[PULL 00/12] Net patches
[PULL V2 00/15] Net patches
1
The following changes since commit 7fe7fae8b48e3f9c647fd685e5155ebc8e6fb84d:
1
The following changes since commit d9ccf33f9479201e5add8db0af68ca9ca8da358b:
2
2
3
Merge remote-tracking branch 'remotes/dgilbert-gitlab/tags/pull-migration-20210609a' into staging (2021-06-09 16:40:21 +0100)
3
Merge remote-tracking branch 'remotes/lvivier-gitlab/tags/linux-user-for-7.0-pull-request' into staging (2022-03-09 20:01:17 +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/jasowang/qemu.git tags/net-pull-request
7
https://github.com/jasowang/qemu.git tags/net-pull-request
8
8
9
for you to fetch changes up to 5a2d9929ac1f01a1e8ef2a3f56f69e6069863dad:
9
for you to fetch changes up to eea40402ecf895ed345f8e8eb07dbb484f4542c5:
10
10
11
Fixed calculation error of pkt->header_size in fill_pkt_tcp_info() (2021-06-11 10:30:13 +0800)
11
vdpa: Expose VHOST_F_LOG_ALL on SVQ (2022-03-10 10:26:32 +0800)
12
12
13
----------------------------------------------------------------
13
----------------------------------------------------------------
14
14
15
----------------------------------------------------------------
15
----------------------------------------------------------------
16
Jason Wang (4):
16
Eugenio Pérez (14):
17
vhost-vdpa: skip ram device from the IOTLB mapping
17
vhost: Add VhostShadowVirtqueue
18
vhost-vdpa: map virtqueue notification area if possible
18
vhost: Add Shadow VirtQueue kick forwarding capabilities
19
vhost-vdpa: don't initialize backend_features
19
vhost: Add Shadow VirtQueue call forwarding capabilities
20
vhost-vdpa: remove the unused vhost_vdpa_get_acked_features()
20
vhost: Add vhost_svq_valid_features to shadow vq
21
virtio: Add vhost_svq_get_vring_addr
22
vdpa: adapt vhost_ops callbacks to svq
23
vhost: Shadow virtqueue buffers forwarding
24
util: Add iova_tree_alloc_map
25
util: add iova_tree_find_iova
26
vhost: Add VhostIOVATree
27
vdpa: Add custom IOTLB translations to SVQ
28
vdpa: Adapt vhost_vdpa_get_vring_base to SVQ
29
vdpa: Never set log_base addr if SVQ is enabled
30
vdpa: Expose VHOST_F_LOG_ALL on SVQ
21
31
22
Paolo Bonzini (1):
32
Jason Wang (1):
23
netdev: add more commands to preconfig mode
33
virtio-net: fix map leaking on error during receive
24
34
25
Rao, Lei (7):
35
hw/net/virtio-net.c | 1 +
26
Remove some duplicate trace code.
36
hw/virtio/meson.build | 2 +-
27
Fix the qemu crash when guest shutdown during checkpoint
37
hw/virtio/vhost-iova-tree.c | 110 +++++++
28
Optimize the function of filter_send
38
hw/virtio/vhost-iova-tree.h | 27 ++
29
Remove migrate_set_block_enabled in checkpoint
39
hw/virtio/vhost-shadow-virtqueue.c | 638 +++++++++++++++++++++++++++++++++++++
30
Add a function named packet_new_nocopy for COLO.
40
hw/virtio/vhost-shadow-virtqueue.h | 87 +++++
31
Add the function of colo_compare_cleanup
41
hw/virtio/vhost-vdpa.c | 525 +++++++++++++++++++++++++++++-
32
Fixed calculation error of pkt->header_size in fill_pkt_tcp_info()
42
include/hw/virtio/vhost-vdpa.h | 8 +
33
43
include/qemu/iova-tree.h | 38 ++-
34
hmp-commands.hx | 2 +
44
util/iova-tree.c | 169 ++++++++++
35
hw/virtio/vhost-vdpa.c | 100 +++++++++++++++++++++++++++++++++++------
45
10 files changed, 1588 insertions(+), 17 deletions(-)
36
include/hw/virtio/vhost-vdpa.h | 6 +++
46
create mode 100644 hw/virtio/vhost-iova-tree.c
37
include/net/vhost-vdpa.h | 1 -
47
create mode 100644 hw/virtio/vhost-iova-tree.h
38
migration/colo.c | 6 ---
48
create mode 100644 hw/virtio/vhost-shadow-virtqueue.c
39
migration/migration.c | 4 ++
49
create mode 100644 hw/virtio/vhost-shadow-virtqueue.h
40
net/colo-compare.c | 25 +++++------
41
net/colo-compare.h | 1 +
42
net/colo.c | 25 +++++++----
43
net/colo.h | 1 +
44
net/filter-mirror.c | 8 ++--
45
net/filter-rewriter.c | 3 +-
46
net/net.c | 4 ++
47
net/vhost-vdpa.c | 9 ----
48
qapi/net.json | 6 ++-
49
softmmu/runstate.c | 1 +
50
16 files changed, 143 insertions(+), 59 deletions(-)
51
50
52
51
53
54
diff view generated by jsdifflib
New patch
[PULL V2 01/15] virtio-net: fix map leaking on error during receive
1
Commit bedd7e93d0196 ("virtio-net: fix use after unmap/free for sg")
2
tries to fix the use after free of the sg by caching the virtqueue
3
elements in an array and unmap them at once after receiving the
4
packets, But it forgot to unmap the cached elements on error which
5
will lead to leaking of mapping and other unexpected results.
1
6
7
Fixing this by detaching the cached elements on error. This addresses
8
CVE-2022-26353.
9
10
Reported-by: Victor Tom <vv474172261@gmail.com>
11
Cc: qemu-stable@nongnu.org
12
Fixes: CVE-2022-26353
13
Fixes: bedd7e93d0196 ("virtio-net: fix use after unmap/free for sg")
14
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
15
Signed-off-by: Jason Wang <jasowang@redhat.com>
16
---
17
hw/net/virtio-net.c | 1 +
18
1 file changed, 1 insertion(+)
19
20
diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c
21
index XXXXXXX..XXXXXXX 100644
22
--- a/hw/net/virtio-net.c
23
+++ b/hw/net/virtio-net.c
24
@@ -XXX,XX +XXX,XX @@ static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf,
25
26
err:
27
for (j = 0; j < i; j++) {
28
+ virtqueue_detach_element(q->rx_vq, elems[j], lens[j]);
29
g_free(elems[j]);
30
}
31
32
--
33
2.7.4
diff view generated by jsdifflib
New patch
[PULL V2 02/15] vhost: Add VhostShadowVirtqueue
1
From: Eugenio Pérez <eperezma@redhat.com>
1
2
3
Vhost shadow virtqueue (SVQ) is an intermediate jump for virtqueue
4
notifications and buffers, allowing qemu to track them. While qemu is
5
forwarding the buffers and virtqueue changes, it is able to commit the
6
memory it's being dirtied, the same way regular qemu's VirtIO devices
7
do.
8
9
This commit only exposes basic SVQ allocation and free. Next patches of
10
the series add functionality like notifications and buffers forwarding.
11
12
Acked-by: Michael S. Tsirkin <mst@redhat.com>
13
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
14
Signed-off-by: Jason Wang <jasowang@redhat.com>
15
---
16
hw/virtio/meson.build | 2 +-
17
hw/virtio/vhost-shadow-virtqueue.c | 62 ++++++++++++++++++++++++++++++++++++++
18
hw/virtio/vhost-shadow-virtqueue.h | 28 +++++++++++++++++
19
3 files changed, 91 insertions(+), 1 deletion(-)
20
create mode 100644 hw/virtio/vhost-shadow-virtqueue.c
21
create mode 100644 hw/virtio/vhost-shadow-virtqueue.h
22
23
diff --git a/hw/virtio/meson.build b/hw/virtio/meson.build
24
index XXXXXXX..XXXXXXX 100644
25
--- a/hw/virtio/meson.build
26
+++ b/hw/virtio/meson.build
27
@@ -XXX,XX +XXX,XX @@ softmmu_ss.add(when: 'CONFIG_ALL', if_true: files('vhost-stub.c'))
28
29
virtio_ss = ss.source_set()
30
virtio_ss.add(files('virtio.c'))
31
-virtio_ss.add(when: 'CONFIG_VHOST', if_true: files('vhost.c', 'vhost-backend.c'))
32
+virtio_ss.add(when: 'CONFIG_VHOST', if_true: files('vhost.c', 'vhost-backend.c', 'vhost-shadow-virtqueue.c'))
33
virtio_ss.add(when: 'CONFIG_VHOST_USER', if_true: files('vhost-user.c'))
34
virtio_ss.add(when: 'CONFIG_VHOST_VDPA', if_true: files('vhost-vdpa.c'))
35
virtio_ss.add(when: 'CONFIG_VIRTIO_BALLOON', if_true: files('virtio-balloon.c'))
36
diff --git a/hw/virtio/vhost-shadow-virtqueue.c b/hw/virtio/vhost-shadow-virtqueue.c
37
new file mode 100644
38
index XXXXXXX..XXXXXXX
39
--- /dev/null
40
+++ b/hw/virtio/vhost-shadow-virtqueue.c
41
@@ -XXX,XX +XXX,XX @@
42
+/*
43
+ * vhost shadow virtqueue
44
+ *
45
+ * SPDX-FileCopyrightText: Red Hat, Inc. 2021
46
+ * SPDX-FileContributor: Author: Eugenio Pérez <eperezma@redhat.com>
47
+ *
48
+ * SPDX-License-Identifier: GPL-2.0-or-later
49
+ */
50
+
51
+#include "qemu/osdep.h"
52
+#include "hw/virtio/vhost-shadow-virtqueue.h"
53
+
54
+#include "qemu/error-report.h"
55
+
56
+/**
57
+ * Creates vhost shadow virtqueue, and instructs the vhost device to use the
58
+ * shadow methods and file descriptors.
59
+ *
60
+ * Returns the new virtqueue or NULL.
61
+ *
62
+ * In case of error, reason is reported through error_report.
63
+ */
64
+VhostShadowVirtqueue *vhost_svq_new(void)
65
+{
66
+ g_autofree VhostShadowVirtqueue *svq = g_new0(VhostShadowVirtqueue, 1);
67
+ int r;
68
+
69
+ r = event_notifier_init(&svq->hdev_kick, 0);
70
+ if (r != 0) {
71
+ error_report("Couldn't create kick event notifier: %s (%d)",
72
+ g_strerror(errno), errno);
73
+ goto err_init_hdev_kick;
74
+ }
75
+
76
+ r = event_notifier_init(&svq->hdev_call, 0);
77
+ if (r != 0) {
78
+ error_report("Couldn't create call event notifier: %s (%d)",
79
+ g_strerror(errno), errno);
80
+ goto err_init_hdev_call;
81
+ }
82
+
83
+ return g_steal_pointer(&svq);
84
+
85
+err_init_hdev_call:
86
+ event_notifier_cleanup(&svq->hdev_kick);
87
+
88
+err_init_hdev_kick:
89
+ return NULL;
90
+}
91
+
92
+/**
93
+ * Free the resources of the shadow virtqueue.
94
+ *
95
+ * @pvq: gpointer to SVQ so it can be used by autofree functions.
96
+ */
97
+void vhost_svq_free(gpointer pvq)
98
+{
99
+ VhostShadowVirtqueue *vq = pvq;
100
+ event_notifier_cleanup(&vq->hdev_kick);
101
+ event_notifier_cleanup(&vq->hdev_call);
102
+ g_free(vq);
103
+}
104
diff --git a/hw/virtio/vhost-shadow-virtqueue.h b/hw/virtio/vhost-shadow-virtqueue.h
105
new file mode 100644
106
index XXXXXXX..XXXXXXX
107
--- /dev/null
108
+++ b/hw/virtio/vhost-shadow-virtqueue.h
109
@@ -XXX,XX +XXX,XX @@
110
+/*
111
+ * vhost shadow virtqueue
112
+ *
113
+ * SPDX-FileCopyrightText: Red Hat, Inc. 2021
114
+ * SPDX-FileContributor: Author: Eugenio Pérez <eperezma@redhat.com>
115
+ *
116
+ * SPDX-License-Identifier: GPL-2.0-or-later
117
+ */
118
+
119
+#ifndef VHOST_SHADOW_VIRTQUEUE_H
120
+#define VHOST_SHADOW_VIRTQUEUE_H
121
+
122
+#include "qemu/event_notifier.h"
123
+
124
+/* Shadow virtqueue to relay notifications */
125
+typedef struct VhostShadowVirtqueue {
126
+ /* Shadow kick notifier, sent to vhost */
127
+ EventNotifier hdev_kick;
128
+ /* Shadow call notifier, sent to vhost */
129
+ EventNotifier hdev_call;
130
+} VhostShadowVirtqueue;
131
+
132
+VhostShadowVirtqueue *vhost_svq_new(void);
133
+
134
+void vhost_svq_free(gpointer vq);
135
+G_DEFINE_AUTOPTR_CLEANUP_FUNC(VhostShadowVirtqueue, vhost_svq_free);
136
+
137
+#endif
138
--
139
2.7.4
140
141
diff view generated by jsdifflib
[PULL 03/12] vhost-vdpa: don't initialize backend_features
[PULL V2 03/15] vhost: Add Shadow VirtQueue kick forwarding capabilities
1
We used to initialize backend_features during vhost_vdpa_init()
1
From: Eugenio Pérez <eperezma@redhat.com>
2
regardless whether or not it was supported by vhost. This will lead
2
3
the unsupported features like VIRTIO_F_IN_ORDER to be included and set
3
At this mode no buffer forwarding will be performed in SVQ mode: Qemu
4
to the vhost-vdpa during vhost_dev_start. Because the
4
will just forward the guest's kicks to the device.
5
VIRTIO_F_IN_ORDER is not supported by vhost-vdpa so it won't be
5
6
advertised to guest which will break the datapath.
6
Host memory notifiers regions are left out for simplicity, and they will
7
7
not be addressed in this series.
8
Fix this by not initializing the backend_features, so the
8
9
acked_features could be built only from guest features via
9
Acked-by: Michael S. Tsirkin <mst@redhat.com>
10
vhost_net_ack_features().
10
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
11
12
Fixes: 108a64818e69b ("vhost-vdpa: introduce vhost-vdpa backend")
13
Cc: qemu-stable@nongnu.org
14
Cc: Gautam Dawar <gdawar@xilinx.com>
15
Signed-off-by: Jason Wang <jasowang@redhat.com>
11
Signed-off-by: Jason Wang <jasowang@redhat.com>
16
---
12
---
17
hw/virtio/vhost-vdpa.c | 3 ---
13
hw/virtio/vhost-shadow-virtqueue.c | 56 ++++++++++++++
18
1 file changed, 3 deletions(-)
14
hw/virtio/vhost-shadow-virtqueue.h | 14 ++++
19
15
hw/virtio/vhost-vdpa.c | 145 ++++++++++++++++++++++++++++++++++++-
16
include/hw/virtio/vhost-vdpa.h | 4 +
17
4 files changed, 217 insertions(+), 2 deletions(-)
18
19
diff --git a/hw/virtio/vhost-shadow-virtqueue.c b/hw/virtio/vhost-shadow-virtqueue.c
20
index XXXXXXX..XXXXXXX 100644
21
--- a/hw/virtio/vhost-shadow-virtqueue.c
22
+++ b/hw/virtio/vhost-shadow-virtqueue.c
23
@@ -XXX,XX +XXX,XX @@
24
#include "hw/virtio/vhost-shadow-virtqueue.h"
25
26
#include "qemu/error-report.h"
27
+#include "qemu/main-loop.h"
28
+#include "linux-headers/linux/vhost.h"
29
+
30
+/**
31
+ * Forward guest notifications.
32
+ *
33
+ * @n: guest kick event notifier, the one that guest set to notify svq.
34
+ */
35
+static void vhost_handle_guest_kick(EventNotifier *n)
36
+{
37
+ VhostShadowVirtqueue *svq = container_of(n, VhostShadowVirtqueue,
38
+ svq_kick);
39
+ event_notifier_test_and_clear(n);
40
+ event_notifier_set(&svq->hdev_kick);
41
+}
42
+
43
+/**
44
+ * Set a new file descriptor for the guest to kick the SVQ and notify for avail
45
+ *
46
+ * @svq: The svq
47
+ * @svq_kick_fd: The svq kick fd
48
+ *
49
+ * Note that the SVQ will never close the old file descriptor.
50
+ */
51
+void vhost_svq_set_svq_kick_fd(VhostShadowVirtqueue *svq, int svq_kick_fd)
52
+{
53
+ EventNotifier *svq_kick = &svq->svq_kick;
54
+ bool poll_stop = VHOST_FILE_UNBIND != event_notifier_get_fd(svq_kick);
55
+ bool poll_start = svq_kick_fd != VHOST_FILE_UNBIND;
56
+
57
+ if (poll_stop) {
58
+ event_notifier_set_handler(svq_kick, NULL);
59
+ }
60
+
61
+ /*
62
+ * event_notifier_set_handler already checks for guest's notifications if
63
+ * they arrive at the new file descriptor in the switch, so there is no
64
+ * need to explicitly check for them.
65
+ */
66
+ if (poll_start) {
67
+ event_notifier_init_fd(svq_kick, svq_kick_fd);
68
+ event_notifier_set(svq_kick);
69
+ event_notifier_set_handler(svq_kick, vhost_handle_guest_kick);
70
+ }
71
+}
72
+
73
+/**
74
+ * Stop the shadow virtqueue operation.
75
+ * @svq: Shadow Virtqueue
76
+ */
77
+void vhost_svq_stop(VhostShadowVirtqueue *svq)
78
+{
79
+ event_notifier_set_handler(&svq->svq_kick, NULL);
80
+}
81
82
/**
83
* Creates vhost shadow virtqueue, and instructs the vhost device to use the
84
@@ -XXX,XX +XXX,XX @@ VhostShadowVirtqueue *vhost_svq_new(void)
85
goto err_init_hdev_call;
86
}
87
88
+ event_notifier_init_fd(&svq->svq_kick, VHOST_FILE_UNBIND);
89
return g_steal_pointer(&svq);
90
91
err_init_hdev_call:
92
@@ -XXX,XX +XXX,XX @@ err_init_hdev_kick:
93
void vhost_svq_free(gpointer pvq)
94
{
95
VhostShadowVirtqueue *vq = pvq;
96
+ vhost_svq_stop(vq);
97
event_notifier_cleanup(&vq->hdev_kick);
98
event_notifier_cleanup(&vq->hdev_call);
99
g_free(vq);
100
diff --git a/hw/virtio/vhost-shadow-virtqueue.h b/hw/virtio/vhost-shadow-virtqueue.h
101
index XXXXXXX..XXXXXXX 100644
102
--- a/hw/virtio/vhost-shadow-virtqueue.h
103
+++ b/hw/virtio/vhost-shadow-virtqueue.h
104
@@ -XXX,XX +XXX,XX @@ typedef struct VhostShadowVirtqueue {
105
EventNotifier hdev_kick;
106
/* Shadow call notifier, sent to vhost */
107
EventNotifier hdev_call;
108
+
109
+ /*
110
+ * Borrowed virtqueue's guest to host notifier. To borrow it in this event
111
+ * notifier allows to recover the VhostShadowVirtqueue from the event loop
112
+ * easily. If we use the VirtQueue's one, we don't have an easy way to
113
+ * retrieve VhostShadowVirtqueue.
114
+ *
115
+ * So shadow virtqueue must not clean it, or we would lose VirtQueue one.
116
+ */
117
+ EventNotifier svq_kick;
118
} VhostShadowVirtqueue;
119
120
+void vhost_svq_set_svq_kick_fd(VhostShadowVirtqueue *svq, int svq_kick_fd);
121
+
122
+void vhost_svq_stop(VhostShadowVirtqueue *svq);
123
+
124
VhostShadowVirtqueue *vhost_svq_new(void);
125
126
void vhost_svq_free(gpointer vq);
20
diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c
127
diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c
21
index XXXXXXX..XXXXXXX 100644
128
index XXXXXXX..XXXXXXX 100644
22
--- a/hw/virtio/vhost-vdpa.c
129
--- a/hw/virtio/vhost-vdpa.c
23
+++ b/hw/virtio/vhost-vdpa.c
130
+++ b/hw/virtio/vhost-vdpa.c
24
@@ -XXX,XX +XXX,XX @@ static void vhost_vdpa_add_status(struct vhost_dev *dev, uint8_t status)
131
@@ -XXX,XX +XXX,XX @@
25
static int vhost_vdpa_init(struct vhost_dev *dev, void *opaque)
132
#include "hw/virtio/vhost.h"
133
#include "hw/virtio/vhost-backend.h"
134
#include "hw/virtio/virtio-net.h"
135
+#include "hw/virtio/vhost-shadow-virtqueue.h"
136
#include "hw/virtio/vhost-vdpa.h"
137
#include "exec/address-spaces.h"
138
#include "qemu/main-loop.h"
139
#include "cpu.h"
140
#include "trace.h"
141
#include "qemu-common.h"
142
+#include "qapi/error.h"
143
144
/*
145
* Return one past the end of the end of section. Be careful with uint64_t
146
@@ -XXX,XX +XXX,XX @@ static bool vhost_vdpa_one_time_request(struct vhost_dev *dev)
147
return v->index != 0;
148
}
149
150
+static int vhost_vdpa_init_svq(struct vhost_dev *hdev, struct vhost_vdpa *v,
151
+ Error **errp)
152
+{
153
+ g_autoptr(GPtrArray) shadow_vqs = NULL;
154
+
155
+ if (!v->shadow_vqs_enabled) {
156
+ return 0;
157
+ }
158
+
159
+ shadow_vqs = g_ptr_array_new_full(hdev->nvqs, vhost_svq_free);
160
+ for (unsigned n = 0; n < hdev->nvqs; ++n) {
161
+ g_autoptr(VhostShadowVirtqueue) svq = vhost_svq_new();
162
+
163
+ if (unlikely(!svq)) {
164
+ error_setg(errp, "Cannot create svq %u", n);
165
+ return -1;
166
+ }
167
+ g_ptr_array_add(shadow_vqs, g_steal_pointer(&svq));
168
+ }
169
+
170
+ v->shadow_vqs = g_steal_pointer(&shadow_vqs);
171
+ return 0;
172
+}
173
+
174
static int vhost_vdpa_init(struct vhost_dev *dev, void *opaque, Error **errp)
26
{
175
{
27
struct vhost_vdpa *v;
176
struct vhost_vdpa *v;
28
- uint64_t features;
177
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_init(struct vhost_dev *dev, void *opaque, Error **errp)
29
assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_VDPA);
30
trace_vhost_vdpa_init(dev, opaque);
31
32
v = opaque;
33
v->dev = dev;
34
dev->opaque = opaque ;
178
dev->opaque = opaque ;
35
- vhost_vdpa_call(dev, VHOST_GET_FEATURES, &features);
36
- dev->backend_features = features;
37
v->listener = vhost_vdpa_memory_listener;
179
v->listener = vhost_vdpa_memory_listener;
38
v->msg_type = VHOST_IOTLB_MSG_V2;
180
v->msg_type = VHOST_IOTLB_MSG_V2;
39
181
+ ret = vhost_vdpa_init_svq(dev, v, errp);
182
+ if (ret) {
183
+ goto err;
184
+ }
185
186
vhost_vdpa_get_iova_range(v);
187
188
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_init(struct vhost_dev *dev, void *opaque, Error **errp)
189
VIRTIO_CONFIG_S_DRIVER);
190
191
return 0;
192
+
193
+err:
194
+ ram_block_discard_disable(false);
195
+ return ret;
196
}
197
198
static void vhost_vdpa_host_notifier_uninit(struct vhost_dev *dev,
199
@@ -XXX,XX +XXX,XX @@ static void vhost_vdpa_host_notifiers_uninit(struct vhost_dev *dev, int n)
200
201
static void vhost_vdpa_host_notifiers_init(struct vhost_dev *dev)
202
{
203
+ struct vhost_vdpa *v = dev->opaque;
204
int i;
205
206
+ if (v->shadow_vqs_enabled) {
207
+ /* FIXME SVQ is not compatible with host notifiers mr */
208
+ return;
209
+ }
210
+
211
for (i = dev->vq_index; i < dev->vq_index + dev->nvqs; i++) {
212
if (vhost_vdpa_host_notifier_init(dev, i)) {
213
goto err;
214
@@ -XXX,XX +XXX,XX @@ err:
215
return;
216
}
217
218
+static void vhost_vdpa_svq_cleanup(struct vhost_dev *dev)
219
+{
220
+ struct vhost_vdpa *v = dev->opaque;
221
+ size_t idx;
222
+
223
+ if (!v->shadow_vqs) {
224
+ return;
225
+ }
226
+
227
+ for (idx = 0; idx < v->shadow_vqs->len; ++idx) {
228
+ vhost_svq_stop(g_ptr_array_index(v->shadow_vqs, idx));
229
+ }
230
+ g_ptr_array_free(v->shadow_vqs, true);
231
+}
232
+
233
static int vhost_vdpa_cleanup(struct vhost_dev *dev)
234
{
235
struct vhost_vdpa *v;
236
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_cleanup(struct vhost_dev *dev)
237
trace_vhost_vdpa_cleanup(dev, v);
238
vhost_vdpa_host_notifiers_uninit(dev, dev->nvqs);
239
memory_listener_unregister(&v->listener);
240
+ vhost_vdpa_svq_cleanup(dev);
241
242
dev->opaque = NULL;
243
ram_block_discard_disable(false);
244
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_get_device_id(struct vhost_dev *dev,
245
return ret;
246
}
247
248
+static void vhost_vdpa_reset_svq(struct vhost_vdpa *v)
249
+{
250
+ if (!v->shadow_vqs_enabled) {
251
+ return;
252
+ }
253
+
254
+ for (unsigned i = 0; i < v->shadow_vqs->len; ++i) {
255
+ VhostShadowVirtqueue *svq = g_ptr_array_index(v->shadow_vqs, i);
256
+ vhost_svq_stop(svq);
257
+ }
258
+}
259
+
260
static int vhost_vdpa_reset_device(struct vhost_dev *dev)
261
{
262
+ struct vhost_vdpa *v = dev->opaque;
263
int ret;
264
uint8_t status = 0;
265
266
+ vhost_vdpa_reset_svq(v);
267
+
268
ret = vhost_vdpa_call(dev, VHOST_VDPA_SET_STATUS, &status);
269
trace_vhost_vdpa_reset_device(dev, status);
270
return ret;
271
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_get_config(struct vhost_dev *dev, uint8_t *config,
272
return ret;
273
}
274
275
+static int vhost_vdpa_set_vring_dev_kick(struct vhost_dev *dev,
276
+ struct vhost_vring_file *file)
277
+{
278
+ trace_vhost_vdpa_set_vring_kick(dev, file->index, file->fd);
279
+ return vhost_vdpa_call(dev, VHOST_SET_VRING_KICK, file);
280
+}
281
+
282
+/**
283
+ * Set the shadow virtqueue descriptors to the device
284
+ *
285
+ * @dev: The vhost device model
286
+ * @svq: The shadow virtqueue
287
+ * @idx: The index of the virtqueue in the vhost device
288
+ * @errp: Error
289
+ */
290
+static bool vhost_vdpa_svq_setup(struct vhost_dev *dev,
291
+ VhostShadowVirtqueue *svq,
292
+ unsigned idx,
293
+ Error **errp)
294
+{
295
+ struct vhost_vring_file file = {
296
+ .index = dev->vq_index + idx,
297
+ };
298
+ const EventNotifier *event_notifier = &svq->hdev_kick;
299
+ int r;
300
+
301
+ file.fd = event_notifier_get_fd(event_notifier);
302
+ r = vhost_vdpa_set_vring_dev_kick(dev, &file);
303
+ if (unlikely(r != 0)) {
304
+ error_setg_errno(errp, -r, "Can't set device kick fd");
305
+ }
306
+
307
+ return r == 0;
308
+}
309
+
310
+static bool vhost_vdpa_svqs_start(struct vhost_dev *dev)
311
+{
312
+ struct vhost_vdpa *v = dev->opaque;
313
+ Error *err = NULL;
314
+ unsigned i;
315
+
316
+ if (!v->shadow_vqs) {
317
+ return true;
318
+ }
319
+
320
+ for (i = 0; i < v->shadow_vqs->len; ++i) {
321
+ VhostShadowVirtqueue *svq = g_ptr_array_index(v->shadow_vqs, i);
322
+ bool ok = vhost_vdpa_svq_setup(dev, svq, i, &err);
323
+ if (unlikely(!ok)) {
324
+ error_reportf_err(err, "Cannot setup SVQ %u: ", i);
325
+ return false;
326
+ }
327
+ }
328
+
329
+ return true;
330
+}
331
+
332
static int vhost_vdpa_dev_start(struct vhost_dev *dev, bool started)
333
{
334
struct vhost_vdpa *v = dev->opaque;
335
+ bool ok;
336
trace_vhost_vdpa_dev_start(dev, started);
337
338
if (started) {
339
vhost_vdpa_host_notifiers_init(dev);
340
+ ok = vhost_vdpa_svqs_start(dev);
341
+ if (unlikely(!ok)) {
342
+ return -1;
343
+ }
344
vhost_vdpa_set_vring_ready(dev);
345
} else {
346
vhost_vdpa_host_notifiers_uninit(dev, dev->nvqs);
347
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_get_vring_base(struct vhost_dev *dev,
348
static int vhost_vdpa_set_vring_kick(struct vhost_dev *dev,
349
struct vhost_vring_file *file)
350
{
351
- trace_vhost_vdpa_set_vring_kick(dev, file->index, file->fd);
352
- return vhost_vdpa_call(dev, VHOST_SET_VRING_KICK, file);
353
+ struct vhost_vdpa *v = dev->opaque;
354
+ int vdpa_idx = file->index - dev->vq_index;
355
+
356
+ if (v->shadow_vqs_enabled) {
357
+ VhostShadowVirtqueue *svq = g_ptr_array_index(v->shadow_vqs, vdpa_idx);
358
+ vhost_svq_set_svq_kick_fd(svq, file->fd);
359
+ return 0;
360
+ } else {
361
+ return vhost_vdpa_set_vring_dev_kick(dev, file);
362
+ }
363
}
364
365
static int vhost_vdpa_set_vring_call(struct vhost_dev *dev,
366
diff --git a/include/hw/virtio/vhost-vdpa.h b/include/hw/virtio/vhost-vdpa.h
367
index XXXXXXX..XXXXXXX 100644
368
--- a/include/hw/virtio/vhost-vdpa.h
369
+++ b/include/hw/virtio/vhost-vdpa.h
370
@@ -XXX,XX +XXX,XX @@
371
#ifndef HW_VIRTIO_VHOST_VDPA_H
372
#define HW_VIRTIO_VHOST_VDPA_H
373
374
+#include <gmodule.h>
375
+
376
#include "hw/virtio/virtio.h"
377
#include "standard-headers/linux/vhost_types.h"
378
379
@@ -XXX,XX +XXX,XX @@ typedef struct vhost_vdpa {
380
bool iotlb_batch_begin_sent;
381
MemoryListener listener;
382
struct vhost_vdpa_iova_range iova_range;
383
+ bool shadow_vqs_enabled;
384
+ GPtrArray *shadow_vqs;
385
struct vhost_dev *dev;
386
VhostVDPAHostNotifier notifier[VIRTIO_QUEUE_MAX];
387
} VhostVDPA;
40
--
388
--
41
2.7.4
389
2.7.4
42
390
43
391
diff view generated by jsdifflib
[PULL 11/12] Add the function of colo_compare_cleanup
[PULL V2 04/15] vhost: Add Shadow VirtQueue call forwarding capabilities
1
From: "Rao, Lei" <lei.rao@intel.com>
1
From: Eugenio Pérez <eperezma@redhat.com>
2
2
3
This patch fixes the following:
3
This will make qemu aware of the device used buffers, allowing it to
4
#0 __GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:50
4
write the guest memory with its contents if needed.
5
#1 0x00007f6ae4559859 in __GI_abort () at abort.c:79
6
#2 0x0000559aaa386720 in error_exit (err=16, msg=0x559aaa5973d0 <__func__.16227> "qemu_mutex_destroy") at util/qemu-thread-posix.c:36
7
#3 0x0000559aaa3868c5 in qemu_mutex_destroy (mutex=0x559aabffe828) at util/qemu-thread-posix.c:69
8
#4 0x0000559aaa2f93a8 in char_finalize (obj=0x559aabffe800) at chardev/char.c:285
9
#5 0x0000559aaa23318a in object_deinit (obj=0x559aabffe800, type=0x559aabfd7d20) at qom/object.c:606
10
#6 0x0000559aaa2331b8 in object_deinit (obj=0x559aabffe800, type=0x559aabfd9060) at qom/object.c:610
11
#7 0x0000559aaa233200 in object_finalize (data=0x559aabffe800) at qom/object.c:620
12
#8 0x0000559aaa234202 in object_unref (obj=0x559aabffe800) at qom/object.c:1074
13
#9 0x0000559aaa2356b6 in object_finalize_child_property (obj=0x559aac0dac10, name=0x559aac778760 "compare0-0", opaque=0x559aabffe800) at qom/object.c:1584
14
#10 0x0000559aaa232f70 in object_property_del_all (obj=0x559aac0dac10) at qom/object.c:557
15
#11 0x0000559aaa2331ed in object_finalize (data=0x559aac0dac10) at qom/object.c:619
16
#12 0x0000559aaa234202 in object_unref (obj=0x559aac0dac10) at qom/object.c:1074
17
#13 0x0000559aaa2356b6 in object_finalize_child_property (obj=0x559aac0c75c0, name=0x559aac0dadc0 "chardevs", opaque=0x559aac0dac10) at qom/object.c:1584
18
#14 0x0000559aaa233071 in object_property_del_child (obj=0x559aac0c75c0, child=0x559aac0dac10, errp=0x0) at qom/object.c:580
19
#15 0x0000559aaa233155 in object_unparent (obj=0x559aac0dac10) at qom/object.c:599
20
#16 0x0000559aaa2fb721 in qemu_chr_cleanup () at chardev/char.c:1159
21
#17 0x0000559aa9f9b110 in main (argc=54, argv=0x7ffeb62fa998, envp=0x7ffeb62fab50) at vl.c:4539
22
5
23
When chardev is cleaned up, chr_write_lock needs to be destroyed. But
6
Acked-by: Michael S. Tsirkin <mst@redhat.com>
24
the colo-compare module is not cleaned up normally before it when the
7
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
25
guest poweroff. It is holding chr_write_lock at this time. This will
26
cause qemu crash.So we add the function of colo_compare_cleanup() before
27
qemu_chr_cleanup() to fix the bug.
28
29
Signed-off-by: Lei Rao <lei.rao@intel.com>
30
Reviewed-by: Zhang Chen <chen.zhang@intel.com>
31
Reviewed-by: Lukas Straub <lukasstraub2@web.de>
32
Tested-by: Lukas Straub <lukasstraub2@web.de>
33
Signed-off-by: Zhang Chen <chen.zhang@intel.com>
34
Signed-off-by: Jason Wang <jasowang@redhat.com>
8
Signed-off-by: Jason Wang <jasowang@redhat.com>
35
---
9
---
36
net/colo-compare.c | 10 ++++++++++
10
hw/virtio/vhost-shadow-virtqueue.c | 38 ++++++++++++++++++++++++++++++++++++++
37
net/colo-compare.h | 1 +
11
hw/virtio/vhost-shadow-virtqueue.h | 4 ++++
38
net/net.c | 4 ++++
12
hw/virtio/vhost-vdpa.c | 31 +++++++++++++++++++++++++++++--
39
3 files changed, 15 insertions(+)
13
3 files changed, 71 insertions(+), 2 deletions(-)
40
14
41
diff --git a/net/colo-compare.c b/net/colo-compare.c
15
diff --git a/hw/virtio/vhost-shadow-virtqueue.c b/hw/virtio/vhost-shadow-virtqueue.c
42
index XXXXXXX..XXXXXXX 100644
16
index XXXXXXX..XXXXXXX 100644
43
--- a/net/colo-compare.c
17
--- a/hw/virtio/vhost-shadow-virtqueue.c
44
+++ b/net/colo-compare.c
18
+++ b/hw/virtio/vhost-shadow-virtqueue.c
45
@@ -XXX,XX +XXX,XX @@ static void colo_compare_init(Object *obj)
19
@@ -XXX,XX +XXX,XX @@ static void vhost_handle_guest_kick(EventNotifier *n)
46
compare_set_vnet_hdr);
47
}
20
}
48
21
49
+void colo_compare_cleanup(void)
22
/**
23
+ * Forward vhost notifications
24
+ *
25
+ * @n: hdev call event notifier, the one that device set to notify svq.
26
+ */
27
+static void vhost_svq_handle_call(EventNotifier *n)
50
+{
28
+{
51
+ CompareState *tmp = NULL;
29
+ VhostShadowVirtqueue *svq = container_of(n, VhostShadowVirtqueue,
52
+ CompareState *n = NULL;
30
+ hdev_call);
31
+ event_notifier_test_and_clear(n);
32
+ event_notifier_set(&svq->svq_call);
33
+}
53
+
34
+
54
+ QTAILQ_FOREACH_SAFE(tmp, &net_compares, next, n) {
35
+/**
55
+ object_unparent(OBJECT(tmp));
36
+ * Set the call notifier for the SVQ to call the guest
37
+ *
38
+ * @svq: Shadow virtqueue
39
+ * @call_fd: call notifier
40
+ *
41
+ * Called on BQL context.
42
+ */
43
+void vhost_svq_set_svq_call_fd(VhostShadowVirtqueue *svq, int call_fd)
44
+{
45
+ if (call_fd == VHOST_FILE_UNBIND) {
46
+ /*
47
+ * Fail event_notifier_set if called handling device call.
48
+ *
49
+ * SVQ still needs device notifications, since it needs to keep
50
+ * forwarding used buffers even with the unbind.
51
+ */
52
+ memset(&svq->svq_call, 0, sizeof(svq->svq_call));
53
+ } else {
54
+ event_notifier_init_fd(&svq->svq_call, call_fd);
56
+ }
55
+ }
57
+}
56
+}
58
+
57
+
59
static void colo_compare_finalize(Object *obj)
58
+/**
59
* Set a new file descriptor for the guest to kick the SVQ and notify for avail
60
*
61
* @svq: The svq
62
@@ -XXX,XX +XXX,XX @@ VhostShadowVirtqueue *vhost_svq_new(void)
63
}
64
65
event_notifier_init_fd(&svq->svq_kick, VHOST_FILE_UNBIND);
66
+ event_notifier_set_handler(&svq->hdev_call, vhost_svq_handle_call);
67
return g_steal_pointer(&svq);
68
69
err_init_hdev_call:
70
@@ -XXX,XX +XXX,XX @@ void vhost_svq_free(gpointer pvq)
71
VhostShadowVirtqueue *vq = pvq;
72
vhost_svq_stop(vq);
73
event_notifier_cleanup(&vq->hdev_kick);
74
+ event_notifier_set_handler(&vq->hdev_call, NULL);
75
event_notifier_cleanup(&vq->hdev_call);
76
g_free(vq);
77
}
78
diff --git a/hw/virtio/vhost-shadow-virtqueue.h b/hw/virtio/vhost-shadow-virtqueue.h
79
index XXXXXXX..XXXXXXX 100644
80
--- a/hw/virtio/vhost-shadow-virtqueue.h
81
+++ b/hw/virtio/vhost-shadow-virtqueue.h
82
@@ -XXX,XX +XXX,XX @@ typedef struct VhostShadowVirtqueue {
83
* So shadow virtqueue must not clean it, or we would lose VirtQueue one.
84
*/
85
EventNotifier svq_kick;
86
+
87
+ /* Guest's call notifier, where the SVQ calls guest. */
88
+ EventNotifier svq_call;
89
} VhostShadowVirtqueue;
90
91
void vhost_svq_set_svq_kick_fd(VhostShadowVirtqueue *svq, int svq_kick_fd);
92
+void vhost_svq_set_svq_call_fd(VhostShadowVirtqueue *svq, int call_fd);
93
94
void vhost_svq_stop(VhostShadowVirtqueue *svq);
95
96
diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c
97
index XXXXXXX..XXXXXXX 100644
98
--- a/hw/virtio/vhost-vdpa.c
99
+++ b/hw/virtio/vhost-vdpa.c
100
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_set_vring_dev_kick(struct vhost_dev *dev,
101
return vhost_vdpa_call(dev, VHOST_SET_VRING_KICK, file);
102
}
103
104
+static int vhost_vdpa_set_vring_dev_call(struct vhost_dev *dev,
105
+ struct vhost_vring_file *file)
106
+{
107
+ trace_vhost_vdpa_set_vring_call(dev, file->index, file->fd);
108
+ return vhost_vdpa_call(dev, VHOST_SET_VRING_CALL, file);
109
+}
110
+
111
/**
112
* Set the shadow virtqueue descriptors to the device
113
*
114
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_set_vring_dev_kick(struct vhost_dev *dev,
115
* @svq: The shadow virtqueue
116
* @idx: The index of the virtqueue in the vhost device
117
* @errp: Error
118
+ *
119
+ * Note that this function does not rewind kick file descriptor if cannot set
120
+ * call one.
121
*/
122
static bool vhost_vdpa_svq_setup(struct vhost_dev *dev,
123
VhostShadowVirtqueue *svq,
124
@@ -XXX,XX +XXX,XX @@ static bool vhost_vdpa_svq_setup(struct vhost_dev *dev,
125
r = vhost_vdpa_set_vring_dev_kick(dev, &file);
126
if (unlikely(r != 0)) {
127
error_setg_errno(errp, -r, "Can't set device kick fd");
128
+ return false;
129
+ }
130
+
131
+ event_notifier = &svq->hdev_call;
132
+ file.fd = event_notifier_get_fd(event_notifier);
133
+ r = vhost_vdpa_set_vring_dev_call(dev, &file);
134
+ if (unlikely(r != 0)) {
135
+ error_setg_errno(errp, -r, "Can't set device call fd");
136
}
137
138
return r == 0;
139
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_set_vring_kick(struct vhost_dev *dev,
140
static int vhost_vdpa_set_vring_call(struct vhost_dev *dev,
141
struct vhost_vring_file *file)
60
{
142
{
61
CompareState *s = COLO_COMPARE(obj);
143
- trace_vhost_vdpa_set_vring_call(dev, file->index, file->fd);
62
diff --git a/net/colo-compare.h b/net/colo-compare.h
144
- return vhost_vdpa_call(dev, VHOST_SET_VRING_CALL, file);
63
index XXXXXXX..XXXXXXX 100644
145
+ struct vhost_vdpa *v = dev->opaque;
64
--- a/net/colo-compare.h
65
+++ b/net/colo-compare.h
66
@@ -XXX,XX +XXX,XX @@
67
void colo_notify_compares_event(void *opaque, int event, Error **errp);
68
void colo_compare_register_notifier(Notifier *notify);
69
void colo_compare_unregister_notifier(Notifier *notify);
70
+void colo_compare_cleanup(void);
71
72
#endif /* QEMU_COLO_COMPARE_H */
73
diff --git a/net/net.c b/net/net.c
74
index XXXXXXX..XXXXXXX 100644
75
--- a/net/net.c
76
+++ b/net/net.c
77
@@ -XXX,XX +XXX,XX @@
78
#include "qapi/error.h"
79
#include "qapi/opts-visitor.h"
80
#include "sysemu/runstate.h"
81
+#include "net/colo-compare.h"
82
#include "net/filter.h"
83
#include "qapi/string-output-visitor.h"
84
85
@@ -XXX,XX +XXX,XX @@ void net_cleanup(void)
86
{
87
NetClientState *nc;
88
89
+ /*cleanup colo compare module for COLO*/
90
+ colo_compare_cleanup();
91
+
146
+
92
/* We may del multiple entries during qemu_del_net_client(),
147
+ if (v->shadow_vqs_enabled) {
93
* so QTAILQ_FOREACH_SAFE() is also not safe here.
148
+ int vdpa_idx = file->index - dev->vq_index;
94
*/
149
+ VhostShadowVirtqueue *svq = g_ptr_array_index(v->shadow_vqs, vdpa_idx);
150
+
151
+ vhost_svq_set_svq_call_fd(svq, file->fd);
152
+ return 0;
153
+ } else {
154
+ return vhost_vdpa_set_vring_dev_call(dev, file);
155
+ }
156
}
157
158
static int vhost_vdpa_get_features(struct vhost_dev *dev,
95
--
159
--
96
2.7.4
160
2.7.4
97
161
98
162
diff view generated by jsdifflib
[PULL 10/12] Add a function named packet_new_nocopy for COLO.
[PULL V2 05/15] vhost: Add vhost_svq_valid_features to shadow vq
1
From: "Rao, Lei" <lei.rao@intel.com>
1
From: Eugenio Pérez <eperezma@redhat.com>
2
2
3
Use the packet_new_nocopy instead of packet_new in the
3
This allows SVQ to negotiate features with the guest and the device. For
4
filter-rewriter module. There will be one less memory
4
the device, SVQ is a driver. While this function bypasses all
5
copy in the processing of each network packet.
5
non-transport features, it needs to disable the features that SVQ does
6
not support when forwarding buffers. This includes packed vq layout,
7
indirect descriptors or event idx.
6
8
7
Signed-off-by: Lei Rao <lei.rao@intel.com>
9
Future changes can add support to offer more features to the guest,
8
Signed-off-by: Zhang Chen <chen.zhang@intel.com>
10
since the use of VirtQueue gives this for free. This is left out at the
9
Reviewed-by: Zhang Chen <chen.zhang@intel.com>
11
moment for simplicity.
12
13
Acked-by: Michael S. Tsirkin <mst@redhat.com>
14
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
10
Signed-off-by: Jason Wang <jasowang@redhat.com>
15
Signed-off-by: Jason Wang <jasowang@redhat.com>
11
---
16
---
12
net/colo.c | 25 +++++++++++++++++--------
17
hw/virtio/vhost-shadow-virtqueue.c | 44 ++++++++++++++++++++++++++++++++++++++
13
net/colo.h | 1 +
18
hw/virtio/vhost-shadow-virtqueue.h | 2 ++
14
net/filter-rewriter.c | 3 +--
19
hw/virtio/vhost-vdpa.c | 15 +++++++++++++
15
3 files changed, 19 insertions(+), 10 deletions(-)
20
3 files changed, 61 insertions(+)
16
21
17
diff --git a/net/colo.c b/net/colo.c
22
diff --git a/hw/virtio/vhost-shadow-virtqueue.c b/hw/virtio/vhost-shadow-virtqueue.c
18
index XXXXXXX..XXXXXXX 100644
23
index XXXXXXX..XXXXXXX 100644
19
--- a/net/colo.c
24
--- a/hw/virtio/vhost-shadow-virtqueue.c
20
+++ b/net/colo.c
25
+++ b/hw/virtio/vhost-shadow-virtqueue.c
21
@@ -XXX,XX +XXX,XX @@ void connection_destroy(void *opaque)
26
@@ -XXX,XX +XXX,XX @@
22
27
#include "hw/virtio/vhost-shadow-virtqueue.h"
23
Packet *packet_new(const void *data, int size, int vnet_hdr_len)
28
24
{
29
#include "qemu/error-report.h"
25
- Packet *pkt = g_slice_new(Packet);
30
+#include "qapi/error.h"
26
+ Packet *pkt = g_slice_new0(Packet);
31
#include "qemu/main-loop.h"
27
32
#include "linux-headers/linux/vhost.h"
28
pkt->data = g_memdup(data, size);
33
29
pkt->size = size;
34
/**
30
pkt->creation_ms = qemu_clock_get_ms(QEMU_CLOCK_HOST);
35
+ * Validate the transport device features that both guests can use with the SVQ
31
pkt->vnet_hdr_len = vnet_hdr_len;
36
+ * and SVQs can use with the device.
32
- pkt->tcp_seq = 0;
37
+ *
33
- pkt->tcp_ack = 0;
38
+ * @dev_features: The features
34
- pkt->seq_end = 0;
39
+ * @errp: Error pointer
35
- pkt->header_size = 0;
40
+ */
36
- pkt->payload_size = 0;
41
+bool vhost_svq_valid_features(uint64_t features, Error **errp)
37
- pkt->offset = 0;
42
+{
38
- pkt->flags = 0;
43
+ bool ok = true;
44
+ uint64_t svq_features = features;
39
+
45
+
40
+ return pkt;
46
+ for (uint64_t b = VIRTIO_TRANSPORT_F_START; b <= VIRTIO_TRANSPORT_F_END;
47
+ ++b) {
48
+ switch (b) {
49
+ case VIRTIO_F_ANY_LAYOUT:
50
+ continue;
51
+
52
+ case VIRTIO_F_ACCESS_PLATFORM:
53
+ /* SVQ trust in the host's IOMMU to translate addresses */
54
+ case VIRTIO_F_VERSION_1:
55
+ /* SVQ trust that the guest vring is little endian */
56
+ if (!(svq_features & BIT_ULL(b))) {
57
+ set_bit(b, &svq_features);
58
+ ok = false;
59
+ }
60
+ continue;
61
+
62
+ default:
63
+ if (svq_features & BIT_ULL(b)) {
64
+ clear_bit(b, &svq_features);
65
+ ok = false;
66
+ }
67
+ }
68
+ }
69
+
70
+ if (!ok) {
71
+ error_setg(errp, "SVQ Invalid device feature flags, offer: 0x%"PRIx64
72
+ ", ok: 0x%"PRIx64, features, svq_features);
73
+ }
74
+ return ok;
41
+}
75
+}
42
+
76
+
43
+/*
77
+/**
44
+ * packet_new_nocopy will not copy data, so the caller can't release
78
* Forward guest notifications.
45
+ * the data. And it will be released in packet_destroy.
79
*
46
+ */
80
* @n: guest kick event notifier, the one that guest set to notify svq.
47
+Packet *packet_new_nocopy(void *data, int size, int vnet_hdr_len)
81
diff --git a/hw/virtio/vhost-shadow-virtqueue.h b/hw/virtio/vhost-shadow-virtqueue.h
48
+{
82
index XXXXXXX..XXXXXXX 100644
49
+ Packet *pkt = g_slice_new0(Packet);
83
--- a/hw/virtio/vhost-shadow-virtqueue.h
84
+++ b/hw/virtio/vhost-shadow-virtqueue.h
85
@@ -XXX,XX +XXX,XX @@ typedef struct VhostShadowVirtqueue {
86
EventNotifier svq_call;
87
} VhostShadowVirtqueue;
88
89
+bool vhost_svq_valid_features(uint64_t features, Error **errp);
50
+
90
+
51
+ pkt->data = data;
91
void vhost_svq_set_svq_kick_fd(VhostShadowVirtqueue *svq, int svq_kick_fd);
52
+ pkt->size = size;
92
void vhost_svq_set_svq_call_fd(VhostShadowVirtqueue *svq, int call_fd);
53
+ pkt->creation_ms = qemu_clock_get_ms(QEMU_CLOCK_HOST);
93
54
+ pkt->vnet_hdr_len = vnet_hdr_len;
94
diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c
55
56
return pkt;
57
}
58
diff --git a/net/colo.h b/net/colo.h
59
index XXXXXXX..XXXXXXX 100644
95
index XXXXXXX..XXXXXXX 100644
60
--- a/net/colo.h
96
--- a/hw/virtio/vhost-vdpa.c
61
+++ b/net/colo.h
97
+++ b/hw/virtio/vhost-vdpa.c
62
@@ -XXX,XX +XXX,XX @@ bool connection_has_tracked(GHashTable *connection_track_table,
98
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_init_svq(struct vhost_dev *hdev, struct vhost_vdpa *v,
63
ConnectionKey *key);
99
Error **errp)
64
void connection_hashtable_reset(GHashTable *connection_track_table);
100
{
65
Packet *packet_new(const void *data, int size, int vnet_hdr_len);
101
g_autoptr(GPtrArray) shadow_vqs = NULL;
66
+Packet *packet_new_nocopy(void *data, int size, int vnet_hdr_len);
102
+ uint64_t dev_features, svq_features;
67
void packet_destroy(void *opaque, void *user_data);
103
+ int r;
68
void packet_destroy_partial(void *opaque, void *user_data);
104
+ bool ok;
69
105
70
diff --git a/net/filter-rewriter.c b/net/filter-rewriter.c
106
if (!v->shadow_vqs_enabled) {
71
index XXXXXXX..XXXXXXX 100644
107
return 0;
72
--- a/net/filter-rewriter.c
73
+++ b/net/filter-rewriter.c
74
@@ -XXX,XX +XXX,XX @@ static ssize_t colo_rewriter_receive_iov(NetFilterState *nf,
75
vnet_hdr_len = nf->netdev->vnet_hdr_len;
76
}
108
}
77
109
78
- pkt = packet_new(buf, size, vnet_hdr_len);
110
+ r = hdev->vhost_ops->vhost_get_features(hdev, &dev_features);
79
- g_free(buf);
111
+ if (r != 0) {
80
+ pkt = packet_new_nocopy(buf, size, vnet_hdr_len);
112
+ error_setg_errno(errp, -r, "Can't get vdpa device features");
81
113
+ return r;
82
/*
114
+ }
83
* if we get tcp packet
115
+
116
+ svq_features = dev_features;
117
+ ok = vhost_svq_valid_features(svq_features, errp);
118
+ if (unlikely(!ok)) {
119
+ return -1;
120
+ }
121
+
122
shadow_vqs = g_ptr_array_new_full(hdev->nvqs, vhost_svq_free);
123
for (unsigned n = 0; n < hdev->nvqs; ++n) {
124
g_autoptr(VhostShadowVirtqueue) svq = vhost_svq_new();
84
--
125
--
85
2.7.4
126
2.7.4
86
127
87
128
diff view generated by jsdifflib
New patch
[PULL V2 06/15] virtio: Add vhost_svq_get_vring_addr
1
From: Eugenio Pérez <eperezma@redhat.com>
1
2
3
It reports the shadow virtqueue address from qemu virtual address space.
4
5
Since this will be different from the guest's vaddr, but the device can
6
access it, SVQ takes special care about its alignment & lack of garbage
7
data. It assumes that IOMMU will work in host_page_size ranges for that.
8
9
Acked-by: Michael S. Tsirkin <mst@redhat.com>
10
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
11
Signed-off-by: Jason Wang <jasowang@redhat.com>
12
---
13
hw/virtio/vhost-shadow-virtqueue.c | 29 +++++++++++++++++++++++++++++
14
hw/virtio/vhost-shadow-virtqueue.h | 9 +++++++++
15
2 files changed, 38 insertions(+)
16
17
diff --git a/hw/virtio/vhost-shadow-virtqueue.c b/hw/virtio/vhost-shadow-virtqueue.c
18
index XXXXXXX..XXXXXXX 100644
19
--- a/hw/virtio/vhost-shadow-virtqueue.c
20
+++ b/hw/virtio/vhost-shadow-virtqueue.c
21
@@ -XXX,XX +XXX,XX @@ void vhost_svq_set_svq_call_fd(VhostShadowVirtqueue *svq, int call_fd)
22
}
23
24
/**
25
+ * Get the shadow vq vring address.
26
+ * @svq: Shadow virtqueue
27
+ * @addr: Destination to store address
28
+ */
29
+void vhost_svq_get_vring_addr(const VhostShadowVirtqueue *svq,
30
+ struct vhost_vring_addr *addr)
31
+{
32
+ addr->desc_user_addr = (uint64_t)svq->vring.desc;
33
+ addr->avail_user_addr = (uint64_t)svq->vring.avail;
34
+ addr->used_user_addr = (uint64_t)svq->vring.used;
35
+}
36
+
37
+size_t vhost_svq_driver_area_size(const VhostShadowVirtqueue *svq)
38
+{
39
+ size_t desc_size = sizeof(vring_desc_t) * svq->vring.num;
40
+ size_t avail_size = offsetof(vring_avail_t, ring) +
41
+ sizeof(uint16_t) * svq->vring.num;
42
+
43
+ return ROUND_UP(desc_size + avail_size, qemu_real_host_page_size);
44
+}
45
+
46
+size_t vhost_svq_device_area_size(const VhostShadowVirtqueue *svq)
47
+{
48
+ size_t used_size = offsetof(vring_used_t, ring) +
49
+ sizeof(vring_used_elem_t) * svq->vring.num;
50
+ return ROUND_UP(used_size, qemu_real_host_page_size);
51
+}
52
+
53
+/**
54
* Set a new file descriptor for the guest to kick the SVQ and notify for avail
55
*
56
* @svq: The svq
57
diff --git a/hw/virtio/vhost-shadow-virtqueue.h b/hw/virtio/vhost-shadow-virtqueue.h
58
index XXXXXXX..XXXXXXX 100644
59
--- a/hw/virtio/vhost-shadow-virtqueue.h
60
+++ b/hw/virtio/vhost-shadow-virtqueue.h
61
@@ -XXX,XX +XXX,XX @@
62
#define VHOST_SHADOW_VIRTQUEUE_H
63
64
#include "qemu/event_notifier.h"
65
+#include "hw/virtio/virtio.h"
66
+#include "standard-headers/linux/vhost_types.h"
67
68
/* Shadow virtqueue to relay notifications */
69
typedef struct VhostShadowVirtqueue {
70
+ /* Shadow vring */
71
+ struct vring vring;
72
+
73
/* Shadow kick notifier, sent to vhost */
74
EventNotifier hdev_kick;
75
/* Shadow call notifier, sent to vhost */
76
@@ -XXX,XX +XXX,XX @@ bool vhost_svq_valid_features(uint64_t features, Error **errp);
77
78
void vhost_svq_set_svq_kick_fd(VhostShadowVirtqueue *svq, int svq_kick_fd);
79
void vhost_svq_set_svq_call_fd(VhostShadowVirtqueue *svq, int call_fd);
80
+void vhost_svq_get_vring_addr(const VhostShadowVirtqueue *svq,
81
+ struct vhost_vring_addr *addr);
82
+size_t vhost_svq_driver_area_size(const VhostShadowVirtqueue *svq);
83
+size_t vhost_svq_device_area_size(const VhostShadowVirtqueue *svq);
84
85
void vhost_svq_stop(VhostShadowVirtqueue *svq);
86
87
--
88
2.7.4
89
90
diff view generated by jsdifflib
[PULL 12/12] Fixed calculation error of pkt->header_size in fill_pkt_tcp_info()
[PULL V2 07/15] vdpa: adapt vhost_ops callbacks to svq
1
From: "Rao, Lei" <lei.rao@intel.com>
1
From: Eugenio Pérez <eperezma@redhat.com>
2
2
3
The data pointer has skipped vnet_hdr_len in the function of
3
First half of the buffers forwarding part, preparing vhost-vdpa
4
parse_packet_early().So, we can not subtract vnet_hdr_len again
4
callbacks to SVQ to offer it. QEMU cannot enable it at this moment, so
5
when calculating pkt->header_size in fill_pkt_tcp_info(). Otherwise,
5
this is effectively dead code at the moment, but it helps to reduce
6
it will cause network packet comparsion errors and greatly increase
6
patch size.
7
the frequency of checkpoints.
8
7
9
Signed-off-by: Lei Rao <lei.rao@intel.com>
8
Acked-by: Michael S. Tsirkin <mst@redhat.com>
10
Signed-off-by: Zhang Chen <chen.zhang@intel.com>
9
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
11
Reviewed-by: Li Zhijian <lizhijian@fujitsu.com>
12
Reviewed-by: Zhang Chen <chen.zhang@intel.com>
13
Reviewed-by: Lukas Straub <lukasstraub2@web.de>
14
Tested-by: Lukas Straub <lukasstraub2@web.de>
15
Signed-off-by: Jason Wang <jasowang@redhat.com>
10
Signed-off-by: Jason Wang <jasowang@redhat.com>
16
---
11
---
17
net/colo-compare.c | 2 +-
12
hw/virtio/vhost-vdpa.c | 48 +++++++++++++++++++++++++++++++++++++++++-------
18
1 file changed, 1 insertion(+), 1 deletion(-)
13
1 file changed, 41 insertions(+), 7 deletions(-)
19
14
20
diff --git a/net/colo-compare.c b/net/colo-compare.c
15
diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c
21
index XXXXXXX..XXXXXXX 100644
16
index XXXXXXX..XXXXXXX 100644
22
--- a/net/colo-compare.c
17
--- a/hw/virtio/vhost-vdpa.c
23
+++ b/net/colo-compare.c
18
+++ b/hw/virtio/vhost-vdpa.c
24
@@ -XXX,XX +XXX,XX @@ static void fill_pkt_tcp_info(void *data, uint32_t *max_ack)
19
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_get_config(struct vhost_dev *dev, uint8_t *config,
25
pkt->tcp_ack = ntohl(tcphd->th_ack);
20
return ret;
26
*max_ack = *max_ack > pkt->tcp_ack ? *max_ack : pkt->tcp_ack;
21
}
27
pkt->header_size = pkt->transport_header - (uint8_t *)pkt->data
22
28
- + (tcphd->th_off << 2) - pkt->vnet_hdr_len;
23
+static int vhost_vdpa_set_dev_vring_base(struct vhost_dev *dev,
29
+ + (tcphd->th_off << 2);
24
+ struct vhost_vring_state *ring)
30
pkt->payload_size = pkt->size - pkt->header_size;
25
+{
31
pkt->seq_end = pkt->tcp_seq + pkt->payload_size;
26
+ trace_vhost_vdpa_set_vring_base(dev, ring->index, ring->num);
32
pkt->flags = tcphd->th_flags;
27
+ return vhost_vdpa_call(dev, VHOST_SET_VRING_BASE, ring);
28
+}
29
+
30
static int vhost_vdpa_set_vring_dev_kick(struct vhost_dev *dev,
31
struct vhost_vring_file *file)
32
{
33
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_set_vring_dev_call(struct vhost_dev *dev,
34
return vhost_vdpa_call(dev, VHOST_SET_VRING_CALL, file);
35
}
36
37
+static int vhost_vdpa_set_vring_dev_addr(struct vhost_dev *dev,
38
+ struct vhost_vring_addr *addr)
39
+{
40
+ trace_vhost_vdpa_set_vring_addr(dev, addr->index, addr->flags,
41
+ addr->desc_user_addr, addr->used_user_addr,
42
+ addr->avail_user_addr,
43
+ addr->log_guest_addr);
44
+
45
+ return vhost_vdpa_call(dev, VHOST_SET_VRING_ADDR, addr);
46
+
47
+}
48
+
49
/**
50
* Set the shadow virtqueue descriptors to the device
51
*
52
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_set_log_base(struct vhost_dev *dev, uint64_t base,
53
static int vhost_vdpa_set_vring_addr(struct vhost_dev *dev,
54
struct vhost_vring_addr *addr)
55
{
56
- trace_vhost_vdpa_set_vring_addr(dev, addr->index, addr->flags,
57
- addr->desc_user_addr, addr->used_user_addr,
58
- addr->avail_user_addr,
59
- addr->log_guest_addr);
60
- return vhost_vdpa_call(dev, VHOST_SET_VRING_ADDR, addr);
61
+ struct vhost_vdpa *v = dev->opaque;
62
+
63
+ if (v->shadow_vqs_enabled) {
64
+ /*
65
+ * Device vring addr was set at device start. SVQ base is handled by
66
+ * VirtQueue code.
67
+ */
68
+ return 0;
69
+ }
70
+
71
+ return vhost_vdpa_set_vring_dev_addr(dev, addr);
72
}
73
74
static int vhost_vdpa_set_vring_num(struct vhost_dev *dev,
75
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_set_vring_num(struct vhost_dev *dev,
76
static int vhost_vdpa_set_vring_base(struct vhost_dev *dev,
77
struct vhost_vring_state *ring)
78
{
79
- trace_vhost_vdpa_set_vring_base(dev, ring->index, ring->num);
80
- return vhost_vdpa_call(dev, VHOST_SET_VRING_BASE, ring);
81
+ struct vhost_vdpa *v = dev->opaque;
82
+
83
+ if (v->shadow_vqs_enabled) {
84
+ /*
85
+ * Device vring base was set at device start. SVQ base is handled by
86
+ * VirtQueue code.
87
+ */
88
+ return 0;
89
+ }
90
+
91
+ return vhost_vdpa_set_dev_vring_base(dev, ring);
92
}
93
94
static int vhost_vdpa_get_vring_base(struct vhost_dev *dev,
33
--
95
--
34
2.7.4
96
2.7.4
35
97
36
98
diff view generated by jsdifflib
[PULL 09/12] Remove migrate_set_block_enabled in checkpoint
[PULL V2 08/15] vhost: Shadow virtqueue buffers forwarding
1
From: "Rao, Lei" <lei.rao@intel.com>
1
From: Eugenio Pérez <eperezma@redhat.com>
2
2
3
We can detect disk migration in migrate_prepare, if disk migration
3
Initial version of shadow virtqueue that actually forward buffers. There
4
is enabled in COLO mode, we can directly report an error.and there
4
is no iommu support at the moment, and that will be addressed in future
5
is no need to disable block migration at every checkpoint.
5
patches of this series. Since all vhost-vdpa devices use forced IOMMU,
6
6
this means that SVQ is not usable at this point of the series on any
7
Signed-off-by: Lei Rao <lei.rao@intel.com>
7
device.
8
Signed-off-by: Zhang Chen <chen.zhang@intel.com>
8
9
Reviewed-by: Li Zhijian <lizhijian@fujitsu.com>
9
For simplicity it only supports modern devices, that expects vring
10
Reviewed-by: Zhang Chen <chen.zhang@intel.com>
10
in little endian, with split ring and no event idx or indirect
11
Reviewed-by: Lukas Straub <lukasstraub2@web.de>
11
descriptors. Support for them will not be added in this series.
12
Tested-by: Lukas Straub <lukasstraub2@web.de>
12
13
It reuses the VirtQueue code for the device part. The driver part is
14
based on Linux's virtio_ring driver, but with stripped functionality
15
and optimizations so it's easier to review.
16
17
However, forwarding buffers have some particular pieces: One of the most
18
unexpected ones is that a guest's buffer can expand through more than
19
one descriptor in SVQ. While this is handled gracefully by qemu's
20
emulated virtio devices, it may cause unexpected SVQ queue full. This
21
patch also solves it by checking for this condition at both guest's
22
kicks and device's calls. The code may be more elegant in the future if
23
SVQ code runs in its own iocontext.
24
25
Acked-by: Michael S. Tsirkin <mst@redhat.com>
26
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
13
Signed-off-by: Jason Wang <jasowang@redhat.com>
27
Signed-off-by: Jason Wang <jasowang@redhat.com>
14
---
28
---
15
migration/colo.c | 6 ------
29
hw/virtio/vhost-shadow-virtqueue.c | 354 ++++++++++++++++++++++++++++++++++++-
16
migration/migration.c | 4 ++++
30
hw/virtio/vhost-shadow-virtqueue.h | 26 +++
17
2 files changed, 4 insertions(+), 6 deletions(-)
31
hw/virtio/vhost-vdpa.c | 159 ++++++++++++++++-
18
32
3 files changed, 527 insertions(+), 12 deletions(-)
19
diff --git a/migration/colo.c b/migration/colo.c
33
34
diff --git a/hw/virtio/vhost-shadow-virtqueue.c b/hw/virtio/vhost-shadow-virtqueue.c
20
index XXXXXXX..XXXXXXX 100644
35
index XXXXXXX..XXXXXXX 100644
21
--- a/migration/colo.c
36
--- a/hw/virtio/vhost-shadow-virtqueue.c
22
+++ b/migration/colo.c
37
+++ b/hw/virtio/vhost-shadow-virtqueue.c
23
@@ -XXX,XX +XXX,XX @@ static int colo_do_checkpoint_transaction(MigrationState *s,
38
@@ -XXX,XX +XXX,XX @@
24
if (failover_get_state() != FAILOVER_STATUS_NONE) {
39
#include "qemu/error-report.h"
25
goto out;
40
#include "qapi/error.h"
41
#include "qemu/main-loop.h"
42
+#include "qemu/log.h"
43
+#include "qemu/memalign.h"
44
#include "linux-headers/linux/vhost.h"
45
46
/**
47
@@ -XXX,XX +XXX,XX @@ bool vhost_svq_valid_features(uint64_t features, Error **errp)
48
}
49
50
/**
51
- * Forward guest notifications.
52
+ * Number of descriptors that the SVQ can make available from the guest.
53
+ *
54
+ * @svq: The svq
55
+ */
56
+static uint16_t vhost_svq_available_slots(const VhostShadowVirtqueue *svq)
57
+{
58
+ return svq->vring.num - (svq->shadow_avail_idx - svq->shadow_used_idx);
59
+}
60
+
61
+static void vhost_vring_write_descs(VhostShadowVirtqueue *svq,
62
+ const struct iovec *iovec,
63
+ size_t num, bool more_descs, bool write)
64
+{
65
+ uint16_t i = svq->free_head, last = svq->free_head;
66
+ unsigned n;
67
+ uint16_t flags = write ? cpu_to_le16(VRING_DESC_F_WRITE) : 0;
68
+ vring_desc_t *descs = svq->vring.desc;
69
+
70
+ if (num == 0) {
71
+ return;
72
+ }
73
+
74
+ for (n = 0; n < num; n++) {
75
+ if (more_descs || (n + 1 < num)) {
76
+ descs[i].flags = flags | cpu_to_le16(VRING_DESC_F_NEXT);
77
+ } else {
78
+ descs[i].flags = flags;
79
+ }
80
+ descs[i].addr = cpu_to_le64((hwaddr)iovec[n].iov_base);
81
+ descs[i].len = cpu_to_le32(iovec[n].iov_len);
82
+
83
+ last = i;
84
+ i = cpu_to_le16(descs[i].next);
85
+ }
86
+
87
+ svq->free_head = le16_to_cpu(descs[last].next);
88
+}
89
+
90
+static bool vhost_svq_add_split(VhostShadowVirtqueue *svq,
91
+ VirtQueueElement *elem,
92
+ unsigned *head)
93
+{
94
+ unsigned avail_idx;
95
+ vring_avail_t *avail = svq->vring.avail;
96
+
97
+ *head = svq->free_head;
98
+
99
+ /* We need some descriptors here */
100
+ if (unlikely(!elem->out_num && !elem->in_num)) {
101
+ qemu_log_mask(LOG_GUEST_ERROR,
102
+ "Guest provided element with no descriptors");
103
+ return false;
104
+ }
105
+
106
+ vhost_vring_write_descs(svq, elem->out_sg, elem->out_num,
107
+ elem->in_num > 0, false);
108
+ vhost_vring_write_descs(svq, elem->in_sg, elem->in_num, false, true);
109
+
110
+ /*
111
+ * Put the entry in the available array (but don't update avail->idx until
112
+ * they do sync).
113
+ */
114
+ avail_idx = svq->shadow_avail_idx & (svq->vring.num - 1);
115
+ avail->ring[avail_idx] = cpu_to_le16(*head);
116
+ svq->shadow_avail_idx++;
117
+
118
+ /* Update the avail index after write the descriptor */
119
+ smp_wmb();
120
+ avail->idx = cpu_to_le16(svq->shadow_avail_idx);
121
+
122
+ return true;
123
+}
124
+
125
+static bool vhost_svq_add(VhostShadowVirtqueue *svq, VirtQueueElement *elem)
126
+{
127
+ unsigned qemu_head;
128
+ bool ok = vhost_svq_add_split(svq, elem, &qemu_head);
129
+ if (unlikely(!ok)) {
130
+ return false;
131
+ }
132
+
133
+ svq->ring_id_maps[qemu_head] = elem;
134
+ return true;
135
+}
136
+
137
+static void vhost_svq_kick(VhostShadowVirtqueue *svq)
138
+{
139
+ /*
140
+ * We need to expose the available array entries before checking the used
141
+ * flags
142
+ */
143
+ smp_mb();
144
+ if (svq->vring.used->flags & VRING_USED_F_NO_NOTIFY) {
145
+ return;
146
+ }
147
+
148
+ event_notifier_set(&svq->hdev_kick);
149
+}
150
+
151
+/**
152
+ * Forward available buffers.
153
+ *
154
+ * @svq: Shadow VirtQueue
155
+ *
156
+ * Note that this function does not guarantee that all guest's available
157
+ * buffers are available to the device in SVQ avail ring. The guest may have
158
+ * exposed a GPA / GIOVA contiguous buffer, but it may not be contiguous in
159
+ * qemu vaddr.
160
+ *
161
+ * If that happens, guest's kick notifications will be disabled until the
162
+ * device uses some buffers.
163
+ */
164
+static void vhost_handle_guest_kick(VhostShadowVirtqueue *svq)
165
+{
166
+ /* Clear event notifier */
167
+ event_notifier_test_and_clear(&svq->svq_kick);
168
+
169
+ /* Forward to the device as many available buffers as possible */
170
+ do {
171
+ virtio_queue_set_notification(svq->vq, false);
172
+
173
+ while (true) {
174
+ VirtQueueElement *elem;
175
+ bool ok;
176
+
177
+ if (svq->next_guest_avail_elem) {
178
+ elem = g_steal_pointer(&svq->next_guest_avail_elem);
179
+ } else {
180
+ elem = virtqueue_pop(svq->vq, sizeof(*elem));
181
+ }
182
+
183
+ if (!elem) {
184
+ break;
185
+ }
186
+
187
+ if (elem->out_num + elem->in_num >
188
+ vhost_svq_available_slots(svq)) {
189
+ /*
190
+ * This condition is possible since a contiguous buffer in GPA
191
+ * does not imply a contiguous buffer in qemu's VA
192
+ * scatter-gather segments. If that happens, the buffer exposed
193
+ * to the device needs to be a chain of descriptors at this
194
+ * moment.
195
+ *
196
+ * SVQ cannot hold more available buffers if we are here:
197
+ * queue the current guest descriptor and ignore further kicks
198
+ * until some elements are used.
199
+ */
200
+ svq->next_guest_avail_elem = elem;
201
+ return;
202
+ }
203
+
204
+ ok = vhost_svq_add(svq, elem);
205
+ if (unlikely(!ok)) {
206
+ /* VQ is broken, just return and ignore any other kicks */
207
+ return;
208
+ }
209
+ vhost_svq_kick(svq);
210
+ }
211
+
212
+ virtio_queue_set_notification(svq->vq, true);
213
+ } while (!virtio_queue_empty(svq->vq));
214
+}
215
+
216
+/**
217
+ * Handle guest's kick.
218
*
219
* @n: guest kick event notifier, the one that guest set to notify svq.
220
*/
221
-static void vhost_handle_guest_kick(EventNotifier *n)
222
+static void vhost_handle_guest_kick_notifier(EventNotifier *n)
223
{
224
VhostShadowVirtqueue *svq = container_of(n, VhostShadowVirtqueue,
225
svq_kick);
226
event_notifier_test_and_clear(n);
227
- event_notifier_set(&svq->hdev_kick);
228
+ vhost_handle_guest_kick(svq);
229
+}
230
+
231
+static bool vhost_svq_more_used(VhostShadowVirtqueue *svq)
232
+{
233
+ if (svq->last_used_idx != svq->shadow_used_idx) {
234
+ return true;
235
+ }
236
+
237
+ svq->shadow_used_idx = cpu_to_le16(svq->vring.used->idx);
238
+
239
+ return svq->last_used_idx != svq->shadow_used_idx;
240
}
241
242
/**
243
- * Forward vhost notifications
244
+ * Enable vhost device calls after disable them.
245
+ *
246
+ * @svq: The svq
247
+ *
248
+ * It returns false if there are pending used buffers from the vhost device,
249
+ * avoiding the possible races between SVQ checking for more work and enabling
250
+ * callbacks. True if SVQ used vring has no more pending buffers.
251
+ */
252
+static bool vhost_svq_enable_notification(VhostShadowVirtqueue *svq)
253
+{
254
+ svq->vring.avail->flags &= ~cpu_to_le16(VRING_AVAIL_F_NO_INTERRUPT);
255
+ /* Make sure the flag is written before the read of used_idx */
256
+ smp_mb();
257
+ return !vhost_svq_more_used(svq);
258
+}
259
+
260
+static void vhost_svq_disable_notification(VhostShadowVirtqueue *svq)
261
+{
262
+ svq->vring.avail->flags |= cpu_to_le16(VRING_AVAIL_F_NO_INTERRUPT);
263
+}
264
+
265
+static VirtQueueElement *vhost_svq_get_buf(VhostShadowVirtqueue *svq,
266
+ uint32_t *len)
267
+{
268
+ vring_desc_t *descs = svq->vring.desc;
269
+ const vring_used_t *used = svq->vring.used;
270
+ vring_used_elem_t used_elem;
271
+ uint16_t last_used;
272
+
273
+ if (!vhost_svq_more_used(svq)) {
274
+ return NULL;
275
+ }
276
+
277
+ /* Only get used array entries after they have been exposed by dev */
278
+ smp_rmb();
279
+ last_used = svq->last_used_idx & (svq->vring.num - 1);
280
+ used_elem.id = le32_to_cpu(used->ring[last_used].id);
281
+ used_elem.len = le32_to_cpu(used->ring[last_used].len);
282
+
283
+ svq->last_used_idx++;
284
+ if (unlikely(used_elem.id >= svq->vring.num)) {
285
+ qemu_log_mask(LOG_GUEST_ERROR, "Device %s says index %u is used",
286
+ svq->vdev->name, used_elem.id);
287
+ return NULL;
288
+ }
289
+
290
+ if (unlikely(!svq->ring_id_maps[used_elem.id])) {
291
+ qemu_log_mask(LOG_GUEST_ERROR,
292
+ "Device %s says index %u is used, but it was not available",
293
+ svq->vdev->name, used_elem.id);
294
+ return NULL;
295
+ }
296
+
297
+ descs[used_elem.id].next = svq->free_head;
298
+ svq->free_head = used_elem.id;
299
+
300
+ *len = used_elem.len;
301
+ return g_steal_pointer(&svq->ring_id_maps[used_elem.id]);
302
+}
303
+
304
+static void vhost_svq_flush(VhostShadowVirtqueue *svq,
305
+ bool check_for_avail_queue)
306
+{
307
+ VirtQueue *vq = svq->vq;
308
+
309
+ /* Forward as many used buffers as possible. */
310
+ do {
311
+ unsigned i = 0;
312
+
313
+ vhost_svq_disable_notification(svq);
314
+ while (true) {
315
+ uint32_t len;
316
+ g_autofree VirtQueueElement *elem = vhost_svq_get_buf(svq, &len);
317
+ if (!elem) {
318
+ break;
319
+ }
320
+
321
+ if (unlikely(i >= svq->vring.num)) {
322
+ qemu_log_mask(LOG_GUEST_ERROR,
323
+ "More than %u used buffers obtained in a %u size SVQ",
324
+ i, svq->vring.num);
325
+ virtqueue_fill(vq, elem, len, i);
326
+ virtqueue_flush(vq, i);
327
+ return;
328
+ }
329
+ virtqueue_fill(vq, elem, len, i++);
330
+ }
331
+
332
+ virtqueue_flush(vq, i);
333
+ event_notifier_set(&svq->svq_call);
334
+
335
+ if (check_for_avail_queue && svq->next_guest_avail_elem) {
336
+ /*
337
+ * Avail ring was full when vhost_svq_flush was called, so it's a
338
+ * good moment to make more descriptors available if possible.
339
+ */
340
+ vhost_handle_guest_kick(svq);
341
+ }
342
+ } while (!vhost_svq_enable_notification(svq));
343
+}
344
+
345
+/**
346
+ * Forward used buffers.
347
*
348
* @n: hdev call event notifier, the one that device set to notify svq.
349
+ *
350
+ * Note that we are not making any buffers available in the loop, there is no
351
+ * way that it runs more than virtqueue size times.
352
*/
353
static void vhost_svq_handle_call(EventNotifier *n)
354
{
355
VhostShadowVirtqueue *svq = container_of(n, VhostShadowVirtqueue,
356
hdev_call);
357
event_notifier_test_and_clear(n);
358
- event_notifier_set(&svq->svq_call);
359
+ vhost_svq_flush(svq, true);
360
}
361
362
/**
363
@@ -XXX,XX +XXX,XX @@ void vhost_svq_set_svq_kick_fd(VhostShadowVirtqueue *svq, int svq_kick_fd)
364
if (poll_start) {
365
event_notifier_init_fd(svq_kick, svq_kick_fd);
366
event_notifier_set(svq_kick);
367
- event_notifier_set_handler(svq_kick, vhost_handle_guest_kick);
368
+ event_notifier_set_handler(svq_kick, vhost_handle_guest_kick_notifier);
369
+ }
370
+}
371
+
372
+/**
373
+ * Start the shadow virtqueue operation.
374
+ *
375
+ * @svq: Shadow Virtqueue
376
+ * @vdev: VirtIO device
377
+ * @vq: Virtqueue to shadow
378
+ */
379
+void vhost_svq_start(VhostShadowVirtqueue *svq, VirtIODevice *vdev,
380
+ VirtQueue *vq)
381
+{
382
+ size_t desc_size, driver_size, device_size;
383
+
384
+ svq->next_guest_avail_elem = NULL;
385
+ svq->shadow_avail_idx = 0;
386
+ svq->shadow_used_idx = 0;
387
+ svq->last_used_idx = 0;
388
+ svq->vdev = vdev;
389
+ svq->vq = vq;
390
+
391
+ svq->vring.num = virtio_queue_get_num(vdev, virtio_get_queue_index(vq));
392
+ driver_size = vhost_svq_driver_area_size(svq);
393
+ device_size = vhost_svq_device_area_size(svq);
394
+ svq->vring.desc = qemu_memalign(qemu_real_host_page_size, driver_size);
395
+ desc_size = sizeof(vring_desc_t) * svq->vring.num;
396
+ svq->vring.avail = (void *)((char *)svq->vring.desc + desc_size);
397
+ memset(svq->vring.desc, 0, driver_size);
398
+ svq->vring.used = qemu_memalign(qemu_real_host_page_size, device_size);
399
+ memset(svq->vring.used, 0, device_size);
400
+ svq->ring_id_maps = g_new0(VirtQueueElement *, svq->vring.num);
401
+ for (unsigned i = 0; i < svq->vring.num - 1; i++) {
402
+ svq->vring.desc[i].next = cpu_to_le16(i + 1);
26
}
403
}
27
-
404
}
28
- /* Disable block migration */
405
29
- migrate_set_block_enabled(false, &local_err);
406
@@ -XXX,XX +XXX,XX @@ void vhost_svq_set_svq_kick_fd(VhostShadowVirtqueue *svq, int svq_kick_fd)
30
- if (local_err) {
407
void vhost_svq_stop(VhostShadowVirtqueue *svq)
31
- goto out;
408
{
32
- }
409
event_notifier_set_handler(&svq->svq_kick, NULL);
33
qemu_mutex_lock_iothread();
410
+ g_autofree VirtQueueElement *next_avail_elem = NULL;
34
411
+
35
#ifdef CONFIG_REPLICATION
412
+ if (!svq->vq) {
36
diff --git a/migration/migration.c b/migration/migration.c
413
+ return;
414
+ }
415
+
416
+ /* Send all pending used descriptors to guest */
417
+ vhost_svq_flush(svq, false);
418
+
419
+ for (unsigned i = 0; i < svq->vring.num; ++i) {
420
+ g_autofree VirtQueueElement *elem = NULL;
421
+ elem = g_steal_pointer(&svq->ring_id_maps[i]);
422
+ if (elem) {
423
+ virtqueue_detach_element(svq->vq, elem, 0);
424
+ }
425
+ }
426
+
427
+ next_avail_elem = g_steal_pointer(&svq->next_guest_avail_elem);
428
+ if (next_avail_elem) {
429
+ virtqueue_detach_element(svq->vq, next_avail_elem, 0);
430
+ }
431
+ svq->vq = NULL;
432
+ g_free(svq->ring_id_maps);
433
+ qemu_vfree(svq->vring.desc);
434
+ qemu_vfree(svq->vring.used);
435
}
436
437
/**
438
diff --git a/hw/virtio/vhost-shadow-virtqueue.h b/hw/virtio/vhost-shadow-virtqueue.h
37
index XXXXXXX..XXXXXXX 100644
439
index XXXXXXX..XXXXXXX 100644
38
--- a/migration/migration.c
440
--- a/hw/virtio/vhost-shadow-virtqueue.h
39
+++ b/migration/migration.c
441
+++ b/hw/virtio/vhost-shadow-virtqueue.h
40
@@ -XXX,XX +XXX,XX @@ static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
442
@@ -XXX,XX +XXX,XX @@ typedef struct VhostShadowVirtqueue {
443
444
/* Guest's call notifier, where the SVQ calls guest. */
445
EventNotifier svq_call;
446
+
447
+ /* Virtio queue shadowing */
448
+ VirtQueue *vq;
449
+
450
+ /* Virtio device */
451
+ VirtIODevice *vdev;
452
+
453
+ /* Map for use the guest's descriptors */
454
+ VirtQueueElement **ring_id_maps;
455
+
456
+ /* Next VirtQueue element that guest made available */
457
+ VirtQueueElement *next_guest_avail_elem;
458
+
459
+ /* Next head to expose to the device */
460
+ uint16_t shadow_avail_idx;
461
+
462
+ /* Next free descriptor */
463
+ uint16_t free_head;
464
+
465
+ /* Last seen used idx */
466
+ uint16_t shadow_used_idx;
467
+
468
+ /* Next head to consume from the device */
469
+ uint16_t last_used_idx;
470
} VhostShadowVirtqueue;
471
472
bool vhost_svq_valid_features(uint64_t features, Error **errp);
473
@@ -XXX,XX +XXX,XX @@ void vhost_svq_get_vring_addr(const VhostShadowVirtqueue *svq,
474
size_t vhost_svq_driver_area_size(const VhostShadowVirtqueue *svq);
475
size_t vhost_svq_device_area_size(const VhostShadowVirtqueue *svq);
476
477
+void vhost_svq_start(VhostShadowVirtqueue *svq, VirtIODevice *vdev,
478
+ VirtQueue *vq);
479
void vhost_svq_stop(VhostShadowVirtqueue *svq);
480
481
VhostShadowVirtqueue *vhost_svq_new(void);
482
diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c
483
index XXXXXXX..XXXXXXX 100644
484
--- a/hw/virtio/vhost-vdpa.c
485
+++ b/hw/virtio/vhost-vdpa.c
486
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_set_vring_dev_addr(struct vhost_dev *dev,
487
* Note that this function does not rewind kick file descriptor if cannot set
488
* call one.
489
*/
490
-static bool vhost_vdpa_svq_setup(struct vhost_dev *dev,
491
- VhostShadowVirtqueue *svq,
492
- unsigned idx,
493
- Error **errp)
494
+static int vhost_vdpa_svq_set_fds(struct vhost_dev *dev,
495
+ VhostShadowVirtqueue *svq,
496
+ unsigned idx,
497
+ Error **errp)
498
{
499
struct vhost_vring_file file = {
500
.index = dev->vq_index + idx,
501
@@ -XXX,XX +XXX,XX @@ static bool vhost_vdpa_svq_setup(struct vhost_dev *dev,
502
r = vhost_vdpa_set_vring_dev_kick(dev, &file);
503
if (unlikely(r != 0)) {
504
error_setg_errno(errp, -r, "Can't set device kick fd");
505
- return false;
506
+ return r;
41
}
507
}
42
508
43
if (blk || blk_inc) {
509
event_notifier = &svq->hdev_call;
44
+ if (migrate_colo_enabled()) {
510
@@ -XXX,XX +XXX,XX @@ static bool vhost_vdpa_svq_setup(struct vhost_dev *dev,
45
+ error_setg(errp, "No disk migration is required in COLO mode");
511
error_setg_errno(errp, -r, "Can't set device call fd");
46
+ return false;
512
}
47
+ }
513
48
if (migrate_use_block() || migrate_use_block_incremental()) {
514
+ return r;
49
error_setg(errp, "Command options are incompatible with "
515
+}
50
"current migration capabilities");
516
+
517
+/**
518
+ * Unmap a SVQ area in the device
519
+ */
520
+static bool vhost_vdpa_svq_unmap_ring(struct vhost_vdpa *v, hwaddr iova,
521
+ hwaddr size)
522
+{
523
+ int r;
524
+
525
+ size = ROUND_UP(size, qemu_real_host_page_size);
526
+ r = vhost_vdpa_dma_unmap(v, iova, size);
527
+ return r == 0;
528
+}
529
+
530
+static bool vhost_vdpa_svq_unmap_rings(struct vhost_dev *dev,
531
+ const VhostShadowVirtqueue *svq)
532
+{
533
+ struct vhost_vdpa *v = dev->opaque;
534
+ struct vhost_vring_addr svq_addr;
535
+ size_t device_size = vhost_svq_device_area_size(svq);
536
+ size_t driver_size = vhost_svq_driver_area_size(svq);
537
+ bool ok;
538
+
539
+ vhost_svq_get_vring_addr(svq, &svq_addr);
540
+
541
+ ok = vhost_vdpa_svq_unmap_ring(v, svq_addr.desc_user_addr, driver_size);
542
+ if (unlikely(!ok)) {
543
+ return false;
544
+ }
545
+
546
+ return vhost_vdpa_svq_unmap_ring(v, svq_addr.used_user_addr, device_size);
547
+}
548
+
549
+/**
550
+ * Map the shadow virtqueue rings in the device
551
+ *
552
+ * @dev: The vhost device
553
+ * @svq: The shadow virtqueue
554
+ * @addr: Assigned IOVA addresses
555
+ * @errp: Error pointer
556
+ */
557
+static bool vhost_vdpa_svq_map_rings(struct vhost_dev *dev,
558
+ const VhostShadowVirtqueue *svq,
559
+ struct vhost_vring_addr *addr,
560
+ Error **errp)
561
+{
562
+ struct vhost_vdpa *v = dev->opaque;
563
+ size_t device_size = vhost_svq_device_area_size(svq);
564
+ size_t driver_size = vhost_svq_driver_area_size(svq);
565
+ int r;
566
+
567
+ ERRP_GUARD();
568
+ vhost_svq_get_vring_addr(svq, addr);
569
+
570
+ r = vhost_vdpa_dma_map(v, addr->desc_user_addr, driver_size,
571
+ (void *)addr->desc_user_addr, true);
572
+ if (unlikely(r != 0)) {
573
+ error_setg_errno(errp, -r, "Cannot create vq driver region: ");
574
+ return false;
575
+ }
576
+
577
+ r = vhost_vdpa_dma_map(v, addr->used_user_addr, device_size,
578
+ (void *)addr->used_user_addr, false);
579
+ if (unlikely(r != 0)) {
580
+ error_setg_errno(errp, -r, "Cannot create vq device region: ");
581
+ }
582
+
583
+ return r == 0;
584
+}
585
+
586
+static bool vhost_vdpa_svq_setup(struct vhost_dev *dev,
587
+ VhostShadowVirtqueue *svq,
588
+ unsigned idx,
589
+ Error **errp)
590
+{
591
+ uint16_t vq_index = dev->vq_index + idx;
592
+ struct vhost_vring_state s = {
593
+ .index = vq_index,
594
+ };
595
+ int r;
596
+
597
+ r = vhost_vdpa_set_dev_vring_base(dev, &s);
598
+ if (unlikely(r)) {
599
+ error_setg_errno(errp, -r, "Cannot set vring base");
600
+ return false;
601
+ }
602
+
603
+ r = vhost_vdpa_svq_set_fds(dev, svq, idx, errp);
604
return r == 0;
605
}
606
607
@@ -XXX,XX +XXX,XX @@ static bool vhost_vdpa_svqs_start(struct vhost_dev *dev)
608
}
609
610
for (i = 0; i < v->shadow_vqs->len; ++i) {
611
+ VirtQueue *vq = virtio_get_queue(dev->vdev, dev->vq_index + i);
612
VhostShadowVirtqueue *svq = g_ptr_array_index(v->shadow_vqs, i);
613
+ struct vhost_vring_addr addr = {
614
+ .index = i,
615
+ };
616
+ int r;
617
bool ok = vhost_vdpa_svq_setup(dev, svq, i, &err);
618
if (unlikely(!ok)) {
619
- error_reportf_err(err, "Cannot setup SVQ %u: ", i);
620
+ goto err;
621
+ }
622
+
623
+ vhost_svq_start(svq, dev->vdev, vq);
624
+ ok = vhost_vdpa_svq_map_rings(dev, svq, &addr, &err);
625
+ if (unlikely(!ok)) {
626
+ goto err_map;
627
+ }
628
+
629
+ /* Override vring GPA set by vhost subsystem */
630
+ r = vhost_vdpa_set_vring_dev_addr(dev, &addr);
631
+ if (unlikely(r != 0)) {
632
+ error_setg_errno(&err, -r, "Cannot set device address");
633
+ goto err_set_addr;
634
+ }
635
+ }
636
+
637
+ return true;
638
+
639
+err_set_addr:
640
+ vhost_vdpa_svq_unmap_rings(dev, g_ptr_array_index(v->shadow_vqs, i));
641
+
642
+err_map:
643
+ vhost_svq_stop(g_ptr_array_index(v->shadow_vqs, i));
644
+
645
+err:
646
+ error_reportf_err(err, "Cannot setup SVQ %u: ", i);
647
+ for (unsigned j = 0; j < i; ++j) {
648
+ VhostShadowVirtqueue *svq = g_ptr_array_index(v->shadow_vqs, j);
649
+ vhost_vdpa_svq_unmap_rings(dev, svq);
650
+ vhost_svq_stop(svq);
651
+ }
652
+
653
+ return false;
654
+}
655
+
656
+static bool vhost_vdpa_svqs_stop(struct vhost_dev *dev)
657
+{
658
+ struct vhost_vdpa *v = dev->opaque;
659
+
660
+ if (!v->shadow_vqs) {
661
+ return true;
662
+ }
663
+
664
+ for (unsigned i = 0; i < v->shadow_vqs->len; ++i) {
665
+ VhostShadowVirtqueue *svq = g_ptr_array_index(v->shadow_vqs,
666
+ i);
667
+ bool ok = vhost_vdpa_svq_unmap_rings(dev, svq);
668
+ if (unlikely(!ok)) {
669
return false;
670
}
671
}
672
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_dev_start(struct vhost_dev *dev, bool started)
673
}
674
vhost_vdpa_set_vring_ready(dev);
675
} else {
676
+ ok = vhost_vdpa_svqs_stop(dev);
677
+ if (unlikely(!ok)) {
678
+ return -1;
679
+ }
680
vhost_vdpa_host_notifiers_uninit(dev, dev->nvqs);
681
}
682
51
--
683
--
52
2.7.4
684
2.7.4
53
685
54
686
diff view generated by jsdifflib
[PULL 07/12] Fix the qemu crash when guest shutdown during checkpoint
[PULL V2 09/15] util: Add iova_tree_alloc_map
1
From: "Rao, Lei" <lei.rao@intel.com>
1
From: Eugenio Pérez <eperezma@redhat.com>
2
2
3
This patch fixes the following:
3
This iova tree function allows it to look for a hole in allocated
4
qemu-system-x86_64: invalid runstate transition: 'colo' ->'shutdown'
4
regions and return a totally new translation for a given translated
5
Aborted (core dumped)
5
address.
6
6
7
Signed-off-by: Lei Rao <lei.rao@intel.com>
7
It's usage is mainly to allow devices to access qemu address space,
8
Reviewed-by: Li Zhijian <lizhijian@fujitsu.com>
8
remapping guest's one into a new iova space where qemu can add chunks of
9
Reviewed-by: Zhang Chen <chen.zhang@intel.com>
9
addresses.
10
Reviewed-by: Lukas Straub <lukasstraub2@web.de>
10
11
Tested-by: Lukas Straub <lukasstraub2@web.de>
11
Acked-by: Michael S. Tsirkin <mst@redhat.com>
12
Signed-off-by: Zhang Chen <chen.zhang@intel.com>
12
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
13
Reviewed-by: Peter Xu <peterx@redhat.com>
13
Signed-off-by: Jason Wang <jasowang@redhat.com>
14
Signed-off-by: Jason Wang <jasowang@redhat.com>
14
---
15
---
15
softmmu/runstate.c | 1 +
16
include/qemu/iova-tree.h | 18 +++++++
16
1 file changed, 1 insertion(+)
17
util/iova-tree.c | 135 +++++++++++++++++++++++++++++++++++++++++++++++
17
18
2 files changed, 153 insertions(+)
18
diff --git a/softmmu/runstate.c b/softmmu/runstate.c
19
20
diff --git a/include/qemu/iova-tree.h b/include/qemu/iova-tree.h
19
index XXXXXXX..XXXXXXX 100644
21
index XXXXXXX..XXXXXXX 100644
20
--- a/softmmu/runstate.c
22
--- a/include/qemu/iova-tree.h
21
+++ b/softmmu/runstate.c
23
+++ b/include/qemu/iova-tree.h
22
@@ -XXX,XX +XXX,XX @@ static const RunStateTransition runstate_transitions_def[] = {
24
@@ -XXX,XX +XXX,XX @@
23
{ RUN_STATE_RESTORE_VM, RUN_STATE_PRELAUNCH },
25
#define IOVA_OK (0)
24
26
#define IOVA_ERR_INVALID (-1) /* Invalid parameters */
25
{ RUN_STATE_COLO, RUN_STATE_RUNNING },
27
#define IOVA_ERR_OVERLAP (-2) /* IOVA range overlapped */
26
+ { RUN_STATE_COLO, RUN_STATE_SHUTDOWN},
28
+#define IOVA_ERR_NOMEM (-3) /* Cannot allocate */
27
29
28
{ RUN_STATE_RUNNING, RUN_STATE_DEBUG },
30
typedef struct IOVATree IOVATree;
29
{ RUN_STATE_RUNNING, RUN_STATE_INTERNAL_ERROR },
31
typedef struct DMAMap {
32
@@ -XXX,XX +XXX,XX @@ const DMAMap *iova_tree_find_address(const IOVATree *tree, hwaddr iova);
33
void iova_tree_foreach(IOVATree *tree, iova_tree_iterator iterator);
34
35
/**
36
+ * iova_tree_alloc_map:
37
+ *
38
+ * @tree: the iova tree to allocate from
39
+ * @map: the new map (as translated addr & size) to allocate in the iova region
40
+ * @iova_begin: the minimum address of the allocation
41
+ * @iova_end: the maximum addressable direction of the allocation
42
+ *
43
+ * Allocates a new region of a given size, between iova_min and iova_max.
44
+ *
45
+ * Return: Same as iova_tree_insert, but cannot overlap and can return error if
46
+ * iova tree is out of free contiguous range. The caller gets the assigned iova
47
+ * in map->iova.
48
+ */
49
+int iova_tree_alloc_map(IOVATree *tree, DMAMap *map, hwaddr iova_begin,
50
+ hwaddr iova_end);
51
+
52
+/**
53
* iova_tree_destroy:
54
*
55
* @tree: the iova tree to destroy
56
diff --git a/util/iova-tree.c b/util/iova-tree.c
57
index XXXXXXX..XXXXXXX 100644
58
--- a/util/iova-tree.c
59
+++ b/util/iova-tree.c
60
@@ -XXX,XX +XXX,XX @@ struct IOVATree {
61
GTree *tree;
62
};
63
64
+/* Args to pass to iova_tree_alloc foreach function. */
65
+struct IOVATreeAllocArgs {
66
+ /* Size of the desired allocation */
67
+ size_t new_size;
68
+
69
+ /* The minimum address allowed in the allocation */
70
+ hwaddr iova_begin;
71
+
72
+ /* Map at the left of the hole, can be NULL if "this" is first one */
73
+ const DMAMap *prev;
74
+
75
+ /* Map at the right of the hole, can be NULL if "prev" is the last one */
76
+ const DMAMap *this;
77
+
78
+ /* If found, we fill in the IOVA here */
79
+ hwaddr iova_result;
80
+
81
+ /* Whether have we found a valid IOVA */
82
+ bool iova_found;
83
+};
84
+
85
+/**
86
+ * Iterate args to the next hole
87
+ *
88
+ * @args: The alloc arguments
89
+ * @next: The next mapping in the tree. Can be NULL to signal the last one
90
+ */
91
+static void iova_tree_alloc_args_iterate(struct IOVATreeAllocArgs *args,
92
+ const DMAMap *next) {
93
+ args->prev = args->this;
94
+ args->this = next;
95
+}
96
+
97
static int iova_tree_compare(gconstpointer a, gconstpointer b, gpointer data)
98
{
99
const DMAMap *m1 = a, *m2 = b;
100
@@ -XXX,XX +XXX,XX @@ int iova_tree_remove(IOVATree *tree, const DMAMap *map)
101
return IOVA_OK;
102
}
103
104
+/**
105
+ * Try to find an unallocated IOVA range between prev and this elements.
106
+ *
107
+ * @args: Arguments to allocation
108
+ *
109
+ * Cases:
110
+ *
111
+ * (1) !prev, !this: No entries allocated, always succeed
112
+ *
113
+ * (2) !prev, this: We're iterating at the 1st element.
114
+ *
115
+ * (3) prev, !this: We're iterating at the last element.
116
+ *
117
+ * (4) prev, this: this is the most common case, we'll try to find a hole
118
+ * between "prev" and "this" mapping.
119
+ *
120
+ * Note that this function assumes the last valid iova is HWADDR_MAX, but it
121
+ * searches linearly so it's easy to discard the result if it's not the case.
122
+ */
123
+static void iova_tree_alloc_map_in_hole(struct IOVATreeAllocArgs *args)
124
+{
125
+ const DMAMap *prev = args->prev, *this = args->this;
126
+ uint64_t hole_start, hole_last;
127
+
128
+ if (this && this->iova + this->size < args->iova_begin) {
129
+ return;
130
+ }
131
+
132
+ hole_start = MAX(prev ? prev->iova + prev->size + 1 : 0, args->iova_begin);
133
+ hole_last = this ? this->iova : HWADDR_MAX;
134
+
135
+ if (hole_last - hole_start > args->new_size) {
136
+ args->iova_result = hole_start;
137
+ args->iova_found = true;
138
+ }
139
+}
140
+
141
+/**
142
+ * Foreach dma node in the tree, compare if there is a hole with its previous
143
+ * node (or minimum iova address allowed) and the node.
144
+ *
145
+ * @key: Node iterating
146
+ * @value: Node iterating
147
+ * @pargs: Struct to communicate with the outside world
148
+ *
149
+ * Return: false to keep iterating, true if needs break.
150
+ */
151
+static gboolean iova_tree_alloc_traverse(gpointer key, gpointer value,
152
+ gpointer pargs)
153
+{
154
+ struct IOVATreeAllocArgs *args = pargs;
155
+ DMAMap *node = value;
156
+
157
+ assert(key == value);
158
+
159
+ iova_tree_alloc_args_iterate(args, node);
160
+ iova_tree_alloc_map_in_hole(args);
161
+ return args->iova_found;
162
+}
163
+
164
+int iova_tree_alloc_map(IOVATree *tree, DMAMap *map, hwaddr iova_begin,
165
+ hwaddr iova_last)
166
+{
167
+ struct IOVATreeAllocArgs args = {
168
+ .new_size = map->size,
169
+ .iova_begin = iova_begin,
170
+ };
171
+
172
+ if (unlikely(iova_last < iova_begin)) {
173
+ return IOVA_ERR_INVALID;
174
+ }
175
+
176
+ /*
177
+ * Find a valid hole for the mapping
178
+ *
179
+ * Assuming low iova_begin, so no need to do a binary search to
180
+ * locate the first node.
181
+ *
182
+ * TODO: Replace all this with g_tree_node_first/next/last when available
183
+ * (from glib since 2.68). To do it with g_tree_foreach complicates the
184
+ * code a lot.
185
+ *
186
+ */
187
+ g_tree_foreach(tree->tree, iova_tree_alloc_traverse, &args);
188
+ if (!args.iova_found) {
189
+ /*
190
+ * Either tree is empty or the last hole is still not checked.
191
+ * g_tree_foreach does not compare (last, iova_last] range, so we check
192
+ * it here.
193
+ */
194
+ iova_tree_alloc_args_iterate(&args, NULL);
195
+ iova_tree_alloc_map_in_hole(&args);
196
+ }
197
+
198
+ if (!args.iova_found || args.iova_result + map->size > iova_last) {
199
+ return IOVA_ERR_NOMEM;
200
+ }
201
+
202
+ map->iova = args.iova_result;
203
+ return iova_tree_insert(tree, map);
204
+}
205
+
206
void iova_tree_destroy(IOVATree *tree)
207
{
208
g_tree_destroy(tree->tree);
30
--
209
--
31
2.7.4
210
2.7.4
32
211
33
212
diff view generated by jsdifflib
[PULL 04/12] vhost-vdpa: remove the unused vhost_vdpa_get_acked_features()
[PULL V2 10/15] util: add iova_tree_find_iova
1
No user for this helper, let's remove it.
1
From: Eugenio Pérez <eperezma@redhat.com>
2
2
3
This function does the reverse operation of iova_tree_find: To look for
4
a mapping that match a translated address so we can do the reverse.
5
6
This have linear complexity instead of logarithmic, but it supports
7
overlapping HVA. Future developments could reduce it.
8
9
Acked-by: Michael S. Tsirkin <mst@redhat.com>
10
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
3
Signed-off-by: Jason Wang <jasowang@redhat.com>
11
Signed-off-by: Jason Wang <jasowang@redhat.com>
4
---
12
---
5
include/net/vhost-vdpa.h | 1 -
13
include/qemu/iova-tree.h | 20 +++++++++++++++++++-
6
net/vhost-vdpa.c | 9 ---------
14
util/iova-tree.c | 34 ++++++++++++++++++++++++++++++++++
7
2 files changed, 10 deletions(-)
15
2 files changed, 53 insertions(+), 1 deletion(-)
8
16
9
diff --git a/include/net/vhost-vdpa.h b/include/net/vhost-vdpa.h
17
diff --git a/include/qemu/iova-tree.h b/include/qemu/iova-tree.h
10
index XXXXXXX..XXXXXXX 100644
18
index XXXXXXX..XXXXXXX 100644
11
--- a/include/net/vhost-vdpa.h
19
--- a/include/qemu/iova-tree.h
12
+++ b/include/net/vhost-vdpa.h
20
+++ b/include/qemu/iova-tree.h
13
@@ -XXX,XX +XXX,XX @@
21
@@ -XXX,XX +XXX,XX @@ int iova_tree_remove(IOVATree *tree, const DMAMap *map);
14
#define TYPE_VHOST_VDPA "vhost-vdpa"
22
* @tree: the iova tree to search from
15
23
* @map: the mapping to search
16
struct vhost_net *vhost_vdpa_get_vhost_net(NetClientState *nc);
24
*
17
-uint64_t vhost_vdpa_get_acked_features(NetClientState *nc);
25
- * Search for a mapping in the iova tree that overlaps with the
18
26
+ * Search for a mapping in the iova tree that iova overlaps with the
19
extern const int vdpa_feature_bits[];
27
* mapping range specified. Only the first found mapping will be
20
28
* returned.
21
diff --git a/net/vhost-vdpa.c b/net/vhost-vdpa.c
29
*
30
@@ -XXX,XX +XXX,XX @@ int iova_tree_remove(IOVATree *tree, const DMAMap *map);
31
const DMAMap *iova_tree_find(const IOVATree *tree, const DMAMap *map);
32
33
/**
34
+ * iova_tree_find_iova:
35
+ *
36
+ * @tree: the iova tree to search from
37
+ * @map: the mapping to search
38
+ *
39
+ * Search for a mapping in the iova tree that translated_addr overlaps with the
40
+ * mapping range specified. Only the first found mapping will be
41
+ * returned.
42
+ *
43
+ * Return: DMAMap pointer if found, or NULL if not found. Note that
44
+ * the returned DMAMap pointer is maintained internally. User should
45
+ * only read the content but never modify or free the content. Also,
46
+ * user is responsible to make sure the pointer is valid (say, no
47
+ * concurrent deletion in progress).
48
+ */
49
+const DMAMap *iova_tree_find_iova(const IOVATree *tree, const DMAMap *map);
50
+
51
+/**
52
* iova_tree_find_address:
53
*
54
* @tree: the iova tree to search from
55
diff --git a/util/iova-tree.c b/util/iova-tree.c
22
index XXXXXXX..XXXXXXX 100644
56
index XXXXXXX..XXXXXXX 100644
23
--- a/net/vhost-vdpa.c
57
--- a/util/iova-tree.c
24
+++ b/net/vhost-vdpa.c
58
+++ b/util/iova-tree.c
25
@@ -XXX,XX +XXX,XX @@ VHostNetState *vhost_vdpa_get_vhost_net(NetClientState *nc)
59
@@ -XXX,XX +XXX,XX @@ struct IOVATreeAllocArgs {
26
return s->vhost_net;
60
bool iova_found;
61
};
62
63
+typedef struct IOVATreeFindIOVAArgs {
64
+ const DMAMap *needle;
65
+ const DMAMap *result;
66
+} IOVATreeFindIOVAArgs;
67
+
68
/**
69
* Iterate args to the next hole
70
*
71
@@ -XXX,XX +XXX,XX @@ const DMAMap *iova_tree_find(const IOVATree *tree, const DMAMap *map)
72
return g_tree_lookup(tree->tree, map);
27
}
73
}
28
74
29
-uint64_t vhost_vdpa_get_acked_features(NetClientState *nc)
75
+static gboolean iova_tree_find_address_iterator(gpointer key, gpointer value,
30
-{
76
+ gpointer data)
31
- VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc);
77
+{
32
- assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA);
78
+ const DMAMap *map = key;
33
- s->acked_features = vhost_net_get_acked_features(s->vhost_net);
79
+ IOVATreeFindIOVAArgs *args = data;
34
-
80
+ const DMAMap *needle;
35
- return s->acked_features;
81
+
36
-}
82
+ g_assert(key == value);
37
-
83
+
38
static int vhost_vdpa_net_check_device_id(struct vhost_net *net)
84
+ needle = args->needle;
85
+ if (map->translated_addr + map->size < needle->translated_addr ||
86
+ needle->translated_addr + needle->size < map->translated_addr) {
87
+ return false;
88
+ }
89
+
90
+ args->result = map;
91
+ return true;
92
+}
93
+
94
+const DMAMap *iova_tree_find_iova(const IOVATree *tree, const DMAMap *map)
95
+{
96
+ IOVATreeFindIOVAArgs args = {
97
+ .needle = map,
98
+ };
99
+
100
+ g_tree_foreach(tree->tree, iova_tree_find_address_iterator, &args);
101
+ return args.result;
102
+}
103
+
104
const DMAMap *iova_tree_find_address(const IOVATree *tree, hwaddr iova)
39
{
105
{
40
uint32_t device_id;
106
const DMAMap map = { .iova = iova, .size = 0 };
41
--
107
--
42
2.7.4
108
2.7.4
43
109
44
110
diff view generated by jsdifflib
[PULL 06/12] Remove some duplicate trace code.
[PULL V2 11/15] vhost: Add VhostIOVATree
1
From: "Rao, Lei" <lei.rao@intel.com>
1
From: Eugenio Pérez <eperezma@redhat.com>
2
2
3
There is the same trace code in the colo_compare_packet_payload.
3
This tree is able to look for a translated address from an IOVA address.
4
4
5
Signed-off-by: Lei Rao <lei.rao@intel.com>
5
At first glance it is similar to util/iova-tree. However, SVQ working on
6
Reviewed-by: Li Zhijian <lizhijian@fujitsu.com>
6
devices with limited IOVA space need more capabilities, like allocating
7
Reviewed-by: Zhang Chen <chen.zhang@intel.com>
7
IOVA chunks or performing reverse translations (qemu addresses to iova).
8
Reviewed-by: Lukas Straub <lukasstraub2@web.de>
8
9
Tested-by: Lukas Straub <lukasstraub2@web.de>
9
The allocation capability, as "assign a free IOVA address to this chunk
10
Signed-off-by: Zhang Chen <chen.zhang@intel.com>
10
of memory in qemu's address space" allows shadow virtqueue to create a
11
new address space that is not restricted by guest's addressable one, so
12
we can allocate shadow vqs vrings outside of it.
13
14
It duplicates the tree so it can search efficiently in both directions,
15
and it will signal overlap if iova or the translated address is present
16
in any tree.
17
18
Acked-by: Michael S. Tsirkin <mst@redhat.com>
19
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
11
Signed-off-by: Jason Wang <jasowang@redhat.com>
20
Signed-off-by: Jason Wang <jasowang@redhat.com>
12
---
21
---
13
net/colo-compare.c | 13 -------------
22
hw/virtio/meson.build | 2 +-
14
1 file changed, 13 deletions(-)
23
hw/virtio/vhost-iova-tree.c | 110 ++++++++++++++++++++++++++++++++++++++++++++
24
hw/virtio/vhost-iova-tree.h | 27 +++++++++++
25
3 files changed, 138 insertions(+), 1 deletion(-)
26
create mode 100644 hw/virtio/vhost-iova-tree.c
27
create mode 100644 hw/virtio/vhost-iova-tree.h
15
28
16
diff --git a/net/colo-compare.c b/net/colo-compare.c
29
diff --git a/hw/virtio/meson.build b/hw/virtio/meson.build
17
index XXXXXXX..XXXXXXX 100644
30
index XXXXXXX..XXXXXXX 100644
18
--- a/net/colo-compare.c
31
--- a/hw/virtio/meson.build
19
+++ b/net/colo-compare.c
32
+++ b/hw/virtio/meson.build
20
@@ -XXX,XX +XXX,XX @@ static int colo_packet_compare_other(Packet *spkt, Packet *ppkt)
33
@@ -XXX,XX +XXX,XX @@ softmmu_ss.add(when: 'CONFIG_ALL', if_true: files('vhost-stub.c'))
21
uint16_t offset = ppkt->vnet_hdr_len;
34
22
35
virtio_ss = ss.source_set()
23
trace_colo_compare_main("compare other");
36
virtio_ss.add(files('virtio.c'))
24
- if (trace_event_get_state_backends(TRACE_COLO_COMPARE_IP_INFO)) {
37
-virtio_ss.add(when: 'CONFIG_VHOST', if_true: files('vhost.c', 'vhost-backend.c', 'vhost-shadow-virtqueue.c'))
25
- char pri_ip_src[20], pri_ip_dst[20], sec_ip_src[20], sec_ip_dst[20];
38
+virtio_ss.add(when: 'CONFIG_VHOST', if_true: files('vhost.c', 'vhost-backend.c', 'vhost-shadow-virtqueue.c', 'vhost-iova-tree.c'))
26
-
39
virtio_ss.add(when: 'CONFIG_VHOST_USER', if_true: files('vhost-user.c'))
27
- strcpy(pri_ip_src, inet_ntoa(ppkt->ip->ip_src));
40
virtio_ss.add(when: 'CONFIG_VHOST_VDPA', if_true: files('vhost-vdpa.c'))
28
- strcpy(pri_ip_dst, inet_ntoa(ppkt->ip->ip_dst));
41
virtio_ss.add(when: 'CONFIG_VIRTIO_BALLOON', if_true: files('virtio-balloon.c'))
29
- strcpy(sec_ip_src, inet_ntoa(spkt->ip->ip_src));
42
diff --git a/hw/virtio/vhost-iova-tree.c b/hw/virtio/vhost-iova-tree.c
30
- strcpy(sec_ip_dst, inet_ntoa(spkt->ip->ip_dst));
43
new file mode 100644
31
-
44
index XXXXXXX..XXXXXXX
32
- trace_colo_compare_ip_info(ppkt->size, pri_ip_src,
45
--- /dev/null
33
- pri_ip_dst, spkt->size,
46
+++ b/hw/virtio/vhost-iova-tree.c
34
- sec_ip_src, sec_ip_dst);
47
@@ -XXX,XX +XXX,XX @@
35
- }
48
+/*
36
-
49
+ * vhost software live migration iova tree
37
if (ppkt->size != spkt->size) {
50
+ *
38
trace_colo_compare_main("Other: payload size of packets are different");
51
+ * SPDX-FileCopyrightText: Red Hat, Inc. 2021
39
return -1;
52
+ * SPDX-FileContributor: Author: Eugenio Pérez <eperezma@redhat.com>
53
+ *
54
+ * SPDX-License-Identifier: GPL-2.0-or-later
55
+ */
56
+
57
+#include "qemu/osdep.h"
58
+#include "qemu/iova-tree.h"
59
+#include "vhost-iova-tree.h"
60
+
61
+#define iova_min_addr qemu_real_host_page_size
62
+
63
+/**
64
+ * VhostIOVATree, able to:
65
+ * - Translate iova address
66
+ * - Reverse translate iova address (from translated to iova)
67
+ * - Allocate IOVA regions for translated range (linear operation)
68
+ */
69
+struct VhostIOVATree {
70
+ /* First addressable iova address in the device */
71
+ uint64_t iova_first;
72
+
73
+ /* Last addressable iova address in the device */
74
+ uint64_t iova_last;
75
+
76
+ /* IOVA address to qemu memory maps. */
77
+ IOVATree *iova_taddr_map;
78
+};
79
+
80
+/**
81
+ * Create a new IOVA tree
82
+ *
83
+ * Returns the new IOVA tree
84
+ */
85
+VhostIOVATree *vhost_iova_tree_new(hwaddr iova_first, hwaddr iova_last)
86
+{
87
+ VhostIOVATree *tree = g_new(VhostIOVATree, 1);
88
+
89
+ /* Some devices do not like 0 addresses */
90
+ tree->iova_first = MAX(iova_first, iova_min_addr);
91
+ tree->iova_last = iova_last;
92
+
93
+ tree->iova_taddr_map = iova_tree_new();
94
+ return tree;
95
+}
96
+
97
+/**
98
+ * Delete an iova tree
99
+ */
100
+void vhost_iova_tree_delete(VhostIOVATree *iova_tree)
101
+{
102
+ iova_tree_destroy(iova_tree->iova_taddr_map);
103
+ g_free(iova_tree);
104
+}
105
+
106
+/**
107
+ * Find the IOVA address stored from a memory address
108
+ *
109
+ * @tree: The iova tree
110
+ * @map: The map with the memory address
111
+ *
112
+ * Return the stored mapping, or NULL if not found.
113
+ */
114
+const DMAMap *vhost_iova_tree_find_iova(const VhostIOVATree *tree,
115
+ const DMAMap *map)
116
+{
117
+ return iova_tree_find_iova(tree->iova_taddr_map, map);
118
+}
119
+
120
+/**
121
+ * Allocate a new mapping
122
+ *
123
+ * @tree: The iova tree
124
+ * @map: The iova map
125
+ *
126
+ * Returns:
127
+ * - IOVA_OK if the map fits in the container
128
+ * - IOVA_ERR_INVALID if the map does not make sense (like size overflow)
129
+ * - IOVA_ERR_NOMEM if tree cannot allocate more space.
130
+ *
131
+ * It returns assignated iova in map->iova if return value is VHOST_DMA_MAP_OK.
132
+ */
133
+int vhost_iova_tree_map_alloc(VhostIOVATree *tree, DMAMap *map)
134
+{
135
+ /* Some vhost devices do not like addr 0. Skip first page */
136
+ hwaddr iova_first = tree->iova_first ?: qemu_real_host_page_size;
137
+
138
+ if (map->translated_addr + map->size < map->translated_addr ||
139
+ map->perm == IOMMU_NONE) {
140
+ return IOVA_ERR_INVALID;
141
+ }
142
+
143
+ /* Allocate a node in IOVA address */
144
+ return iova_tree_alloc_map(tree->iova_taddr_map, map, iova_first,
145
+ tree->iova_last);
146
+}
147
+
148
+/**
149
+ * Remove existing mappings from iova tree
150
+ *
151
+ * @iova_tree: The vhost iova tree
152
+ * @map: The map to remove
153
+ */
154
+void vhost_iova_tree_remove(VhostIOVATree *iova_tree, const DMAMap *map)
155
+{
156
+ iova_tree_remove(iova_tree->iova_taddr_map, map);
157
+}
158
diff --git a/hw/virtio/vhost-iova-tree.h b/hw/virtio/vhost-iova-tree.h
159
new file mode 100644
160
index XXXXXXX..XXXXXXX
161
--- /dev/null
162
+++ b/hw/virtio/vhost-iova-tree.h
163
@@ -XXX,XX +XXX,XX @@
164
+/*
165
+ * vhost software live migration iova tree
166
+ *
167
+ * SPDX-FileCopyrightText: Red Hat, Inc. 2021
168
+ * SPDX-FileContributor: Author: Eugenio Pérez <eperezma@redhat.com>
169
+ *
170
+ * SPDX-License-Identifier: GPL-2.0-or-later
171
+ */
172
+
173
+#ifndef HW_VIRTIO_VHOST_IOVA_TREE_H
174
+#define HW_VIRTIO_VHOST_IOVA_TREE_H
175
+
176
+#include "qemu/iova-tree.h"
177
+#include "exec/memory.h"
178
+
179
+typedef struct VhostIOVATree VhostIOVATree;
180
+
181
+VhostIOVATree *vhost_iova_tree_new(uint64_t iova_first, uint64_t iova_last);
182
+void vhost_iova_tree_delete(VhostIOVATree *iova_tree);
183
+G_DEFINE_AUTOPTR_CLEANUP_FUNC(VhostIOVATree, vhost_iova_tree_delete);
184
+
185
+const DMAMap *vhost_iova_tree_find_iova(const VhostIOVATree *iova_tree,
186
+ const DMAMap *map);
187
+int vhost_iova_tree_map_alloc(VhostIOVATree *iova_tree, DMAMap *map);
188
+void vhost_iova_tree_remove(VhostIOVATree *iova_tree, const DMAMap *map);
189
+
190
+#endif
40
--
191
--
41
2.7.4
192
2.7.4
42
193
43
194
diff view generated by jsdifflib
[PULL 08/12] Optimize the function of filter_send
[PULL V2 12/15] vdpa: Add custom IOTLB translations to SVQ
1
From: "Rao, Lei" <lei.rao@intel.com>
1
From: Eugenio Pérez <eperezma@redhat.com>
2
2
3
The iov_size has been calculated in filter_send(). we can directly
3
Use translations added in VhostIOVATree in SVQ.
4
return the size.In this way, this is no need to repeat calculations
4
5
in filter_redirector_receive_iov();
5
Only introduce usage here, not allocation and deallocation. As with
6
6
previous patches, we use the dead code paths of shadow_vqs_enabled to
7
Signed-off-by: Lei Rao <lei.rao@intel.com>
7
avoid commiting too many changes at once. These are impossible to take
8
Reviewed-by: Li Zhijian <lizhijian@fujitsu.com>
8
at the moment.
9
Reviewed-by: Zhang Chen <chen.zhang@intel.com>
9
10
Reviewed-by: Lukas Straub <lukasstraub2@web.de>
10
Acked-by: Michael S. Tsirkin <mst@redhat.com>
11
Tested-by: Lukas Straub <lukasstraub2@web.de>
11
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
12
Signed-off-by: Zhang Chen <chen.zhang@intel.com>
13
Signed-off-by: Jason Wang <jasowang@redhat.com>
12
Signed-off-by: Jason Wang <jasowang@redhat.com>
14
---
13
---
15
net/filter-mirror.c | 8 ++++----
14
hw/virtio/vhost-shadow-virtqueue.c | 75 +++++++++++++++++++++--
16
1 file changed, 4 insertions(+), 4 deletions(-)
15
hw/virtio/vhost-shadow-virtqueue.h | 6 +-
17
16
hw/virtio/vhost-vdpa.c | 122 +++++++++++++++++++++++++++++++------
18
diff --git a/net/filter-mirror.c b/net/filter-mirror.c
17
include/hw/virtio/vhost-vdpa.h | 3 +
18
4 files changed, 181 insertions(+), 25 deletions(-)
19
20
diff --git a/hw/virtio/vhost-shadow-virtqueue.c b/hw/virtio/vhost-shadow-virtqueue.c
19
index XXXXXXX..XXXXXXX 100644
21
index XXXXXXX..XXXXXXX 100644
20
--- a/net/filter-mirror.c
22
--- a/hw/virtio/vhost-shadow-virtqueue.c
21
+++ b/net/filter-mirror.c
23
+++ b/hw/virtio/vhost-shadow-virtqueue.c
22
@@ -XXX,XX +XXX,XX @@ static int filter_send(MirrorState *s,
24
@@ -XXX,XX +XXX,XX @@ static uint16_t vhost_svq_available_slots(const VhostShadowVirtqueue *svq)
23
goto err;
25
return svq->vring.num - (svq->shadow_avail_idx - svq->shadow_used_idx);
26
}
27
28
+/**
29
+ * Translate addresses between the qemu's virtual address and the SVQ IOVA
30
+ *
31
+ * @svq: Shadow VirtQueue
32
+ * @vaddr: Translated IOVA addresses
33
+ * @iovec: Source qemu's VA addresses
34
+ * @num: Length of iovec and minimum length of vaddr
35
+ */
36
+static bool vhost_svq_translate_addr(const VhostShadowVirtqueue *svq,
37
+ void **addrs, const struct iovec *iovec,
38
+ size_t num)
39
+{
40
+ if (num == 0) {
41
+ return true;
42
+ }
43
+
44
+ for (size_t i = 0; i < num; ++i) {
45
+ DMAMap needle = {
46
+ .translated_addr = (hwaddr)iovec[i].iov_base,
47
+ .size = iovec[i].iov_len,
48
+ };
49
+ size_t off;
50
+
51
+ const DMAMap *map = vhost_iova_tree_find_iova(svq->iova_tree, &needle);
52
+ /*
53
+ * Map cannot be NULL since iova map contains all guest space and
54
+ * qemu already has a physical address mapped
55
+ */
56
+ if (unlikely(!map)) {
57
+ qemu_log_mask(LOG_GUEST_ERROR,
58
+ "Invalid address 0x%"HWADDR_PRIx" given by guest",
59
+ needle.translated_addr);
60
+ return false;
61
+ }
62
+
63
+ off = needle.translated_addr - map->translated_addr;
64
+ addrs[i] = (void *)(map->iova + off);
65
+
66
+ if (unlikely(int128_gt(int128_add(needle.translated_addr,
67
+ iovec[i].iov_len),
68
+ map->translated_addr + map->size))) {
69
+ qemu_log_mask(LOG_GUEST_ERROR,
70
+ "Guest buffer expands over iova range");
71
+ return false;
72
+ }
73
+ }
74
+
75
+ return true;
76
+}
77
+
78
static void vhost_vring_write_descs(VhostShadowVirtqueue *svq,
79
+ void * const *sg,
80
const struct iovec *iovec,
81
size_t num, bool more_descs, bool write)
82
{
83
@@ -XXX,XX +XXX,XX @@ static void vhost_vring_write_descs(VhostShadowVirtqueue *svq,
84
} else {
85
descs[i].flags = flags;
86
}
87
- descs[i].addr = cpu_to_le64((hwaddr)iovec[n].iov_base);
88
+ descs[i].addr = cpu_to_le64((hwaddr)sg[n]);
89
descs[i].len = cpu_to_le32(iovec[n].iov_len);
90
91
last = i;
92
@@ -XXX,XX +XXX,XX @@ static bool vhost_svq_add_split(VhostShadowVirtqueue *svq,
93
{
94
unsigned avail_idx;
95
vring_avail_t *avail = svq->vring.avail;
96
+ bool ok;
97
+ g_autofree void **sgs = g_new(void *, MAX(elem->out_num, elem->in_num));
98
99
*head = svq->free_head;
100
101
@@ -XXX,XX +XXX,XX @@ static bool vhost_svq_add_split(VhostShadowVirtqueue *svq,
102
return false;
24
}
103
}
25
104
26
- return 0;
105
- vhost_vring_write_descs(svq, elem->out_sg, elem->out_num,
27
+ return size;
106
+ ok = vhost_svq_translate_addr(svq, sgs, elem->out_sg, elem->out_num);
28
107
+ if (unlikely(!ok)) {
29
err:
108
+ return false;
30
return ret < 0 ? ret : -EIO;
109
+ }
31
@@ -XXX,XX +XXX,XX @@ static ssize_t filter_mirror_receive_iov(NetFilterState *nf,
110
+ vhost_vring_write_descs(svq, sgs, elem->out_sg, elem->out_num,
32
int ret;
111
elem->in_num > 0, false);
33
112
- vhost_vring_write_descs(svq, elem->in_sg, elem->in_num, false, true);
34
ret = filter_send(s, iov, iovcnt);
113
+
35
- if (ret) {
114
+
36
+ if (ret < 0) {
115
+ ok = vhost_svq_translate_addr(svq, sgs, elem->in_sg, elem->in_num);
37
error_report("filter mirror send failed(%s)", strerror(-ret));
116
+ if (unlikely(!ok)) {
117
+ return false;
118
+ }
119
+
120
+ vhost_vring_write_descs(svq, sgs, elem->in_sg, elem->in_num, false, true);
121
122
/*
123
* Put the entry in the available array (but don't update avail->idx until
124
@@ -XXX,XX +XXX,XX @@ void vhost_svq_stop(VhostShadowVirtqueue *svq)
125
* Creates vhost shadow virtqueue, and instructs the vhost device to use the
126
* shadow methods and file descriptors.
127
*
128
+ * @iova_tree: Tree to perform descriptors translations
129
+ *
130
* Returns the new virtqueue or NULL.
131
*
132
* In case of error, reason is reported through error_report.
133
*/
134
-VhostShadowVirtqueue *vhost_svq_new(void)
135
+VhostShadowVirtqueue *vhost_svq_new(VhostIOVATree *iova_tree)
136
{
137
g_autofree VhostShadowVirtqueue *svq = g_new0(VhostShadowVirtqueue, 1);
138
int r;
139
@@ -XXX,XX +XXX,XX @@ VhostShadowVirtqueue *vhost_svq_new(void)
140
141
event_notifier_init_fd(&svq->svq_kick, VHOST_FILE_UNBIND);
142
event_notifier_set_handler(&svq->hdev_call, vhost_svq_handle_call);
143
+ svq->iova_tree = iova_tree;
144
return g_steal_pointer(&svq);
145
146
err_init_hdev_call:
147
diff --git a/hw/virtio/vhost-shadow-virtqueue.h b/hw/virtio/vhost-shadow-virtqueue.h
148
index XXXXXXX..XXXXXXX 100644
149
--- a/hw/virtio/vhost-shadow-virtqueue.h
150
+++ b/hw/virtio/vhost-shadow-virtqueue.h
151
@@ -XXX,XX +XXX,XX @@
152
#include "qemu/event_notifier.h"
153
#include "hw/virtio/virtio.h"
154
#include "standard-headers/linux/vhost_types.h"
155
+#include "hw/virtio/vhost-iova-tree.h"
156
157
/* Shadow virtqueue to relay notifications */
158
typedef struct VhostShadowVirtqueue {
159
@@ -XXX,XX +XXX,XX @@ typedef struct VhostShadowVirtqueue {
160
/* Virtio device */
161
VirtIODevice *vdev;
162
163
+ /* IOVA mapping */
164
+ VhostIOVATree *iova_tree;
165
+
166
/* Map for use the guest's descriptors */
167
VirtQueueElement **ring_id_maps;
168
169
@@ -XXX,XX +XXX,XX @@ void vhost_svq_start(VhostShadowVirtqueue *svq, VirtIODevice *vdev,
170
VirtQueue *vq);
171
void vhost_svq_stop(VhostShadowVirtqueue *svq);
172
173
-VhostShadowVirtqueue *vhost_svq_new(void);
174
+VhostShadowVirtqueue *vhost_svq_new(VhostIOVATree *iova_tree);
175
176
void vhost_svq_free(gpointer vq);
177
G_DEFINE_AUTOPTR_CLEANUP_FUNC(VhostShadowVirtqueue, vhost_svq_free);
178
diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c
179
index XXXXXXX..XXXXXXX 100644
180
--- a/hw/virtio/vhost-vdpa.c
181
+++ b/hw/virtio/vhost-vdpa.c
182
@@ -XXX,XX +XXX,XX @@ static void vhost_vdpa_listener_region_add(MemoryListener *listener,
183
vaddr, section->readonly);
184
185
llsize = int128_sub(llend, int128_make64(iova));
186
+ if (v->shadow_vqs_enabled) {
187
+ DMAMap mem_region = {
188
+ .translated_addr = (hwaddr)vaddr,
189
+ .size = int128_get64(llsize) - 1,
190
+ .perm = IOMMU_ACCESS_FLAG(true, section->readonly),
191
+ };
192
+
193
+ int r = vhost_iova_tree_map_alloc(v->iova_tree, &mem_region);
194
+ if (unlikely(r != IOVA_OK)) {
195
+ error_report("Can't allocate a mapping (%d)", r);
196
+ goto fail;
197
+ }
198
+
199
+ iova = mem_region.iova;
200
+ }
201
202
vhost_vdpa_iotlb_batch_begin_once(v);
203
ret = vhost_vdpa_dma_map(v, iova, int128_get64(llsize),
204
@@ -XXX,XX +XXX,XX @@ static void vhost_vdpa_listener_region_del(MemoryListener *listener,
205
206
llsize = int128_sub(llend, int128_make64(iova));
207
208
+ if (v->shadow_vqs_enabled) {
209
+ const DMAMap *result;
210
+ const void *vaddr = memory_region_get_ram_ptr(section->mr) +
211
+ section->offset_within_region +
212
+ (iova - section->offset_within_address_space);
213
+ DMAMap mem_region = {
214
+ .translated_addr = (hwaddr)vaddr,
215
+ .size = int128_get64(llsize) - 1,
216
+ };
217
+
218
+ result = vhost_iova_tree_find_iova(v->iova_tree, &mem_region);
219
+ iova = result->iova;
220
+ vhost_iova_tree_remove(v->iova_tree, &mem_region);
221
+ }
222
vhost_vdpa_iotlb_batch_begin_once(v);
223
ret = vhost_vdpa_dma_unmap(v, iova, int128_get64(llsize));
224
if (ret) {
225
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_init_svq(struct vhost_dev *hdev, struct vhost_vdpa *v,
226
227
shadow_vqs = g_ptr_array_new_full(hdev->nvqs, vhost_svq_free);
228
for (unsigned n = 0; n < hdev->nvqs; ++n) {
229
- g_autoptr(VhostShadowVirtqueue) svq = vhost_svq_new();
230
+ g_autoptr(VhostShadowVirtqueue) svq = vhost_svq_new(v->iova_tree);
231
232
if (unlikely(!svq)) {
233
error_setg(errp, "Cannot create svq %u", n);
234
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_svq_set_fds(struct vhost_dev *dev,
235
/**
236
* Unmap a SVQ area in the device
237
*/
238
-static bool vhost_vdpa_svq_unmap_ring(struct vhost_vdpa *v, hwaddr iova,
239
- hwaddr size)
240
+static bool vhost_vdpa_svq_unmap_ring(struct vhost_vdpa *v,
241
+ const DMAMap *needle)
242
{
243
+ const DMAMap *result = vhost_iova_tree_find_iova(v->iova_tree, needle);
244
+ hwaddr size;
245
int r;
246
247
- size = ROUND_UP(size, qemu_real_host_page_size);
248
- r = vhost_vdpa_dma_unmap(v, iova, size);
249
+ if (unlikely(!result)) {
250
+ error_report("Unable to find SVQ address to unmap");
251
+ return false;
252
+ }
253
+
254
+ size = ROUND_UP(result->size, qemu_real_host_page_size);
255
+ r = vhost_vdpa_dma_unmap(v, result->iova, size);
256
return r == 0;
257
}
258
259
static bool vhost_vdpa_svq_unmap_rings(struct vhost_dev *dev,
260
const VhostShadowVirtqueue *svq)
261
{
262
+ DMAMap needle = {};
263
struct vhost_vdpa *v = dev->opaque;
264
struct vhost_vring_addr svq_addr;
265
- size_t device_size = vhost_svq_device_area_size(svq);
266
- size_t driver_size = vhost_svq_driver_area_size(svq);
267
bool ok;
268
269
vhost_svq_get_vring_addr(svq, &svq_addr);
270
271
- ok = vhost_vdpa_svq_unmap_ring(v, svq_addr.desc_user_addr, driver_size);
272
+ needle.translated_addr = svq_addr.desc_user_addr;
273
+ ok = vhost_vdpa_svq_unmap_ring(v, &needle);
274
if (unlikely(!ok)) {
275
return false;
38
}
276
}
39
277
40
@@ -XXX,XX +XXX,XX @@ static ssize_t filter_redirector_receive_iov(NetFilterState *nf,
278
- return vhost_vdpa_svq_unmap_ring(v, svq_addr.used_user_addr, device_size);
41
279
+ needle.translated_addr = svq_addr.used_user_addr;
42
if (qemu_chr_fe_backend_connected(&s->chr_out)) {
280
+ return vhost_vdpa_svq_unmap_ring(v, &needle);
43
ret = filter_send(s, iov, iovcnt);
281
+}
44
- if (ret) {
282
+
45
+ if (ret < 0) {
283
+/**
46
error_report("filter redirector send failed(%s)", strerror(-ret));
284
+ * Map the SVQ area in the device
47
}
285
+ *
48
- return iov_size(iov, iovcnt);
286
+ * @v: Vhost-vdpa device
49
+ return ret;
287
+ * @needle: The area to search iova
50
} else {
288
+ * @errorp: Error pointer
51
return 0;
289
+ */
290
+static bool vhost_vdpa_svq_map_ring(struct vhost_vdpa *v, DMAMap *needle,
291
+ Error **errp)
292
+{
293
+ int r;
294
+
295
+ r = vhost_iova_tree_map_alloc(v->iova_tree, needle);
296
+ if (unlikely(r != IOVA_OK)) {
297
+ error_setg(errp, "Cannot allocate iova (%d)", r);
298
+ return false;
299
+ }
300
+
301
+ r = vhost_vdpa_dma_map(v, needle->iova, needle->size + 1,
302
+ (void *)needle->translated_addr,
303
+ needle->perm == IOMMU_RO);
304
+ if (unlikely(r != 0)) {
305
+ error_setg_errno(errp, -r, "Cannot map region to device");
306
+ vhost_iova_tree_remove(v->iova_tree, needle);
307
+ }
308
+
309
+ return r == 0;
310
}
311
312
/**
313
@@ -XXX,XX +XXX,XX @@ static bool vhost_vdpa_svq_map_rings(struct vhost_dev *dev,
314
struct vhost_vring_addr *addr,
315
Error **errp)
316
{
317
+ DMAMap device_region, driver_region;
318
+ struct vhost_vring_addr svq_addr;
319
struct vhost_vdpa *v = dev->opaque;
320
size_t device_size = vhost_svq_device_area_size(svq);
321
size_t driver_size = vhost_svq_driver_area_size(svq);
322
- int r;
323
+ size_t avail_offset;
324
+ bool ok;
325
326
ERRP_GUARD();
327
- vhost_svq_get_vring_addr(svq, addr);
328
+ vhost_svq_get_vring_addr(svq, &svq_addr);
329
330
- r = vhost_vdpa_dma_map(v, addr->desc_user_addr, driver_size,
331
- (void *)addr->desc_user_addr, true);
332
- if (unlikely(r != 0)) {
333
- error_setg_errno(errp, -r, "Cannot create vq driver region: ");
334
+ driver_region = (DMAMap) {
335
+ .translated_addr = svq_addr.desc_user_addr,
336
+ .size = driver_size - 1,
337
+ .perm = IOMMU_RO,
338
+ };
339
+ ok = vhost_vdpa_svq_map_ring(v, &driver_region, errp);
340
+ if (unlikely(!ok)) {
341
+ error_prepend(errp, "Cannot create vq driver region: ");
342
return false;
52
}
343
}
344
+ addr->desc_user_addr = driver_region.iova;
345
+ avail_offset = svq_addr.avail_user_addr - svq_addr.desc_user_addr;
346
+ addr->avail_user_addr = driver_region.iova + avail_offset;
347
348
- r = vhost_vdpa_dma_map(v, addr->used_user_addr, device_size,
349
- (void *)addr->used_user_addr, false);
350
- if (unlikely(r != 0)) {
351
- error_setg_errno(errp, -r, "Cannot create vq device region: ");
352
+ device_region = (DMAMap) {
353
+ .translated_addr = svq_addr.used_user_addr,
354
+ .size = device_size - 1,
355
+ .perm = IOMMU_RW,
356
+ };
357
+ ok = vhost_vdpa_svq_map_ring(v, &device_region, errp);
358
+ if (unlikely(!ok)) {
359
+ error_prepend(errp, "Cannot create vq device region: ");
360
+ vhost_vdpa_svq_unmap_ring(v, &driver_region);
361
}
362
+ addr->used_user_addr = device_region.iova;
363
364
- return r == 0;
365
+ return ok;
366
}
367
368
static bool vhost_vdpa_svq_setup(struct vhost_dev *dev,
369
diff --git a/include/hw/virtio/vhost-vdpa.h b/include/hw/virtio/vhost-vdpa.h
370
index XXXXXXX..XXXXXXX 100644
371
--- a/include/hw/virtio/vhost-vdpa.h
372
+++ b/include/hw/virtio/vhost-vdpa.h
373
@@ -XXX,XX +XXX,XX @@
374
375
#include <gmodule.h>
376
377
+#include "hw/virtio/vhost-iova-tree.h"
378
#include "hw/virtio/virtio.h"
379
#include "standard-headers/linux/vhost_types.h"
380
381
@@ -XXX,XX +XXX,XX @@ typedef struct vhost_vdpa {
382
MemoryListener listener;
383
struct vhost_vdpa_iova_range iova_range;
384
bool shadow_vqs_enabled;
385
+ /* IOVA mapping used by the Shadow Virtqueue */
386
+ VhostIOVATree *iova_tree;
387
GPtrArray *shadow_vqs;
388
struct vhost_dev *dev;
389
VhostVDPAHostNotifier notifier[VIRTIO_QUEUE_MAX];
53
--
390
--
54
2.7.4
391
2.7.4
55
392
56
393
diff view generated by jsdifflib
[PULL 05/12] netdev: add more commands to preconfig mode
[PULL V2 13/15] vdpa: Adapt vhost_vdpa_get_vring_base to SVQ
1
From: Paolo Bonzini <pbonzini@redhat.com>
1
From: Eugenio Pérez <eperezma@redhat.com>
2
2
3
Creating and destroying network backend does not require a fully
3
This is needed to achieve migration, so the destination can restore its
4
constructed machine. Allow the related monitor commands to run before
4
index.
5
machine initialization has concluded.
6
5
7
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
6
Setting base as last used idx, so destination will see as available all
8
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
7
the entries that the device did not use, including the in-flight
8
processing ones.
9
10
This is ok for networking, but other kinds of devices might have
11
problems with these retransmissions.
12
13
Acked-by: Michael S. Tsirkin <mst@redhat.com>
14
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
9
Signed-off-by: Jason Wang <jasowang@redhat.com>
15
Signed-off-by: Jason Wang <jasowang@redhat.com>
10
---
16
---
11
hmp-commands.hx | 2 ++
17
hw/virtio/vhost-vdpa.c | 17 +++++++++++++++++
12
qapi/net.json | 6 ++++--
18
1 file changed, 17 insertions(+)
13
2 files changed, 6 insertions(+), 2 deletions(-)
14
19
15
diff --git a/hmp-commands.hx b/hmp-commands.hx
20
diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c
16
index XXXXXXX..XXXXXXX 100644
21
index XXXXXXX..XXXXXXX 100644
17
--- a/hmp-commands.hx
22
--- a/hw/virtio/vhost-vdpa.c
18
+++ b/hmp-commands.hx
23
+++ b/hw/virtio/vhost-vdpa.c
19
@@ -XXX,XX +XXX,XX @@ ERST
24
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_set_vring_base(struct vhost_dev *dev,
20
.help = "add host network device",
25
static int vhost_vdpa_get_vring_base(struct vhost_dev *dev,
21
.cmd = hmp_netdev_add,
26
struct vhost_vring_state *ring)
22
.command_completion = netdev_add_completion,
27
{
23
+ .flags = "p",
28
+ struct vhost_vdpa *v = dev->opaque;
24
},
29
int ret;
25
30
26
SRST
31
+ if (v->shadow_vqs_enabled) {
27
@@ -XXX,XX +XXX,XX @@ ERST
32
+ VhostShadowVirtqueue *svq = g_ptr_array_index(v->shadow_vqs,
28
.help = "remove host network device",
33
+ ring->index);
29
.cmd = hmp_netdev_del,
34
+
30
.command_completion = netdev_del_completion,
35
+ /*
31
+ .flags = "p",
36
+ * Setting base as last used idx, so destination will see as available
32
},
37
+ * all the entries that the device did not use, including the in-flight
33
38
+ * processing ones.
34
SRST
39
+ *
35
diff --git a/qapi/net.json b/qapi/net.json
40
+ * TODO: This is ok for networking, but other kinds of devices might
36
index XXXXXXX..XXXXXXX 100644
41
+ * have problems with these retransmissions.
37
--- a/qapi/net.json
42
+ */
38
+++ b/qapi/net.json
43
+ ring->num = svq->last_used_idx;
39
@@ -XXX,XX +XXX,XX @@
44
+ return 0;
40
# <- { "return": {} }
45
+ }
41
#
46
+
42
##
47
ret = vhost_vdpa_call(dev, VHOST_GET_VRING_BASE, ring);
43
-{ 'command': 'netdev_add', 'data': 'Netdev', 'boxed': true }
48
trace_vhost_vdpa_get_vring_base(dev, ring->index, ring->num);
44
+{ 'command': 'netdev_add', 'data': 'Netdev', 'boxed': true,
49
return ret;
45
+ 'allow-preconfig': true }
46
47
##
48
# @netdev_del:
49
@@ -XXX,XX +XXX,XX @@
50
# <- { "return": {} }
51
#
52
##
53
-{ 'command': 'netdev_del', 'data': {'id': 'str'} }
54
+{ 'command': 'netdev_del', 'data': {'id': 'str'},
55
+ 'allow-preconfig': true }
56
57
##
58
# @NetLegacyNicOptions:
59
--
50
--
60
2.7.4
51
2.7.4
61
52
62
53
diff view generated by jsdifflib
[PULL 01/12] vhost-vdpa: skip ram device from the IOTLB mapping
[PULL V2 14/15] vdpa: Never set log_base addr if SVQ is enabled
1
vDPA is not tie to any specific hardware, for safety and simplicity,
1
From: Eugenio Pérez <eperezma@redhat.com>
2
vhost-vDPA doesn't allow MMIO area to be mapped via IOTLB. Only the
3
doorbell could be mapped via mmap(). So this patch exclude skip the
4
ram device from the IOTLB mapping.
5
2
6
Reviewed-by: Si-Wei Liu <si-wei.liu@oracle.com>
3
Setting the log address would make the device start reporting invalid
4
dirty memory because the SVQ vrings are located in qemu's memory.
5
6
Acked-by: Michael S. Tsirkin <mst@redhat.com>
7
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
7
Signed-off-by: Jason Wang <jasowang@redhat.com>
8
Signed-off-by: Jason Wang <jasowang@redhat.com>
8
---
9
---
9
hw/virtio/vhost-vdpa.c | 12 ++----------
10
hw/virtio/vhost-vdpa.c | 3 ++-
10
1 file changed, 2 insertions(+), 10 deletions(-)
11
1 file changed, 2 insertions(+), 1 deletion(-)
11
12
12
diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c
13
diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c
13
index XXXXXXX..XXXXXXX 100644
14
index XXXXXXX..XXXXXXX 100644
14
--- a/hw/virtio/vhost-vdpa.c
15
--- a/hw/virtio/vhost-vdpa.c
15
+++ b/hw/virtio/vhost-vdpa.c
16
+++ b/hw/virtio/vhost-vdpa.c
16
@@ -XXX,XX +XXX,XX @@ static bool vhost_vdpa_listener_skipped_section(MemoryRegionSection *section)
17
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_dev_start(struct vhost_dev *dev, bool started)
18
static int vhost_vdpa_set_log_base(struct vhost_dev *dev, uint64_t base,
19
struct vhost_log *log)
17
{
20
{
18
return (!memory_region_is_ram(section->mr) &&
21
- if (vhost_vdpa_one_time_request(dev)) {
19
!memory_region_is_iommu(section->mr)) ||
22
+ struct vhost_vdpa *v = dev->opaque;
20
+ /* vhost-vDPA doesn't allow MMIO to be mapped */
23
+ if (v->shadow_vqs_enabled || vhost_vdpa_one_time_request(dev)) {
21
+ memory_region_is_ram_device(section->mr) ||
24
return 0;
22
/*
23
* Sizing an enabled 64-bit BAR can cause spurious mappings to
24
* addresses in the upper part of the 64-bit address space. These
25
@@ -XXX,XX +XXX,XX @@ static void vhost_vdpa_listener_region_add(MemoryListener *listener,
26
vaddr, section->readonly);
27
if (ret) {
28
error_report("vhost vdpa map fail!");
29
- if (memory_region_is_ram_device(section->mr)) {
30
- /* Allow unexpected mappings not to be fatal for RAM devices */
31
- error_report("map ram fail!");
32
- return ;
33
- }
34
goto fail;
35
}
25
}
36
26
37
return;
38
39
fail:
40
- if (memory_region_is_ram_device(section->mr)) {
41
- error_report("failed to vdpa_dma_map. pci p2p may not work");
42
- return;
43
-
44
- }
45
/*
46
* On the initfn path, store the first error in the container so we
47
* can gracefully fail. Runtime, there's not much we can do other
48
--
27
--
49
2.7.4
28
2.7.4
50
29
51
30
diff view generated by jsdifflib
[PULL 02/12] vhost-vdpa: map virtqueue notification area if possible
[PULL V2 15/15] vdpa: Expose VHOST_F_LOG_ALL on SVQ
1
This patch implements the vq notification mapping support for
1
From: Eugenio Pérez <eperezma@redhat.com>
2
vhost-vDPA. This is simply done by using mmap()/munmap() for the
3
vhost-vDPA fd during device start/stop. For the device without
4
notification mapping support, we fall back to eventfd based
5
notification gracefully.
6
2
7
Reviewed-by: Si-Wei Liu <si-wei.liu@oracle.com>
3
SVQ is able to log the dirty bits by itself, so let's use it to not
4
block migration.
5
6
Also, ignore set and clear of VHOST_F_LOG_ALL on set_features if SVQ is
7
enabled. Even if the device supports it, the reports would be nonsense
8
because SVQ memory is in the qemu region.
9
10
The log region is still allocated. Future changes might skip that, but
11
this series is already long enough.
12
13
Acked-by: Michael S. Tsirkin <mst@redhat.com>
14
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
8
Signed-off-by: Jason Wang <jasowang@redhat.com>
15
Signed-off-by: Jason Wang <jasowang@redhat.com>
9
---
16
---
10
hw/virtio/vhost-vdpa.c | 85 ++++++++++++++++++++++++++++++++++++++++++
17
hw/virtio/vhost-vdpa.c | 39 +++++++++++++++++++++++++++++++++++----
11
include/hw/virtio/vhost-vdpa.h | 6 +++
18
include/hw/virtio/vhost-vdpa.h | 1 +
12
2 files changed, 91 insertions(+)
19
2 files changed, 36 insertions(+), 4 deletions(-)
13
20
14
diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c
21
diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c
15
index XXXXXXX..XXXXXXX 100644
22
index XXXXXXX..XXXXXXX 100644
16
--- a/hw/virtio/vhost-vdpa.c
23
--- a/hw/virtio/vhost-vdpa.c
17
+++ b/hw/virtio/vhost-vdpa.c
24
+++ b/hw/virtio/vhost-vdpa.c
18
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_init(struct vhost_dev *dev, void *opaque)
25
@@ -XXX,XX +XXX,XX @@ static bool vhost_vdpa_one_time_request(struct vhost_dev *dev)
19
return 0;
26
return v->index != 0;
20
}
27
}
21
28
22
+static void vhost_vdpa_host_notifier_uninit(struct vhost_dev *dev,
29
+static int vhost_vdpa_get_dev_features(struct vhost_dev *dev,
23
+ int queue_index)
30
+ uint64_t *features)
24
+{
31
+{
25
+ size_t page_size = qemu_real_host_page_size;
32
+ int ret;
26
+ struct vhost_vdpa *v = dev->opaque;
27
+ VirtIODevice *vdev = dev->vdev;
28
+ VhostVDPAHostNotifier *n;
29
+
33
+
30
+ n = &v->notifier[queue_index];
34
+ ret = vhost_vdpa_call(dev, VHOST_GET_FEATURES, features);
31
+
35
+ trace_vhost_vdpa_get_features(dev, *features);
32
+ if (n->addr) {
36
+ return ret;
33
+ virtio_queue_set_host_notifier_mr(vdev, queue_index, &n->mr, false);
34
+ object_unparent(OBJECT(&n->mr));
35
+ munmap(n->addr, page_size);
36
+ n->addr = NULL;
37
+ }
38
+}
37
+}
39
+
38
+
40
+static void vhost_vdpa_host_notifiers_uninit(struct vhost_dev *dev, int n)
39
static int vhost_vdpa_init_svq(struct vhost_dev *hdev, struct vhost_vdpa *v,
41
+{
40
Error **errp)
42
+ int i;
41
{
42
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_init_svq(struct vhost_dev *hdev, struct vhost_vdpa *v,
43
return 0;
44
}
45
46
- r = hdev->vhost_ops->vhost_get_features(hdev, &dev_features);
47
+ r = vhost_vdpa_get_dev_features(hdev, &dev_features);
48
if (r != 0) {
49
error_setg_errno(errp, -r, "Can't get vdpa device features");
50
return r;
51
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_set_mem_table(struct vhost_dev *dev,
52
static int vhost_vdpa_set_features(struct vhost_dev *dev,
53
uint64_t features)
54
{
55
+ struct vhost_vdpa *v = dev->opaque;
56
int ret;
57
58
if (vhost_vdpa_one_time_request(dev)) {
59
return 0;
60
}
61
62
+ if (v->shadow_vqs_enabled) {
63
+ if ((v->acked_features ^ features) == BIT_ULL(VHOST_F_LOG_ALL)) {
64
+ /*
65
+ * QEMU is just trying to enable or disable logging. SVQ handles
66
+ * this sepparately, so no need to forward this.
67
+ */
68
+ v->acked_features = features;
69
+ return 0;
70
+ }
43
+
71
+
44
+ for (i = 0; i < n; i++) {
72
+ v->acked_features = features;
45
+ vhost_vdpa_host_notifier_uninit(dev, i);
46
+ }
47
+}
48
+
73
+
49
+static int vhost_vdpa_host_notifier_init(struct vhost_dev *dev, int queue_index)
74
+ /* We must not ack _F_LOG if SVQ is enabled */
50
+{
75
+ features &= ~BIT_ULL(VHOST_F_LOG_ALL);
51
+ size_t page_size = qemu_real_host_page_size;
52
+ struct vhost_vdpa *v = dev->opaque;
53
+ VirtIODevice *vdev = dev->vdev;
54
+ VhostVDPAHostNotifier *n;
55
+ int fd = v->device_fd;
56
+ void *addr;
57
+ char *name;
58
+
59
+ vhost_vdpa_host_notifier_uninit(dev, queue_index);
60
+
61
+ n = &v->notifier[queue_index];
62
+
63
+ addr = mmap(NULL, page_size, PROT_WRITE, MAP_SHARED, fd,
64
+ queue_index * page_size);
65
+ if (addr == MAP_FAILED) {
66
+ goto err;
67
+ }
76
+ }
68
+
77
+
69
+ name = g_strdup_printf("vhost-vdpa/host-notifier@%p mmaps[%d]",
78
trace_vhost_vdpa_set_features(dev, features);
70
+ v, queue_index);
79
ret = vhost_vdpa_call(dev, VHOST_SET_FEATURES, &features);
71
+ memory_region_init_ram_device_ptr(&n->mr, OBJECT(vdev), name,
80
if (ret) {
72
+ page_size, addr);
81
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_set_vring_call(struct vhost_dev *dev,
73
+ g_free(name);
82
static int vhost_vdpa_get_features(struct vhost_dev *dev,
83
uint64_t *features)
84
{
85
- int ret;
86
+ struct vhost_vdpa *v = dev->opaque;
87
+ int ret = vhost_vdpa_get_dev_features(dev, features);
74
+
88
+
75
+ if (virtio_queue_set_host_notifier_mr(vdev, queue_index, &n->mr, true)) {
89
+ if (ret == 0 && v->shadow_vqs_enabled) {
76
+ munmap(addr, page_size);
90
+ /* Add SVQ logging capabilities */
77
+ goto err;
91
+ *features |= BIT_ULL(VHOST_F_LOG_ALL);
78
+ }
92
+ }
79
+ n->addr = addr;
93
80
+
94
- ret = vhost_vdpa_call(dev, VHOST_GET_FEATURES, features);
81
+ return 0;
95
- trace_vhost_vdpa_get_features(dev, *features);
82
+
96
return ret;
83
+err:
97
}
84
+ return -1;
98
85
+}
86
+
87
+static void vhost_vdpa_host_notifiers_init(struct vhost_dev *dev)
88
+{
89
+ int i;
90
+
91
+ for (i = dev->vq_index; i < dev->vq_index + dev->nvqs; i++) {
92
+ if (vhost_vdpa_host_notifier_init(dev, i)) {
93
+ goto err;
94
+ }
95
+ }
96
+
97
+ return;
98
+
99
+err:
100
+ vhost_vdpa_host_notifiers_uninit(dev, i);
101
+ return;
102
+}
103
+
104
static int vhost_vdpa_cleanup(struct vhost_dev *dev)
105
{
106
struct vhost_vdpa *v;
107
assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_VDPA);
108
v = dev->opaque;
109
trace_vhost_vdpa_cleanup(dev, v);
110
+ vhost_vdpa_host_notifiers_uninit(dev, dev->nvqs);
111
memory_listener_unregister(&v->listener);
112
113
dev->opaque = NULL;
114
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_dev_start(struct vhost_dev *dev, bool started)
115
if (started) {
116
uint8_t status = 0;
117
memory_listener_register(&v->listener, &address_space_memory);
118
+ vhost_vdpa_host_notifiers_init(dev);
119
vhost_vdpa_set_vring_ready(dev);
120
vhost_vdpa_add_status(dev, VIRTIO_CONFIG_S_DRIVER_OK);
121
vhost_vdpa_call(dev, VHOST_VDPA_GET_STATUS, &status);
122
@@ -XXX,XX +XXX,XX @@ static int vhost_vdpa_dev_start(struct vhost_dev *dev, bool started)
123
vhost_vdpa_reset_device(dev);
124
vhost_vdpa_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE |
125
VIRTIO_CONFIG_S_DRIVER);
126
+ vhost_vdpa_host_notifiers_uninit(dev, dev->nvqs);
127
memory_listener_unregister(&v->listener);
128
129
return 0;
130
diff --git a/include/hw/virtio/vhost-vdpa.h b/include/hw/virtio/vhost-vdpa.h
99
diff --git a/include/hw/virtio/vhost-vdpa.h b/include/hw/virtio/vhost-vdpa.h
131
index XXXXXXX..XXXXXXX 100644
100
index XXXXXXX..XXXXXXX 100644
132
--- a/include/hw/virtio/vhost-vdpa.h
101
--- a/include/hw/virtio/vhost-vdpa.h
133
+++ b/include/hw/virtio/vhost-vdpa.h
102
+++ b/include/hw/virtio/vhost-vdpa.h
134
@@ -XXX,XX +XXX,XX @@
103
@@ -XXX,XX +XXX,XX @@ typedef struct vhost_vdpa {
135
104
bool iotlb_batch_begin_sent;
136
#include "hw/virtio/virtio.h"
137
138
+typedef struct VhostVDPAHostNotifier {
139
+ MemoryRegion mr;
140
+ void *addr;
141
+} VhostVDPAHostNotifier;
142
+
143
typedef struct vhost_vdpa {
144
int device_fd;
145
uint32_t msg_type;
146
MemoryListener listener;
105
MemoryListener listener;
147
struct vhost_dev *dev;
106
struct vhost_vdpa_iova_range iova_range;
148
+ VhostVDPAHostNotifier notifier[VIRTIO_QUEUE_MAX];
107
+ uint64_t acked_features;
149
} VhostVDPA;
108
bool shadow_vqs_enabled;
150
109
/* IOVA mapping used by the Shadow Virtqueue */
151
#endif
110
VhostIOVATree *iova_tree;
152
--
111
--
153
2.7.4
112
2.7.4
154
113
155
114
diff view generated by jsdifflib

Patchew 2.1-devel-b67e4c2

© 2016 - 2025 Red Hat, Inc.