1 | The following changes since commit 7260438b7056469610ee166f7abe9ff8a26b8b16: | 1 | The following changes since commit e47f81b617684c4546af286d307b69014a83538a: |
---|---|---|---|
2 | 2 | ||
3 | Merge remote-tracking branch 'remotes/palmer/tags/riscv-for-master-3.2-part2' into staging (2019-01-14 11:41:43 +0000) | 3 | Merge remote-tracking branch 'remotes/thibault/tags/samuel-thibault' into staging (2019-02-07 18:53:25 +0000) |
4 | 4 | ||
5 | are available in the Git repository at: | 5 | are available in the Git repository at: |
6 | 6 | ||
7 | git://github.com/stefanha/qemu.git tags/block-pull-request | 7 | git://github.com/stefanha/qemu.git tags/block-pull-request |
8 | 8 | ||
9 | for you to fetch changes up to fef1660132b0f25bf2d275d7f986ddcfe19a4426: | 9 | for you to fetch changes up to 55140166667bb555c5d05165b93b25557d2e6397: |
10 | 10 | ||
11 | aio-posix: Fix concurrent aio_poll/set_fd_handler. (2019-01-14 14:09:41 +0000) | 11 | tests/virtio-blk: add test for WRITE_ZEROES command (2019-02-11 11:58:17 +0800) |
12 | 12 | ||
13 | ---------------------------------------------------------------- | 13 | ---------------------------------------------------------------- |
14 | Pull request | 14 | Pull request |
15 | 15 | ||
16 | No user-visible changes. | 16 | User-visible changes: |
17 | |||
18 | * virtio-blk: DISCARD and WRITE_ZEROES support | ||
17 | 19 | ||
18 | ---------------------------------------------------------------- | 20 | ---------------------------------------------------------------- |
19 | 21 | ||
20 | Remy Noel (2): | 22 | Peter Xu (1): |
21 | aio-posix: Unregister fd from ctx epoll when removing fd_handler. | 23 | iothread: fix iothread hang when stop too soon |
22 | aio-posix: Fix concurrent aio_poll/set_fd_handler. | ||
23 | 24 | ||
24 | util/aio-posix.c | 90 +++++++++++++++++++++++++++++------------------- | 25 | Stefano Garzarella (7): |
25 | util/aio-win32.c | 67 ++++++++++++++++------------------- | 26 | virtio-blk: cleanup using VirtIOBlock *s and VirtIODevice *vdev |
26 | 2 files changed, 84 insertions(+), 73 deletions(-) | 27 | virtio-blk: add acct_failed param to virtio_blk_handle_rw_error() |
28 | virtio-blk: add host_features field in VirtIOBlock | ||
29 | virtio-blk: add "discard" and "write-zeroes" properties | ||
30 | virtio-blk: add DISCARD and WRITE_ZEROES features | ||
31 | tests/virtio-blk: change assert on data_size in virtio_blk_request() | ||
32 | tests/virtio-blk: add test for WRITE_ZEROES command | ||
33 | |||
34 | Vladimir Sementsov-Ogievskiy (1): | ||
35 | qemugdb/coroutine: fix arch_prctl has unknown return type | ||
36 | |||
37 | include/hw/virtio/virtio-blk.h | 5 +- | ||
38 | hw/block/virtio-blk.c | 236 +++++++++++++++++++++++++++++---- | ||
39 | hw/core/machine.c | 2 + | ||
40 | iothread.c | 6 +- | ||
41 | tests/virtio-blk-test.c | 75 ++++++++++- | ||
42 | scripts/qemugdb/coroutine.py | 2 +- | ||
43 | 6 files changed, 297 insertions(+), 29 deletions(-) | ||
27 | 44 | ||
28 | -- | 45 | -- |
29 | 2.20.1 | 46 | 2.20.1 |
30 | 47 | ||
31 | 48 | diff view generated by jsdifflib |
New patch | |||
---|---|---|---|
1 | From: Peter Xu <peterx@redhat.com> | ||
1 | 2 | ||
3 | Lukas reported an hard to reproduce QMP iothread hang on s390 that | ||
4 | QEMU might hang at pthread_join() of the QMP monitor iothread before | ||
5 | quitting: | ||
6 | |||
7 | Thread 1 | ||
8 | #0 0x000003ffad10932c in pthread_join | ||
9 | #1 0x0000000109e95750 in qemu_thread_join | ||
10 | at /home/thuth/devel/qemu/util/qemu-thread-posix.c:570 | ||
11 | #2 0x0000000109c95a1c in iothread_stop | ||
12 | #3 0x0000000109bb0874 in monitor_cleanup | ||
13 | #4 0x0000000109b55042 in main | ||
14 | |||
15 | While the iothread is still in the main loop: | ||
16 | |||
17 | Thread 4 | ||
18 | #0 0x000003ffad0010e4 in ?? | ||
19 | #1 0x000003ffad553958 in g_main_context_iterate.isra.19 | ||
20 | #2 0x000003ffad553d90 in g_main_loop_run | ||
21 | #3 0x0000000109c9585a in iothread_run | ||
22 | at /home/thuth/devel/qemu/iothread.c:74 | ||
23 | #4 0x0000000109e94752 in qemu_thread_start | ||
24 | at /home/thuth/devel/qemu/util/qemu-thread-posix.c:502 | ||
25 | #5 0x000003ffad10825a in start_thread | ||
26 | #6 0x000003ffad00dcf2 in thread_start | ||
27 | |||
28 | IMHO it's because there's a race between the main thread and iothread | ||
29 | when stopping the thread in following sequence: | ||
30 | |||
31 | main thread iothread | ||
32 | =========== ============== | ||
33 | aio_poll() | ||
34 | iothread_get_g_main_context | ||
35 | set iothread->worker_context | ||
36 | iothread_stop | ||
37 | schedule iothread_stop_bh | ||
38 | execute iothread_stop_bh [1] | ||
39 | set iothread->running=false | ||
40 | (since main_loop==NULL so | ||
41 | skip to quit main loop. | ||
42 | Note: although main_loop is | ||
43 | NULL but worker_context is | ||
44 | not!) | ||
45 | atomic_read(&iothread->worker_context) [2] | ||
46 | create main_loop object | ||
47 | g_main_loop_run() [3] | ||
48 | pthread_join() [4] | ||
49 | |||
50 | We can see that when execute iothread_stop_bh() at [1] it's possible | ||
51 | that main_loop is still NULL because it's only created until the first | ||
52 | check of the worker_context later at [2]. Then the iothread will hang | ||
53 | in the main loop [3] and it'll starve the main thread too [4]. | ||
54 | |||
55 | Here the simple solution should be that we check again the "running" | ||
56 | variable before check against worker_context. | ||
57 | |||
58 | CC: Thomas Huth <thuth@redhat.com> | ||
59 | CC: Dr. David Alan Gilbert <dgilbert@redhat.com> | ||
60 | CC: Stefan Hajnoczi <stefanha@redhat.com> | ||
61 | CC: Lukáš Doktor <ldoktor@redhat.com> | ||
62 | CC: Markus Armbruster <armbru@redhat.com> | ||
63 | CC: Eric Blake <eblake@redhat.com> | ||
64 | CC: Paolo Bonzini <pbonzini@redhat.com> | ||
65 | Reported-by: Lukáš Doktor <ldoktor@redhat.com> | ||
66 | Signed-off-by: Peter Xu <peterx@redhat.com> | ||
67 | Tested-by: Thomas Huth <thuth@redhat.com> | ||
68 | Message-id: 20190129051432.22023-1-peterx@redhat.com | ||
69 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
70 | --- | ||
71 | iothread.c | 6 +++++- | ||
72 | 1 file changed, 5 insertions(+), 1 deletion(-) | ||
73 | |||
74 | diff --git a/iothread.c b/iothread.c | ||
75 | index XXXXXXX..XXXXXXX 100644 | ||
76 | --- a/iothread.c | ||
77 | +++ b/iothread.c | ||
78 | @@ -XXX,XX +XXX,XX @@ static void *iothread_run(void *opaque) | ||
79 | while (iothread->running) { | ||
80 | aio_poll(iothread->ctx, true); | ||
81 | |||
82 | - if (atomic_read(&iothread->worker_context)) { | ||
83 | + /* | ||
84 | + * We must check the running state again in case it was | ||
85 | + * changed in previous aio_poll() | ||
86 | + */ | ||
87 | + if (iothread->running && atomic_read(&iothread->worker_context)) { | ||
88 | GMainLoop *loop; | ||
89 | |||
90 | g_main_context_push_thread_default(iothread->worker_context); | ||
91 | -- | ||
92 | 2.20.1 | ||
93 | |||
94 | diff view generated by jsdifflib |
New patch | |||
---|---|---|---|
1 | From: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> | ||
1 | 2 | ||
3 | qemu coroutine command results in following error output: | ||
4 | |||
5 | Python Exception <class 'gdb.error'> 'arch_prctl' has unknown return | ||
6 | type; cast the call to its declared return type: Error occurred in | ||
7 | Python command: 'arch_prctl' has unknown return type; cast the call to | ||
8 | its declared return type | ||
9 | |||
10 | Fix it by giving it what it wants: arch_prctl return type. | ||
11 | |||
12 | Information on the topic: | ||
13 | https://sourceware.org/gdb/onlinedocs/gdb/Calling.html | ||
14 | |||
15 | Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> | ||
16 | Message-id: 20190206151425.105871-1-vsementsov@virtuozzo.com | ||
17 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
18 | --- | ||
19 | scripts/qemugdb/coroutine.py | 2 +- | ||
20 | 1 file changed, 1 insertion(+), 1 deletion(-) | ||
21 | |||
22 | diff --git a/scripts/qemugdb/coroutine.py b/scripts/qemugdb/coroutine.py | ||
23 | index XXXXXXX..XXXXXXX 100644 | ||
24 | --- a/scripts/qemugdb/coroutine.py | ||
25 | +++ b/scripts/qemugdb/coroutine.py | ||
26 | @@ -XXX,XX +XXX,XX @@ def get_fs_base(): | ||
27 | pthread_self().''' | ||
28 | # %rsp - 120 is scratch space according to the SystemV ABI | ||
29 | old = gdb.parse_and_eval('*(uint64_t*)($rsp - 120)') | ||
30 | - gdb.execute('call arch_prctl(0x1003, $rsp - 120)', False, True) | ||
31 | + gdb.execute('call (int)arch_prctl(0x1003, $rsp - 120)', False, True) | ||
32 | fs_base = gdb.parse_and_eval('*(uint64_t*)($rsp - 120)') | ||
33 | gdb.execute('set *(uint64_t*)($rsp - 120) = %s' % old, False, True) | ||
34 | return fs_base | ||
35 | -- | ||
36 | 2.20.1 | ||
37 | |||
38 | diff view generated by jsdifflib |
New patch | |||
---|---|---|---|
1 | From: Stefano Garzarella <sgarzare@redhat.com> | ||
1 | 2 | ||
3 | In several part we still using req->dev or VIRTIO_DEVICE(req->dev) | ||
4 | when we have already defined s and vdev pointers: | ||
5 | VirtIOBlock *s = req->dev; | ||
6 | VirtIODevice *vdev = VIRTIO_DEVICE(s); | ||
7 | |||
8 | Signed-off-by: Stefano Garzarella <sgarzare@redhat.com> | ||
9 | Reviewed-by: Liam Merwick <liam.merwick@oracle.com> | ||
10 | Message-id: 20190208142347.214815-1-sgarzare@redhat.com | ||
11 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
12 | --- | ||
13 | hw/block/virtio-blk.c | 22 +++++++++------------- | ||
14 | 1 file changed, 9 insertions(+), 13 deletions(-) | ||
15 | |||
16 | diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c | ||
17 | index XXXXXXX..XXXXXXX 100644 | ||
18 | --- a/hw/block/virtio-blk.c | ||
19 | +++ b/hw/block/virtio-blk.c | ||
20 | @@ -XXX,XX +XXX,XX @@ static void virtio_blk_req_complete(VirtIOBlockReq *req, unsigned char status) | ||
21 | static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error, | ||
22 | bool is_read) | ||
23 | { | ||
24 | - BlockErrorAction action = blk_get_error_action(req->dev->blk, | ||
25 | - is_read, error); | ||
26 | VirtIOBlock *s = req->dev; | ||
27 | + BlockErrorAction action = blk_get_error_action(s->blk, is_read, error); | ||
28 | |||
29 | if (action == BLOCK_ERROR_ACTION_STOP) { | ||
30 | /* Break the link as the next request is going to be parsed from the | ||
31 | @@ -XXX,XX +XXX,XX @@ static void virtio_blk_flush_complete(void *opaque, int ret) | ||
32 | } | ||
33 | |||
34 | virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); | ||
35 | - block_acct_done(blk_get_stats(req->dev->blk), &req->acct); | ||
36 | + block_acct_done(blk_get_stats(s->blk), &req->acct); | ||
37 | virtio_blk_free_request(req); | ||
38 | |||
39 | out: | ||
40 | @@ -XXX,XX +XXX,XX @@ static int virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb) | ||
41 | - sizeof(struct virtio_blk_inhdr); | ||
42 | iov_discard_back(in_iov, &in_num, sizeof(struct virtio_blk_inhdr)); | ||
43 | |||
44 | - type = virtio_ldl_p(VIRTIO_DEVICE(req->dev), &req->out.type); | ||
45 | + type = virtio_ldl_p(vdev, &req->out.type); | ||
46 | |||
47 | /* VIRTIO_BLK_T_OUT defines the command direction. VIRTIO_BLK_T_BARRIER | ||
48 | * is an optional flag. Although a guest should not send this flag if | ||
49 | @@ -XXX,XX +XXX,XX @@ static int virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb) | ||
50 | case VIRTIO_BLK_T_IN: | ||
51 | { | ||
52 | bool is_write = type & VIRTIO_BLK_T_OUT; | ||
53 | - req->sector_num = virtio_ldq_p(VIRTIO_DEVICE(req->dev), | ||
54 | - &req->out.sector); | ||
55 | + req->sector_num = virtio_ldq_p(vdev, &req->out.sector); | ||
56 | |||
57 | if (is_write) { | ||
58 | qemu_iovec_init_external(&req->qiov, out_iov, out_num); | ||
59 | @@ -XXX,XX +XXX,XX @@ static int virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb) | ||
60 | req->qiov.size / BDRV_SECTOR_SIZE); | ||
61 | } | ||
62 | |||
63 | - if (!virtio_blk_sect_range_ok(req->dev, req->sector_num, | ||
64 | - req->qiov.size)) { | ||
65 | + if (!virtio_blk_sect_range_ok(s, req->sector_num, req->qiov.size)) { | ||
66 | virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR); | ||
67 | - block_acct_invalid(blk_get_stats(req->dev->blk), | ||
68 | + block_acct_invalid(blk_get_stats(s->blk), | ||
69 | is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ); | ||
70 | virtio_blk_free_request(req); | ||
71 | return 0; | ||
72 | } | ||
73 | |||
74 | - block_acct_start(blk_get_stats(req->dev->blk), | ||
75 | - &req->acct, req->qiov.size, | ||
76 | + block_acct_start(blk_get_stats(s->blk), &req->acct, req->qiov.size, | ||
77 | is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ); | ||
78 | |||
79 | /* merge would exceed maximum number of requests or IO direction | ||
80 | * changes */ | ||
81 | if (mrb->num_reqs > 0 && (mrb->num_reqs == VIRTIO_BLK_MAX_MERGE_REQS || | ||
82 | is_write != mrb->is_write || | ||
83 | - !req->dev->conf.request_merging)) { | ||
84 | - virtio_blk_submit_multireq(req->dev->blk, mrb); | ||
85 | + !s->conf.request_merging)) { | ||
86 | + virtio_blk_submit_multireq(s->blk, mrb); | ||
87 | } | ||
88 | |||
89 | assert(mrb->num_reqs < VIRTIO_BLK_MAX_MERGE_REQS); | ||
90 | -- | ||
91 | 2.20.1 | ||
92 | |||
93 | diff view generated by jsdifflib |
New patch | |||
---|---|---|---|
1 | From: Stefano Garzarella <sgarzare@redhat.com> | ||
1 | 2 | ||
3 | We add acct_failed param in order to use virtio_blk_handle_rw_error() | ||
4 | also when is not required to call block_acct_failed(). (eg. a discard | ||
5 | operation is failed) | ||
6 | |||
7 | Reviewed-by: Michael S. Tsirkin <mst@redhat.com> | ||
8 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
9 | Signed-off-by: Stefano Garzarella <sgarzare@redhat.com> | ||
10 | Acked-by: Pankaj Gupta <pagupta@redhat.com> | ||
11 | Message-id: 20190208134950.187665-2-sgarzare@redhat.com | ||
12 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
13 | --- | ||
14 | hw/block/virtio-blk.c | 10 ++++++---- | ||
15 | 1 file changed, 6 insertions(+), 4 deletions(-) | ||
16 | |||
17 | diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c | ||
18 | index XXXXXXX..XXXXXXX 100644 | ||
19 | --- a/hw/block/virtio-blk.c | ||
20 | +++ b/hw/block/virtio-blk.c | ||
21 | @@ -XXX,XX +XXX,XX @@ static void virtio_blk_req_complete(VirtIOBlockReq *req, unsigned char status) | ||
22 | } | ||
23 | |||
24 | static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error, | ||
25 | - bool is_read) | ||
26 | + bool is_read, bool acct_failed) | ||
27 | { | ||
28 | VirtIOBlock *s = req->dev; | ||
29 | BlockErrorAction action = blk_get_error_action(s->blk, is_read, error); | ||
30 | @@ -XXX,XX +XXX,XX @@ static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error, | ||
31 | s->rq = req; | ||
32 | } else if (action == BLOCK_ERROR_ACTION_REPORT) { | ||
33 | virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR); | ||
34 | - block_acct_failed(blk_get_stats(s->blk), &req->acct); | ||
35 | + if (acct_failed) { | ||
36 | + block_acct_failed(blk_get_stats(s->blk), &req->acct); | ||
37 | + } | ||
38 | virtio_blk_free_request(req); | ||
39 | } | ||
40 | |||
41 | @@ -XXX,XX +XXX,XX @@ static void virtio_blk_rw_complete(void *opaque, int ret) | ||
42 | * the memory until the request is completed (which will | ||
43 | * happen on the other side of the migration). | ||
44 | */ | ||
45 | - if (virtio_blk_handle_rw_error(req, -ret, is_read)) { | ||
46 | + if (virtio_blk_handle_rw_error(req, -ret, is_read, true)) { | ||
47 | continue; | ||
48 | } | ||
49 | } | ||
50 | @@ -XXX,XX +XXX,XX @@ static void virtio_blk_flush_complete(void *opaque, int ret) | ||
51 | |||
52 | aio_context_acquire(blk_get_aio_context(s->conf.conf.blk)); | ||
53 | if (ret) { | ||
54 | - if (virtio_blk_handle_rw_error(req, -ret, 0)) { | ||
55 | + if (virtio_blk_handle_rw_error(req, -ret, 0, true)) { | ||
56 | goto out; | ||
57 | } | ||
58 | } | ||
59 | -- | ||
60 | 2.20.1 | ||
61 | |||
62 | diff view generated by jsdifflib |
New patch | |||
---|---|---|---|
1 | From: Stefano Garzarella <sgarzare@redhat.com> | ||
1 | 2 | ||
3 | Since configurable features for virtio-blk are growing, this patch | ||
4 | adds host_features field in the struct VirtIOBlock. (as in virtio-net) | ||
5 | In this way, we can avoid to add new fields for new properties and | ||
6 | we can directly set VIRTIO_BLK_F* flags in the host_features. | ||
7 | |||
8 | We update "config-wce" and "scsi" property definition to use the new | ||
9 | host_features field without change the behaviour. | ||
10 | |||
11 | Suggested-by: Michael S. Tsirkin <mst@redhat.com> | ||
12 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
13 | Signed-off-by: Stefano Garzarella <sgarzare@redhat.com> | ||
14 | Acked-by: Pankaj Gupta <pagupta@redhat.com> | ||
15 | Message-id: 20190208134950.187665-3-sgarzare@redhat.com | ||
16 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
17 | --- | ||
18 | include/hw/virtio/virtio-blk.h | 3 +-- | ||
19 | hw/block/virtio-blk.c | 16 +++++++++------- | ||
20 | 2 files changed, 10 insertions(+), 9 deletions(-) | ||
21 | |||
22 | diff --git a/include/hw/virtio/virtio-blk.h b/include/hw/virtio/virtio-blk.h | ||
23 | index XXXXXXX..XXXXXXX 100644 | ||
24 | --- a/include/hw/virtio/virtio-blk.h | ||
25 | +++ b/include/hw/virtio/virtio-blk.h | ||
26 | @@ -XXX,XX +XXX,XX @@ struct VirtIOBlkConf | ||
27 | BlockConf conf; | ||
28 | IOThread *iothread; | ||
29 | char *serial; | ||
30 | - uint32_t scsi; | ||
31 | - uint32_t config_wce; | ||
32 | uint32_t request_merging; | ||
33 | uint16_t num_queues; | ||
34 | uint16_t queue_size; | ||
35 | @@ -XXX,XX +XXX,XX @@ typedef struct VirtIOBlock { | ||
36 | bool dataplane_disabled; | ||
37 | bool dataplane_started; | ||
38 | struct VirtIOBlockDataPlane *dataplane; | ||
39 | + uint64_t host_features; | ||
40 | } VirtIOBlock; | ||
41 | |||
42 | typedef struct VirtIOBlockReq { | ||
43 | diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c | ||
44 | index XXXXXXX..XXXXXXX 100644 | ||
45 | --- a/hw/block/virtio-blk.c | ||
46 | +++ b/hw/block/virtio-blk.c | ||
47 | @@ -XXX,XX +XXX,XX @@ static int virtio_blk_handle_scsi_req(VirtIOBlockReq *req) | ||
48 | */ | ||
49 | scsi = (void *)elem->in_sg[elem->in_num - 2].iov_base; | ||
50 | |||
51 | - if (!blk->conf.scsi) { | ||
52 | + if (!virtio_has_feature(blk->host_features, VIRTIO_BLK_F_SCSI)) { | ||
53 | status = VIRTIO_BLK_S_UNSUPP; | ||
54 | goto fail; | ||
55 | } | ||
56 | @@ -XXX,XX +XXX,XX @@ static uint64_t virtio_blk_get_features(VirtIODevice *vdev, uint64_t features, | ||
57 | { | ||
58 | VirtIOBlock *s = VIRTIO_BLK(vdev); | ||
59 | |||
60 | + /* Firstly sync all virtio-blk possible supported features */ | ||
61 | + features |= s->host_features; | ||
62 | + | ||
63 | virtio_add_feature(&features, VIRTIO_BLK_F_SEG_MAX); | ||
64 | virtio_add_feature(&features, VIRTIO_BLK_F_GEOMETRY); | ||
65 | virtio_add_feature(&features, VIRTIO_BLK_F_TOPOLOGY); | ||
66 | virtio_add_feature(&features, VIRTIO_BLK_F_BLK_SIZE); | ||
67 | if (virtio_has_feature(features, VIRTIO_F_VERSION_1)) { | ||
68 | - if (s->conf.scsi) { | ||
69 | + if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_SCSI)) { | ||
70 | error_setg(errp, "Please set scsi=off for virtio-blk devices in order to use virtio 1.0"); | ||
71 | return 0; | ||
72 | } | ||
73 | @@ -XXX,XX +XXX,XX @@ static uint64_t virtio_blk_get_features(VirtIODevice *vdev, uint64_t features, | ||
74 | virtio_add_feature(&features, VIRTIO_BLK_F_SCSI); | ||
75 | } | ||
76 | |||
77 | - if (s->conf.config_wce) { | ||
78 | - virtio_add_feature(&features, VIRTIO_BLK_F_CONFIG_WCE); | ||
79 | - } | ||
80 | if (blk_enable_write_cache(s->blk)) { | ||
81 | virtio_add_feature(&features, VIRTIO_BLK_F_WCE); | ||
82 | } | ||
83 | @@ -XXX,XX +XXX,XX @@ static Property virtio_blk_properties[] = { | ||
84 | DEFINE_BLOCK_ERROR_PROPERTIES(VirtIOBlock, conf.conf), | ||
85 | DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock, conf.conf), | ||
86 | DEFINE_PROP_STRING("serial", VirtIOBlock, conf.serial), | ||
87 | - DEFINE_PROP_BIT("config-wce", VirtIOBlock, conf.config_wce, 0, true), | ||
88 | + DEFINE_PROP_BIT64("config-wce", VirtIOBlock, host_features, | ||
89 | + VIRTIO_BLK_F_CONFIG_WCE, true), | ||
90 | #ifdef __linux__ | ||
91 | - DEFINE_PROP_BIT("scsi", VirtIOBlock, conf.scsi, 0, false), | ||
92 | + DEFINE_PROP_BIT64("scsi", VirtIOBlock, host_features, | ||
93 | + VIRTIO_BLK_F_SCSI, false), | ||
94 | #endif | ||
95 | DEFINE_PROP_BIT("request-merging", VirtIOBlock, conf.request_merging, 0, | ||
96 | true), | ||
97 | -- | ||
98 | 2.20.1 | ||
99 | |||
100 | diff view generated by jsdifflib |
New patch | |||
---|---|---|---|
1 | From: Stefano Garzarella <sgarzare@redhat.com> | ||
1 | 2 | ||
3 | In order to avoid migration issues, we enable DISCARD and | ||
4 | WRITE_ZEROES features only for machine type >= 4.0 | ||
5 | |||
6 | As discussed with Michael S. Tsirkin and Stefan Hajnoczi on the | ||
7 | list [1], DISCARD operation should not have security implications | ||
8 | (eg. page cache attacks), so we can enable it by default. | ||
9 | |||
10 | [1] https://lists.gnu.org/archive/html/qemu-devel/2019-02/msg00504.html | ||
11 | |||
12 | Suggested-by: Dr. David Alan Gilbert <dgilbert@redhat.com> | ||
13 | Reviewed-by: Michael S. Tsirkin <mst@redhat.com> | ||
14 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
15 | Signed-off-by: Stefano Garzarella <sgarzare@redhat.com> | ||
16 | Acked-by: Pankaj Gupta <pagupta@redhat.com> | ||
17 | Message-id: 20190208134950.187665-4-sgarzare@redhat.com | ||
18 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
19 | --- | ||
20 | hw/block/virtio-blk.c | 4 ++++ | ||
21 | hw/core/machine.c | 2 ++ | ||
22 | 2 files changed, 6 insertions(+) | ||
23 | |||
24 | diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c | ||
25 | index XXXXXXX..XXXXXXX 100644 | ||
26 | --- a/hw/block/virtio-blk.c | ||
27 | +++ b/hw/block/virtio-blk.c | ||
28 | @@ -XXX,XX +XXX,XX @@ static Property virtio_blk_properties[] = { | ||
29 | DEFINE_PROP_UINT16("queue-size", VirtIOBlock, conf.queue_size, 128), | ||
30 | DEFINE_PROP_LINK("iothread", VirtIOBlock, conf.iothread, TYPE_IOTHREAD, | ||
31 | IOThread *), | ||
32 | + DEFINE_PROP_BIT64("discard", VirtIOBlock, host_features, | ||
33 | + VIRTIO_BLK_F_DISCARD, true), | ||
34 | + DEFINE_PROP_BIT64("write-zeroes", VirtIOBlock, host_features, | ||
35 | + VIRTIO_BLK_F_WRITE_ZEROES, true), | ||
36 | DEFINE_PROP_END_OF_LIST(), | ||
37 | }; | ||
38 | |||
39 | diff --git a/hw/core/machine.c b/hw/core/machine.c | ||
40 | index XXXXXXX..XXXXXXX 100644 | ||
41 | --- a/hw/core/machine.c | ||
42 | +++ b/hw/core/machine.c | ||
43 | @@ -XXX,XX +XXX,XX @@ GlobalProperty hw_compat_3_1[] = { | ||
44 | { "usb-kbd", "serial", "42" }, | ||
45 | { "usb-mouse", "serial", "42" }, | ||
46 | { "usb-kbd", "serial", "42" }, | ||
47 | + { "virtio-blk-device", "discard", "false" }, | ||
48 | + { "virtio-blk-device", "write-zeroes", "false" }, | ||
49 | }; | ||
50 | const size_t hw_compat_3_1_len = G_N_ELEMENTS(hw_compat_3_1); | ||
51 | |||
52 | -- | ||
53 | 2.20.1 | ||
54 | |||
55 | diff view generated by jsdifflib |
1 | From: Remy Noel <remy.noel@blade-group.com> | 1 | From: Stefano Garzarella <sgarzare@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | It is possible for an io_poll callback to be concurrently executed along | 3 | This patch adds the support of DISCARD and WRITE_ZEROES commands, |
4 | with an aio_set_fd_handlers. This can cause all sorts of problems, like | 4 | that have been introduced in the virtio-blk protocol to have |
5 | a NULL callback or a bad opaque pointer. | 5 | better performance when using SSD backend. |
6 | 6 | ||
7 | This changes set_fd_handlers so that it no longer modify existing handlers | 7 | We support only one segment per request since multiple segments |
8 | entries and instead, always insert those after having proper initialisation. | 8 | are not widely used and there are no userspace APIs that allow |
9 | 9 | applications to submit multiple segments in a single call. | |
10 | Tested-by: Stefan Hajnoczi <stefanha@redhat.com> | 10 | |
11 | Signed-off-by: Remy Noel <remy.noel@blade-group.com> | 11 | Reviewed-by: Michael S. Tsirkin <mst@redhat.com> |
12 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> | 12 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> |
13 | Reviewed-by: Paolo Bonzini <pbonzini@redhat.com> | 13 | Signed-off-by: Stefano Garzarella <sgarzare@redhat.com> |
14 | Message-id: 20181220152030.28035-3-remy.noel@blade-group.com | 14 | Acked-by: Pankaj Gupta <pagupta@redhat.com> |
15 | Message-id: 20190208134950.187665-5-sgarzare@redhat.com | ||
15 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> | 16 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> |
16 | --- | 17 | --- |
17 | util/aio-posix.c | 89 ++++++++++++++++++++++++++++-------------------- | 18 | include/hw/virtio/virtio-blk.h | 2 + |
18 | util/aio-win32.c | 67 ++++++++++++++++-------------------- | 19 | hw/block/virtio-blk.c | 184 +++++++++++++++++++++++++++++++++ |
19 | 2 files changed, 82 insertions(+), 74 deletions(-) | 20 | 2 files changed, 186 insertions(+) |
20 | 21 | ||
21 | diff --git a/util/aio-posix.c b/util/aio-posix.c | 22 | diff --git a/include/hw/virtio/virtio-blk.h b/include/hw/virtio/virtio-blk.h |
22 | index XXXXXXX..XXXXXXX 100644 | 23 | index XXXXXXX..XXXXXXX 100644 |
23 | --- a/util/aio-posix.c | 24 | --- a/include/hw/virtio/virtio-blk.h |
24 | +++ b/util/aio-posix.c | 25 | +++ b/include/hw/virtio/virtio-blk.h |
25 | @@ -XXX,XX +XXX,XX @@ static AioHandler *find_aio_handler(AioContext *ctx, int fd) | 26 | @@ -XXX,XX +XXX,XX @@ struct VirtIOBlkConf |
26 | return NULL; | 27 | uint32_t request_merging; |
28 | uint16_t num_queues; | ||
29 | uint16_t queue_size; | ||
30 | + uint32_t max_discard_sectors; | ||
31 | + uint32_t max_write_zeroes_sectors; | ||
32 | }; | ||
33 | |||
34 | struct VirtIOBlockDataPlane; | ||
35 | diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c | ||
36 | index XXXXXXX..XXXXXXX 100644 | ||
37 | --- a/hw/block/virtio-blk.c | ||
38 | +++ b/hw/block/virtio-blk.c | ||
39 | @@ -XXX,XX +XXX,XX @@ out: | ||
40 | aio_context_release(blk_get_aio_context(s->conf.conf.blk)); | ||
27 | } | 41 | } |
28 | 42 | ||
29 | +static bool aio_remove_fd_handler(AioContext *ctx, AioHandler *node) | 43 | +static void virtio_blk_discard_write_zeroes_complete(void *opaque, int ret) |
30 | +{ | 44 | +{ |
31 | + /* If the GSource is in the process of being destroyed then | 45 | + VirtIOBlockReq *req = opaque; |
32 | + * g_source_remove_poll() causes an assertion failure. Skip | 46 | + VirtIOBlock *s = req->dev; |
33 | + * removal in that case, because glib cleans up its state during | 47 | + bool is_write_zeroes = (virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type) & |
34 | + * destruction anyway. | 48 | + ~VIRTIO_BLK_T_BARRIER) == VIRTIO_BLK_T_WRITE_ZEROES; |
49 | + | ||
50 | + aio_context_acquire(blk_get_aio_context(s->conf.conf.blk)); | ||
51 | + if (ret) { | ||
52 | + if (virtio_blk_handle_rw_error(req, -ret, false, is_write_zeroes)) { | ||
53 | + goto out; | ||
54 | + } | ||
55 | + } | ||
56 | + | ||
57 | + virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); | ||
58 | + if (is_write_zeroes) { | ||
59 | + block_acct_done(blk_get_stats(s->blk), &req->acct); | ||
60 | + } | ||
61 | + virtio_blk_free_request(req); | ||
62 | + | ||
63 | +out: | ||
64 | + aio_context_release(blk_get_aio_context(s->conf.conf.blk)); | ||
65 | +} | ||
66 | + | ||
67 | #ifdef __linux__ | ||
68 | |||
69 | typedef struct { | ||
70 | @@ -XXX,XX +XXX,XX @@ static bool virtio_blk_sect_range_ok(VirtIOBlock *dev, | ||
71 | return true; | ||
72 | } | ||
73 | |||
74 | +static uint8_t virtio_blk_handle_discard_write_zeroes(VirtIOBlockReq *req, | ||
75 | + struct virtio_blk_discard_write_zeroes *dwz_hdr, bool is_write_zeroes) | ||
76 | +{ | ||
77 | + VirtIOBlock *s = req->dev; | ||
78 | + VirtIODevice *vdev = VIRTIO_DEVICE(s); | ||
79 | + uint64_t sector; | ||
80 | + uint32_t num_sectors, flags, max_sectors; | ||
81 | + uint8_t err_status; | ||
82 | + int bytes; | ||
83 | + | ||
84 | + sector = virtio_ldq_p(vdev, &dwz_hdr->sector); | ||
85 | + num_sectors = virtio_ldl_p(vdev, &dwz_hdr->num_sectors); | ||
86 | + flags = virtio_ldl_p(vdev, &dwz_hdr->flags); | ||
87 | + max_sectors = is_write_zeroes ? s->conf.max_write_zeroes_sectors : | ||
88 | + s->conf.max_discard_sectors; | ||
89 | + | ||
90 | + /* | ||
91 | + * max_sectors is at most BDRV_REQUEST_MAX_SECTORS, this check | ||
92 | + * make us sure that "num_sectors << BDRV_SECTOR_BITS" can fit in | ||
93 | + * the integer variable. | ||
35 | + */ | 94 | + */ |
36 | + if (!g_source_is_destroyed(&ctx->source)) { | 95 | + if (unlikely(num_sectors > max_sectors)) { |
37 | + g_source_remove_poll(&ctx->source, &node->pfd); | 96 | + err_status = VIRTIO_BLK_S_IOERR; |
38 | + } | 97 | + goto err; |
39 | + | 98 | + } |
40 | + /* If a read is in progress, just mark the node as deleted */ | 99 | + |
41 | + if (qemu_lockcnt_count(&ctx->list_lock)) { | 100 | + bytes = num_sectors << BDRV_SECTOR_BITS; |
42 | + node->deleted = 1; | 101 | + |
43 | + node->pfd.revents = 0; | 102 | + if (unlikely(!virtio_blk_sect_range_ok(s, sector, bytes))) { |
44 | + return false; | 103 | + err_status = VIRTIO_BLK_S_IOERR; |
45 | + } | 104 | + goto err; |
46 | + /* Otherwise, delete it for real. We can't just mark it as | 105 | + } |
47 | + * deleted because deleted nodes are only cleaned up while | 106 | + |
48 | + * no one is walking the handlers list. | 107 | + /* |
108 | + * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for discard | ||
109 | + * and write zeroes commands if any unknown flag is set. | ||
49 | + */ | 110 | + */ |
50 | + QLIST_REMOVE(node, node); | 111 | + if (unlikely(flags & ~VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) { |
51 | + return true; | 112 | + err_status = VIRTIO_BLK_S_UNSUPP; |
113 | + goto err; | ||
114 | + } | ||
115 | + | ||
116 | + if (is_write_zeroes) { /* VIRTIO_BLK_T_WRITE_ZEROES */ | ||
117 | + int blk_aio_flags = 0; | ||
118 | + | ||
119 | + if (flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP) { | ||
120 | + blk_aio_flags |= BDRV_REQ_MAY_UNMAP; | ||
121 | + } | ||
122 | + | ||
123 | + block_acct_start(blk_get_stats(s->blk), &req->acct, bytes, | ||
124 | + BLOCK_ACCT_WRITE); | ||
125 | + | ||
126 | + blk_aio_pwrite_zeroes(s->blk, sector << BDRV_SECTOR_BITS, | ||
127 | + bytes, blk_aio_flags, | ||
128 | + virtio_blk_discard_write_zeroes_complete, req); | ||
129 | + } else { /* VIRTIO_BLK_T_DISCARD */ | ||
130 | + /* | ||
131 | + * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for | ||
132 | + * discard commands if the unmap flag is set. | ||
133 | + */ | ||
134 | + if (unlikely(flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) { | ||
135 | + err_status = VIRTIO_BLK_S_UNSUPP; | ||
136 | + goto err; | ||
137 | + } | ||
138 | + | ||
139 | + blk_aio_pdiscard(s->blk, sector << BDRV_SECTOR_BITS, bytes, | ||
140 | + virtio_blk_discard_write_zeroes_complete, req); | ||
141 | + } | ||
142 | + | ||
143 | + return VIRTIO_BLK_S_OK; | ||
144 | + | ||
145 | +err: | ||
146 | + if (is_write_zeroes) { | ||
147 | + block_acct_invalid(blk_get_stats(s->blk), BLOCK_ACCT_WRITE); | ||
148 | + } | ||
149 | + return err_status; | ||
52 | +} | 150 | +} |
53 | + | 151 | + |
54 | void aio_set_fd_handler(AioContext *ctx, | 152 | static int virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb) |
55 | int fd, | ||
56 | bool is_external, | ||
57 | @@ -XXX,XX +XXX,XX @@ void aio_set_fd_handler(AioContext *ctx, | ||
58 | void *opaque) | ||
59 | { | 153 | { |
60 | AioHandler *node; | 154 | uint32_t type; |
61 | + AioHandler *new_node = NULL; | 155 | @@ -XXX,XX +XXX,XX @@ static int virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb) |
62 | bool is_new = false; | 156 | virtio_blk_free_request(req); |
63 | bool deleted = false; | 157 | break; |
64 | int poll_disable_change; | ||
65 | @@ -XXX,XX +XXX,XX @@ void aio_set_fd_handler(AioContext *ctx, | ||
66 | qemu_lockcnt_unlock(&ctx->list_lock); | ||
67 | return; | ||
68 | } | ||
69 | - | ||
70 | - /* If the GSource is in the process of being destroyed then | ||
71 | - * g_source_remove_poll() causes an assertion failure. Skip | ||
72 | - * removal in that case, because glib cleans up its state during | ||
73 | - * destruction anyway. | ||
74 | - */ | ||
75 | - if (!g_source_is_destroyed(&ctx->source)) { | ||
76 | - g_source_remove_poll(&ctx->source, &node->pfd); | ||
77 | - } | ||
78 | - | ||
79 | - /* If a read is in progress, just mark the node as deleted */ | ||
80 | - if (qemu_lockcnt_count(&ctx->list_lock)) { | ||
81 | - node->deleted = 1; | ||
82 | - node->pfd.revents = 0; | ||
83 | - } else { | ||
84 | - /* Otherwise, delete it for real. We can't just mark it as | ||
85 | - * deleted because deleted nodes are only cleaned up while | ||
86 | - * no one is walking the handlers list. | ||
87 | - */ | ||
88 | - QLIST_REMOVE(node, node); | ||
89 | - deleted = true; | ||
90 | - } | ||
91 | /* Clean events in order to unregister fd from the ctx epoll. */ | ||
92 | node->pfd.events = 0; | ||
93 | |||
94 | @@ -XXX,XX +XXX,XX @@ void aio_set_fd_handler(AioContext *ctx, | ||
95 | } else { | ||
96 | poll_disable_change = !io_poll - (node && !node->io_poll); | ||
97 | if (node == NULL) { | ||
98 | - /* Alloc and insert if it's not already there */ | ||
99 | - node = g_new0(AioHandler, 1); | ||
100 | - node->pfd.fd = fd; | ||
101 | - QLIST_INSERT_HEAD_RCU(&ctx->aio_handlers, node, node); | ||
102 | - | ||
103 | - g_source_add_poll(&ctx->source, &node->pfd); | ||
104 | is_new = true; | ||
105 | } | ||
106 | + /* Alloc and insert if it's not already there */ | ||
107 | + new_node = g_new0(AioHandler, 1); | ||
108 | |||
109 | /* Update handler with latest information */ | ||
110 | - node->io_read = io_read; | ||
111 | - node->io_write = io_write; | ||
112 | - node->io_poll = io_poll; | ||
113 | - node->opaque = opaque; | ||
114 | - node->is_external = is_external; | ||
115 | + new_node->io_read = io_read; | ||
116 | + new_node->io_write = io_write; | ||
117 | + new_node->io_poll = io_poll; | ||
118 | + new_node->opaque = opaque; | ||
119 | + new_node->is_external = is_external; | ||
120 | |||
121 | - node->pfd.events = (io_read ? G_IO_IN | G_IO_HUP | G_IO_ERR : 0); | ||
122 | - node->pfd.events |= (io_write ? G_IO_OUT | G_IO_ERR : 0); | ||
123 | + if (is_new) { | ||
124 | + new_node->pfd.fd = fd; | ||
125 | + } else { | ||
126 | + new_node->pfd = node->pfd; | ||
127 | + } | ||
128 | + g_source_add_poll(&ctx->source, &new_node->pfd); | ||
129 | + | ||
130 | + new_node->pfd.events = (io_read ? G_IO_IN | G_IO_HUP | G_IO_ERR : 0); | ||
131 | + new_node->pfd.events |= (io_write ? G_IO_OUT | G_IO_ERR : 0); | ||
132 | + | ||
133 | + QLIST_INSERT_HEAD_RCU(&ctx->aio_handlers, new_node, node); | ||
134 | + } | ||
135 | + if (node) { | ||
136 | + deleted = aio_remove_fd_handler(ctx, node); | ||
137 | } | 158 | } |
138 | 159 | + /* | |
139 | /* No need to order poll_disable_cnt writes against other updates; | 160 | + * VIRTIO_BLK_T_DISCARD and VIRTIO_BLK_T_WRITE_ZEROES are defined with |
140 | @@ -XXX,XX +XXX,XX @@ void aio_set_fd_handler(AioContext *ctx, | 161 | + * VIRTIO_BLK_T_OUT flag set. We masked this flag in the switch statement, |
141 | atomic_set(&ctx->poll_disable_cnt, | 162 | + * so we must mask it for these requests, then we will check if it is set. |
142 | atomic_read(&ctx->poll_disable_cnt) + poll_disable_change); | 163 | + */ |
143 | 164 | + case VIRTIO_BLK_T_DISCARD & ~VIRTIO_BLK_T_OUT: | |
144 | - aio_epoll_update(ctx, node, is_new); | 165 | + case VIRTIO_BLK_T_WRITE_ZEROES & ~VIRTIO_BLK_T_OUT: |
145 | + if (new_node) { | 166 | + { |
146 | + aio_epoll_update(ctx, new_node, is_new); | 167 | + struct virtio_blk_discard_write_zeroes dwz_hdr; |
147 | + } else if (node) { | 168 | + size_t out_len = iov_size(out_iov, out_num); |
148 | + /* Unregister deleted fd_handler */ | 169 | + bool is_write_zeroes = (type & ~VIRTIO_BLK_T_BARRIER) == |
149 | + aio_epoll_update(ctx, node, false); | 170 | + VIRTIO_BLK_T_WRITE_ZEROES; |
150 | + } | 171 | + uint8_t err_status; |
151 | qemu_lockcnt_unlock(&ctx->list_lock); | 172 | + |
152 | aio_notify(ctx); | 173 | + /* |
153 | 174 | + * Unsupported if VIRTIO_BLK_T_OUT is not set or the request contains | |
154 | diff --git a/util/aio-win32.c b/util/aio-win32.c | 175 | + * more than one segment. |
155 | index XXXXXXX..XXXXXXX 100644 | 176 | + */ |
156 | --- a/util/aio-win32.c | 177 | + if (unlikely(!(type & VIRTIO_BLK_T_OUT) || |
157 | +++ b/util/aio-win32.c | 178 | + out_len > sizeof(dwz_hdr))) { |
158 | @@ -XXX,XX +XXX,XX @@ struct AioHandler { | 179 | + virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP); |
159 | QLIST_ENTRY(AioHandler) node; | 180 | + virtio_blk_free_request(req); |
181 | + return 0; | ||
182 | + } | ||
183 | + | ||
184 | + if (unlikely(iov_to_buf(out_iov, out_num, 0, &dwz_hdr, | ||
185 | + sizeof(dwz_hdr)) != sizeof(dwz_hdr))) { | ||
186 | + virtio_error(vdev, "virtio-blk discard/write_zeroes header" | ||
187 | + " too short"); | ||
188 | + return -1; | ||
189 | + } | ||
190 | + | ||
191 | + err_status = virtio_blk_handle_discard_write_zeroes(req, &dwz_hdr, | ||
192 | + is_write_zeroes); | ||
193 | + if (err_status != VIRTIO_BLK_S_OK) { | ||
194 | + virtio_blk_req_complete(req, err_status); | ||
195 | + virtio_blk_free_request(req); | ||
196 | + } | ||
197 | + | ||
198 | + break; | ||
199 | + } | ||
200 | default: | ||
201 | virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP); | ||
202 | virtio_blk_free_request(req); | ||
203 | @@ -XXX,XX +XXX,XX @@ static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config) | ||
204 | blkcfg.alignment_offset = 0; | ||
205 | blkcfg.wce = blk_enable_write_cache(s->blk); | ||
206 | virtio_stw_p(vdev, &blkcfg.num_queues, s->conf.num_queues); | ||
207 | + if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD)) { | ||
208 | + virtio_stl_p(vdev, &blkcfg.max_discard_sectors, | ||
209 | + s->conf.max_discard_sectors); | ||
210 | + virtio_stl_p(vdev, &blkcfg.discard_sector_alignment, | ||
211 | + blk_size >> BDRV_SECTOR_BITS); | ||
212 | + /* | ||
213 | + * We support only one segment per request since multiple segments | ||
214 | + * are not widely used and there are no userspace APIs that allow | ||
215 | + * applications to submit multiple segments in a single call. | ||
216 | + */ | ||
217 | + virtio_stl_p(vdev, &blkcfg.max_discard_seg, 1); | ||
218 | + } | ||
219 | + if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES)) { | ||
220 | + virtio_stl_p(vdev, &blkcfg.max_write_zeroes_sectors, | ||
221 | + s->conf.max_write_zeroes_sectors); | ||
222 | + blkcfg.write_zeroes_may_unmap = 1; | ||
223 | + virtio_stl_p(vdev, &blkcfg.max_write_zeroes_seg, 1); | ||
224 | + } | ||
225 | memcpy(config, &blkcfg, sizeof(struct virtio_blk_config)); | ||
226 | } | ||
227 | |||
228 | @@ -XXX,XX +XXX,XX @@ static void virtio_blk_device_realize(DeviceState *dev, Error **errp) | ||
229 | return; | ||
230 | } | ||
231 | |||
232 | + if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD) && | ||
233 | + (!conf->max_discard_sectors || | ||
234 | + conf->max_discard_sectors > BDRV_REQUEST_MAX_SECTORS)) { | ||
235 | + error_setg(errp, "invalid max-discard-sectors property (%" PRIu32 ")" | ||
236 | + ", must be between 1 and %d", | ||
237 | + conf->max_discard_sectors, (int)BDRV_REQUEST_MAX_SECTORS); | ||
238 | + return; | ||
239 | + } | ||
240 | + | ||
241 | + if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES) && | ||
242 | + (!conf->max_write_zeroes_sectors || | ||
243 | + conf->max_write_zeroes_sectors > BDRV_REQUEST_MAX_SECTORS)) { | ||
244 | + error_setg(errp, "invalid max-write-zeroes-sectors property (%" PRIu32 | ||
245 | + "), must be between 1 and %d", | ||
246 | + conf->max_write_zeroes_sectors, | ||
247 | + (int)BDRV_REQUEST_MAX_SECTORS); | ||
248 | + return; | ||
249 | + } | ||
250 | + | ||
251 | virtio_init(vdev, "virtio-blk", VIRTIO_ID_BLOCK, | ||
252 | sizeof(struct virtio_blk_config)); | ||
253 | |||
254 | @@ -XXX,XX +XXX,XX @@ static Property virtio_blk_properties[] = { | ||
255 | VIRTIO_BLK_F_DISCARD, true), | ||
256 | DEFINE_PROP_BIT64("write-zeroes", VirtIOBlock, host_features, | ||
257 | VIRTIO_BLK_F_WRITE_ZEROES, true), | ||
258 | + DEFINE_PROP_UINT32("max-discard-sectors", VirtIOBlock, | ||
259 | + conf.max_discard_sectors, BDRV_REQUEST_MAX_SECTORS), | ||
260 | + DEFINE_PROP_UINT32("max-write-zeroes-sectors", VirtIOBlock, | ||
261 | + conf.max_write_zeroes_sectors, BDRV_REQUEST_MAX_SECTORS), | ||
262 | DEFINE_PROP_END_OF_LIST(), | ||
160 | }; | 263 | }; |
161 | 264 | ||
162 | +static void aio_remove_fd_handler(AioContext *ctx, AioHandler *node) | ||
163 | +{ | ||
164 | + /* If aio_poll is in progress, just mark the node as deleted */ | ||
165 | + if (qemu_lockcnt_count(&ctx->list_lock)) { | ||
166 | + node->deleted = 1; | ||
167 | + node->pfd.revents = 0; | ||
168 | + } else { | ||
169 | + /* Otherwise, delete it for real. We can't just mark it as | ||
170 | + * deleted because deleted nodes are only cleaned up after | ||
171 | + * releasing the list_lock. | ||
172 | + */ | ||
173 | + QLIST_REMOVE(node, node); | ||
174 | + g_free(node); | ||
175 | + } | ||
176 | +} | ||
177 | + | ||
178 | void aio_set_fd_handler(AioContext *ctx, | ||
179 | int fd, | ||
180 | bool is_external, | ||
181 | @@ -XXX,XX +XXX,XX @@ void aio_set_fd_handler(AioContext *ctx, | ||
182 | void *opaque) | ||
183 | { | ||
184 | /* fd is a SOCKET in our case */ | ||
185 | - AioHandler *node; | ||
186 | + AioHandler *old_node; | ||
187 | + AioHandler *node = NULL; | ||
188 | |||
189 | qemu_lockcnt_lock(&ctx->list_lock); | ||
190 | - QLIST_FOREACH(node, &ctx->aio_handlers, node) { | ||
191 | - if (node->pfd.fd == fd && !node->deleted) { | ||
192 | + QLIST_FOREACH(old_node, &ctx->aio_handlers, node) { | ||
193 | + if (old_node->pfd.fd == fd && !old_node->deleted) { | ||
194 | break; | ||
195 | } | ||
196 | } | ||
197 | |||
198 | - /* Are we deleting the fd handler? */ | ||
199 | - if (!io_read && !io_write) { | ||
200 | - if (node) { | ||
201 | - /* If aio_poll is in progress, just mark the node as deleted */ | ||
202 | - if (qemu_lockcnt_count(&ctx->list_lock)) { | ||
203 | - node->deleted = 1; | ||
204 | - node->pfd.revents = 0; | ||
205 | - } else { | ||
206 | - /* Otherwise, delete it for real. We can't just mark it as | ||
207 | - * deleted because deleted nodes are only cleaned up after | ||
208 | - * releasing the list_lock. | ||
209 | - */ | ||
210 | - QLIST_REMOVE(node, node); | ||
211 | - g_free(node); | ||
212 | - } | ||
213 | - } | ||
214 | - } else { | ||
215 | + if (io_read || io_write) { | ||
216 | HANDLE event; | ||
217 | long bitmask = 0; | ||
218 | |||
219 | - if (node == NULL) { | ||
220 | - /* Alloc and insert if it's not already there */ | ||
221 | - node = g_new0(AioHandler, 1); | ||
222 | - node->pfd.fd = fd; | ||
223 | - QLIST_INSERT_HEAD_RCU(&ctx->aio_handlers, node, node); | ||
224 | - } | ||
225 | + /* Alloc and insert if it's not already there */ | ||
226 | + node = g_new0(AioHandler, 1); | ||
227 | + node->pfd.fd = fd; | ||
228 | |||
229 | node->pfd.events = 0; | ||
230 | if (node->io_read) { | ||
231 | @@ -XXX,XX +XXX,XX @@ void aio_set_fd_handler(AioContext *ctx, | ||
232 | bitmask |= FD_WRITE | FD_CONNECT; | ||
233 | } | ||
234 | |||
235 | + QLIST_INSERT_HEAD_RCU(&ctx->aio_handlers, node, node); | ||
236 | event = event_notifier_get_handle(&ctx->notifier); | ||
237 | WSAEventSelect(node->pfd.fd, event, bitmask); | ||
238 | } | ||
239 | + if (old_node) { | ||
240 | + aio_remove_fd_handler(ctx, old_node); | ||
241 | + } | ||
242 | |||
243 | qemu_lockcnt_unlock(&ctx->list_lock); | ||
244 | aio_notify(ctx); | ||
245 | @@ -XXX,XX +XXX,XX @@ void aio_set_event_notifier(AioContext *ctx, | ||
246 | if (node) { | ||
247 | g_source_remove_poll(&ctx->source, &node->pfd); | ||
248 | |||
249 | - /* aio_poll is in progress, just mark the node as deleted */ | ||
250 | - if (qemu_lockcnt_count(&ctx->list_lock)) { | ||
251 | - node->deleted = 1; | ||
252 | - node->pfd.revents = 0; | ||
253 | - } else { | ||
254 | - /* Otherwise, delete it for real. We can't just mark it as | ||
255 | - * deleted because deleted nodes are only cleaned up after | ||
256 | - * releasing the list_lock. | ||
257 | - */ | ||
258 | - QLIST_REMOVE(node, node); | ||
259 | - g_free(node); | ||
260 | - } | ||
261 | + aio_remove_fd_handler(ctx, node); | ||
262 | } | ||
263 | } else { | ||
264 | if (node == NULL) { | ||
265 | -- | 265 | -- |
266 | 2.20.1 | 266 | 2.20.1 |
267 | 267 | ||
268 | 268 | diff view generated by jsdifflib |
New patch | |||
---|---|---|---|
1 | From: Stefano Garzarella <sgarzare@redhat.com> | ||
1 | 2 | ||
3 | The size of data in the virtio_blk_request must be a multiple | ||
4 | of 512 bytes for IN and OUT requests, or a multiple of the size | ||
5 | of struct virtio_blk_discard_write_zeroes for DISCARD and | ||
6 | WRITE_ZEROES requests. | ||
7 | |||
8 | Reviewed-by: Michael S. Tsirkin <mst@redhat.com> | ||
9 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
10 | Reviewed-by: Thomas Huth <thuth@redhat.com> | ||
11 | Signed-off-by: Stefano Garzarella <sgarzare@redhat.com> | ||
12 | Acked-by: Pankaj Gupta <pagupta@redhat.com> | ||
13 | Message-id: 20190208134950.187665-6-sgarzare@redhat.com | ||
14 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> | ||
15 | --- | ||
16 | tests/virtio-blk-test.c | 15 ++++++++++++++- | ||
17 | 1 file changed, 14 insertions(+), 1 deletion(-) | ||
18 | |||
19 | diff --git a/tests/virtio-blk-test.c b/tests/virtio-blk-test.c | ||
20 | index XXXXXXX..XXXXXXX 100644 | ||
21 | --- a/tests/virtio-blk-test.c | ||
22 | +++ b/tests/virtio-blk-test.c | ||
23 | @@ -XXX,XX +XXX,XX @@ static uint64_t virtio_blk_request(QGuestAllocator *alloc, QVirtioDevice *d, | ||
24 | uint64_t addr; | ||
25 | uint8_t status = 0xFF; | ||
26 | |||
27 | - g_assert_cmpuint(data_size % 512, ==, 0); | ||
28 | + switch (req->type) { | ||
29 | + case VIRTIO_BLK_T_IN: | ||
30 | + case VIRTIO_BLK_T_OUT: | ||
31 | + g_assert_cmpuint(data_size % 512, ==, 0); | ||
32 | + break; | ||
33 | + case VIRTIO_BLK_T_DISCARD: | ||
34 | + case VIRTIO_BLK_T_WRITE_ZEROES: | ||
35 | + g_assert_cmpuint(data_size % | ||
36 | + sizeof(struct virtio_blk_discard_write_zeroes), ==, 0); | ||
37 | + break; | ||
38 | + default: | ||
39 | + g_assert_cmpuint(data_size, ==, 0); | ||
40 | + } | ||
41 | + | ||
42 | addr = guest_alloc(alloc, sizeof(*req) + data_size); | ||
43 | |||
44 | virtio_blk_fix_request(d, req); | ||
45 | -- | ||
46 | 2.20.1 | ||
47 | |||
48 | diff view generated by jsdifflib |
1 | From: Remy Noel <remy.noel@blade-group.com> | 1 | From: Stefano Garzarella <sgarzare@redhat.com> |
---|---|---|---|
2 | 2 | ||
3 | Cleaning the events will cause aio_epoll_update to unregister the fd. | 3 | If the WRITE_ZEROES feature is enabled, we check this command |
4 | Otherwise, the fd is kept registered until it is destroyed. | 4 | in the test_basic(). |
5 | 5 | ||
6 | Signed-off-by: Remy Noel <remy.noel@blade-group.com> | 6 | Reviewed-by: Michael S. Tsirkin <mst@redhat.com> |
7 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> | 7 | Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> |
8 | Reviewed-by: Paolo Bonzini <pbonzini@redhat.com> | 8 | Acked-by: Thomas Huth <thuth@redhat.com> |
9 | Message-id: 20181220152030.28035-2-remy.noel@blade-group.com | 9 | Signed-off-by: Stefano Garzarella <sgarzare@redhat.com> |
10 | Acked-by: Pankaj Gupta <pagupta@redhat.com> | ||
11 | Message-id: 20190208134950.187665-7-sgarzare@redhat.com | ||
10 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> | 12 | Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com> |
11 | --- | 13 | --- |
12 | util/aio-posix.c | 3 +++ | 14 | tests/virtio-blk-test.c | 60 +++++++++++++++++++++++++++++++++++++++++ |
13 | 1 file changed, 3 insertions(+) | 15 | 1 file changed, 60 insertions(+) |
14 | 16 | ||
15 | diff --git a/util/aio-posix.c b/util/aio-posix.c | 17 | diff --git a/tests/virtio-blk-test.c b/tests/virtio-blk-test.c |
16 | index XXXXXXX..XXXXXXX 100644 | 18 | index XXXXXXX..XXXXXXX 100644 |
17 | --- a/util/aio-posix.c | 19 | --- a/tests/virtio-blk-test.c |
18 | +++ b/util/aio-posix.c | 20 | +++ b/tests/virtio-blk-test.c |
19 | @@ -XXX,XX +XXX,XX @@ void aio_set_fd_handler(AioContext *ctx, | 21 | @@ -XXX,XX +XXX,XX @@ static void test_basic(QVirtioDevice *dev, QGuestAllocator *alloc, |
20 | QLIST_REMOVE(node, node); | 22 | |
21 | deleted = true; | 23 | guest_free(alloc, req_addr); |
22 | } | 24 | |
23 | + /* Clean events in order to unregister fd from the ctx epoll. */ | 25 | + if (features & (1u << VIRTIO_BLK_F_WRITE_ZEROES)) { |
24 | + node->pfd.events = 0; | 26 | + struct virtio_blk_discard_write_zeroes dwz_hdr; |
27 | + void *expected; | ||
25 | + | 28 | + |
26 | poll_disable_change = -!node->io_poll; | 29 | + /* |
27 | } else { | 30 | + * WRITE_ZEROES request on the same sector of previous test where |
28 | poll_disable_change = !io_poll - (node && !node->io_poll); | 31 | + * we wrote "TEST". |
32 | + */ | ||
33 | + req.type = VIRTIO_BLK_T_WRITE_ZEROES; | ||
34 | + req.data = (char *) &dwz_hdr; | ||
35 | + dwz_hdr.sector = 0; | ||
36 | + dwz_hdr.num_sectors = 1; | ||
37 | + dwz_hdr.flags = 0; | ||
38 | + | ||
39 | + req_addr = virtio_blk_request(alloc, dev, &req, sizeof(dwz_hdr)); | ||
40 | + | ||
41 | + free_head = qvirtqueue_add(vq, req_addr, 16, false, true); | ||
42 | + qvirtqueue_add(vq, req_addr + 16, sizeof(dwz_hdr), false, true); | ||
43 | + qvirtqueue_add(vq, req_addr + 16 + sizeof(dwz_hdr), 1, true, false); | ||
44 | + | ||
45 | + qvirtqueue_kick(dev, vq, free_head); | ||
46 | + | ||
47 | + qvirtio_wait_used_elem(dev, vq, free_head, NULL, | ||
48 | + QVIRTIO_BLK_TIMEOUT_US); | ||
49 | + status = readb(req_addr + 16 + sizeof(dwz_hdr)); | ||
50 | + g_assert_cmpint(status, ==, 0); | ||
51 | + | ||
52 | + guest_free(alloc, req_addr); | ||
53 | + | ||
54 | + /* Read request to check if the sector contains all zeroes */ | ||
55 | + req.type = VIRTIO_BLK_T_IN; | ||
56 | + req.ioprio = 1; | ||
57 | + req.sector = 0; | ||
58 | + req.data = g_malloc0(512); | ||
59 | + | ||
60 | + req_addr = virtio_blk_request(alloc, dev, &req, 512); | ||
61 | + | ||
62 | + g_free(req.data); | ||
63 | + | ||
64 | + free_head = qvirtqueue_add(vq, req_addr, 16, false, true); | ||
65 | + qvirtqueue_add(vq, req_addr + 16, 512, true, true); | ||
66 | + qvirtqueue_add(vq, req_addr + 528, 1, true, false); | ||
67 | + | ||
68 | + qvirtqueue_kick(dev, vq, free_head); | ||
69 | + | ||
70 | + qvirtio_wait_used_elem(dev, vq, free_head, NULL, | ||
71 | + QVIRTIO_BLK_TIMEOUT_US); | ||
72 | + status = readb(req_addr + 528); | ||
73 | + g_assert_cmpint(status, ==, 0); | ||
74 | + | ||
75 | + data = g_malloc(512); | ||
76 | + expected = g_malloc0(512); | ||
77 | + memread(req_addr + 16, data, 512); | ||
78 | + g_assert_cmpmem(data, 512, expected, 512); | ||
79 | + g_free(expected); | ||
80 | + g_free(data); | ||
81 | + | ||
82 | + guest_free(alloc, req_addr); | ||
83 | + } | ||
84 | + | ||
85 | if (features & (1u << VIRTIO_F_ANY_LAYOUT)) { | ||
86 | /* Write and read with 2 descriptor layout */ | ||
87 | /* Write request */ | ||
29 | -- | 88 | -- |
30 | 2.20.1 | 89 | 2.20.1 |
31 | 90 | ||
32 | 91 | diff view generated by jsdifflib |