion(struct qemu_laiocb *laiocb)
153
}
154
155
laiocb->ret = ret;
156
- aio_context_acquire(s->aio_context);
157
if (laiocb->co) {
158
/* If the coroutine is already entered it must be in ioq_submit() and
159
* will notice laio->ret has been filled in when it eventually runs
160
@@ -XXX,XX +XXX,XX @@ static void qemu_laio_process_completion(struct qemu_laiocb *laiocb)
161
* that!
162
*/
163
if (!qemu_coroutine_entered(laiocb->co)) {
164
- qemu_coroutine_enter(laiocb->co);
165
+ aio_co_wake(laiocb->co);
166
}
167
} else {
168
laiocb->common.cb(laiocb->common.opaque, ret);
169
qemu_aio_unref(laiocb);
170
}
171
- aio_context_release(s->aio_context);
172
}
173
174
/**
175
diff --git a/block/mirror.c b/block/mirror.c
176
index XXXXXXX..XXXXXXX 100644
177
--- a/block/mirror.c
178
+++ b/block/mirror.c
179
@@ -XXX,XX +XXX,XX @@ static void mirror_write_complete(void *opaque, int ret)
180
{
181
MirrorOp *op = opaque;
182
MirrorBlockJob *s = op->s;
183
+
184
+ aio_context_acquire(blk_get_aio_context(s->common.blk));
185
if (ret < 0) {
186
BlockErrorAction action;
187
188
@@ -XXX,XX +XXX,XX @@ static void mirror_write_complete(void *opaque, int ret)
189
}
190
}
191
mirror_iteration_done(op, ret);
192
+ aio_context_release(blk_get_aio_context(s->common.blk));
193
}
194
195
static void mirror_read_complete(void *opaque, int ret)
196
{
197
MirrorOp *op = opaque;
198
MirrorBlockJob *s = op->s;
199
+
200
+ aio_context_acquire(blk_get_aio_context(s->common.blk));
201
if (ret < 0) {
202
BlockErrorAction action;
203
204
@@ -XXX,XX +XXX,XX @@ static void mirror_read_complete(void *opaque, int ret)
205
}
206
207
mirror_iteration_done(op, ret);
208
- return;
209
+ } else {
210
+ blk_aio_pwritev(s->target, op->sector_num * BDRV_SECTOR_SIZE, &op->qiov,
211
+ 0, mirror_write_complete, op);
212
}
213
- blk_aio_pwritev(s->target, op->sector_num * BDRV_SECTOR_SIZE, &op->qiov,
214
- 0, mirror_write_complete, op);
215
+ aio_context_release(blk_get_aio_context(s->common.blk));
216
}
217
218
static inline void mirror_clip_sectors(MirrorBlockJob *s,
219
diff --git a/block/null.c b/block/null.c
220
index XXXXXXX..XXXXXXX 100644
221
--- a/block/null.c
222
+++ b/block/null.c
223
@@ -XXX,XX +XXX,XX @@ static const AIOCBInfo null_aiocb_info = {
224
static void null_bh_cb(void *opaque)
225
{
226
NullAIOCB *acb = opaque;
227
- AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
228
-
229
- aio_context_acquire(ctx);
230
acb->common.cb(acb->common.opaque, 0);
231
- aio_context_release(ctx);
232
qemu_aio_unref(acb);
233
}
234
235
static void null_timer_cb(void *opaque)
236
{
237
NullAIOCB *acb = opaque;
238
- AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
239
-
240
- aio_context_acquire(ctx);
241
acb->common.cb(acb->common.opaque, 0);
242
- aio_context_release(ctx);
243
timer_deinit(&acb->timer);
244
qemu_aio_unref(acb);
245
}
246
diff --git a/block/qed-cluster.c b/block/qed-cluster.c
247
index XXXXXXX..XXXXXXX 100644
248
--- a/block/qed-cluster.c
249
+++ b/block/qed-cluster.c
250
@@ -XXX,XX +XXX,XX @@ static void qed_find_cluster_cb(void *opaque, int ret)
251
unsigned int index;
252
unsigned int n;
253
254
+ qed_acquire(s);
255
if (ret) {
256
goto out;
257
}
258
@@ -XXX,XX +XXX,XX @@ static void qed_find_cluster_cb(void *opaque, int ret)
259
260
out:
261
find_cluster_cb->cb(find_cluster_cb->opaque, ret, offset, len);
262
+ qed_release(s);
263
g_free(find_cluster_cb);
264
}
265
266
diff --git a/block/qed-table.c b/block/qed-table.c
267
index XXXXXXX..XXXXXXX 100644
268
--- a/block/qed-table.c
269
+++ b/block/qed-table.c
270
@@ -XXX,XX +XXX,XX @@ static void qed_read_table_cb(void *opaque, int ret)
271
{
272
QEDReadTableCB *read_table_cb = opaque;
273
QEDTable *table = read_table_cb->table;
274
+ BDRVQEDState *s = read_table_cb->s;
275
int noffsets = read_table_cb->qiov.size / sizeof(uint64_t);
276
int i;
277
278
@@ -XXX,XX +XXX,XX @@ static void qed_read_table_cb(void *opaque, int ret)
279
}
280
281
/* Byteswap offsets */
282
+ qed_acquire(s);
283
for (i = 0; i < noffsets; i++) {
284
table->offsets[i] = le64_to_cpu(table->offsets[i]);
285
}
286
+ qed_release(s);
287
288
out:
289
/* Completion */
290
- trace_qed_read_table_cb(read_table_cb->s, read_table_cb->table, ret);
291
+ trace_qed_read_table_cb(s, read_table_cb->table, ret);
292
gencb_complete(&read_table_cb->gencb, ret);
293
}
294
295
@@ -XXX,XX +XXX,XX @@ typedef struct {
296
static void qed_write_table_cb(void *opaque, int ret)
297
{
298
QEDWriteTableCB *write_table_cb = opaque;
299
+ BDRVQEDState *s = write_table_cb->s;
300
301
- trace_qed_write_table_cb(write_table_cb->s,
302
+ trace_qed_write_table_cb(s,
303
write_table_cb->orig_table,
304
write_table_cb->flush,
305
ret);
306
@@ -XXX,XX +XXX,XX @@ static void qed_write_table_cb(void *opaque, int ret)
307
if (write_table_cb->flush) {
308
/* We still need to flush first */
309
write_table_cb->flush = false;
310
+ qed_acquire(s);
311
bdrv_aio_flush(write_table_cb->s->bs, qed_write_table_cb,
312
write_table_cb);
313
+ qed_release(s);
314
return;
315
}
316
317
@@ -XXX,XX +XXX,XX @@ static void qed_read_l2_table_cb(void *opaque, int ret)
318
CachedL2Table *l2_table = request->l2_table;
319
uint64_t l2_offset = read_l2_table_cb->l2_offset;
320
321
+ qed_acquire(s);
322
if (ret) {
323
/* can't trust loaded L2 table anymore */
324
qed_unref_l2_cache_entry(l2_table);
325
@@ -XXX,XX +XXX,XX @@ static void qed_read_l2_table_cb(void *opaque, int ret)
326
request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, l2_offset);
327
assert(request->l2_table != NULL);
328
}
329
+ qed_release(s);
330
331
gencb_complete(&read_l2_table_cb->gencb, ret);
332
}
333
diff --git a/block/qed.c b/block/qed.c
334
index XXXXXXX..XXXXXXX 100644
335
--- a/block/qed.c
336
+++ b/block/qed.c
337
@@ -XXX,XX +XXX,XX @@ static void qed_is_allocated_cb(void *opaque, int ret, uint64_t offset, size_t l
338
}
339
340
if (cb->co) {
341
- qemu_coroutine_enter(cb->co);
342
+ aio_co_wake(cb->co);
343
}
344
}
345
346
@@ -XXX,XX +XXX,XX @@ static void coroutine_fn qed_co_pwrite_zeroes_cb(void *opaque, int ret)
347
cb->done = true;
348
cb->ret = ret;
349
if (cb->co) {
350
- qemu_coroutine_enter(cb->co);
351
+ aio_co_wake(cb->co);
352
}
353
}
354
355
diff --git a/block/rbd.c b/block/rbd.c
356
index XXXXXXX..XXXXXXX 100644
357
--- a/block/rbd.c
358
+++ b/block/rbd.c
359
@@ -XXX,XX +XXX,XX @@ shutdown:
360
static void qemu_rbd_complete_aio(RADOSCB *rcb)
361
{
362
RBDAIOCB *acb = rcb->acb;
363
- AioContext *ctx = bdrv_get_aio_context(acb->common.bs);
364
int64_t r;
365
366
r = rcb->ret;
367
@@ -XXX,XX +XXX,XX @@ static void qemu_rbd_complete_aio(RADOSCB *rcb)
368
qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
369
}
370
qemu_vfree(acb->bounce);
371
-
372
- aio_context_acquire(ctx);
373
acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
374
- aio_context_release(ctx);
375
376
qemu_aio_unref(acb);
377
}
378
diff --git a/block/win32-aio.c b/block/win32-aio.c
379
index XXXXXXX..XXXXXXX 100644
380
--- a/block/win32-aio.c
381
+++ b/block/win32-aio.c
382
@@ -XXX,XX +XXX,XX @@ static void win32_aio_process_completion(QEMUWin32AIOState *s,
383
qemu_vfree(waiocb->buf);
384
}
385
386
-
387
- aio_context_acquire(s->aio_ctx);
388
waiocb->common.cb(waiocb->common.opaque, ret);
389
- aio_context_release(s->aio_ctx);
390
qemu_aio_unref(waiocb);
391
}
392
393
diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c
394
index XXXXXXX..XXXXXXX 100644
395
--- a/hw/block/virtio-blk.c
396
+++ b/hw/block/virtio-blk.c
397
@@ -XXX,XX +XXX,XX @@ static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
398
static void virtio_blk_rw_complete(void *opaque, int ret)
399
{
400
VirtIOBlockReq *next = opaque;
401
+ VirtIOBlock *s = next->dev;
402
403
+ aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
404
while (next) {
405
VirtIOBlockReq *req = next;
406
next = req->mr_next;
407
@@ -XXX,XX +XXX,XX @@ static void virtio_blk_rw_complete(void *opaque, int ret)
408
block_acct_done(blk_get_stats(req->dev->blk), &req->acct);
409
virtio_blk_free_request(req);
410
}
411
+ aio_context_release(blk_get_aio_context(s->conf.conf.blk));
412
}
413
414
static void virtio_blk_flush_complete(void *opaque, int ret)
415
{
416
VirtIOBlockReq *req = opaque;
417
+ VirtIOBlock *s = req->dev;
418
419
+ aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
420
if (ret) {
421
if (virtio_blk_handle_rw_error(req, -ret, 0)) {
422
- return;
423
+ goto out;
424
}
425
}
426
427
virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
428
block_acct_done(blk_get_stats(req->dev->blk), &req->acct);
429
virtio_blk_free_request(req);
430
+
431
+out:
432
+ aio_context_release(blk_get_aio_context(s->conf.conf.blk));
433
}
434
435
#ifdef __linux__
436
@@ -XXX,XX +XXX,XX @@ static void virtio_blk_ioctl_complete(void *opaque, int status)
437
virtio_stl_p(vdev, &scsi->data_len, hdr->dxfer_len);
438
439
out:
440
+ aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
441
virtio_blk_req_complete(req, status);
442
virtio_blk_free_request(req);
443
+ aio_context_release(blk_get_aio_context(s->conf.conf.blk));
444
g_free(ioctl_req);
445
}
446
447
diff --git a/hw/scsi/scsi-disk.c b/hw/scsi/scsi-disk.c
448
index XXXXXXX..XXXXXXX 100644
449
--- a/hw/scsi/scsi-disk.c
450
+++ b/hw/scsi/scsi-disk.c
451
@@ -XXX,XX +XXX,XX @@ static void scsi_aio_complete(void *opaque, int ret)
452
453
assert(r->req.aiocb != NULL);
454
r->req.aiocb = NULL;
455
+ aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
456
if (scsi_disk_req_check_error(r, ret, true)) {
457
goto done;
458
}
459
@@ -XXX,XX +XXX,XX @@ static void scsi_aio_complete(void *opaque, int ret)
460
scsi_req_complete(&r->req, GOOD);
461
462
done:
463
+ aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
464
scsi_req_unref(&r->req);
465
}
466
467
@@ -XXX,XX +XXX,XX @@ static void scsi_dma_complete(void *opaque, int ret)
468
assert(r->req.aiocb != NULL);
469
r->req.aiocb = NULL;
470
471
+ aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
472
if (ret < 0) {
473
block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
474
} else {
475
block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
476
}
477
scsi_dma_complete_noio(r, ret);
478
+ aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
479
}
480
481
static void scsi_read_complete(void * opaque, int ret)
482
@@ -XXX,XX +XXX,XX @@ static void scsi_read_complete(void * opaque, int ret)
483
484
assert(r->req.aiocb != NULL);
485
r->req.aiocb = NULL;
486
+ aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
487
if (scsi_disk_req_check_error(r, ret, true)) {
488
goto done;
489
}
490
@@ -XXX,XX +XXX,XX @@ static void scsi_read_complete(void * opaque, int ret)
491
492
done:
493
scsi_req_unref(&r->req);
494
+ aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
495
}
496
497
/* Actually issue a read to the block device. */
498
@@ -XXX,XX +XXX,XX @@ static void scsi_do_read_cb(void *opaque, int ret)
499
assert (r->req.aiocb != NULL);
500
r->req.aiocb = NULL;
501
502
+ aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
503
if (ret < 0) {
504
block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
505
} else {
506
block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
507
}
508
scsi_do_read(opaque, ret);
509
+ aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
510
}
511
512
/* Read more data from scsi device into buffer. */
513
@@ -XXX,XX +XXX,XX @@ static void scsi_write_complete(void * opaque, int ret)
514
assert (r->req.aiocb != NULL);
515
r->req.aiocb = NULL;
516
517
+ aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
518
if (ret < 0) {
519
block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
520
} else {
521
block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
522
}
523
scsi_write_complete_noio(r, ret);
524
+ aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
525
}
526
527
static void scsi_write_data(SCSIRequest *req)
528
@@ -XXX,XX +XXX,XX @@ static void scsi_unmap_complete(void *opaque, int ret)
529
{
530
UnmapCBData *data = opaque;
531
SCSIDiskReq *r = data->r;
532
+ SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
533
534
assert(r->req.aiocb != NULL);
535
r->req.aiocb = NULL;
536
537
+ aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
538
scsi_unmap_complete_noio(data, ret);
539
+ aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
540
}
541
542
static void scsi_disk_emulate_unmap(SCSIDiskReq *r, uint8_t *inbuf)
543
@@ -XXX,XX +XXX,XX @@ static void scsi_write_same_complete(void *opaque, int ret)
544
545
assert(r->req.aiocb != NULL);
546
r->req.aiocb = NULL;
547
+ aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
548
if (scsi_disk_req_check_error(r, ret, true)) {
549
goto done;
550
}
551
@@ -XXX,XX +XXX,XX @@ done:
552
scsi_req_unref(&r->req);
553
qemu_vfree(data->iov.iov_base);
554
g_free(data);
555
+ aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
556
}
557
558
static void scsi_disk_emulate_write_same(SCSIDiskReq *r, uint8_t *inbuf)
559
diff --git a/hw/scsi/scsi-generic.c b/hw/scsi/scsi-generic.c
560
index XXXXXXX..XXXXXXX 100644
561
--- a/hw/scsi/scsi-generic.c
562
+++ b/hw/scsi/scsi-generic.c
563
@@ -XXX,XX +XXX,XX @@ done:
564
static void scsi_command_complete(void *opaque, int ret)
565
{
566
SCSIGenericReq *r = (SCSIGenericReq *)opaque;
567
+ SCSIDevice *s = r->req.dev;
568
569
assert(r->req.aiocb != NULL);
570
r->req.aiocb = NULL;
571
+
572
+ aio_context_acquire(blk_get_aio_context(s->conf.blk));
573
scsi_command_complete_noio(r, ret);
574
+ aio_context_release(blk_get_aio_context(s->conf.blk));
575
}
576
577
static int execute_command(BlockBackend *blk,
578
@@ -XXX,XX +XXX,XX @@ static void scsi_read_complete(void * opaque, int ret)
579
assert(r->req.aiocb != NULL);
580
r->req.aiocb = NULL;
581
582
+ aio_context_acquire(blk_get_aio_context(s->conf.blk));
583
+
584
if (ret || r->req.io_canceled) {
585
scsi_command_complete_noio(r, ret);
586
- return;
587
+ goto done;
588
}
589
590
len = r->io_header.dxfer_len - r->io_header.resid;
591
@@ -XXX,XX +XXX,XX @@ static void scsi_read_complete(void * opaque, int ret)
592
r->len = -1;
593
if (len == 0) {
594
scsi_command_complete_noio(r, 0);
595
- return;
596
+ goto done;
597
}
598
599
/* Snoop READ CAPACITY output to set the blocksize. */
600
@@ -XXX,XX +XXX,XX @@ static void scsi_read_complete(void * opaque, int ret)
601
}
602
scsi_req_data(&r->req, len);
603
scsi_req_unref(&r->req);
604
+
605
+done:
606
+ aio_context_release(blk_get_aio_context(s->conf.blk));
607
}
608
609
/* Read more data from scsi device into buffer. */
610
@@ -XXX,XX +XXX,XX @@ static void scsi_write_complete(void * opaque, int ret)
611
assert(r->req.aiocb != NULL);
612
r->req.aiocb = NULL;
613
614
+ aio_context_acquire(blk_get_aio_context(s->conf.blk));
615
+
616
if (ret || r->req.io_canceled) {
617
scsi_command_complete_noio(r, ret);
618
- return;
619
+ goto done;
620
}
621
622
if (r->req.cmd.buf[0] == MODE_SELECT && r->req.cmd.buf[4] == 12 &&
623
@@ -XXX,XX +XXX,XX @@ static void scsi_write_complete(void * opaque, int ret)
624
}
625
626
scsi_command_complete_noio(r, ret);
627
+
628
+done:
629
+ aio_context_release(blk_get_aio_context(s->conf.blk));
630
}
631
632
/* Write data to a scsi device. Returns nonzero on failure.
633
diff --git a/util/thread-pool.c b/util/thread-pool.c
634
index XXXXXXX..XXXXXXX 100644
635
--- a/util/thread-pool.c
636
+++ b/util/thread-pool.c
637
@@ -XXX,XX +XXX,XX @@ restart:
638
*/
639
qemu_bh_schedule(pool->completion_bh);
640
641
+ aio_context_release(pool->ctx);
642
elem->common.cb(elem->common.opaque, elem->ret);
643
+ aio_context_acquire(pool->ctx);
644
qemu_aio_unref(elem);
645
goto restart;
646
} else {
647
@@ -XXX,XX +XXX,XX @@ static void thread_pool_co_cb(void *opaque, int ret)
648
ThreadPoolCo *co = opaque;
649
650
co->ret = ret;
651
- qemu_coroutine_enter(co->co);
652
+ aio_co_wake(co->co);
653
}
654
655
int coroutine_fn thread_pool_submit_co(ThreadPool *pool, ThreadPoolFunc *func,
656
--
657
2.9.3
658
659
diff view generated by jsdifflib
New patch
1
From: Paolo Bonzini <pbonzini@redhat.com>
1
2
3
This patch prepares for the removal of unnecessary lockcnt inc/dec pairs.
4
Extract the dispatching loop for file descriptor handlers into a new
5
function aio_dispatch_handlers, and then inline aio_dispatch into
6
aio_poll.
7
8
aio_dispatch can now become void.
9
10
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
11
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
12
Reviewed-by: Fam Zheng <famz@redhat.com>
13
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
14
Message-id: 20170213135235.12274-17-pbonzini@redhat.com
15
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
16
---
17
include/block/aio.h | 6 +-----
18
util/aio-posix.c | 44 ++++++++++++++------------------------------
19
util/aio-win32.c | 13 ++++---------
20
util/async.c | 2 +-
21
4 files changed, 20 insertions(+), 45 deletions(-)
22
23
diff --git a/include/block/aio.h b/include/block/aio.h
24
index XXXXXXX..XXXXXXX 100644
25
--- a/include/block/aio.h
26
+++ b/include/block/aio.h
27
@@ -XXX,XX +XXX,XX @@ bool aio_pending(AioContext *ctx);
28
/* Dispatch any pending callbacks from the GSource attached to the AioContext.
29
*
30
* This is used internally in the implementation of the GSource.
31
- *
32
- * @dispatch_fds: true to process fds, false to skip them
33
- * (can be used as an optimization by callers that know there
34
- * are no fds ready)
35
*/
36
-bool aio_dispatch(AioContext *ctx, bool dispatch_fds);
37
+void aio_dispatch(AioContext *ctx);
38
39
/* Progress in completing AIO work to occur. This can issue new pending
40
* aio as a result of executing I/O completion or bh callbacks.
41
diff --git a/util/aio-posix.c b/util/aio-posix.c
42
index XXXXXXX..XXXXXXX 100644
43
--- a/util/aio-posix.c
44
+++ b/util/aio-posix.c
45
@@ -XXX,XX +XXX,XX @@ static bool aio_dispatch_handlers(AioContext *ctx)
46
AioHandler *node, *tmp;
47
bool progress = false;
48
49
- /*
50
- * We have to walk very carefully in case aio_set_fd_handler is
51
- * called while we're walking.
52
- */
53
- qemu_lockcnt_inc(&ctx->list_lock);
54
-
55
QLIST_FOREACH_SAFE_RCU(node, &ctx->aio_handlers, node, tmp) {
56
int revents;
57
58
@@ -XXX,XX +XXX,XX @@ static bool aio_dispatch_handlers(AioContext *ctx)
59
}
60
}
61
62
- qemu_lockcnt_dec(&ctx->list_lock);
63
return progress;
64
}
65
66
-/*
67
- * Note that dispatch_fds == false has the side-effect of post-poning the
68
- * freeing of deleted handlers.
69
- */
70
-bool aio_dispatch(AioContext *ctx, bool dispatch_fds)
71
+void aio_dispatch(AioContext *ctx)
72
{
73
- bool progress;
74
+ aio_bh_poll(ctx);
75
76
- /*
77
- * If there are callbacks left that have been queued, we need to call them.
78
- * Do not call select in this case, because it is possible that the caller
79
- * does not need a complete flush (as is the case for aio_poll loops).
80
- */
81
- progress = aio_bh_poll(ctx);
82
+ qemu_lockcnt_inc(&ctx->list_lock);
83
+ aio_dispatch_handlers(ctx);
84
+ qemu_lockcnt_dec(&ctx->list_lock);
85
86
- if (dispatch_fds) {
87
- progress |= aio_dispatch_handlers(ctx);
88
- }
89
-
90
- /* Run our timers */
91
- progress |= timerlistgroup_run_timers(&ctx->tlg);
92
-
93
- return progress;
94
+ timerlistgroup_run_timers(&ctx->tlg);
95
}
96
97
/* These thread-local variables are used only in a small part of aio_poll
98
@@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking)
99
npfd = 0;
100
qemu_lockcnt_dec(&ctx->list_lock);
101
102
- /* Run dispatch even if there were no readable fds to run timers */
103
- if (aio_dispatch(ctx, ret > 0)) {
104
- progress = true;
105
+ progress |= aio_bh_poll(ctx);
106
+
107
+ if (ret > 0) {
108
+ qemu_lockcnt_inc(&ctx->list_lock);
109
+ progress |= aio_dispatch_handlers(ctx);
110
+ qemu_lockcnt_dec(&ctx->list_lock);
111
}
112
113
+ progress |= timerlistgroup_run_timers(&ctx->tlg);
114
+
115
return progress;
116
}
117
118
diff --git a/util/aio-win32.c b/util/aio-win32.c
119
index XXXXXXX..XXXXXXX 100644
120
--- a/util/aio-win32.c
121
+++ b/util/aio-win32.c
122
@@ -XXX,XX +XXX,XX @@ static bool aio_dispatch_handlers(AioContext *ctx, HANDLE event)
123
return progress;
124
}
125
126
-bool aio_dispatch(AioContext *ctx, bool dispatch_fds)
127
+void aio_dispatch(AioContext *ctx)
128
{
129
- bool progress;
130
-
131
- progress = aio_bh_poll(ctx);
132
- if (dispatch_fds) {
133
- progress |= aio_dispatch_handlers(ctx, INVALID_HANDLE_VALUE);
134
- }
135
- progress |= timerlistgroup_run_timers(&ctx->tlg);
136
- return progress;
137
+ aio_bh_poll(ctx);
138
+ aio_dispatch_handlers(ctx, INVALID_HANDLE_VALUE);
139
+ timerlistgroup_run_timers(&ctx->tlg);
140
}
141
142
bool aio_poll(AioContext *ctx, bool blocking)
143
diff --git a/util/async.c b/util/async.c
144
index XXXXXXX..XXXXXXX 100644
145
--- a/util/async.c
146
+++ b/util/async.c
147
@@ -XXX,XX +XXX,XX @@ aio_ctx_dispatch(GSource *source,
148
AioContext *ctx = (AioContext *) source;
149
150
assert(callback == NULL);
151
- aio_dispatch(ctx, true);
152
+ aio_dispatch(ctx);
153
return true;
154
}
155
156
--
157
2.9.3
158
159
diff view generated by jsdifflib
1
From: Fabiano Rosas <farosas@suse.de>
1
From: Paolo Bonzini <pbonzini@redhat.com>
2
2
3
Having a name in the source helps with debugging core dumps when one
3
Pull the increment/decrement pair out of aio_bh_poll and into the
4
might not have access to TLS data to cross-reference AioContexts with
4
callers.
5
their addresses.
6
5
7
Signed-off-by: Fabiano Rosas <farosas@suse.de>
6
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
8
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
7
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9
Message-id: 20230905180359.14083-1-farosas@suse.de
8
Reviewed-by: Fam Zheng <famz@redhat.com>
9
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
10
Message-id: 20170213135235.12274-18-pbonzini@redhat.com
10
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11
---
12
---
12
iothread.c | 14 ++++++++------
13
util/aio-posix.c | 8 +++-----
13
1 file changed, 8 insertions(+), 6 deletions(-)
14
util/aio-win32.c | 8 ++++----
15
util/async.c | 12 ++++++------
16
3 files changed, 13 insertions(+), 15 deletions(-)
14
17
15
diff --git a/iothread.c b/iothread.c
18
diff --git a/util/aio-posix.c b/util/aio-posix.c
16
index XXXXXXX..XXXXXXX 100644
19
index XXXXXXX..XXXXXXX 100644
17
--- a/iothread.c
20
--- a/util/aio-posix.c
18
+++ b/iothread.c
21
+++ b/util/aio-posix.c
19
@@ -XXX,XX +XXX,XX @@ static void iothread_instance_finalize(Object *obj)
22
@@ -XXX,XX +XXX,XX @@ static bool aio_dispatch_handlers(AioContext *ctx)
20
qemu_sem_destroy(&iothread->init_done_sem);
23
24
void aio_dispatch(AioContext *ctx)
25
{
26
+ qemu_lockcnt_inc(&ctx->list_lock);
27
aio_bh_poll(ctx);
28
-
29
- qemu_lockcnt_inc(&ctx->list_lock);
30
aio_dispatch_handlers(ctx);
31
qemu_lockcnt_dec(&ctx->list_lock);
32
33
@@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking)
34
}
35
36
npfd = 0;
37
- qemu_lockcnt_dec(&ctx->list_lock);
38
39
progress |= aio_bh_poll(ctx);
40
41
if (ret > 0) {
42
- qemu_lockcnt_inc(&ctx->list_lock);
43
progress |= aio_dispatch_handlers(ctx);
44
- qemu_lockcnt_dec(&ctx->list_lock);
45
}
46
47
+ qemu_lockcnt_dec(&ctx->list_lock);
48
+
49
progress |= timerlistgroup_run_timers(&ctx->tlg);
50
51
return progress;
52
diff --git a/util/aio-win32.c b/util/aio-win32.c
53
index XXXXXXX..XXXXXXX 100644
54
--- a/util/aio-win32.c
55
+++ b/util/aio-win32.c
56
@@ -XXX,XX +XXX,XX @@ static bool aio_dispatch_handlers(AioContext *ctx, HANDLE event)
57
bool progress = false;
58
AioHandler *tmp;
59
60
- qemu_lockcnt_inc(&ctx->list_lock);
61
-
62
/*
63
* We have to walk very carefully in case aio_set_fd_handler is
64
* called while we're walking.
65
@@ -XXX,XX +XXX,XX @@ static bool aio_dispatch_handlers(AioContext *ctx, HANDLE event)
66
}
67
}
68
69
- qemu_lockcnt_dec(&ctx->list_lock);
70
return progress;
21
}
71
}
22
72
23
-static void iothread_init_gcontext(IOThread *iothread)
73
void aio_dispatch(AioContext *ctx)
24
+static void iothread_init_gcontext(IOThread *iothread, const char *thread_name)
25
{
74
{
26
GSource *source;
75
+ qemu_lockcnt_inc(&ctx->list_lock);
27
+ g_autofree char *name = g_strdup_printf("%s aio-context", thread_name);
76
aio_bh_poll(ctx);
28
77
aio_dispatch_handlers(ctx, INVALID_HANDLE_VALUE);
29
iothread->worker_context = g_main_context_new();
78
+ qemu_lockcnt_dec(&ctx->list_lock);
30
source = aio_get_g_source(iothread_get_aio_context(iothread));
79
timerlistgroup_run_timers(&ctx->tlg);
31
+ g_source_set_name(source, name);
80
}
32
g_source_attach(source, iothread->worker_context);
81
33
g_source_unref(source);
82
@@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking)
34
iothread->main_loop = g_main_loop_new(iothread->worker_context, TRUE);
83
}
35
@@ -XXX,XX +XXX,XX @@ static void iothread_init(EventLoopBase *base, Error **errp)
84
}
85
86
- qemu_lockcnt_dec(&ctx->list_lock);
87
first = true;
88
89
/* ctx->notifier is always registered. */
90
@@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking)
91
progress |= aio_dispatch_handlers(ctx, event);
92
} while (count > 0);
93
94
+ qemu_lockcnt_dec(&ctx->list_lock);
95
+
96
progress |= timerlistgroup_run_timers(&ctx->tlg);
97
return progress;
98
}
99
diff --git a/util/async.c b/util/async.c
100
index XXXXXXX..XXXXXXX 100644
101
--- a/util/async.c
102
+++ b/util/async.c
103
@@ -XXX,XX +XXX,XX @@ void aio_bh_call(QEMUBH *bh)
104
bh->cb(bh->opaque);
105
}
106
107
-/* Multiple occurrences of aio_bh_poll cannot be called concurrently */
108
+/* Multiple occurrences of aio_bh_poll cannot be called concurrently.
109
+ * The count in ctx->list_lock is incremented before the call, and is
110
+ * not affected by the call.
111
+ */
112
int aio_bh_poll(AioContext *ctx)
36
{
113
{
37
Error *local_error = NULL;
114
QEMUBH *bh, **bhp, *next;
38
IOThread *iothread = IOTHREAD(base);
115
int ret;
39
- char *thread_name;
116
bool deleted = false;
40
+ g_autofree char *thread_name = NULL;
117
41
118
- qemu_lockcnt_inc(&ctx->list_lock);
42
iothread->stopping = false;
119
-
43
iothread->running = true;
120
ret = 0;
44
@@ -XXX,XX +XXX,XX @@ static void iothread_init(EventLoopBase *base, Error **errp)
121
for (bh = atomic_rcu_read(&ctx->first_bh); bh; bh = next) {
45
return;
122
next = atomic_rcu_read(&bh->next);
123
@@ -XXX,XX +XXX,XX @@ int aio_bh_poll(AioContext *ctx)
124
125
/* remove deleted bhs */
126
if (!deleted) {
127
- qemu_lockcnt_dec(&ctx->list_lock);
128
return ret;
46
}
129
}
47
130
48
+ thread_name = g_strdup_printf("IO %s",
131
- if (qemu_lockcnt_dec_and_lock(&ctx->list_lock)) {
49
+ object_get_canonical_path_component(OBJECT(base)));
132
+ if (qemu_lockcnt_dec_if_lock(&ctx->list_lock)) {
50
+
133
bhp = &ctx->first_bh;
51
/*
134
while (*bhp) {
52
* Init one GMainContext for the iothread unconditionally, even if
135
bh = *bhp;
53
* it's not used
136
@@ -XXX,XX +XXX,XX @@ int aio_bh_poll(AioContext *ctx)
54
*/
137
bhp = &bh->next;
55
- iothread_init_gcontext(iothread);
138
}
56
+ iothread_init_gcontext(iothread, thread_name);
139
}
57
140
- qemu_lockcnt_unlock(&ctx->list_lock);
58
iothread_set_aio_context_params(base, &local_error);
141
+ qemu_lockcnt_inc_and_unlock(&ctx->list_lock);
59
if (local_error) {
142
}
60
@@ -XXX,XX +XXX,XX @@ static void iothread_init(EventLoopBase *base, Error **errp)
143
return ret;
61
/* This assumes we are called from a thread with useful CPU affinity for us
144
}
62
* to inherit.
63
*/
64
- thread_name = g_strdup_printf("IO %s",
65
- object_get_canonical_path_component(OBJECT(base)));
66
qemu_thread_create(&iothread->thread, thread_name, iothread_run,
67
iothread, QEMU_THREAD_JOINABLE);
68
- g_free(thread_name);
69
70
/* Wait for initialization to complete */
71
while (iothread->thread_id == -1) {
72
--
145
--
73
2.41.0
146
2.9.3
74
147
75
148
diff view generated by jsdifflib
New patch
1
From: Paolo Bonzini <pbonzini@redhat.com>
1
2
3
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
4
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5
Reviewed-by: Fam Zheng <famz@redhat.com>
6
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
7
Message-id: 20170213135235.12274-19-pbonzini@redhat.com
8
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
9
---
10
include/block/block_int.h | 64 +++++++++++++++++++++++++-----------------
11
include/sysemu/block-backend.h | 14 ++++++---
12
2 files changed, 49 insertions(+), 29 deletions(-)
13
14
diff --git a/include/block/block_int.h b/include/block/block_int.h
15
index XXXXXXX..XXXXXXX 100644
16
--- a/include/block/block_int.h
17
+++ b/include/block/block_int.h
18
@@ -XXX,XX +XXX,XX @@ struct BdrvChild {
19
* copied as well.
20
*/
21
struct BlockDriverState {
22
- int64_t total_sectors; /* if we are reading a disk image, give its
23
- size in sectors */
24
+ /* Protected by big QEMU lock or read-only after opening. No special
25
+ * locking needed during I/O...
26
+ */
27
int open_flags; /* flags used to open the file, re-used for re-open */
28
bool read_only; /* if true, the media is read only */
29
bool encrypted; /* if true, the media is encrypted */
30
@@ -XXX,XX +XXX,XX @@ struct BlockDriverState {
31
bool sg; /* if true, the device is a /dev/sg* */
32
bool probed; /* if true, format was probed rather than specified */
33
34
- int copy_on_read; /* if nonzero, copy read backing sectors into image.
35
- note this is a reference count */
36
-
37
- CoQueue flush_queue; /* Serializing flush queue */
38
- bool active_flush_req; /* Flush request in flight? */
39
- unsigned int write_gen; /* Current data generation */
40
- unsigned int flushed_gen; /* Flushed write generation */
41
-
42
BlockDriver *drv; /* NULL means no media */
43
void *opaque;
44
45
@@ -XXX,XX +XXX,XX @@ struct BlockDriverState {
46
BdrvChild *backing;
47
BdrvChild *file;
48
49
- /* Callback before write request is processed */
50
- NotifierWithReturnList before_write_notifiers;
51
-
52
- /* number of in-flight requests; overall and serialising */
53
- unsigned int in_flight;
54
- unsigned int serialising_in_flight;
55
-
56
- bool wakeup;
57
-
58
- /* Offset after the highest byte written to */
59
- uint64_t wr_highest_offset;
60
-
61
/* I/O Limits */
62
BlockLimits bl;
63
64
@@ -XXX,XX +XXX,XX @@ struct BlockDriverState {
65
QTAILQ_ENTRY(BlockDriverState) bs_list;
66
/* element of the list of monitor-owned BDS */
67
QTAILQ_ENTRY(BlockDriverState) monitor_list;
68
- QLIST_HEAD(, BdrvDirtyBitmap) dirty_bitmaps;
69
int refcnt;
70
71
- QLIST_HEAD(, BdrvTrackedRequest) tracked_requests;
72
-
73
/* operation blockers */
74
QLIST_HEAD(, BdrvOpBlocker) op_blockers[BLOCK_OP_TYPE_MAX];
75
76
@@ -XXX,XX +XXX,XX @@ struct BlockDriverState {
77
/* The error object in use for blocking operations on backing_hd */
78
Error *backing_blocker;
79
80
+ /* Protected by AioContext lock */
81
+
82
+ /* If true, copy read backing sectors into image. Can be >1 if more
83
+ * than one client has requested copy-on-read.
84
+ */
85
+ int copy_on_read;
86
+
87
+ /* If we are reading a disk image, give its size in sectors.
88
+ * Generally read-only; it is written to by load_vmstate and save_vmstate,
89
+ * but the block layer is quiescent during those.
90
+ */
91
+ int64_t total_sectors;
92
+
93
+ /* Callback before write request is processed */
94
+ NotifierWithReturnList before_write_notifiers;
95
+
96
+ /* number of in-flight requests; overall and serialising */
97
+ unsigned int in_flight;
98
+ unsigned int serialising_in_flight;
99
+
100
+ bool wakeup;
101
+
102
+ /* Offset after the highest byte written to */
103
+ uint64_t wr_highest_offset;
104
+
105
/* threshold limit for writes, in bytes. "High water mark". */
106
uint64_t write_threshold_offset;
107
NotifierWithReturn write_threshold_notifier;
108
@@ -XXX,XX +XXX,XX @@ struct BlockDriverState {
109
/* counter for nested bdrv_io_plug */
110
unsigned io_plugged;
111
112
+ QLIST_HEAD(, BdrvTrackedRequest) tracked_requests;
113
+ CoQueue flush_queue; /* Serializing flush queue */
114
+ bool active_flush_req; /* Flush request in flight? */
115
+ unsigned int write_gen; /* Current data generation */
116
+ unsigned int flushed_gen; /* Flushed write generation */
117
+
118
+ QLIST_HEAD(, BdrvDirtyBitmap) dirty_bitmaps;
119
+
120
+ /* do we need to tell the quest if we have a volatile write cache? */
121
+ int enable_write_cache;
122
+
123
int quiesce_counter;
124
};
125
126
diff --git a/include/sysemu/block-backend.h b/include/sysemu/block-backend.h
127
index XXXXXXX..XXXXXXX 100644
128
--- a/include/sysemu/block-backend.h
129
+++ b/include/sysemu/block-backend.h
130
@@ -XXX,XX +XXX,XX @@ typedef struct BlockDevOps {
131
* fields that must be public. This is in particular for QLIST_ENTRY() and
132
* friends so that BlockBackends can be kept in lists outside block-backend.c */
133
typedef struct BlockBackendPublic {
134
- /* I/O throttling.
135
- * throttle_state tells us if this BlockBackend has I/O limits configured.
136
- * io_limits_disabled tells us if they are currently being enforced */
137
+ /* I/O throttling has its own locking, but also some fields are
138
+ * protected by the AioContext lock.
139
+ */
140
+
141
+ /* Protected by AioContext lock. */
142
CoQueue throttled_reqs[2];
143
+
144
+ /* Nonzero if the I/O limits are currently being ignored; generally
145
+ * it is zero. */
146
unsigned int io_limits_disabled;
147
148
/* The following fields are protected by the ThrottleGroup lock.
149
- * See the ThrottleGroup documentation for details. */
150
+ * See the ThrottleGroup documentation for details.
151
+ * throttle_state tells us if I/O limits are configured. */
152
ThrottleState *throttle_state;
153
ThrottleTimers throttle_timers;
154
unsigned pending_reqs[2];
155
--
156
2.9.3
157
158
diff view generated by jsdifflib
New patch
1
1
From: Paolo Bonzini <pbonzini@redhat.com>
2
3
This uses the lock-free mutex described in the paper '"Blocking without
4
Locking", or LFTHREADS: A lock-free thread library' by Gidenstam and
5
Papatriantafilou. The same technique is used in OSv, and in fact
6
the code is essentially a conversion to C of OSv's code.
7
8
[Added missing coroutine_fn in tests/test-aio-multithread.c.
9
--Stefan]
10
11
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
12
Reviewed-by: Fam Zheng <famz@redhat.com>
13
Message-id: 20170213181244.16297-2-pbonzini@redhat.com
14
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
15
---
16
include/qemu/coroutine.h | 17 ++++-
17
tests/test-aio-multithread.c | 86 ++++++++++++++++++++++++
18
util/qemu-coroutine-lock.c | 155 ++++++++++++++++++++++++++++++++++++++++---
19
util/trace-events | 1 +
20
4 files changed, 246 insertions(+), 13 deletions(-)
21
22
diff --git a/include/qemu/coroutine.h b/include/qemu/coroutine.h
23
index XXXXXXX..XXXXXXX 100644
24
--- a/include/qemu/coroutine.h
25
+++ b/include/qemu/coroutine.h
26
@@ -XXX,XX +XXX,XX @@ bool qemu_co_queue_empty(CoQueue *queue);
27
/**
28
* Provides a mutex that can be used to synchronise coroutines
29
*/
30
+struct CoWaitRecord;
31
typedef struct CoMutex {
32
- bool locked;
33
+ /* Count of pending lockers; 0 for a free mutex, 1 for an
34
+ * uncontended mutex.
35
+ */
36
+ unsigned locked;
37
+
38
+ /* A queue of waiters. Elements are added atomically in front of
39
+ * from_push. to_pop is only populated, and popped from, by whoever
40
+ * is in charge of the next wakeup. This can be an unlocker or,
41
+ * through the handoff protocol, a locker that is about to go to sleep.
42
+ */
43
+ QSLIST_HEAD(, CoWaitRecord) from_push, to_pop;
44
+
45
+ unsigned handoff, sequence;
46
+
47
Coroutine *holder;
48
- CoQueue queue;
49
} CoMutex;
50
51
/**
52
diff --git a/tests/test-aio-multithread.c b/tests/test-aio-multithread.c
53
index XXXXXXX..XXXXXXX 100644
54
--- a/tests/test-aio-multithread.c
55
+++ b/tests/test-aio-multithread.c
56
@@ -XXX,XX +XXX,XX @@ static void test_multi_co_schedule_10(void)
57
test_multi_co_schedule(10);
58
}
59
60
+/* CoMutex thread-safety. */
61
+
62
+static uint32_t atomic_counter;
63
+static uint32_t running;
64
+static uint32_t counter;
65
+static CoMutex comutex;
66
+
67
+static void coroutine_fn test_multi_co_mutex_entry(void *opaque)
68
+{
69
+ while (!atomic_mb_read(&now_stopping)) {
70
+ qemu_co_mutex_lock(&comutex);
71
+ counter++;
72
+ qemu_co_mutex_unlock(&comutex);
73
+
74
+ /* Increase atomic_counter *after* releasing the mutex. Otherwise
75
+ * there is a chance (it happens about 1 in 3 runs) that the iothread
76
+ * exits before the coroutine is woken up, causing a spurious
77
+ * assertion failure.
78
+ */
79
+ atomic_inc(&atomic_counter);
80
+ }
81
+ atomic_dec(&running);
82
+}
83
+
84
+static void test_multi_co_mutex(int threads, int seconds)
85
+{
86
+ int i;
87
+
88
+ qemu_co_mutex_init(&comutex);
89
+ counter = 0;
90
+ atomic_counter = 0;
91
+ now_stopping = false;
92
+
93
+ create_aio_contexts();
94
+ assert(threads <= NUM_CONTEXTS);
95
+ running = threads;
96
+ for (i = 0; i < threads; i++) {
97
+ Coroutine *co1 = qemu_coroutine_create(test_multi_co_mutex_entry, NULL);
98
+ aio_co_schedule(ctx[i], co1);
99
+ }
100
+
101
+ g_usleep(seconds * 1000000);
102
+
103
+ atomic_mb_set(&now_stopping, true);
104
+ while (running > 0) {
105
+ g_usleep(100000);
106
+ }
107
+
108
+ join_aio_contexts();
109
+ g_test_message("%d iterations/second\n", counter / seconds);
110
+ g_assert_cmpint(counter, ==, atomic_counter);
111
+}
112
+
113
+/* Testing with NUM_CONTEXTS threads focuses on the queue. The mutex however
114
+ * is too contended (and the threads spend too much time in aio_poll)
115
+ * to actually stress the handoff protocol.
116
+ */
117
+static void test_multi_co_mutex_1(void)
118
+{
119
+ test_multi_co_mutex(NUM_CONTEXTS, 1);
120
+}
121
+
122
+static void test_multi_co_mutex_10(void)
123
+{
124
+ test_multi_co_mutex(NUM_CONTEXTS, 10);
125
+}
126
+
127
+/* Testing with fewer threads stresses the handoff protocol too. Still, the
128
+ * case where the locker _can_ pick up a handoff is very rare, happening
129
+ * about 10 times in 1 million, so increase the runtime a bit compared to
130
+ * other "quick" testcases that only run for 1 second.
131
+ */
132
+static void test_multi_co_mutex_2_3(void)
133
+{
134
+ test_multi_co_mutex(2, 3);
135
+}
136
+
137
+static void test_multi_co_mutex_2_30(void)
138
+{
139
+ test_multi_co_mutex(2, 30);
140
+}
141
+
142
/* End of tests. */
143
144
int main(int argc, char **argv)
145
@@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv)
146
g_test_add_func("/aio/multi/lifecycle", test_lifecycle);
147
if (g_test_quick()) {
148
g_test_add_func("/aio/multi/schedule", test_multi_co_schedule_1);
149
+ g_test_add_func("/aio/multi/mutex/contended", test_multi_co_mutex_1);
150
+ g_test_add_func("/aio/multi/mutex/handoff", test_multi_co_mutex_2_3);
151
} else {
152
g_test_add_func("/aio/multi/schedule", test_multi_co_schedule_10);
153
+ g_test_add_func("/aio/multi/mutex/contended", test_multi_co_mutex_10);
154
+ g_test_add_func("/aio/multi/mutex/handoff", test_multi_co_mutex_2_30);
155
}
156
return g_test_run();
157
}
158
diff --git a/util/qemu-coroutine-lock.c b/util/qemu-coroutine-lock.c
159
index XXXXXXX..XXXXXXX 100644
160
--- a/util/qemu-coroutine-lock.c
161
+++ b/util/qemu-coroutine-lock.c
162
@@ -XXX,XX +XXX,XX @@
163
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
164
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
165
* THE SOFTWARE.
166
+ *
167
+ * The lock-free mutex implementation is based on OSv
168
+ * (core/lfmutex.cc, include/lockfree/mutex.hh).
169
+ * Copyright (C) 2013 Cloudius Systems, Ltd.
170
*/
171
172
#include "qemu/osdep.h"
173
@@ -XXX,XX +XXX,XX @@ bool qemu_co_queue_empty(CoQueue *queue)
174
return QSIMPLEQ_FIRST(&queue->entries) == NULL;
175
}
176
177
+/* The wait records are handled with a multiple-producer, single-consumer
178
+ * lock-free queue. There cannot be two concurrent pop_waiter() calls
179
+ * because pop_waiter() can only be called while mutex->handoff is zero.
180
+ * This can happen in three cases:
181
+ * - in qemu_co_mutex_unlock, before the hand-off protocol has started.
182
+ * In this case, qemu_co_mutex_lock will see mutex->handoff == 0 and
183
+ * not take part in the handoff.
184
+ * - in qemu_co_mutex_lock, if it steals the hand-off responsibility from
185
+ * qemu_co_mutex_unlock. In this case, qemu_co_mutex_unlock will fail
186
+ * the cmpxchg (it will see either 0 or the next sequence value) and
187
+ * exit. The next hand-off cannot begin until qemu_co_mutex_lock has
188
+ * woken up someone.
189
+ * - in qemu_co_mutex_unlock, if it takes the hand-off token itself.
190
+ * In this case another iteration starts with mutex->handoff == 0;
191
+ * a concurrent qemu_co_mutex_lock will fail the cmpxchg, and
192
+ * qemu_co_mutex_unlock will go back to case (1).
193
+ *
194
+ * The following functions manage this queue.
195
+ */
196
+typedef struct CoWaitRecord {
197
+ Coroutine *co;
198
+ QSLIST_ENTRY(CoWaitRecord) next;
199
+} CoWaitRecord;
200
+
201
+static void push_waiter(CoMutex *mutex, CoWaitRecord *w)
202
+{
203
+ w->co = qemu_coroutine_self();
204
+ QSLIST_INSERT_HEAD_ATOMIC(&mutex->from_push, w, next);
205
+}
206
+
207
+static void move_waiters(CoMutex *mutex)
208
+{
209
+ QSLIST_HEAD(, CoWaitRecord) reversed;
210
+ QSLIST_MOVE_ATOMIC(&reversed, &mutex->from_push);
211
+ while (!QSLIST_EMPTY(&reversed)) {
212
+ CoWaitRecord *w = QSLIST_FIRST(&reversed);
213
+ QSLIST_REMOVE_HEAD(&reversed, next);
214
+ QSLIST_INSERT_HEAD(&mutex->to_pop, w, next);
215
+ }
216
+}
217
+
218
+static CoWaitRecord *pop_waiter(CoMutex *mutex)
219
+{
220
+ CoWaitRecord *w;
221
+
222
+ if (QSLIST_EMPTY(&mutex->to_pop)) {
223
+ move_waiters(mutex);
224
+ if (QSLIST_EMPTY(&mutex->to_pop)) {
225
+ return NULL;
226
+ }
227
+ }
228
+ w = QSLIST_FIRST(&mutex->to_pop);
229
+ QSLIST_REMOVE_HEAD(&mutex->to_pop, next);
230
+ return w;
231
+}
232
+
233
+static bool has_waiters(CoMutex *mutex)
234
+{
235
+ return QSLIST_EMPTY(&mutex->to_pop) || QSLIST_EMPTY(&mutex->from_push);
236
+}
237
+
238
void qemu_co_mutex_init(CoMutex *mutex)
239
{
240
memset(mutex, 0, sizeof(*mutex));
241
- qemu_co_queue_init(&mutex->queue);
242
}
243
244
-void coroutine_fn qemu_co_mutex_lock(CoMutex *mutex)
245
+static void coroutine_fn qemu_co_mutex_lock_slowpath(CoMutex *mutex)
246
{
247
Coroutine *self = qemu_coroutine_self();
248
+ CoWaitRecord w;
249
+ unsigned old_handoff;
250
251
trace_qemu_co_mutex_lock_entry(mutex, self);
252
+ w.co = self;
253
+ push_waiter(mutex, &w);
254
255
- while (mutex->locked) {
256
- qemu_co_queue_wait(&mutex->queue);
257
+ /* This is the "Responsibility Hand-Off" protocol; a lock() picks from
258
+ * a concurrent unlock() the responsibility of waking somebody up.
259
+ */
260
+ old_handoff = atomic_mb_read(&mutex->handoff);
261
+ if (old_handoff &&
262
+ has_waiters(mutex) &&
263
+ atomic_cmpxchg(&mutex->handoff, old_handoff, 0) == old_handoff) {
264
+ /* There can be no concurrent pops, because there can be only
265
+ * one active handoff at a time.
266
+ */
267
+ CoWaitRecord *to_wake = pop_waiter(mutex);
268
+ Coroutine *co = to_wake->co;
269
+ if (co == self) {
270
+ /* We got the lock ourselves! */
271
+ assert(to_wake == &w);
272
+ return;
273
+ }
274
+
275
+ aio_co_wake(co);
276
}
277
278
- mutex->locked = true;
279
- mutex->holder = self;
280
- self->locks_held++;
281
-
282
+ qemu_coroutine_yield();
283
trace_qemu_co_mutex_lock_return(mutex, self);
284
}
285
286
+void coroutine_fn qemu_co_mutex_lock(CoMutex *mutex)
287
+{
288
+ Coroutine *self = qemu_coroutine_self();
289
+
290
+ if (atomic_fetch_inc(&mutex->locked) == 0) {
291
+ /* Uncontended. */
292
+ trace_qemu_co_mutex_lock_uncontended(mutex, self);
293
+ } else {
294
+ qemu_co_mutex_lock_slowpath(mutex);
295
+ }
296
+ mutex->holder = self;
297
+ self->locks_held++;
298
+}
299
+
300
void coroutine_fn qemu_co_mutex_unlock(CoMutex *mutex)
301
{
302
Coroutine *self = qemu_coroutine_self();
303
304
trace_qemu_co_mutex_unlock_entry(mutex, self);
305
306
- assert(mutex->locked == true);
307
+ assert(mutex->locked);
308
assert(mutex->holder == self);
309
assert(qemu_in_coroutine());
310
311
- mutex->locked = false;
312
mutex->holder = NULL;
313
self->locks_held--;
314
- qemu_co_queue_next(&mutex->queue);
315
+ if (atomic_fetch_dec(&mutex->locked) == 1) {
316
+ /* No waiting qemu_co_mutex_lock(). Pfew, that was easy! */
317
+ return;
318
+ }
319
+
320
+ for (;;) {
321
+ CoWaitRecord *to_wake = pop_waiter(mutex);
322
+ unsigned our_handoff;
323
+
324
+ if (to_wake) {
325
+ Coroutine *co = to_wake->co;
326
+ aio_co_wake(co);
327
+ break;
328
+ }
329
+
330
+ /* Some concurrent lock() is in progress (we know this because
331
+ * mutex->locked was >1) but it hasn't yet put itself on the wait
332
+ * queue. Pick a sequence number for the handoff protocol (not 0).
333
+ */
334
+ if (++mutex->sequence == 0) {
335
+ mutex->sequence = 1;
336
+ }
337
+
338
+ our_handoff = mutex->sequence;
339
+ atomic_mb_set(&mutex->handoff, our_handoff);
340
+ if (!has_waiters(mutex)) {
341
+ /* The concurrent lock has not added itself yet, so it
342
+ * will be able to pick our handoff.
343
+ */
344
+ break;
345
+ }
346
+
347
+ /* Try to do the handoff protocol ourselves; if somebody else has
348
+ * already taken it, however, we're done and they're responsible.
349
+ */
350
+ if (atomic_cmpxchg(&mutex->handoff, our_handoff, 0) != our_handoff) {
351
+ break;
352
+ }
353
+ }
354
355
trace_qemu_co_mutex_unlock_return(mutex, self);
356
}
357
diff --git a/util/trace-events b/util/trace-events
358
index XXXXXXX..XXXXXXX 100644
359
--- a/util/trace-events
360
+++ b/util/trace-events
361
@@ -XXX,XX +XXX,XX @@ qemu_coroutine_terminate(void *co) "self %p"
362
363
# util/qemu-coroutine-lock.c
364
qemu_co_queue_run_restart(void *co) "co %p"
365
+qemu_co_mutex_lock_uncontended(void *mutex, void *self) "mutex %p self %p"
366
qemu_co_mutex_lock_entry(void *mutex, void *self) "mutex %p self %p"
367
qemu_co_mutex_lock_return(void *mutex, void *self) "mutex %p self %p"
368
qemu_co_mutex_unlock_entry(void *mutex, void *self) "mutex %p self %p"
369
--
370
2.9.3
371
372
diff view generated by jsdifflib
New patch
1
From: Paolo Bonzini <pbonzini@redhat.com>
1
2
3
Running a very small critical section on pthread_mutex_t and CoMutex
4
shows that pthread_mutex_t is much faster because it doesn't actually
5
go to sleep. What happens is that the critical section is shorter
6
than the latency of entering the kernel and thus FUTEX_WAIT always
7
fails. With CoMutex there is no such latency but you still want to
8
avoid wait and wakeup. So introduce it artificially.
9
10
This only works with one waiters; because CoMutex is fair, it will
11
always have more waits and wakeups than a pthread_mutex_t.
12
13
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
14
Reviewed-by: Fam Zheng <famz@redhat.com>
15
Message-id: 20170213181244.16297-3-pbonzini@redhat.com
16
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
17
---
18
include/qemu/coroutine.h | 5 +++++
19
util/qemu-coroutine-lock.c | 51 ++++++++++++++++++++++++++++++++++++++++------
20
util/qemu-coroutine.c | 2 +-
21
3 files changed, 51 insertions(+), 7 deletions(-)
22
23
diff --git a/include/qemu/coroutine.h b/include/qemu/coroutine.h
24
index XXXXXXX..XXXXXXX 100644
25
--- a/include/qemu/coroutine.h
26
+++ b/include/qemu/coroutine.h
27
@@ -XXX,XX +XXX,XX @@ typedef struct CoMutex {
28
*/
29
unsigned locked;
30
31
+ /* Context that is holding the lock. Useful to avoid spinning
32
+ * when two coroutines on the same AioContext try to get the lock. :)
33
+ */
34
+ AioContext *ctx;
35
+
36
/* A queue of waiters. Elements are added atomically in front of
37
* from_push. to_pop is only populated, and popped from, by whoever
38
* is in charge of the next wakeup. This can be an unlocker or,
39
diff --git a/util/qemu-coroutine-lock.c b/util/qemu-coroutine-lock.c
40
index XXXXXXX..XXXXXXX 100644
41
--- a/util/qemu-coroutine-lock.c
42
+++ b/util/qemu-coroutine-lock.c
43
@@ -XXX,XX +XXX,XX @@
44
#include "qemu-common.h"
45
#include "qemu/coroutine.h"
46
#include "qemu/coroutine_int.h"
47
+#include "qemu/processor.h"
48
#include "qemu/queue.h"
49
#include "block/aio.h"
50
#include "trace.h"
51
@@ -XXX,XX +XXX,XX @@ void qemu_co_mutex_init(CoMutex *mutex)
52
memset(mutex, 0, sizeof(*mutex));
53
}
54
55
-static void coroutine_fn qemu_co_mutex_lock_slowpath(CoMutex *mutex)
56
+static void coroutine_fn qemu_co_mutex_wake(CoMutex *mutex, Coroutine *co)
57
+{
58
+ /* Read co before co->ctx; pairs with smp_wmb() in
59
+ * qemu_coroutine_enter().
60
+ */
61
+ smp_read_barrier_depends();
62
+ mutex->ctx = co->ctx;
63
+ aio_co_wake(co);
64
+}
65
+
66
+static void coroutine_fn qemu_co_mutex_lock_slowpath(AioContext *ctx,
67
+ CoMutex *mutex)
68
{
69
Coroutine *self = qemu_coroutine_self();
70
CoWaitRecord w;
71
@@ -XXX,XX +XXX,XX @@ static void coroutine_fn qemu_co_mutex_lock_slowpath(CoMutex *mutex)
72
if (co == self) {
73
/* We got the lock ourselves! */
74
assert(to_wake == &w);
75
+ mutex->ctx = ctx;
76
return;
77
}
78
79
- aio_co_wake(co);
80
+ qemu_co_mutex_wake(mutex, co);
81
}
82
83
qemu_coroutine_yield();
84
@@ -XXX,XX +XXX,XX @@ static void coroutine_fn qemu_co_mutex_lock_slowpath(CoMutex *mutex)
85
86
void coroutine_fn qemu_co_mutex_lock(CoMutex *mutex)
87
{
88
+ AioContext *ctx = qemu_get_current_aio_context();
89
Coroutine *self = qemu_coroutine_self();
90
+ int waiters, i;
91
92
- if (atomic_fetch_inc(&mutex->locked) == 0) {
93
+ /* Running a very small critical section on pthread_mutex_t and CoMutex
94
+ * shows that pthread_mutex_t is much faster because it doesn't actually
95
+ * go to sleep. What happens is that the critical section is shorter
96
+ * than the latency of entering the kernel and thus FUTEX_WAIT always
97
+ * fails. With CoMutex there is no such latency but you still want to
98
+ * avoid wait and wakeup. So introduce it artificially.
99
+ */
100
+ i = 0;
101
+retry_fast_path:
102
+ waiters = atomic_cmpxchg(&mutex->locked, 0, 1);
103
+ if (waiters != 0) {
104
+ while (waiters == 1 && ++i < 1000) {
105
+ if (atomic_read(&mutex->ctx) == ctx) {
106
+ break;
107
+ }
108
+ if (atomic_read(&mutex->locked) == 0) {
109
+ goto retry_fast_path;
110
+ }
111
+ cpu_relax();
112
+ }
113
+ waiters = atomic_fetch_inc(&mutex->locked);
114
+ }
115
+
116
+ if (waiters == 0) {
117
/* Uncontended. */
118
trace_qemu_co_mutex_lock_uncontended(mutex, self);
119
+ mutex->ctx = ctx;
120
} else {
121
- qemu_co_mutex_lock_slowpath(mutex);
122
+ qemu_co_mutex_lock_slowpath(ctx, mutex);
123
}
124
mutex->holder = self;
125
self->locks_held++;
126
@@ -XXX,XX +XXX,XX @@ void coroutine_fn qemu_co_mutex_unlock(CoMutex *mutex)
127
assert(mutex->holder == self);
128
assert(qemu_in_coroutine());
129
130
+ mutex->ctx = NULL;
131
mutex->holder = NULL;
132
self->locks_held--;
133
if (atomic_fetch_dec(&mutex->locked) == 1) {
134
@@ -XXX,XX +XXX,XX @@ void coroutine_fn qemu_co_mutex_unlock(CoMutex *mutex)
135
unsigned our_handoff;
136
137
if (to_wake) {
138
- Coroutine *co = to_wake->co;
139
- aio_co_wake(co);
140
+ qemu_co_mutex_wake(mutex, to_wake->co);
141
break;
142
}
143
144
diff --git a/util/qemu-coroutine.c b/util/qemu-coroutine.c
145
index XXXXXXX..XXXXXXX 100644
146
--- a/util/qemu-coroutine.c
147
+++ b/util/qemu-coroutine.c
148
@@ -XXX,XX +XXX,XX @@ void qemu_coroutine_enter(Coroutine *co)
149
co->ctx = qemu_get_current_aio_context();
150
151
/* Store co->ctx before anything that stores co. Matches
152
- * barrier in aio_co_wake.
153
+ * barrier in aio_co_wake and qemu_co_mutex_wake.
154
*/
155
smp_wmb();
156
157
--
158
2.9.3
159
160
diff view generated by jsdifflib
New patch
1
1
From: Paolo Bonzini <pbonzini@redhat.com>
2
3
Add two implementations of the same benchmark as the previous patch,
4
but using pthreads. One uses a normal QemuMutex, the other is Linux
5
only and implements a fair mutex based on MCS locks and futexes.
6
This shows that the slower performance of the 5-thread case is due to
7
the fairness of CoMutex, rather than to coroutines. If fairness does
8
not matter, as is the case with two threads, CoMutex can actually be
9
faster than pthreads.
10
11
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
12
Reviewed-by: Fam Zheng <famz@redhat.com>
13
Message-id: 20170213181244.16297-4-pbonzini@redhat.com
14
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
15
---
16
tests/test-aio-multithread.c | 164 +++++++++++++++++++++++++++++++++++++++++++
17
1 file changed, 164 insertions(+)
18
19
diff --git a/tests/test-aio-multithread.c b/tests/test-aio-multithread.c
20
index XXXXXXX..XXXXXXX 100644
21
--- a/tests/test-aio-multithread.c
22
+++ b/tests/test-aio-multithread.c
23
@@ -XXX,XX +XXX,XX @@ static void test_multi_co_mutex_2_30(void)
24
test_multi_co_mutex(2, 30);
25
}
26
27
+/* Same test with fair mutexes, for performance comparison. */
28
+
29
+#ifdef CONFIG_LINUX
30
+#include "qemu/futex.h"
31
+
32
+/* The nodes for the mutex reside in this structure (on which we try to avoid
33
+ * false sharing). The head of the mutex is in the "mutex_head" variable.
34
+ */
35
+static struct {
36
+ int next, locked;
37
+ int padding[14];
38
+} nodes[NUM_CONTEXTS] __attribute__((__aligned__(64)));
39
+
40
+static int mutex_head = -1;
41
+
42
+static void mcs_mutex_lock(void)
43
+{
44
+ int prev;
45
+
46
+ nodes[id].next = -1;
47
+ nodes[id].locked = 1;
48
+ prev = atomic_xchg(&mutex_head, id);
49
+ if (prev != -1) {
50
+ atomic_set(&nodes[prev].next, id);
51
+ qemu_futex_wait(&nodes[id].locked, 1);
52
+ }
53
+}
54
+
55
+static void mcs_mutex_unlock(void)
56
+{
57
+ int next;
58
+ if (nodes[id].next == -1) {
59
+ if (atomic_read(&mutex_head) == id &&
60
+ atomic_cmpxchg(&mutex_head, id, -1) == id) {
61
+ /* Last item in the list, exit. */
62
+ return;
63
+ }
64
+ while (atomic_read(&nodes[id].next) == -1) {
65
+ /* mcs_mutex_lock did the xchg, but has not updated
66
+ * nodes[prev].next yet.
67
+ */
68
+ }
69
+ }
70
+
71
+ /* Wake up the next in line. */
72
+ next = nodes[id].next;
73
+ nodes[next].locked = 0;
74
+ qemu_futex_wake(&nodes[next].locked, 1);
75
+}
76
+
77
+static void test_multi_fair_mutex_entry(void *opaque)
78
+{
79
+ while (!atomic_mb_read(&now_stopping)) {
80
+ mcs_mutex_lock();
81
+ counter++;
82
+ mcs_mutex_unlock();
83
+ atomic_inc(&atomic_counter);
84
+ }
85
+ atomic_dec(&running);
86
+}
87
+
88
+static void test_multi_fair_mutex(int threads, int seconds)
89
+{
90
+ int i;
91
+
92
+ assert(mutex_head == -1);
93
+ counter = 0;
94
+ atomic_counter = 0;
95
+ now_stopping = false;
96
+
97
+ create_aio_contexts();
98
+ assert(threads <= NUM_CONTEXTS);
99
+ running = threads;
100
+ for (i = 0; i < threads; i++) {
101
+ Coroutine *co1 = qemu_coroutine_create(test_multi_fair_mutex_entry, NULL);
102
+ aio_co_schedule(ctx[i], co1);
103
+ }
104
+
105
+ g_usleep(seconds * 1000000);
106
+
107
+ atomic_mb_set(&now_stopping, true);
108
+ while (running > 0) {
109
+ g_usleep(100000);
110
+ }
111
+
112
+ join_aio_contexts();
113
+ g_test_message("%d iterations/second\n", counter / seconds);
114
+ g_assert_cmpint(counter, ==, atomic_counter);
115
+}
116
+
117
+static void test_multi_fair_mutex_1(void)
118
+{
119
+ test_multi_fair_mutex(NUM_CONTEXTS, 1);
120
+}
121
+
122
+static void test_multi_fair_mutex_10(void)
123
+{
124
+ test_multi_fair_mutex(NUM_CONTEXTS, 10);
125
+}
126
+#endif
127
+
128
+/* Same test with pthread mutexes, for performance comparison and
129
+ * portability. */
130
+
131
+static QemuMutex mutex;
132
+
133
+static void test_multi_mutex_entry(void *opaque)
134
+{
135
+ while (!atomic_mb_read(&now_stopping)) {
136
+ qemu_mutex_lock(&mutex);
137
+ counter++;
138
+ qemu_mutex_unlock(&mutex);
139
+ atomic_inc(&atomic_counter);
140
+ }
141
+ atomic_dec(&running);
142
+}
143
+
144
+static void test_multi_mutex(int threads, int seconds)
145
+{
146
+ int i;
147
+
148
+ qemu_mutex_init(&mutex);
149
+ counter = 0;
150
+ atomic_counter = 0;
151
+ now_stopping = false;
152
+
153
+ create_aio_contexts();
154
+ assert(threads <= NUM_CONTEXTS);
155
+ running = threads;
156
+ for (i = 0; i < threads; i++) {
157
+ Coroutine *co1 = qemu_coroutine_create(test_multi_mutex_entry, NULL);
158
+ aio_co_schedule(ctx[i], co1);
159
+ }
160
+
161
+ g_usleep(seconds * 1000000);
162
+
163
+ atomic_mb_set(&now_stopping, true);
164
+ while (running > 0) {
165
+ g_usleep(100000);
166
+ }
167
+
168
+ join_aio_contexts();
169
+ g_test_message("%d iterations/second\n", counter / seconds);
170
+ g_assert_cmpint(counter, ==, atomic_counter);
171
+}
172
+
173
+static void test_multi_mutex_1(void)
174
+{
175
+ test_multi_mutex(NUM_CONTEXTS, 1);
176
+}
177
+
178
+static void test_multi_mutex_10(void)
179
+{
180
+ test_multi_mutex(NUM_CONTEXTS, 10);
181
+}
182
+
183
/* End of tests. */
184
185
int main(int argc, char **argv)
186
@@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv)
187
g_test_add_func("/aio/multi/schedule", test_multi_co_schedule_1);
188
g_test_add_func("/aio/multi/mutex/contended", test_multi_co_mutex_1);
189
g_test_add_func("/aio/multi/mutex/handoff", test_multi_co_mutex_2_3);
190
+#ifdef CONFIG_LINUX
191
+ g_test_add_func("/aio/multi/mutex/mcs", test_multi_fair_mutex_1);
192
+#endif
193
+ g_test_add_func("/aio/multi/mutex/pthread", test_multi_mutex_1);
194
} else {
195
g_test_add_func("/aio/multi/schedule", test_multi_co_schedule_10);
196
g_test_add_func("/aio/multi/mutex/contended", test_multi_co_mutex_10);
197
g_test_add_func("/aio/multi/mutex/handoff", test_multi_co_mutex_2_30);
198
+#ifdef CONFIG_LINUX
199
+ g_test_add_func("/aio/multi/mutex/mcs", test_multi_fair_mutex_10);
200
+#endif
201
+ g_test_add_func("/aio/multi/mutex/pthread", test_multi_mutex_10);
202
}
203
return g_test_run();
204
}
205
--
206
2.9.3
207
208
diff view generated by jsdifflib
New patch
1
From: Paolo Bonzini <pbonzini@redhat.com>
1
2
3
This will avoid forward references in the next patch. It is also
4
more logical because CoQueue is not anymore the basic primitive.
5
6
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
7
Reviewed-by: Fam Zheng <famz@redhat.com>
8
Message-id: 20170213181244.16297-5-pbonzini@redhat.com
9
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
10
---
11
include/qemu/coroutine.h | 89 ++++++++++++++++++++++++------------------------
12
1 file changed, 44 insertions(+), 45 deletions(-)
13
14
diff --git a/include/qemu/coroutine.h b/include/qemu/coroutine.h
15
index XXXXXXX..XXXXXXX 100644
16
--- a/include/qemu/coroutine.h
17
+++ b/include/qemu/coroutine.h
18
@@ -XXX,XX +XXX,XX @@ bool qemu_in_coroutine(void);
19
*/
20
bool qemu_coroutine_entered(Coroutine *co);
21
22
-
23
-/**
24
- * CoQueues are a mechanism to queue coroutines in order to continue executing
25
- * them later. They provide the fundamental primitives on which coroutine locks
26
- * are built.
27
- */
28
-typedef struct CoQueue {
29
- QSIMPLEQ_HEAD(, Coroutine) entries;
30
-} CoQueue;
31
-
32
-/**
33
- * Initialise a CoQueue. This must be called before any other operation is used
34
- * on the CoQueue.
35
- */
36
-void qemu_co_queue_init(CoQueue *queue);
37
-
38
-/**
39
- * Adds the current coroutine to the CoQueue and transfers control to the
40
- * caller of the coroutine.
41
- */
42
-void coroutine_fn qemu_co_queue_wait(CoQueue *queue);
43
-
44
-/**
45
- * Restarts the next coroutine in the CoQueue and removes it from the queue.
46
- *
47
- * Returns true if a coroutine was restarted, false if the queue is empty.
48
- */
49
-bool coroutine_fn qemu_co_queue_next(CoQueue *queue);
50
-
51
-/**
52
- * Restarts all coroutines in the CoQueue and leaves the queue empty.
53
- */
54
-void coroutine_fn qemu_co_queue_restart_all(CoQueue *queue);
55
-
56
-/**
57
- * Enter the next coroutine in the queue
58
- */
59
-bool qemu_co_enter_next(CoQueue *queue);
60
-
61
-/**
62
- * Checks if the CoQueue is empty.
63
- */
64
-bool qemu_co_queue_empty(CoQueue *queue);
65
-
66
-
67
/**
68
* Provides a mutex that can be used to synchronise coroutines
69
*/
70
@@ -XXX,XX +XXX,XX @@ void coroutine_fn qemu_co_mutex_lock(CoMutex *mutex);
71
*/
72
void coroutine_fn qemu_co_mutex_unlock(CoMutex *mutex);
73
74
+
75
+/**
76
+ * CoQueues are a mechanism to queue coroutines in order to continue executing
77
+ * them later.
78
+ */
79
+typedef struct CoQueue {
80
+ QSIMPLEQ_HEAD(, Coroutine) entries;
81
+} CoQueue;
82
+
83
+/**
84
+ * Initialise a CoQueue. This must be called before any other operation is used
85
+ * on the CoQueue.
86
+ */
87
+void qemu_co_queue_init(CoQueue *queue);
88
+
89
+/**
90
+ * Adds the current coroutine to the CoQueue and transfers control to the
91
+ * caller of the coroutine.
92
+ */
93
+void coroutine_fn qemu_co_queue_wait(CoQueue *queue);
94
+
95
+/**
96
+ * Restarts the next coroutine in the CoQueue and removes it from the queue.
97
+ *
98
+ * Returns true if a coroutine was restarted, false if the queue is empty.
99
+ */
100
+bool coroutine_fn qemu_co_queue_next(CoQueue *queue);
101
+
102
+/**
103
+ * Restarts all coroutines in the CoQueue and leaves the queue empty.
104
+ */
105
+void coroutine_fn qemu_co_queue_restart_all(CoQueue *queue);
106
+
107
+/**
108
+ * Enter the next coroutine in the queue
109
+ */
110
+bool qemu_co_enter_next(CoQueue *queue);
111
+
112
+/**
113
+ * Checks if the CoQueue is empty.
114
+ */
115
+bool qemu_co_queue_empty(CoQueue *queue);
116
+
117
+
118
typedef struct CoRwlock {
119
bool writer;
120
int reader;
121
--
122
2.9.3
123
124
diff view generated by jsdifflib
1
From: Jeuk Kim <jeuk20.kim@samsung.com>
1
From: Paolo Bonzini <pbonzini@redhat.com>
2
2
3
This commit adds support for ufs logical unit.
3
All that CoQueue needs in order to become thread-safe is help
4
The LU handles processing for the SCSI command,
4
from an external mutex. Add this to the API.
5
unit descriptor query request.
5
6
6
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
7
This commit enables the UFS device to process
7
Reviewed-by: Fam Zheng <famz@redhat.com>
8
IO requests.
8
Message-id: 20170213181244.16297-6-pbonzini@redhat.com
9
10
Signed-off-by: Jeuk Kim <jeuk20.kim@samsung.com>
11
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
12
Message-id: beacc504376ab6a14b1a3830bb3c69382cf6aebc.1693980783.git.jeuk20.kim@gmail.com
13
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
9
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
14
---
10
---
15
hw/ufs/ufs.h | 43 ++
11
include/qemu/coroutine.h | 8 +++++---
16
include/scsi/constants.h | 1 +
12
block/backup.c | 2 +-
17
hw/ufs/lu.c | 1445 ++++++++++++++++++++++++++++++++++++++
13
block/io.c | 4 ++--
18
hw/ufs/ufs.c | 252 ++++++-
14
block/nbd-client.c | 2 +-
19
hw/ufs/meson.build | 2 +-
15
block/qcow2-cluster.c | 4 +---
20
hw/ufs/trace-events | 25 +
16
block/sheepdog.c | 2 +-
21
6 files changed, 1761 insertions(+), 7 deletions(-)
17
block/throttle-groups.c | 2 +-
22
create mode 100644 hw/ufs/lu.c
18
hw/9pfs/9p.c | 2 +-
23
19
util/qemu-coroutine-lock.c | 24 +++++++++++++++++++++---
24
diff --git a/hw/ufs/ufs.h b/hw/ufs/ufs.h
20
9 files changed, 34 insertions(+), 16 deletions(-)
25
index XXXXXXX..XXXXXXX 100644
21
26
--- a/hw/ufs/ufs.h
22
diff --git a/include/qemu/coroutine.h b/include/qemu/coroutine.h
27
+++ b/hw/ufs/ufs.h
23
index XXXXXXX..XXXXXXX 100644
28
@@ -XXX,XX +XXX,XX @@
24
--- a/include/qemu/coroutine.h
29
#define UFS_MAX_LUS 32
25
+++ b/include/qemu/coroutine.h
30
#define UFS_BLOCK_SIZE 4096
26
@@ -XXX,XX +XXX,XX @@ void coroutine_fn qemu_co_mutex_unlock(CoMutex *mutex);
31
27
32
+typedef struct UfsBusClass {
28
/**
33
+ BusClass parent_class;
29
* CoQueues are a mechanism to queue coroutines in order to continue executing
34
+ bool (*parent_check_address)(BusState *bus, DeviceState *dev, Error **errp);
30
- * them later.
35
+} UfsBusClass;
31
+ * them later. They are similar to condition variables, but they need help
32
+ * from an external mutex in order to maintain thread-safety.
33
*/
34
typedef struct CoQueue {
35
QSIMPLEQ_HEAD(, Coroutine) entries;
36
@@ -XXX,XX +XXX,XX @@ void qemu_co_queue_init(CoQueue *queue);
37
38
/**
39
* Adds the current coroutine to the CoQueue and transfers control to the
40
- * caller of the coroutine.
41
+ * caller of the coroutine. The mutex is unlocked during the wait and
42
+ * locked again afterwards.
43
*/
44
-void coroutine_fn qemu_co_queue_wait(CoQueue *queue);
45
+void coroutine_fn qemu_co_queue_wait(CoQueue *queue, CoMutex *mutex);
46
47
/**
48
* Restarts the next coroutine in the CoQueue and removes it from the queue.
49
diff --git a/block/backup.c b/block/backup.c
50
index XXXXXXX..XXXXXXX 100644
51
--- a/block/backup.c
52
+++ b/block/backup.c
53
@@ -XXX,XX +XXX,XX @@ static void coroutine_fn wait_for_overlapping_requests(BackupBlockJob *job,
54
retry = false;
55
QLIST_FOREACH(req, &job->inflight_reqs, list) {
56
if (end > req->start && start < req->end) {
57
- qemu_co_queue_wait(&req->wait_queue);
58
+ qemu_co_queue_wait(&req->wait_queue, NULL);
59
retry = true;
60
break;
61
}
62
diff --git a/block/io.c b/block/io.c
63
index XXXXXXX..XXXXXXX 100644
64
--- a/block/io.c
65
+++ b/block/io.c
66
@@ -XXX,XX +XXX,XX @@ static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
67
* (instead of producing a deadlock in the former case). */
68
if (!req->waiting_for) {
69
self->waiting_for = req;
70
- qemu_co_queue_wait(&req->wait_queue);
71
+ qemu_co_queue_wait(&req->wait_queue, NULL);
72
self->waiting_for = NULL;
73
retry = true;
74
waited = true;
75
@@ -XXX,XX +XXX,XX @@ int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
76
77
/* Wait until any previous flushes are completed */
78
while (bs->active_flush_req) {
79
- qemu_co_queue_wait(&bs->flush_queue);
80
+ qemu_co_queue_wait(&bs->flush_queue, NULL);
81
}
82
83
bs->active_flush_req = true;
84
diff --git a/block/nbd-client.c b/block/nbd-client.c
85
index XXXXXXX..XXXXXXX 100644
86
--- a/block/nbd-client.c
87
+++ b/block/nbd-client.c
88
@@ -XXX,XX +XXX,XX @@ static void nbd_coroutine_start(NBDClientSession *s,
89
/* Poor man semaphore. The free_sema is locked when no other request
90
* can be accepted, and unlocked after receiving one reply. */
91
if (s->in_flight == MAX_NBD_REQUESTS) {
92
- qemu_co_queue_wait(&s->free_sema);
93
+ qemu_co_queue_wait(&s->free_sema, NULL);
94
assert(s->in_flight < MAX_NBD_REQUESTS);
95
}
96
s->in_flight++;
97
diff --git a/block/qcow2-cluster.c b/block/qcow2-cluster.c
98
index XXXXXXX..XXXXXXX 100644
99
--- a/block/qcow2-cluster.c
100
+++ b/block/qcow2-cluster.c
101
@@ -XXX,XX +XXX,XX @@ static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset,
102
if (bytes == 0) {
103
/* Wait for the dependency to complete. We need to recheck
104
* the free/allocated clusters when we continue. */
105
- qemu_co_mutex_unlock(&s->lock);
106
- qemu_co_queue_wait(&old_alloc->dependent_requests);
107
- qemu_co_mutex_lock(&s->lock);
108
+ qemu_co_queue_wait(&old_alloc->dependent_requests, &s->lock);
109
return -EAGAIN;
110
}
111
}
112
diff --git a/block/sheepdog.c b/block/sheepdog.c
113
index XXXXXXX..XXXXXXX 100644
114
--- a/block/sheepdog.c
115
+++ b/block/sheepdog.c
116
@@ -XXX,XX +XXX,XX @@ static void wait_for_overlapping_aiocb(BDRVSheepdogState *s, SheepdogAIOCB *acb)
117
retry:
118
QLIST_FOREACH(cb, &s->inflight_aiocb_head, aiocb_siblings) {
119
if (AIOCBOverlapping(acb, cb)) {
120
- qemu_co_queue_wait(&s->overlapping_queue);
121
+ qemu_co_queue_wait(&s->overlapping_queue, NULL);
122
goto retry;
123
}
124
}
125
diff --git a/block/throttle-groups.c b/block/throttle-groups.c
126
index XXXXXXX..XXXXXXX 100644
127
--- a/block/throttle-groups.c
128
+++ b/block/throttle-groups.c
129
@@ -XXX,XX +XXX,XX @@ void coroutine_fn throttle_group_co_io_limits_intercept(BlockBackend *blk,
130
if (must_wait || blkp->pending_reqs[is_write]) {
131
blkp->pending_reqs[is_write]++;
132
qemu_mutex_unlock(&tg->lock);
133
- qemu_co_queue_wait(&blkp->throttled_reqs[is_write]);
134
+ qemu_co_queue_wait(&blkp->throttled_reqs[is_write], NULL);
135
qemu_mutex_lock(&tg->lock);
136
blkp->pending_reqs[is_write]--;
137
}
138
diff --git a/hw/9pfs/9p.c b/hw/9pfs/9p.c
139
index XXXXXXX..XXXXXXX 100644
140
--- a/hw/9pfs/9p.c
141
+++ b/hw/9pfs/9p.c
142
@@ -XXX,XX +XXX,XX @@ static void coroutine_fn v9fs_flush(void *opaque)
143
/*
144
* Wait for pdu to complete.
145
*/
146
- qemu_co_queue_wait(&cancel_pdu->complete);
147
+ qemu_co_queue_wait(&cancel_pdu->complete, NULL);
148
cancel_pdu->cancelled = 0;
149
pdu_free(cancel_pdu);
150
}
151
diff --git a/util/qemu-coroutine-lock.c b/util/qemu-coroutine-lock.c
152
index XXXXXXX..XXXXXXX 100644
153
--- a/util/qemu-coroutine-lock.c
154
+++ b/util/qemu-coroutine-lock.c
155
@@ -XXX,XX +XXX,XX @@ void qemu_co_queue_init(CoQueue *queue)
156
QSIMPLEQ_INIT(&queue->entries);
157
}
158
159
-void coroutine_fn qemu_co_queue_wait(CoQueue *queue)
160
+void coroutine_fn qemu_co_queue_wait(CoQueue *queue, CoMutex *mutex)
161
{
162
Coroutine *self = qemu_coroutine_self();
163
QSIMPLEQ_INSERT_TAIL(&queue->entries, self, co_queue_next);
36
+
164
+
37
+typedef struct UfsBus {
165
+ if (mutex) {
38
+ SCSIBus parent_bus;
166
+ qemu_co_mutex_unlock(mutex);
39
+} UfsBus;
40
+
41
+#define TYPE_UFS_BUS "ufs-bus"
42
+DECLARE_OBJ_CHECKERS(UfsBus, UfsBusClass, UFS_BUS, TYPE_UFS_BUS)
43
+
44
typedef enum UfsRequestState {
45
UFS_REQUEST_IDLE = 0,
46
UFS_REQUEST_READY = 1,
47
@@ -XXX,XX +XXX,XX @@ typedef enum UfsRequestState {
48
typedef enum UfsReqResult {
49
UFS_REQUEST_SUCCESS = 0,
50
UFS_REQUEST_FAIL = 1,
51
+ UFS_REQUEST_NO_COMPLETE = 2,
52
} UfsReqResult;
53
54
typedef struct UfsRequest {
55
@@ -XXX,XX +XXX,XX @@ typedef struct UfsRequest {
56
QEMUSGList *sg;
57
} UfsRequest;
58
59
+typedef struct UfsLu {
60
+ SCSIDevice qdev;
61
+ uint8_t lun;
62
+ UnitDescriptor unit_desc;
63
+} UfsLu;
64
+
65
+typedef struct UfsWLu {
66
+ SCSIDevice qdev;
67
+ uint8_t lun;
68
+} UfsWLu;
69
+
70
typedef struct UfsParams {
71
char *serial;
72
uint8_t nutrs; /* Number of UTP Transfer Request Slots */
73
@@ -XXX,XX +XXX,XX @@ typedef struct UfsParams {
74
75
typedef struct UfsHc {
76
PCIDevice parent_obj;
77
+ UfsBus bus;
78
MemoryRegion iomem;
79
UfsReg reg;
80
UfsParams params;
81
uint32_t reg_size;
82
UfsRequest *req_list;
83
84
+ UfsLu *lus[UFS_MAX_LUS];
85
+ UfsWLu *report_wlu;
86
+ UfsWLu *dev_wlu;
87
+ UfsWLu *boot_wlu;
88
+ UfsWLu *rpmb_wlu;
89
DeviceDescriptor device_desc;
90
GeometryDescriptor geometry_desc;
91
Attributes attributes;
92
@@ -XXX,XX +XXX,XX @@ typedef struct UfsHc {
93
#define TYPE_UFS "ufs"
94
#define UFS(obj) OBJECT_CHECK(UfsHc, (obj), TYPE_UFS)
95
96
+#define TYPE_UFS_LU "ufs-lu"
97
+#define UFSLU(obj) OBJECT_CHECK(UfsLu, (obj), TYPE_UFS_LU)
98
+
99
+#define TYPE_UFS_WLU "ufs-wlu"
100
+#define UFSWLU(obj) OBJECT_CHECK(UfsWLu, (obj), TYPE_UFS_WLU)
101
+
102
typedef enum UfsQueryFlagPerm {
103
UFS_QUERY_FLAG_NONE = 0x0,
104
UFS_QUERY_FLAG_READ = 0x1,
105
@@ -XXX,XX +XXX,XX @@ typedef enum UfsQueryAttrPerm {
106
UFS_QUERY_ATTR_WRITE = 0x2,
107
} UfsQueryAttrPerm;
108
109
+static inline bool is_wlun(uint8_t lun)
110
+{
111
+ return (lun == UFS_UPIU_REPORT_LUNS_WLUN ||
112
+ lun == UFS_UPIU_UFS_DEVICE_WLUN || lun == UFS_UPIU_BOOT_WLUN ||
113
+ lun == UFS_UPIU_RPMB_WLUN);
114
+}
115
+
116
#endif /* HW_UFS_UFS_H */
117
diff --git a/include/scsi/constants.h b/include/scsi/constants.h
118
index XXXXXXX..XXXXXXX 100644
119
--- a/include/scsi/constants.h
120
+++ b/include/scsi/constants.h
121
@@ -XXX,XX +XXX,XX @@
122
#define MODE_PAGE_FLEXIBLE_DISK_GEOMETRY 0x05
123
#define MODE_PAGE_CACHING 0x08
124
#define MODE_PAGE_AUDIO_CTL 0x0e
125
+#define MODE_PAGE_CONTROL 0x0a
126
#define MODE_PAGE_POWER 0x1a
127
#define MODE_PAGE_FAULT_FAIL 0x1c
128
#define MODE_PAGE_TO_PROTECT 0x1d
129
diff --git a/hw/ufs/lu.c b/hw/ufs/lu.c
130
new file mode 100644
131
index XXXXXXX..XXXXXXX
132
--- /dev/null
133
+++ b/hw/ufs/lu.c
134
@@ -XXX,XX +XXX,XX @@
135
+/*
136
+ * QEMU UFS Logical Unit
137
+ *
138
+ * Copyright (c) 2023 Samsung Electronics Co., Ltd. All rights reserved.
139
+ *
140
+ * Written by Jeuk Kim <jeuk20.kim@samsung.com>
141
+ *
142
+ * This code is licensed under the GNU GPL v2 or later.
143
+ */
144
+
145
+#include "qemu/osdep.h"
146
+#include "qemu/units.h"
147
+#include "qapi/error.h"
148
+#include "qemu/memalign.h"
149
+#include "hw/scsi/scsi.h"
150
+#include "scsi/constants.h"
151
+#include "sysemu/block-backend.h"
152
+#include "qemu/cutils.h"
153
+#include "trace.h"
154
+#include "ufs.h"
155
+
156
+/*
157
+ * The code below handling SCSI commands is copied from hw/scsi/scsi-disk.c,
158
+ * with minor adjustments to make it work for UFS.
159
+ */
160
+
161
+#define SCSI_DMA_BUF_SIZE (128 * KiB)
162
+#define SCSI_MAX_INQUIRY_LEN 256
163
+#define SCSI_INQUIRY_DATA_SIZE 36
164
+#define SCSI_MAX_MODE_LEN 256
165
+
166
+typedef struct UfsSCSIReq {
167
+ SCSIRequest req;
168
+ /* Both sector and sector_count are in terms of BDRV_SECTOR_SIZE bytes. */
169
+ uint64_t sector;
170
+ uint32_t sector_count;
171
+ uint32_t buflen;
172
+ bool started;
173
+ bool need_fua_emulation;
174
+ struct iovec iov;
175
+ QEMUIOVector qiov;
176
+ BlockAcctCookie acct;
177
+} UfsSCSIReq;
178
+
179
+static void ufs_scsi_free_request(SCSIRequest *req)
180
+{
181
+ UfsSCSIReq *r = DO_UPCAST(UfsSCSIReq, req, req);
182
+
183
+ qemu_vfree(r->iov.iov_base);
184
+}
185
+
186
+static void scsi_check_condition(UfsSCSIReq *r, SCSISense sense)
187
+{
188
+ trace_ufs_scsi_check_condition(r->req.tag, sense.key, sense.asc,
189
+ sense.ascq);
190
+ scsi_req_build_sense(&r->req, sense);
191
+ scsi_req_complete(&r->req, CHECK_CONDITION);
192
+}
193
+
194
+static int ufs_scsi_emulate_vpd_page(SCSIRequest *req, uint8_t *outbuf,
195
+ uint32_t outbuf_len)
196
+{
197
+ UfsHc *u = UFS(req->bus->qbus.parent);
198
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, req->dev);
199
+ uint8_t page_code = req->cmd.buf[2];
200
+ int start, buflen = 0;
201
+
202
+ if (outbuf_len < SCSI_INQUIRY_DATA_SIZE) {
203
+ return -1;
204
+ }
167
+ }
205
+
168
+
206
+ outbuf[buflen++] = lu->qdev.type & 0x1f;
169
+ /* There is no race condition here. Other threads will call
207
+ outbuf[buflen++] = page_code;
170
+ * aio_co_schedule on our AioContext, which can reenter this
208
+ outbuf[buflen++] = 0x00;
171
+ * coroutine but only after this yield and after the main loop
209
+ outbuf[buflen++] = 0x00;
172
+ * has gone through the next iteration.
210
+ start = buflen;
173
+ */
174
qemu_coroutine_yield();
175
assert(qemu_in_coroutine());
211
+
176
+
212
+ switch (page_code) {
177
+ /* TODO: OSv implements wait morphing here, where the wakeup
213
+ case 0x00: /* Supported page codes, mandatory */
178
+ * primitive automatically places the woken coroutine on the
214
+ {
179
+ * mutex's queue. This avoids the thundering herd effect.
215
+ trace_ufs_scsi_emulate_vpd_page_00(req->cmd.xfer);
216
+ outbuf[buflen++] = 0x00; /* list of supported pages (this page) */
217
+ if (u->params.serial) {
218
+ outbuf[buflen++] = 0x80; /* unit serial number */
219
+ }
220
+ outbuf[buflen++] = 0x87; /* mode page policy */
221
+ break;
222
+ }
223
+ case 0x80: /* Device serial number, optional */
224
+ {
225
+ int l;
226
+
227
+ if (!u->params.serial) {
228
+ trace_ufs_scsi_emulate_vpd_page_80_not_supported();
229
+ return -1;
230
+ }
231
+
232
+ l = strlen(u->params.serial);
233
+ if (l > SCSI_INQUIRY_DATA_SIZE) {
234
+ l = SCSI_INQUIRY_DATA_SIZE;
235
+ }
236
+
237
+ trace_ufs_scsi_emulate_vpd_page_80(req->cmd.xfer);
238
+ memcpy(outbuf + buflen, u->params.serial, l);
239
+ buflen += l;
240
+ break;
241
+ }
242
+ case 0x87: /* Mode Page Policy, mandatory */
243
+ {
244
+ trace_ufs_scsi_emulate_vpd_page_87(req->cmd.xfer);
245
+ outbuf[buflen++] = 0x3f; /* apply to all mode pages and subpages */
246
+ outbuf[buflen++] = 0xff;
247
+ outbuf[buflen++] = 0; /* shared */
248
+ outbuf[buflen++] = 0;
249
+ break;
250
+ }
251
+ default:
252
+ return -1;
253
+ }
254
+ /* done with EVPD */
255
+ assert(buflen - start <= 255);
256
+ outbuf[start - 1] = buflen - start;
257
+ return buflen;
258
+}
259
+
260
+static int ufs_scsi_emulate_inquiry(SCSIRequest *req, uint8_t *outbuf,
261
+ uint32_t outbuf_len)
262
+{
263
+ int buflen = 0;
264
+
265
+ if (outbuf_len < SCSI_INQUIRY_DATA_SIZE) {
266
+ return -1;
267
+ }
268
+
269
+ if (req->cmd.buf[1] & 0x1) {
270
+ /* Vital product data */
271
+ return ufs_scsi_emulate_vpd_page(req, outbuf, outbuf_len);
272
+ }
273
+
274
+ /* Standard INQUIRY data */
275
+ if (req->cmd.buf[2] != 0) {
276
+ return -1;
277
+ }
278
+
279
+ /* PAGE CODE == 0 */
280
+ buflen = req->cmd.xfer;
281
+ if (buflen > SCSI_MAX_INQUIRY_LEN) {
282
+ buflen = SCSI_MAX_INQUIRY_LEN;
283
+ }
284
+
285
+ if (is_wlun(req->lun)) {
286
+ outbuf[0] = TYPE_WLUN;
287
+ } else {
288
+ outbuf[0] = 0;
289
+ }
290
+ outbuf[1] = 0;
291
+
292
+ strpadcpy((char *)&outbuf[16], 16, "QEMU UFS", ' ');
293
+ strpadcpy((char *)&outbuf[8], 8, "QEMU", ' ');
294
+
295
+ memset(&outbuf[32], 0, 4);
296
+
297
+ outbuf[2] = 0x06; /* SPC-4 */
298
+ outbuf[3] = 0x2;
299
+
300
+ if (buflen > SCSI_INQUIRY_DATA_SIZE) {
301
+ outbuf[4] = buflen - 5; /* Additional Length = (Len - 1) - 4 */
302
+ } else {
303
+ /*
304
+ * If the allocation length of CDB is too small, the additional
305
+ * length is not adjusted
306
+ */
307
+ outbuf[4] = SCSI_INQUIRY_DATA_SIZE - 5;
308
+ }
309
+
310
+ /* Support TCQ. */
311
+ outbuf[7] = req->bus->info->tcq ? 0x02 : 0;
312
+ return buflen;
313
+}
314
+
315
+static int mode_sense_page(UfsLu *lu, int page, uint8_t **p_outbuf,
316
+ int page_control)
317
+{
318
+ static const int mode_sense_valid[0x3f] = {
319
+ [MODE_PAGE_CACHING] = 1,
320
+ [MODE_PAGE_R_W_ERROR] = 1,
321
+ [MODE_PAGE_CONTROL] = 1,
322
+ };
323
+
324
+ uint8_t *p = *p_outbuf + 2;
325
+ int length;
326
+
327
+ assert(page < ARRAY_SIZE(mode_sense_valid));
328
+ if ((mode_sense_valid[page]) == 0) {
329
+ return -1;
330
+ }
331
+
332
+ /*
333
+ * If Changeable Values are requested, a mask denoting those mode parameters
334
+ * that are changeable shall be returned. As we currently don't support
335
+ * parameter changes via MODE_SELECT all bits are returned set to zero.
336
+ * The buffer was already memset to zero by the caller of this function.
337
+ */
180
+ */
338
+ switch (page) {
181
+ if (mutex) {
339
+ case MODE_PAGE_CACHING:
182
+ qemu_co_mutex_lock(mutex);
340
+ length = 0x12;
341
+ if (page_control == 1 || /* Changeable Values */
342
+ blk_enable_write_cache(lu->qdev.conf.blk)) {
343
+ p[0] = 4; /* WCE */
344
+ }
345
+ break;
346
+
347
+ case MODE_PAGE_R_W_ERROR:
348
+ length = 10;
349
+ if (page_control == 1) { /* Changeable Values */
350
+ break;
351
+ }
352
+ p[0] = 0x80; /* Automatic Write Reallocation Enabled */
353
+ break;
354
+
355
+ case MODE_PAGE_CONTROL:
356
+ length = 10;
357
+ if (page_control == 1) { /* Changeable Values */
358
+ break;
359
+ }
360
+ p[1] = 0x10; /* Queue Algorithm modifier */
361
+ p[8] = 0xff; /* Busy Timeout Period */
362
+ p[9] = 0xff;
363
+ break;
364
+
365
+ default:
366
+ return -1;
367
+ }
368
+
369
+ assert(length < 256);
370
+ (*p_outbuf)[0] = page;
371
+ (*p_outbuf)[1] = length;
372
+ *p_outbuf += length + 2;
373
+ return length + 2;
374
+}
375
+
376
+static int ufs_scsi_emulate_mode_sense(UfsSCSIReq *r, uint8_t *outbuf)
377
+{
378
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, r->req.dev);
379
+ bool dbd;
380
+ int page, buflen, ret, page_control;
381
+ uint8_t *p;
382
+ uint8_t dev_specific_param = 0;
383
+
384
+ dbd = (r->req.cmd.buf[1] & 0x8) != 0;
385
+ if (!dbd) {
386
+ return -1;
387
+ }
388
+
389
+ page = r->req.cmd.buf[2] & 0x3f;
390
+ page_control = (r->req.cmd.buf[2] & 0xc0) >> 6;
391
+
392
+ trace_ufs_scsi_emulate_mode_sense((r->req.cmd.buf[0] == MODE_SENSE) ? 6 :
393
+ 10,
394
+ page, r->req.cmd.xfer, page_control);
395
+ memset(outbuf, 0, r->req.cmd.xfer);
396
+ p = outbuf;
397
+
398
+ if (!blk_is_writable(lu->qdev.conf.blk)) {
399
+ dev_specific_param |= 0x80; /* Readonly. */
400
+ }
401
+
402
+ p[2] = 0; /* Medium type. */
403
+ p[3] = dev_specific_param;
404
+ p[6] = p[7] = 0; /* Block descriptor length. */
405
+ p += 8;
406
+
407
+ if (page_control == 3) {
408
+ /* Saved Values */
409
+ scsi_check_condition(r, SENSE_CODE(SAVING_PARAMS_NOT_SUPPORTED));
410
+ return -1;
411
+ }
412
+
413
+ if (page == 0x3f) {
414
+ for (page = 0; page <= 0x3e; page++) {
415
+ mode_sense_page(lu, page, &p, page_control);
416
+ }
417
+ } else {
418
+ ret = mode_sense_page(lu, page, &p, page_control);
419
+ if (ret == -1) {
420
+ return -1;
421
+ }
422
+ }
423
+
424
+ buflen = p - outbuf;
425
+ /*
426
+ * The mode data length field specifies the length in bytes of the
427
+ * following data that is available to be transferred. The mode data
428
+ * length does not include itself.
429
+ */
430
+ outbuf[0] = ((buflen - 2) >> 8) & 0xff;
431
+ outbuf[1] = (buflen - 2) & 0xff;
432
+ return buflen;
433
+}
434
+
435
+/*
436
+ * scsi_handle_rw_error has two return values. False means that the error
437
+ * must be ignored, true means that the error has been processed and the
438
+ * caller should not do anything else for this request. Note that
439
+ * scsi_handle_rw_error always manages its reference counts, independent
440
+ * of the return value.
441
+ */
442
+static bool scsi_handle_rw_error(UfsSCSIReq *r, int ret, bool acct_failed)
443
+{
444
+ bool is_read = (r->req.cmd.mode == SCSI_XFER_FROM_DEV);
445
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, r->req.dev);
446
+ SCSISense sense = SENSE_CODE(NO_SENSE);
447
+ int error = 0;
448
+ bool req_has_sense = false;
449
+ BlockErrorAction action;
450
+ int status;
451
+
452
+ if (ret < 0) {
453
+ status = scsi_sense_from_errno(-ret, &sense);
454
+ error = -ret;
455
+ } else {
456
+ /* A passthrough command has completed with nonzero status. */
457
+ status = ret;
458
+ if (status == CHECK_CONDITION) {
459
+ req_has_sense = true;
460
+ error = scsi_sense_buf_to_errno(r->req.sense, sizeof(r->req.sense));
461
+ } else {
462
+ error = EINVAL;
463
+ }
464
+ }
465
+
466
+ /*
467
+ * Check whether the error has to be handled by the guest or should
468
+ * rather follow the rerror=/werror= settings. Guest-handled errors
469
+ * are usually retried immediately, so do not post them to QMP and
470
+ * do not account them as failed I/O.
471
+ */
472
+ if (req_has_sense && scsi_sense_buf_is_guest_recoverable(
473
+ r->req.sense, sizeof(r->req.sense))) {
474
+ action = BLOCK_ERROR_ACTION_REPORT;
475
+ acct_failed = false;
476
+ } else {
477
+ action = blk_get_error_action(lu->qdev.conf.blk, is_read, error);
478
+ blk_error_action(lu->qdev.conf.blk, action, is_read, error);
479
+ }
480
+
481
+ switch (action) {
482
+ case BLOCK_ERROR_ACTION_REPORT:
483
+ if (acct_failed) {
484
+ block_acct_failed(blk_get_stats(lu->qdev.conf.blk), &r->acct);
485
+ }
486
+ if (!req_has_sense && status == CHECK_CONDITION) {
487
+ scsi_req_build_sense(&r->req, sense);
488
+ }
489
+ scsi_req_complete(&r->req, status);
490
+ return true;
491
+
492
+ case BLOCK_ERROR_ACTION_IGNORE:
493
+ return false;
494
+
495
+ case BLOCK_ERROR_ACTION_STOP:
496
+ scsi_req_retry(&r->req);
497
+ return true;
498
+
499
+ default:
500
+ g_assert_not_reached();
501
+ }
502
+}
503
+
504
+static bool ufs_scsi_req_check_error(UfsSCSIReq *r, int ret, bool acct_failed)
505
+{
506
+ if (r->req.io_canceled) {
507
+ scsi_req_cancel_complete(&r->req);
508
+ return true;
509
+ }
510
+
511
+ if (ret < 0) {
512
+ return scsi_handle_rw_error(r, ret, acct_failed);
513
+ }
514
+
515
+ return false;
516
+}
517
+
518
+static void scsi_aio_complete(void *opaque, int ret)
519
+{
520
+ UfsSCSIReq *r = (UfsSCSIReq *)opaque;
521
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, r->req.dev);
522
+
523
+ assert(r->req.aiocb != NULL);
524
+ r->req.aiocb = NULL;
525
+ aio_context_acquire(blk_get_aio_context(lu->qdev.conf.blk));
526
+ if (ufs_scsi_req_check_error(r, ret, true)) {
527
+ goto done;
528
+ }
529
+
530
+ block_acct_done(blk_get_stats(lu->qdev.conf.blk), &r->acct);
531
+ scsi_req_complete(&r->req, GOOD);
532
+
533
+done:
534
+ aio_context_release(blk_get_aio_context(lu->qdev.conf.blk));
535
+ scsi_req_unref(&r->req);
536
+}
537
+
538
+static int32_t ufs_scsi_emulate_command(SCSIRequest *req, uint8_t *buf)
539
+{
540
+ UfsSCSIReq *r = DO_UPCAST(UfsSCSIReq, req, req);
541
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, req->dev);
542
+ uint32_t last_block = 0;
543
+ uint8_t *outbuf;
544
+ int buflen;
545
+
546
+ switch (req->cmd.buf[0]) {
547
+ case INQUIRY:
548
+ case MODE_SENSE_10:
549
+ case START_STOP:
550
+ case REQUEST_SENSE:
551
+ break;
552
+
553
+ default:
554
+ if (!blk_is_available(lu->qdev.conf.blk)) {
555
+ scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
556
+ return 0;
557
+ }
558
+ break;
559
+ }
560
+
561
+ /*
562
+ * FIXME: we shouldn't return anything bigger than 4k, but the code
563
+ * requires the buffer to be as big as req->cmd.xfer in several
564
+ * places. So, do not allow CDBs with a very large ALLOCATION
565
+ * LENGTH. The real fix would be to modify scsi_read_data and
566
+ * dma_buf_read, so that they return data beyond the buflen
567
+ * as all zeros.
568
+ */
569
+ if (req->cmd.xfer > 65536) {
570
+ goto illegal_request;
571
+ }
572
+ r->buflen = MAX(4096, req->cmd.xfer);
573
+
574
+ if (!r->iov.iov_base) {
575
+ r->iov.iov_base = blk_blockalign(lu->qdev.conf.blk, r->buflen);
576
+ }
577
+
578
+ outbuf = r->iov.iov_base;
579
+ memset(outbuf, 0, r->buflen);
580
+ switch (req->cmd.buf[0]) {
581
+ case TEST_UNIT_READY:
582
+ assert(blk_is_available(lu->qdev.conf.blk));
583
+ break;
584
+ case INQUIRY:
585
+ buflen = ufs_scsi_emulate_inquiry(req, outbuf, r->buflen);
586
+ if (buflen < 0) {
587
+ goto illegal_request;
588
+ }
589
+ break;
590
+ case MODE_SENSE_10:
591
+ buflen = ufs_scsi_emulate_mode_sense(r, outbuf);
592
+ if (buflen < 0) {
593
+ goto illegal_request;
594
+ }
595
+ break;
596
+ case READ_CAPACITY_10:
597
+ /* The normal LEN field for this command is zero. */
598
+ memset(outbuf, 0, 8);
599
+ if (lu->qdev.max_lba > 0) {
600
+ last_block = lu->qdev.max_lba - 1;
601
+ };
602
+ outbuf[0] = (last_block >> 24) & 0xff;
603
+ outbuf[1] = (last_block >> 16) & 0xff;
604
+ outbuf[2] = (last_block >> 8) & 0xff;
605
+ outbuf[3] = last_block & 0xff;
606
+ outbuf[4] = (lu->qdev.blocksize >> 24) & 0xff;
607
+ outbuf[5] = (lu->qdev.blocksize >> 16) & 0xff;
608
+ outbuf[6] = (lu->qdev.blocksize >> 8) & 0xff;
609
+ outbuf[7] = lu->qdev.blocksize & 0xff;
610
+ break;
611
+ case REQUEST_SENSE:
612
+ /* Just return "NO SENSE". */
613
+ buflen = scsi_convert_sense(NULL, 0, outbuf, r->buflen,
614
+ (req->cmd.buf[1] & 1) == 0);
615
+ if (buflen < 0) {
616
+ goto illegal_request;
617
+ }
618
+ break;
619
+ case SYNCHRONIZE_CACHE:
620
+ /* The request is used as the AIO opaque value, so add a ref. */
621
+ scsi_req_ref(&r->req);
622
+ block_acct_start(blk_get_stats(lu->qdev.conf.blk), &r->acct, 0,
623
+ BLOCK_ACCT_FLUSH);
624
+ r->req.aiocb = blk_aio_flush(lu->qdev.conf.blk, scsi_aio_complete, r);
625
+ return 0;
626
+ case VERIFY_10:
627
+ trace_ufs_scsi_emulate_command_VERIFY((req->cmd.buf[1] >> 1) & 3);
628
+ if (req->cmd.buf[1] & 6) {
629
+ goto illegal_request;
630
+ }
631
+ break;
632
+ case SERVICE_ACTION_IN_16:
633
+ /* Service Action In subcommands. */
634
+ if ((req->cmd.buf[1] & 31) == SAI_READ_CAPACITY_16) {
635
+ trace_ufs_scsi_emulate_command_SAI_16();
636
+ memset(outbuf, 0, req->cmd.xfer);
637
+
638
+ if (lu->qdev.max_lba > 0) {
639
+ last_block = lu->qdev.max_lba - 1;
640
+ };
641
+ outbuf[0] = 0;
642
+ outbuf[1] = 0;
643
+ outbuf[2] = 0;
644
+ outbuf[3] = 0;
645
+ outbuf[4] = (last_block >> 24) & 0xff;
646
+ outbuf[5] = (last_block >> 16) & 0xff;
647
+ outbuf[6] = (last_block >> 8) & 0xff;
648
+ outbuf[7] = last_block & 0xff;
649
+ outbuf[8] = (lu->qdev.blocksize >> 24) & 0xff;
650
+ outbuf[9] = (lu->qdev.blocksize >> 16) & 0xff;
651
+ outbuf[10] = (lu->qdev.blocksize >> 8) & 0xff;
652
+ outbuf[11] = lu->qdev.blocksize & 0xff;
653
+ outbuf[12] = 0;
654
+ outbuf[13] = get_physical_block_exp(&lu->qdev.conf);
655
+
656
+ if (lu->unit_desc.provisioning_type == 2 ||
657
+ lu->unit_desc.provisioning_type == 3) {
658
+ outbuf[14] = 0x80;
659
+ }
660
+ /* Protection, exponent and lowest lba field left blank. */
661
+ break;
662
+ }
663
+ trace_ufs_scsi_emulate_command_SAI_unsupported();
664
+ goto illegal_request;
665
+ case MODE_SELECT_10:
666
+ trace_ufs_scsi_emulate_command_MODE_SELECT_10(r->req.cmd.xfer);
667
+ break;
668
+ case START_STOP:
669
+ /*
670
+ * TODO: START_STOP is not yet implemented. It always returns success.
671
+ * Revisit it when ufs power management is implemented.
672
+ */
673
+ trace_ufs_scsi_emulate_command_START_STOP();
674
+ break;
675
+ case FORMAT_UNIT:
676
+ trace_ufs_scsi_emulate_command_FORMAT_UNIT();
677
+ break;
678
+ case SEND_DIAGNOSTIC:
679
+ trace_ufs_scsi_emulate_command_SEND_DIAGNOSTIC();
680
+ break;
681
+ default:
682
+ trace_ufs_scsi_emulate_command_UNKNOWN(buf[0],
683
+ scsi_command_name(buf[0]));
684
+ scsi_check_condition(r, SENSE_CODE(INVALID_OPCODE));
685
+ return 0;
686
+ }
687
+ assert(!r->req.aiocb);
688
+ r->iov.iov_len = MIN(r->buflen, req->cmd.xfer);
689
+ if (r->iov.iov_len == 0) {
690
+ scsi_req_complete(&r->req, GOOD);
691
+ }
692
+ if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
693
+ assert(r->iov.iov_len == req->cmd.xfer);
694
+ return -r->iov.iov_len;
695
+ } else {
696
+ return r->iov.iov_len;
697
+ }
698
+
699
+illegal_request:
700
+ if (r->req.status == -1) {
701
+ scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
702
+ }
703
+ return 0;
704
+}
705
+
706
+static void ufs_scsi_emulate_read_data(SCSIRequest *req)
707
+{
708
+ UfsSCSIReq *r = DO_UPCAST(UfsSCSIReq, req, req);
709
+ int buflen = r->iov.iov_len;
710
+
711
+ if (buflen) {
712
+ trace_ufs_scsi_emulate_read_data(buflen);
713
+ r->iov.iov_len = 0;
714
+ r->started = true;
715
+ scsi_req_data(&r->req, buflen);
716
+ return;
717
+ }
718
+
719
+ /* This also clears the sense buffer for REQUEST SENSE. */
720
+ scsi_req_complete(&r->req, GOOD);
721
+}
722
+
723
+static int ufs_scsi_check_mode_select(UfsLu *lu, int page, uint8_t *inbuf,
724
+ int inlen)
725
+{
726
+ uint8_t mode_current[SCSI_MAX_MODE_LEN];
727
+ uint8_t mode_changeable[SCSI_MAX_MODE_LEN];
728
+ uint8_t *p;
729
+ int len, expected_len, changeable_len, i;
730
+
731
+ /*
732
+ * The input buffer does not include the page header, so it is
733
+ * off by 2 bytes.
734
+ */
735
+ expected_len = inlen + 2;
736
+ if (expected_len > SCSI_MAX_MODE_LEN) {
737
+ return -1;
738
+ }
739
+
740
+ /* MODE_PAGE_ALLS is only valid for MODE SENSE commands */
741
+ if (page == MODE_PAGE_ALLS) {
742
+ return -1;
743
+ }
744
+
745
+ p = mode_current;
746
+ memset(mode_current, 0, inlen + 2);
747
+ len = mode_sense_page(lu, page, &p, 0);
748
+ if (len < 0 || len != expected_len) {
749
+ return -1;
750
+ }
751
+
752
+ p = mode_changeable;
753
+ memset(mode_changeable, 0, inlen + 2);
754
+ changeable_len = mode_sense_page(lu, page, &p, 1);
755
+ assert(changeable_len == len);
756
+
757
+ /*
758
+ * Check that unchangeable bits are the same as what MODE SENSE
759
+ * would return.
760
+ */
761
+ for (i = 2; i < len; i++) {
762
+ if (((mode_current[i] ^ inbuf[i - 2]) & ~mode_changeable[i]) != 0) {
763
+ return -1;
764
+ }
765
+ }
766
+ return 0;
767
+}
768
+
769
+static void ufs_scsi_apply_mode_select(UfsLu *lu, int page, uint8_t *p)
770
+{
771
+ switch (page) {
772
+ case MODE_PAGE_CACHING:
773
+ blk_set_enable_write_cache(lu->qdev.conf.blk, (p[0] & 4) != 0);
774
+ break;
775
+
776
+ default:
777
+ break;
778
+ }
779
+}
780
+
781
+static int mode_select_pages(UfsSCSIReq *r, uint8_t *p, int len, bool change)
782
+{
783
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, r->req.dev);
784
+
785
+ while (len > 0) {
786
+ int page, page_len;
787
+
788
+ page = p[0] & 0x3f;
789
+ if (p[0] & 0x40) {
790
+ goto invalid_param;
791
+ } else {
792
+ if (len < 2) {
793
+ goto invalid_param_len;
794
+ }
795
+ page_len = p[1];
796
+ p += 2;
797
+ len -= 2;
798
+ }
799
+
800
+ if (page_len > len) {
801
+ goto invalid_param_len;
802
+ }
803
+
804
+ if (!change) {
805
+ if (ufs_scsi_check_mode_select(lu, page, p, page_len) < 0) {
806
+ goto invalid_param;
807
+ }
808
+ } else {
809
+ ufs_scsi_apply_mode_select(lu, page, p);
810
+ }
811
+
812
+ p += page_len;
813
+ len -= page_len;
814
+ }
815
+ return 0;
816
+
817
+invalid_param:
818
+ scsi_check_condition(r, SENSE_CODE(INVALID_PARAM));
819
+ return -1;
820
+
821
+invalid_param_len:
822
+ scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
823
+ return -1;
824
+}
825
+
826
+static void ufs_scsi_emulate_mode_select(UfsSCSIReq *r, uint8_t *inbuf)
827
+{
828
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, r->req.dev);
829
+ uint8_t *p = inbuf;
830
+ int len = r->req.cmd.xfer;
831
+ int hdr_len = 8;
832
+ int bd_len;
833
+ int pass;
834
+
835
+ /* We only support PF=1, SP=0. */
836
+ if ((r->req.cmd.buf[1] & 0x11) != 0x10) {
837
+ goto invalid_field;
838
+ }
839
+
840
+ if (len < hdr_len) {
841
+ goto invalid_param_len;
842
+ }
843
+
844
+ bd_len = lduw_be_p(&p[6]);
845
+ if (bd_len != 0) {
846
+ goto invalid_param;
847
+ }
848
+
849
+ len -= hdr_len;
850
+ p += hdr_len;
851
+
852
+ /* Ensure no change is made if there is an error! */
853
+ for (pass = 0; pass < 2; pass++) {
854
+ if (mode_select_pages(r, p, len, pass == 1) < 0) {
855
+ assert(pass == 0);
856
+ return;
857
+ }
858
+ }
859
+
860
+ if (!blk_enable_write_cache(lu->qdev.conf.blk)) {
861
+ /* The request is used as the AIO opaque value, so add a ref. */
862
+ scsi_req_ref(&r->req);
863
+ block_acct_start(blk_get_stats(lu->qdev.conf.blk), &r->acct, 0,
864
+ BLOCK_ACCT_FLUSH);
865
+ r->req.aiocb = blk_aio_flush(lu->qdev.conf.blk, scsi_aio_complete, r);
866
+ return;
867
+ }
868
+
869
+ scsi_req_complete(&r->req, GOOD);
870
+ return;
871
+
872
+invalid_param:
873
+ scsi_check_condition(r, SENSE_CODE(INVALID_PARAM));
874
+ return;
875
+
876
+invalid_param_len:
877
+ scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
878
+ return;
879
+
880
+invalid_field:
881
+ scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
882
+}
883
+
884
+/* block_num and nb_blocks expected to be in qdev blocksize */
885
+static inline bool check_lba_range(UfsLu *lu, uint64_t block_num,
886
+ uint32_t nb_blocks)
887
+{
888
+ /*
889
+ * The first line tests that no overflow happens when computing the last
890
+ * block. The second line tests that the last accessed block is in
891
+ * range.
892
+ *
893
+ * Careful, the computations should not underflow for nb_blocks == 0,
894
+ * and a 0-block read to the first LBA beyond the end of device is
895
+ * valid.
896
+ */
897
+ return (block_num <= block_num + nb_blocks &&
898
+ block_num + nb_blocks <= lu->qdev.max_lba + 1);
899
+}
900
+
901
+static void ufs_scsi_emulate_write_data(SCSIRequest *req)
902
+{
903
+ UfsSCSIReq *r = DO_UPCAST(UfsSCSIReq, req, req);
904
+
905
+ if (r->iov.iov_len) {
906
+ int buflen = r->iov.iov_len;
907
+ trace_ufs_scsi_emulate_write_data(buflen);
908
+ r->iov.iov_len = 0;
909
+ scsi_req_data(&r->req, buflen);
910
+ return;
911
+ }
912
+
913
+ switch (req->cmd.buf[0]) {
914
+ case MODE_SELECT_10:
915
+ /* This also clears the sense buffer for REQUEST SENSE. */
916
+ ufs_scsi_emulate_mode_select(r, r->iov.iov_base);
917
+ break;
918
+ default:
919
+ abort();
920
+ }
921
+}
922
+
923
+/* Return a pointer to the data buffer. */
924
+static uint8_t *ufs_scsi_get_buf(SCSIRequest *req)
925
+{
926
+ UfsSCSIReq *r = DO_UPCAST(UfsSCSIReq, req, req);
927
+
928
+ return (uint8_t *)r->iov.iov_base;
929
+}
930
+
931
+static int32_t ufs_scsi_dma_command(SCSIRequest *req, uint8_t *buf)
932
+{
933
+ UfsSCSIReq *r = DO_UPCAST(UfsSCSIReq, req, req);
934
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, req->dev);
935
+ uint32_t len;
936
+ uint8_t command;
937
+
938
+ command = buf[0];
939
+
940
+ if (!blk_is_available(lu->qdev.conf.blk)) {
941
+ scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
942
+ return 0;
943
+ }
944
+
945
+ len = scsi_data_cdb_xfer(r->req.cmd.buf);
946
+ switch (command) {
947
+ case READ_6:
948
+ case READ_10:
949
+ trace_ufs_scsi_dma_command_READ(r->req.cmd.lba, len);
950
+ if (r->req.cmd.buf[1] & 0xe0) {
951
+ goto illegal_request;
952
+ }
953
+ if (!check_lba_range(lu, r->req.cmd.lba, len)) {
954
+ goto illegal_lba;
955
+ }
956
+ r->sector = r->req.cmd.lba * (lu->qdev.blocksize / BDRV_SECTOR_SIZE);
957
+ r->sector_count = len * (lu->qdev.blocksize / BDRV_SECTOR_SIZE);
958
+ break;
959
+ case WRITE_6:
960
+ case WRITE_10:
961
+ trace_ufs_scsi_dma_command_WRITE(r->req.cmd.lba, len);
962
+ if (!blk_is_writable(lu->qdev.conf.blk)) {
963
+ scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
964
+ return 0;
965
+ }
966
+ if (r->req.cmd.buf[1] & 0xe0) {
967
+ goto illegal_request;
968
+ }
969
+ if (!check_lba_range(lu, r->req.cmd.lba, len)) {
970
+ goto illegal_lba;
971
+ }
972
+ r->sector = r->req.cmd.lba * (lu->qdev.blocksize / BDRV_SECTOR_SIZE);
973
+ r->sector_count = len * (lu->qdev.blocksize / BDRV_SECTOR_SIZE);
974
+ break;
975
+ default:
976
+ abort();
977
+ illegal_request:
978
+ scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
979
+ return 0;
980
+ illegal_lba:
981
+ scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
982
+ return 0;
983
+ }
984
+ r->need_fua_emulation = ((r->req.cmd.buf[1] & 8) != 0);
985
+ if (r->sector_count == 0) {
986
+ scsi_req_complete(&r->req, GOOD);
987
+ }
988
+ assert(r->iov.iov_len == 0);
989
+ if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
990
+ return -r->sector_count * BDRV_SECTOR_SIZE;
991
+ } else {
992
+ return r->sector_count * BDRV_SECTOR_SIZE;
993
+ }
994
+}
995
+
996
+static void scsi_write_do_fua(UfsSCSIReq *r)
997
+{
998
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, r->req.dev);
999
+
1000
+ assert(r->req.aiocb == NULL);
1001
+ assert(!r->req.io_canceled);
1002
+
1003
+ if (r->need_fua_emulation) {
1004
+ block_acct_start(blk_get_stats(lu->qdev.conf.blk), &r->acct, 0,
1005
+ BLOCK_ACCT_FLUSH);
1006
+ r->req.aiocb = blk_aio_flush(lu->qdev.conf.blk, scsi_aio_complete, r);
1007
+ return;
1008
+ }
1009
+
1010
+ scsi_req_complete(&r->req, GOOD);
1011
+ scsi_req_unref(&r->req);
1012
+}
1013
+
1014
+static void scsi_dma_complete_noio(UfsSCSIReq *r, int ret)
1015
+{
1016
+ assert(r->req.aiocb == NULL);
1017
+ if (ufs_scsi_req_check_error(r, ret, false)) {
1018
+ goto done;
1019
+ }
1020
+
1021
+ r->sector += r->sector_count;
1022
+ r->sector_count = 0;
1023
+ if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
1024
+ scsi_write_do_fua(r);
1025
+ return;
1026
+ } else {
1027
+ scsi_req_complete(&r->req, GOOD);
1028
+ }
1029
+
1030
+done:
1031
+ scsi_req_unref(&r->req);
1032
+}
1033
+
1034
+static void scsi_dma_complete(void *opaque, int ret)
1035
+{
1036
+ UfsSCSIReq *r = (UfsSCSIReq *)opaque;
1037
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, r->req.dev);
1038
+
1039
+ assert(r->req.aiocb != NULL);
1040
+ r->req.aiocb = NULL;
1041
+
1042
+ aio_context_acquire(blk_get_aio_context(lu->qdev.conf.blk));
1043
+ if (ret < 0) {
1044
+ block_acct_failed(blk_get_stats(lu->qdev.conf.blk), &r->acct);
1045
+ } else {
1046
+ block_acct_done(blk_get_stats(lu->qdev.conf.blk), &r->acct);
1047
+ }
1048
+ scsi_dma_complete_noio(r, ret);
1049
+ aio_context_release(blk_get_aio_context(lu->qdev.conf.blk));
1050
+}
1051
+
1052
+static BlockAIOCB *scsi_dma_readv(int64_t offset, QEMUIOVector *iov,
1053
+ BlockCompletionFunc *cb, void *cb_opaque,
1054
+ void *opaque)
1055
+{
1056
+ UfsSCSIReq *r = opaque;
1057
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, r->req.dev);
1058
+ return blk_aio_preadv(lu->qdev.conf.blk, offset, iov, 0, cb, cb_opaque);
1059
+}
1060
+
1061
+static void scsi_init_iovec(UfsSCSIReq *r, size_t size)
1062
+{
1063
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, r->req.dev);
1064
+
1065
+ if (!r->iov.iov_base) {
1066
+ r->buflen = size;
1067
+ r->iov.iov_base = blk_blockalign(lu->qdev.conf.blk, r->buflen);
1068
+ }
1069
+ r->iov.iov_len = MIN(r->sector_count * BDRV_SECTOR_SIZE, r->buflen);
1070
+ qemu_iovec_init_external(&r->qiov, &r->iov, 1);
1071
+}
1072
+
1073
+static void scsi_read_complete_noio(UfsSCSIReq *r, int ret)
1074
+{
1075
+ uint32_t n;
1076
+
1077
+ assert(r->req.aiocb == NULL);
1078
+ if (ufs_scsi_req_check_error(r, ret, false)) {
1079
+ goto done;
1080
+ }
1081
+
1082
+ n = r->qiov.size / BDRV_SECTOR_SIZE;
1083
+ r->sector += n;
1084
+ r->sector_count -= n;
1085
+ scsi_req_data(&r->req, r->qiov.size);
1086
+
1087
+done:
1088
+ scsi_req_unref(&r->req);
1089
+}
1090
+
1091
+static void scsi_read_complete(void *opaque, int ret)
1092
+{
1093
+ UfsSCSIReq *r = (UfsSCSIReq *)opaque;
1094
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, r->req.dev);
1095
+
1096
+ assert(r->req.aiocb != NULL);
1097
+ r->req.aiocb = NULL;
1098
+ trace_ufs_scsi_read_data_count(r->sector_count);
1099
+ aio_context_acquire(blk_get_aio_context(lu->qdev.conf.blk));
1100
+ if (ret < 0) {
1101
+ block_acct_failed(blk_get_stats(lu->qdev.conf.blk), &r->acct);
1102
+ } else {
1103
+ block_acct_done(blk_get_stats(lu->qdev.conf.blk), &r->acct);
1104
+ trace_ufs_scsi_read_complete(r->req.tag, r->qiov.size);
1105
+ }
1106
+ scsi_read_complete_noio(r, ret);
1107
+ aio_context_release(blk_get_aio_context(lu->qdev.conf.blk));
1108
+}
1109
+
1110
+/* Actually issue a read to the block device. */
1111
+static void scsi_do_read(UfsSCSIReq *r, int ret)
1112
+{
1113
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, r->req.dev);
1114
+
1115
+ assert(r->req.aiocb == NULL);
1116
+ if (ufs_scsi_req_check_error(r, ret, false)) {
1117
+ goto done;
1118
+ }
1119
+
1120
+ /* The request is used as the AIO opaque value, so add a ref. */
1121
+ scsi_req_ref(&r->req);
1122
+
1123
+ if (r->req.sg) {
1124
+ dma_acct_start(lu->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_READ);
1125
+ r->req.residual -= r->req.sg->size;
1126
+ r->req.aiocb = dma_blk_io(
1127
+ blk_get_aio_context(lu->qdev.conf.blk), r->req.sg,
1128
+ r->sector << BDRV_SECTOR_BITS, BDRV_SECTOR_SIZE, scsi_dma_readv, r,
1129
+ scsi_dma_complete, r, DMA_DIRECTION_FROM_DEVICE);
1130
+ } else {
1131
+ scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
1132
+ block_acct_start(blk_get_stats(lu->qdev.conf.blk), &r->acct,
1133
+ r->qiov.size, BLOCK_ACCT_READ);
1134
+ r->req.aiocb = scsi_dma_readv(r->sector << BDRV_SECTOR_BITS, &r->qiov,
1135
+ scsi_read_complete, r, r);
1136
+ }
1137
+
1138
+done:
1139
+ scsi_req_unref(&r->req);
1140
+}
1141
+
1142
+static void scsi_do_read_cb(void *opaque, int ret)
1143
+{
1144
+ UfsSCSIReq *r = (UfsSCSIReq *)opaque;
1145
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, r->req.dev);
1146
+
1147
+ assert(r->req.aiocb != NULL);
1148
+ r->req.aiocb = NULL;
1149
+
1150
+ aio_context_acquire(blk_get_aio_context(lu->qdev.conf.blk));
1151
+ if (ret < 0) {
1152
+ block_acct_failed(blk_get_stats(lu->qdev.conf.blk), &r->acct);
1153
+ } else {
1154
+ block_acct_done(blk_get_stats(lu->qdev.conf.blk), &r->acct);
1155
+ }
1156
+ scsi_do_read(opaque, ret);
1157
+ aio_context_release(blk_get_aio_context(lu->qdev.conf.blk));
1158
+}
1159
+
1160
+/* Read more data from scsi device into buffer. */
1161
+static void scsi_read_data(SCSIRequest *req)
1162
+{
1163
+ UfsSCSIReq *r = DO_UPCAST(UfsSCSIReq, req, req);
1164
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, r->req.dev);
1165
+ bool first;
1166
+
1167
+ trace_ufs_scsi_read_data_count(r->sector_count);
1168
+ if (r->sector_count == 0) {
1169
+ /* This also clears the sense buffer for REQUEST SENSE. */
1170
+ scsi_req_complete(&r->req, GOOD);
1171
+ return;
1172
+ }
1173
+
1174
+ /* No data transfer may already be in progress */
1175
+ assert(r->req.aiocb == NULL);
1176
+
1177
+ /* The request is used as the AIO opaque value, so add a ref. */
1178
+ scsi_req_ref(&r->req);
1179
+ if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
1180
+ trace_ufs_scsi_read_data_invalid();
1181
+ scsi_read_complete_noio(r, -EINVAL);
1182
+ return;
1183
+ }
1184
+
1185
+ if (!blk_is_available(req->dev->conf.blk)) {
1186
+ scsi_read_complete_noio(r, -ENOMEDIUM);
1187
+ return;
1188
+ }
1189
+
1190
+ first = !r->started;
1191
+ r->started = true;
1192
+ if (first && r->need_fua_emulation) {
1193
+ block_acct_start(blk_get_stats(lu->qdev.conf.blk), &r->acct, 0,
1194
+ BLOCK_ACCT_FLUSH);
1195
+ r->req.aiocb = blk_aio_flush(lu->qdev.conf.blk, scsi_do_read_cb, r);
1196
+ } else {
1197
+ scsi_do_read(r, 0);
1198
+ }
1199
+}
1200
+
1201
+static void scsi_write_complete_noio(UfsSCSIReq *r, int ret)
1202
+{
1203
+ uint32_t n;
1204
+
1205
+ assert(r->req.aiocb == NULL);
1206
+ if (ufs_scsi_req_check_error(r, ret, false)) {
1207
+ goto done;
1208
+ }
1209
+
1210
+ n = r->qiov.size / BDRV_SECTOR_SIZE;
1211
+ r->sector += n;
1212
+ r->sector_count -= n;
1213
+ if (r->sector_count == 0) {
1214
+ scsi_write_do_fua(r);
1215
+ return;
1216
+ } else {
1217
+ scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
1218
+ trace_ufs_scsi_write_complete_noio(r->req.tag, r->qiov.size);
1219
+ scsi_req_data(&r->req, r->qiov.size);
1220
+ }
1221
+
1222
+done:
1223
+ scsi_req_unref(&r->req);
1224
+}
1225
+
1226
+static void scsi_write_complete(void *opaque, int ret)
1227
+{
1228
+ UfsSCSIReq *r = (UfsSCSIReq *)opaque;
1229
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, r->req.dev);
1230
+
1231
+ assert(r->req.aiocb != NULL);
1232
+ r->req.aiocb = NULL;
1233
+
1234
+ aio_context_acquire(blk_get_aio_context(lu->qdev.conf.blk));
1235
+ if (ret < 0) {
1236
+ block_acct_failed(blk_get_stats(lu->qdev.conf.blk), &r->acct);
1237
+ } else {
1238
+ block_acct_done(blk_get_stats(lu->qdev.conf.blk), &r->acct);
1239
+ }
1240
+ scsi_write_complete_noio(r, ret);
1241
+ aio_context_release(blk_get_aio_context(lu->qdev.conf.blk));
1242
+}
1243
+
1244
+static BlockAIOCB *scsi_dma_writev(int64_t offset, QEMUIOVector *iov,
1245
+ BlockCompletionFunc *cb, void *cb_opaque,
1246
+ void *opaque)
1247
+{
1248
+ UfsSCSIReq *r = opaque;
1249
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, r->req.dev);
1250
+ return blk_aio_pwritev(lu->qdev.conf.blk, offset, iov, 0, cb, cb_opaque);
1251
+}
1252
+
1253
+static void scsi_write_data(SCSIRequest *req)
1254
+{
1255
+ UfsSCSIReq *r = DO_UPCAST(UfsSCSIReq, req, req);
1256
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, r->req.dev);
1257
+
1258
+ /* No data transfer may already be in progress */
1259
+ assert(r->req.aiocb == NULL);
1260
+
1261
+ /* The request is used as the AIO opaque value, so add a ref. */
1262
+ scsi_req_ref(&r->req);
1263
+ if (r->req.cmd.mode != SCSI_XFER_TO_DEV) {
1264
+ trace_ufs_scsi_write_data_invalid();
1265
+ scsi_write_complete_noio(r, -EINVAL);
1266
+ return;
1267
+ }
1268
+
1269
+ if (!r->req.sg && !r->qiov.size) {
1270
+ /* Called for the first time. Ask the driver to send us more data. */
1271
+ r->started = true;
1272
+ scsi_write_complete_noio(r, 0);
1273
+ return;
1274
+ }
1275
+ if (!blk_is_available(req->dev->conf.blk)) {
1276
+ scsi_write_complete_noio(r, -ENOMEDIUM);
1277
+ return;
1278
+ }
1279
+
1280
+ if (r->req.sg) {
1281
+ dma_acct_start(lu->qdev.conf.blk, &r->acct, r->req.sg,
1282
+ BLOCK_ACCT_WRITE);
1283
+ r->req.residual -= r->req.sg->size;
1284
+ r->req.aiocb = dma_blk_io(
1285
+ blk_get_aio_context(lu->qdev.conf.blk), r->req.sg,
1286
+ r->sector << BDRV_SECTOR_BITS, BDRV_SECTOR_SIZE, scsi_dma_writev, r,
1287
+ scsi_dma_complete, r, DMA_DIRECTION_TO_DEVICE);
1288
+ } else {
1289
+ block_acct_start(blk_get_stats(lu->qdev.conf.blk), &r->acct,
1290
+ r->qiov.size, BLOCK_ACCT_WRITE);
1291
+ r->req.aiocb = scsi_dma_writev(r->sector << BDRV_SECTOR_BITS, &r->qiov,
1292
+ scsi_write_complete, r, r);
1293
+ }
1294
+}
1295
+
1296
+static const SCSIReqOps ufs_scsi_emulate_reqops = {
1297
+ .size = sizeof(UfsSCSIReq),
1298
+ .free_req = ufs_scsi_free_request,
1299
+ .send_command = ufs_scsi_emulate_command,
1300
+ .read_data = ufs_scsi_emulate_read_data,
1301
+ .write_data = ufs_scsi_emulate_write_data,
1302
+ .get_buf = ufs_scsi_get_buf,
1303
+};
1304
+
1305
+static const SCSIReqOps ufs_scsi_dma_reqops = {
1306
+ .size = sizeof(UfsSCSIReq),
1307
+ .free_req = ufs_scsi_free_request,
1308
+ .send_command = ufs_scsi_dma_command,
1309
+ .read_data = scsi_read_data,
1310
+ .write_data = scsi_write_data,
1311
+ .get_buf = ufs_scsi_get_buf,
1312
+};
1313
+
1314
+/*
1315
+ * Following commands are not yet supported
1316
+ * PRE_FETCH(10),
1317
+ * UNMAP,
1318
+ * WRITE_BUFFER, READ_BUFFER,
1319
+ * SECURITY_PROTOCOL_IN, SECURITY_PROTOCOL_OUT
1320
+ */
1321
+static const SCSIReqOps *const ufs_scsi_reqops_dispatch[256] = {
1322
+ [TEST_UNIT_READY] = &ufs_scsi_emulate_reqops,
1323
+ [INQUIRY] = &ufs_scsi_emulate_reqops,
1324
+ [MODE_SENSE_10] = &ufs_scsi_emulate_reqops,
1325
+ [START_STOP] = &ufs_scsi_emulate_reqops,
1326
+ [READ_CAPACITY_10] = &ufs_scsi_emulate_reqops,
1327
+ [REQUEST_SENSE] = &ufs_scsi_emulate_reqops,
1328
+ [SYNCHRONIZE_CACHE] = &ufs_scsi_emulate_reqops,
1329
+ [MODE_SELECT_10] = &ufs_scsi_emulate_reqops,
1330
+ [VERIFY_10] = &ufs_scsi_emulate_reqops,
1331
+ [FORMAT_UNIT] = &ufs_scsi_emulate_reqops,
1332
+ [SERVICE_ACTION_IN_16] = &ufs_scsi_emulate_reqops,
1333
+ [SEND_DIAGNOSTIC] = &ufs_scsi_emulate_reqops,
1334
+
1335
+ [READ_6] = &ufs_scsi_dma_reqops,
1336
+ [READ_10] = &ufs_scsi_dma_reqops,
1337
+ [WRITE_6] = &ufs_scsi_dma_reqops,
1338
+ [WRITE_10] = &ufs_scsi_dma_reqops,
1339
+};
1340
+
1341
+static SCSIRequest *scsi_new_request(SCSIDevice *dev, uint32_t tag,
1342
+ uint32_t lun, uint8_t *buf,
1343
+ void *hba_private)
1344
+{
1345
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, dev);
1346
+ SCSIRequest *req;
1347
+ const SCSIReqOps *ops;
1348
+ uint8_t command;
1349
+
1350
+ command = buf[0];
1351
+ ops = ufs_scsi_reqops_dispatch[command];
1352
+ if (!ops) {
1353
+ ops = &ufs_scsi_emulate_reqops;
1354
+ }
1355
+ req = scsi_req_alloc(ops, &lu->qdev, tag, lun, hba_private);
1356
+
1357
+ return req;
1358
+}
1359
+
1360
+static Property ufs_lu_props[] = {
1361
+ DEFINE_PROP_DRIVE("drive", UfsLu, qdev.conf.blk),
1362
+ DEFINE_PROP_END_OF_LIST(),
1363
+};
1364
+
1365
+static bool ufs_lu_brdv_init(UfsLu *lu, Error **errp)
1366
+{
1367
+ SCSIDevice *dev = &lu->qdev;
1368
+ bool read_only;
1369
+
1370
+ if (!lu->qdev.conf.blk) {
1371
+ error_setg(errp, "drive property not set");
1372
+ return false;
1373
+ }
1374
+
1375
+ if (!blkconf_blocksizes(&lu->qdev.conf, errp)) {
1376
+ return false;
1377
+ }
1378
+
1379
+ if (blk_get_aio_context(lu->qdev.conf.blk) != qemu_get_aio_context() &&
1380
+ !lu->qdev.hba_supports_iothread) {
1381
+ error_setg(errp, "HBA does not support iothreads");
1382
+ return false;
1383
+ }
1384
+
1385
+ read_only = !blk_supports_write_perm(lu->qdev.conf.blk);
1386
+
1387
+ if (!blkconf_apply_backend_options(&dev->conf, read_only,
1388
+ dev->type == TYPE_DISK, errp)) {
1389
+ return false;
1390
+ }
1391
+
1392
+ if (blk_is_sg(lu->qdev.conf.blk)) {
1393
+ error_setg(errp, "unwanted /dev/sg*");
1394
+ return false;
1395
+ }
1396
+
1397
+ blk_iostatus_enable(lu->qdev.conf.blk);
1398
+ return true;
1399
+}
1400
+
1401
+static bool ufs_add_lu(UfsHc *u, UfsLu *lu, Error **errp)
1402
+{
1403
+ BlockBackend *blk = lu->qdev.conf.blk;
1404
+ int64_t brdv_len = blk_getlength(blk);
1405
+ uint64_t raw_dev_cap =
1406
+ be64_to_cpu(u->geometry_desc.total_raw_device_capacity);
1407
+
1408
+ if (u->device_desc.number_lu >= UFS_MAX_LUS) {
1409
+ error_setg(errp, "ufs host controller has too many logical units.");
1410
+ return false;
1411
+ }
1412
+
1413
+ if (u->lus[lu->lun] != NULL) {
1414
+ error_setg(errp, "ufs logical unit %d already exists.", lu->lun);
1415
+ return false;
1416
+ }
1417
+
1418
+ u->lus[lu->lun] = lu;
1419
+ u->device_desc.number_lu++;
1420
+ raw_dev_cap += (brdv_len >> UFS_GEOMETRY_CAPACITY_SHIFT);
1421
+ u->geometry_desc.total_raw_device_capacity = cpu_to_be64(raw_dev_cap);
1422
+ return true;
1423
+}
1424
+
1425
+static inline uint8_t ufs_log2(uint64_t input)
1426
+{
1427
+ int log = 0;
1428
+ while (input >>= 1) {
1429
+ log++;
1430
+ }
1431
+ return log;
1432
+}
1433
+
1434
+static void ufs_init_lu(UfsLu *lu)
1435
+{
1436
+ BlockBackend *blk = lu->qdev.conf.blk;
1437
+ int64_t brdv_len = blk_getlength(blk);
1438
+
1439
+ lu->lun = lu->qdev.lun;
1440
+ memset(&lu->unit_desc, 0, sizeof(lu->unit_desc));
1441
+ lu->unit_desc.length = sizeof(UnitDescriptor);
1442
+ lu->unit_desc.descriptor_idn = UFS_QUERY_DESC_IDN_UNIT;
1443
+ lu->unit_desc.lu_enable = 0x01;
1444
+ lu->unit_desc.logical_block_size = ufs_log2(lu->qdev.blocksize);
1445
+ lu->unit_desc.unit_index = lu->qdev.lun;
1446
+ lu->unit_desc.logical_block_count =
1447
+ cpu_to_be64(brdv_len / (1 << lu->unit_desc.logical_block_size));
1448
+}
1449
+
1450
+static bool ufs_lu_check_constraints(UfsLu *lu, Error **errp)
1451
+{
1452
+ if (!lu->qdev.conf.blk) {
1453
+ error_setg(errp, "drive property not set");
1454
+ return false;
1455
+ }
1456
+
1457
+ if (lu->qdev.channel != 0) {
1458
+ error_setg(errp, "ufs logical unit does not support channel");
1459
+ return false;
1460
+ }
1461
+
1462
+ if (lu->qdev.lun >= UFS_MAX_LUS) {
1463
+ error_setg(errp, "lun must be between 1 and %d", UFS_MAX_LUS - 1);
1464
+ return false;
1465
+ }
1466
+
1467
+ return true;
1468
+}
1469
+
1470
+static void ufs_lu_realize(SCSIDevice *dev, Error **errp)
1471
+{
1472
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, dev);
1473
+ BusState *s = qdev_get_parent_bus(&dev->qdev);
1474
+ UfsHc *u = UFS(s->parent);
1475
+ AioContext *ctx = NULL;
1476
+ uint64_t nb_sectors, nb_blocks;
1477
+
1478
+ if (!ufs_lu_check_constraints(lu, errp)) {
1479
+ return;
1480
+ }
1481
+
1482
+ if (lu->qdev.conf.blk) {
1483
+ ctx = blk_get_aio_context(lu->qdev.conf.blk);
1484
+ aio_context_acquire(ctx);
1485
+ if (!blkconf_blocksizes(&lu->qdev.conf, errp)) {
1486
+ goto out;
1487
+ }
1488
+ }
1489
+ lu->qdev.blocksize = UFS_BLOCK_SIZE;
1490
+ blk_get_geometry(lu->qdev.conf.blk, &nb_sectors);
1491
+ nb_blocks = nb_sectors / (lu->qdev.blocksize / BDRV_SECTOR_SIZE);
1492
+ if (nb_blocks > UINT32_MAX) {
1493
+ nb_blocks = UINT32_MAX;
1494
+ }
1495
+ lu->qdev.max_lba = nb_blocks;
1496
+ lu->qdev.type = TYPE_DISK;
1497
+
1498
+ ufs_init_lu(lu);
1499
+ if (!ufs_add_lu(u, lu, errp)) {
1500
+ goto out;
1501
+ }
1502
+
1503
+ ufs_lu_brdv_init(lu, errp);
1504
+out:
1505
+ if (ctx) {
1506
+ aio_context_release(ctx);
1507
+ }
1508
+}
1509
+
1510
+static void ufs_lu_unrealize(SCSIDevice *dev)
1511
+{
1512
+ UfsLu *lu = DO_UPCAST(UfsLu, qdev, dev);
1513
+
1514
+ blk_drain(lu->qdev.conf.blk);
1515
+}
1516
+
1517
+static void ufs_wlu_realize(DeviceState *qdev, Error **errp)
1518
+{
1519
+ UfsWLu *wlu = UFSWLU(qdev);
1520
+ SCSIDevice *dev = &wlu->qdev;
1521
+
1522
+ if (!is_wlun(dev->lun)) {
1523
+ error_setg(errp, "not well-known logical unit number");
1524
+ return;
1525
+ }
1526
+
1527
+ QTAILQ_INIT(&dev->requests);
1528
+}
1529
+
1530
+static void ufs_lu_class_init(ObjectClass *oc, void *data)
1531
+{
1532
+ DeviceClass *dc = DEVICE_CLASS(oc);
1533
+ SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(oc);
1534
+
1535
+ sc->realize = ufs_lu_realize;
1536
+ sc->unrealize = ufs_lu_unrealize;
1537
+ sc->alloc_req = scsi_new_request;
1538
+ dc->bus_type = TYPE_UFS_BUS;
1539
+ device_class_set_props(dc, ufs_lu_props);
1540
+ dc->desc = "Virtual UFS logical unit";
1541
+}
1542
+
1543
+static void ufs_wlu_class_init(ObjectClass *oc, void *data)
1544
+{
1545
+ DeviceClass *dc = DEVICE_CLASS(oc);
1546
+ SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(oc);
1547
+
1548
+ /*
1549
+ * The realize() function of TYPE_SCSI_DEVICE causes a segmentation fault
1550
+ * if a block drive does not exist. Define a new realize function for
1551
+ * well-known LUs that do not have a block drive.
1552
+ */
1553
+ dc->realize = ufs_wlu_realize;
1554
+ sc->alloc_req = scsi_new_request;
1555
+ dc->bus_type = TYPE_UFS_BUS;
1556
+ dc->desc = "Virtual UFS well-known logical unit";
1557
+}
1558
+
1559
+static const TypeInfo ufs_lu_info = {
1560
+ .name = TYPE_UFS_LU,
1561
+ .parent = TYPE_SCSI_DEVICE,
1562
+ .class_init = ufs_lu_class_init,
1563
+ .instance_size = sizeof(UfsLu),
1564
+};
1565
+
1566
+static const TypeInfo ufs_wlu_info = {
1567
+ .name = TYPE_UFS_WLU,
1568
+ .parent = TYPE_SCSI_DEVICE,
1569
+ .class_init = ufs_wlu_class_init,
1570
+ .instance_size = sizeof(UfsWLu),
1571
+};
1572
+
1573
+static void ufs_lu_register_types(void)
1574
+{
1575
+ type_register_static(&ufs_lu_info);
1576
+ type_register_static(&ufs_wlu_info);
1577
+}
1578
+
1579
+type_init(ufs_lu_register_types)
1580
diff --git a/hw/ufs/ufs.c b/hw/ufs/ufs.c
1581
index XXXXXXX..XXXXXXX 100644
1582
--- a/hw/ufs/ufs.c
1583
+++ b/hw/ufs/ufs.c
1584
@@ -XXX,XX +XXX,XX @@
1585
* SPDX-License-Identifier: GPL-2.0-or-later
1586
*/
1587
1588
+/**
1589
+ * Reference Specs: https://www.jedec.org/, 3.1
1590
+ *
1591
+ * Usage
1592
+ * -----
1593
+ *
1594
+ * Add options:
1595
+ * -drive file=<file>,if=none,id=<drive_id>
1596
+ * -device ufs,serial=<serial>,id=<bus_name>, \
1597
+ * nutrs=<N[optional]>,nutmrs=<N[optional]>
1598
+ * -device ufs-lu,drive=<drive_id>,bus=<bus_name>
1599
+ */
1600
+
1601
#include "qemu/osdep.h"
1602
#include "qapi/error.h"
1603
#include "migration/vmstate.h"
1604
@@ -XXX,XX +XXX,XX @@ static const MemoryRegionOps ufs_mmio_ops = {
1605
},
1606
};
1607
1608
+static QEMUSGList *ufs_get_sg_list(SCSIRequest *scsi_req)
1609
+{
1610
+ UfsRequest *req = scsi_req->hba_private;
1611
+ return req->sg;
1612
+}
1613
+
1614
+static void ufs_build_upiu_sense_data(UfsRequest *req, SCSIRequest *scsi_req)
1615
+{
1616
+ req->rsp_upiu.sr.sense_data_len = cpu_to_be16(scsi_req->sense_len);
1617
+ assert(scsi_req->sense_len <= SCSI_SENSE_LEN);
1618
+ memcpy(req->rsp_upiu.sr.sense_data, scsi_req->sense, scsi_req->sense_len);
1619
+}
1620
+
1621
static void ufs_build_upiu_header(UfsRequest *req, uint8_t trans_type,
1622
uint8_t flags, uint8_t response,
1623
uint8_t scsi_status,
1624
@@ -XXX,XX +XXX,XX @@ static void ufs_build_upiu_header(UfsRequest *req, uint8_t trans_type,
1625
req->rsp_upiu.header.data_segment_length = cpu_to_be16(data_segment_length);
1626
}
1627
1628
+static void ufs_scsi_command_complete(SCSIRequest *scsi_req, size_t resid)
1629
+{
1630
+ UfsRequest *req = scsi_req->hba_private;
1631
+ int16_t status = scsi_req->status;
1632
+ uint32_t expected_len = be32_to_cpu(req->req_upiu.sc.exp_data_transfer_len);
1633
+ uint32_t transfered_len = scsi_req->cmd.xfer - resid;
1634
+ uint8_t flags = 0, response = UFS_COMMAND_RESULT_SUCESS;
1635
+ uint16_t data_segment_length;
1636
+
1637
+ if (expected_len > transfered_len) {
1638
+ req->rsp_upiu.sr.residual_transfer_count =
1639
+ cpu_to_be32(expected_len - transfered_len);
1640
+ flags |= UFS_UPIU_FLAG_UNDERFLOW;
1641
+ } else if (expected_len < transfered_len) {
1642
+ req->rsp_upiu.sr.residual_transfer_count =
1643
+ cpu_to_be32(transfered_len - expected_len);
1644
+ flags |= UFS_UPIU_FLAG_OVERFLOW;
1645
+ }
1646
+
1647
+ if (status != 0) {
1648
+ ufs_build_upiu_sense_data(req, scsi_req);
1649
+ response = UFS_COMMAND_RESULT_FAIL;
1650
+ }
1651
+
1652
+ data_segment_length = cpu_to_be16(scsi_req->sense_len +
1653
+ sizeof(req->rsp_upiu.sr.sense_data_len));
1654
+ ufs_build_upiu_header(req, UFS_UPIU_TRANSACTION_RESPONSE, flags, response,
1655
+ status, data_segment_length);
1656
+
1657
+ ufs_complete_req(req, UFS_REQUEST_SUCCESS);
1658
+
1659
+ scsi_req->hba_private = NULL;
1660
+ scsi_req_unref(scsi_req);
1661
+}
1662
+
1663
+static const struct SCSIBusInfo ufs_scsi_info = {
1664
+ .tcq = true,
1665
+ .max_target = 0,
1666
+ .max_lun = UFS_MAX_LUS,
1667
+ .max_channel = 0,
1668
+
1669
+ .get_sg_list = ufs_get_sg_list,
1670
+ .complete = ufs_scsi_command_complete,
1671
+};
1672
+
1673
+static UfsReqResult ufs_exec_scsi_cmd(UfsRequest *req)
1674
+{
1675
+ UfsHc *u = req->hc;
1676
+ uint8_t lun = req->req_upiu.header.lun;
1677
+ uint8_t task_tag = req->req_upiu.header.task_tag;
1678
+ SCSIDevice *dev = NULL;
1679
+
1680
+ trace_ufs_exec_scsi_cmd(req->slot, lun, req->req_upiu.sc.cdb[0]);
1681
+
1682
+ if (!is_wlun(lun)) {
1683
+ if (lun >= u->device_desc.number_lu) {
1684
+ trace_ufs_err_scsi_cmd_invalid_lun(lun);
1685
+ return UFS_REQUEST_FAIL;
1686
+ } else if (u->lus[lun] == NULL) {
1687
+ trace_ufs_err_scsi_cmd_invalid_lun(lun);
1688
+ return UFS_REQUEST_FAIL;
1689
+ }
1690
+ }
1691
+
1692
+ switch (lun) {
1693
+ case UFS_UPIU_REPORT_LUNS_WLUN:
1694
+ dev = &u->report_wlu->qdev;
1695
+ break;
1696
+ case UFS_UPIU_UFS_DEVICE_WLUN:
1697
+ dev = &u->dev_wlu->qdev;
1698
+ break;
1699
+ case UFS_UPIU_BOOT_WLUN:
1700
+ dev = &u->boot_wlu->qdev;
1701
+ break;
1702
+ case UFS_UPIU_RPMB_WLUN:
1703
+ dev = &u->rpmb_wlu->qdev;
1704
+ break;
1705
+ default:
1706
+ dev = &u->lus[lun]->qdev;
1707
+ }
1708
+
1709
+ SCSIRequest *scsi_req = scsi_req_new(
1710
+ dev, task_tag, lun, req->req_upiu.sc.cdb, UFS_CDB_SIZE, req);
1711
+
1712
+ uint32_t len = scsi_req_enqueue(scsi_req);
1713
+ if (len) {
1714
+ scsi_req_continue(scsi_req);
1715
+ }
1716
+
1717
+ return UFS_REQUEST_NO_COMPLETE;
1718
+}
1719
+
1720
static UfsReqResult ufs_exec_nop_cmd(UfsRequest *req)
1721
{
1722
trace_ufs_exec_nop_cmd(req->slot);
1723
@@ -XXX,XX +XXX,XX @@ static const RpmbUnitDescriptor rpmb_unit_desc = {
1724
1725
static QueryRespCode ufs_read_unit_desc(UfsRequest *req)
1726
{
1727
+ UfsHc *u = req->hc;
1728
uint8_t lun = req->req_upiu.qr.index;
1729
1730
- if (lun != UFS_UPIU_RPMB_WLUN && lun > UFS_MAX_LUS) {
1731
+ if (lun != UFS_UPIU_RPMB_WLUN &&
1732
+ (lun > UFS_MAX_LUS || u->lus[lun] == NULL)) {
1733
trace_ufs_err_query_invalid_index(req->req_upiu.qr.opcode, lun);
1734
return UFS_QUERY_RESULT_INVALID_INDEX;
1735
}
1736
@@ -XXX,XX +XXX,XX @@ static QueryRespCode ufs_read_unit_desc(UfsRequest *req)
1737
if (lun == UFS_UPIU_RPMB_WLUN) {
1738
memcpy(&req->rsp_upiu.qr.data, &rpmb_unit_desc, rpmb_unit_desc.length);
1739
} else {
1740
- /* unit descriptor is not yet supported */
1741
- return UFS_QUERY_RESULT_INVALID_INDEX;
1742
+ memcpy(&req->rsp_upiu.qr.data, &u->lus[lun]->unit_desc,
1743
+ sizeof(u->lus[lun]->unit_desc));
1744
}
1745
1746
return UFS_QUERY_RESULT_SUCCESS;
1747
@@ -XXX,XX +XXX,XX @@ static void ufs_exec_req(UfsRequest *req)
1748
req_result = ufs_exec_nop_cmd(req);
1749
break;
1750
case UFS_UPIU_TRANSACTION_COMMAND:
1751
- /* Not yet implemented */
1752
- req_result = UFS_REQUEST_FAIL;
1753
+ req_result = ufs_exec_scsi_cmd(req);
1754
break;
1755
case UFS_UPIU_TRANSACTION_QUERY_REQ:
1756
req_result = ufs_exec_query_cmd(req);
1757
@@ -XXX,XX +XXX,XX @@ static void ufs_exec_req(UfsRequest *req)
1758
req_result = UFS_REQUEST_FAIL;
1759
}
1760
1761
- ufs_complete_req(req, req_result);
1762
+ /*
1763
+ * The ufs_complete_req for scsi commands is handled by the
1764
+ * ufs_scsi_command_complete() callback function. Therefore, to avoid
1765
+ * duplicate processing, ufs_complete_req() is not called for scsi commands.
1766
+ */
1767
+ if (req_result != UFS_REQUEST_NO_COMPLETE) {
1768
+ ufs_complete_req(req, req_result);
1769
+ }
183
+ }
1770
}
184
}
1771
185
1772
static void ufs_process_req(void *opaque)
186
/**
1773
@@ -XXX,XX +XXX,XX @@ static void ufs_init_hc(UfsHc *u)
187
@@ -XXX,XX +XXX,XX @@ void qemu_co_rwlock_rdlock(CoRwlock *lock)
1774
u->flags.permanently_disable_fw_update = 1;
188
Coroutine *self = qemu_coroutine_self();
1775
}
189
1776
190
while (lock->writer) {
1777
+static bool ufs_init_wlu(UfsHc *u, UfsWLu **wlu, uint8_t wlun, Error **errp)
191
- qemu_co_queue_wait(&lock->queue);
1778
+{
192
+ qemu_co_queue_wait(&lock->queue, NULL);
1779
+ UfsWLu *new_wlu = UFSWLU(qdev_new(TYPE_UFS_WLU));
193
}
1780
+
194
lock->reader++;
1781
+ qdev_prop_set_uint32(DEVICE(new_wlu), "lun", wlun);
195
self->locks_held++;
1782
+
196
@@ -XXX,XX +XXX,XX @@ void qemu_co_rwlock_wrlock(CoRwlock *lock)
1783
+ /*
197
Coroutine *self = qemu_coroutine_self();
1784
+ * The well-known lu shares the same bus as the normal lu. If the well-known
198
1785
+ * lu writes the same channel value as the normal lu, the report will be
199
while (lock->writer || lock->reader) {
1786
+ * made not only for the normal lu but also for the well-known lu at
200
- qemu_co_queue_wait(&lock->queue);
1787
+ * REPORT_LUN time. To prevent this, the channel value of normal lu is fixed
201
+ qemu_co_queue_wait(&lock->queue, NULL);
1788
+ * to 0 and the channel value of well-known lu is fixed to 1.
202
}
1789
+ */
203
lock->writer = true;
1790
+ qdev_prop_set_uint32(DEVICE(new_wlu), "channel", 1);
204
self->locks_held++;
1791
+ if (!qdev_realize_and_unref(DEVICE(new_wlu), BUS(&u->bus), errp)) {
1792
+ return false;
1793
+ }
1794
+
1795
+ *wlu = new_wlu;
1796
+ return true;
1797
+}
1798
+
1799
static void ufs_realize(PCIDevice *pci_dev, Error **errp)
1800
{
1801
UfsHc *u = UFS(pci_dev);
1802
@@ -XXX,XX +XXX,XX @@ static void ufs_realize(PCIDevice *pci_dev, Error **errp)
1803
return;
1804
}
1805
1806
+ qbus_init(&u->bus, sizeof(UfsBus), TYPE_UFS_BUS, &pci_dev->qdev,
1807
+ u->parent_obj.qdev.id);
1808
+ u->bus.parent_bus.info = &ufs_scsi_info;
1809
+
1810
ufs_init_state(u);
1811
ufs_init_hc(u);
1812
ufs_init_pci(u, pci_dev);
1813
+
1814
+ if (!ufs_init_wlu(u, &u->report_wlu, UFS_UPIU_REPORT_LUNS_WLUN, errp)) {
1815
+ return;
1816
+ }
1817
+
1818
+ if (!ufs_init_wlu(u, &u->dev_wlu, UFS_UPIU_UFS_DEVICE_WLUN, errp)) {
1819
+ return;
1820
+ }
1821
+
1822
+ if (!ufs_init_wlu(u, &u->boot_wlu, UFS_UPIU_BOOT_WLUN, errp)) {
1823
+ return;
1824
+ }
1825
+
1826
+ if (!ufs_init_wlu(u, &u->rpmb_wlu, UFS_UPIU_RPMB_WLUN, errp)) {
1827
+ return;
1828
+ }
1829
}
1830
1831
static void ufs_exit(PCIDevice *pci_dev)
1832
{
1833
UfsHc *u = UFS(pci_dev);
1834
1835
+ if (u->dev_wlu) {
1836
+ object_unref(OBJECT(u->dev_wlu));
1837
+ u->dev_wlu = NULL;
1838
+ }
1839
+
1840
+ if (u->report_wlu) {
1841
+ object_unref(OBJECT(u->report_wlu));
1842
+ u->report_wlu = NULL;
1843
+ }
1844
+
1845
+ if (u->rpmb_wlu) {
1846
+ object_unref(OBJECT(u->rpmb_wlu));
1847
+ u->rpmb_wlu = NULL;
1848
+ }
1849
+
1850
+ if (u->boot_wlu) {
1851
+ object_unref(OBJECT(u->boot_wlu));
1852
+ u->boot_wlu = NULL;
1853
+ }
1854
+
1855
qemu_bh_delete(u->doorbell_bh);
1856
qemu_bh_delete(u->complete_bh);
1857
1858
@@ -XXX,XX +XXX,XX @@ static void ufs_class_init(ObjectClass *oc, void *data)
1859
dc->vmsd = &ufs_vmstate;
1860
}
1861
1862
+static bool ufs_bus_check_address(BusState *qbus, DeviceState *qdev,
1863
+ Error **errp)
1864
+{
1865
+ SCSIDevice *dev = SCSI_DEVICE(qdev);
1866
+ UfsBusClass *ubc = UFS_BUS_GET_CLASS(qbus);
1867
+ UfsHc *u = UFS(qbus->parent);
1868
+
1869
+ if (strcmp(object_get_typename(OBJECT(dev)), TYPE_UFS_WLU) == 0) {
1870
+ if (dev->lun != UFS_UPIU_REPORT_LUNS_WLUN &&
1871
+ dev->lun != UFS_UPIU_UFS_DEVICE_WLUN &&
1872
+ dev->lun != UFS_UPIU_BOOT_WLUN && dev->lun != UFS_UPIU_RPMB_WLUN) {
1873
+ error_setg(errp, "bad well-known lun: %d", dev->lun);
1874
+ return false;
1875
+ }
1876
+
1877
+ if ((dev->lun == UFS_UPIU_REPORT_LUNS_WLUN && u->report_wlu != NULL) ||
1878
+ (dev->lun == UFS_UPIU_UFS_DEVICE_WLUN && u->dev_wlu != NULL) ||
1879
+ (dev->lun == UFS_UPIU_BOOT_WLUN && u->boot_wlu != NULL) ||
1880
+ (dev->lun == UFS_UPIU_RPMB_WLUN && u->rpmb_wlu != NULL)) {
1881
+ error_setg(errp, "well-known lun %d already exists", dev->lun);
1882
+ return false;
1883
+ }
1884
+
1885
+ return true;
1886
+ }
1887
+
1888
+ if (strcmp(object_get_typename(OBJECT(dev)), TYPE_UFS_LU) != 0) {
1889
+ error_setg(errp, "%s cannot be connected to ufs-bus",
1890
+ object_get_typename(OBJECT(dev)));
1891
+ return false;
1892
+ }
1893
+
1894
+ return ubc->parent_check_address(qbus, qdev, errp);
1895
+}
1896
+
1897
+static void ufs_bus_class_init(ObjectClass *class, void *data)
1898
+{
1899
+ BusClass *bc = BUS_CLASS(class);
1900
+ UfsBusClass *ubc = UFS_BUS_CLASS(class);
1901
+ ubc->parent_check_address = bc->check_address;
1902
+ bc->check_address = ufs_bus_check_address;
1903
+}
1904
+
1905
static const TypeInfo ufs_info = {
1906
.name = TYPE_UFS,
1907
.parent = TYPE_PCI_DEVICE,
1908
@@ -XXX,XX +XXX,XX @@ static const TypeInfo ufs_info = {
1909
.interfaces = (InterfaceInfo[]){ { INTERFACE_PCIE_DEVICE }, {} },
1910
};
1911
1912
+static const TypeInfo ufs_bus_info = {
1913
+ .name = TYPE_UFS_BUS,
1914
+ .parent = TYPE_SCSI_BUS,
1915
+ .class_init = ufs_bus_class_init,
1916
+ .class_size = sizeof(UfsBusClass),
1917
+ .instance_size = sizeof(UfsBus),
1918
+};
1919
+
1920
static void ufs_register_types(void)
1921
{
1922
type_register_static(&ufs_info);
1923
+ type_register_static(&ufs_bus_info);
1924
}
1925
1926
type_init(ufs_register_types)
1927
diff --git a/hw/ufs/meson.build b/hw/ufs/meson.build
1928
index XXXXXXX..XXXXXXX 100644
1929
--- a/hw/ufs/meson.build
1930
+++ b/hw/ufs/meson.build
1931
@@ -1 +1 @@
1932
-system_ss.add(when: 'CONFIG_UFS_PCI', if_true: files('ufs.c'))
1933
+system_ss.add(when: 'CONFIG_UFS_PCI', if_true: files('ufs.c', 'lu.c'))
1934
diff --git a/hw/ufs/trace-events b/hw/ufs/trace-events
1935
index XXXXXXX..XXXXXXX 100644
1936
--- a/hw/ufs/trace-events
1937
+++ b/hw/ufs/trace-events
1938
@@ -XXX,XX +XXX,XX @@ ufs_exec_scsi_cmd(uint32_t slot, uint8_t lun, uint8_t opcode) "slot %"PRIu32", l
1939
ufs_exec_query_cmd(uint32_t slot, uint8_t opcode) "slot %"PRIu32", opcode 0x%"PRIx8""
1940
ufs_process_uiccmd(uint32_t uiccmd, uint32_t ucmdarg1, uint32_t ucmdarg2, uint32_t ucmdarg3) "uiccmd 0x%"PRIx32", ucmdarg1 0x%"PRIx32", ucmdarg2 0x%"PRIx32", ucmdarg3 0x%"PRIx32""
1941
1942
+# lu.c
1943
+ufs_scsi_check_condition(uint32_t tag, uint8_t key, uint8_t asc, uint8_t ascq) "Command complete tag=0x%x sense=%d/%d/%d"
1944
+ufs_scsi_read_complete(uint32_t tag, size_t size) "Data ready tag=0x%x len=%zd"
1945
+ufs_scsi_read_data_count(uint32_t sector_count) "Read sector_count=%d"
1946
+ufs_scsi_read_data_invalid(void) "Data transfer direction invalid"
1947
+ufs_scsi_write_complete_noio(uint32_t tag, size_t size) "Write complete tag=0x%x more=%zd"
1948
+ufs_scsi_write_data_invalid(void) "Data transfer direction invalid"
1949
+ufs_scsi_emulate_vpd_page_00(size_t xfer) "Inquiry EVPD[Supported pages] buffer size %zd"
1950
+ufs_scsi_emulate_vpd_page_80_not_supported(void) "Inquiry EVPD[Serial number] not supported"
1951
+ufs_scsi_emulate_vpd_page_80(size_t xfer) "Inquiry EVPD[Serial number] buffer size %zd"
1952
+ufs_scsi_emulate_vpd_page_87(size_t xfer) "Inquiry EVPD[Mode Page Policy] buffer size %zd"
1953
+ufs_scsi_emulate_mode_sense(int cmd, int page, size_t xfer, int control) "Mode Sense(%d) (page %d, xfer %zd, page_control %d)"
1954
+ufs_scsi_emulate_read_data(int buflen) "Read buf_len=%d"
1955
+ufs_scsi_emulate_write_data(int buflen) "Write buf_len=%d"
1956
+ufs_scsi_emulate_command_START_STOP(void) "START STOP UNIT"
1957
+ufs_scsi_emulate_command_FORMAT_UNIT(void) "FORMAT UNIT"
1958
+ufs_scsi_emulate_command_SEND_DIAGNOSTIC(void) "SEND DIAGNOSTIC"
1959
+ufs_scsi_emulate_command_SAI_16(void) "SAI READ CAPACITY(16)"
1960
+ufs_scsi_emulate_command_SAI_unsupported(void) "Unsupported Service Action In"
1961
+ufs_scsi_emulate_command_MODE_SELECT_10(size_t xfer) "Mode Select(10) (len %zd)"
1962
+ufs_scsi_emulate_command_VERIFY(int bytchk) "Verify (bytchk %d)"
1963
+ufs_scsi_emulate_command_UNKNOWN(int cmd, const char *name) "Unknown SCSI command (0x%2.2x=%s)"
1964
+ufs_scsi_dma_command_READ(uint64_t lba, uint32_t len) "Read (block %" PRIu64 ", count %u)"
1965
+ufs_scsi_dma_command_WRITE(uint64_t lba, int len) "Write (block %" PRIu64 ", count %u)"
1966
+
1967
# error condition
1968
ufs_err_dma_read_utrd(uint32_t slot, uint64_t addr) "failed to read utrd. UTRLDBR slot %"PRIu32", UTRD dma addr %"PRIu64""
1969
ufs_err_dma_read_req_upiu(uint32_t slot, uint64_t addr) "failed to read req upiu. UTRLDBR slot %"PRIu32", request upiu addr %"PRIu64""
1970
--
205
--
1971
2.41.0
206
2.9.3
207
208
diff view generated by jsdifflib
New patch
1
From: Paolo Bonzini <pbonzini@redhat.com>
1
2
3
This adds a CoMutex around the existing CoQueue. Because the write-side
4
can just take CoMutex, the old "writer" field is not necessary anymore.
5
Instead of removing it altogether, count the number of pending writers
6
during a read-side critical section and forbid further readers from
7
entering.
8
9
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
10
Reviewed-by: Fam Zheng <famz@redhat.com>
11
Message-id: 20170213181244.16297-7-pbonzini@redhat.com
12
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
13
---
14
include/qemu/coroutine.h | 3 ++-
15
util/qemu-coroutine-lock.c | 35 ++++++++++++++++++++++++-----------
16
2 files changed, 26 insertions(+), 12 deletions(-)
17
18
diff --git a/include/qemu/coroutine.h b/include/qemu/coroutine.h
19
index XXXXXXX..XXXXXXX 100644
20
--- a/include/qemu/coroutine.h
21
+++ b/include/qemu/coroutine.h
22
@@ -XXX,XX +XXX,XX @@ bool qemu_co_queue_empty(CoQueue *queue);
23
24
25
typedef struct CoRwlock {
26
- bool writer;
27
+ int pending_writer;
28
int reader;
29
+ CoMutex mutex;
30
CoQueue queue;
31
} CoRwlock;
32
33
diff --git a/util/qemu-coroutine-lock.c b/util/qemu-coroutine-lock.c
34
index XXXXXXX..XXXXXXX 100644
35
--- a/util/qemu-coroutine-lock.c
36
+++ b/util/qemu-coroutine-lock.c
37
@@ -XXX,XX +XXX,XX @@ void qemu_co_rwlock_init(CoRwlock *lock)
38
{
39
memset(lock, 0, sizeof(*lock));
40
qemu_co_queue_init(&lock->queue);
41
+ qemu_co_mutex_init(&lock->mutex);
42
}
43
44
void qemu_co_rwlock_rdlock(CoRwlock *lock)
45
{
46
Coroutine *self = qemu_coroutine_self();
47
48
- while (lock->writer) {
49
- qemu_co_queue_wait(&lock->queue, NULL);
50
+ qemu_co_mutex_lock(&lock->mutex);
51
+ /* For fairness, wait if a writer is in line. */
52
+ while (lock->pending_writer) {
53
+ qemu_co_queue_wait(&lock->queue, &lock->mutex);
54
}
55
lock->reader++;
56
+ qemu_co_mutex_unlock(&lock->mutex);
57
+
58
+ /* The rest of the read-side critical section is run without the mutex. */
59
self->locks_held++;
60
}
61
62
@@ -XXX,XX +XXX,XX @@ void qemu_co_rwlock_unlock(CoRwlock *lock)
63
Coroutine *self = qemu_coroutine_self();
64
65
assert(qemu_in_coroutine());
66
- if (lock->writer) {
67
- lock->writer = false;
68
+ if (!lock->reader) {
69
+ /* The critical section started in qemu_co_rwlock_wrlock. */
70
qemu_co_queue_restart_all(&lock->queue);
71
} else {
72
+ self->locks_held--;
73
+
74
+ qemu_co_mutex_lock(&lock->mutex);
75
lock->reader--;
76
assert(lock->reader >= 0);
77
/* Wakeup only one waiting writer */
78
@@ -XXX,XX +XXX,XX @@ void qemu_co_rwlock_unlock(CoRwlock *lock)
79
qemu_co_queue_next(&lock->queue);
80
}
81
}
82
- self->locks_held--;
83
+ qemu_co_mutex_unlock(&lock->mutex);
84
}
85
86
void qemu_co_rwlock_wrlock(CoRwlock *lock)
87
{
88
- Coroutine *self = qemu_coroutine_self();
89
-
90
- while (lock->writer || lock->reader) {
91
- qemu_co_queue_wait(&lock->queue, NULL);
92
+ qemu_co_mutex_lock(&lock->mutex);
93
+ lock->pending_writer++;
94
+ while (lock->reader) {
95
+ qemu_co_queue_wait(&lock->queue, &lock->mutex);
96
}
97
- lock->writer = true;
98
- self->locks_held++;
99
+ lock->pending_writer--;
100
+
101
+ /* The rest of the write-side critical section is run with
102
+ * the mutex taken, so that lock->reader remains zero.
103
+ * There is no need to update self->locks_held.
104
+ */
105
}
106
--
107
2.9.3
108
109
diff view generated by jsdifflib