1 | The following changes since commit 5ec2eca83dc478ddf24077e02a8b34dd26cd3ff9: | 1 | The following changes since commit 56f9e46b841c7be478ca038d8d4085d776ab4b0d: |
---|---|---|---|
2 | 2 | ||
3 | Merge remote-tracking branch 'remotes/awilliam/tags/vfio-updates-20190613.0' into staging (2019-06-14 09:33:55 +0100) | 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://github.com/XanClic/qemu.git tags/pull-block-2019-06-14 | 7 | git://github.com/stefanha/qemu.git tags/block-pull-request |
8 | 8 | ||
9 | for you to fetch changes up to 21c1ce592a144188dfe59b9e156a97da412a59a2: | 9 | for you to fetch changes up to a7b91d35bab97a2d3e779d0c64c9b837b52a6cf7: |
10 | 10 | ||
11 | iotests: Test qemu-img convert -C --salvage (2019-06-14 15:09:42 +0200) | 11 | coroutine-lock: make CoRwlock thread-safe and fair (2017-02-21 11:39:40 +0000) |
12 | 12 | ||
13 | ---------------------------------------------------------------- | 13 | ---------------------------------------------------------------- |
14 | Block patches: | 14 | Pull request |
15 | - Allow blockdev-backup from nodes that are not in qemu's main AIO | 15 | |
16 | context to newly added nodes | 16 | v2: |
17 | - Add salvaging mode to qemu-img convert | 17 | * Rebased to resolve scsi conflicts |
18 | - Minor fixes to tests, documentation, and for less Valgrind annoyance | ||
19 | 18 | ||
20 | ---------------------------------------------------------------- | 19 | ---------------------------------------------------------------- |
21 | Andrey Shinkevich (1): | ||
22 | hw/block/fdc: floppy command FIFO memory initialization | ||
23 | 20 | ||
24 | John Snow (6): | 21 | Paolo Bonzini (24): |
25 | blockdev-backup: don't check aio_context too early | 22 | block: move AioContext, QEMUTimer, main-loop to libqemuutil |
26 | iotests.py: do not use infinite waits | 23 | aio: introduce aio_co_schedule and aio_co_wake |
27 | QEMUMachine: add events_wait method | 24 | block-backend: allow blk_prw from coroutine context |
28 | iotests.py: rewrite run_job to be pickier | 25 | test-thread-pool: use generic AioContext infrastructure |
29 | iotests: add iotest 256 for testing blockdev-backup across iothread | 26 | io: add methods to set I/O handlers on AioContext |
30 | contexts | 27 | io: make qio_channel_yield aware of AioContexts |
31 | event_match: always match on None value | 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 | ||
32 | 48 | ||
33 | Max Reitz (12): | 49 | Makefile.objs | 4 - |
34 | iotests: Filter 175's allocation information | 50 | stubs/Makefile.objs | 1 + |
35 | iotests: Fix intermittent failure in 219 | 51 | tests/Makefile.include | 19 +- |
36 | qemu-img: Fix options leakage in img_rebase() | 52 | util/Makefile.objs | 6 +- |
37 | qapi/block-core: Overlays are not snapshots | 53 | block/nbd-client.h | 2 +- |
38 | blockdev: Overlays are not snapshots | 54 | block/qed.h | 3 + |
39 | qemu-img: Move quiet into ImgConvertState | 55 | include/block/aio.h | 38 ++- |
40 | qemu-img: Add salvaging mode to convert | 56 | include/block/block_int.h | 64 +++-- |
41 | blkdebug: Add @iotype error option | 57 | include/io/channel.h | 72 +++++- |
42 | blkdebug: Add "none" event | 58 | include/qemu/coroutine.h | 84 ++++--- |
43 | blkdebug: Inject errors on .bdrv_co_block_status() | 59 | include/qemu/coroutine_int.h | 11 +- |
44 | iotests: Test qemu-img convert --salvage | 60 | include/sysemu/block-backend.h | 14 +- |
45 | iotests: Test qemu-img convert -C --salvage | 61 | tests/iothread.h | 25 ++ |
46 | 62 | block/backup.c | 2 +- | |
47 | Vladimir Sementsov-Ogievskiy (1): | 63 | block/blkdebug.c | 9 +- |
48 | iotests: restrict 254 to support only qcow2 | 64 | block/blkreplay.c | 2 +- |
49 | 65 | block/block-backend.c | 13 +- | |
50 | qapi/block-core.json | 53 ++++++++--- | 66 | block/curl.c | 44 +++- |
51 | block/blkdebug.c | 60 ++++++++++-- | 67 | block/gluster.c | 9 +- |
52 | blockdev.c | 14 +-- | 68 | block/io.c | 42 +--- |
53 | hw/block/fdc.c | 1 + | 69 | block/iscsi.c | 15 +- |
54 | qemu-img.c | 106 +++++++++++++++------ | 70 | block/linux-aio.c | 10 +- |
55 | python/qemu/__init__.py | 67 ++++++++++---- | 71 | block/mirror.c | 12 +- |
56 | qemu-img-cmds.hx | 4 +- | 72 | block/nbd-client.c | 119 +++++---- |
57 | qemu-img.texi | 4 + | 73 | block/nfs.c | 9 +- |
58 | tests/qemu-iotests/082 | 1 + | 74 | block/qcow2-cluster.c | 4 +- |
59 | tests/qemu-iotests/082.out | 3 + | 75 | block/qed-cluster.c | 2 + |
60 | tests/qemu-iotests/085.out | 10 +- | 76 | block/qed-table.c | 12 +- |
61 | tests/qemu-iotests/175 | 26 +++++- | 77 | block/qed.c | 58 +++-- |
62 | tests/qemu-iotests/175.out | 8 +- | 78 | block/sheepdog.c | 31 +-- |
63 | tests/qemu-iotests/219 | 13 ++- | 79 | block/ssh.c | 29 +-- |
64 | tests/qemu-iotests/251 | 170 ++++++++++++++++++++++++++++++++++ | 80 | block/throttle-groups.c | 4 +- |
65 | tests/qemu-iotests/251.out | 43 +++++++++ | 81 | block/win32-aio.c | 9 +- |
66 | tests/qemu-iotests/254 | 2 + | 82 | dma-helpers.c | 2 + |
67 | tests/qemu-iotests/256 | 122 ++++++++++++++++++++++++ | 83 | hw/9pfs/9p.c | 2 +- |
68 | tests/qemu-iotests/256.out | 119 ++++++++++++++++++++++++ | 84 | hw/block/virtio-blk.c | 19 +- |
69 | tests/qemu-iotests/group | 2 + | 85 | hw/scsi/scsi-bus.c | 2 + |
70 | tests/qemu-iotests/iotests.py | 60 +++++++----- | 86 | hw/scsi/scsi-disk.c | 15 ++ |
71 | 21 files changed, 772 insertions(+), 116 deletions(-) | 87 | hw/scsi/scsi-generic.c | 20 +- |
72 | create mode 100755 tests/qemu-iotests/251 | 88 | hw/scsi/virtio-scsi.c | 7 + |
73 | create mode 100644 tests/qemu-iotests/251.out | 89 | io/channel-command.c | 13 + |
74 | create mode 100755 tests/qemu-iotests/256 | 90 | io/channel-file.c | 11 + |
75 | create mode 100644 tests/qemu-iotests/256.out | 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%) | ||
76 | 129 | ||
77 | -- | 130 | -- |
78 | 2.21.0 | 131 | 2.9.3 |
79 | 132 | ||
80 | 133 | diff view generated by jsdifflib |
1 | A snapshot is something that reflects the state of something at a | 1 | From: Paolo Bonzini <pbonzini@redhat.com> |
---|---|---|---|
2 | certain point in time. It does not change. | 2 | |
3 | 3 | AioContext is fairly self contained, the only dependency is QEMUTimer but | |
4 | The file our snapshot commands create (or the node they install) is not | 4 | that in turn doesn't need anything else. So move them out of block-obj-y |
5 | a snapshot, as it does change over time. It is an overlay. We cannot | 5 | to avoid introducing a dependency from io/ to block-obj-y. |
6 | do anything about the parameter names, but we can at least adjust the | 6 | |
7 | descriptions to reflect that fact. | 7 | main-loop and its dependency iohandler also need to be moved, because |
8 | 8 | later in this series io/ will call iohandler_get_aio_context. | |
9 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 9 | |
10 | Reviewed-by: Eric Blake <eblake@redhat.com> | 10 | [Changed copyright "the QEMU team" to "other QEMU contributors" as |
11 | Message-id: 20190603202236.1342-2-mreitz@redhat.com | 11 | suggested by Daniel Berrange and agreed by Paolo. |
12 | Reviewed-by: John Snow <jsnow@redhat.com> | 12 | --Stefan] |
13 | Reviewed-by: Alberto Garcia <berto@igalia.com> | 13 | |
14 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 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 | ||
17 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
15 | --- | 18 | --- |
16 | qapi/block-core.json | 20 ++++++++++---------- | 19 | Makefile.objs | 4 --- |
17 | 1 file changed, 10 insertions(+), 10 deletions(-) | 20 | stubs/Makefile.objs | 1 + |
18 | 21 | tests/Makefile.include | 11 ++++---- | |
19 | diff --git a/qapi/block-core.json b/qapi/block-core.json | 22 | util/Makefile.objs | 6 +++- |
20 | index XXXXXXX..XXXXXXX 100644 | 23 | block/io.c | 29 ------------------- |
21 | --- a/qapi/block-core.json | 24 | stubs/linux-aio.c | 32 +++++++++++++++++++++ |
22 | +++ b/qapi/block-core.json | 25 | stubs/set-fd-handler.c | 11 -------- |
26 | aio-posix.c => util/aio-posix.c | 2 +- | ||
27 | aio-win32.c => util/aio-win32.c | 0 | ||
28 | util/aiocb.c | 55 +++++++++++++++++++++++++++++++++++++ | ||
29 | async.c => util/async.c | 3 +- | ||
30 | iohandler.c => util/iohandler.c | 0 | ||
31 | main-loop.c => util/main-loop.c | 0 | ||
32 | qemu-timer.c => util/qemu-timer.c | 0 | ||
33 | thread-pool.c => util/thread-pool.c | 2 +- | ||
34 | trace-events | 11 -------- | ||
35 | util/trace-events | 11 ++++++++ | ||
36 | 17 files changed, 114 insertions(+), 64 deletions(-) | ||
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 | ||
184 | new file mode 100644 | ||
185 | index XXXXXXX..XXXXXXX | ||
186 | --- /dev/null | ||
187 | +++ b/stubs/linux-aio.c | ||
188 | @@ -XXX,XX +XXX,XX @@ | ||
189 | +/* | ||
190 | + * Linux native AIO support. | ||
191 | + * | ||
192 | + * Copyright (C) 2009 IBM, Corp. | ||
193 | + * Copyright (C) 2009 Red Hat, Inc. | ||
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. | ||
197 | + */ | ||
198 | +#include "qemu/osdep.h" | ||
199 | +#include "block/aio.h" | ||
200 | +#include "block/raw-aio.h" | ||
201 | + | ||
202 | +void laio_detach_aio_context(LinuxAioState *s, AioContext *old_context) | ||
203 | +{ | ||
204 | + abort(); | ||
205 | +} | ||
206 | + | ||
207 | +void laio_attach_aio_context(LinuxAioState *s, AioContext *new_context) | ||
208 | +{ | ||
209 | + abort(); | ||
210 | +} | ||
211 | + | ||
212 | +LinuxAioState *laio_init(void) | ||
213 | +{ | ||
214 | + abort(); | ||
215 | +} | ||
216 | + | ||
217 | +void laio_cleanup(LinuxAioState *s) | ||
218 | +{ | ||
219 | + abort(); | ||
220 | +} | ||
221 | diff --git a/stubs/set-fd-handler.c b/stubs/set-fd-handler.c | ||
222 | index XXXXXXX..XXXXXXX 100644 | ||
223 | --- a/stubs/set-fd-handler.c | ||
224 | +++ b/stubs/set-fd-handler.c | ||
225 | @@ -XXX,XX +XXX,XX @@ void qemu_set_fd_handler(int fd, | ||
226 | { | ||
227 | abort(); | ||
228 | } | ||
229 | - | ||
230 | -void aio_set_fd_handler(AioContext *ctx, | ||
231 | - int fd, | ||
232 | - bool is_external, | ||
233 | - IOHandler *io_read, | ||
234 | - IOHandler *io_write, | ||
235 | - AioPollFn *io_poll, | ||
236 | - void *opaque) | ||
237 | -{ | ||
238 | - abort(); | ||
239 | -} | ||
240 | diff --git a/aio-posix.c b/util/aio-posix.c | ||
241 | similarity index 99% | ||
242 | rename from aio-posix.c | ||
243 | rename to util/aio-posix.c | ||
244 | index XXXXXXX..XXXXXXX 100644 | ||
245 | --- a/aio-posix.c | ||
246 | +++ b/util/aio-posix.c | ||
247 | @@ -XXX,XX +XXX,XX @@ | ||
248 | #include "qemu/rcu_queue.h" | ||
249 | #include "qemu/sockets.h" | ||
250 | #include "qemu/cutils.h" | ||
251 | -#include "trace-root.h" | ||
252 | +#include "trace.h" | ||
253 | #ifdef CONFIG_EPOLL_CREATE1 | ||
254 | #include <sys/epoll.h> | ||
255 | #endif | ||
256 | diff --git a/aio-win32.c b/util/aio-win32.c | ||
257 | similarity index 100% | ||
258 | rename from aio-win32.c | ||
259 | rename to util/aio-win32.c | ||
260 | diff --git a/util/aiocb.c b/util/aiocb.c | ||
261 | new file mode 100644 | ||
262 | index XXXXXXX..XXXXXXX | ||
263 | --- /dev/null | ||
264 | +++ b/util/aiocb.c | ||
265 | @@ -XXX,XX +XXX,XX @@ | ||
266 | +/* | ||
267 | + * BlockAIOCB allocation | ||
268 | + * | ||
269 | + * Copyright (c) 2003-2017 Fabrice Bellard and other QEMU contributors | ||
270 | + * | ||
271 | + * Permission is hereby granted, free of charge, to any person obtaining a copy | ||
272 | + * of this software and associated documentation files (the "Software"), to deal | ||
273 | + * in the Software without restriction, including without limitation the rights | ||
274 | + * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | ||
275 | + * copies of the Software, and to permit persons to whom the Software is | ||
276 | + * furnished to do so, subject to the following conditions: | ||
277 | + * | ||
278 | + * The above copyright notice and this permission notice shall be included in | ||
279 | + * all copies or substantial portions of the Software. | ||
280 | + * | ||
281 | + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | ||
282 | + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | ||
283 | + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | ||
284 | + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | ||
285 | + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | ||
286 | + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | ||
287 | + * THE SOFTWARE. | ||
288 | + */ | ||
289 | + | ||
290 | +#include "qemu/osdep.h" | ||
291 | +#include "block/aio.h" | ||
292 | + | ||
293 | +void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs, | ||
294 | + BlockCompletionFunc *cb, void *opaque) | ||
295 | +{ | ||
296 | + BlockAIOCB *acb; | ||
297 | + | ||
298 | + acb = g_malloc(aiocb_info->aiocb_size); | ||
299 | + acb->aiocb_info = aiocb_info; | ||
300 | + acb->bs = bs; | ||
301 | + acb->cb = cb; | ||
302 | + acb->opaque = opaque; | ||
303 | + acb->refcnt = 1; | ||
304 | + return acb; | ||
305 | +} | ||
306 | + | ||
307 | +void qemu_aio_ref(void *p) | ||
308 | +{ | ||
309 | + BlockAIOCB *acb = p; | ||
310 | + acb->refcnt++; | ||
311 | +} | ||
312 | + | ||
313 | +void qemu_aio_unref(void *p) | ||
314 | +{ | ||
315 | + BlockAIOCB *acb = p; | ||
316 | + assert(acb->refcnt > 0); | ||
317 | + if (--acb->refcnt == 0) { | ||
318 | + g_free(acb); | ||
319 | + } | ||
320 | +} | ||
321 | diff --git a/async.c b/util/async.c | ||
322 | similarity index 99% | ||
323 | rename from async.c | ||
324 | rename to util/async.c | ||
325 | index XXXXXXX..XXXXXXX 100644 | ||
326 | --- a/async.c | ||
327 | +++ b/util/async.c | ||
328 | @@ -XXX,XX +XXX,XX @@ | ||
329 | /* | ||
330 | - * QEMU System Emulator | ||
331 | + * Data plane event loop | ||
332 | * | ||
333 | * Copyright (c) 2003-2008 Fabrice Bellard | ||
334 | + * Copyright (c) 2009-2017 QEMU contributors | ||
335 | * | ||
336 | * Permission is hereby granted, free of charge, to any person obtaining a copy | ||
337 | * of this software and associated documentation files (the "Software"), to deal | ||
338 | diff --git a/iohandler.c b/util/iohandler.c | ||
339 | similarity index 100% | ||
340 | rename from iohandler.c | ||
341 | rename to util/iohandler.c | ||
342 | diff --git a/main-loop.c b/util/main-loop.c | ||
343 | similarity index 100% | ||
344 | rename from main-loop.c | ||
345 | rename to util/main-loop.c | ||
346 | diff --git a/qemu-timer.c b/util/qemu-timer.c | ||
347 | similarity index 100% | ||
348 | rename from qemu-timer.c | ||
349 | rename to util/qemu-timer.c | ||
350 | diff --git a/thread-pool.c b/util/thread-pool.c | ||
351 | similarity index 99% | ||
352 | rename from thread-pool.c | ||
353 | rename to util/thread-pool.c | ||
354 | index XXXXXXX..XXXXXXX 100644 | ||
355 | --- a/thread-pool.c | ||
356 | +++ b/util/thread-pool.c | ||
357 | @@ -XXX,XX +XXX,XX @@ | ||
358 | #include "qemu/queue.h" | ||
359 | #include "qemu/thread.h" | ||
360 | #include "qemu/coroutine.h" | ||
361 | -#include "trace-root.h" | ||
362 | +#include "trace.h" | ||
363 | #include "block/thread-pool.h" | ||
364 | #include "qemu/main-loop.h" | ||
365 | |||
366 | diff --git a/trace-events b/trace-events | ||
367 | index XXXXXXX..XXXXXXX 100644 | ||
368 | --- a/trace-events | ||
369 | +++ b/trace-events | ||
23 | @@ -XXX,XX +XXX,XX @@ | 370 | @@ -XXX,XX +XXX,XX @@ |
24 | # | 371 | # |
25 | # Either @device or @node-name must be set but not both. | 372 | # The <format-string> should be a sprintf()-compatible format string. |
26 | # | 373 | |
27 | -# @device: the name of the device to generate the snapshot from. | 374 | -# aio-posix.c |
28 | +# @device: the name of the device to take a snapshot of. | 375 | -run_poll_handlers_begin(void *ctx, int64_t max_ns) "ctx %p max_ns %"PRId64 |
29 | # | 376 | -run_poll_handlers_end(void *ctx, bool progress) "ctx %p progress %d" |
30 | # @node-name: graph node name to generate the snapshot from (Since 2.0) | 377 | -poll_shrink(void *ctx, int64_t old, int64_t new) "ctx %p old %"PRId64" new %"PRId64 |
31 | # | 378 | -poll_grow(void *ctx, int64_t old, int64_t new) "ctx %p old %"PRId64" new %"PRId64 |
32 | -# @snapshot-file: the target of the new image. If the file exists, or | 379 | - |
33 | -# if it is a device, the snapshot will be created in the existing | 380 | -# thread-pool.c |
34 | -# file/device. Otherwise, a new file will be created. | 381 | -thread_pool_submit(void *pool, void *req, void *opaque) "pool %p req %p opaque %p" |
35 | +# @snapshot-file: the target of the new overlay image. If the file | 382 | -thread_pool_complete(void *pool, void *req, void *opaque, int ret) "pool %p req %p opaque %p ret %d" |
36 | +# exists, or if it is a device, the overlay will be created in the | 383 | -thread_pool_cancel(void *req, void *opaque) "req %p opaque %p" |
37 | +# existing file/device. Otherwise, a new file will be created. | 384 | - |
38 | # | 385 | # ioport.c |
39 | # @snapshot-node-name: the graph node name of the new image (Since 2.0) | 386 | cpu_in(unsigned int addr, char size, unsigned int val) "addr %#x(%c) value %u" |
40 | # | 387 | cpu_out(unsigned int addr, char size, unsigned int val) "addr %#x(%c) value %u" |
41 | -# @format: the format of the snapshot image, default is 'qcow2'. | 388 | diff --git a/util/trace-events b/util/trace-events |
42 | +# @format: the format of the overlay image, default is 'qcow2'. | 389 | index XXXXXXX..XXXXXXX 100644 |
43 | # | 390 | --- a/util/trace-events |
44 | # @mode: whether and how QEMU should create a new image, default is | 391 | +++ b/util/trace-events |
45 | # 'absolute-paths'. | 392 | @@ -XXX,XX +XXX,XX @@ |
46 | @@ -XXX,XX +XXX,XX @@ | 393 | # See docs/tracing.txt for syntax documentation. |
47 | ## | 394 | |
48 | # @BlockdevSnapshot: | 395 | +# util/aio-posix.c |
49 | # | 396 | +run_poll_handlers_begin(void *ctx, int64_t max_ns) "ctx %p max_ns %"PRId64 |
50 | -# @node: device or node name that will have a snapshot created. | 397 | +run_poll_handlers_end(void *ctx, bool progress) "ctx %p progress %d" |
51 | +# @node: device or node name that will have a snapshot taken. | 398 | +poll_shrink(void *ctx, int64_t old, int64_t new) "ctx %p old %"PRId64" new %"PRId64 |
52 | # | 399 | +poll_grow(void *ctx, int64_t old, int64_t new) "ctx %p old %"PRId64" new %"PRId64 |
53 | # @overlay: reference to the existing block device that will become | 400 | + |
54 | -# the overlay of @node, as part of creating the snapshot. | 401 | +# util/thread-pool.c |
55 | +# the overlay of @node, as part of taking the snapshot. | 402 | +thread_pool_submit(void *pool, void *req, void *opaque) "pool %p req %p opaque %p" |
56 | # It must not have a current backing file (this can be | 403 | +thread_pool_complete(void *pool, void *req, void *opaque, int ret) "pool %p req %p opaque %p ret %d" |
57 | # achieved by passing "backing": null to blockdev-add). | 404 | +thread_pool_cancel(void *req, void *opaque) "req %p opaque %p" |
58 | # | 405 | + |
59 | @@ -XXX,XX +XXX,XX @@ | 406 | # util/buffer.c |
60 | ## | 407 | buffer_resize(const char *buf, size_t olen, size_t len) "%s: old %zd, new %zd" |
61 | # @blockdev-snapshot-sync: | 408 | buffer_move_empty(const char *buf, size_t len, const char *from) "%s: %zd bytes from %s" |
62 | # | ||
63 | -# Generates a synchronous snapshot of a block device. | ||
64 | +# Takes a synchronous snapshot of a block device. | ||
65 | # | ||
66 | # For the arguments, see the documentation of BlockdevSnapshotSync. | ||
67 | # | ||
68 | @@ -XXX,XX +XXX,XX @@ | ||
69 | ## | ||
70 | # @blockdev-snapshot: | ||
71 | # | ||
72 | -# Generates a snapshot of a block device. | ||
73 | +# Takes a snapshot of a block device. | ||
74 | # | ||
75 | -# Create a snapshot, by installing 'node' as the backing image of | ||
76 | +# Take a snapshot, by installing 'node' as the backing image of | ||
77 | # 'overlay'. Additionally, if 'node' is associated with a block | ||
78 | # device, the block device changes to using 'overlay' as its new active | ||
79 | # image. | ||
80 | -- | 409 | -- |
81 | 2.21.0 | 410 | 2.9.3 |
82 | 411 | ||
83 | 412 | diff view generated by jsdifflib |
1 | From: John Snow <jsnow@redhat.com> | 1 | From: Paolo Bonzini <pbonzini@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | Signed-off-by: John Snow <jsnow@redhat.com> | 3 | aio_co_wake provides the infrastructure to start a coroutine on a "home" |
4 | Message-id: 20190523170643.20794-6-jsnow@redhat.com | 4 | AioContext. It will be used by CoMutex and CoQueue, so that coroutines |
5 | Reviewed-by: Max Reitz <mreitz@redhat.com> | 5 | don't jump from one context to another when they go to sleep on a |
6 | [mreitz: Moved from 250 to 256] | 6 | mutex or waitqueue. However, it can also be used as a more efficient |
7 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 7 | alternative to one-shot bottom halves, and saves the effort of tracking |
8 | which AioContext a coroutine is running on. | ||
9 | |||
10 | aio_co_schedule is the part of aio_co_wake that starts a coroutine | ||
11 | on a remove AioContext, but it is also useful to implement e.g. | ||
12 | bdrv_set_aio_context callbacks. | ||
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 | ||
27 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
8 | --- | 28 | --- |
9 | tests/qemu-iotests/256 | 122 +++++++++++++++++++++++++++++++++++++ | 29 | tests/Makefile.include | 8 +- |
10 | tests/qemu-iotests/256.out | 119 ++++++++++++++++++++++++++++++++++++ | 30 | include/block/aio.h | 32 +++++++ |
11 | tests/qemu-iotests/group | 1 + | 31 | include/qemu/coroutine_int.h | 11 ++- |
12 | 3 files changed, 242 insertions(+) | 32 | tests/iothread.h | 25 +++++ |
13 | create mode 100755 tests/qemu-iotests/256 | 33 | tests/iothread.c | 91 ++++++++++++++++++ |
14 | create mode 100644 tests/qemu-iotests/256.out | 34 | tests/test-aio-multithread.c | 213 +++++++++++++++++++++++++++++++++++++++++++ |
15 | 35 | util/async.c | 65 +++++++++++++ | |
16 | diff --git a/tests/qemu-iotests/256 b/tests/qemu-iotests/256 | 36 | util/qemu-coroutine.c | 8 ++ |
17 | new file mode 100755 | 37 | util/trace-events | 4 + |
18 | index XXXXXXX..XXXXXXX | 38 | 9 files changed, 453 insertions(+), 4 deletions(-) |
19 | --- /dev/null | 39 | create mode 100644 tests/iothread.h |
20 | +++ b/tests/qemu-iotests/256 | 40 | create mode 100644 tests/iothread.c |
21 | @@ -XXX,XX +XXX,XX @@ | 41 | create mode 100644 tests/test-aio-multithread.c |
22 | +#!/usr/bin/env python | 42 | |
23 | +# | 43 | diff --git a/tests/Makefile.include b/tests/Makefile.include |
24 | +# Test incremental/backup across iothread contexts | 44 | index XXXXXXX..XXXXXXX 100644 |
25 | +# | 45 | --- a/tests/Makefile.include |
26 | +# Copyright (c) 2019 John Snow for Red Hat, Inc. | 46 | +++ b/tests/Makefile.include |
27 | +# | 47 | @@ -XXX,XX +XXX,XX @@ check-unit-y += tests/test-aio$(EXESUF) |
28 | +# This program is free software; you can redistribute it and/or modify | 48 | gcov-files-test-aio-y = util/async.c util/qemu-timer.o |
29 | +# it under the terms of the GNU General Public License as published by | 49 | gcov-files-test-aio-$(CONFIG_WIN32) += util/aio-win32.c |
30 | +# the Free Software Foundation; either version 2 of the License, or | 50 | gcov-files-test-aio-$(CONFIG_POSIX) += util/aio-posix.c |
31 | +# (at your option) any later version. | 51 | +check-unit-y += tests/test-aio-multithread$(EXESUF) |
32 | +# | 52 | +gcov-files-test-aio-multithread-y = $(gcov-files-test-aio-y) |
33 | +# This program is distributed in the hope that it will be useful, | 53 | +gcov-files-test-aio-multithread-y += util/qemu-coroutine.c tests/iothread.c |
34 | +# but WITHOUT ANY WARRANTY; without even the implied warranty of | 54 | check-unit-y += tests/test-throttle$(EXESUF) |
35 | +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | 55 | -gcov-files-test-aio-$(CONFIG_WIN32) = aio-win32.c |
36 | +# GNU General Public License for more details. | 56 | -gcov-files-test-aio-$(CONFIG_POSIX) = aio-posix.c |
37 | +# | 57 | check-unit-y += tests/test-thread-pool$(EXESUF) |
38 | +# You should have received a copy of the GNU General Public License | 58 | gcov-files-test-thread-pool-y = thread-pool.c |
39 | +# along with this program. If not, see <http://www.gnu.org/licenses/>. | 59 | gcov-files-test-hbitmap-y = util/hbitmap.c |
40 | +# | 60 | @@ -XXX,XX +XXX,XX @@ test-qapi-obj-y = tests/test-qapi-visit.o tests/test-qapi-types.o \ |
41 | +# owner=jsnow@redhat.com | 61 | $(test-qom-obj-y) |
42 | + | 62 | test-crypto-obj-y = $(crypto-obj-y) $(test-qom-obj-y) |
43 | +import os | 63 | test-io-obj-y = $(io-obj-y) $(test-crypto-obj-y) |
44 | +import iotests | 64 | -test-block-obj-y = $(block-obj-y) $(test-io-obj-y) |
45 | +from iotests import log | 65 | +test-block-obj-y = $(block-obj-y) $(test-io-obj-y) tests/iothread.o |
46 | + | 66 | |
47 | +iotests.verify_image_format(supported_fmts=['qcow2']) | 67 | tests/check-qint$(EXESUF): tests/check-qint.o $(test-util-obj-y) |
48 | +size = 64 * 1024 * 1024 | 68 | tests/check-qstring$(EXESUF): tests/check-qstring.o $(test-util-obj-y) |
49 | + | 69 | @@ -XXX,XX +XXX,XX @@ tests/check-qom-proplist$(EXESUF): tests/check-qom-proplist.o $(test-qom-obj-y) |
50 | +with iotests.FilePath('img0') as img0_path, \ | 70 | tests/test-char$(EXESUF): tests/test-char.o $(test-util-obj-y) $(qtest-obj-y) $(test-io-obj-y) $(chardev-obj-y) |
51 | + iotests.FilePath('img1') as img1_path, \ | 71 | tests/test-coroutine$(EXESUF): tests/test-coroutine.o $(test-block-obj-y) |
52 | + iotests.FilePath('img0-full') as img0_full_path, \ | 72 | tests/test-aio$(EXESUF): tests/test-aio.o $(test-block-obj-y) |
53 | + iotests.FilePath('img1-full') as img1_full_path, \ | 73 | +tests/test-aio-multithread$(EXESUF): tests/test-aio-multithread.o $(test-block-obj-y) |
54 | + iotests.FilePath('img0-incr') as img0_incr_path, \ | 74 | tests/test-throttle$(EXESUF): tests/test-throttle.o $(test-block-obj-y) |
55 | + iotests.FilePath('img1-incr') as img1_incr_path, \ | 75 | tests/test-blockjob$(EXESUF): tests/test-blockjob.o $(test-block-obj-y) $(test-util-obj-y) |
56 | + iotests.VM() as vm: | 76 | tests/test-blockjob-txn$(EXESUF): tests/test-blockjob-txn.o $(test-block-obj-y) $(test-util-obj-y) |
57 | + | 77 | diff --git a/include/block/aio.h b/include/block/aio.h |
58 | + def create_target(filepath, name, size): | 78 | index XXXXXXX..XXXXXXX 100644 |
59 | + basename = os.path.basename(filepath) | 79 | --- a/include/block/aio.h |
60 | + nodename = "file_{}".format(basename) | 80 | +++ b/include/block/aio.h |
61 | + log(vm.command('blockdev-create', job_id='job1', | 81 | @@ -XXX,XX +XXX,XX @@ typedef void QEMUBHFunc(void *opaque); |
62 | + options={ | 82 | typedef bool AioPollFn(void *opaque); |
63 | + 'driver': 'file', | 83 | typedef void IOHandler(void *opaque); |
64 | + 'filename': filepath, | 84 | |
65 | + 'size': 0, | 85 | +struct Coroutine; |
66 | + })) | 86 | struct ThreadPool; |
67 | + vm.run_job('job1') | 87 | struct LinuxAioState; |
68 | + log(vm.command('blockdev-add', driver='file', | 88 | |
69 | + node_name=nodename, filename=filepath)) | 89 | @@ -XXX,XX +XXX,XX @@ struct AioContext { |
70 | + log(vm.command('blockdev-create', job_id='job2', | 90 | bool notified; |
71 | + options={ | 91 | EventNotifier notifier; |
72 | + 'driver': iotests.imgfmt, | 92 | |
73 | + 'file': nodename, | 93 | + QSLIST_HEAD(, Coroutine) scheduled_coroutines; |
74 | + 'size': size, | 94 | + QEMUBH *co_schedule_bh; |
75 | + })) | 95 | + |
76 | + vm.run_job('job2') | 96 | /* Thread pool for performing work and receiving completion callbacks. |
77 | + log(vm.command('blockdev-add', driver=iotests.imgfmt, | 97 | * Has its own locking. |
78 | + node_name=name, | 98 | */ |
79 | + file=nodename)) | 99 | @@ -XXX,XX +XXX,XX @@ static inline bool aio_node_check(AioContext *ctx, bool is_external) |
80 | + | 100 | } |
81 | + log('--- Preparing images & VM ---\n') | 101 | |
82 | + vm.add_object('iothread,id=iothread0') | 102 | /** |
83 | + vm.add_object('iothread,id=iothread1') | 103 | + * aio_co_schedule: |
84 | + vm.add_device('virtio-scsi-pci,id=scsi0,iothread=iothread0') | 104 | + * @ctx: the aio context |
85 | + vm.add_device('virtio-scsi-pci,id=scsi1,iothread=iothread1') | 105 | + * @co: the coroutine |
86 | + iotests.qemu_img_create('-f', iotests.imgfmt, img0_path, str(size)) | 106 | + * |
87 | + iotests.qemu_img_create('-f', iotests.imgfmt, img1_path, str(size)) | 107 | + * Start a coroutine on a remote AioContext. |
88 | + vm.add_drive(img0_path, interface='none') | 108 | + * |
89 | + vm.add_device('scsi-hd,id=device0,drive=drive0,bus=scsi0.0') | 109 | + * The coroutine must not be entered by anyone else while aio_co_schedule() |
90 | + vm.add_drive(img1_path, interface='none') | 110 | + * is active. In addition the coroutine must have yielded unless ctx |
91 | + vm.add_device('scsi-hd,id=device1,drive=drive1,bus=scsi1.0') | 111 | + * is the context in which the coroutine is running (i.e. the value of |
92 | + | 112 | + * qemu_get_current_aio_context() from the coroutine itself). |
93 | + log('--- Starting VM ---\n') | 113 | + */ |
94 | + vm.launch() | 114 | +void aio_co_schedule(AioContext *ctx, struct Coroutine *co); |
95 | + | 115 | + |
96 | + log('--- Create Targets & Full Backups ---\n') | 116 | +/** |
97 | + create_target(img0_full_path, 'img0-full', size) | 117 | + * aio_co_wake: |
98 | + create_target(img1_full_path, 'img1-full', size) | 118 | + * @co: the coroutine |
99 | + ret = vm.qmp_log('transaction', indent=2, actions=[ | 119 | + * |
100 | + { 'type': 'block-dirty-bitmap-add', | 120 | + * Restart a coroutine on the AioContext where it was running last, thus |
101 | + 'data': { 'node': 'drive0', 'name': 'bitmap0' }}, | 121 | + * preventing coroutines from jumping from one context to another when they |
102 | + { 'type': 'block-dirty-bitmap-add', | 122 | + * go to sleep. |
103 | + 'data': { 'node': 'drive1', 'name': 'bitmap1' }}, | 123 | + * |
104 | + { 'type': 'blockdev-backup', | 124 | + * aio_co_wake may be executed either in coroutine or non-coroutine |
105 | + 'data': { 'device': 'drive0', | 125 | + * context. The coroutine must not be entered by anyone else while |
106 | + 'target': 'img0-full', | 126 | + * aio_co_wake() is active. |
107 | + 'sync': 'full', | 127 | + */ |
108 | + 'job-id': 'j0' }}, | 128 | +void aio_co_wake(struct Coroutine *co); |
109 | + { 'type': 'blockdev-backup', | 129 | + |
110 | + 'data': { 'device': 'drive1', | 130 | +/** |
111 | + 'target': 'img1-full', | 131 | * Return the AioContext whose event loop runs in the current thread. |
112 | + 'sync': 'full', | 132 | * |
113 | + 'job-id': 'j1' }} | 133 | * If called from an IOThread this will be the IOThread's AioContext. If |
114 | + ]) | 134 | diff --git a/include/qemu/coroutine_int.h b/include/qemu/coroutine_int.h |
115 | + if "error" in ret: | 135 | index XXXXXXX..XXXXXXX 100644 |
116 | + raise Exception(ret['error']['desc']) | 136 | --- a/include/qemu/coroutine_int.h |
117 | + vm.run_job('j0', auto_dismiss=True) | 137 | +++ b/include/qemu/coroutine_int.h |
118 | + vm.run_job('j1', auto_dismiss=True) | 138 | @@ -XXX,XX +XXX,XX @@ struct Coroutine { |
119 | + | 139 | CoroutineEntry *entry; |
120 | + log('\n--- Create Targets & Incremental Backups ---\n') | 140 | void *entry_arg; |
121 | + create_target(img0_incr_path, 'img0-incr', size) | 141 | Coroutine *caller; |
122 | + create_target(img1_incr_path, 'img1-incr', size) | 142 | + |
123 | + ret = vm.qmp_log('transaction', indent=2, actions=[ | 143 | + /* Only used when the coroutine has terminated. */ |
124 | + { 'type': 'blockdev-backup', | 144 | QSLIST_ENTRY(Coroutine) pool_next; |
125 | + 'data': { 'device': 'drive0', | 145 | + |
126 | + 'target': 'img0-incr', | 146 | size_t locks_held; |
127 | + 'sync': 'incremental', | 147 | |
128 | + 'bitmap': 'bitmap0', | 148 | - /* Coroutines that should be woken up when we yield or terminate */ |
129 | + 'job-id': 'j2' }}, | 149 | + /* Coroutines that should be woken up when we yield or terminate. |
130 | + { 'type': 'blockdev-backup', | 150 | + * Only used when the coroutine is running. |
131 | + 'data': { 'device': 'drive1', | 151 | + */ |
132 | + 'target': 'img1-incr', | 152 | QSIMPLEQ_HEAD(, Coroutine) co_queue_wakeup; |
133 | + 'sync': 'incremental', | 153 | + |
134 | + 'bitmap': 'bitmap1', | 154 | + /* Only used when the coroutine has yielded. */ |
135 | + 'job-id': 'j3' }} | 155 | + AioContext *ctx; |
136 | + ]) | 156 | QSIMPLEQ_ENTRY(Coroutine) co_queue_next; |
137 | + if "error" in ret: | 157 | + QSLIST_ENTRY(Coroutine) co_scheduled_next; |
138 | + raise Exception(ret['error']['desc']) | 158 | }; |
139 | + vm.run_job('j2', auto_dismiss=True) | 159 | |
140 | + vm.run_job('j3', auto_dismiss=True) | 160 | Coroutine *qemu_coroutine_new(void); |
141 | + | 161 | diff --git a/tests/iothread.h b/tests/iothread.h |
142 | + log('\n--- Done ---') | ||
143 | + vm.shutdown() | ||
144 | diff --git a/tests/qemu-iotests/256.out b/tests/qemu-iotests/256.out | ||
145 | new file mode 100644 | 162 | new file mode 100644 |
146 | index XXXXXXX..XXXXXXX | 163 | index XXXXXXX..XXXXXXX |
147 | --- /dev/null | 164 | --- /dev/null |
148 | +++ b/tests/qemu-iotests/256.out | 165 | +++ b/tests/iothread.h |
149 | @@ -XXX,XX +XXX,XX @@ | 166 | @@ -XXX,XX +XXX,XX @@ |
150 | +--- Preparing images & VM --- | 167 | +/* |
151 | + | 168 | + * Event loop thread implementation for unit tests |
152 | +--- Starting VM --- | 169 | + * |
153 | + | 170 | + * Copyright Red Hat Inc., 2013, 2016 |
154 | +--- Create Targets & Full Backups --- | 171 | + * |
155 | + | 172 | + * Authors: |
156 | +{} | 173 | + * Stefan Hajnoczi <stefanha@redhat.com> |
157 | +{"execute": "job-dismiss", "arguments": {"id": "job1"}} | 174 | + * Paolo Bonzini <pbonzini@redhat.com> |
158 | +{"return": {}} | 175 | + * |
159 | +{} | 176 | + * This work is licensed under the terms of the GNU GPL, version 2 or later. |
160 | +{} | 177 | + * See the COPYING file in the top-level directory. |
161 | +{"execute": "job-dismiss", "arguments": {"id": "job2"}} | 178 | + */ |
162 | +{"return": {}} | 179 | +#ifndef TEST_IOTHREAD_H |
163 | +{} | 180 | +#define TEST_IOTHREAD_H |
164 | +{} | 181 | + |
165 | +{"execute": "job-dismiss", "arguments": {"id": "job1"}} | 182 | +#include "block/aio.h" |
166 | +{"return": {}} | 183 | +#include "qemu/thread.h" |
167 | +{} | 184 | + |
168 | +{} | 185 | +typedef struct IOThread IOThread; |
169 | +{"execute": "job-dismiss", "arguments": {"id": "job2"}} | 186 | + |
170 | +{"return": {}} | 187 | +IOThread *iothread_new(void); |
171 | +{} | 188 | +void iothread_join(IOThread *iothread); |
172 | +{ | 189 | +AioContext *iothread_get_aio_context(IOThread *iothread); |
173 | + "execute": "transaction", | 190 | + |
174 | + "arguments": { | 191 | +#endif |
175 | + "actions": [ | 192 | diff --git a/tests/iothread.c b/tests/iothread.c |
176 | + { | 193 | new file mode 100644 |
177 | + "data": { | 194 | index XXXXXXX..XXXXXXX |
178 | + "name": "bitmap0", | 195 | --- /dev/null |
179 | + "node": "drive0" | 196 | +++ b/tests/iothread.c |
180 | + }, | 197 | @@ -XXX,XX +XXX,XX @@ |
181 | + "type": "block-dirty-bitmap-add" | 198 | +/* |
182 | + }, | 199 | + * Event loop thread implementation for unit tests |
183 | + { | 200 | + * |
184 | + "data": { | 201 | + * Copyright Red Hat Inc., 2013, 2016 |
185 | + "name": "bitmap1", | 202 | + * |
186 | + "node": "drive1" | 203 | + * Authors: |
187 | + }, | 204 | + * Stefan Hajnoczi <stefanha@redhat.com> |
188 | + "type": "block-dirty-bitmap-add" | 205 | + * Paolo Bonzini <pbonzini@redhat.com> |
189 | + }, | 206 | + * |
190 | + { | 207 | + * This work is licensed under the terms of the GNU GPL, version 2 or later. |
191 | + "data": { | 208 | + * See the COPYING file in the top-level directory. |
192 | + "device": "drive0", | 209 | + * |
193 | + "job-id": "j0", | 210 | + */ |
194 | + "sync": "full", | 211 | + |
195 | + "target": "img0-full" | 212 | +#include "qemu/osdep.h" |
196 | + }, | 213 | +#include "qapi/error.h" |
197 | + "type": "blockdev-backup" | 214 | +#include "block/aio.h" |
198 | + }, | 215 | +#include "qemu/main-loop.h" |
199 | + { | 216 | +#include "qemu/rcu.h" |
200 | + "data": { | 217 | +#include "iothread.h" |
201 | + "device": "drive1", | 218 | + |
202 | + "job-id": "j1", | 219 | +struct IOThread { |
203 | + "sync": "full", | 220 | + AioContext *ctx; |
204 | + "target": "img1-full" | 221 | + |
205 | + }, | 222 | + QemuThread thread; |
206 | + "type": "blockdev-backup" | 223 | + QemuMutex init_done_lock; |
207 | + } | 224 | + QemuCond init_done_cond; /* is thread initialization done? */ |
208 | + ] | 225 | + bool stopping; |
209 | + } | 226 | +}; |
210 | +} | 227 | + |
211 | +{ | 228 | +static __thread IOThread *my_iothread; |
212 | + "return": {} | 229 | + |
213 | +} | 230 | +AioContext *qemu_get_current_aio_context(void) |
214 | +{"data": {"device": "j0", "len": 67108864, "offset": 67108864, "speed": 0, "type": "backup"}, "event": "BLOCK_JOB_COMPLETED", "timestamp": {"microseconds": "USECS", "seconds": "SECS"}} | 231 | +{ |
215 | +{"data": {"device": "j1", "len": 67108864, "offset": 67108864, "speed": 0, "type": "backup"}, "event": "BLOCK_JOB_COMPLETED", "timestamp": {"microseconds": "USECS", "seconds": "SECS"}} | 232 | + return my_iothread ? my_iothread->ctx : qemu_get_aio_context(); |
216 | + | 233 | +} |
217 | +--- Create Targets & Incremental Backups --- | 234 | + |
218 | + | 235 | +static void *iothread_run(void *opaque) |
219 | +{} | 236 | +{ |
220 | +{"execute": "job-dismiss", "arguments": {"id": "job1"}} | 237 | + IOThread *iothread = opaque; |
221 | +{"return": {}} | 238 | + |
222 | +{} | 239 | + rcu_register_thread(); |
223 | +{} | 240 | + |
224 | +{"execute": "job-dismiss", "arguments": {"id": "job2"}} | 241 | + my_iothread = iothread; |
225 | +{"return": {}} | 242 | + qemu_mutex_lock(&iothread->init_done_lock); |
226 | +{} | 243 | + iothread->ctx = aio_context_new(&error_abort); |
227 | +{} | 244 | + qemu_cond_signal(&iothread->init_done_cond); |
228 | +{"execute": "job-dismiss", "arguments": {"id": "job1"}} | 245 | + qemu_mutex_unlock(&iothread->init_done_lock); |
229 | +{"return": {}} | 246 | + |
230 | +{} | 247 | + while (!atomic_read(&iothread->stopping)) { |
231 | +{} | 248 | + aio_poll(iothread->ctx, true); |
232 | +{"execute": "job-dismiss", "arguments": {"id": "job2"}} | 249 | + } |
233 | +{"return": {}} | 250 | + |
234 | +{} | 251 | + rcu_unregister_thread(); |
235 | +{ | 252 | + return NULL; |
236 | + "execute": "transaction", | 253 | +} |
237 | + "arguments": { | 254 | + |
238 | + "actions": [ | 255 | +void iothread_join(IOThread *iothread) |
239 | + { | 256 | +{ |
240 | + "data": { | 257 | + iothread->stopping = true; |
241 | + "bitmap": "bitmap0", | 258 | + aio_notify(iothread->ctx); |
242 | + "device": "drive0", | 259 | + qemu_thread_join(&iothread->thread); |
243 | + "job-id": "j2", | 260 | + qemu_cond_destroy(&iothread->init_done_cond); |
244 | + "sync": "incremental", | 261 | + qemu_mutex_destroy(&iothread->init_done_lock); |
245 | + "target": "img0-incr" | 262 | + aio_context_unref(iothread->ctx); |
246 | + }, | 263 | + g_free(iothread); |
247 | + "type": "blockdev-backup" | 264 | +} |
248 | + }, | 265 | + |
249 | + { | 266 | +IOThread *iothread_new(void) |
250 | + "data": { | 267 | +{ |
251 | + "bitmap": "bitmap1", | 268 | + IOThread *iothread = g_new0(IOThread, 1); |
252 | + "device": "drive1", | 269 | + |
253 | + "job-id": "j3", | 270 | + qemu_mutex_init(&iothread->init_done_lock); |
254 | + "sync": "incremental", | 271 | + qemu_cond_init(&iothread->init_done_cond); |
255 | + "target": "img1-incr" | 272 | + qemu_thread_create(&iothread->thread, NULL, iothread_run, |
256 | + }, | 273 | + iothread, QEMU_THREAD_JOINABLE); |
257 | + "type": "blockdev-backup" | 274 | + |
258 | + } | 275 | + /* Wait for initialization to complete */ |
259 | + ] | 276 | + qemu_mutex_lock(&iothread->init_done_lock); |
260 | + } | 277 | + while (iothread->ctx == NULL) { |
261 | +} | 278 | + qemu_cond_wait(&iothread->init_done_cond, |
262 | +{ | 279 | + &iothread->init_done_lock); |
263 | + "return": {} | 280 | + } |
264 | +} | 281 | + qemu_mutex_unlock(&iothread->init_done_lock); |
265 | +{"data": {"device": "j2", "len": 67108864, "offset": 67108864, "speed": 0, "type": "backup"}, "event": "BLOCK_JOB_COMPLETED", "timestamp": {"microseconds": "USECS", "seconds": "SECS"}} | 282 | + return iothread; |
266 | +{"data": {"device": "j3", "len": 67108864, "offset": 67108864, "speed": 0, "type": "backup"}, "event": "BLOCK_JOB_COMPLETED", "timestamp": {"microseconds": "USECS", "seconds": "SECS"}} | 283 | +} |
267 | + | 284 | + |
268 | +--- Done --- | 285 | +AioContext *iothread_get_aio_context(IOThread *iothread) |
269 | diff --git a/tests/qemu-iotests/group b/tests/qemu-iotests/group | 286 | +{ |
287 | + return iothread->ctx; | ||
288 | +} | ||
289 | diff --git a/tests/test-aio-multithread.c b/tests/test-aio-multithread.c | ||
290 | new file mode 100644 | ||
291 | index XXXXXXX..XXXXXXX | ||
292 | --- /dev/null | ||
293 | +++ b/tests/test-aio-multithread.c | ||
294 | @@ -XXX,XX +XXX,XX @@ | ||
295 | +/* | ||
296 | + * AioContext multithreading tests | ||
297 | + * | ||
298 | + * Copyright Red Hat, Inc. 2016 | ||
299 | + * | ||
300 | + * Authors: | ||
301 | + * Paolo Bonzini <pbonzini@redhat.com> | ||
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. | ||
305 | + */ | ||
306 | + | ||
307 | +#include "qemu/osdep.h" | ||
308 | +#include <glib.h> | ||
309 | +#include "block/aio.h" | ||
310 | +#include "qapi/error.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 | ||
270 | index XXXXXXX..XXXXXXX 100644 | 509 | index XXXXXXX..XXXXXXX 100644 |
271 | --- a/tests/qemu-iotests/group | 510 | --- a/util/async.c |
272 | +++ b/tests/qemu-iotests/group | 511 | +++ b/util/async.c |
273 | @@ -XXX,XX +XXX,XX @@ | 512 | @@ -XXX,XX +XXX,XX @@ |
274 | 253 rw auto quick | 513 | #include "qemu/main-loop.h" |
275 | 254 rw auto backing quick | 514 | #include "qemu/atomic.h" |
276 | 255 rw auto quick | 515 | #include "block/raw-aio.h" |
277 | +256 rw auto quick | 516 | +#include "qemu/coroutine_int.h" |
517 | +#include "trace.h" | ||
518 | |||
519 | /***********************************************************/ | ||
520 | /* bottom halves (can be seen as timers which expire ASAP) */ | ||
521 | @@ -XXX,XX +XXX,XX @@ aio_ctx_finalize(GSource *source) | ||
522 | } | ||
523 | #endif | ||
524 | |||
525 | + assert(QSLIST_EMPTY(&ctx->scheduled_coroutines)); | ||
526 | + qemu_bh_delete(ctx->co_schedule_bh); | ||
527 | + | ||
528 | qemu_lockcnt_lock(&ctx->list_lock); | ||
529 | assert(!qemu_lockcnt_count(&ctx->list_lock)); | ||
530 | while (ctx->first_bh) { | ||
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); | ||
602 | + } else { | ||
603 | + aio_context_acquire(ctx); | ||
604 | + qemu_coroutine_enter(co); | ||
605 | + aio_context_release(ctx); | ||
606 | + } | ||
607 | +} | ||
608 | + | ||
609 | void aio_context_ref(AioContext *ctx) | ||
610 | { | ||
611 | g_source_ref(&ctx->source); | ||
612 | diff --git a/util/qemu-coroutine.c b/util/qemu-coroutine.c | ||
613 | index XXXXXXX..XXXXXXX 100644 | ||
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. | ||
632 | + */ | ||
633 | + smp_wmb(); | ||
634 | + | ||
635 | ret = qemu_coroutine_switch(self, co, COROUTINE_ENTER); | ||
636 | |||
637 | qemu_co_queue_run_restart(co); | ||
638 | diff --git a/util/trace-events b/util/trace-events | ||
639 | index XXXXXXX..XXXXXXX 100644 | ||
640 | --- a/util/trace-events | ||
641 | +++ b/util/trace-events | ||
642 | @@ -XXX,XX +XXX,XX @@ run_poll_handlers_end(void *ctx, bool progress) "ctx %p progress %d" | ||
643 | poll_shrink(void *ctx, int64_t old, int64_t new) "ctx %p old %"PRId64" new %"PRId64 | ||
644 | poll_grow(void *ctx, int64_t old, int64_t new) "ctx %p old %"PRId64" new %"PRId64 | ||
645 | |||
646 | +# util/async.c | ||
647 | +aio_co_schedule(void *ctx, void *co) "ctx %p co %p" | ||
648 | +aio_co_schedule_bh_cb(void *ctx, void *co) "ctx %p co %p" | ||
649 | + | ||
650 | # util/thread-pool.c | ||
651 | thread_pool_submit(void *pool, void *req, void *opaque) "pool %p req %p opaque %p" | ||
652 | thread_pool_complete(void *pool, void *req, void *opaque, int ret) "pool %p req %p opaque %p ret %d" | ||
278 | -- | 653 | -- |
279 | 2.21.0 | 654 | 2.9.3 |
280 | 655 | ||
281 | 656 | diff view generated by jsdifflib |
1 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 1 | From: Paolo Bonzini <pbonzini@redhat.com> |
---|---|---|---|
2 | Reviewed-by: Eric Blake <eblake@redhat.com> | 2 | |
3 | Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> | 3 | qcow2_create2 calls this. Do not run a nested event loop, as that |
4 | Message-id: 20190507203508.18026-6-mreitz@redhat.com | 4 | breaks when aio_co_wake tries to queue the coroutine on the co_queue_wakeup |
5 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 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> | ||
6 | --- | 12 | --- |
7 | qapi/block-core.json | 5 ++++- | 13 | block/block-backend.c | 12 ++++++++---- |
8 | block/blkdebug.c | 8 ++++++++ | 14 | 1 file changed, 8 insertions(+), 4 deletions(-) |
9 | 2 files changed, 12 insertions(+), 1 deletion(-) | ||
10 | 15 | ||
11 | diff --git a/qapi/block-core.json b/qapi/block-core.json | 16 | diff --git a/block/block-backend.c b/block/block-backend.c |
12 | index XXXXXXX..XXXXXXX 100644 | 17 | index XXXXXXX..XXXXXXX 100644 |
13 | --- a/qapi/block-core.json | 18 | --- a/block/block-backend.c |
14 | +++ b/qapi/block-core.json | 19 | +++ b/block/block-backend.c |
15 | @@ -XXX,XX +XXX,XX @@ | 20 | @@ -XXX,XX +XXX,XX @@ static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf, |
16 | # | ||
17 | # @flush: .bdrv_co_flush_to_disk() | ||
18 | # | ||
19 | +# @block-status: .bdrv_co_block_status() | ||
20 | +# | ||
21 | # Since: 4.1 | ||
22 | ## | ||
23 | { 'enum': 'BlkdebugIOType', 'prefix': 'BLKDEBUG_IO_TYPE', | ||
24 | - 'data': [ 'read', 'write', 'write-zeroes', 'discard', 'flush' ] } | ||
25 | + 'data': [ 'read', 'write', 'write-zeroes', 'discard', 'flush', | ||
26 | + 'block-status' ] } | ||
27 | |||
28 | ## | ||
29 | # @BlkdebugInjectErrorOptions: | ||
30 | diff --git a/block/blkdebug.c b/block/blkdebug.c | ||
31 | index XXXXXXX..XXXXXXX 100644 | ||
32 | --- a/block/blkdebug.c | ||
33 | +++ b/block/blkdebug.c | ||
34 | @@ -XXX,XX +XXX,XX @@ static int coroutine_fn blkdebug_co_block_status(BlockDriverState *bs, | ||
35 | int64_t *map, | ||
36 | BlockDriverState **file) | ||
37 | { | 21 | { |
38 | + int err; | 22 | QEMUIOVector qiov; |
39 | + | 23 | struct iovec iov; |
40 | assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment)); | 24 | - Coroutine *co; |
41 | + | 25 | BlkRwCo rwco; |
42 | + err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_BLOCK_STATUS); | 26 | |
43 | + if (err) { | 27 | iov = (struct iovec) { |
44 | + return err; | 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); | ||
45 | + } | 42 | + } |
46 | + | 43 | |
47 | return bdrv_co_block_status_from_file(bs, want_zero, offset, bytes, | 44 | return rwco.ret; |
48 | pnum, map, file); | ||
49 | } | 45 | } |
50 | -- | 46 | -- |
51 | 2.21.0 | 47 | 2.9.3 |
52 | 48 | ||
53 | 49 | diff view generated by jsdifflib |
1 | We do not support this combination (yet), so this should yield an error | 1 | From: Paolo Bonzini <pbonzini@redhat.com> |
---|---|---|---|
2 | message. | ||
3 | 2 | ||
4 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 3 | Once the thread pool starts using aio_co_wake, it will also need |
5 | Tested-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> | 4 | qemu_get_current_aio_context(). Make test-thread-pool create |
6 | Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> | 5 | an AioContext with qemu_init_main_loop, so that stubs/iothread.c |
7 | Message-id: 20190507203508.18026-8-mreitz@redhat.com | 6 | and tests/iothread.c can provide the rest. |
8 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 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> | ||
9 | --- | 13 | --- |
10 | tests/qemu-iotests/082 | 1 + | 14 | tests/test-thread-pool.c | 12 +++--------- |
11 | tests/qemu-iotests/082.out | 3 +++ | 15 | 1 file changed, 3 insertions(+), 9 deletions(-) |
12 | 2 files changed, 4 insertions(+) | ||
13 | 16 | ||
14 | diff --git a/tests/qemu-iotests/082 b/tests/qemu-iotests/082 | 17 | diff --git a/tests/test-thread-pool.c b/tests/test-thread-pool.c |
15 | index XXXXXXX..XXXXXXX 100755 | ||
16 | --- a/tests/qemu-iotests/082 | ||
17 | +++ b/tests/qemu-iotests/082 | ||
18 | @@ -XXX,XX +XXX,XX @@ echo === convert: -C and other options === | ||
19 | run_qemu_img convert -C -S 4k -O $IMGFMT "$TEST_IMG" "$TEST_IMG".target | ||
20 | run_qemu_img convert -C -S 8k -O $IMGFMT "$TEST_IMG" "$TEST_IMG".target | ||
21 | run_qemu_img convert -C -c -O $IMGFMT "$TEST_IMG" "$TEST_IMG".target | ||
22 | +run_qemu_img convert -C --salvage -O $IMGFMT "$TEST_IMG" "$TEST_IMG".target | ||
23 | |||
24 | echo | ||
25 | echo === amend: Options specified more than once === | ||
26 | diff --git a/tests/qemu-iotests/082.out b/tests/qemu-iotests/082.out | ||
27 | index XXXXXXX..XXXXXXX 100644 | 18 | index XXXXXXX..XXXXXXX 100644 |
28 | --- a/tests/qemu-iotests/082.out | 19 | --- a/tests/test-thread-pool.c |
29 | +++ b/tests/qemu-iotests/082.out | 20 | +++ b/tests/test-thread-pool.c |
30 | @@ -XXX,XX +XXX,XX @@ qemu-img: Cannot enable copy offloading when -S is used | 21 | @@ -XXX,XX +XXX,XX @@ |
31 | Testing: convert -C -c -O qcow2 TEST_DIR/t.qcow2 TEST_DIR/t.qcow2.target | 22 | #include "qapi/error.h" |
32 | qemu-img: Cannot enable copy offloading when -c is used | 23 | #include "qemu/timer.h" |
33 | 24 | #include "qemu/error-report.h" | |
34 | +Testing: convert -C --salvage -O qcow2 TEST_DIR/t.qcow2 TEST_DIR/t.qcow2.target | 25 | +#include "qemu/main-loop.h" |
35 | +qemu-img: Cannot use copy offloading in salvaging mode | 26 | |
36 | + | 27 | static AioContext *ctx; |
37 | === amend: Options specified more than once === | 28 | static ThreadPool *pool; |
38 | 29 | @@ -XXX,XX +XXX,XX @@ static void test_cancel_async(void) | |
39 | Testing: amend -f foo -f qcow2 -o lazy_refcounts=on TEST_DIR/t.qcow2 | 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 | } | ||
40 | -- | 54 | -- |
41 | 2.21.0 | 55 | 2.9.3 |
42 | 56 | ||
43 | 57 | diff view generated by jsdifflib |
1 | img_rebase() can leak a QDict in two occasions. Fix it. | 1 | From: Paolo Bonzini <pbonzini@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | Coverity: CID 1401416 | 3 | This is in preparation for making qio_channel_yield work on |
4 | Fixes: d16699b64671466b42079c45b89127aeea1ca565 | 4 | AioContexts other than the main one. |
5 | Fixes: 330c72957196e0ae382abcaa97ebf4eb9bc8574f | 5 | |
6 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 6 | Reviewed-by: Daniel P. Berrange <berrange@redhat.com> |
7 | Message-id: 20190528195338.12376-1-mreitz@redhat.com | 7 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> |
8 | Reviewed-by: John Snow <jsnow@redhat.com> | 8 | Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> |
9 | Signed-off-by: Max Reitz <mreitz@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> | ||
10 | --- | 12 | --- |
11 | qemu-img.c | 3 +++ | 13 | include/io/channel.h | 25 +++++++++++++++++++++++++ |
12 | 1 file changed, 3 insertions(+) | 14 | io/channel-command.c | 13 +++++++++++++ |
13 | 15 | io/channel-file.c | 11 +++++++++++ | |
14 | diff --git a/qemu-img.c b/qemu-img.c | 16 | io/channel-socket.c | 16 +++++++++++----- |
15 | index XXXXXXX..XXXXXXX 100644 | 17 | io/channel-tls.c | 12 ++++++++++++ |
16 | --- a/qemu-img.c | 18 | io/channel-watch.c | 6 ++++++ |
17 | +++ b/qemu-img.c | 19 | io/channel.c | 11 +++++++++++ |
18 | @@ -XXX,XX +XXX,XX @@ static int img_rebase(int argc, char **argv) | 20 | 7 files changed, 89 insertions(+), 5 deletions(-) |
19 | out_baseimg, | 21 | |
20 | &local_err); | 22 | diff --git a/include/io/channel.h b/include/io/channel.h |
21 | if (local_err) { | 23 | index XXXXXXX..XXXXXXX 100644 |
22 | + qobject_unref(options); | 24 | --- a/include/io/channel.h |
23 | error_reportf_err(local_err, | 25 | +++ b/include/io/channel.h |
24 | "Could not resolve backing filename: "); | 26 | @@ -XXX,XX +XXX,XX @@ |
25 | ret = -1; | 27 | |
26 | @@ -XXX,XX +XXX,XX @@ static int img_rebase(int argc, char **argv) | 28 | #include "qemu-common.h" |
27 | */ | 29 | #include "qom/object.h" |
28 | prefix_chain_bs = bdrv_find_backing_image(bs, out_real_path); | 30 | +#include "block/aio.h" |
29 | if (prefix_chain_bs) { | 31 | |
30 | + qobject_unref(options); | 32 | #define TYPE_QIO_CHANNEL "qio-channel" |
31 | g_free(out_real_path); | 33 | #define QIO_CHANNEL(obj) \ |
32 | + | 34 | @@ -XXX,XX +XXX,XX @@ struct QIOChannelClass { |
33 | blk_new_backing = blk_new(qemu_get_aio_context(), | 35 | off_t offset, |
34 | BLK_PERM_CONSISTENT_READ, | 36 | int whence, |
35 | BLK_PERM_ALL); | 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, | ||
36 | -- | 240 | -- |
37 | 2.21.0 | 241 | 2.9.3 |
38 | 242 | ||
39 | 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 |
1 | Together with @iotypes and @sector, this can be used to trap e.g. the | 1 | From: Paolo Bonzini <pbonzini@redhat.com> |
---|---|---|---|
2 | first read or write access to a certain sector without having to know | ||
3 | what happens internally in the block layer, i.e. which "real" events | ||
4 | happen right before such an access. | ||
5 | 2 | ||
6 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 3 | In the client, read the reply headers from a coroutine, switching the |
7 | Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> | 4 | read side between the "read header" coroutine and the I/O coroutine that |
8 | Message-id: 20190507203508.18026-5-mreitz@redhat.com | 5 | reads the body of the reply. |
9 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 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> | ||
10 | --- | 17 | --- |
11 | qapi/block-core.json | 4 +++- | 18 | block/nbd-client.h | 2 +- |
12 | block/blkdebug.c | 2 ++ | 19 | block/nbd-client.c | 117 ++++++++++++++++++++++++----------------------------- |
13 | 2 files changed, 5 insertions(+), 1 deletion(-) | 20 | nbd/client.c | 2 +- |
21 | nbd/common.c | 9 +---- | ||
22 | nbd/server.c | 94 +++++++++++++----------------------------- | ||
23 | 5 files changed, 83 insertions(+), 141 deletions(-) | ||
14 | 24 | ||
15 | diff --git a/qapi/block-core.json b/qapi/block-core.json | 25 | diff --git a/block/nbd-client.h b/block/nbd-client.h |
16 | index XXXXXXX..XXXXXXX 100644 | 26 | index XXXXXXX..XXXXXXX 100644 |
17 | --- a/qapi/block-core.json | 27 | --- a/block/nbd-client.h |
18 | +++ b/qapi/block-core.json | 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 | ||
19 | @@ -XXX,XX +XXX,XX @@ | 42 | @@ -XXX,XX +XXX,XX @@ |
20 | # | 43 | #define HANDLE_TO_INDEX(bs, handle) ((handle) ^ ((uint64_t)(intptr_t)bs)) |
21 | # @cluster_alloc_space: an allocation of file space for a cluster (since 4.1) | 44 | #define INDEX_TO_HANDLE(bs, index) ((index) ^ ((uint64_t)(intptr_t)bs)) |
22 | # | 45 | |
23 | +# @none: triggers once at creation of the blkdebug node (since 4.1) | 46 | -static void nbd_recv_coroutines_enter_all(NBDClientSession *s) |
24 | +# | 47 | +static void nbd_recv_coroutines_enter_all(BlockDriverState *bs) |
25 | # Since: 2.9 | 48 | { |
26 | ## | 49 | + NBDClientSession *s = nbd_get_client_session(bs); |
27 | { 'enum': 'BlkdebugEvent', 'prefix': 'BLKDBG', | 50 | int i; |
28 | @@ -XXX,XX +XXX,XX @@ | 51 | |
29 | 'pwritev_rmw_tail', 'pwritev_rmw_after_tail', 'pwritev', | 52 | for (i = 0; i < MAX_NBD_REQUESTS; i++) { |
30 | 'pwritev_zero', 'pwritev_done', 'empty_image_prepare', | 53 | @@ -XXX,XX +XXX,XX @@ static void nbd_recv_coroutines_enter_all(NBDClientSession *s) |
31 | 'l1_shrink_write_table', 'l1_shrink_free_l2_clusters', | 54 | qemu_coroutine_enter(s->recv_coroutine[i]); |
32 | - 'cor_write', 'cluster_alloc_space'] } | 55 | } |
33 | + 'cor_write', 'cluster_alloc_space', 'none'] } | 56 | } |
34 | 57 | + BDRV_POLL_WHILE(bs, s->read_reply_co); | |
35 | ## | 58 | } |
36 | # @BlkdebugIOType: | 59 | |
37 | diff --git a/block/blkdebug.c b/block/blkdebug.c | 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 | ||
38 | index XXXXXXX..XXXXXXX 100644 | 291 | index XXXXXXX..XXXXXXX 100644 |
39 | --- a/block/blkdebug.c | 292 | --- a/nbd/client.c |
40 | +++ b/block/blkdebug.c | 293 | +++ b/nbd/client.c |
41 | @@ -XXX,XX +XXX,XX @@ static int blkdebug_open(BlockDriverState *bs, QDict *options, int flags, | 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) | ||
42 | goto out; | 523 | goto out; |
43 | } | 524 | } |
44 | 525 | qemu_co_mutex_init(&client->send_lock); | |
45 | + bdrv_debug_event(bs, BLKDBG_NONE); | 526 | - nbd_set_handlers(client); |
527 | |||
528 | if (exp) { | ||
529 | QTAILQ_INSERT_TAIL(&exp->clients, client, next); | ||
530 | } | ||
46 | + | 531 | + |
47 | ret = 0; | 532 | + nbd_client_receive_next_request(client); |
533 | + | ||
48 | out: | 534 | out: |
49 | if (ret < 0) { | 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; | ||
50 | -- | 545 | -- |
51 | 2.21.0 | 546 | 2.9.3 |
52 | 547 | ||
53 | 548 | diff view generated by jsdifflib |
1 | From: Andrey Shinkevich <andrey.shinkevich@virtuozzo.com> | 1 | From: Paolo Bonzini <pbonzini@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | The uninitialized memory allocated for the command FIFO of the | 3 | As a small step towards the introduction of multiqueue, we want |
4 | floppy controller during the VM hardware initialization incurs | 4 | coroutines to remain on the same AioContext that started them, |
5 | many unwanted reports by Valgrind when VM state is being saved. | 5 | unless they are moved explicitly with e.g. aio_co_schedule. This patch |
6 | That verbosity hardens a search for the real memory issues when | 6 | avoids that coroutines switch AioContext when they use a CoMutex. |
7 | the iotests run. Particularly, the patch eliminates 20 unnecessary | 7 | For now it does not make much of a difference, because the CoMutex |
8 | reports of the Valgrind tool in the iotest #169. | 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. | ||
9 | 10 | ||
10 | Signed-off-by: Andrey Shinkevich <andrey.shinkevich@virtuozzo.com> | 11 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> |
11 | Message-id: 1559154027-282547-1-git-send-email-andrey.shinkevich@virtuozzo.com | 12 | Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> |
12 | Reviewed-by: John Snow <jsnow@redhat.com> | 13 | Reviewed-by: Fam Zheng <famz@redhat.com> |
13 | Signed-off-by: Max Reitz <mreitz@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> | ||
14 | --- | 17 | --- |
15 | hw/block/fdc.c | 1 + | 18 | util/qemu-coroutine-lock.c | 5 ++--- |
16 | 1 file changed, 1 insertion(+) | 19 | util/trace-events | 1 - |
20 | 2 files changed, 2 insertions(+), 4 deletions(-) | ||
17 | 21 | ||
18 | diff --git a/hw/block/fdc.c b/hw/block/fdc.c | 22 | diff --git a/util/qemu-coroutine-lock.c b/util/qemu-coroutine-lock.c |
19 | index XXXXXXX..XXXXXXX 100644 | 23 | index XXXXXXX..XXXXXXX 100644 |
20 | --- a/hw/block/fdc.c | 24 | --- a/util/qemu-coroutine-lock.c |
21 | +++ b/hw/block/fdc.c | 25 | +++ b/util/qemu-coroutine-lock.c |
22 | @@ -XXX,XX +XXX,XX @@ static void fdctrl_realize_common(DeviceState *dev, FDCtrl *fdctrl, | 26 | @@ -XXX,XX +XXX,XX @@ |
23 | 27 | #include "qemu/coroutine.h" | |
24 | FLOPPY_DPRINTF("init controller\n"); | 28 | #include "qemu/coroutine_int.h" |
25 | fdctrl->fifo = qemu_memalign(512, FD_SECTOR_LEN); | 29 | #include "qemu/queue.h" |
26 | + memset(fdctrl->fifo, 0, FD_SECTOR_LEN); | 30 | +#include "block/aio.h" |
27 | fdctrl->fifo_size = 512; | 31 | #include "trace.h" |
28 | fdctrl->result_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, | 32 | |
29 | fdctrl_result_timer, fdctrl); | 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" | ||
30 | -- | 64 | -- |
31 | 2.21.0 | 65 | 2.9.3 |
32 | 66 | ||
33 | 67 | diff view generated by jsdifflib |
1 | This new error option allows users of blkdebug to inject errors only on | 1 | From: Paolo Bonzini <pbonzini@redhat.com> |
---|---|---|---|
2 | certain kinds of I/O operations. Users usually want to make a very | ||
3 | specific operation fail, not just any; but right now they simply hope | ||
4 | that the event that triggers the error injection is followed up with | ||
5 | that very operation. That may not be true, however, because the block | ||
6 | layer is changing (including blkdebug, which may increase the number of | ||
7 | types of I/O operations on which to inject errors). | ||
8 | 2 | ||
9 | The new option's default has been chosen to keep backwards | 3 | Keep the coroutine on the same AioContext. Without this change, |
10 | compatibility. | 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. | ||
11 | 7 | ||
12 | Note that similar to the internal representation, we could choose to | 8 | While doing the change, replace custom bottom half with aio_co_schedule. |
13 | expose this option as a list of I/O types. But there is no practical | ||
14 | use for this, because as described above, users usually know exactly | ||
15 | which kind of operation they want to make fail, so there is no need to | ||
16 | specify multiple I/O types at once. In addition, exposing this option | ||
17 | as a list would require non-trivial changes to qemu_opts_absorb_qdict(). | ||
18 | 9 | ||
19 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 10 | Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> |
20 | Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> | 11 | Reviewed-by: Fam Zheng <famz@redhat.com> |
21 | Message-id: 20190507203508.18026-4-mreitz@redhat.com | 12 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> |
22 | Signed-off-by: Max Reitz <mreitz@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> | ||
23 | --- | 16 | --- |
24 | qapi/block-core.json | 26 +++++++++++++++++++++++ | 17 | block/blkdebug.c | 9 +-------- |
25 | block/blkdebug.c | 50 ++++++++++++++++++++++++++++++++++++-------- | 18 | 1 file changed, 1 insertion(+), 8 deletions(-) |
26 | 2 files changed, 67 insertions(+), 9 deletions(-) | ||
27 | 19 | ||
28 | diff --git a/qapi/block-core.json b/qapi/block-core.json | ||
29 | index XXXXXXX..XXXXXXX 100644 | ||
30 | --- a/qapi/block-core.json | ||
31 | +++ b/qapi/block-core.json | ||
32 | @@ -XXX,XX +XXX,XX @@ | ||
33 | 'l1_shrink_write_table', 'l1_shrink_free_l2_clusters', | ||
34 | 'cor_write', 'cluster_alloc_space'] } | ||
35 | |||
36 | +## | ||
37 | +# @BlkdebugIOType: | ||
38 | +# | ||
39 | +# Kinds of I/O that blkdebug can inject errors in. | ||
40 | +# | ||
41 | +# @read: .bdrv_co_preadv() | ||
42 | +# | ||
43 | +# @write: .bdrv_co_pwritev() | ||
44 | +# | ||
45 | +# @write-zeroes: .bdrv_co_pwrite_zeroes() | ||
46 | +# | ||
47 | +# @discard: .bdrv_co_pdiscard() | ||
48 | +# | ||
49 | +# @flush: .bdrv_co_flush_to_disk() | ||
50 | +# | ||
51 | +# Since: 4.1 | ||
52 | +## | ||
53 | +{ 'enum': 'BlkdebugIOType', 'prefix': 'BLKDEBUG_IO_TYPE', | ||
54 | + 'data': [ 'read', 'write', 'write-zeroes', 'discard', 'flush' ] } | ||
55 | + | ||
56 | ## | ||
57 | # @BlkdebugInjectErrorOptions: | ||
58 | # | ||
59 | @@ -XXX,XX +XXX,XX @@ | ||
60 | # @state: the state identifier blkdebug needs to be in to | ||
61 | # actually trigger the event; defaults to "any" | ||
62 | # | ||
63 | +# @iotype: the type of I/O operations on which this error should | ||
64 | +# be injected; defaults to "all read, write, | ||
65 | +# write-zeroes, discard, and flush operations" | ||
66 | +# (since: 4.1) | ||
67 | +# | ||
68 | # @errno: error identifier (errno) to be returned; defaults to | ||
69 | # EIO | ||
70 | # | ||
71 | @@ -XXX,XX +XXX,XX @@ | ||
72 | { 'struct': 'BlkdebugInjectErrorOptions', | ||
73 | 'data': { 'event': 'BlkdebugEvent', | ||
74 | '*state': 'int', | ||
75 | + '*iotype': 'BlkdebugIOType', | ||
76 | '*errno': 'int', | ||
77 | '*sector': 'int', | ||
78 | '*once': 'bool', | ||
79 | diff --git a/block/blkdebug.c b/block/blkdebug.c | 20 | diff --git a/block/blkdebug.c b/block/blkdebug.c |
80 | index XXXXXXX..XXXXXXX 100644 | 21 | index XXXXXXX..XXXXXXX 100644 |
81 | --- a/block/blkdebug.c | 22 | --- a/block/blkdebug.c |
82 | +++ b/block/blkdebug.c | 23 | +++ b/block/blkdebug.c |
83 | @@ -XXX,XX +XXX,XX @@ typedef struct BlkdebugRule { | ||
84 | int state; | ||
85 | union { | ||
86 | struct { | ||
87 | + uint64_t iotype_mask; | ||
88 | int error; | ||
89 | int immediately; | ||
90 | int once; | ||
91 | @@ -XXX,XX +XXX,XX @@ typedef struct BlkdebugRule { | ||
92 | QSIMPLEQ_ENTRY(BlkdebugRule) active_next; | ||
93 | } BlkdebugRule; | ||
94 | |||
95 | +QEMU_BUILD_BUG_MSG(BLKDEBUG_IO_TYPE__MAX > 64, | ||
96 | + "BlkdebugIOType mask does not fit into an uint64_t"); | ||
97 | + | ||
98 | static QemuOptsList inject_error_opts = { | ||
99 | .name = "inject-error", | ||
100 | .head = QTAILQ_HEAD_INITIALIZER(inject_error_opts.head), | ||
101 | @@ -XXX,XX +XXX,XX @@ static QemuOptsList inject_error_opts = { | ||
102 | .name = "state", | ||
103 | .type = QEMU_OPT_NUMBER, | ||
104 | }, | ||
105 | + { | ||
106 | + .name = "iotype", | ||
107 | + .type = QEMU_OPT_STRING, | ||
108 | + }, | ||
109 | { | ||
110 | .name = "errno", | ||
111 | .type = QEMU_OPT_NUMBER, | ||
112 | @@ -XXX,XX +XXX,XX @@ static int add_rule(void *opaque, QemuOpts *opts, Error **errp) | ||
113 | int event; | ||
114 | struct BlkdebugRule *rule; | ||
115 | int64_t sector; | ||
116 | + BlkdebugIOType iotype; | ||
117 | + Error *local_error = NULL; | ||
118 | |||
119 | /* Find the right event for the rule */ | ||
120 | event_name = qemu_opt_get(opts, "event"); | ||
121 | @@ -XXX,XX +XXX,XX @@ static int add_rule(void *opaque, QemuOpts *opts, Error **errp) | ||
122 | sector = qemu_opt_get_number(opts, "sector", -1); | ||
123 | rule->options.inject.offset = | ||
124 | sector == -1 ? -1 : sector * BDRV_SECTOR_SIZE; | ||
125 | + | ||
126 | + iotype = qapi_enum_parse(&BlkdebugIOType_lookup, | ||
127 | + qemu_opt_get(opts, "iotype"), | ||
128 | + BLKDEBUG_IO_TYPE__MAX, &local_error); | ||
129 | + if (local_error) { | ||
130 | + error_propagate(errp, local_error); | ||
131 | + return -1; | ||
132 | + } | ||
133 | + if (iotype != BLKDEBUG_IO_TYPE__MAX) { | ||
134 | + rule->options.inject.iotype_mask = (1ull << iotype); | ||
135 | + } else { | ||
136 | + /* Apply the default */ | ||
137 | + rule->options.inject.iotype_mask = | ||
138 | + (1ull << BLKDEBUG_IO_TYPE_READ) | ||
139 | + | (1ull << BLKDEBUG_IO_TYPE_WRITE) | ||
140 | + | (1ull << BLKDEBUG_IO_TYPE_WRITE_ZEROES) | ||
141 | + | (1ull << BLKDEBUG_IO_TYPE_DISCARD) | ||
142 | + | (1ull << BLKDEBUG_IO_TYPE_FLUSH); | ||
143 | + } | ||
144 | + | ||
145 | break; | ||
146 | |||
147 | case ACTION_SET_STATE: | ||
148 | @@ -XXX,XX +XXX,XX @@ out: | 24 | @@ -XXX,XX +XXX,XX @@ out: |
149 | return ret; | 25 | return ret; |
150 | } | 26 | } |
151 | 27 | ||
152 | -static int rule_check(BlockDriverState *bs, uint64_t offset, uint64_t bytes) | 28 | -static void error_callback_bh(void *opaque) |
153 | +static int rule_check(BlockDriverState *bs, uint64_t offset, uint64_t bytes, | 29 | -{ |
154 | + BlkdebugIOType iotype) | 30 | - Coroutine *co = opaque; |
31 | - qemu_coroutine_enter(co); | ||
32 | -} | ||
33 | - | ||
34 | static int inject_error(BlockDriverState *bs, BlkdebugRule *rule) | ||
155 | { | 35 | { |
156 | BDRVBlkdebugState *s = bs->opaque; | 36 | BDRVBlkdebugState *s = bs->opaque; |
157 | BlkdebugRule *rule = NULL; | 37 | @@ -XXX,XX +XXX,XX @@ static int inject_error(BlockDriverState *bs, BlkdebugRule *rule) |
158 | @@ -XXX,XX +XXX,XX @@ static int rule_check(BlockDriverState *bs, uint64_t offset, uint64_t bytes) | ||
159 | QSIMPLEQ_FOREACH(rule, &s->active_rules, active_next) { | ||
160 | uint64_t inject_offset = rule->options.inject.offset; | ||
161 | |||
162 | - if (inject_offset == -1 || | ||
163 | - (bytes && inject_offset >= offset && | ||
164 | - inject_offset < offset + bytes)) | ||
165 | + if ((inject_offset == -1 || | ||
166 | + (bytes && inject_offset >= offset && | ||
167 | + inject_offset < offset + bytes)) && | ||
168 | + (rule->options.inject.iotype_mask & (1ull << iotype))) | ||
169 | { | ||
170 | break; | ||
171 | } | ||
172 | @@ -XXX,XX +XXX,XX @@ blkdebug_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes, | ||
173 | assert(bytes <= bs->bl.max_transfer); | ||
174 | } | 38 | } |
175 | 39 | ||
176 | - err = rule_check(bs, offset, bytes); | 40 | if (!immediately) { |
177 | + err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_READ); | 41 | - aio_bh_schedule_oneshot(bdrv_get_aio_context(bs), error_callback_bh, |
178 | if (err) { | 42 | - qemu_coroutine_self()); |
179 | return err; | 43 | + aio_co_schedule(qemu_get_current_aio_context(), qemu_coroutine_self()); |
44 | qemu_coroutine_yield(); | ||
180 | } | 45 | } |
181 | @@ -XXX,XX +XXX,XX @@ blkdebug_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes, | 46 | |
182 | assert(bytes <= bs->bl.max_transfer); | ||
183 | } | ||
184 | |||
185 | - err = rule_check(bs, offset, bytes); | ||
186 | + err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_WRITE); | ||
187 | if (err) { | ||
188 | return err; | ||
189 | } | ||
190 | @@ -XXX,XX +XXX,XX @@ blkdebug_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes, | ||
191 | |||
192 | static int blkdebug_co_flush(BlockDriverState *bs) | ||
193 | { | ||
194 | - int err = rule_check(bs, 0, 0); | ||
195 | + int err = rule_check(bs, 0, 0, BLKDEBUG_IO_TYPE_FLUSH); | ||
196 | |||
197 | if (err) { | ||
198 | return err; | ||
199 | @@ -XXX,XX +XXX,XX @@ static int coroutine_fn blkdebug_co_pwrite_zeroes(BlockDriverState *bs, | ||
200 | assert(bytes <= bs->bl.max_pwrite_zeroes); | ||
201 | } | ||
202 | |||
203 | - err = rule_check(bs, offset, bytes); | ||
204 | + err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_WRITE_ZEROES); | ||
205 | if (err) { | ||
206 | return err; | ||
207 | } | ||
208 | @@ -XXX,XX +XXX,XX @@ static int coroutine_fn blkdebug_co_pdiscard(BlockDriverState *bs, | ||
209 | assert(bytes <= bs->bl.max_pdiscard); | ||
210 | } | ||
211 | |||
212 | - err = rule_check(bs, offset, bytes); | ||
213 | + err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_DISCARD); | ||
214 | if (err) { | ||
215 | return err; | ||
216 | } | ||
217 | -- | 47 | -- |
218 | 2.21.0 | 48 | 2.9.3 |
219 | 49 | ||
220 | 50 | diff view generated by jsdifflib |
1 | From: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> | 1 | From: Paolo Bonzini <pbonzini@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | Test fails at least for qcow, because of different cluster sizes in | 3 | qed_aio_start_io and qed_aio_next_io will not have to acquire/release |
4 | base and top (and therefore different granularities of bitmaps we are | 4 | the AioContext, while qed_aio_next_io_cb will. Split the functionality |
5 | trying to merge). | 5 | and gain a little type-safety in the process. |
6 | 6 | ||
7 | The test aim is to check block-dirty-bitmap-merge between different | 7 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> |
8 | nodes functionality, no needs to check all formats. So, let's just drop | 8 | Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> |
9 | support for anything except qcow2. | 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> | ||
13 | --- | ||
14 | block/qed.c | 39 +++++++++++++++++++++++++-------------- | ||
15 | 1 file changed, 25 insertions(+), 14 deletions(-) | ||
10 | 16 | ||
11 | Reported-by: Max Reitz <mreitz@redhat.com> | 17 | diff --git a/block/qed.c b/block/qed.c |
12 | Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> | 18 | index XXXXXXX..XXXXXXX 100644 |
13 | Message-id: 20190605155405.104384-1-vsementsov@virtuozzo.com | 19 | --- a/block/qed.c |
14 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 20 | +++ b/block/qed.c |
15 | --- | 21 | @@ -XXX,XX +XXX,XX @@ static CachedL2Table *qed_new_l2_table(BDRVQEDState *s) |
16 | tests/qemu-iotests/254 | 2 ++ | 22 | return l2_table; |
17 | 1 file changed, 2 insertions(+) | 23 | } |
18 | 24 | ||
19 | diff --git a/tests/qemu-iotests/254 b/tests/qemu-iotests/254 | 25 | -static void qed_aio_next_io(void *opaque, int ret); |
20 | index XXXXXXX..XXXXXXX 100755 | 26 | +static void qed_aio_next_io(QEDAIOCB *acb, int ret); |
21 | --- a/tests/qemu-iotests/254 | ||
22 | +++ b/tests/qemu-iotests/254 | ||
23 | @@ -XXX,XX +XXX,XX @@ | ||
24 | import iotests | ||
25 | from iotests import qemu_img_create, file_path, log | ||
26 | |||
27 | +iotests.verify_image_format(supported_fmts=['qcow2']) | ||
28 | + | 27 | + |
29 | disk, top = file_path('disk', 'top') | 28 | +static void qed_aio_start_io(QEDAIOCB *acb) |
30 | size = 1024 * 1024 | 29 | +{ |
30 | + qed_aio_next_io(acb, 0); | ||
31 | +} | ||
32 | + | ||
33 | +static void qed_aio_next_io_cb(void *opaque, int ret) | ||
34 | +{ | ||
35 | + QEDAIOCB *acb = opaque; | ||
36 | + | ||
37 | + qed_aio_next_io(acb, ret); | ||
38 | +} | ||
39 | |||
40 | static void qed_plug_allocating_write_reqs(BDRVQEDState *s) | ||
41 | { | ||
42 | @@ -XXX,XX +XXX,XX @@ static void qed_unplug_allocating_write_reqs(BDRVQEDState *s) | ||
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); | ||
48 | } | ||
49 | } | ||
50 | |||
51 | @@ -XXX,XX +XXX,XX @@ static void qed_aio_complete(QEDAIOCB *acb, int ret) | ||
52 | QSIMPLEQ_REMOVE_HEAD(&s->allocating_write_reqs, next); | ||
53 | acb = QSIMPLEQ_FIRST(&s->allocating_write_reqs); | ||
54 | if (acb) { | ||
55 | - qed_aio_next_io(acb, 0); | ||
56 | + qed_aio_start_io(acb); | ||
57 | } else if (s->header.features & QED_F_NEED_CHECK) { | ||
58 | qed_start_need_check_timer(s); | ||
59 | } | ||
60 | @@ -XXX,XX +XXX,XX @@ static void qed_commit_l2_update(void *opaque, int ret) | ||
61 | acb->request.l2_table = qed_find_l2_cache_entry(&s->l2_cache, l2_offset); | ||
62 | assert(acb->request.l2_table != NULL); | ||
63 | |||
64 | - qed_aio_next_io(opaque, ret); | ||
65 | + qed_aio_next_io(acb, ret); | ||
66 | } | ||
67 | |||
68 | /** | ||
69 | @@ -XXX,XX +XXX,XX @@ static void qed_aio_write_l2_update(QEDAIOCB *acb, int ret, uint64_t offset) | ||
70 | if (need_alloc) { | ||
71 | /* Write out the whole new L2 table */ | ||
72 | qed_write_l2_table(s, &acb->request, 0, s->table_nelems, true, | ||
73 | - qed_aio_write_l1_update, acb); | ||
74 | + qed_aio_write_l1_update, acb); | ||
75 | } else { | ||
76 | /* Write out only the updated part of the L2 table */ | ||
77 | qed_write_l2_table(s, &acb->request, index, acb->cur_nclusters, false, | ||
78 | - qed_aio_next_io, acb); | ||
79 | + qed_aio_next_io_cb, acb); | ||
80 | } | ||
81 | return; | ||
82 | |||
83 | @@ -XXX,XX +XXX,XX @@ static void qed_aio_write_main(void *opaque, int ret) | ||
84 | } | ||
85 | |||
86 | if (acb->find_cluster_ret == QED_CLUSTER_FOUND) { | ||
87 | - next_fn = qed_aio_next_io; | ||
88 | + next_fn = qed_aio_next_io_cb; | ||
89 | } else { | ||
90 | if (s->bs->backing) { | ||
91 | next_fn = qed_aio_write_flush_before_l2_update; | ||
92 | @@ -XXX,XX +XXX,XX @@ static void qed_aio_write_alloc(QEDAIOCB *acb, size_t len) | ||
93 | if (acb->flags & QED_AIOCB_ZERO) { | ||
94 | /* Skip ahead if the clusters are already zero */ | ||
95 | if (acb->find_cluster_ret == QED_CLUSTER_ZERO) { | ||
96 | - qed_aio_next_io(acb, 0); | ||
97 | + qed_aio_start_io(acb); | ||
98 | return; | ||
99 | } | ||
100 | |||
101 | @@ -XXX,XX +XXX,XX @@ static void qed_aio_read_data(void *opaque, int ret, | ||
102 | /* Handle zero cluster and backing file reads */ | ||
103 | if (ret == QED_CLUSTER_ZERO) { | ||
104 | qemu_iovec_memset(&acb->cur_qiov, 0, 0, acb->cur_qiov.size); | ||
105 | - qed_aio_next_io(acb, 0); | ||
106 | + qed_aio_start_io(acb); | ||
107 | return; | ||
108 | } else if (ret != QED_CLUSTER_FOUND) { | ||
109 | qed_read_backing_file(s, acb->cur_pos, &acb->cur_qiov, | ||
110 | - &acb->backing_qiov, qed_aio_next_io, acb); | ||
111 | + &acb->backing_qiov, qed_aio_next_io_cb, acb); | ||
112 | return; | ||
113 | } | ||
114 | |||
115 | BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); | ||
116 | bdrv_aio_readv(bs->file, offset / BDRV_SECTOR_SIZE, | ||
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; | ||
141 | } | ||
31 | 142 | ||
32 | -- | 143 | -- |
33 | 2.21.0 | 144 | 2.9.3 |
34 | 145 | ||
35 | 146 | diff view generated by jsdifflib |
1 | There are error messages which refer to an overlay node as the snapshot. | 1 | From: Paolo Bonzini <pbonzini@redhat.com> |
---|---|---|---|
2 | That is wrong, those are two different things. | ||
3 | 2 | ||
4 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 3 | The AioContext data structures are now protected by list_lock and/or |
5 | Reviewed-by: Eric Blake <eblake@redhat.com> | 4 | they are walked with FOREACH_RCU primitives. There is no need anymore |
6 | Message-id: 20190603202236.1342-3-mreitz@redhat.com | 5 | to acquire the AioContext for the entire duration of aio_dispatch. |
7 | Reviewed-by: John Snow <jsnow@redhat.com> | 6 | Instead, just acquire it before and after invoking the callbacks. |
8 | Reviewed-by: Alberto Garcia <berto@igalia.com> | 7 | The next step is then to push it further down. |
9 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 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> | ||
10 | --- | 15 | --- |
11 | blockdev.c | 10 +++++----- | 16 | util/aio-posix.c | 25 +++++++++++-------------- |
12 | tests/qemu-iotests/085.out | 10 +++++----- | 17 | util/aio-win32.c | 15 +++++++-------- |
13 | 2 files changed, 10 insertions(+), 10 deletions(-) | 18 | util/async.c | 2 ++ |
19 | 3 files changed, 20 insertions(+), 22 deletions(-) | ||
14 | 20 | ||
15 | diff --git a/blockdev.c b/blockdev.c | 21 | diff --git a/util/aio-posix.c b/util/aio-posix.c |
16 | index XXXXXXX..XXXXXXX 100644 | 22 | index XXXXXXX..XXXXXXX 100644 |
17 | --- a/blockdev.c | 23 | --- a/util/aio-posix.c |
18 | +++ b/blockdev.c | 24 | +++ b/util/aio-posix.c |
19 | @@ -XXX,XX +XXX,XX @@ static void external_snapshot_prepare(BlkActionState *common, | 25 | @@ -XXX,XX +XXX,XX @@ static bool aio_dispatch_handlers(AioContext *ctx) |
20 | s->has_snapshot_node_name ? s->snapshot_node_name : NULL; | 26 | (revents & (G_IO_IN | G_IO_HUP | G_IO_ERR)) && |
21 | 27 | aio_node_check(ctx, node->is_external) && | |
22 | if (node_name && !snapshot_node_name) { | 28 | node->io_read) { |
23 | - error_setg(errp, "New snapshot node name missing"); | 29 | + aio_context_acquire(ctx); |
24 | + error_setg(errp, "New overlay node name missing"); | 30 | node->io_read(node->opaque); |
25 | goto out; | 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; | ||
26 | } | 43 | } |
27 | 44 | ||
28 | if (snapshot_node_name && | 45 | @@ -XXX,XX +XXX,XX @@ bool aio_dispatch(AioContext *ctx, bool dispatch_fds) |
29 | bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) { | 46 | } |
30 | - error_setg(errp, "New snapshot node name already in use"); | 47 | |
31 | + error_setg(errp, "New overlay node name already in use"); | 48 | /* Run our timers */ |
32 | goto out; | 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); | ||
33 | } | 93 | } |
34 | 94 | - if (timeout) { | |
35 | @@ -XXX,XX +XXX,XX @@ static void external_snapshot_prepare(BlkActionState *common, | 95 | - aio_context_acquire(ctx); |
96 | - } | ||
36 | } | 97 | } |
37 | 98 | ||
38 | if (bdrv_has_blk(state->new_bs)) { | 99 | if (blocking) { |
39 | - error_setg(errp, "The snapshot is already in use"); | 100 | @@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking) |
40 | + error_setg(errp, "The overlay is already in use"); | 101 | progress = true; |
41 | goto out; | ||
42 | } | 102 | } |
43 | 103 | ||
44 | @@ -XXX,XX +XXX,XX @@ static void external_snapshot_prepare(BlkActionState *common, | 104 | - aio_context_release(ctx); |
45 | } | 105 | - |
46 | 106 | return progress; | |
47 | if (state->new_bs->backing != NULL) { | 107 | } |
48 | - error_setg(errp, "The snapshot already has a backing image"); | 108 | |
49 | + error_setg(errp, "The overlay already has a backing image"); | 109 | diff --git a/util/aio-win32.c b/util/aio-win32.c |
50 | goto out; | ||
51 | } | ||
52 | |||
53 | if (!state->new_bs->drv->supports_backing) { | ||
54 | - error_setg(errp, "The snapshot does not support backing images"); | ||
55 | + error_setg(errp, "The overlay does not support backing images"); | ||
56 | goto out; | ||
57 | } | ||
58 | |||
59 | diff --git a/tests/qemu-iotests/085.out b/tests/qemu-iotests/085.out | ||
60 | index XXXXXXX..XXXXXXX 100644 | 110 | index XXXXXXX..XXXXXXX 100644 |
61 | --- a/tests/qemu-iotests/085.out | 111 | --- a/util/aio-win32.c |
62 | +++ b/tests/qemu-iotests/085.out | 112 | +++ b/util/aio-win32.c |
63 | @@ -XXX,XX +XXX,XX @@ Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=134217728 backing_file=TEST_DIR/ | 113 | @@ -XXX,XX +XXX,XX @@ static bool aio_dispatch_handlers(AioContext *ctx, HANDLE event) |
64 | 114 | (revents || event_notifier_get_handle(node->e) == event) && | |
65 | === Invalid command - cannot create a snapshot using a file BDS === | 115 | node->io_notify) { |
66 | 116 | node->pfd.revents = 0; | |
67 | -{"error": {"class": "GenericError", "desc": "The snapshot does not support backing images"}} | 117 | + aio_context_acquire(ctx); |
68 | +{"error": {"class": "GenericError", "desc": "The overlay does not support backing images"}} | 118 | node->io_notify(node->e); |
69 | 119 | + aio_context_release(ctx); | |
70 | === Invalid command - snapshot node used as active layer === | 120 | |
71 | 121 | /* aio_notify() does not count as progress */ | |
72 | -{"error": {"class": "GenericError", "desc": "The snapshot is already in use"}} | 122 | if (node->e != &ctx->notifier) { |
73 | -{"error": {"class": "GenericError", "desc": "The snapshot is already in use"}} | 123 | @@ -XXX,XX +XXX,XX @@ static bool aio_dispatch_handlers(AioContext *ctx, HANDLE event) |
74 | -{"error": {"class": "GenericError", "desc": "The snapshot is already in use"}} | 124 | (node->io_read || node->io_write)) { |
75 | +{"error": {"class": "GenericError", "desc": "The overlay is already in use"}} | 125 | node->pfd.revents = 0; |
76 | +{"error": {"class": "GenericError", "desc": "The overlay is already in use"}} | 126 | if ((revents & G_IO_IN) && node->io_read) { |
77 | +{"error": {"class": "GenericError", "desc": "The overlay is already in use"}} | 127 | + aio_context_acquire(ctx); |
78 | 128 | node->io_read(node->opaque); | |
79 | === Invalid command - snapshot node used as backing hd === | 129 | + aio_context_release(ctx); |
80 | 130 | progress = true; | |
81 | @@ -XXX,XX +XXX,XX @@ Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=134217728 backing_file=TEST_DIR/ | 131 | } |
82 | Formatting 'TEST_DIR/t.IMGFMT.base', fmt=IMGFMT size=134217728 | 132 | if ((revents & G_IO_OUT) && node->io_write) { |
83 | Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=134217728 backing_file=TEST_DIR/t.IMGFMT.base | 133 | + aio_context_acquire(ctx); |
84 | {"return": {}} | 134 | node->io_write(node->opaque); |
85 | -{"error": {"class": "GenericError", "desc": "The snapshot already has a backing image"}} | 135 | + aio_context_release(ctx); |
86 | +{"error": {"class": "GenericError", "desc": "The overlay already has a backing image"}} | 136 | progress = true; |
87 | 137 | } | |
88 | === Invalid command - The node does not exist === | 138 | |
89 | 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; | ||
90 | -- | 189 | -- |
91 | 2.21.0 | 190 | 2.9.3 |
92 | 191 | ||
93 | 192 | diff view generated by jsdifflib |
1 | It is possible for an empty file to take up blocks on a filesystem, for | 1 | From: Paolo Bonzini <pbonzini@redhat.com> |
---|---|---|---|
2 | example: | 2 | |
3 | 3 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> | |
4 | $ qemu-img create -f raw test.img 1G | 4 | Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> |
5 | Formatting 'test.img', fmt=raw size=1073741824 | 5 | Reviewed-by: Fam Zheng <famz@redhat.com> |
6 | $ mkfs.ext4 -I 128 -q test.img | 6 | Reviewed-by: Daniel P. Berrange <berrange@redhat.com> |
7 | $ mkdir test-mount | 7 | Message-id: 20170213135235.12274-13-pbonzini@redhat.com |
8 | $ sudo mount -o loop test.img test-mount | 8 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> |
9 | $ sudo touch test-mount/test-file | ||
10 | $ stat -c 'blocks=%b' test-mount/test-file | ||
11 | blocks=8 | ||
12 | |||
13 | These extra blocks (one cluster) are apparently used for metadata, | ||
14 | because they are always there, on top of blocks used for data: | ||
15 | |||
16 | $ sudo dd if=/dev/zero of=test-mount/test-file bs=1M count=1 | ||
17 | 1+0 records in | ||
18 | 1+0 records out | ||
19 | 1048576 bytes (1.0 MB, 1.0 MiB) copied, 0.00135339 s, 775 MB/s | ||
20 | $ stat -c 'blocks=%b' test-mount/test-file | ||
21 | blocks=2056 | ||
22 | |||
23 | Make iotest 175 take this into account. | ||
24 | |||
25 | Reported-by: Thomas Huth <thuth@redhat.com> | ||
26 | Signed-off-by: Max Reitz <mreitz@redhat.com> | ||
27 | Reviewed-by: Eric Blake <eblake@redhat.com> | ||
28 | Reviewed-by: Nir Soffer <nsoffer@redhat.com> | ||
29 | Message-id: 20190516144319.12570-1-mreitz@redhat.com | ||
30 | Signed-off-by: Max Reitz <mreitz@redhat.com> | ||
31 | --- | 9 | --- |
32 | tests/qemu-iotests/175 | 26 ++++++++++++++++++++++---- | 10 | block/qed.h | 3 +++ |
33 | tests/qemu-iotests/175.out | 8 ++++---- | 11 | block/curl.c | 2 ++ |
34 | 2 files changed, 26 insertions(+), 8 deletions(-) | 12 | block/io.c | 5 +++++ |
35 | 13 | block/iscsi.c | 8 ++++++-- | |
36 | diff --git a/tests/qemu-iotests/175 b/tests/qemu-iotests/175 | 14 | block/null.c | 4 ++++ |
37 | index XXXXXXX..XXXXXXX 100755 | 15 | block/qed.c | 12 ++++++++++++ |
38 | --- a/tests/qemu-iotests/175 | 16 | block/throttle-groups.c | 2 ++ |
39 | +++ b/tests/qemu-iotests/175 | 17 | util/aio-posix.c | 2 -- |
40 | @@ -XXX,XX +XXX,XX @@ status=1 # failure is the default! | 18 | util/aio-win32.c | 2 -- |
41 | 19 | util/qemu-coroutine-sleep.c | 2 +- | |
42 | _cleanup() | 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) | ||
43 | { | 82 | { |
44 | - _cleanup_test_img | 83 | IscsiLun *iscsilun = opaque; |
45 | + _cleanup_test_img | 84 | |
46 | + rm -f "$TEST_DIR/empty" | 85 | + aio_context_acquire(iscsilun->aio_context); |
47 | } | 86 | if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) { |
48 | trap "_cleanup; exit \$status" 0 1 2 3 15 | 87 | error_report("iSCSI: NOP timeout. Reconnecting..."); |
49 | 88 | iscsilun->request_timed_out = true; | |
50 | +# Some file systems sometimes allocate extra blocks independently of | 89 | } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) { |
51 | +# the file size. This function hides the resulting difference in the | 90 | error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages."); |
52 | +# stat -c '%b' output. | 91 | - return; |
53 | +# Parameter 1: Number of blocks an empty file occupies | 92 | + goto out; |
54 | +# Parameter 2: Image size in bytes | 93 | } |
55 | +_filter_blocks() | 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) | ||
56 | +{ | 136 | +{ |
57 | + extra_blocks=$1 | 137 | + aio_context_acquire(bdrv_get_aio_context(s->bs)); |
58 | + img_size=$2 | ||
59 | + | ||
60 | + sed -e "s/blocks=$extra_blocks\\(\$\\|[^0-9]\\)/nothing allocated/" \ | ||
61 | + -e "s/blocks=$((extra_blocks + img_size / 512))\\(\$\\|[^0-9]\\)/everything allocated/" | ||
62 | +} | 138 | +} |
63 | + | 139 | + |
64 | # get standard environment, filters and checks | 140 | +void qed_release(BDRVQEDState *s) |
65 | . ./common.rc | 141 | +{ |
66 | . ./common.filter | 142 | + aio_context_release(bdrv_get_aio_context(s->bs)); |
67 | @@ -XXX,XX +XXX,XX @@ _supported_fmt raw | 143 | } |
68 | _supported_proto file | 144 | |
69 | _supported_os Linux | 145 | static void qed_start_need_check_timer(BDRVQEDState *s) |
70 | 146 | diff --git a/block/throttle-groups.c b/block/throttle-groups.c | |
71 | -size=1m | 147 | index XXXXXXX..XXXXXXX 100644 |
72 | +size=$((1 * 1024 * 1024)) | 148 | --- a/block/throttle-groups.c |
73 | + | 149 | +++ b/block/throttle-groups.c |
74 | +touch "$TEST_DIR/empty" | 150 | @@ -XXX,XX +XXX,XX @@ static void timer_cb(BlockBackend *blk, bool is_write) |
75 | +extra_blocks=$(stat -c '%b' "$TEST_DIR/empty") | 151 | qemu_mutex_unlock(&tg->lock); |
76 | 152 | ||
77 | echo | 153 | /* Run the request that was waiting for this timer */ |
78 | echo "== creating image with default preallocation ==" | 154 | + aio_context_acquire(blk_get_aio_context(blk)); |
79 | _make_test_img $size | _filter_imgfmt | 155 | empty_queue = !qemu_co_enter_next(&blkp->throttled_reqs[is_write]); |
80 | -stat -c "size=%s, blocks=%b" $TEST_IMG | 156 | + aio_context_release(blk_get_aio_context(blk)); |
81 | +stat -c "size=%s, blocks=%b" $TEST_IMG | _filter_blocks $extra_blocks $size | 157 | |
82 | 158 | /* If the request queue was empty then we have to take care of | |
83 | for mode in off full falloc; do | 159 | * scheduling the next one */ |
84 | echo | 160 | diff --git a/util/aio-posix.c b/util/aio-posix.c |
85 | echo "== creating image with preallocation $mode ==" | 161 | index XXXXXXX..XXXXXXX 100644 |
86 | IMGOPTS=preallocation=$mode _make_test_img $size | _filter_imgfmt | 162 | --- a/util/aio-posix.c |
87 | - stat -c "size=%s, blocks=%b" $TEST_IMG | 163 | +++ b/util/aio-posix.c |
88 | + stat -c "size=%s, blocks=%b" $TEST_IMG | _filter_blocks $extra_blocks $size | 164 | @@ -XXX,XX +XXX,XX @@ bool aio_dispatch(AioContext *ctx, bool dispatch_fds) |
89 | done | 165 | } |
90 | 166 | ||
91 | # success, all done | 167 | /* Run our timers */ |
92 | diff --git a/tests/qemu-iotests/175.out b/tests/qemu-iotests/175.out | 168 | - aio_context_acquire(ctx); |
93 | index XXXXXXX..XXXXXXX 100644 | 169 | progress |= timerlistgroup_run_timers(&ctx->tlg); |
94 | --- a/tests/qemu-iotests/175.out | 170 | - aio_context_release(ctx); |
95 | +++ b/tests/qemu-iotests/175.out | 171 | |
96 | @@ -XXX,XX +XXX,XX @@ QA output created by 175 | 172 | return progress; |
97 | 173 | } | |
98 | == creating image with default preallocation == | 174 | diff --git a/util/aio-win32.c b/util/aio-win32.c |
99 | Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=1048576 | 175 | index XXXXXXX..XXXXXXX 100644 |
100 | -size=1048576, blocks=0 | 176 | --- a/util/aio-win32.c |
101 | +size=1048576, nothing allocated | 177 | +++ b/util/aio-win32.c |
102 | 178 | @@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking) | |
103 | == creating image with preallocation off == | 179 | progress |= aio_dispatch_handlers(ctx, event); |
104 | Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=1048576 preallocation=off | 180 | } while (count > 0); |
105 | -size=1048576, blocks=0 | 181 | |
106 | +size=1048576, nothing allocated | 182 | - aio_context_acquire(ctx); |
107 | 183 | progress |= timerlistgroup_run_timers(&ctx->tlg); | |
108 | == creating image with preallocation full == | 184 | - aio_context_release(ctx); |
109 | Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=1048576 preallocation=full | 185 | return progress; |
110 | -size=1048576, blocks=2048 | 186 | } |
111 | +size=1048576, everything allocated | 187 | |
112 | 188 | diff --git a/util/qemu-coroutine-sleep.c b/util/qemu-coroutine-sleep.c | |
113 | == creating image with preallocation falloc == | 189 | index XXXXXXX..XXXXXXX 100644 |
114 | Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=1048576 preallocation=falloc | 190 | --- a/util/qemu-coroutine-sleep.c |
115 | -size=1048576, blocks=2048 | 191 | +++ b/util/qemu-coroutine-sleep.c |
116 | +size=1048576, everything allocated | 192 | @@ -XXX,XX +XXX,XX @@ static void co_sleep_cb(void *opaque) |
117 | *** done | 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, | ||
118 | -- | 201 | -- |
119 | 2.21.0 | 202 | 2.9.3 |
120 | 203 | ||
121 | 204 | diff view generated by jsdifflib |
1 | From: John Snow <jsnow@redhat.com> | 1 | From: Paolo Bonzini <pbonzini@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | Instead of event_wait which looks for a single event, add an events_wait | 3 | This covers both file descriptor callbacks and polling callbacks, |
4 | which can look for any number of events simultaneously. However, it | 4 | since they execute related code. |
5 | will still only return one at a time, whichever happens first. | ||
6 | 5 | ||
7 | Signed-off-by: John Snow <jsnow@redhat.com> | 6 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> |
8 | Message-id: 20190523170643.20794-4-jsnow@redhat.com | 7 | Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> |
9 | Reviewed-by: Max Reitz <mreitz@redhat.com> | 8 | Reviewed-by: Fam Zheng <famz@redhat.com> |
10 | Signed-off-by: Max Reitz <mreitz@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> | ||
11 | --- | 12 | --- |
12 | python/qemu/__init__.py | 69 +++++++++++++++++++++++++++++------------ | 13 | block/curl.c | 16 +++++++++++++--- |
13 | 1 file changed, 49 insertions(+), 20 deletions(-) | 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(-) | ||
14 | 25 | ||
15 | diff --git a/python/qemu/__init__.py b/python/qemu/__init__.py | 26 | diff --git a/block/curl.c b/block/curl.c |
16 | index XXXXXXX..XXXXXXX 100644 | 27 | index XXXXXXX..XXXXXXX 100644 |
17 | --- a/python/qemu/__init__.py | 28 | --- a/block/curl.c |
18 | +++ b/python/qemu/__init__.py | 29 | +++ b/block/curl.c |
19 | @@ -XXX,XX +XXX,XX @@ class QEMUMachine(object): | 30 | @@ -XXX,XX +XXX,XX @@ static void curl_multi_check_completion(BDRVCURLState *s) |
20 | self._qmp.clear_events() | 31 | } |
21 | return events | 32 | } |
22 | 33 | ||
23 | - def event_wait(self, name, timeout=60.0, match=None): | 34 | -static void curl_multi_do(void *arg) |
24 | + @staticmethod | 35 | +static void curl_multi_do_locked(CURLState *s) |
25 | + def event_match(event, match=None): | 36 | { |
26 | """ | 37 | - CURLState *s = (CURLState *)arg; |
27 | - Wait for specified timeout on named event in QMP; optionally filter | 38 | CURLSocket *socket, *next_socket; |
28 | - results by match. | 39 | int running; |
29 | + Check if an event matches optional match criteria. | 40 | int r; |
30 | 41 | @@ -XXX,XX +XXX,XX @@ static void curl_multi_do(void *arg) | |
31 | - The 'match' is checked to be a recursive subset of the 'event'; skips | 42 | } |
32 | - branch processing on match's value None | 43 | } |
33 | - {"foo": {"bar": 1}} matches {"foo": None} | 44 | |
34 | - {"foo": {"bar": 1}} does not matches {"foo": {"baz": None}} | 45 | +static void curl_multi_do(void *arg) |
35 | + The match criteria takes the form of a matching subdict. The event is | 46 | +{ |
36 | + checked to be a superset of the subdict, recursively, with matching | 47 | + CURLState *s = (CURLState *)arg; |
37 | + values whenever those values are not None. | 48 | + |
38 | + | 49 | + aio_context_acquire(s->s->aio_context); |
39 | + Examples, with the subdict queries on the left: | 50 | + curl_multi_do_locked(s); |
40 | + - None matches any object. | 51 | + aio_context_release(s->s->aio_context); |
41 | + - {"foo": None} matches {"foo": {"bar": 1}} | 52 | +} |
42 | + - {"foo": {"baz": None}} does not match {"foo": {"bar": 1}} | 53 | + |
43 | + - {"foo": {"baz": 2}} matches {"foo": {"bar": 1, "baz": 2}} | 54 | static void curl_multi_read(void *arg) |
44 | """ | 55 | { |
45 | - def event_match(event, match=None): | 56 | CURLState *s = (CURLState *)arg; |
46 | - if match is None: | 57 | |
47 | - return True | 58 | - curl_multi_do(arg); |
48 | + if match is None: | 59 | + aio_context_acquire(s->s->aio_context); |
49 | + return True | 60 | + curl_multi_do_locked(s); |
50 | 61 | curl_multi_check_completion(s->s); | |
51 | - for key in match: | 62 | + aio_context_release(s->s->aio_context); |
52 | - if key in event: | 63 | } |
53 | - if isinstance(event[key], dict): | 64 | |
54 | - if not event_match(event[key], match[key]): | 65 | static void curl_multi_timeout_do(void *arg) |
55 | - return False | 66 | diff --git a/block/iscsi.c b/block/iscsi.c |
56 | - elif event[key] != match[key]: | 67 | index XXXXXXX..XXXXXXX 100644 |
57 | + for key in match: | 68 | --- a/block/iscsi.c |
58 | + if key in event: | 69 | +++ b/block/iscsi.c |
59 | + if isinstance(event[key], dict): | 70 | @@ -XXX,XX +XXX,XX @@ iscsi_process_read(void *arg) |
60 | + if not QEMUMachine.event_match(event[key], match[key]): | 71 | IscsiLun *iscsilun = arg; |
61 | return False | 72 | struct iscsi_context *iscsi = iscsilun->iscsi; |
62 | - else: | 73 | |
63 | + elif event[key] != match[key]: | 74 | + aio_context_acquire(iscsilun->aio_context); |
64 | return False | 75 | iscsi_service(iscsi, POLLIN); |
65 | + else: | 76 | iscsi_set_events(iscsilun); |
66 | + return False | 77 | + aio_context_release(iscsilun->aio_context); |
67 | + return True | 78 | } |
68 | 79 | ||
69 | - return True | 80 | static void |
70 | + def event_wait(self, name, timeout=60.0, match=None): | 81 | @@ -XXX,XX +XXX,XX @@ iscsi_process_write(void *arg) |
71 | + """ | 82 | IscsiLun *iscsilun = arg; |
72 | + event_wait waits for and returns a named event from QMP with a timeout. | 83 | struct iscsi_context *iscsi = iscsilun->iscsi; |
73 | + | 84 | |
74 | + name: The event to wait for. | 85 | + aio_context_acquire(iscsilun->aio_context); |
75 | + timeout: QEMUMonitorProtocol.pull_event timeout parameter. | 86 | iscsi_service(iscsi, POLLOUT); |
76 | + match: Optional match criteria. See event_match for details. | 87 | iscsi_set_events(iscsilun); |
77 | + """ | 88 | + aio_context_release(iscsilun->aio_context); |
78 | + return self.events_wait([(name, match)], timeout) | 89 | } |
79 | + | 90 | |
80 | + def events_wait(self, events, timeout=60.0): | 91 | static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun) |
81 | + """ | 92 | diff --git a/block/linux-aio.c b/block/linux-aio.c |
82 | + events_wait waits for and returns a named event from QMP with a timeout. | 93 | index XXXXXXX..XXXXXXX 100644 |
83 | + | 94 | --- a/block/linux-aio.c |
84 | + events: a sequence of (name, match_criteria) tuples. | 95 | +++ b/block/linux-aio.c |
85 | + The match criteria are optional and may be None. | 96 | @@ -XXX,XX +XXX,XX @@ static void qemu_laio_completion_cb(EventNotifier *e) |
86 | + See event_match for details. | 97 | LinuxAioState *s = container_of(e, LinuxAioState, e); |
87 | + timeout: QEMUMonitorProtocol.pull_event timeout parameter. | 98 | |
88 | + """ | 99 | if (event_notifier_test_and_clear(&s->e)) { |
89 | + def _match(event): | 100 | + aio_context_acquire(s->aio_context); |
90 | + for name, match in events: | 101 | qemu_laio_process_completions_and_submit(s); |
91 | + if (event['event'] == name and | 102 | + aio_context_release(s->aio_context); |
92 | + self.event_match(event, match)): | 103 | } |
93 | + return True | 104 | } |
94 | + return False | 105 | |
95 | 106 | @@ -XXX,XX +XXX,XX @@ static bool qemu_laio_poll_cb(void *opaque) | |
96 | # Search cached events | 107 | return false; |
97 | for event in self._events: | 108 | } |
98 | - if (event['event'] == name) and event_match(event, match): | 109 | |
99 | + if _match(event): | 110 | + aio_context_acquire(s->aio_context); |
100 | self._events.remove(event) | 111 | qemu_laio_process_completions_and_submit(s); |
101 | return event | 112 | + aio_context_release(s->aio_context); |
102 | 113 | return true; | |
103 | # Poll for new events | 114 | } |
104 | while True: | 115 | |
105 | event = self._qmp.pull_event(wait=timeout) | 116 | diff --git a/block/nfs.c b/block/nfs.c |
106 | - if (event['event'] == name) and event_match(event, match): | 117 | index XXXXXXX..XXXXXXX 100644 |
107 | + if _match(event): | 118 | --- a/block/nfs.c |
108 | return event | 119 | +++ b/block/nfs.c |
109 | self._events.append(event) | 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 | } | ||
110 | 475 | ||
111 | -- | 476 | -- |
112 | 2.21.0 | 477 | 2.9.3 |
113 | 478 | ||
114 | 479 | diff view generated by jsdifflib |
1 | In 219, we wait for the job to make progress before we emit its status. | 1 | From: Paolo Bonzini <pbonzini@redhat.com> |
---|---|---|---|
2 | This makes the output reliable. We do not wait for any more progress if | ||
3 | the job's current-progress already matches its total-progress. | ||
4 | 2 | ||
5 | Unfortunately, there is a bug: Right after the job has been started, | 3 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> |
6 | it's possible that total-progress is still 0. In that case, we may skip | 4 | Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> |
7 | the first progress-making step and keep ending up 64 kB short. | 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(-) | ||
8 | 28 | ||
9 | To fix that bug, we can simply wait for total-progress to reach 4 MB | 29 | diff --git a/block/archipelago.c b/block/archipelago.c |
10 | (the image size) after starting the job. | 30 | index XXXXXXX..XXXXXXX 100644 |
11 | 31 | --- a/block/archipelago.c | |
12 | Reported-by: Karen Mezick <kmezick@redhat.com> | 32 | +++ b/block/archipelago.c |
13 | Buglink: https://bugzilla.redhat.com/show_bug.cgi?id=1686651 | 33 | @@ -XXX,XX +XXX,XX @@ static void qemu_archipelago_complete_aio(void *opaque) |
14 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 34 | { |
15 | Message-id: 20190516161114.27596-1-mreitz@redhat.com | 35 | AIORequestData *reqdata = (AIORequestData *) opaque; |
16 | Reviewed-by: John Snow <jsnow@redhat.com> | 36 | ArchipelagoAIOCB *aio_cb = (ArchipelagoAIOCB *) reqdata->aio_cb; |
17 | [mreitz: Adjusted commit message as per John's proposal] | 37 | + AioContext *ctx = bdrv_get_aio_context(aio_cb->common.bs); |
18 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 38 | |
19 | --- | 39 | + aio_context_acquire(ctx); |
20 | tests/qemu-iotests/219 | 13 ++++++++++--- | 40 | aio_cb->common.cb(aio_cb->common.opaque, aio_cb->ret); |
21 | 1 file changed, 10 insertions(+), 3 deletions(-) | 41 | + aio_context_release(ctx); |
22 | 42 | aio_cb->status = 0; | |
23 | diff --git a/tests/qemu-iotests/219 b/tests/qemu-iotests/219 | 43 | |
44 | qemu_aio_unref(aio_cb); | ||
45 | diff --git a/block/blkreplay.c b/block/blkreplay.c | ||
24 | index XXXXXXX..XXXXXXX 100755 | 46 | index XXXXXXX..XXXXXXX 100755 |
25 | --- a/tests/qemu-iotests/219 | 47 | --- a/block/blkreplay.c |
26 | +++ b/tests/qemu-iotests/219 | 48 | +++ b/block/blkreplay.c |
27 | @@ -XXX,XX +XXX,XX @@ import iotests | 49 | @@ -XXX,XX +XXX,XX @@ static int64_t blkreplay_getlength(BlockDriverState *bs) |
28 | 50 | static void blkreplay_bh_cb(void *opaque) | |
29 | iotests.verify_image_format(supported_fmts=['qcow2']) | 51 | { |
30 | 52 | Request *req = opaque; | |
31 | +img_size = 4 * 1024 * 1024 | 53 | - qemu_coroutine_enter(req->co); |
32 | + | 54 | + aio_co_wake(req->co); |
33 | def pause_wait(vm, job_id): | 55 | qemu_bh_delete(req->bh); |
34 | with iotests.Timeout(3, "Timeout waiting for job to pause"): | 56 | g_free(req); |
35 | while True: | 57 | } |
36 | @@ -XXX,XX +XXX,XX @@ def test_pause_resume(vm): | 58 | diff --git a/block/block-backend.c b/block/block-backend.c |
37 | iotests.log(vm.qmp('query-jobs')) | 59 | index XXXXXXX..XXXXXXX 100644 |
38 | 60 | --- a/block/block-backend.c | |
39 | def test_job_lifecycle(vm, job, job_args, has_ready=False): | 61 | +++ b/block/block-backend.c |
40 | + global img_size | 62 | @@ -XXX,XX +XXX,XX @@ int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags) |
41 | + | 63 | static void error_callback_bh(void *opaque) |
42 | iotests.log('') | 64 | { |
43 | iotests.log('') | 65 | struct BlockBackendAIOCB *acb = opaque; |
44 | iotests.log('Starting block job: %s (auto-finalize: %s; auto-dismiss: %s)' % | 66 | + AioContext *ctx = bdrv_get_aio_context(acb->common.bs); |
45 | @@ -XXX,XX +XXX,XX @@ def test_job_lifecycle(vm, job, job_args, has_ready=False): | 67 | |
46 | iotests.log(iotests.filter_qmp_event(vm.event_wait('JOB_STATUS_CHANGE'))) | 68 | bdrv_dec_in_flight(acb->common.bs); |
47 | iotests.log(iotests.filter_qmp_event(vm.event_wait('JOB_STATUS_CHANGE'))) | 69 | + aio_context_acquire(ctx); |
48 | 70 | acb->common.cb(acb->common.opaque, acb->ret); | |
49 | + # Wait for total-progress to stabilize | 71 | + aio_context_release(ctx); |
50 | + while vm.qmp('query-jobs')['return'][0]['total-progress'] < img_size: | 72 | qemu_aio_unref(acb); |
51 | + pass | 73 | } |
52 | + | 74 | |
53 | # RUNNING state: | 75 | @@ -XXX,XX +XXX,XX @@ static void blk_aio_complete(BlkAioEmAIOCB *acb) |
54 | # pause/resume should work, complete/finalize/dismiss should error out | 76 | static void blk_aio_complete_bh(void *opaque) |
55 | iotests.log('') | 77 | { |
56 | @@ -XXX,XX +XXX,XX @@ with iotests.FilePath('disk.img') as disk_path, \ | 78 | BlkAioEmAIOCB *acb = opaque; |
57 | iotests.FilePath('copy.img') as copy_path, \ | 79 | + AioContext *ctx = bdrv_get_aio_context(acb->common.bs); |
58 | iotests.VM() as vm: | 80 | |
59 | 81 | assert(acb->has_returned); | |
60 | - img_size = '4M' | 82 | + aio_context_acquire(ctx); |
61 | - iotests.qemu_img_create('-f', iotests.imgfmt, disk_path, img_size) | 83 | blk_aio_complete(acb); |
62 | - iotests.qemu_io('-c', 'write 0 %s' % (img_size), | 84 | + aio_context_release(ctx); |
63 | + iotests.qemu_img_create('-f', iotests.imgfmt, disk_path, str(img_size)) | 85 | } |
64 | + iotests.qemu_io('-c', 'write 0 %i' % (img_size), | 86 | |
65 | '-f', iotests.imgfmt, disk_path) | 87 | static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes, |
66 | 88 | diff --git a/block/curl.c b/block/curl.c | |
67 | iotests.log('Launching VM...') | 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) | ||
68 | -- | 502 | -- |
69 | 2.21.0 | 503 | 2.9.3 |
70 | 504 | ||
71 | 505 | diff view generated by jsdifflib |
1 | From: John Snow <jsnow@redhat.com> | 1 | From: Paolo Bonzini <pbonzini@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | Before, event_match didn't always recurse if the event value was not a | 3 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> |
4 | dictionary, and would instead check for equality immediately. | 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(-) | ||
5 | 28 | ||
6 | By delaying equality checking to post-recursion, we can allow leaf | 29 | diff --git a/block/archipelago.c b/block/archipelago.c |
7 | values like "5" to match "None" and take advantage of the generic | 30 | index XXXXXXX..XXXXXXX 100644 |
8 | None-returns-True clause. | 31 | --- a/block/archipelago.c |
9 | 32 | +++ b/block/archipelago.c | |
10 | This makes the matching a little more obviously consistent at the | 33 | @@ -XXX,XX +XXX,XX @@ static void qemu_archipelago_complete_aio(void *opaque) |
11 | expense of being able to check for explicit None values, which is | 34 | { |
12 | probably not that important given what this function is used for. | 35 | AIORequestData *reqdata = (AIORequestData *) opaque; |
13 | 36 | ArchipelagoAIOCB *aio_cb = (ArchipelagoAIOCB *) reqdata->aio_cb; | |
14 | Signed-off-by: John Snow <jsnow@redhat.com> | 37 | - AioContext *ctx = bdrv_get_aio_context(aio_cb->common.bs); |
15 | Message-id: 20190528183857.26167-1-jsnow@redhat.com | 38 | |
16 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 39 | - aio_context_acquire(ctx); |
17 | --- | 40 | aio_cb->common.cb(aio_cb->common.opaque, aio_cb->ret); |
18 | python/qemu/__init__.py | 24 ++++++++++++++---------- | 41 | - aio_context_release(ctx); |
19 | 1 file changed, 14 insertions(+), 10 deletions(-) | 42 | aio_cb->status = 0; |
20 | 43 | ||
21 | diff --git a/python/qemu/__init__.py b/python/qemu/__init__.py | 44 | qemu_aio_unref(aio_cb); |
22 | index XXXXXXX..XXXXXXX 100644 | 45 | diff --git a/block/block-backend.c b/block/block-backend.c |
23 | --- a/python/qemu/__init__.py | 46 | index XXXXXXX..XXXXXXX 100644 |
24 | +++ b/python/qemu/__init__.py | 47 | --- a/block/block-backend.c |
25 | @@ -XXX,XX +XXX,XX @@ class QEMUMachine(object): | 48 | +++ b/block/block-backend.c |
26 | 49 | @@ -XXX,XX +XXX,XX @@ int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags) | |
27 | The match criteria takes the form of a matching subdict. The event is | 50 | static void error_callback_bh(void *opaque) |
28 | checked to be a superset of the subdict, recursively, with matching | 51 | { |
29 | - values whenever those values are not None. | 52 | struct BlockBackendAIOCB *acb = opaque; |
30 | + values whenever the subdict values are not None. | 53 | - AioContext *ctx = bdrv_get_aio_context(acb->common.bs); |
31 | + | 54 | |
32 | + This has a limitation that you cannot explicitly check for None values. | 55 | bdrv_dec_in_flight(acb->common.bs); |
33 | 56 | - aio_context_acquire(ctx); | |
34 | Examples, with the subdict queries on the left: | 57 | acb->common.cb(acb->common.opaque, acb->ret); |
35 | - None matches any object. | 58 | - aio_context_release(ctx); |
36 | - {"foo": None} matches {"foo": {"bar": 1}} | 59 | qemu_aio_unref(acb); |
37 | - - {"foo": {"baz": None}} does not match {"foo": {"bar": 1}} | 60 | } |
38 | - - {"foo": {"baz": 2}} matches {"foo": {"bar": 1, "baz": 2}} | 61 | |
39 | + - {"foo": None} matches {"foo": 5} | 62 | @@ -XXX,XX +XXX,XX @@ static void blk_aio_complete(BlkAioEmAIOCB *acb) |
40 | + - {"foo": {"abc": None}} does not match {"foo": {"bar": 1}} | 63 | static void blk_aio_complete_bh(void *opaque) |
41 | + - {"foo": {"rab": 2}} matches {"foo": {"bar": 1, "rab": 2}} | 64 | { |
42 | """ | 65 | BlkAioEmAIOCB *acb = opaque; |
43 | if match is None: | 66 | - AioContext *ctx = bdrv_get_aio_context(acb->common.bs); |
44 | return True | 67 | - |
45 | 68 | assert(acb->has_returned); | |
46 | - for key in match: | 69 | - aio_context_acquire(ctx); |
47 | - if key in event: | 70 | blk_aio_complete(acb); |
48 | - if isinstance(event[key], dict): | 71 | - aio_context_release(ctx); |
49 | + try: | 72 | } |
50 | + for key in match: | 73 | |
51 | + if key in event: | 74 | static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes, |
52 | if not QEMUMachine.event_match(event[key], match[key]): | 75 | diff --git a/block/curl.c b/block/curl.c |
53 | return False | 76 | index XXXXXXX..XXXXXXX 100644 |
54 | - elif event[key] != match[key]: | 77 | --- a/block/curl.c |
55 | + else: | 78 | +++ b/block/curl.c |
56 | return False | 79 | @@ -XXX,XX +XXX,XX @@ static void curl_readv_bh_cb(void *p) |
57 | - else: | 80 | curl_multi_socket_action(s->multi, CURL_SOCKET_TIMEOUT, 0, &running); |
58 | - return False | 81 | |
59 | - return True | 82 | out: |
60 | + return True | 83 | + aio_context_release(ctx); |
61 | + except TypeError: | 84 | if (ret != -EINPROGRESS) { |
62 | + # either match or event wasn't iterable (not a dict) | 85 | acb->common.cb(acb->common.opaque, ret); |
63 | + return match == event | 86 | qemu_aio_unref(acb); |
64 | 87 | } | |
65 | def event_wait(self, name, timeout=60.0, match=None): | 88 | - aio_context_release(ctx); |
66 | """ | 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, | ||
67 | -- | 656 | -- |
68 | 2.21.0 | 657 | 2.9.3 |
69 | 658 | ||
70 | 659 | diff view generated by jsdifflib |
1 | From: John Snow <jsnow@redhat.com> | 1 | From: Paolo Bonzini <pbonzini@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | in blockdev_backup_prepare, we check to make sure that the target is | 3 | This patch prepares for the removal of unnecessary lockcnt inc/dec pairs. |
4 | associated with a compatible aio context. However, do_blockdev_backup is | 4 | Extract the dispatching loop for file descriptor handlers into a new |
5 | called later and has some logic to move the target to a compatible | 5 | function aio_dispatch_handlers, and then inline aio_dispatch into |
6 | aio_context. The transaction version will fail certain commands | 6 | aio_poll. |
7 | needlessly early as a result. | ||
8 | 7 | ||
9 | Allow blockdev_backup_prepare to simply call do_blockdev_backup, which | 8 | aio_dispatch can now become void. |
10 | will ultimately decide if the contexts are compatible or not. | ||
11 | 9 | ||
12 | Note: the transaction version has always disallowed this operation since | 10 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> |
13 | its initial commit bd8baecd (2014), whereas the version of | 11 | Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> |
14 | qmp_blockdev_backup at the time, from commit c29c1dd312f, tried to | 12 | Reviewed-by: Fam Zheng <famz@redhat.com> |
15 | enforce the aio_context switch instead. It's not clear, and I can't see | 13 | Reviewed-by: Daniel P. Berrange <berrange@redhat.com> |
16 | from the mailing list archives at the time, why the two functions take a | 14 | Message-id: 20170213135235.12274-17-pbonzini@redhat.com |
17 | different approach. It wasn't until later in efd7556708b (2016) that the | 15 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> |
18 | standalone version tried to determine if it could set the context or | 16 | --- |
19 | not. | 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(-) | ||
20 | 22 | ||
21 | Reported-by: aihua liang <aliang@redhat.com> | 23 | diff --git a/include/block/aio.h b/include/block/aio.h |
22 | Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1683498 | ||
23 | Signed-off-by: John Snow <jsnow@redhat.com> | ||
24 | Message-id: 20190523170643.20794-2-jsnow@redhat.com | ||
25 | Reviewed-by: Max Reitz <mreitz@redhat.com> | ||
26 | Signed-off-by: Max Reitz <mreitz@redhat.com> | ||
27 | --- | ||
28 | blockdev.c | 4 ---- | ||
29 | 1 file changed, 4 deletions(-) | ||
30 | |||
31 | diff --git a/blockdev.c b/blockdev.c | ||
32 | index XXXXXXX..XXXXXXX 100644 | 24 | index XXXXXXX..XXXXXXX 100644 |
33 | --- a/blockdev.c | 25 | --- a/include/block/aio.h |
34 | +++ b/blockdev.c | 26 | +++ b/include/block/aio.h |
35 | @@ -XXX,XX +XXX,XX @@ static void blockdev_backup_prepare(BlkActionState *common, Error **errp) | 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 | } | ||
36 | } | 60 | } |
37 | 61 | ||
38 | aio_context = bdrv_get_aio_context(bs); | 62 | - qemu_lockcnt_dec(&ctx->list_lock); |
39 | - if (aio_context != bdrv_get_aio_context(target)) { | 63 | return progress; |
40 | - error_setg(errp, "Backup between two IO threads is not implemented"); | 64 | } |
41 | - return; | 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); | ||
42 | - } | 88 | - } |
43 | aio_context_acquire(aio_context); | 89 | - |
44 | state->bs = bs; | 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 | } | ||
45 | 155 | ||
46 | -- | 156 | -- |
47 | 2.21.0 | 157 | 2.9.3 |
48 | 158 | ||
49 | 159 | diff view generated by jsdifflib |
1 | Move img_convert()'s quiet flag into the ImgConvertState so it is | 1 | From: Paolo Bonzini <pbonzini@redhat.com> |
---|---|---|---|
2 | accessible by nested functions. -q dictates that it suppresses anything | ||
3 | but errors, so if those functions want to emit warnings, they need to | ||
4 | query this flag first. (There currently are no such warnings, but there | ||
5 | will be as of the next patch.) | ||
6 | 2 | ||
7 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 3 | Pull the increment/decrement pair out of aio_bh_poll and into the |
8 | Reviewed-by: Eric Blake <eblake@redhat.com> | 4 | callers. |
9 | Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> | 5 | |
10 | Message-id: 20190507203508.18026-2-mreitz@redhat.com | 6 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> |
11 | Signed-off-by: Max Reitz <mreitz@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: 20170213135235.12274-18-pbonzini@redhat.com | ||
11 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
12 | --- | 12 | --- |
13 | qemu-img.c | 13 +++++++------ | 13 | util/aio-posix.c | 8 +++----- |
14 | 1 file changed, 7 insertions(+), 6 deletions(-) | 14 | util/aio-win32.c | 8 ++++---- |
15 | util/async.c | 12 ++++++------ | ||
16 | 3 files changed, 13 insertions(+), 15 deletions(-) | ||
15 | 17 | ||
16 | diff --git a/qemu-img.c b/qemu-img.c | 18 | diff --git a/util/aio-posix.c b/util/aio-posix.c |
17 | index XXXXXXX..XXXXXXX 100644 | 19 | index XXXXXXX..XXXXXXX 100644 |
18 | --- a/qemu-img.c | 20 | --- a/util/aio-posix.c |
19 | +++ b/qemu-img.c | 21 | +++ b/util/aio-posix.c |
20 | @@ -XXX,XX +XXX,XX @@ typedef struct ImgConvertState { | 22 | @@ -XXX,XX +XXX,XX @@ static bool aio_dispatch_handlers(AioContext *ctx) |
21 | int64_t target_backing_sectors; /* negative if unknown */ | 23 | |
22 | bool wr_in_order; | 24 | void aio_dispatch(AioContext *ctx) |
23 | bool copy_range; | 25 | { |
24 | + bool quiet; | 26 | + qemu_lockcnt_inc(&ctx->list_lock); |
25 | int min_sparse; | 27 | aio_bh_poll(ctx); |
26 | int alignment; | 28 | - |
27 | size_t cluster_sectors; | 29 | - qemu_lockcnt_inc(&ctx->list_lock); |
28 | @@ -XXX,XX +XXX,XX @@ static int img_convert(int argc, char **argv) | 30 | aio_dispatch_handlers(ctx); |
29 | QDict *open_opts = NULL; | 31 | qemu_lockcnt_dec(&ctx->list_lock); |
30 | char *options = NULL; | 32 | |
31 | Error *local_err = NULL; | 33 | @@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking) |
32 | - bool writethrough, src_writethrough, quiet = false, image_opts = false, | ||
33 | + bool writethrough, src_writethrough, image_opts = false, | ||
34 | skip_create = false, progress = false, tgt_image_opts = false; | ||
35 | int64_t ret = -EINVAL; | ||
36 | bool force_share = false; | ||
37 | @@ -XXX,XX +XXX,XX @@ static int img_convert(int argc, char **argv) | ||
38 | src_cache = optarg; | ||
39 | break; | ||
40 | case 'q': | ||
41 | - quiet = true; | ||
42 | + s.quiet = true; | ||
43 | break; | ||
44 | case 'n': | ||
45 | skip_create = true; | ||
46 | @@ -XXX,XX +XXX,XX @@ static int img_convert(int argc, char **argv) | ||
47 | } | 34 | } |
48 | 35 | ||
49 | /* Initialize before goto out */ | 36 | npfd = 0; |
50 | - if (quiet) { | 37 | - qemu_lockcnt_dec(&ctx->list_lock); |
51 | + if (s.quiet) { | 38 | |
52 | progress = false; | 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); | ||
53 | } | 45 | } |
54 | qemu_progress_init(progress, 1.0); | 46 | |
55 | @@ -XXX,XX +XXX,XX @@ static int img_convert(int argc, char **argv) | 47 | + qemu_lockcnt_dec(&ctx->list_lock); |
56 | 48 | + | |
57 | for (bs_i = 0; bs_i < s.src_num; bs_i++) { | 49 | progress |= timerlistgroup_run_timers(&ctx->tlg); |
58 | s.src[bs_i] = img_open(image_opts, argv[optind + bs_i], | 50 | |
59 | - fmt, src_flags, src_writethrough, quiet, | 51 | return progress; |
60 | + fmt, src_flags, src_writethrough, s.quiet, | 52 | diff --git a/util/aio-win32.c b/util/aio-win32.c |
61 | force_share); | 53 | index XXXXXXX..XXXXXXX 100644 |
62 | if (!s.src[bs_i]) { | 54 | --- a/util/aio-win32.c |
63 | ret = -1; | 55 | +++ b/util/aio-win32.c |
64 | @@ -XXX,XX +XXX,XX @@ static int img_convert(int argc, char **argv) | 56 | @@ -XXX,XX +XXX,XX @@ static bool aio_dispatch_handlers(AioContext *ctx, HANDLE event) |
65 | 57 | bool progress = false; | |
66 | if (skip_create) { | 58 | AioHandler *tmp; |
67 | s.target = img_open(tgt_image_opts, out_filename, out_fmt, | 59 | |
68 | - flags, writethrough, quiet, false); | 60 | - qemu_lockcnt_inc(&ctx->list_lock); |
69 | + flags, writethrough, s.quiet, false); | 61 | - |
70 | } else { | 62 | /* |
71 | /* TODO ultimately we should allow --target-image-opts | 63 | * We have to walk very carefully in case aio_set_fd_handler is |
72 | * to be used even when -n is not given. | 64 | * called while we're walking. |
73 | @@ -XXX,XX +XXX,XX @@ static int img_convert(int argc, char **argv) | 65 | @@ -XXX,XX +XXX,XX @@ static bool aio_dispatch_handlers(AioContext *ctx, HANDLE event) |
74 | * to allow filenames in option syntax | 66 | } |
75 | */ | ||
76 | s.target = img_open_file(out_filename, open_opts, out_fmt, | ||
77 | - flags, writethrough, quiet, false); | ||
78 | + flags, writethrough, s.quiet, false); | ||
79 | open_opts = NULL; /* blk_new_open will have freed it */ | ||
80 | } | 67 | } |
81 | if (!s.target) { | 68 | |
69 | - qemu_lockcnt_dec(&ctx->list_lock); | ||
70 | return progress; | ||
71 | } | ||
72 | |||
73 | void aio_dispatch(AioContext *ctx) | ||
74 | { | ||
75 | + qemu_lockcnt_inc(&ctx->list_lock); | ||
76 | aio_bh_poll(ctx); | ||
77 | aio_dispatch_handlers(ctx, INVALID_HANDLE_VALUE); | ||
78 | + qemu_lockcnt_dec(&ctx->list_lock); | ||
79 | timerlistgroup_run_timers(&ctx->tlg); | ||
80 | } | ||
81 | |||
82 | @@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking) | ||
83 | } | ||
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) | ||
113 | { | ||
114 | QEMUBH *bh, **bhp, *next; | ||
115 | int ret; | ||
116 | bool deleted = false; | ||
117 | |||
118 | - qemu_lockcnt_inc(&ctx->list_lock); | ||
119 | - | ||
120 | ret = 0; | ||
121 | for (bh = atomic_rcu_read(&ctx->first_bh); bh; bh = next) { | ||
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; | ||
129 | } | ||
130 | |||
131 | - if (qemu_lockcnt_dec_and_lock(&ctx->list_lock)) { | ||
132 | + if (qemu_lockcnt_dec_if_lock(&ctx->list_lock)) { | ||
133 | bhp = &ctx->first_bh; | ||
134 | while (*bhp) { | ||
135 | bh = *bhp; | ||
136 | @@ -XXX,XX +XXX,XX @@ int aio_bh_poll(AioContext *ctx) | ||
137 | bhp = &bh->next; | ||
138 | } | ||
139 | } | ||
140 | - qemu_lockcnt_unlock(&ctx->list_lock); | ||
141 | + qemu_lockcnt_inc_and_unlock(&ctx->list_lock); | ||
142 | } | ||
143 | return ret; | ||
144 | } | ||
82 | -- | 145 | -- |
83 | 2.21.0 | 146 | 2.9.3 |
84 | 147 | ||
85 | 148 | diff view generated by jsdifflib |
1 | From: John Snow <jsnow@redhat.com> | 1 | From: Paolo Bonzini <pbonzini@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | Don't pull events out of the queue that don't belong to us; | 3 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> |
4 | be choosier so that we can use this method to drive jobs that | 4 | Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> |
5 | were launched by transactions that may have more jobs. | 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(-) | ||
6 | 13 | ||
7 | Signed-off-by: John Snow <jsnow@redhat.com> | 14 | diff --git a/include/block/block_int.h b/include/block/block_int.h |
8 | Message-id: 20190523170643.20794-5-jsnow@redhat.com | ||
9 | Reviewed-by: Max Reitz <mreitz@redhat.com> | ||
10 | Signed-off-by: Max Reitz <mreitz@redhat.com> | ||
11 | --- | ||
12 | tests/qemu-iotests/iotests.py | 48 +++++++++++++++++++++-------------- | ||
13 | 1 file changed, 29 insertions(+), 19 deletions(-) | ||
14 | |||
15 | diff --git a/tests/qemu-iotests/iotests.py b/tests/qemu-iotests/iotests.py | ||
16 | index XXXXXXX..XXXXXXX 100644 | 15 | index XXXXXXX..XXXXXXX 100644 |
17 | --- a/tests/qemu-iotests/iotests.py | 16 | --- a/include/block/block_int.h |
18 | +++ b/tests/qemu-iotests/iotests.py | 17 | +++ b/include/block/block_int.h |
19 | @@ -XXX,XX +XXX,XX @@ class VM(qtest.QEMUQtestMachine): | 18 | @@ -XXX,XX +XXX,XX @@ struct BdrvChild { |
20 | # Returns None on success, and an error string on failure | 19 | * copied as well. |
21 | def run_job(self, job, auto_finalize=True, auto_dismiss=False, | 20 | */ |
22 | pre_finalize=None, wait=60.0): | 21 | struct BlockDriverState { |
23 | + match_device = {'data': {'device': job}} | 22 | - int64_t total_sectors; /* if we are reading a disk image, give its |
24 | + match_id = {'data': {'id': job}} | 23 | - size in sectors */ |
25 | + events = [ | 24 | + /* Protected by big QEMU lock or read-only after opening. No special |
26 | + ('BLOCK_JOB_COMPLETED', match_device), | 25 | + * locking needed during I/O... |
27 | + ('BLOCK_JOB_CANCELLED', match_device), | 26 | + */ |
28 | + ('BLOCK_JOB_ERROR', match_device), | 27 | int open_flags; /* flags used to open the file, re-used for re-open */ |
29 | + ('BLOCK_JOB_READY', match_device), | 28 | bool read_only; /* if true, the media is read only */ |
30 | + ('BLOCK_JOB_PENDING', match_id), | 29 | bool encrypted; /* if true, the media is encrypted */ |
31 | + ('JOB_STATUS_CHANGE', match_id) | 30 | @@ -XXX,XX +XXX,XX @@ struct BlockDriverState { |
32 | + ] | 31 | bool sg; /* if true, the device is a /dev/sg* */ |
33 | error = None | 32 | bool probed; /* if true, format was probed rather than specified */ |
34 | while True: | 33 | |
35 | - for ev in self.get_qmp_events_filtered(wait=wait): | 34 | - int copy_on_read; /* if nonzero, copy read backing sectors into image. |
36 | - if ev['event'] == 'JOB_STATUS_CHANGE': | 35 | - note this is a reference count */ |
37 | - status = ev['data']['status'] | 36 | - |
38 | - if status == 'aborting': | 37 | - CoQueue flush_queue; /* Serializing flush queue */ |
39 | - result = self.qmp('query-jobs') | 38 | - bool active_flush_req; /* Flush request in flight? */ |
40 | - for j in result['return']: | 39 | - unsigned int write_gen; /* Current data generation */ |
41 | - if j['id'] == job: | 40 | - unsigned int flushed_gen; /* Flushed write generation */ |
42 | - error = j['error'] | 41 | - |
43 | - log('Job failed: %s' % (j['error'])) | 42 | BlockDriver *drv; /* NULL means no media */ |
44 | - elif status == 'pending' and not auto_finalize: | 43 | void *opaque; |
45 | - if pre_finalize: | 44 | |
46 | - pre_finalize() | 45 | @@ -XXX,XX +XXX,XX @@ struct BlockDriverState { |
47 | - self.qmp_log('job-finalize', id=job) | 46 | BdrvChild *backing; |
48 | - elif status == 'concluded' and not auto_dismiss: | 47 | BdrvChild *file; |
49 | - self.qmp_log('job-dismiss', id=job) | 48 | |
50 | - elif status == 'null': | 49 | - /* Callback before write request is processed */ |
51 | - return error | 50 | - NotifierWithReturnList before_write_notifiers; |
52 | - else: | 51 | - |
53 | - log(ev) | 52 | - /* number of in-flight requests; overall and serialising */ |
54 | + ev = filter_qmp_event(self.events_wait(events)) | 53 | - unsigned int in_flight; |
55 | + if ev['event'] != 'JOB_STATUS_CHANGE': | 54 | - unsigned int serialising_in_flight; |
56 | + log(ev) | 55 | - |
57 | + continue | 56 | - bool wakeup; |
58 | + status = ev['data']['status'] | 57 | - |
59 | + if status == 'aborting': | 58 | - /* Offset after the highest byte written to */ |
60 | + result = self.qmp('query-jobs') | 59 | - uint64_t wr_highest_offset; |
61 | + for j in result['return']: | 60 | - |
62 | + if j['id'] == job: | 61 | /* I/O Limits */ |
63 | + error = j['error'] | 62 | BlockLimits bl; |
64 | + log('Job failed: %s' % (j['error'])) | 63 | |
65 | + elif status == 'pending' and not auto_finalize: | 64 | @@ -XXX,XX +XXX,XX @@ struct BlockDriverState { |
66 | + if pre_finalize: | 65 | QTAILQ_ENTRY(BlockDriverState) bs_list; |
67 | + pre_finalize() | 66 | /* element of the list of monitor-owned BDS */ |
68 | + self.qmp_log('job-finalize', id=job) | 67 | QTAILQ_ENTRY(BlockDriverState) monitor_list; |
69 | + elif status == 'concluded' and not auto_dismiss: | 68 | - QLIST_HEAD(, BdrvDirtyBitmap) dirty_bitmaps; |
70 | + self.qmp_log('job-dismiss', id=job) | 69 | int refcnt; |
71 | + elif status == 'null': | 70 | |
72 | + return error | 71 | - QLIST_HEAD(, BdrvTrackedRequest) tracked_requests; |
73 | 72 | - | |
74 | def node_info(self, node_name): | 73 | /* operation blockers */ |
75 | nodes = self.qmp('query-named-block-nodes') | 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]; | ||
76 | -- | 155 | -- |
77 | 2.21.0 | 156 | 2.9.3 |
78 | 157 | ||
79 | 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 |
1 | This adds a salvaging mode (--salvage) to qemu-img convert which ignores | 1 | From: Paolo Bonzini <pbonzini@redhat.com> |
---|---|---|---|
2 | read errors and treats the respective areas as containing only zeroes. | ||
3 | This can be used for instance to at least partially recover the data | ||
4 | from terminally corrupted qcow2 images. | ||
5 | 2 | ||
6 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 3 | Running a very small critical section on pthread_mutex_t and CoMutex |
7 | Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> | 4 | shows that pthread_mutex_t is much faster because it doesn't actually |
8 | Message-id: 20190507203508.18026-3-mreitz@redhat.com | 5 | go to sleep. What happens is that the critical section is shorter |
9 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 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> | ||
10 | --- | 17 | --- |
11 | qemu-img.c | 90 +++++++++++++++++++++++++++++++++++++----------- | 18 | include/qemu/coroutine.h | 5 +++++ |
12 | qemu-img-cmds.hx | 4 +-- | 19 | util/qemu-coroutine-lock.c | 51 ++++++++++++++++++++++++++++++++++++++++------ |
13 | qemu-img.texi | 4 +++ | 20 | util/qemu-coroutine.c | 2 +- |
14 | 3 files changed, 75 insertions(+), 23 deletions(-) | 21 | 3 files changed, 51 insertions(+), 7 deletions(-) |
15 | 22 | ||
16 | diff --git a/qemu-img.c b/qemu-img.c | 23 | diff --git a/include/qemu/coroutine.h b/include/qemu/coroutine.h |
17 | index XXXXXXX..XXXXXXX 100644 | 24 | index XXXXXXX..XXXXXXX 100644 |
18 | --- a/qemu-img.c | 25 | --- a/include/qemu/coroutine.h |
19 | +++ b/qemu-img.c | 26 | +++ b/include/qemu/coroutine.h |
20 | @@ -XXX,XX +XXX,XX @@ enum { | 27 | @@ -XXX,XX +XXX,XX @@ typedef struct CoMutex { |
21 | OPTION_SIZE = 264, | 28 | */ |
22 | OPTION_PREALLOCATION = 265, | 29 | unsigned locked; |
23 | OPTION_SHRINK = 266, | 30 | |
24 | + OPTION_SALVAGE = 267, | 31 | + /* Context that is holding the lock. Useful to avoid spinning |
25 | }; | 32 | + * when two coroutines on the same AioContext try to get the lock. :) |
26 | 33 | + */ | |
27 | typedef enum OutputFormat { | 34 | + AioContext *ctx; |
28 | @@ -XXX,XX +XXX,XX @@ typedef struct ImgConvertState { | 35 | + |
29 | int64_t target_backing_sectors; /* negative if unknown */ | 36 | /* A queue of waiters. Elements are added atomically in front of |
30 | bool wr_in_order; | 37 | * from_push. to_pop is only populated, and popped from, by whoever |
31 | bool copy_range; | 38 | * is in charge of the next wakeup. This can be an unlocker or, |
32 | + bool salvage; | 39 | diff --git a/util/qemu-coroutine-lock.c b/util/qemu-coroutine-lock.c |
33 | bool quiet; | 40 | index XXXXXXX..XXXXXXX 100644 |
34 | int min_sparse; | 41 | --- a/util/qemu-coroutine-lock.c |
35 | int alignment; | 42 | +++ b/util/qemu-coroutine-lock.c |
36 | @@ -XXX,XX +XXX,XX @@ static int convert_iteration_sectors(ImgConvertState *s, int64_t sector_num) | 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); | ||
37 | } | 81 | } |
38 | 82 | ||
39 | if (s->sector_next_status <= sector_num) { | 83 | qemu_coroutine_yield(); |
40 | - int64_t count = n * BDRV_SECTOR_SIZE; | 84 | @@ -XXX,XX +XXX,XX @@ static void coroutine_fn qemu_co_mutex_lock_slowpath(CoMutex *mutex) |
41 | + uint64_t offset = (sector_num - src_cur_offset) * BDRV_SECTOR_SIZE; | 85 | |
42 | + int64_t count; | 86 | void coroutine_fn qemu_co_mutex_lock(CoMutex *mutex) |
43 | 87 | { | |
44 | - if (s->target_has_backing) { | 88 | + AioContext *ctx = qemu_get_current_aio_context(); |
45 | + do { | 89 | Coroutine *self = qemu_coroutine_self(); |
46 | + count = n * BDRV_SECTOR_SIZE; | 90 | + int waiters, i; |
47 | + | 91 | |
48 | + if (s->target_has_backing) { | 92 | - if (atomic_fetch_inc(&mutex->locked) == 0) { |
49 | + ret = bdrv_block_status(blk_bs(s->src[src_cur]), offset, | 93 | + /* Running a very small critical section on pthread_mutex_t and CoMutex |
50 | + count, &count, NULL, NULL); | 94 | + * shows that pthread_mutex_t is much faster because it doesn't actually |
51 | + } else { | 95 | + * go to sleep. What happens is that the critical section is shorter |
52 | + ret = bdrv_block_status_above(blk_bs(s->src[src_cur]), NULL, | 96 | + * than the latency of entering the kernel and thus FUTEX_WAIT always |
53 | + offset, count, &count, NULL, | 97 | + * fails. With CoMutex there is no such latency but you still want to |
54 | + NULL); | 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; | ||
55 | + } | 107 | + } |
56 | + | 108 | + if (atomic_read(&mutex->locked) == 0) { |
57 | + if (ret < 0) { | 109 | + goto retry_fast_path; |
58 | + if (s->salvage) { | ||
59 | + if (n == 1) { | ||
60 | + if (!s->quiet) { | ||
61 | + warn_report("error while reading block status at " | ||
62 | + "offset %" PRIu64 ": %s", offset, | ||
63 | + strerror(-ret)); | ||
64 | + } | ||
65 | + /* Just try to read the data, then */ | ||
66 | + ret = BDRV_BLOCK_DATA; | ||
67 | + count = BDRV_SECTOR_SIZE; | ||
68 | + } else { | ||
69 | + /* Retry on a shorter range */ | ||
70 | + n = DIV_ROUND_UP(n, 4); | ||
71 | + } | ||
72 | + } else { | ||
73 | + error_report("error while reading block status at offset " | ||
74 | + "%" PRIu64 ": %s", offset, strerror(-ret)); | ||
75 | + return ret; | ||
76 | + } | ||
77 | + } | 110 | + } |
78 | + } while (ret < 0); | 111 | + cpu_relax(); |
79 | |||
80 | - ret = bdrv_block_status(blk_bs(s->src[src_cur]), | ||
81 | - (sector_num - src_cur_offset) * | ||
82 | - BDRV_SECTOR_SIZE, | ||
83 | - count, &count, NULL, NULL); | ||
84 | - } else { | ||
85 | - ret = bdrv_block_status_above(blk_bs(s->src[src_cur]), NULL, | ||
86 | - (sector_num - src_cur_offset) * | ||
87 | - BDRV_SECTOR_SIZE, | ||
88 | - count, &count, NULL, NULL); | ||
89 | - } | ||
90 | - if (ret < 0) { | ||
91 | - error_report("error while reading block status of sector %" PRId64 | ||
92 | - ": %s", sector_num, strerror(-ret)); | ||
93 | - return ret; | ||
94 | - } | ||
95 | n = DIV_ROUND_UP(count, BDRV_SECTOR_SIZE); | ||
96 | |||
97 | if (ret & BDRV_BLOCK_ZERO) { | ||
98 | @@ -XXX,XX +XXX,XX @@ static int convert_iteration_sectors(ImgConvertState *s, int64_t sector_num) | ||
99 | static int coroutine_fn convert_co_read(ImgConvertState *s, int64_t sector_num, | ||
100 | int nb_sectors, uint8_t *buf) | ||
101 | { | ||
102 | + uint64_t single_read_until = 0; | ||
103 | int n, ret; | ||
104 | |||
105 | assert(nb_sectors <= s->buf_sectors); | ||
106 | @@ -XXX,XX +XXX,XX @@ static int coroutine_fn convert_co_read(ImgConvertState *s, int64_t sector_num, | ||
107 | BlockBackend *blk; | ||
108 | int src_cur; | ||
109 | int64_t bs_sectors, src_cur_offset; | ||
110 | + uint64_t offset; | ||
111 | |||
112 | /* In the case of compression with multiple source files, we can get a | ||
113 | * nb_sectors that spreads into the next part. So we must be able to | ||
114 | @@ -XXX,XX +XXX,XX @@ static int coroutine_fn convert_co_read(ImgConvertState *s, int64_t sector_num, | ||
115 | blk = s->src[src_cur]; | ||
116 | bs_sectors = s->src_sectors[src_cur]; | ||
117 | |||
118 | + offset = (sector_num - src_cur_offset) << BDRV_SECTOR_BITS; | ||
119 | + | ||
120 | n = MIN(nb_sectors, bs_sectors - (sector_num - src_cur_offset)); | ||
121 | + if (single_read_until > offset) { | ||
122 | + n = 1; | ||
123 | + } | 112 | + } |
124 | 113 | + waiters = atomic_fetch_inc(&mutex->locked); | |
125 | - ret = blk_co_pread( | ||
126 | - blk, (sector_num - src_cur_offset) << BDRV_SECTOR_BITS, | ||
127 | - n << BDRV_SECTOR_BITS, buf, 0); | ||
128 | + ret = blk_co_pread(blk, offset, n << BDRV_SECTOR_BITS, buf, 0); | ||
129 | if (ret < 0) { | ||
130 | - return ret; | ||
131 | + if (s->salvage) { | ||
132 | + if (n > 1) { | ||
133 | + single_read_until = offset + (n << BDRV_SECTOR_BITS); | ||
134 | + continue; | ||
135 | + } else { | ||
136 | + if (!s->quiet) { | ||
137 | + warn_report("error while reading offset %" PRIu64 | ||
138 | + ": %s", offset, strerror(-ret)); | ||
139 | + } | ||
140 | + memset(buf, 0, BDRV_SECTOR_SIZE); | ||
141 | + } | ||
142 | + } else { | ||
143 | + return ret; | ||
144 | + } | ||
145 | } | ||
146 | |||
147 | sector_num += n; | ||
148 | @@ -XXX,XX +XXX,XX @@ static int img_convert(int argc, char **argv) | ||
149 | {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS}, | ||
150 | {"force-share", no_argument, 0, 'U'}, | ||
151 | {"target-image-opts", no_argument, 0, OPTION_TARGET_IMAGE_OPTS}, | ||
152 | + {"salvage", no_argument, 0, OPTION_SALVAGE}, | ||
153 | {0, 0, 0, 0} | ||
154 | }; | ||
155 | c = getopt_long(argc, argv, ":hf:O:B:Cco:l:S:pt:T:qnm:WU", | ||
156 | @@ -XXX,XX +XXX,XX @@ static int img_convert(int argc, char **argv) | ||
157 | case OPTION_IMAGE_OPTS: | ||
158 | image_opts = true; | ||
159 | break; | ||
160 | + case OPTION_SALVAGE: | ||
161 | + s.salvage = true; | ||
162 | + break; | ||
163 | case OPTION_TARGET_IMAGE_OPTS: | ||
164 | tgt_image_opts = true; | ||
165 | break; | ||
166 | @@ -XXX,XX +XXX,XX @@ static int img_convert(int argc, char **argv) | ||
167 | goto fail_getopt; | ||
168 | } | ||
169 | |||
170 | + if (s.copy_range && s.salvage) { | ||
171 | + error_report("Cannot use copy offloading in salvaging mode"); | ||
172 | + goto fail_getopt; | ||
173 | + } | 114 | + } |
174 | + | 115 | + |
175 | if (tgt_image_opts && !skip_create) { | 116 | + if (waiters == 0) { |
176 | error_report("--target-image-opts requires use of -n flag"); | 117 | /* Uncontended. */ |
177 | goto fail_getopt; | 118 | trace_qemu_co_mutex_lock_uncontended(mutex, self); |
178 | diff --git a/qemu-img-cmds.hx b/qemu-img-cmds.hx | 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 | ||
179 | index XXXXXXX..XXXXXXX 100644 | 145 | index XXXXXXX..XXXXXXX 100644 |
180 | --- a/qemu-img-cmds.hx | 146 | --- a/util/qemu-coroutine.c |
181 | +++ b/qemu-img-cmds.hx | 147 | +++ b/util/qemu-coroutine.c |
182 | @@ -XXX,XX +XXX,XX @@ STEXI | 148 | @@ -XXX,XX +XXX,XX @@ void qemu_coroutine_enter(Coroutine *co) |
183 | ETEXI | 149 | co->ctx = qemu_get_current_aio_context(); |
184 | 150 | ||
185 | DEF("convert", img_convert, | 151 | /* Store co->ctx before anything that stores co. Matches |
186 | - "convert [--object objectdef] [--image-opts] [--target-image-opts] [-U] [-C] [-c] [-p] [-q] [-n] [-f fmt] [-t cache] [-T src_cache] [-O output_fmt] [-B backing_file] [-o options] [-l snapshot_param] [-S sparse_size] [-m num_coroutines] [-W] filename [filename2 [...]] output_filename") | 152 | - * barrier in aio_co_wake. |
187 | + "convert [--object objectdef] [--image-opts] [--target-image-opts] [-U] [-C] [-c] [-p] [-q] [-n] [-f fmt] [-t cache] [-T src_cache] [-O output_fmt] [-B backing_file] [-o options] [-l snapshot_param] [-S sparse_size] [-m num_coroutines] [-W] [--salvage] filename [filename2 [...]] output_filename") | 153 | + * barrier in aio_co_wake and qemu_co_mutex_wake. |
188 | STEXI | 154 | */ |
189 | -@item convert [--object @var{objectdef}] [--image-opts] [--target-image-opts] [-U] [-C] [-c] [-p] [-q] [-n] [-f @var{fmt}] [-t @var{cache}] [-T @var{src_cache}] [-O @var{output_fmt}] [-B @var{backing_file}] [-o @var{options}] [-l @var{snapshot_param}] [-S @var{sparse_size}] [-m @var{num_coroutines}] [-W] @var{filename} [@var{filename2} [...]] @var{output_filename} | 155 | smp_wmb(); |
190 | +@item convert [--object @var{objectdef}] [--image-opts] [--target-image-opts] [-U] [-C] [-c] [-p] [-q] [-n] [-f @var{fmt}] [-t @var{cache}] [-T @var{src_cache}] [-O @var{output_fmt}] [-B @var{backing_file}] [-o @var{options}] [-l @var{snapshot_param}] [-S @var{sparse_size}] [-m @var{num_coroutines}] [-W] [--salvage] @var{filename} [@var{filename2} [...]] @var{output_filename} | 156 | |
191 | ETEXI | ||
192 | |||
193 | DEF("create", img_create, | ||
194 | diff --git a/qemu-img.texi b/qemu-img.texi | ||
195 | index XXXXXXX..XXXXXXX 100644 | ||
196 | --- a/qemu-img.texi | ||
197 | +++ b/qemu-img.texi | ||
198 | @@ -XXX,XX +XXX,XX @@ improve performance if the data is remote, such as with NFS or iSCSI backends, | ||
199 | but will not automatically sparsify zero sectors, and may result in a fully | ||
200 | allocated target image depending on the host support for getting allocation | ||
201 | information. | ||
202 | +@item --salvage | ||
203 | +Try to ignore I/O errors when reading. Unless in quiet mode (@code{-q}), errors | ||
204 | +will still be printed. Areas that cannot be read from the source will be | ||
205 | +treated as containing only zeroes. | ||
206 | @end table | ||
207 | |||
208 | Parameters to dd subcommand: | ||
209 | -- | 157 | -- |
210 | 2.21.0 | 158 | 2.9.3 |
211 | 159 | ||
212 | 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 |
1 | This test converts a simple image to another, but blkdebug injects | 1 | From: Paolo Bonzini <pbonzini@redhat.com> |
---|---|---|---|
2 | block_status and read faults at some offsets. The resulting image | ||
3 | should be the same as the input image, except that sectors that could | ||
4 | not be read have to be 0. | ||
5 | 2 | ||
6 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 3 | This will avoid forward references in the next patch. It is also |
7 | Message-id: 20190507203508.18026-7-mreitz@redhat.com | 4 | more logical because CoQueue is not anymore the basic primitive. |
8 | Tested-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> | 5 | |
9 | Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> | 6 | Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> |
10 | [mreitz: Dropped superfluous printf from _filter_offsets, as suggested | 7 | Reviewed-by: Fam Zheng <famz@redhat.com> |
11 | by Vladimir; disable test for VDI and IMGOPTSSYNTAX] | 8 | Message-id: 20170213181244.16297-5-pbonzini@redhat.com |
12 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 9 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> |
13 | --- | 10 | --- |
14 | tests/qemu-iotests/251 | 170 +++++++++++++++++++++++++++++++++++++ | 11 | include/qemu/coroutine.h | 89 ++++++++++++++++++++++++------------------------ |
15 | tests/qemu-iotests/251.out | 43 ++++++++++ | 12 | 1 file changed, 44 insertions(+), 45 deletions(-) |
16 | tests/qemu-iotests/group | 1 + | ||
17 | 3 files changed, 214 insertions(+) | ||
18 | create mode 100755 tests/qemu-iotests/251 | ||
19 | create mode 100644 tests/qemu-iotests/251.out | ||
20 | 13 | ||
21 | diff --git a/tests/qemu-iotests/251 b/tests/qemu-iotests/251 | 14 | diff --git a/include/qemu/coroutine.h b/include/qemu/coroutine.h |
22 | new file mode 100755 | 15 | index XXXXXXX..XXXXXXX 100644 |
23 | index XXXXXXX..XXXXXXX | 16 | --- a/include/qemu/coroutine.h |
24 | --- /dev/null | 17 | +++ b/include/qemu/coroutine.h |
25 | +++ b/tests/qemu-iotests/251 | 18 | @@ -XXX,XX +XXX,XX @@ bool qemu_in_coroutine(void); |
26 | @@ -XXX,XX +XXX,XX @@ | 19 | */ |
27 | +#!/usr/bin/env bash | 20 | bool qemu_coroutine_entered(Coroutine *co); |
28 | +# | 21 | |
29 | +# Test qemu-img convert --salvage | 22 | - |
30 | +# | 23 | -/** |
31 | +# Copyright (C) 2019 Red Hat, Inc. | 24 | - * CoQueues are a mechanism to queue coroutines in order to continue executing |
32 | +# | 25 | - * them later. They provide the fundamental primitives on which coroutine locks |
33 | +# This program is free software; you can redistribute it and/or modify | 26 | - * are built. |
34 | +# it under the terms of the GNU General Public License as published by | 27 | - */ |
35 | +# the Free Software Foundation; either version 2 of the License, or | 28 | -typedef struct CoQueue { |
36 | +# (at your option) any later version. | 29 | - QSIMPLEQ_HEAD(, Coroutine) entries; |
37 | +# | 30 | -} CoQueue; |
38 | +# This program is distributed in the hope that it will be useful, | 31 | - |
39 | +# but WITHOUT ANY WARRANTY; without even the implied warranty of | 32 | -/** |
40 | +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | 33 | - * Initialise a CoQueue. This must be called before any other operation is used |
41 | +# GNU General Public License for more details. | 34 | - * on the CoQueue. |
42 | +# | 35 | - */ |
43 | +# You should have received a copy of the GNU General Public License | 36 | -void qemu_co_queue_init(CoQueue *queue); |
44 | +# along with this program. If not, see <http://www.gnu.org/licenses/>. | 37 | - |
45 | +# | 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 | |||
46 | + | 74 | + |
47 | +# creator | 75 | +/** |
48 | +owner=mreitz@redhat.com | 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; | ||
49 | + | 82 | + |
50 | +seq=$(basename $0) | 83 | +/** |
51 | +echo "QA output created by $seq" | 84 | + * Initialise a CoQueue. This must be called before any other operation is used |
85 | + * on the CoQueue. | ||
86 | + */ | ||
87 | +void qemu_co_queue_init(CoQueue *queue); | ||
52 | + | 88 | + |
53 | +status=1 # failure is the default! | 89 | +/** |
90 | + * Adds the current coroutine to the CoQueue and transfers control to the | ||
91 | + * caller of the coroutine. | ||
92 | + */ | ||
93 | +void coroutine_fn qemu_co_queue_wait(CoQueue *queue); | ||
54 | + | 94 | + |
55 | +_cleanup() | 95 | +/** |
56 | +{ | 96 | + * Restarts the next coroutine in the CoQueue and removes it from the queue. |
57 | + _cleanup_test_img | 97 | + * |
58 | +} | 98 | + * Returns true if a coroutine was restarted, false if the queue is empty. |
59 | +trap "_cleanup; exit \$status" 0 1 2 3 15 | 99 | + */ |
100 | +bool coroutine_fn qemu_co_queue_next(CoQueue *queue); | ||
60 | + | 101 | + |
61 | +# get standard environment, filters and checks | 102 | +/** |
62 | +. ./common.rc | 103 | + * Restarts all coroutines in the CoQueue and leaves the queue empty. |
63 | +. ./common.filter | 104 | + */ |
64 | +. ./common.qemu | 105 | +void coroutine_fn qemu_co_queue_restart_all(CoQueue *queue); |
65 | + | 106 | + |
66 | +_supported_fmt generic | 107 | +/** |
67 | +_supported_proto file | 108 | + * Enter the next coroutine in the queue |
68 | +_supported_os Linux | 109 | + */ |
110 | +bool qemu_co_enter_next(CoQueue *queue); | ||
69 | + | 111 | + |
70 | +if [ "$IMGOPTSSYNTAX" = "true" ]; then | 112 | +/** |
71 | + # We use json:{} filenames here, so we cannot work with additional options. | 113 | + * Checks if the CoQueue is empty. |
72 | + _unsupported_fmt $IMGFMT | 114 | + */ |
73 | +else | 115 | +bool qemu_co_queue_empty(CoQueue *queue); |
74 | + # With VDI, the output is ordered differently. Just disable it. | ||
75 | + _unsupported_fmt vdi | ||
76 | +fi | ||
77 | + | 116 | + |
78 | + | 117 | + |
79 | +TEST_IMG="$TEST_IMG.orig" _make_test_img 64M | 118 | typedef struct CoRwlock { |
80 | + | 119 | bool writer; |
81 | +$QEMU_IO -c 'write -P 42 0 64M' "$TEST_IMG.orig" | _filter_qemu_io | 120 | int reader; |
82 | + | ||
83 | + | ||
84 | +sector_size=512 | ||
85 | + | ||
86 | +# Offsets on which to fail block-status. Keep in ascending order so | ||
87 | +# the indexing done by _filter_offsets will appear in ascending order | ||
88 | +# in the output as well. | ||
89 | +status_fail_offsets="$((16 * 1024 * 1024 + 8192)) | ||
90 | + $((33 * 1024 * 1024 + 512))" | ||
91 | + | ||
92 | +# Offsets on which to fail reads. Keep in ascending order for the | ||
93 | +# same reason. | ||
94 | +# The second element is shared with $status_fail_offsets on purpose. | ||
95 | +# Starting with the third element, we test what happens when a | ||
96 | +# continuous range of sectors is inaccessible. | ||
97 | +read_fail_offsets="$((32 * 1024 * 1024 - 65536)) | ||
98 | + $((33 * 1024 * 1024 + 512)) | ||
99 | + $(seq $((34 * 1024 * 1024)) $sector_size \ | ||
100 | + $((34 * 1024 * 1024 + 4096 - $sector_size)))" | ||
101 | + | ||
102 | + | ||
103 | +# blkdebug must be above the format layer so it can intercept all | ||
104 | +# block-status events | ||
105 | +source_img="json:{'driver': 'blkdebug', | ||
106 | + 'image': { | ||
107 | + 'driver': '$IMGFMT', | ||
108 | + 'file': { | ||
109 | + 'driver': 'file', | ||
110 | + 'filename': '$TEST_IMG.orig' | ||
111 | + } | ||
112 | + }, | ||
113 | + 'inject-error': [" | ||
114 | + | ||
115 | +for ofs in $status_fail_offsets | ||
116 | +do | ||
117 | + source_img+="{ 'event': 'none', | ||
118 | + 'iotype': 'block-status', | ||
119 | + 'errno': 5, | ||
120 | + 'sector': $((ofs / sector_size)) }," | ||
121 | +done | ||
122 | + | ||
123 | +for ofs in $read_fail_offsets | ||
124 | +do | ||
125 | + source_img+="{ 'event': 'none', | ||
126 | + 'iotype': 'read', | ||
127 | + 'errno': 5, | ||
128 | + 'sector': $((ofs / sector_size)) }," | ||
129 | +done | ||
130 | + | ||
131 | +# Remove the trailing comma and terminate @inject-error and json:{} | ||
132 | +source_img="${source_img%,} ] }" | ||
133 | + | ||
134 | + | ||
135 | +echo | ||
136 | + | ||
137 | + | ||
138 | +_filter_offsets() { | ||
139 | + filters= | ||
140 | + | ||
141 | + index=0 | ||
142 | + for ofs in $1 | ||
143 | + do | ||
144 | + filters+=" -e s/$ofs/status_fail_offset_$index/" | ||
145 | + index=$((index + 1)) | ||
146 | + done | ||
147 | + | ||
148 | + index=0 | ||
149 | + for ofs in $2 | ||
150 | + do | ||
151 | + filters+=" -e s/$ofs/read_fail_offset_$index/" | ||
152 | + index=$((index + 1)) | ||
153 | + done | ||
154 | + | ||
155 | + sed $filters | ||
156 | +} | ||
157 | + | ||
158 | +# While determining the number of allocated sectors in the input | ||
159 | +# image, we should see one block status warning per element of | ||
160 | +# $status_fail_offsets. | ||
161 | +# | ||
162 | +# Then, the image is read. Since the block status is queried in | ||
163 | +# basically the same way, the same warnings as in the previous step | ||
164 | +# should reappear. Interleaved with those we should see a read | ||
165 | +# warning per element of $read_fail_offsets. | ||
166 | +# Note that $read_fail_offsets and $status_fail_offsets share an | ||
167 | +# element (read_fail_offset_1 == status_fail_offset_1), so | ||
168 | +# "status_fail_offset_1" in the output is the same as | ||
169 | +# "read_fail_offset_1". | ||
170 | +$QEMU_IMG convert --salvage "$source_img" "$TEST_IMG" 2>&1 \ | ||
171 | + | _filter_offsets "$status_fail_offsets" "$read_fail_offsets" | ||
172 | + | ||
173 | +echo | ||
174 | + | ||
175 | +# The offsets where the block status could not be determined should | ||
176 | +# have been treated as containing data and thus should be correct in | ||
177 | +# the output image. | ||
178 | +# The offsets where reading failed altogether should be 0. Make them | ||
179 | +# 0 in the input image, too, so we can compare both images. | ||
180 | +for ofs in $read_fail_offsets | ||
181 | +do | ||
182 | + $QEMU_IO -c "write -z $ofs $sector_size" "$TEST_IMG.orig" \ | ||
183 | + | _filter_qemu_io \ | ||
184 | + | _filter_offsets '' "$read_fail_offsets" | ||
185 | +done | ||
186 | + | ||
187 | +echo | ||
188 | + | ||
189 | +# These should be equal now. | ||
190 | +$QEMU_IMG compare "$TEST_IMG.orig" "$TEST_IMG" | ||
191 | + | ||
192 | + | ||
193 | +# success, all done | ||
194 | +echo "*** done" | ||
195 | +rm -f $seq.full | ||
196 | +status=0 | ||
197 | diff --git a/tests/qemu-iotests/251.out b/tests/qemu-iotests/251.out | ||
198 | new file mode 100644 | ||
199 | index XXXXXXX..XXXXXXX | ||
200 | --- /dev/null | ||
201 | +++ b/tests/qemu-iotests/251.out | ||
202 | @@ -XXX,XX +XXX,XX @@ | ||
203 | +QA output created by 251 | ||
204 | +Formatting 'TEST_DIR/t.IMGFMT.orig', fmt=IMGFMT size=67108864 | ||
205 | +wrote 67108864/67108864 bytes at offset 0 | ||
206 | +64 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) | ||
207 | + | ||
208 | +qemu-img: warning: error while reading block status at offset status_fail_offset_0: Input/output error | ||
209 | +qemu-img: warning: error while reading block status at offset status_fail_offset_1: Input/output error | ||
210 | +qemu-img: warning: error while reading block status at offset status_fail_offset_0: Input/output error | ||
211 | +qemu-img: warning: error while reading offset read_fail_offset_0: Input/output error | ||
212 | +qemu-img: warning: error while reading block status at offset status_fail_offset_1: Input/output error | ||
213 | +qemu-img: warning: error while reading offset status_fail_offset_1: Input/output error | ||
214 | +qemu-img: warning: error while reading offset read_fail_offset_2: Input/output error | ||
215 | +qemu-img: warning: error while reading offset read_fail_offset_3: Input/output error | ||
216 | +qemu-img: warning: error while reading offset read_fail_offset_4: Input/output error | ||
217 | +qemu-img: warning: error while reading offset read_fail_offset_5: Input/output error | ||
218 | +qemu-img: warning: error while reading offset read_fail_offset_6: Input/output error | ||
219 | +qemu-img: warning: error while reading offset read_fail_offset_7: Input/output error | ||
220 | +qemu-img: warning: error while reading offset read_fail_offset_8: Input/output error | ||
221 | +qemu-img: warning: error while reading offset read_fail_offset_9: Input/output error | ||
222 | + | ||
223 | +wrote 512/512 bytes at offset read_fail_offset_0 | ||
224 | +512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) | ||
225 | +wrote 512/512 bytes at offset read_fail_offset_1 | ||
226 | +512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) | ||
227 | +wrote 512/512 bytes at offset read_fail_offset_2 | ||
228 | +512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) | ||
229 | +wrote 512/512 bytes at offset read_fail_offset_3 | ||
230 | +512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) | ||
231 | +wrote 512/512 bytes at offset read_fail_offset_4 | ||
232 | +512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) | ||
233 | +wrote 512/512 bytes at offset read_fail_offset_5 | ||
234 | +512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) | ||
235 | +wrote 512/512 bytes at offset read_fail_offset_6 | ||
236 | +512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) | ||
237 | +wrote 512/512 bytes at offset read_fail_offset_7 | ||
238 | +512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) | ||
239 | +wrote 512/512 bytes at offset read_fail_offset_8 | ||
240 | +512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) | ||
241 | +wrote 512/512 bytes at offset read_fail_offset_9 | ||
242 | +512 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec) | ||
243 | + | ||
244 | +Images are identical. | ||
245 | +*** done | ||
246 | diff --git a/tests/qemu-iotests/group b/tests/qemu-iotests/group | ||
247 | index XXXXXXX..XXXXXXX 100644 | ||
248 | --- a/tests/qemu-iotests/group | ||
249 | +++ b/tests/qemu-iotests/group | ||
250 | @@ -XXX,XX +XXX,XX @@ | ||
251 | 248 rw quick | ||
252 | 249 rw auto quick | ||
253 | 250 rw auto quick | ||
254 | +251 rw auto quick | ||
255 | 252 rw auto backing quick | ||
256 | 253 rw auto quick | ||
257 | 254 rw auto backing quick | ||
258 | -- | 121 | -- |
259 | 2.21.0 | 122 | 2.9.3 |
260 | 123 | ||
261 | 124 | diff view generated by jsdifflib |
New patch | |||
---|---|---|---|
1 | 1 | From: Paolo Bonzini <pbonzini@redhat.com> | |
2 | |||
3 | All that CoQueue needs in order to become thread-safe is help | ||
4 | from an external mutex. Add this to the API. | ||
5 | |||
6 | Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> | ||
7 | Reviewed-by: Fam Zheng <famz@redhat.com> | ||
8 | Message-id: 20170213181244.16297-6-pbonzini@redhat.com | ||
9 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
10 | --- | ||
11 | include/qemu/coroutine.h | 8 +++++--- | ||
12 | block/backup.c | 2 +- | ||
13 | block/io.c | 4 ++-- | ||
14 | block/nbd-client.c | 2 +- | ||
15 | block/qcow2-cluster.c | 4 +--- | ||
16 | block/sheepdog.c | 2 +- | ||
17 | block/throttle-groups.c | 2 +- | ||
18 | hw/9pfs/9p.c | 2 +- | ||
19 | util/qemu-coroutine-lock.c | 24 +++++++++++++++++++++--- | ||
20 | 9 files changed, 34 insertions(+), 16 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 @@ void coroutine_fn qemu_co_mutex_unlock(CoMutex *mutex); | ||
27 | |||
28 | /** | ||
29 | * CoQueues are a mechanism to queue coroutines in order to continue executing | ||
30 | - * them later. | ||
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); | ||
164 | + | ||
165 | + if (mutex) { | ||
166 | + qemu_co_mutex_unlock(mutex); | ||
167 | + } | ||
168 | + | ||
169 | + /* There is no race condition here. Other threads will call | ||
170 | + * aio_co_schedule on our AioContext, which can reenter this | ||
171 | + * coroutine but only after this yield and after the main loop | ||
172 | + * has gone through the next iteration. | ||
173 | + */ | ||
174 | qemu_coroutine_yield(); | ||
175 | assert(qemu_in_coroutine()); | ||
176 | + | ||
177 | + /* TODO: OSv implements wait morphing here, where the wakeup | ||
178 | + * primitive automatically places the woken coroutine on the | ||
179 | + * mutex's queue. This avoids the thundering herd effect. | ||
180 | + */ | ||
181 | + if (mutex) { | ||
182 | + qemu_co_mutex_lock(mutex); | ||
183 | + } | ||
184 | } | ||
185 | |||
186 | /** | ||
187 | @@ -XXX,XX +XXX,XX @@ void qemu_co_rwlock_rdlock(CoRwlock *lock) | ||
188 | Coroutine *self = qemu_coroutine_self(); | ||
189 | |||
190 | while (lock->writer) { | ||
191 | - qemu_co_queue_wait(&lock->queue); | ||
192 | + qemu_co_queue_wait(&lock->queue, NULL); | ||
193 | } | ||
194 | lock->reader++; | ||
195 | self->locks_held++; | ||
196 | @@ -XXX,XX +XXX,XX @@ void qemu_co_rwlock_wrlock(CoRwlock *lock) | ||
197 | Coroutine *self = qemu_coroutine_self(); | ||
198 | |||
199 | while (lock->writer || lock->reader) { | ||
200 | - qemu_co_queue_wait(&lock->queue); | ||
201 | + qemu_co_queue_wait(&lock->queue, NULL); | ||
202 | } | ||
203 | lock->writer = true; | ||
204 | self->locks_held++; | ||
205 | -- | ||
206 | 2.9.3 | ||
207 | |||
208 | diff view generated by jsdifflib |
1 | From: John Snow <jsnow@redhat.com> | 1 | From: Paolo Bonzini <pbonzini@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | Cap waits to 60 seconds so that iotests can fail gracefully if something | 3 | This adds a CoMutex around the existing CoQueue. Because the write-side |
4 | goes wrong. | 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. | ||
5 | 8 | ||
6 | Signed-off-by: John Snow <jsnow@redhat.com> | 9 | Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> |
7 | Message-id: 20190523170643.20794-3-jsnow@redhat.com | 10 | Reviewed-by: Fam Zheng <famz@redhat.com> |
8 | Reviewed-by: Max Reitz <mreitz@redhat.com> | 11 | Message-id: 20170213181244.16297-7-pbonzini@redhat.com |
9 | Signed-off-by: Max Reitz <mreitz@redhat.com> | 12 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> |
10 | --- | 13 | --- |
11 | tests/qemu-iotests/iotests.py | 14 +++++++------- | 14 | include/qemu/coroutine.h | 3 ++- |
12 | 1 file changed, 7 insertions(+), 7 deletions(-) | 15 | util/qemu-coroutine-lock.c | 35 ++++++++++++++++++++++++----------- |
16 | 2 files changed, 26 insertions(+), 12 deletions(-) | ||
13 | 17 | ||
14 | diff --git a/tests/qemu-iotests/iotests.py b/tests/qemu-iotests/iotests.py | 18 | diff --git a/include/qemu/coroutine.h b/include/qemu/coroutine.h |
15 | index XXXXXXX..XXXXXXX 100644 | 19 | index XXXXXXX..XXXXXXX 100644 |
16 | --- a/tests/qemu-iotests/iotests.py | 20 | --- a/include/qemu/coroutine.h |
17 | +++ b/tests/qemu-iotests/iotests.py | 21 | +++ b/include/qemu/coroutine.h |
18 | @@ -XXX,XX +XXX,XX @@ class VM(qtest.QEMUQtestMachine): | 22 | @@ -XXX,XX +XXX,XX @@ bool qemu_co_queue_empty(CoQueue *queue); |
19 | output_list += [key + '=' + obj[key]] | 23 | |
20 | return ','.join(output_list) | 24 | |
21 | 25 | typedef struct CoRwlock { | |
22 | - def get_qmp_events_filtered(self, wait=True): | 26 | - bool writer; |
23 | + def get_qmp_events_filtered(self, wait=60.0): | 27 | + int pending_writer; |
24 | result = [] | 28 | int reader; |
25 | for ev in self.get_qmp_events(wait=wait): | 29 | + CoMutex mutex; |
26 | result.append(filter_qmp_event(ev)) | 30 | CoQueue queue; |
27 | @@ -XXX,XX +XXX,XX @@ class VM(qtest.QEMUQtestMachine): | 31 | } CoRwlock; |
28 | 32 | ||
29 | # Returns None on success, and an error string on failure | 33 | diff --git a/util/qemu-coroutine-lock.c b/util/qemu-coroutine-lock.c |
30 | def run_job(self, job, auto_finalize=True, auto_dismiss=False, | 34 | index XXXXXXX..XXXXXXX 100644 |
31 | - pre_finalize=None): | 35 | --- a/util/qemu-coroutine-lock.c |
32 | + pre_finalize=None, wait=60.0): | 36 | +++ b/util/qemu-coroutine-lock.c |
33 | error = None | 37 | @@ -XXX,XX +XXX,XX @@ void qemu_co_rwlock_init(CoRwlock *lock) |
34 | while True: | 38 | { |
35 | - for ev in self.get_qmp_events_filtered(wait=True): | 39 | memset(lock, 0, sizeof(*lock)); |
36 | + for ev in self.get_qmp_events_filtered(wait=wait): | 40 | qemu_co_queue_init(&lock->queue); |
37 | if ev['event'] == 'JOB_STATUS_CHANGE': | 41 | + qemu_co_mutex_init(&lock->mutex); |
38 | status = ev['data']['status'] | 42 | } |
39 | if status == 'aborting': | 43 | |
40 | @@ -XXX,XX +XXX,XX @@ class QMPTestCase(unittest.TestCase): | 44 | void qemu_co_rwlock_rdlock(CoRwlock *lock) |
41 | self.assertEqual(self.vm.flatten_qmp_object(json.loads(json_filename[5:])), | 45 | { |
42 | self.vm.flatten_qmp_object(reference)) | 46 | Coroutine *self = qemu_coroutine_self(); |
43 | 47 | ||
44 | - def cancel_and_wait(self, drive='drive0', force=False, resume=False): | 48 | - while (lock->writer) { |
45 | + def cancel_and_wait(self, drive='drive0', force=False, resume=False, wait=60.0): | 49 | - qemu_co_queue_wait(&lock->queue, NULL); |
46 | '''Cancel a block job and wait for it to finish, returning the event''' | 50 | + qemu_co_mutex_lock(&lock->mutex); |
47 | result = self.vm.qmp('block-job-cancel', device=drive, force=force) | 51 | + /* For fairness, wait if a writer is in line. */ |
48 | self.assert_qmp(result, 'return', {}) | 52 | + while (lock->pending_writer) { |
49 | @@ -XXX,XX +XXX,XX @@ class QMPTestCase(unittest.TestCase): | 53 | + qemu_co_queue_wait(&lock->queue, &lock->mutex); |
50 | cancelled = False | 54 | } |
51 | result = None | 55 | lock->reader++; |
52 | while not cancelled: | 56 | + qemu_co_mutex_unlock(&lock->mutex); |
53 | - for event in self.vm.get_qmp_events(wait=True): | 57 | + |
54 | + for event in self.vm.get_qmp_events(wait=wait): | 58 | + /* The rest of the read-side critical section is run without the mutex. */ |
55 | if event['event'] == 'BLOCK_JOB_COMPLETED' or \ | 59 | self->locks_held++; |
56 | event['event'] == 'BLOCK_JOB_CANCELLED': | 60 | } |
57 | self.assert_qmp(event, 'data/device', drive) | 61 | |
58 | @@ -XXX,XX +XXX,XX @@ class QMPTestCase(unittest.TestCase): | 62 | @@ -XXX,XX +XXX,XX @@ void qemu_co_rwlock_unlock(CoRwlock *lock) |
59 | self.assert_no_active_block_jobs() | 63 | Coroutine *self = qemu_coroutine_self(); |
60 | return result | 64 | |
61 | 65 | assert(qemu_in_coroutine()); | |
62 | - def wait_until_completed(self, drive='drive0', check_offset=True): | 66 | - if (lock->writer) { |
63 | + def wait_until_completed(self, drive='drive0', check_offset=True, wait=60.0): | 67 | - lock->writer = false; |
64 | '''Wait for a block job to finish, returning the event''' | 68 | + if (!lock->reader) { |
65 | while True: | 69 | + /* The critical section started in qemu_co_rwlock_wrlock. */ |
66 | - for event in self.vm.get_qmp_events(wait=True): | 70 | qemu_co_queue_restart_all(&lock->queue); |
67 | + for event in self.vm.get_qmp_events(wait=wait): | 71 | } else { |
68 | if event['event'] == 'BLOCK_JOB_COMPLETED': | 72 | + self->locks_held--; |
69 | self.assert_qmp(event, 'data/device', drive) | 73 | + |
70 | self.assert_qmp_absent(event, 'data/error') | 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 | } | ||
71 | -- | 106 | -- |
72 | 2.21.0 | 107 | 2.9.3 |
73 | 108 | ||
74 | 109 | diff view generated by jsdifflib |