* [PULL 01/22] block: Remove unused blk_op_is_blocked()
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-11 16:00 ` [PULL 02/22] block: Zero block driver state before reopening Kevin Wolf
` (21 subsequent siblings)
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
Commit fc4e394b28 removed the last caller of blk_op_is_blocked(). Remove
the now unused function.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250206165331.379033-1-kwolf@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
include/system/block-backend-global-state.h | 1 -
block/block-backend.c | 12 ------------
2 files changed, 13 deletions(-)
diff --git a/include/system/block-backend-global-state.h b/include/system/block-backend-global-state.h
index 9cc9b008ec..35b5e8380b 100644
--- a/include/system/block-backend-global-state.h
+++ b/include/system/block-backend-global-state.h
@@ -86,7 +86,6 @@ bool blk_supports_write_perm(BlockBackend *blk);
bool blk_is_sg(BlockBackend *blk);
void blk_set_enable_write_cache(BlockBackend *blk, bool wce);
int blk_get_flags(BlockBackend *blk);
-bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp);
int blk_set_aio_context(BlockBackend *blk, AioContext *new_context,
Error **errp);
void blk_add_aio_context_notifier(BlockBackend *blk,
diff --git a/block/block-backend.c b/block/block-backend.c
index 9288f7e1c6..a402db13f2 100644
--- a/block/block-backend.c
+++ b/block/block-backend.c
@@ -2357,18 +2357,6 @@ void *blk_blockalign(BlockBackend *blk, size_t size)
return qemu_blockalign(blk ? blk_bs(blk) : NULL, size);
}
-bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp)
-{
- BlockDriverState *bs = blk_bs(blk);
- GLOBAL_STATE_CODE();
- GRAPH_RDLOCK_GUARD_MAINLOOP();
-
- if (!bs) {
- return false;
- }
-
- return bdrv_op_is_blocked(bs, op, errp);
-}
/**
* Return BB's current AioContext. Note that this context may change
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* [PULL 02/22] block: Zero block driver state before reopening
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
2025-03-11 16:00 ` [PULL 01/22] block: Remove unused blk_op_is_blocked() Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-11 16:00 ` [PULL 03/22] file-posix: Support FUA writes Kevin Wolf
` (20 subsequent siblings)
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
Block drivers assume in their .bdrv_open() implementation that their
state in bs->opaque has been zeroed; it is initially allocated with
g_malloc0() in bdrv_open_driver().
bdrv_snapshot_goto() needs to make sure that it is zeroed again before
calling drv->bdrv_open() to avoid that block drivers use stale values.
One symptom of this bug is VMDK running into a double free when the user
tries to apply an internal snapshot like 'qemu-img snapshot -a test
test.vmdk'. This should be a graceful error because VMDK doesn't support
internal snapshots.
==25507== Invalid free() / delete / delete[] / realloc()
==25507== at 0x484B347: realloc (vg_replace_malloc.c:1801)
==25507== by 0x54B592A: g_realloc (gmem.c:171)
==25507== by 0x1B221D: vmdk_add_extent (../block/vmdk.c:570)
==25507== by 0x1B1084: vmdk_open_sparse (../block/vmdk.c:1059)
==25507== by 0x1AF3D8: vmdk_open (../block/vmdk.c:1371)
==25507== by 0x1A2AE0: bdrv_snapshot_goto (../block/snapshot.c:299)
==25507== by 0x205C77: img_snapshot (../qemu-img.c:3500)
==25507== by 0x58FA087: (below main) (libc_start_call_main.h:58)
==25507== Address 0x832f3e0 is 0 bytes inside a block of size 272 free'd
==25507== at 0x4846B83: free (vg_replace_malloc.c:989)
==25507== by 0x54AEAC4: g_free (gmem.c:208)
==25507== by 0x1AF629: vmdk_close (../block/vmdk.c:2889)
==25507== by 0x1A2A9C: bdrv_snapshot_goto (../block/snapshot.c:290)
==25507== by 0x205C77: img_snapshot (../qemu-img.c:3500)
==25507== by 0x58FA087: (below main) (libc_start_call_main.h:58)
This error was discovered by fuzzing qemu-img.
Cc: qemu-stable@nongnu.org
Closes: https://gitlab.com/qemu-project/qemu/-/issues/2853
Closes: https://gitlab.com/qemu-project/qemu/-/issues/2851
Reported-by: Denis Rastyogin <gerben@altlinux.org>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250310104858.28221-1-kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
block/snapshot.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/block/snapshot.c b/block/snapshot.c
index 9c44780e96..22567f1fb9 100644
--- a/block/snapshot.c
+++ b/block/snapshot.c
@@ -296,6 +296,7 @@ int bdrv_snapshot_goto(BlockDriverState *bs,
bdrv_graph_wrunlock();
ret = bdrv_snapshot_goto(fallback_bs, snapshot_id, errp);
+ memset(bs->opaque, 0, drv->instance_size);
open_ret = drv->bdrv_open(bs, options, bs->open_flags, &local_err);
qobject_unref(options);
if (open_ret < 0) {
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* [PULL 03/22] file-posix: Support FUA writes
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
2025-03-11 16:00 ` [PULL 01/22] block: Remove unused blk_op_is_blocked() Kevin Wolf
2025-03-11 16:00 ` [PULL 02/22] block: Zero block driver state before reopening Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-11 16:00 ` [PULL 04/22] block/io: Ignore FUA with cache.no-flush=on Kevin Wolf
` (19 subsequent siblings)
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
Until now, FUA was always emulated with a separate flush after the write
for file-posix. The overhead of processing a second request can reduce
performance significantly for a guest disk that has disabled the write
cache, especially if the host disk is already write through, too, and
the flush isn't actually doing anything.
Advertise support for REQ_FUA in write requests and implement it for
Linux AIO and io_uring using the RWF_DSYNC flag for write requests. The
thread pool still performs a separate fdatasync() call. This can be
improved later by using the pwritev2() syscall if available.
As an example, this is how fio numbers can be improved in some scenarios
with this patch (all using virtio-blk with cache=directsync on an nvme
block device for the VM, fio with ioengine=libaio,direct=1,sync=1):
| old | with FUA support
------------------------------+---------------+-------------------
bs=4k, iodepth=1, numjobs=1 | 45.6k iops | 56.1k iops
bs=4k, iodepth=1, numjobs=16 | 183.3k iops | 236.0k iops
bs=4k, iodepth=16, numjobs=1 | 258.4k iops | 311.1k iops
However, not all scenarios are clear wins. On another slower disk I saw
little to no improvment. In fact, in two corner case scenarios, I even
observed a regression, which I however consider acceptable:
1. On slow host disks in a write through cache mode, when the guest is
using virtio-blk in a separate iothread so that polling can be
enabled, and each completion is quickly followed up with a new
request (so that polling gets it), it can happen that enabling FUA
makes things slower - the additional very fast no-op flush we used to
have gave the adaptive polling algorithm a success so that it kept
polling. Without it, we only have the slow write request, which
disables polling. This is a problem in the polling algorithm that
will be fixed later in this series.
2. With a high queue depth, it can be beneficial to have flush requests
for another reason: The optimisation in bdrv_co_flush() that flushes
only once per write generation acts as a synchronisation mechanism
that lets all requests complete at the same time. This can result in
better batching and if the disk is very fast (I only saw this with a
null_blk backend), this can make up for the overhead of the flush and
improve throughput. In theory, we could optionally introduce a
similar artificial latency in the normal completion path to achieve
the same kind of completion batching. This is not implemented in this
series.
Compatibility is not a concern for io_uring, it has supported RWF_DSYNC
from the start. Linux AIO started supporting it in Linux 4.13 and libaio
0.3.111. The kernel is not a problem for any supported build platform,
so it's not necessary to add runtime checks. However, openSUSE is still
stuck with an older libaio version that would break the build. We must
detect this at build time to avoid build failures.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250307221634.71951-2-kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
include/block/raw-aio.h | 8 ++++++--
block/file-posix.c | 26 ++++++++++++++++++--------
block/io_uring.c | 13 ++++++++-----
block/linux-aio.c | 24 +++++++++++++++++++++---
meson.build | 4 ++++
5 files changed, 57 insertions(+), 18 deletions(-)
diff --git a/include/block/raw-aio.h b/include/block/raw-aio.h
index 626706827f..247bdbff13 100644
--- a/include/block/raw-aio.h
+++ b/include/block/raw-aio.h
@@ -17,6 +17,7 @@
#define QEMU_RAW_AIO_H
#include "block/aio.h"
+#include "block/block-common.h"
#include "qemu/iov.h"
/* AIO request types */
@@ -58,9 +59,11 @@ void laio_cleanup(LinuxAioState *s);
/* laio_co_submit: submit I/O requests in the thread's current AioContext. */
int coroutine_fn laio_co_submit(int fd, uint64_t offset, QEMUIOVector *qiov,
- int type, uint64_t dev_max_batch);
+ int type, BdrvRequestFlags flags,
+ uint64_t dev_max_batch);
bool laio_has_fdsync(int);
+bool laio_has_fua(void);
void laio_detach_aio_context(LinuxAioState *s, AioContext *old_context);
void laio_attach_aio_context(LinuxAioState *s, AioContext *new_context);
#endif
@@ -71,7 +74,8 @@ void luring_cleanup(LuringState *s);
/* luring_co_submit: submit I/O requests in the thread's current AioContext. */
int coroutine_fn luring_co_submit(BlockDriverState *bs, int fd, uint64_t offset,
- QEMUIOVector *qiov, int type);
+ QEMUIOVector *qiov, int type,
+ BdrvRequestFlags flags);
void luring_detach_aio_context(LuringState *s, AioContext *old_context);
void luring_attach_aio_context(LuringState *s, AioContext *new_context);
#endif
diff --git a/block/file-posix.c b/block/file-posix.c
index 44e16dda87..0f1c722804 100644
--- a/block/file-posix.c
+++ b/block/file-posix.c
@@ -194,6 +194,7 @@ static int fd_open(BlockDriverState *bs)
}
static int64_t raw_getlength(BlockDriverState *bs);
+static int coroutine_fn raw_co_flush_to_disk(BlockDriverState *bs);
typedef struct RawPosixAIOData {
BlockDriverState *bs;
@@ -804,6 +805,10 @@ static int raw_open_common(BlockDriverState *bs, QDict *options,
#endif
s->needs_alignment = raw_needs_alignment(bs);
+ if (!s->use_linux_aio || laio_has_fua()) {
+ bs->supported_write_flags = BDRV_REQ_FUA;
+ }
+
bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK;
if (S_ISREG(st.st_mode)) {
/* When extending regular files, we get zeros from the OS */
@@ -2477,7 +2482,8 @@ static inline bool raw_check_linux_aio(BDRVRawState *s)
#endif
static int coroutine_fn raw_co_prw(BlockDriverState *bs, int64_t *offset_ptr,
- uint64_t bytes, QEMUIOVector *qiov, int type)
+ uint64_t bytes, QEMUIOVector *qiov, int type,
+ int flags)
{
BDRVRawState *s = bs->opaque;
RawPosixAIOData acb;
@@ -2508,13 +2514,13 @@ static int coroutine_fn raw_co_prw(BlockDriverState *bs, int64_t *offset_ptr,
#ifdef CONFIG_LINUX_IO_URING
} else if (raw_check_linux_io_uring(s)) {
assert(qiov->size == bytes);
- ret = luring_co_submit(bs, s->fd, offset, qiov, type);
+ ret = luring_co_submit(bs, s->fd, offset, qiov, type, flags);
goto out;
#endif
#ifdef CONFIG_LINUX_AIO
} else if (raw_check_linux_aio(s)) {
assert(qiov->size == bytes);
- ret = laio_co_submit(s->fd, offset, qiov, type,
+ ret = laio_co_submit(s->fd, offset, qiov, type, flags,
s->aio_max_batch);
goto out;
#endif
@@ -2534,6 +2540,10 @@ static int coroutine_fn raw_co_prw(BlockDriverState *bs, int64_t *offset_ptr,
assert(qiov->size == bytes);
ret = raw_thread_pool_submit(handle_aiocb_rw, &acb);
+ if (ret == 0 && (flags & BDRV_REQ_FUA)) {
+ /* TODO Use pwritev2() instead if it's available */
+ ret = raw_co_flush_to_disk(bs);
+ }
goto out; /* Avoid the compiler err of unused label */
out:
@@ -2571,14 +2581,14 @@ static int coroutine_fn raw_co_preadv(BlockDriverState *bs, int64_t offset,
int64_t bytes, QEMUIOVector *qiov,
BdrvRequestFlags flags)
{
- return raw_co_prw(bs, &offset, bytes, qiov, QEMU_AIO_READ);
+ return raw_co_prw(bs, &offset, bytes, qiov, QEMU_AIO_READ, flags);
}
static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, int64_t offset,
int64_t bytes, QEMUIOVector *qiov,
BdrvRequestFlags flags)
{
- return raw_co_prw(bs, &offset, bytes, qiov, QEMU_AIO_WRITE);
+ return raw_co_prw(bs, &offset, bytes, qiov, QEMU_AIO_WRITE, flags);
}
static int coroutine_fn raw_co_flush_to_disk(BlockDriverState *bs)
@@ -2600,12 +2610,12 @@ static int coroutine_fn raw_co_flush_to_disk(BlockDriverState *bs)
#ifdef CONFIG_LINUX_IO_URING
if (raw_check_linux_io_uring(s)) {
- return luring_co_submit(bs, s->fd, 0, NULL, QEMU_AIO_FLUSH);
+ return luring_co_submit(bs, s->fd, 0, NULL, QEMU_AIO_FLUSH, 0);
}
#endif
#ifdef CONFIG_LINUX_AIO
if (s->has_laio_fdsync && raw_check_linux_aio(s)) {
- return laio_co_submit(s->fd, 0, NULL, QEMU_AIO_FLUSH, 0);
+ return laio_co_submit(s->fd, 0, NULL, QEMU_AIO_FLUSH, 0, 0);
}
#endif
return raw_thread_pool_submit(handle_aiocb_flush, &acb);
@@ -3540,7 +3550,7 @@ static int coroutine_fn raw_co_zone_append(BlockDriverState *bs,
}
trace_zbd_zone_append(bs, *offset >> BDRV_SECTOR_BITS);
- return raw_co_prw(bs, offset, len, qiov, QEMU_AIO_ZONE_APPEND);
+ return raw_co_prw(bs, offset, len, qiov, QEMU_AIO_ZONE_APPEND, 0);
}
#endif
diff --git a/block/io_uring.c b/block/io_uring.c
index f52b66b340..dc967dbf91 100644
--- a/block/io_uring.c
+++ b/block/io_uring.c
@@ -335,15 +335,17 @@ static void luring_deferred_fn(void *opaque)
*
*/
static int luring_do_submit(int fd, LuringAIOCB *luringcb, LuringState *s,
- uint64_t offset, int type)
+ uint64_t offset, int type, BdrvRequestFlags flags)
{
int ret;
struct io_uring_sqe *sqes = &luringcb->sqeq;
+ int luring_flags;
switch (type) {
case QEMU_AIO_WRITE:
- io_uring_prep_writev(sqes, fd, luringcb->qiov->iov,
- luringcb->qiov->niov, offset);
+ luring_flags = (flags & BDRV_REQ_FUA) ? RWF_DSYNC : 0;
+ io_uring_prep_writev2(sqes, fd, luringcb->qiov->iov,
+ luringcb->qiov->niov, offset, luring_flags);
break;
case QEMU_AIO_ZONE_APPEND:
io_uring_prep_writev(sqes, fd, luringcb->qiov->iov,
@@ -380,7 +382,8 @@ static int luring_do_submit(int fd, LuringAIOCB *luringcb, LuringState *s,
}
int coroutine_fn luring_co_submit(BlockDriverState *bs, int fd, uint64_t offset,
- QEMUIOVector *qiov, int type)
+ QEMUIOVector *qiov, int type,
+ BdrvRequestFlags flags)
{
int ret;
AioContext *ctx = qemu_get_current_aio_context();
@@ -393,7 +396,7 @@ int coroutine_fn luring_co_submit(BlockDriverState *bs, int fd, uint64_t offset,
};
trace_luring_co_submit(bs, s, &luringcb, fd, offset, qiov ? qiov->size : 0,
type);
- ret = luring_do_submit(fd, &luringcb, s, offset, type);
+ ret = luring_do_submit(fd, &luringcb, s, offset, type, flags);
if (ret < 0) {
return ret;
diff --git a/block/linux-aio.c b/block/linux-aio.c
index 194c8f434f..1108ae361c 100644
--- a/block/linux-aio.c
+++ b/block/linux-aio.c
@@ -368,15 +368,23 @@ static void laio_deferred_fn(void *opaque)
}
static int laio_do_submit(int fd, struct qemu_laiocb *laiocb, off_t offset,
- int type, uint64_t dev_max_batch)
+ int type, BdrvRequestFlags flags,
+ uint64_t dev_max_batch)
{
LinuxAioState *s = laiocb->ctx;
struct iocb *iocbs = &laiocb->iocb;
QEMUIOVector *qiov = laiocb->qiov;
+ int laio_flags;
switch (type) {
case QEMU_AIO_WRITE:
+#ifdef HAVE_IO_PREP_PWRITEV2
+ laio_flags = (flags & BDRV_REQ_FUA) ? RWF_DSYNC : 0;
+ io_prep_pwritev2(iocbs, fd, qiov->iov, qiov->niov, offset, laio_flags);
+#else
+ assert(flags == 0);
io_prep_pwritev(iocbs, fd, qiov->iov, qiov->niov, offset);
+#endif
break;
case QEMU_AIO_ZONE_APPEND:
io_prep_pwritev(iocbs, fd, qiov->iov, qiov->niov, offset);
@@ -409,7 +417,8 @@ static int laio_do_submit(int fd, struct qemu_laiocb *laiocb, off_t offset,
}
int coroutine_fn laio_co_submit(int fd, uint64_t offset, QEMUIOVector *qiov,
- int type, uint64_t dev_max_batch)
+ int type, BdrvRequestFlags flags,
+ uint64_t dev_max_batch)
{
int ret;
AioContext *ctx = qemu_get_current_aio_context();
@@ -422,7 +431,7 @@ int coroutine_fn laio_co_submit(int fd, uint64_t offset, QEMUIOVector *qiov,
.qiov = qiov,
};
- ret = laio_do_submit(fd, &laiocb, offset, type, dev_max_batch);
+ ret = laio_do_submit(fd, &laiocb, offset, type, flags, dev_max_batch);
if (ret < 0) {
return ret;
}
@@ -505,3 +514,12 @@ bool laio_has_fdsync(int fd)
io_destroy(ctx);
return (ret == -EINVAL) ? false : true;
}
+
+bool laio_has_fua(void)
+{
+#ifdef HAVE_IO_PREP_PWRITEV2
+ return true;
+#else
+ return false;
+#endif
+}
diff --git a/meson.build b/meson.build
index 9d9c11731f..3dad23794e 100644
--- a/meson.build
+++ b/meson.build
@@ -2727,6 +2727,10 @@ config_host_data.set('HAVE_OPTRESET',
cc.has_header_symbol('getopt.h', 'optreset'))
config_host_data.set('HAVE_IPPROTO_MPTCP',
cc.has_header_symbol('netinet/in.h', 'IPPROTO_MPTCP'))
+if libaio.found()
+ config_host_data.set('HAVE_IO_PREP_PWRITEV2',
+ cc.has_header_symbol('libaio.h', 'io_prep_pwritev2'))
+endif
# has_member
config_host_data.set('HAVE_SIGEV_NOTIFY_THREAD_ID',
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* [PULL 04/22] block/io: Ignore FUA with cache.no-flush=on
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
` (2 preceding siblings ...)
2025-03-11 16:00 ` [PULL 03/22] file-posix: Support FUA writes Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-11 16:00 ` [PULL 05/22] aio: Create AioPolledEvent Kevin Wolf
` (18 subsequent siblings)
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
For block drivers that don't advertise FUA support, we already call
bdrv_co_flush(), which considers BDRV_O_NO_FLUSH. However, drivers that
do support FUA still see the FUA flag with BDRV_O_NO_FLUSH and get the
associated performance penalty that cache.no-flush=on was supposed to
avoid.
Clear FUA for write requests if BDRV_O_NO_FLUSH is set.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250307221634.71951-3-kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
block/io.c | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/block/io.c b/block/io.c
index d369b994df..1ba8d1aeea 100644
--- a/block/io.c
+++ b/block/io.c
@@ -1058,6 +1058,10 @@ bdrv_driver_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
return -ENOMEDIUM;
}
+ if (bs->open_flags & BDRV_O_NO_FLUSH) {
+ flags &= ~BDRV_REQ_FUA;
+ }
+
if ((flags & BDRV_REQ_FUA) &&
(~bs->supported_write_flags & BDRV_REQ_FUA)) {
flags &= ~BDRV_REQ_FUA;
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* [PULL 05/22] aio: Create AioPolledEvent
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
` (3 preceding siblings ...)
2025-03-11 16:00 ` [PULL 04/22] block/io: Ignore FUA with cache.no-flush=on Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-11 16:00 ` [PULL 06/22] aio-posix: Factor out adjust_polling_time() Kevin Wolf
` (17 subsequent siblings)
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
As a preparation for having multiple adaptive polling states per
AioContext, move the 'ns' field into a separate struct.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250307221634.71951-4-kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
include/block/aio.h | 6 +++++-
util/aio-posix.c | 31 ++++++++++++++++---------------
util/async.c | 3 ++-
3 files changed, 23 insertions(+), 17 deletions(-)
diff --git a/include/block/aio.h b/include/block/aio.h
index b2ab3514de..c9fcfe7ccf 100644
--- a/include/block/aio.h
+++ b/include/block/aio.h
@@ -123,6 +123,10 @@ struct BHListSlice {
typedef QSLIST_HEAD(, AioHandler) AioHandlerSList;
+typedef struct AioPolledEvent {
+ int64_t ns; /* current polling time in nanoseconds */
+} AioPolledEvent;
+
struct AioContext {
GSource source;
@@ -229,7 +233,7 @@ struct AioContext {
int poll_disable_cnt;
/* Polling mode parameters */
- int64_t poll_ns; /* current polling time in nanoseconds */
+ AioPolledEvent poll;
int64_t poll_max_ns; /* maximum polling time in nanoseconds */
int64_t poll_grow; /* polling time growth factor */
int64_t poll_shrink; /* polling time shrink factor */
diff --git a/util/aio-posix.c b/util/aio-posix.c
index 06bf9f456c..95bddb9e4b 100644
--- a/util/aio-posix.c
+++ b/util/aio-posix.c
@@ -585,7 +585,7 @@ static bool try_poll_mode(AioContext *ctx, AioHandlerList *ready_list,
return false;
}
- max_ns = qemu_soonest_timeout(*timeout, ctx->poll_ns);
+ max_ns = qemu_soonest_timeout(*timeout, ctx->poll.ns);
if (max_ns && !ctx->fdmon_ops->need_wait(ctx)) {
/*
* Enable poll mode. It pairs with the poll_set_started() in
@@ -683,40 +683,40 @@ bool aio_poll(AioContext *ctx, bool blocking)
if (ctx->poll_max_ns) {
int64_t block_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - start;
- if (block_ns <= ctx->poll_ns) {
+ if (block_ns <= ctx->poll.ns) {
/* This is the sweet spot, no adjustment needed */
} else if (block_ns > ctx->poll_max_ns) {
/* We'd have to poll for too long, poll less */
- int64_t old = ctx->poll_ns;
+ int64_t old = ctx->poll.ns;
if (ctx->poll_shrink) {
- ctx->poll_ns /= ctx->poll_shrink;
+ ctx->poll.ns /= ctx->poll_shrink;
} else {
- ctx->poll_ns = 0;
+ ctx->poll.ns = 0;
}
- trace_poll_shrink(ctx, old, ctx->poll_ns);
- } else if (ctx->poll_ns < ctx->poll_max_ns &&
+ trace_poll_shrink(ctx, old, ctx->poll.ns);
+ } else if (ctx->poll.ns < ctx->poll_max_ns &&
block_ns < ctx->poll_max_ns) {
/* There is room to grow, poll longer */
- int64_t old = ctx->poll_ns;
+ int64_t old = ctx->poll.ns;
int64_t grow = ctx->poll_grow;
if (grow == 0) {
grow = 2;
}
- if (ctx->poll_ns) {
- ctx->poll_ns *= grow;
+ if (ctx->poll.ns) {
+ ctx->poll.ns *= grow;
} else {
- ctx->poll_ns = 4000; /* start polling at 4 microseconds */
+ ctx->poll.ns = 4000; /* start polling at 4 microseconds */
}
- if (ctx->poll_ns > ctx->poll_max_ns) {
- ctx->poll_ns = ctx->poll_max_ns;
+ if (ctx->poll.ns > ctx->poll_max_ns) {
+ ctx->poll.ns = ctx->poll_max_ns;
}
- trace_poll_grow(ctx, old, ctx->poll_ns);
+ trace_poll_grow(ctx, old, ctx->poll.ns);
}
}
@@ -770,8 +770,9 @@ void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns,
/* No thread synchronization here, it doesn't matter if an incorrect value
* is used once.
*/
+ ctx->poll.ns = 0;
+
ctx->poll_max_ns = max_ns;
- ctx->poll_ns = 0;
ctx->poll_grow = grow;
ctx->poll_shrink = shrink;
diff --git a/util/async.c b/util/async.c
index 47e3d35a26..fc8a78aa79 100644
--- a/util/async.c
+++ b/util/async.c
@@ -609,7 +609,8 @@ AioContext *aio_context_new(Error **errp)
qemu_rec_mutex_init(&ctx->lock);
timerlistgroup_init(&ctx->tlg, aio_timerlist_notify, ctx);
- ctx->poll_ns = 0;
+ ctx->poll.ns = 0;
+
ctx->poll_max_ns = 0;
ctx->poll_grow = 0;
ctx->poll_shrink = 0;
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* [PULL 06/22] aio-posix: Factor out adjust_polling_time()
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
` (4 preceding siblings ...)
2025-03-11 16:00 ` [PULL 05/22] aio: Create AioPolledEvent Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-11 16:00 ` [PULL 07/22] aio-posix: Separate AioPolledEvent per AioHandler Kevin Wolf
` (16 subsequent siblings)
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250307221634.71951-5-kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
util/aio-posix.c | 77 ++++++++++++++++++++++++++----------------------
1 file changed, 41 insertions(+), 36 deletions(-)
diff --git a/util/aio-posix.c b/util/aio-posix.c
index 95bddb9e4b..259827c7ad 100644
--- a/util/aio-posix.c
+++ b/util/aio-posix.c
@@ -600,6 +600,46 @@ static bool try_poll_mode(AioContext *ctx, AioHandlerList *ready_list,
return false;
}
+static void adjust_polling_time(AioContext *ctx, AioPolledEvent *poll,
+ int64_t block_ns)
+{
+ if (block_ns <= poll->ns) {
+ /* This is the sweet spot, no adjustment needed */
+ } else if (block_ns > ctx->poll_max_ns) {
+ /* We'd have to poll for too long, poll less */
+ int64_t old = poll->ns;
+
+ if (ctx->poll_shrink) {
+ poll->ns /= ctx->poll_shrink;
+ } else {
+ poll->ns = 0;
+ }
+
+ trace_poll_shrink(ctx, old, poll->ns);
+ } else if (poll->ns < ctx->poll_max_ns &&
+ block_ns < ctx->poll_max_ns) {
+ /* There is room to grow, poll longer */
+ int64_t old = poll->ns;
+ int64_t grow = ctx->poll_grow;
+
+ if (grow == 0) {
+ grow = 2;
+ }
+
+ if (poll->ns) {
+ poll->ns *= grow;
+ } else {
+ poll->ns = 4000; /* start polling at 4 microseconds */
+ }
+
+ if (poll->ns > ctx->poll_max_ns) {
+ poll->ns = ctx->poll_max_ns;
+ }
+
+ trace_poll_grow(ctx, old, poll->ns);
+ }
+}
+
bool aio_poll(AioContext *ctx, bool blocking)
{
AioHandlerList ready_list = QLIST_HEAD_INITIALIZER(ready_list);
@@ -682,42 +722,7 @@ bool aio_poll(AioContext *ctx, bool blocking)
/* Adjust polling time */
if (ctx->poll_max_ns) {
int64_t block_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - start;
-
- if (block_ns <= ctx->poll.ns) {
- /* This is the sweet spot, no adjustment needed */
- } else if (block_ns > ctx->poll_max_ns) {
- /* We'd have to poll for too long, poll less */
- int64_t old = ctx->poll.ns;
-
- if (ctx->poll_shrink) {
- ctx->poll.ns /= ctx->poll_shrink;
- } else {
- ctx->poll.ns = 0;
- }
-
- trace_poll_shrink(ctx, old, ctx->poll.ns);
- } else if (ctx->poll.ns < ctx->poll_max_ns &&
- block_ns < ctx->poll_max_ns) {
- /* There is room to grow, poll longer */
- int64_t old = ctx->poll.ns;
- int64_t grow = ctx->poll_grow;
-
- if (grow == 0) {
- grow = 2;
- }
-
- if (ctx->poll.ns) {
- ctx->poll.ns *= grow;
- } else {
- ctx->poll.ns = 4000; /* start polling at 4 microseconds */
- }
-
- if (ctx->poll.ns > ctx->poll_max_ns) {
- ctx->poll.ns = ctx->poll_max_ns;
- }
-
- trace_poll_grow(ctx, old, ctx->poll.ns);
- }
+ adjust_polling_time(ctx, &ctx->poll, block_ns);
}
progress |= aio_bh_poll(ctx);
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* [PULL 07/22] aio-posix: Separate AioPolledEvent per AioHandler
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
` (5 preceding siblings ...)
2025-03-11 16:00 ` [PULL 06/22] aio-posix: Factor out adjust_polling_time() Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-11 16:00 ` [PULL 08/22] aio-posix: Adjust polling time also for new handlers Kevin Wolf
` (15 subsequent siblings)
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
Adaptive polling has a big problem: It doesn't consider that an event
loop can wait for many different events that may have very different
typical latencies.
For example, think of a guest that tends to send a new I/O request soon
after the previous I/O request completes, but the storage on the host is
rather slow. In this case, getting the new request from guest quickly
means that polling is enabled, but the next thing is performing the I/O
request on the backend, which is slow and disables polling again for the
next guest request. This means that in such a scenario, polling could
help for every other event, but is only ever enabled when it can't
succeed.
In order to fix this, keep a separate AioPolledEvent for each
AioHandler. We will then know that the backend file descriptor always
has a high latency and isn't worth polling for, but we also know that
the guest is always fast and we should poll for it. This solves at least
half of the problem, we can now keep polling for those cases where it
makes sense and get the improved performance from it.
Since the event loop doesn't know which event will be next, we still do
some unnecessary polling while we're waiting for the slow disk. I made
some attempts to be more clever than just randomly growing and shrinking
the polling time, and even to let callers be explicit about when they
expect a new event, but so far this hasn't resulted in improved
performance or even caused performance regressions. For now, let's just
fix the part that is easy enough to fix, we can revisit the rest later.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250307221634.71951-6-kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
include/block/aio.h | 1 -
util/aio-posix.h | 1 +
util/aio-posix.c | 26 ++++++++++++++++++++++----
util/async.c | 2 --
4 files changed, 23 insertions(+), 7 deletions(-)
diff --git a/include/block/aio.h b/include/block/aio.h
index c9fcfe7ccf..99ff48420b 100644
--- a/include/block/aio.h
+++ b/include/block/aio.h
@@ -233,7 +233,6 @@ struct AioContext {
int poll_disable_cnt;
/* Polling mode parameters */
- AioPolledEvent poll;
int64_t poll_max_ns; /* maximum polling time in nanoseconds */
int64_t poll_grow; /* polling time growth factor */
int64_t poll_shrink; /* polling time shrink factor */
diff --git a/util/aio-posix.h b/util/aio-posix.h
index 4264c518be..82a0201ea4 100644
--- a/util/aio-posix.h
+++ b/util/aio-posix.h
@@ -38,6 +38,7 @@ struct AioHandler {
#endif
int64_t poll_idle_timeout; /* when to stop userspace polling */
bool poll_ready; /* has polling detected an event? */
+ AioPolledEvent poll;
};
/* Add a handler to a ready list */
diff --git a/util/aio-posix.c b/util/aio-posix.c
index 259827c7ad..80785c29d2 100644
--- a/util/aio-posix.c
+++ b/util/aio-posix.c
@@ -579,13 +579,19 @@ static bool run_poll_handlers(AioContext *ctx, AioHandlerList *ready_list,
static bool try_poll_mode(AioContext *ctx, AioHandlerList *ready_list,
int64_t *timeout)
{
+ AioHandler *node;
int64_t max_ns;
if (QLIST_EMPTY_RCU(&ctx->poll_aio_handlers)) {
return false;
}
- max_ns = qemu_soonest_timeout(*timeout, ctx->poll.ns);
+ max_ns = 0;
+ QLIST_FOREACH(node, &ctx->poll_aio_handlers, node_poll) {
+ max_ns = MAX(max_ns, node->poll.ns);
+ }
+ max_ns = qemu_soonest_timeout(*timeout, max_ns);
+
if (max_ns && !ctx->fdmon_ops->need_wait(ctx)) {
/*
* Enable poll mode. It pairs with the poll_set_started() in
@@ -721,8 +727,14 @@ bool aio_poll(AioContext *ctx, bool blocking)
/* Adjust polling time */
if (ctx->poll_max_ns) {
+ AioHandler *node;
int64_t block_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - start;
- adjust_polling_time(ctx, &ctx->poll, block_ns);
+
+ QLIST_FOREACH(node, &ctx->poll_aio_handlers, node_poll) {
+ if (QLIST_IS_INSERTED(node, node_ready)) {
+ adjust_polling_time(ctx, &node->poll, block_ns);
+ }
+ }
}
progress |= aio_bh_poll(ctx);
@@ -772,11 +784,17 @@ void aio_context_use_g_source(AioContext *ctx)
void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns,
int64_t grow, int64_t shrink, Error **errp)
{
+ AioHandler *node;
+
+ qemu_lockcnt_inc(&ctx->list_lock);
+ QLIST_FOREACH(node, &ctx->aio_handlers, node) {
+ node->poll.ns = 0;
+ }
+ qemu_lockcnt_dec(&ctx->list_lock);
+
/* No thread synchronization here, it doesn't matter if an incorrect value
* is used once.
*/
- ctx->poll.ns = 0;
-
ctx->poll_max_ns = max_ns;
ctx->poll_grow = grow;
ctx->poll_shrink = shrink;
diff --git a/util/async.c b/util/async.c
index fc8a78aa79..863416dee9 100644
--- a/util/async.c
+++ b/util/async.c
@@ -609,8 +609,6 @@ AioContext *aio_context_new(Error **errp)
qemu_rec_mutex_init(&ctx->lock);
timerlistgroup_init(&ctx->tlg, aio_timerlist_notify, ctx);
- ctx->poll.ns = 0;
-
ctx->poll_max_ns = 0;
ctx->poll_grow = 0;
ctx->poll_shrink = 0;
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* [PULL 08/22] aio-posix: Adjust polling time also for new handlers
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
` (6 preceding siblings ...)
2025-03-11 16:00 ` [PULL 07/22] aio-posix: Separate AioPolledEvent per AioHandler Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-11 16:00 ` [PULL 09/22] iotests: Limit qsd-migrate to working formats Kevin Wolf
` (14 subsequent siblings)
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
aio_dispatch_handler() adds handlers to ctx->poll_aio_handlers if
polling should be enabled. If we call adjust_polling_time() for all
polling handlers before this, new polling handlers are still left at
poll->ns = 0 and polling is only actually enabled after the next event.
Move the adjust_polling_time() call after aio_dispatch_handler().
This fixes test-nested-aio-poll, which expects that polling becomes
effective the first time around.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250311141912.135657-1-kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
util/aio-posix.c | 28 +++++++++++++++++-----------
1 file changed, 17 insertions(+), 11 deletions(-)
diff --git a/util/aio-posix.c b/util/aio-posix.c
index 80785c29d2..2e0a5dadc4 100644
--- a/util/aio-posix.c
+++ b/util/aio-posix.c
@@ -28,6 +28,9 @@
/* Stop userspace polling on a handler if it isn't active for some time */
#define POLL_IDLE_INTERVAL_NS (7 * NANOSECONDS_PER_SECOND)
+static void adjust_polling_time(AioContext *ctx, AioPolledEvent *poll,
+ int64_t block_ns);
+
bool aio_poll_disabled(AioContext *ctx)
{
return qatomic_read(&ctx->poll_disable_cnt);
@@ -392,7 +395,8 @@ static bool aio_dispatch_handler(AioContext *ctx, AioHandler *node)
* scanning all handlers with aio_dispatch_handlers().
*/
static bool aio_dispatch_ready_handlers(AioContext *ctx,
- AioHandlerList *ready_list)
+ AioHandlerList *ready_list,
+ int64_t block_ns)
{
bool progress = false;
AioHandler *node;
@@ -400,6 +404,14 @@ static bool aio_dispatch_ready_handlers(AioContext *ctx,
while ((node = QLIST_FIRST(ready_list))) {
QLIST_REMOVE(node, node_ready);
progress = aio_dispatch_handler(ctx, node) || progress;
+
+ /*
+ * Adjust polling time only after aio_dispatch_handler(), which can
+ * add the handler to ctx->poll_aio_handlers.
+ */
+ if (ctx->poll_max_ns && QLIST_IS_INSERTED(node, node_poll)) {
+ adjust_polling_time(ctx, &node->poll, block_ns);
+ }
}
return progress;
@@ -653,6 +665,7 @@ bool aio_poll(AioContext *ctx, bool blocking)
bool use_notify_me;
int64_t timeout;
int64_t start = 0;
+ int64_t block_ns = 0;
/*
* There cannot be two concurrent aio_poll calls for the same AioContext (or
@@ -725,20 +738,13 @@ bool aio_poll(AioContext *ctx, bool blocking)
aio_notify_accept(ctx);
- /* Adjust polling time */
+ /* Calculate blocked time for adaptive polling */
if (ctx->poll_max_ns) {
- AioHandler *node;
- int64_t block_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - start;
-
- QLIST_FOREACH(node, &ctx->poll_aio_handlers, node_poll) {
- if (QLIST_IS_INSERTED(node, node_ready)) {
- adjust_polling_time(ctx, &node->poll, block_ns);
- }
- }
+ block_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - start;
}
progress |= aio_bh_poll(ctx);
- progress |= aio_dispatch_ready_handlers(ctx, &ready_list);
+ progress |= aio_dispatch_ready_handlers(ctx, &ready_list, block_ns);
aio_free_deleted_handlers(ctx);
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* [PULL 09/22] iotests: Limit qsd-migrate to working formats
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
` (7 preceding siblings ...)
2025-03-11 16:00 ` [PULL 08/22] aio-posix: Adjust polling time also for new handlers Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-11 16:00 ` [PULL 10/22] scsi-disk: drop unused SCSIDiskState->bh field Kevin Wolf
` (13 subsequent siblings)
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
From: Thomas Huth <thuth@redhat.com>
qsd-migrate is currently only working for raw, qcow2 and qed.
Other formats are failing, e.g. because they don't support migration.
Thus let's limit this test to the three usable formats now.
Suggested-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-ID: <20250224214058.205889-1-thuth@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
tests/qemu-iotests/tests/qsd-migrate | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/tests/qemu-iotests/tests/qsd-migrate b/tests/qemu-iotests/tests/qsd-migrate
index de17562cb0..a4c6592420 100755
--- a/tests/qemu-iotests/tests/qsd-migrate
+++ b/tests/qemu-iotests/tests/qsd-migrate
@@ -22,7 +22,7 @@ import iotests
from iotests import filter_qemu_io, filter_qtest
-iotests.script_initialize(supported_fmts=['generic'],
+iotests.script_initialize(supported_fmts=['qcow2', 'qed', 'raw'],
supported_protocols=['file'],
supported_platforms=['linux'])
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* [PULL 10/22] scsi-disk: drop unused SCSIDiskState->bh field
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
` (8 preceding siblings ...)
2025-03-11 16:00 ` [PULL 09/22] iotests: Limit qsd-migrate to working formats Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-11 16:00 ` [PULL 11/22] dma: use current AioContext for dma_blk_io() Kevin Wolf
` (12 subsequent siblings)
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
From: Stefan Hajnoczi <stefanha@redhat.com>
Commit 71544d30a6f8 ("scsi: push request restart to SCSIDevice") removed
the only user of SCSIDiskState->bh.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250311132616.1049687-2-stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
hw/scsi/scsi-disk.c | 1 -
1 file changed, 1 deletion(-)
diff --git a/hw/scsi/scsi-disk.c b/hw/scsi/scsi-disk.c
index e7f738b484..caf6c1437f 100644
--- a/hw/scsi/scsi-disk.c
+++ b/hw/scsi/scsi-disk.c
@@ -106,7 +106,6 @@ struct SCSIDiskState {
uint64_t max_unmap_size;
uint64_t max_io_size;
uint32_t quirks;
- QEMUBH *bh;
char *version;
char *serial;
char *vendor;
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* [PULL 11/22] dma: use current AioContext for dma_blk_io()
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
` (9 preceding siblings ...)
2025-03-11 16:00 ` [PULL 10/22] scsi-disk: drop unused SCSIDiskState->bh field Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-11 16:00 ` [PULL 12/22] scsi: track per-SCSIRequest AioContext Kevin Wolf
` (11 subsequent siblings)
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
From: Stefan Hajnoczi <stefanha@redhat.com>
In the past a single AioContext was used for block I/O and it was
fetched using blk_get_aio_context(). Nowadays the block layer supports
running I/O from any AioContext and multiple AioContexts at the same
time. Remove the dma_blk_io() AioContext argument and use the current
AioContext instead.
This makes calling the function easier and enables multiple IOThreads to
use dma_blk_io() concurrently for the same block device.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250311132616.1049687-3-stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
include/system/dma.h | 3 +--
hw/ide/core.c | 3 +--
hw/ide/macio.c | 3 +--
hw/scsi/scsi-disk.c | 6 ++----
system/dma-helpers.c | 8 ++++----
5 files changed, 9 insertions(+), 14 deletions(-)
diff --git a/include/system/dma.h b/include/system/dma.h
index 5a49a30628..e142f7efa6 100644
--- a/include/system/dma.h
+++ b/include/system/dma.h
@@ -290,8 +290,7 @@ typedef BlockAIOCB *DMAIOFunc(int64_t offset, QEMUIOVector *iov,
BlockCompletionFunc *cb, void *cb_opaque,
void *opaque);
-BlockAIOCB *dma_blk_io(AioContext *ctx,
- QEMUSGList *sg, uint64_t offset, uint32_t align,
+BlockAIOCB *dma_blk_io(QEMUSGList *sg, uint64_t offset, uint32_t align,
DMAIOFunc *io_func, void *io_func_opaque,
BlockCompletionFunc *cb, void *opaque, DMADirection dir);
BlockAIOCB *dma_blk_read(BlockBackend *blk,
diff --git a/hw/ide/core.c b/hw/ide/core.c
index f9baba59e9..b14983ec54 100644
--- a/hw/ide/core.c
+++ b/hw/ide/core.c
@@ -968,8 +968,7 @@ static void ide_dma_cb(void *opaque, int ret)
BDRV_SECTOR_SIZE, ide_dma_cb, s);
break;
case IDE_DMA_TRIM:
- s->bus->dma->aiocb = dma_blk_io(blk_get_aio_context(s->blk),
- &s->sg, offset, BDRV_SECTOR_SIZE,
+ s->bus->dma->aiocb = dma_blk_io(&s->sg, offset, BDRV_SECTOR_SIZE,
ide_issue_trim, s, ide_dma_cb, s,
DMA_DIRECTION_TO_DEVICE);
break;
diff --git a/hw/ide/macio.c b/hw/ide/macio.c
index 5fe764b49b..c8e8e44cc9 100644
--- a/hw/ide/macio.c
+++ b/hw/ide/macio.c
@@ -187,8 +187,7 @@ static void pmac_ide_transfer_cb(void *opaque, int ret)
pmac_ide_transfer_cb, io);
break;
case IDE_DMA_TRIM:
- s->bus->dma->aiocb = dma_blk_io(blk_get_aio_context(s->blk), &s->sg,
- offset, 0x1, ide_issue_trim, s,
+ s->bus->dma->aiocb = dma_blk_io(&s->sg, offset, 0x1, ide_issue_trim, s,
pmac_ide_transfer_cb, io,
DMA_DIRECTION_TO_DEVICE);
break;
diff --git a/hw/scsi/scsi-disk.c b/hw/scsi/scsi-disk.c
index caf6c1437f..f049a20275 100644
--- a/hw/scsi/scsi-disk.c
+++ b/hw/scsi/scsi-disk.c
@@ -487,8 +487,7 @@ static void scsi_do_read(SCSIDiskReq *r, int ret)
if (r->req.sg) {
dma_acct_start(s->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_READ);
r->req.residual -= r->req.sg->size;
- r->req.aiocb = dma_blk_io(blk_get_aio_context(s->qdev.conf.blk),
- r->req.sg, r->sector << BDRV_SECTOR_BITS,
+ r->req.aiocb = dma_blk_io(r->req.sg, r->sector << BDRV_SECTOR_BITS,
BDRV_SECTOR_SIZE,
sdc->dma_readv, r, scsi_dma_complete, r,
DMA_DIRECTION_FROM_DEVICE);
@@ -650,8 +649,7 @@ static void scsi_write_data(SCSIRequest *req)
if (r->req.sg) {
dma_acct_start(s->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_WRITE);
r->req.residual -= r->req.sg->size;
- r->req.aiocb = dma_blk_io(blk_get_aio_context(s->qdev.conf.blk),
- r->req.sg, r->sector << BDRV_SECTOR_BITS,
+ r->req.aiocb = dma_blk_io(r->req.sg, r->sector << BDRV_SECTOR_BITS,
BDRV_SECTOR_SIZE,
sdc->dma_writev, r, scsi_dma_complete, r,
DMA_DIRECTION_TO_DEVICE);
diff --git a/system/dma-helpers.c b/system/dma-helpers.c
index f6403242f5..6bad75876f 100644
--- a/system/dma-helpers.c
+++ b/system/dma-helpers.c
@@ -211,7 +211,7 @@ static const AIOCBInfo dma_aiocb_info = {
.cancel_async = dma_aio_cancel,
};
-BlockAIOCB *dma_blk_io(AioContext *ctx,
+BlockAIOCB *dma_blk_io(
QEMUSGList *sg, uint64_t offset, uint32_t align,
DMAIOFunc *io_func, void *io_func_opaque,
BlockCompletionFunc *cb,
@@ -223,7 +223,7 @@ BlockAIOCB *dma_blk_io(AioContext *ctx,
dbs->acb = NULL;
dbs->sg = sg;
- dbs->ctx = ctx;
+ dbs->ctx = qemu_get_current_aio_context();
dbs->offset = offset;
dbs->align = align;
dbs->sg_cur_index = 0;
@@ -251,7 +251,7 @@ BlockAIOCB *dma_blk_read(BlockBackend *blk,
QEMUSGList *sg, uint64_t offset, uint32_t align,
void (*cb)(void *opaque, int ret), void *opaque)
{
- return dma_blk_io(blk_get_aio_context(blk), sg, offset, align,
+ return dma_blk_io(sg, offset, align,
dma_blk_read_io_func, blk, cb, opaque,
DMA_DIRECTION_FROM_DEVICE);
}
@@ -269,7 +269,7 @@ BlockAIOCB *dma_blk_write(BlockBackend *blk,
QEMUSGList *sg, uint64_t offset, uint32_t align,
void (*cb)(void *opaque, int ret), void *opaque)
{
- return dma_blk_io(blk_get_aio_context(blk), sg, offset, align,
+ return dma_blk_io(sg, offset, align,
dma_blk_write_io_func, blk, cb, opaque,
DMA_DIRECTION_TO_DEVICE);
}
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* [PULL 12/22] scsi: track per-SCSIRequest AioContext
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
` (10 preceding siblings ...)
2025-03-11 16:00 ` [PULL 11/22] dma: use current AioContext for dma_blk_io() Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-11 16:00 ` [PULL 13/22] scsi: introduce requests_lock Kevin Wolf
` (10 subsequent siblings)
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
From: Stefan Hajnoczi <stefanha@redhat.com>
Until now, a SCSIDevice's I/O requests have run in a single AioContext.
In order to support multiple IOThreads it will be necessary to move to
the concept of a per-SCSIRequest AioContext.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250311132616.1049687-4-stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
include/hw/scsi/scsi.h | 1 +
hw/scsi/scsi-bus.c | 1 +
hw/scsi/scsi-disk.c | 17 ++++++-----------
3 files changed, 8 insertions(+), 11 deletions(-)
diff --git a/include/hw/scsi/scsi.h b/include/hw/scsi/scsi.h
index c3d5e17e38..ffc48203f9 100644
--- a/include/hw/scsi/scsi.h
+++ b/include/hw/scsi/scsi.h
@@ -24,6 +24,7 @@ struct SCSIRequest {
SCSIBus *bus;
SCSIDevice *dev;
const SCSIReqOps *ops;
+ AioContext *ctx;
uint32_t refcount;
uint32_t tag;
uint32_t lun;
diff --git a/hw/scsi/scsi-bus.c b/hw/scsi/scsi-bus.c
index 7d4546800f..846bbbf0ec 100644
--- a/hw/scsi/scsi-bus.c
+++ b/hw/scsi/scsi-bus.c
@@ -868,6 +868,7 @@ invalid_opcode:
}
}
+ req->ctx = qemu_get_current_aio_context();
req->cmd = cmd;
req->residual = req->cmd.xfer;
diff --git a/hw/scsi/scsi-disk.c b/hw/scsi/scsi-disk.c
index f049a20275..7cf8c31b98 100644
--- a/hw/scsi/scsi-disk.c
+++ b/hw/scsi/scsi-disk.c
@@ -328,9 +328,8 @@ static void scsi_aio_complete(void *opaque, int ret)
SCSIDiskReq *r = (SCSIDiskReq *)opaque;
SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
- /* The request must only run in the BlockBackend's AioContext */
- assert(blk_get_aio_context(s->qdev.conf.blk) ==
- qemu_get_current_aio_context());
+ /* The request must run in its AioContext */
+ assert(r->req.ctx == qemu_get_current_aio_context());
assert(r->req.aiocb != NULL);
r->req.aiocb = NULL;
@@ -430,12 +429,10 @@ static void scsi_dma_complete(void *opaque, int ret)
static void scsi_read_complete_noio(SCSIDiskReq *r, int ret)
{
- SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
uint32_t n;
- /* The request must only run in the BlockBackend's AioContext */
- assert(blk_get_aio_context(s->qdev.conf.blk) ==
- qemu_get_current_aio_context());
+ /* The request must run in its AioContext */
+ assert(r->req.ctx == qemu_get_current_aio_context());
assert(r->req.aiocb == NULL);
if (scsi_disk_req_check_error(r, ret, ret > 0)) {
@@ -562,12 +559,10 @@ static void scsi_read_data(SCSIRequest *req)
static void scsi_write_complete_noio(SCSIDiskReq *r, int ret)
{
- SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
uint32_t n;
- /* The request must only run in the BlockBackend's AioContext */
- assert(blk_get_aio_context(s->qdev.conf.blk) ==
- qemu_get_current_aio_context());
+ /* The request must run in its AioContext */
+ assert(r->req.ctx == qemu_get_current_aio_context());
assert (r->req.aiocb == NULL);
if (scsi_disk_req_check_error(r, ret, ret > 0)) {
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* [PULL 13/22] scsi: introduce requests_lock
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
` (11 preceding siblings ...)
2025-03-11 16:00 ` [PULL 12/22] scsi: track per-SCSIRequest AioContext Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-11 16:00 ` [PULL 14/22] virtio-scsi: introduce event and ctrl virtqueue locks Kevin Wolf
` (9 subsequent siblings)
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
From: Stefan Hajnoczi <stefanha@redhat.com>
SCSIDevice keeps track of in-flight requests for device reset and Task
Management Functions (TMFs). The request list requires protection so
that multi-threaded SCSI emulation can be implemented in commits that
follow.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250311132616.1049687-5-stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
include/hw/scsi/scsi.h | 7 ++-
hw/scsi/scsi-bus.c | 120 +++++++++++++++++++++++++++++------------
2 files changed, 88 insertions(+), 39 deletions(-)
diff --git a/include/hw/scsi/scsi.h b/include/hw/scsi/scsi.h
index ffc48203f9..90ee192b4d 100644
--- a/include/hw/scsi/scsi.h
+++ b/include/hw/scsi/scsi.h
@@ -49,6 +49,8 @@ struct SCSIRequest {
bool dma_started;
BlockAIOCB *aiocb;
QEMUSGList *sg;
+
+ /* Protected by SCSIDevice->requests_lock */
QTAILQ_ENTRY(SCSIRequest) next;
};
@@ -77,10 +79,7 @@ struct SCSIDevice
uint8_t sense[SCSI_SENSE_BUF_SIZE];
uint32_t sense_len;
- /*
- * The requests list is only accessed from the AioContext that executes
- * requests or from the main loop when IOThread processing is stopped.
- */
+ QemuMutex requests_lock; /* protects the requests list */
QTAILQ_HEAD(, SCSIRequest) requests;
uint32_t channel;
diff --git a/hw/scsi/scsi-bus.c b/hw/scsi/scsi-bus.c
index 846bbbf0ec..ece1107ee8 100644
--- a/hw/scsi/scsi-bus.c
+++ b/hw/scsi/scsi-bus.c
@@ -100,8 +100,15 @@ static void scsi_device_for_each_req_sync(SCSIDevice *s,
assert(!runstate_is_running());
assert(qemu_in_main_thread());
- QTAILQ_FOREACH_SAFE(req, &s->requests, next, next_req) {
- fn(req, opaque);
+ /*
+ * Locking is not necessary because the guest is stopped and no other
+ * threads can be accessing the requests list, but take the lock for
+ * consistency.
+ */
+ WITH_QEMU_LOCK_GUARD(&s->requests_lock) {
+ QTAILQ_FOREACH_SAFE(req, &s->requests, next, next_req) {
+ fn(req, opaque);
+ }
}
}
@@ -115,21 +122,29 @@ static void scsi_device_for_each_req_async_bh(void *opaque)
{
g_autofree SCSIDeviceForEachReqAsyncData *data = opaque;
SCSIDevice *s = data->s;
- AioContext *ctx;
- SCSIRequest *req;
- SCSIRequest *next;
+ g_autoptr(GList) reqs = NULL;
/*
- * The BB cannot have changed contexts between this BH being scheduled and
- * now: BBs' AioContexts, when they have a node attached, can only be
- * changed via bdrv_try_change_aio_context(), in a drained section. While
- * we have the in-flight counter incremented, that drain must block.
+ * Build a list of requests in this AioContext so fn() can be invoked later
+ * outside requests_lock.
*/
- ctx = blk_get_aio_context(s->conf.blk);
- assert(ctx == qemu_get_current_aio_context());
+ WITH_QEMU_LOCK_GUARD(&s->requests_lock) {
+ AioContext *ctx = qemu_get_current_aio_context();
+ SCSIRequest *req;
+ SCSIRequest *next;
+
+ QTAILQ_FOREACH_SAFE(req, &s->requests, next, next) {
+ if (req->ctx == ctx) {
+ scsi_req_ref(req); /* dropped after calling fn() */
+ reqs = g_list_prepend(reqs, req);
+ }
+ }
+ }
- QTAILQ_FOREACH_SAFE(req, &s->requests, next, next) {
- data->fn(req, data->fn_opaque);
+ /* Call fn() on each request */
+ for (GList *elem = g_list_first(reqs); elem; elem = g_list_next(elem)) {
+ data->fn(elem->data, data->fn_opaque);
+ scsi_req_unref(elem->data);
}
/* Drop the reference taken by scsi_device_for_each_req_async() */
@@ -139,9 +154,35 @@ static void scsi_device_for_each_req_async_bh(void *opaque)
blk_dec_in_flight(s->conf.blk);
}
+static void scsi_device_for_each_req_async_do_ctx(gpointer key, gpointer value,
+ gpointer user_data)
+{
+ AioContext *ctx = key;
+ SCSIDeviceForEachReqAsyncData *params = user_data;
+ SCSIDeviceForEachReqAsyncData *data;
+
+ data = g_new(SCSIDeviceForEachReqAsyncData, 1);
+ data->s = params->s;
+ data->fn = params->fn;
+ data->fn_opaque = params->fn_opaque;
+
+ /*
+ * Hold a reference to the SCSIDevice until
+ * scsi_device_for_each_req_async_bh() finishes.
+ */
+ object_ref(OBJECT(data->s));
+
+ /* Paired with scsi_device_for_each_req_async_bh() */
+ blk_inc_in_flight(data->s->conf.blk);
+
+ aio_bh_schedule_oneshot(ctx, scsi_device_for_each_req_async_bh, data);
+}
+
/*
* Schedule @fn() to be invoked for each enqueued request in device @s. @fn()
- * runs in the AioContext that is executing the request.
+ * must be thread-safe because it runs concurrently in each AioContext that is
+ * executing a request.
+ *
* Keeps the BlockBackend's in-flight counter incremented until everything is
* done, so draining it will settle all scheduled @fn() calls.
*/
@@ -151,24 +192,26 @@ static void scsi_device_for_each_req_async(SCSIDevice *s,
{
assert(qemu_in_main_thread());
- SCSIDeviceForEachReqAsyncData *data =
- g_new(SCSIDeviceForEachReqAsyncData, 1);
-
- data->s = s;
- data->fn = fn;
- data->fn_opaque = opaque;
-
- /*
- * Hold a reference to the SCSIDevice until
- * scsi_device_for_each_req_async_bh() finishes.
- */
- object_ref(OBJECT(s));
+ /* The set of AioContexts where the requests are being processed */
+ g_autoptr(GHashTable) aio_contexts = g_hash_table_new(NULL, NULL);
+ WITH_QEMU_LOCK_GUARD(&s->requests_lock) {
+ SCSIRequest *req;
+ QTAILQ_FOREACH(req, &s->requests, next) {
+ g_hash_table_add(aio_contexts, req->ctx);
+ }
+ }
- /* Paired with blk_dec_in_flight() in scsi_device_for_each_req_async_bh() */
- blk_inc_in_flight(s->conf.blk);
- aio_bh_schedule_oneshot(blk_get_aio_context(s->conf.blk),
- scsi_device_for_each_req_async_bh,
- data);
+ /* Schedule a BH for each AioContext */
+ SCSIDeviceForEachReqAsyncData params = {
+ .s = s,
+ .fn = fn,
+ .fn_opaque = opaque,
+ };
+ g_hash_table_foreach(
+ aio_contexts,
+ scsi_device_for_each_req_async_do_ctx,
+ ¶ms
+ );
}
static void scsi_device_realize(SCSIDevice *s, Error **errp)
@@ -349,6 +392,7 @@ static void scsi_qdev_realize(DeviceState *qdev, Error **errp)
dev->lun = lun;
}
+ qemu_mutex_init(&dev->requests_lock);
QTAILQ_INIT(&dev->requests);
scsi_device_realize(dev, &local_err);
if (local_err) {
@@ -369,6 +413,8 @@ static void scsi_qdev_unrealize(DeviceState *qdev)
scsi_device_purge_requests(dev, SENSE_CODE(NO_SENSE));
+ qemu_mutex_destroy(&dev->requests_lock);
+
scsi_device_unrealize(dev);
blockdev_mark_auto_del(dev->conf.blk);
@@ -965,7 +1011,10 @@ static void scsi_req_enqueue_internal(SCSIRequest *req)
req->sg = NULL;
}
req->enqueued = true;
- QTAILQ_INSERT_TAIL(&req->dev->requests, req, next);
+
+ WITH_QEMU_LOCK_GUARD(&req->dev->requests_lock) {
+ QTAILQ_INSERT_TAIL(&req->dev->requests, req, next);
+ }
}
int32_t scsi_req_enqueue(SCSIRequest *req)
@@ -985,7 +1034,9 @@ static void scsi_req_dequeue(SCSIRequest *req)
trace_scsi_req_dequeue(req->dev->id, req->lun, req->tag);
req->retry = false;
if (req->enqueued) {
- QTAILQ_REMOVE(&req->dev->requests, req, next);
+ WITH_QEMU_LOCK_GUARD(&req->dev->requests_lock) {
+ QTAILQ_REMOVE(&req->dev->requests, req, next);
+ }
req->enqueued = false;
scsi_req_unref(req);
}
@@ -1962,8 +2013,7 @@ static void scsi_device_class_init(ObjectClass *klass, void *data)
static void scsi_dev_instance_init(Object *obj)
{
- DeviceState *dev = DEVICE(obj);
- SCSIDevice *s = SCSI_DEVICE(dev);
+ SCSIDevice *s = SCSI_DEVICE(obj);
device_add_bootindex_property(obj, &s->conf.bootindex,
"bootindex", NULL,
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* [PULL 14/22] virtio-scsi: introduce event and ctrl virtqueue locks
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
` (12 preceding siblings ...)
2025-03-11 16:00 ` [PULL 13/22] scsi: introduce requests_lock Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-11 16:00 ` [PULL 15/22] virtio-scsi: protect events_dropped field Kevin Wolf
` (8 subsequent siblings)
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
From: Stefan Hajnoczi <stefanha@redhat.com>
Virtqueues are not thread-safe. Until now this was not a major issue
since all virtqueue processing happened in the same thread. The ctrl
queue's Task Management Function (TMF) requests sometimes need the main
loop, so a BH was used to schedule the virtqueue completion back in the
thread that has virtqueue access.
When IOThread Virtqueue Mapping is introduced in later commits, event
and ctrl virtqueue accesses from other threads will become necessary.
Introduce an optional per-virtqueue lock so the event and ctrl
virtqueues can be protected in the commits that follow.
The addition of the ctrl virtqueue lock makes
virtio_scsi_complete_req_from_main_loop() and its BH unnecessary.
Instead, take the ctrl virtqueue lock from the main loop thread.
The cmd virtqueue does not have a lock because the entirety of SCSI
command processing happens in one thread. Only one thread accesses the
cmd virtqueue and a lock is unnecessary.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250311132616.1049687-6-stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
include/hw/virtio/virtio-scsi.h | 3 ++
hw/scsi/virtio-scsi.c | 84 ++++++++++++++++++---------------
2 files changed, 49 insertions(+), 38 deletions(-)
diff --git a/include/hw/virtio/virtio-scsi.h b/include/hw/virtio/virtio-scsi.h
index be230cd4bf..4ee98ebf63 100644
--- a/include/hw/virtio/virtio-scsi.h
+++ b/include/hw/virtio/virtio-scsi.h
@@ -84,6 +84,9 @@ struct VirtIOSCSI {
int resetting; /* written from main loop thread, read from any thread */
bool events_dropped;
+ QemuMutex ctrl_lock; /* protects ctrl_vq */
+ QemuMutex event_lock; /* protects event_vq */
+
/*
* TMFs deferred to main loop BH. These fields are protected by
* tmf_bh_lock.
diff --git a/hw/scsi/virtio-scsi.c b/hw/scsi/virtio-scsi.c
index 7d094e1881..073ccd3d5b 100644
--- a/hw/scsi/virtio-scsi.c
+++ b/hw/scsi/virtio-scsi.c
@@ -102,13 +102,18 @@ static void virtio_scsi_free_req(VirtIOSCSIReq *req)
g_free(req);
}
-static void virtio_scsi_complete_req(VirtIOSCSIReq *req)
+static void virtio_scsi_complete_req(VirtIOSCSIReq *req, QemuMutex *vq_lock)
{
VirtIOSCSI *s = req->dev;
VirtQueue *vq = req->vq;
VirtIODevice *vdev = VIRTIO_DEVICE(s);
qemu_iovec_from_buf(&req->resp_iov, 0, &req->resp, req->resp_size);
+
+ if (vq_lock) {
+ qemu_mutex_lock(vq_lock);
+ }
+
virtqueue_push(vq, &req->elem, req->qsgl.size + req->resp_iov.size);
if (s->dataplane_started && !s->dataplane_fenced) {
virtio_notify_irqfd(vdev, vq);
@@ -116,6 +121,10 @@ static void virtio_scsi_complete_req(VirtIOSCSIReq *req)
virtio_notify(vdev, vq);
}
+ if (vq_lock) {
+ qemu_mutex_unlock(vq_lock);
+ }
+
if (req->sreq) {
req->sreq->hba_private = NULL;
scsi_req_unref(req->sreq);
@@ -123,34 +132,20 @@ static void virtio_scsi_complete_req(VirtIOSCSIReq *req)
virtio_scsi_free_req(req);
}
-static void virtio_scsi_complete_req_bh(void *opaque)
+static void virtio_scsi_bad_req(VirtIOSCSIReq *req, QemuMutex *vq_lock)
{
- VirtIOSCSIReq *req = opaque;
+ virtio_error(VIRTIO_DEVICE(req->dev), "wrong size for virtio-scsi headers");
- virtio_scsi_complete_req(req);
-}
+ if (vq_lock) {
+ qemu_mutex_lock(vq_lock);
+ }
-/*
- * Called from virtio_scsi_do_one_tmf_bh() in main loop thread. The main loop
- * thread cannot touch the virtqueue since that could race with an IOThread.
- */
-static void virtio_scsi_complete_req_from_main_loop(VirtIOSCSIReq *req)
-{
- VirtIOSCSI *s = req->dev;
+ virtqueue_detach_element(req->vq, &req->elem, 0);
- if (!s->ctx || s->ctx == qemu_get_aio_context()) {
- /* No need to schedule a BH when there is no IOThread */
- virtio_scsi_complete_req(req);
- } else {
- /* Run request completion in the IOThread */
- aio_wait_bh_oneshot(s->ctx, virtio_scsi_complete_req_bh, req);
+ if (vq_lock) {
+ qemu_mutex_unlock(vq_lock);
}
-}
-static void virtio_scsi_bad_req(VirtIOSCSIReq *req)
-{
- virtio_error(VIRTIO_DEVICE(req->dev), "wrong size for virtio-scsi headers");
- virtqueue_detach_element(req->vq, &req->elem, 0);
virtio_scsi_free_req(req);
}
@@ -235,12 +230,21 @@ static int virtio_scsi_parse_req(VirtIOSCSIReq *req,
return 0;
}
-static VirtIOSCSIReq *virtio_scsi_pop_req(VirtIOSCSI *s, VirtQueue *vq)
+static VirtIOSCSIReq *virtio_scsi_pop_req(VirtIOSCSI *s, VirtQueue *vq, QemuMutex *vq_lock)
{
VirtIOSCSICommon *vs = (VirtIOSCSICommon *)s;
VirtIOSCSIReq *req;
+ if (vq_lock) {
+ qemu_mutex_lock(vq_lock);
+ }
+
req = virtqueue_pop(vq, sizeof(VirtIOSCSIReq) + vs->cdb_size);
+
+ if (vq_lock) {
+ qemu_mutex_unlock(vq_lock);
+ }
+
if (!req) {
return NULL;
}
@@ -305,7 +309,7 @@ static void virtio_scsi_cancel_notify(Notifier *notifier, void *data)
trace_virtio_scsi_tmf_resp(virtio_scsi_get_lun(req->req.tmf.lun),
req->req.tmf.tag, req->resp.tmf.response);
- virtio_scsi_complete_req(req);
+ virtio_scsi_complete_req(req, &req->dev->ctrl_lock);
}
g_free(n);
}
@@ -361,7 +365,7 @@ static void virtio_scsi_do_one_tmf_bh(VirtIOSCSIReq *req)
out:
object_unref(OBJECT(d));
- virtio_scsi_complete_req_from_main_loop(req);
+ virtio_scsi_complete_req(req, &s->ctrl_lock);
}
/* Some TMFs must be processed from the main loop thread */
@@ -408,7 +412,7 @@ static void virtio_scsi_reset_tmf_bh(VirtIOSCSI *s)
/* SAM-6 6.3.2 Hard reset */
req->resp.tmf.response = VIRTIO_SCSI_S_TARGET_FAILURE;
- virtio_scsi_complete_req(req);
+ virtio_scsi_complete_req(req, &req->dev->ctrl_lock);
}
}
@@ -562,7 +566,7 @@ static void virtio_scsi_handle_ctrl_req(VirtIOSCSI *s, VirtIOSCSIReq *req)
if (iov_to_buf(req->elem.out_sg, req->elem.out_num, 0,
&type, sizeof(type)) < sizeof(type)) {
- virtio_scsi_bad_req(req);
+ virtio_scsi_bad_req(req, &s->ctrl_lock);
return;
}
@@ -570,7 +574,7 @@ static void virtio_scsi_handle_ctrl_req(VirtIOSCSI *s, VirtIOSCSIReq *req)
if (type == VIRTIO_SCSI_T_TMF) {
if (virtio_scsi_parse_req(req, sizeof(VirtIOSCSICtrlTMFReq),
sizeof(VirtIOSCSICtrlTMFResp)) < 0) {
- virtio_scsi_bad_req(req);
+ virtio_scsi_bad_req(req, &s->ctrl_lock);
return;
} else {
r = virtio_scsi_do_tmf(s, req);
@@ -580,7 +584,7 @@ static void virtio_scsi_handle_ctrl_req(VirtIOSCSI *s, VirtIOSCSIReq *req)
type == VIRTIO_SCSI_T_AN_SUBSCRIBE) {
if (virtio_scsi_parse_req(req, sizeof(VirtIOSCSICtrlANReq),
sizeof(VirtIOSCSICtrlANResp)) < 0) {
- virtio_scsi_bad_req(req);
+ virtio_scsi_bad_req(req, &s->ctrl_lock);
return;
} else {
req->req.an.event_requested =
@@ -600,7 +604,7 @@ static void virtio_scsi_handle_ctrl_req(VirtIOSCSI *s, VirtIOSCSIReq *req)
type == VIRTIO_SCSI_T_AN_SUBSCRIBE)
trace_virtio_scsi_an_resp(virtio_scsi_get_lun(req->req.an.lun),
req->resp.an.response);
- virtio_scsi_complete_req(req);
+ virtio_scsi_complete_req(req, &s->ctrl_lock);
} else {
assert(r == -EINPROGRESS);
}
@@ -610,7 +614,7 @@ static void virtio_scsi_handle_ctrl_vq(VirtIOSCSI *s, VirtQueue *vq)
{
VirtIOSCSIReq *req;
- while ((req = virtio_scsi_pop_req(s, vq))) {
+ while ((req = virtio_scsi_pop_req(s, vq, &s->ctrl_lock))) {
virtio_scsi_handle_ctrl_req(s, req);
}
}
@@ -654,7 +658,7 @@ static void virtio_scsi_complete_cmd_req(VirtIOSCSIReq *req)
* in virtio_scsi_command_complete.
*/
req->resp_size = sizeof(VirtIOSCSICmdResp);
- virtio_scsi_complete_req(req);
+ virtio_scsi_complete_req(req, NULL);
}
static void virtio_scsi_command_failed(SCSIRequest *r)
@@ -788,7 +792,7 @@ static int virtio_scsi_handle_cmd_req_prepare(VirtIOSCSI *s, VirtIOSCSIReq *req)
virtio_scsi_fail_cmd_req(req);
return -ENOTSUP;
} else {
- virtio_scsi_bad_req(req);
+ virtio_scsi_bad_req(req, NULL);
return -EINVAL;
}
}
@@ -843,7 +847,7 @@ static void virtio_scsi_handle_cmd_vq(VirtIOSCSI *s, VirtQueue *vq)
virtio_queue_set_notification(vq, 0);
}
- while ((req = virtio_scsi_pop_req(s, vq))) {
+ while ((req = virtio_scsi_pop_req(s, vq, NULL))) {
ret = virtio_scsi_handle_cmd_req_prepare(s, req);
if (!ret) {
QTAILQ_INSERT_TAIL(&reqs, req, next);
@@ -973,7 +977,7 @@ static void virtio_scsi_push_event(VirtIOSCSI *s,
return;
}
- req = virtio_scsi_pop_req(s, vs->event_vq);
+ req = virtio_scsi_pop_req(s, vs->event_vq, &s->event_lock);
if (!req) {
s->events_dropped = true;
return;
@@ -985,7 +989,7 @@ static void virtio_scsi_push_event(VirtIOSCSI *s,
}
if (virtio_scsi_parse_req(req, 0, sizeof(VirtIOSCSIEvent))) {
- virtio_scsi_bad_req(req);
+ virtio_scsi_bad_req(req, &s->event_lock);
return;
}
@@ -1005,7 +1009,7 @@ static void virtio_scsi_push_event(VirtIOSCSI *s,
}
trace_virtio_scsi_event(virtio_scsi_get_lun(evt->lun), event, reason);
- virtio_scsi_complete_req(req);
+ virtio_scsi_complete_req(req, &s->event_lock);
}
static void virtio_scsi_handle_event_vq(VirtIOSCSI *s, VirtQueue *vq)
@@ -1236,6 +1240,8 @@ static void virtio_scsi_device_realize(DeviceState *dev, Error **errp)
Error *err = NULL;
QTAILQ_INIT(&s->tmf_bh_list);
+ qemu_mutex_init(&s->ctrl_lock);
+ qemu_mutex_init(&s->event_lock);
qemu_mutex_init(&s->tmf_bh_lock);
virtio_scsi_common_realize(dev,
@@ -1280,6 +1286,8 @@ static void virtio_scsi_device_unrealize(DeviceState *dev)
qbus_set_hotplug_handler(BUS(&s->bus), NULL);
virtio_scsi_common_unrealize(dev);
qemu_mutex_destroy(&s->tmf_bh_lock);
+ qemu_mutex_destroy(&s->event_lock);
+ qemu_mutex_destroy(&s->ctrl_lock);
}
static const Property virtio_scsi_properties[] = {
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* [PULL 15/22] virtio-scsi: protect events_dropped field
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
` (13 preceding siblings ...)
2025-03-11 16:00 ` [PULL 14/22] virtio-scsi: introduce event and ctrl virtqueue locks Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-11 16:00 ` [PULL 16/22] virtio-scsi: perform TMFs in appropriate AioContexts Kevin Wolf
` (7 subsequent siblings)
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
From: Stefan Hajnoczi <stefanha@redhat.com>
The block layer can invoke the resize callback from any AioContext that
is processing requests. The virtqueue is already protected but the
events_dropped field also needs to be protected against races. Cover it
using the event virtqueue lock because it is closely associated with
accesses to the virtqueue.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250311132616.1049687-7-stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
include/hw/virtio/virtio-scsi.h | 3 ++-
hw/scsi/virtio-scsi.c | 29 ++++++++++++++++++++---------
2 files changed, 22 insertions(+), 10 deletions(-)
diff --git a/include/hw/virtio/virtio-scsi.h b/include/hw/virtio/virtio-scsi.h
index 4ee98ebf63..7b7e3ced7a 100644
--- a/include/hw/virtio/virtio-scsi.h
+++ b/include/hw/virtio/virtio-scsi.h
@@ -82,10 +82,11 @@ struct VirtIOSCSI {
SCSIBus bus;
int resetting; /* written from main loop thread, read from any thread */
+
+ QemuMutex event_lock; /* protects event_vq and events_dropped */
bool events_dropped;
QemuMutex ctrl_lock; /* protects ctrl_vq */
- QemuMutex event_lock; /* protects event_vq */
/*
* TMFs deferred to main loop BH. These fields are protected by
diff --git a/hw/scsi/virtio-scsi.c b/hw/scsi/virtio-scsi.c
index 073ccd3d5b..2d796a861b 100644
--- a/hw/scsi/virtio-scsi.c
+++ b/hw/scsi/virtio-scsi.c
@@ -948,7 +948,10 @@ static void virtio_scsi_reset(VirtIODevice *vdev)
vs->sense_size = VIRTIO_SCSI_SENSE_DEFAULT_SIZE;
vs->cdb_size = VIRTIO_SCSI_CDB_DEFAULT_SIZE;
- s->events_dropped = false;
+
+ WITH_QEMU_LOCK_GUARD(&s->event_lock) {
+ s->events_dropped = false;
+ }
}
typedef struct {
@@ -978,14 +981,16 @@ static void virtio_scsi_push_event(VirtIOSCSI *s,
}
req = virtio_scsi_pop_req(s, vs->event_vq, &s->event_lock);
- if (!req) {
- s->events_dropped = true;
- return;
- }
+ WITH_QEMU_LOCK_GUARD(&s->event_lock) {
+ if (!req) {
+ s->events_dropped = true;
+ return;
+ }
- if (s->events_dropped) {
- event |= VIRTIO_SCSI_T_EVENTS_MISSED;
- s->events_dropped = false;
+ if (s->events_dropped) {
+ event |= VIRTIO_SCSI_T_EVENTS_MISSED;
+ s->events_dropped = false;
+ }
}
if (virtio_scsi_parse_req(req, 0, sizeof(VirtIOSCSIEvent))) {
@@ -1014,7 +1019,13 @@ static void virtio_scsi_push_event(VirtIOSCSI *s,
static void virtio_scsi_handle_event_vq(VirtIOSCSI *s, VirtQueue *vq)
{
- if (s->events_dropped) {
+ bool events_dropped;
+
+ WITH_QEMU_LOCK_GUARD(&s->event_lock) {
+ events_dropped = s->events_dropped;
+ }
+
+ if (events_dropped) {
VirtIOSCSIEventInfo info = {
.event = VIRTIO_SCSI_T_NO_EVENT,
};
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* [PULL 16/22] virtio-scsi: perform TMFs in appropriate AioContexts
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
` (14 preceding siblings ...)
2025-03-11 16:00 ` [PULL 15/22] virtio-scsi: protect events_dropped field Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-11 16:00 ` [PULL 17/22] virtio-blk: extract cleanup_iothread_vq_mapping() function Kevin Wolf
` (6 subsequent siblings)
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
From: Stefan Hajnoczi <stefanha@redhat.com>
With IOThread Virtqueue Mapping there will be multiple AioContexts
processing SCSI requests. scsi_req_cancel() and other SCSI request
operations must be performed from the AioContext where the request is
running.
Introduce a virtio_scsi_defer_tmf_to_aio_context() function and the
necessary VirtIOSCSIReq->remaining refcount infrastructure to move the
TMF code into the AioContext where the request is running.
For the time being there is still just one AioContext: the main loop or
the IOThread. When the iothread-vq-mapping parameter is added in a later
patch this will be changed to per-virtqueue AioContexts.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250311132616.1049687-8-stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
hw/scsi/virtio-scsi.c | 270 ++++++++++++++++++++++++++++++++----------
1 file changed, 206 insertions(+), 64 deletions(-)
diff --git a/hw/scsi/virtio-scsi.c b/hw/scsi/virtio-scsi.c
index 2d796a861b..2045d27289 100644
--- a/hw/scsi/virtio-scsi.c
+++ b/hw/scsi/virtio-scsi.c
@@ -47,7 +47,7 @@ typedef struct VirtIOSCSIReq {
/* Used for two-stage request submission and TMFs deferred to BH */
QTAILQ_ENTRY(VirtIOSCSIReq) next;
- /* Used for cancellation of request during TMFs */
+ /* Used for cancellation of request during TMFs. Atomic. */
int remaining;
SCSIRequest *sreq;
@@ -298,19 +298,23 @@ typedef struct {
VirtIOSCSIReq *tmf_req;
} VirtIOSCSICancelNotifier;
+static void virtio_scsi_tmf_dec_remaining(VirtIOSCSIReq *tmf)
+{
+ if (qatomic_fetch_dec(&tmf->remaining) == 1) {
+ trace_virtio_scsi_tmf_resp(virtio_scsi_get_lun(tmf->req.tmf.lun),
+ tmf->req.tmf.tag, tmf->resp.tmf.response);
+
+ virtio_scsi_complete_req(tmf, &tmf->dev->ctrl_lock);
+ }
+}
+
static void virtio_scsi_cancel_notify(Notifier *notifier, void *data)
{
VirtIOSCSICancelNotifier *n = container_of(notifier,
VirtIOSCSICancelNotifier,
notifier);
- if (--n->tmf_req->remaining == 0) {
- VirtIOSCSIReq *req = n->tmf_req;
-
- trace_virtio_scsi_tmf_resp(virtio_scsi_get_lun(req->req.tmf.lun),
- req->req.tmf.tag, req->resp.tmf.response);
- virtio_scsi_complete_req(req, &req->dev->ctrl_lock);
- }
+ virtio_scsi_tmf_dec_remaining(n->tmf_req);
g_free(n);
}
@@ -416,7 +420,7 @@ static void virtio_scsi_reset_tmf_bh(VirtIOSCSI *s)
}
}
-static void virtio_scsi_defer_tmf_to_bh(VirtIOSCSIReq *req)
+static void virtio_scsi_defer_tmf_to_main_loop(VirtIOSCSIReq *req)
{
VirtIOSCSI *s = req->dev;
@@ -430,6 +434,137 @@ static void virtio_scsi_defer_tmf_to_bh(VirtIOSCSIReq *req)
}
}
+static void virtio_scsi_tmf_cancel_req(VirtIOSCSIReq *tmf, SCSIRequest *r)
+{
+ VirtIOSCSICancelNotifier *notifier;
+
+ assert(r->ctx == qemu_get_current_aio_context());
+
+ /* Decremented in virtio_scsi_cancel_notify() */
+ qatomic_inc(&tmf->remaining);
+
+ notifier = g_new(VirtIOSCSICancelNotifier, 1);
+ notifier->notifier.notify = virtio_scsi_cancel_notify;
+ notifier->tmf_req = tmf;
+ scsi_req_cancel_async(r, ¬ifier->notifier);
+}
+
+/* Execute a TMF on the requests in the current AioContext */
+static void virtio_scsi_do_tmf_aio_context(void *opaque)
+{
+ AioContext *ctx = qemu_get_current_aio_context();
+ VirtIOSCSIReq *tmf = opaque;
+ VirtIOSCSI *s = tmf->dev;
+ SCSIDevice *d = virtio_scsi_device_get(s, tmf->req.tmf.lun);
+ SCSIRequest *r;
+ bool match_tag;
+
+ if (!d) {
+ tmf->resp.tmf.response = VIRTIO_SCSI_S_BAD_TARGET;
+ virtio_scsi_tmf_dec_remaining(tmf);
+ return;
+ }
+
+ /*
+ * This function could handle other subtypes that need to be processed in
+ * the request's AioContext in the future, but for now only request
+ * cancelation subtypes are performed here.
+ */
+ switch (tmf->req.tmf.subtype) {
+ case VIRTIO_SCSI_T_TMF_ABORT_TASK:
+ match_tag = true;
+ break;
+ case VIRTIO_SCSI_T_TMF_ABORT_TASK_SET:
+ case VIRTIO_SCSI_T_TMF_CLEAR_TASK_SET:
+ match_tag = false;
+ break;
+ default:
+ g_assert_not_reached();
+ }
+
+ WITH_QEMU_LOCK_GUARD(&d->requests_lock) {
+ QTAILQ_FOREACH(r, &d->requests, next) {
+ VirtIOSCSIReq *cmd_req = r->hba_private;
+ assert(cmd_req); /* request has hba_private while enqueued */
+
+ if (r->ctx != ctx) {
+ continue;
+ }
+ if (match_tag && cmd_req->req.cmd.tag != tmf->req.tmf.tag) {
+ continue;
+ }
+ virtio_scsi_tmf_cancel_req(tmf, r);
+ }
+ }
+
+ /* Incremented by virtio_scsi_do_tmf() */
+ virtio_scsi_tmf_dec_remaining(tmf);
+
+ object_unref(d);
+}
+
+static void dummy_bh(void *opaque)
+{
+ /* Do nothing */
+}
+
+/*
+ * Wait for pending virtio_scsi_defer_tmf_to_aio_context() BHs.
+ */
+static void virtio_scsi_flush_defer_tmf_to_aio_context(VirtIOSCSI *s)
+{
+ GLOBAL_STATE_CODE();
+
+ assert(!s->dataplane_started);
+
+ if (s->ctx) {
+ /* Our BH only runs after previously scheduled BHs */
+ aio_wait_bh_oneshot(s->ctx, dummy_bh, NULL);
+ }
+}
+
+/*
+ * Run the TMF in a specific AioContext, handling only requests in that
+ * AioContext. This is necessary because requests can run in different
+ * AioContext and it is only possible to cancel them from the AioContext where
+ * they are running.
+ */
+static void virtio_scsi_defer_tmf_to_aio_context(VirtIOSCSIReq *tmf,
+ AioContext *ctx)
+{
+ /* Decremented in virtio_scsi_do_tmf_aio_context() */
+ qatomic_inc(&tmf->remaining);
+
+ /* See virtio_scsi_flush_defer_tmf_to_aio_context() cleanup during reset */
+ aio_bh_schedule_oneshot(ctx, virtio_scsi_do_tmf_aio_context, tmf);
+}
+
+/*
+ * Returns the AioContext for a given TMF's tag field or NULL. Note that the
+ * request identified by the tag may have completed by the time you can execute
+ * a BH in the AioContext, so don't assume the request still exists in your BH.
+ */
+static AioContext *find_aio_context_for_tmf_tag(SCSIDevice *d,
+ VirtIOSCSIReq *tmf)
+{
+ WITH_QEMU_LOCK_GUARD(&d->requests_lock) {
+ SCSIRequest *r;
+ SCSIRequest *next;
+
+ QTAILQ_FOREACH_SAFE(r, &d->requests, next, next) {
+ VirtIOSCSIReq *cmd_req = r->hba_private;
+
+ /* hba_private is non-NULL while the request is enqueued */
+ assert(cmd_req);
+
+ if (cmd_req->req.cmd.tag == tmf->req.tmf.tag) {
+ return r->ctx;
+ }
+ }
+ }
+ return NULL;
+}
+
/* Return 0 if the request is ready to be completed and return to guest;
* -EINPROGRESS if the request is submitted and will be completed later, in the
* case of async cancellation. */
@@ -437,6 +572,7 @@ static int virtio_scsi_do_tmf(VirtIOSCSI *s, VirtIOSCSIReq *req)
{
SCSIDevice *d = virtio_scsi_device_get(s, req->req.tmf.lun);
SCSIRequest *r, *next;
+ AioContext *ctx;
int ret = 0;
virtio_scsi_ctx_check(s, d);
@@ -454,52 +590,72 @@ static int virtio_scsi_do_tmf(VirtIOSCSI *s, VirtIOSCSIReq *req)
req->req.tmf.tag, req->req.tmf.subtype);
switch (req->req.tmf.subtype) {
- case VIRTIO_SCSI_T_TMF_ABORT_TASK:
- case VIRTIO_SCSI_T_TMF_QUERY_TASK:
+ case VIRTIO_SCSI_T_TMF_ABORT_TASK: {
if (!d) {
goto fail;
}
if (d->lun != virtio_scsi_get_lun(req->req.tmf.lun)) {
goto incorrect_lun;
}
- QTAILQ_FOREACH_SAFE(r, &d->requests, next, next) {
- VirtIOSCSIReq *cmd_req = r->hba_private;
- if (cmd_req && cmd_req->req.cmd.tag == req->req.tmf.tag) {
- break;
- }
+
+ ctx = find_aio_context_for_tmf_tag(d, req);
+ if (ctx) {
+ virtio_scsi_defer_tmf_to_aio_context(req, ctx);
+ ret = -EINPROGRESS;
}
- if (r) {
- /*
- * Assert that the request has not been completed yet, we
- * check for it in the loop above.
- */
- assert(r->hba_private);
- if (req->req.tmf.subtype == VIRTIO_SCSI_T_TMF_QUERY_TASK) {
- /* "If the specified command is present in the task set, then
- * return a service response set to FUNCTION SUCCEEDED".
- */
- req->resp.tmf.response = VIRTIO_SCSI_S_FUNCTION_SUCCEEDED;
- } else {
- VirtIOSCSICancelNotifier *notifier;
-
- req->remaining = 1;
- notifier = g_new(VirtIOSCSICancelNotifier, 1);
- notifier->tmf_req = req;
- notifier->notifier.notify = virtio_scsi_cancel_notify;
- scsi_req_cancel_async(r, ¬ifier->notifier);
- ret = -EINPROGRESS;
+ break;
+ }
+
+ case VIRTIO_SCSI_T_TMF_QUERY_TASK:
+ if (!d) {
+ goto fail;
+ }
+ if (d->lun != virtio_scsi_get_lun(req->req.tmf.lun)) {
+ goto incorrect_lun;
+ }
+
+ WITH_QEMU_LOCK_GUARD(&d->requests_lock) {
+ QTAILQ_FOREACH(r, &d->requests, next) {
+ VirtIOSCSIReq *cmd_req = r->hba_private;
+ assert(cmd_req); /* request has hba_private while enqueued */
+
+ if (cmd_req->req.cmd.tag == req->req.tmf.tag) {
+ /*
+ * "If the specified command is present in the task set,
+ * then return a service response set to FUNCTION
+ * SUCCEEDED".
+ */
+ req->resp.tmf.response = VIRTIO_SCSI_S_FUNCTION_SUCCEEDED;
+ }
}
}
break;
case VIRTIO_SCSI_T_TMF_LOGICAL_UNIT_RESET:
case VIRTIO_SCSI_T_TMF_I_T_NEXUS_RESET:
- virtio_scsi_defer_tmf_to_bh(req);
+ virtio_scsi_defer_tmf_to_main_loop(req);
ret = -EINPROGRESS;
break;
case VIRTIO_SCSI_T_TMF_ABORT_TASK_SET:
- case VIRTIO_SCSI_T_TMF_CLEAR_TASK_SET:
+ case VIRTIO_SCSI_T_TMF_CLEAR_TASK_SET: {
+ if (!d) {
+ goto fail;
+ }
+ if (d->lun != virtio_scsi_get_lun(req->req.tmf.lun)) {
+ goto incorrect_lun;
+ }
+
+ qatomic_inc(&req->remaining);
+
+ ctx = s->ctx ?: qemu_get_aio_context();
+ virtio_scsi_defer_tmf_to_aio_context(req, ctx);
+
+ virtio_scsi_tmf_dec_remaining(req);
+ ret = -EINPROGRESS;
+ break;
+ }
+
case VIRTIO_SCSI_T_TMF_QUERY_TASK_SET:
if (!d) {
goto fail;
@@ -508,34 +664,19 @@ static int virtio_scsi_do_tmf(VirtIOSCSI *s, VirtIOSCSIReq *req)
goto incorrect_lun;
}
- /* Add 1 to "remaining" until virtio_scsi_do_tmf returns.
- * This way, if the bus starts calling back to the notifiers
- * even before we finish the loop, virtio_scsi_cancel_notify
- * will not complete the TMF too early.
- */
- req->remaining = 1;
- QTAILQ_FOREACH_SAFE(r, &d->requests, next, next) {
- if (r->hba_private) {
- if (req->req.tmf.subtype == VIRTIO_SCSI_T_TMF_QUERY_TASK_SET) {
- /* "If there is any command present in the task set, then
- * return a service response set to FUNCTION SUCCEEDED".
- */
- req->resp.tmf.response = VIRTIO_SCSI_S_FUNCTION_SUCCEEDED;
- break;
- } else {
- VirtIOSCSICancelNotifier *notifier;
-
- req->remaining++;
- notifier = g_new(VirtIOSCSICancelNotifier, 1);
- notifier->notifier.notify = virtio_scsi_cancel_notify;
- notifier->tmf_req = req;
- scsi_req_cancel_async(r, ¬ifier->notifier);
- }
+ WITH_QEMU_LOCK_GUARD(&d->requests_lock) {
+ QTAILQ_FOREACH_SAFE(r, &d->requests, next, next) {
+ /* Request has hba_private while enqueued */
+ assert(r->hba_private);
+
+ /*
+ * "If there is any command present in the task set, then
+ * return a service response set to FUNCTION SUCCEEDED".
+ */
+ req->resp.tmf.response = VIRTIO_SCSI_S_FUNCTION_SUCCEEDED;
+ break;
}
}
- if (--req->remaining > 0) {
- ret = -EINPROGRESS;
- }
break;
case VIRTIO_SCSI_T_TMF_CLEAR_ACA:
@@ -941,6 +1082,7 @@ static void virtio_scsi_reset(VirtIODevice *vdev)
assert(!s->dataplane_started);
virtio_scsi_reset_tmf_bh(s);
+ virtio_scsi_flush_defer_tmf_to_aio_context(s);
qatomic_inc(&s->resetting);
bus_cold_reset(BUS(&s->bus));
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* [PULL 17/22] virtio-blk: extract cleanup_iothread_vq_mapping() function
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
` (15 preceding siblings ...)
2025-03-11 16:00 ` [PULL 16/22] virtio-scsi: perform TMFs in appropriate AioContexts Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-11 16:00 ` [PULL 18/22] virtio-blk: tidy up iothread_vq_mapping functions Kevin Wolf
` (5 subsequent siblings)
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
From: Stefan Hajnoczi <stefanha@redhat.com>
This is the cleanup function that must be called after
apply_iothread_vq_mapping() succeeds. virtio-scsi will need this
function too, so extract it.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250311132616.1049687-9-stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
hw/block/virtio-blk.c | 27 +++++++++++++++++++++------
1 file changed, 21 insertions(+), 6 deletions(-)
diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c
index 5135b4d8f1..21b1b768ed 100644
--- a/hw/block/virtio-blk.c
+++ b/hw/block/virtio-blk.c
@@ -1495,6 +1495,9 @@ validate_iothread_vq_mapping_list(IOThreadVirtQueueMappingList *list,
* Fill in the AioContext for each virtqueue in the @vq_aio_context array given
* the iothread-vq-mapping parameter in @iothread_vq_mapping_list.
*
+ * cleanup_iothread_vq_mapping() must be called to free IOThread object
+ * references after this function returns success.
+ *
* Returns: %true on success, %false on failure.
**/
static bool apply_iothread_vq_mapping(
@@ -1545,6 +1548,23 @@ static bool apply_iothread_vq_mapping(
return true;
}
+/**
+ * cleanup_iothread_vq_mapping:
+ * @list: The mapping of virtqueues to IOThreads.
+ *
+ * Release IOThread object references that were acquired by
+ * apply_iothread_vq_mapping().
+ */
+static void cleanup_iothread_vq_mapping(IOThreadVirtQueueMappingList *list)
+{
+ IOThreadVirtQueueMappingList *node;
+
+ for (node = list; node; node = node->next) {
+ IOThread *iothread = iothread_by_id(node->value->iothread);
+ object_unref(OBJECT(iothread));
+ }
+}
+
/* Context: BQL held */
static bool virtio_blk_vq_aio_context_init(VirtIOBlock *s, Error **errp)
{
@@ -1611,12 +1631,7 @@ static void virtio_blk_vq_aio_context_cleanup(VirtIOBlock *s)
assert(!s->ioeventfd_started);
if (conf->iothread_vq_mapping_list) {
- IOThreadVirtQueueMappingList *node;
-
- for (node = conf->iothread_vq_mapping_list; node; node = node->next) {
- IOThread *iothread = iothread_by_id(node->value->iothread);
- object_unref(OBJECT(iothread));
- }
+ cleanup_iothread_vq_mapping(conf->iothread_vq_mapping_list);
}
if (conf->iothread) {
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* [PULL 18/22] virtio-blk: tidy up iothread_vq_mapping functions
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
` (16 preceding siblings ...)
2025-03-11 16:00 ` [PULL 17/22] virtio-blk: extract cleanup_iothread_vq_mapping() function Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-11 16:00 ` [PULL 19/22] virtio: extract iothread-vq-mapping.h API Kevin Wolf
` (4 subsequent siblings)
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
From: Stefan Hajnoczi <stefanha@redhat.com>
Use noun_verb() function naming instead of verb_noun() because the
former is the most common naming style for APIs. The next commit will
move these functions into a header file so that virtio-scsi can call
them.
Shorten iothread_vq_mapping_apply()'s iothread_vq_mapping_list argument
to just "list" like in the other functions.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250311132616.1049687-10-stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
hw/block/virtio-blk.c | 33 ++++++++++++++++-----------------
1 file changed, 16 insertions(+), 17 deletions(-)
diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c
index 21b1b768ed..6bf7b50520 100644
--- a/hw/block/virtio-blk.c
+++ b/hw/block/virtio-blk.c
@@ -1424,8 +1424,8 @@ static const BlockDevOps virtio_block_ops = {
};
static bool
-validate_iothread_vq_mapping_list(IOThreadVirtQueueMappingList *list,
- uint16_t num_queues, Error **errp)
+iothread_vq_mapping_validate(IOThreadVirtQueueMappingList *list, uint16_t
+ num_queues, Error **errp)
{
g_autofree unsigned long *vqs = bitmap_new(num_queues);
g_autoptr(GHashTable) iothreads =
@@ -1486,22 +1486,22 @@ validate_iothread_vq_mapping_list(IOThreadVirtQueueMappingList *list,
}
/**
- * apply_iothread_vq_mapping:
- * @iothread_vq_mapping_list: The mapping of virtqueues to IOThreads.
+ * iothread_vq_mapping_apply:
+ * @list: The mapping of virtqueues to IOThreads.
* @vq_aio_context: The array of AioContext pointers to fill in.
* @num_queues: The length of @vq_aio_context.
* @errp: If an error occurs, a pointer to the area to store the error.
*
* Fill in the AioContext for each virtqueue in the @vq_aio_context array given
- * the iothread-vq-mapping parameter in @iothread_vq_mapping_list.
+ * the iothread-vq-mapping parameter in @list.
*
- * cleanup_iothread_vq_mapping() must be called to free IOThread object
+ * iothread_vq_mapping_cleanup() must be called to free IOThread object
* references after this function returns success.
*
* Returns: %true on success, %false on failure.
**/
-static bool apply_iothread_vq_mapping(
- IOThreadVirtQueueMappingList *iothread_vq_mapping_list,
+static bool iothread_vq_mapping_apply(
+ IOThreadVirtQueueMappingList *list,
AioContext **vq_aio_context,
uint16_t num_queues,
Error **errp)
@@ -1510,16 +1510,15 @@ static bool apply_iothread_vq_mapping(
size_t num_iothreads = 0;
size_t cur_iothread = 0;
- if (!validate_iothread_vq_mapping_list(iothread_vq_mapping_list,
- num_queues, errp)) {
+ if (!iothread_vq_mapping_validate(list, num_queues, errp)) {
return false;
}
- for (node = iothread_vq_mapping_list; node; node = node->next) {
+ for (node = list; node; node = node->next) {
num_iothreads++;
}
- for (node = iothread_vq_mapping_list; node; node = node->next) {
+ for (node = list; node; node = node->next) {
IOThread *iothread = iothread_by_id(node->value->iothread);
AioContext *ctx = iothread_get_aio_context(iothread);
@@ -1549,13 +1548,13 @@ static bool apply_iothread_vq_mapping(
}
/**
- * cleanup_iothread_vq_mapping:
+ * iothread_vq_mapping_cleanup:
* @list: The mapping of virtqueues to IOThreads.
*
* Release IOThread object references that were acquired by
- * apply_iothread_vq_mapping().
+ * iothread_vq_mapping_apply().
*/
-static void cleanup_iothread_vq_mapping(IOThreadVirtQueueMappingList *list)
+static void iothread_vq_mapping_cleanup(IOThreadVirtQueueMappingList *list)
{
IOThreadVirtQueueMappingList *node;
@@ -1597,7 +1596,7 @@ static bool virtio_blk_vq_aio_context_init(VirtIOBlock *s, Error **errp)
s->vq_aio_context = g_new(AioContext *, conf->num_queues);
if (conf->iothread_vq_mapping_list) {
- if (!apply_iothread_vq_mapping(conf->iothread_vq_mapping_list,
+ if (!iothread_vq_mapping_apply(conf->iothread_vq_mapping_list,
s->vq_aio_context,
conf->num_queues,
errp)) {
@@ -1631,7 +1630,7 @@ static void virtio_blk_vq_aio_context_cleanup(VirtIOBlock *s)
assert(!s->ioeventfd_started);
if (conf->iothread_vq_mapping_list) {
- cleanup_iothread_vq_mapping(conf->iothread_vq_mapping_list);
+ iothread_vq_mapping_cleanup(conf->iothread_vq_mapping_list);
}
if (conf->iothread) {
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* [PULL 19/22] virtio: extract iothread-vq-mapping.h API
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
` (17 preceding siblings ...)
2025-03-11 16:00 ` [PULL 18/22] virtio-blk: tidy up iothread_vq_mapping functions Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-11 16:00 ` [PULL 20/22] virtio-scsi: add iothread-vq-mapping parameter Kevin Wolf
` (3 subsequent siblings)
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
From: Stefan Hajnoczi <stefanha@redhat.com>
The code that builds an array of AioContext pointers indexed by the
virtqueue is not specific to virtio-blk. virtio-scsi will need to do the
same thing, so extract the functions.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250311132616.1049687-11-stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
include/hw/virtio/iothread-vq-mapping.h | 45 ++++++++
hw/block/virtio-blk.c | 142 +-----------------------
hw/virtio/iothread-vq-mapping.c | 131 ++++++++++++++++++++++
hw/virtio/meson.build | 1 +
4 files changed, 178 insertions(+), 141 deletions(-)
create mode 100644 include/hw/virtio/iothread-vq-mapping.h
create mode 100644 hw/virtio/iothread-vq-mapping.c
diff --git a/include/hw/virtio/iothread-vq-mapping.h b/include/hw/virtio/iothread-vq-mapping.h
new file mode 100644
index 0000000000..57335c3703
--- /dev/null
+++ b/include/hw/virtio/iothread-vq-mapping.h
@@ -0,0 +1,45 @@
+/*
+ * IOThread Virtqueue Mapping
+ *
+ * Copyright Red Hat, Inc
+ *
+ * SPDX-License-Identifier: GPL-2.0-only
+ */
+
+#ifndef HW_VIRTIO_IOTHREAD_VQ_MAPPING_H
+#define HW_VIRTIO_IOTHREAD_VQ_MAPPING_H
+
+#include "qapi/error.h"
+#include "qapi/qapi-types-virtio.h"
+
+/**
+ * iothread_vq_mapping_apply:
+ * @list: The mapping of virtqueues to IOThreads.
+ * @vq_aio_context: The array of AioContext pointers to fill in.
+ * @num_queues: The length of @vq_aio_context.
+ * @errp: If an error occurs, a pointer to the area to store the error.
+ *
+ * Fill in the AioContext for each virtqueue in the @vq_aio_context array given
+ * the iothread-vq-mapping parameter in @list.
+ *
+ * iothread_vq_mapping_cleanup() must be called to free IOThread object
+ * references after this function returns success.
+ *
+ * Returns: %true on success, %false on failure.
+ **/
+bool iothread_vq_mapping_apply(
+ IOThreadVirtQueueMappingList *list,
+ AioContext **vq_aio_context,
+ uint16_t num_queues,
+ Error **errp);
+
+/**
+ * iothread_vq_mapping_cleanup:
+ * @list: The mapping of virtqueues to IOThreads.
+ *
+ * Release IOThread object references that were acquired by
+ * iothread_vq_mapping_apply().
+ */
+void iothread_vq_mapping_cleanup(IOThreadVirtQueueMappingList *list);
+
+#endif /* HW_VIRTIO_IOTHREAD_VQ_MAPPING_H */
diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c
index 6bf7b50520..5077793e5e 100644
--- a/hw/block/virtio-blk.c
+++ b/hw/block/virtio-blk.c
@@ -33,6 +33,7 @@
#endif
#include "hw/virtio/virtio-bus.h"
#include "migration/qemu-file-types.h"
+#include "hw/virtio/iothread-vq-mapping.h"
#include "hw/virtio/virtio-access.h"
#include "hw/virtio/virtio-blk-common.h"
#include "qemu/coroutine.h"
@@ -1423,147 +1424,6 @@ static const BlockDevOps virtio_block_ops = {
.drained_end = virtio_blk_drained_end,
};
-static bool
-iothread_vq_mapping_validate(IOThreadVirtQueueMappingList *list, uint16_t
- num_queues, Error **errp)
-{
- g_autofree unsigned long *vqs = bitmap_new(num_queues);
- g_autoptr(GHashTable) iothreads =
- g_hash_table_new(g_str_hash, g_str_equal);
-
- for (IOThreadVirtQueueMappingList *node = list; node; node = node->next) {
- const char *name = node->value->iothread;
- uint16List *vq;
-
- if (!iothread_by_id(name)) {
- error_setg(errp, "IOThread \"%s\" object does not exist", name);
- return false;
- }
-
- if (!g_hash_table_add(iothreads, (gpointer)name)) {
- error_setg(errp,
- "duplicate IOThread name \"%s\" in iothread-vq-mapping",
- name);
- return false;
- }
-
- if (node != list) {
- if (!!node->value->vqs != !!list->value->vqs) {
- error_setg(errp, "either all items in iothread-vq-mapping "
- "must have vqs or none of them must have it");
- return false;
- }
- }
-
- for (vq = node->value->vqs; vq; vq = vq->next) {
- if (vq->value >= num_queues) {
- error_setg(errp, "vq index %u for IOThread \"%s\" must be "
- "less than num_queues %u in iothread-vq-mapping",
- vq->value, name, num_queues);
- return false;
- }
-
- if (test_and_set_bit(vq->value, vqs)) {
- error_setg(errp, "cannot assign vq %u to IOThread \"%s\" "
- "because it is already assigned", vq->value, name);
- return false;
- }
- }
- }
-
- if (list->value->vqs) {
- for (uint16_t i = 0; i < num_queues; i++) {
- if (!test_bit(i, vqs)) {
- error_setg(errp,
- "missing vq %u IOThread assignment in iothread-vq-mapping",
- i);
- return false;
- }
- }
- }
-
- return true;
-}
-
-/**
- * iothread_vq_mapping_apply:
- * @list: The mapping of virtqueues to IOThreads.
- * @vq_aio_context: The array of AioContext pointers to fill in.
- * @num_queues: The length of @vq_aio_context.
- * @errp: If an error occurs, a pointer to the area to store the error.
- *
- * Fill in the AioContext for each virtqueue in the @vq_aio_context array given
- * the iothread-vq-mapping parameter in @list.
- *
- * iothread_vq_mapping_cleanup() must be called to free IOThread object
- * references after this function returns success.
- *
- * Returns: %true on success, %false on failure.
- **/
-static bool iothread_vq_mapping_apply(
- IOThreadVirtQueueMappingList *list,
- AioContext **vq_aio_context,
- uint16_t num_queues,
- Error **errp)
-{
- IOThreadVirtQueueMappingList *node;
- size_t num_iothreads = 0;
- size_t cur_iothread = 0;
-
- if (!iothread_vq_mapping_validate(list, num_queues, errp)) {
- return false;
- }
-
- for (node = list; node; node = node->next) {
- num_iothreads++;
- }
-
- for (node = list; node; node = node->next) {
- IOThread *iothread = iothread_by_id(node->value->iothread);
- AioContext *ctx = iothread_get_aio_context(iothread);
-
- /* Released in virtio_blk_vq_aio_context_cleanup() */
- object_ref(OBJECT(iothread));
-
- if (node->value->vqs) {
- uint16List *vq;
-
- /* Explicit vq:IOThread assignment */
- for (vq = node->value->vqs; vq; vq = vq->next) {
- assert(vq->value < num_queues);
- vq_aio_context[vq->value] = ctx;
- }
- } else {
- /* Round-robin vq:IOThread assignment */
- for (unsigned i = cur_iothread; i < num_queues;
- i += num_iothreads) {
- vq_aio_context[i] = ctx;
- }
- }
-
- cur_iothread++;
- }
-
- return true;
-}
-
-/**
- * iothread_vq_mapping_cleanup:
- * @list: The mapping of virtqueues to IOThreads.
- *
- * Release IOThread object references that were acquired by
- * iothread_vq_mapping_apply().
- */
-static void iothread_vq_mapping_cleanup(IOThreadVirtQueueMappingList *list)
-{
- IOThreadVirtQueueMappingList *node;
-
- for (node = list; node; node = node->next) {
- IOThread *iothread = iothread_by_id(node->value->iothread);
- object_unref(OBJECT(iothread));
- }
-}
-
/* Context: BQL held */
static bool virtio_blk_vq_aio_context_init(VirtIOBlock *s, Error **errp)
{
diff --git a/hw/virtio/iothread-vq-mapping.c b/hw/virtio/iothread-vq-mapping.c
new file mode 100644
index 0000000000..15909eb933
--- /dev/null
+++ b/hw/virtio/iothread-vq-mapping.c
@@ -0,0 +1,131 @@
+/*
+ * IOThread Virtqueue Mapping
+ *
+ * Copyright Red Hat, Inc
+ *
+ * SPDX-License-Identifier: GPL-2.0-only
+ */
+
+#include "qemu/osdep.h"
+#include "system/iothread.h"
+#include "hw/virtio/iothread-vq-mapping.h"
+
+static bool
+iothread_vq_mapping_validate(IOThreadVirtQueueMappingList *list, uint16_t
+ num_queues, Error **errp)
+{
+ g_autofree unsigned long *vqs = bitmap_new(num_queues);
+ g_autoptr(GHashTable) iothreads =
+ g_hash_table_new(g_str_hash, g_str_equal);
+
+ for (IOThreadVirtQueueMappingList *node = list; node; node = node->next) {
+ const char *name = node->value->iothread;
+ uint16List *vq;
+
+ if (!iothread_by_id(name)) {
+ error_setg(errp, "IOThread \"%s\" object does not exist", name);
+ return false;
+ }
+
+ if (!g_hash_table_add(iothreads, (gpointer)name)) {
+ error_setg(errp,
+ "duplicate IOThread name \"%s\" in iothread-vq-mapping",
+ name);
+ return false;
+ }
+
+ if (node != list) {
+ if (!!node->value->vqs != !!list->value->vqs) {
+ error_setg(errp, "either all items in iothread-vq-mapping "
+ "must have vqs or none of them must have it");
+ return false;
+ }
+ }
+
+ for (vq = node->value->vqs; vq; vq = vq->next) {
+ if (vq->value >= num_queues) {
+ error_setg(errp, "vq index %u for IOThread \"%s\" must be "
+ "less than num_queues %u in iothread-vq-mapping",
+ vq->value, name, num_queues);
+ return false;
+ }
+
+ if (test_and_set_bit(vq->value, vqs)) {
+ error_setg(errp, "cannot assign vq %u to IOThread \"%s\" "
+ "because it is already assigned", vq->value, name);
+ return false;
+ }
+ }
+ }
+
+ if (list->value->vqs) {
+ for (uint16_t i = 0; i < num_queues; i++) {
+ if (!test_bit(i, vqs)) {
+ error_setg(errp,
+ "missing vq %u IOThread assignment in iothread-vq-mapping",
+ i);
+ return false;
+ }
+ }
+ }
+
+ return true;
+}
+
+bool iothread_vq_mapping_apply(
+ IOThreadVirtQueueMappingList *list,
+ AioContext **vq_aio_context,
+ uint16_t num_queues,
+ Error **errp)
+{
+ IOThreadVirtQueueMappingList *node;
+ size_t num_iothreads = 0;
+ size_t cur_iothread = 0;
+
+ if (!iothread_vq_mapping_validate(list, num_queues, errp)) {
+ return false;
+ }
+
+ for (node = list; node; node = node->next) {
+ num_iothreads++;
+ }
+
+ for (node = list; node; node = node->next) {
+ IOThread *iothread = iothread_by_id(node->value->iothread);
+ AioContext *ctx = iothread_get_aio_context(iothread);
+
+ /* Released in virtio_blk_vq_aio_context_cleanup() */
+ object_ref(OBJECT(iothread));
+
+ if (node->value->vqs) {
+ uint16List *vq;
+
+ /* Explicit vq:IOThread assignment */
+ for (vq = node->value->vqs; vq; vq = vq->next) {
+ assert(vq->value < num_queues);
+ vq_aio_context[vq->value] = ctx;
+ }
+ } else {
+ /* Round-robin vq:IOThread assignment */
+ for (unsigned i = cur_iothread; i < num_queues;
+ i += num_iothreads) {
+ vq_aio_context[i] = ctx;
+ }
+ }
+
+ cur_iothread++;
+ }
+
+ return true;
+}
+
+void iothread_vq_mapping_cleanup(IOThreadVirtQueueMappingList *list)
+{
+ IOThreadVirtQueueMappingList *node;
+
+ for (node = list; node; node = node->next) {
+ IOThread *iothread = iothread_by_id(node->value->iothread);
+ object_unref(OBJECT(iothread));
+ }
+}
+
diff --git a/hw/virtio/meson.build b/hw/virtio/meson.build
index a5f9f7999d..19b04c4d9c 100644
--- a/hw/virtio/meson.build
+++ b/hw/virtio/meson.build
@@ -1,5 +1,6 @@
system_virtio_ss = ss.source_set()
system_virtio_ss.add(files('virtio-bus.c'))
+system_virtio_ss.add(files('iothread-vq-mapping.c'))
system_virtio_ss.add(when: 'CONFIG_VIRTIO_PCI', if_true: files('virtio-pci.c'))
system_virtio_ss.add(when: 'CONFIG_VIRTIO_MMIO', if_true: files('virtio-mmio.c'))
system_virtio_ss.add(when: 'CONFIG_VIRTIO_CRYPTO', if_true: files('virtio-crypto.c'))
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* [PULL 20/22] virtio-scsi: add iothread-vq-mapping parameter
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
` (18 preceding siblings ...)
2025-03-11 16:00 ` [PULL 19/22] virtio: extract iothread-vq-mapping.h API Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-26 10:43 ` Thomas Huth
2025-03-11 16:00 ` [PULL 21/22] virtio-scsi: handle ctrl virtqueue in main loop Kevin Wolf
` (2 subsequent siblings)
22 siblings, 1 reply; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
From: Stefan Hajnoczi <stefanha@redhat.com>
Allow virtio-scsi virtqueues to be assigned to different IOThreads. This
makes it possible to take advantage of host multi-queue block layer
scalability by assigning virtqueues that have affinity with vCPUs to
different IOThreads that have affinity with host CPUs. The same feature
was introduced for virtio-blk in the past:
https://developers.redhat.com/articles/2024/09/05/scaling-virtio-blk-disk-io-iothread-virtqueue-mapping
Here are fio randread 4k iodepth=64 results from a 4 vCPU guest with an
Intel P4800X SSD:
iothreads IOPS
------------------------------
1 189576
2 312698
4 346744
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-ID: <20250311132616.1049687-12-stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
include/hw/virtio/virtio-scsi.h | 5 +-
hw/scsi/virtio-scsi-dataplane.c | 90 ++++++++++++++++++++++++---------
hw/scsi/virtio-scsi.c | 63 ++++++++++++++---------
3 files changed, 107 insertions(+), 51 deletions(-)
diff --git a/include/hw/virtio/virtio-scsi.h b/include/hw/virtio/virtio-scsi.h
index 7b7e3ced7a..086201efa2 100644
--- a/include/hw/virtio/virtio-scsi.h
+++ b/include/hw/virtio/virtio-scsi.h
@@ -22,6 +22,7 @@
#include "hw/virtio/virtio.h"
#include "hw/scsi/scsi.h"
#include "chardev/char-fe.h"
+#include "qapi/qapi-types-virtio.h"
#include "system/iothread.h"
#define TYPE_VIRTIO_SCSI_COMMON "virtio-scsi-common"
@@ -60,6 +61,7 @@ struct VirtIOSCSIConf {
CharBackend chardev;
uint32_t boot_tpgt;
IOThread *iothread;
+ IOThreadVirtQueueMappingList *iothread_vq_mapping_list;
};
struct VirtIOSCSI;
@@ -97,7 +99,7 @@ struct VirtIOSCSI {
QTAILQ_HEAD(, VirtIOSCSIReq) tmf_bh_list;
/* Fields for dataplane below */
- AioContext *ctx; /* one iothread per virtio-scsi-pci for now */
+ AioContext **vq_aio_context; /* per-virtqueue AioContext pointer */
bool dataplane_started;
bool dataplane_starting;
@@ -115,6 +117,7 @@ void virtio_scsi_common_realize(DeviceState *dev,
void virtio_scsi_common_unrealize(DeviceState *dev);
void virtio_scsi_dataplane_setup(VirtIOSCSI *s, Error **errp);
+void virtio_scsi_dataplane_cleanup(VirtIOSCSI *s);
int virtio_scsi_dataplane_start(VirtIODevice *s);
void virtio_scsi_dataplane_stop(VirtIODevice *s);
diff --git a/hw/scsi/virtio-scsi-dataplane.c b/hw/scsi/virtio-scsi-dataplane.c
index f49ab98ecc..6bb368c8a5 100644
--- a/hw/scsi/virtio-scsi-dataplane.c
+++ b/hw/scsi/virtio-scsi-dataplane.c
@@ -18,6 +18,7 @@
#include "system/block-backend.h"
#include "hw/scsi/scsi.h"
#include "scsi/constants.h"
+#include "hw/virtio/iothread-vq-mapping.h"
#include "hw/virtio/virtio-bus.h"
/* Context: BQL held */
@@ -27,8 +28,16 @@ void virtio_scsi_dataplane_setup(VirtIOSCSI *s, Error **errp)
VirtIODevice *vdev = VIRTIO_DEVICE(s);
BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
+ uint16_t num_vqs = vs->conf.num_queues + VIRTIO_SCSI_VQ_NUM_FIXED;
- if (vs->conf.iothread) {
+ if (vs->conf.iothread && vs->conf.iothread_vq_mapping_list) {
+ error_setg(errp,
+ "iothread and iothread-vq-mapping properties cannot be set "
+ "at the same time");
+ return;
+ }
+
+ if (vs->conf.iothread || vs->conf.iothread_vq_mapping_list) {
if (!k->set_guest_notifiers || !k->ioeventfd_assign) {
error_setg(errp,
"device is incompatible with iothread "
@@ -39,13 +48,48 @@ void virtio_scsi_dataplane_setup(VirtIOSCSI *s, Error **errp)
error_setg(errp, "ioeventfd is required for iothread");
return;
}
- s->ctx = iothread_get_aio_context(vs->conf.iothread);
- } else {
- if (!virtio_device_ioeventfd_enabled(vdev)) {
+ }
+
+ s->vq_aio_context = g_new(AioContext *, num_vqs);
+
+ if (vs->conf.iothread_vq_mapping_list) {
+ if (!iothread_vq_mapping_apply(vs->conf.iothread_vq_mapping_list,
+ s->vq_aio_context, num_vqs, errp)) {
+ g_free(s->vq_aio_context);
+ s->vq_aio_context = NULL;
return;
}
- s->ctx = qemu_get_aio_context();
+ } else if (vs->conf.iothread) {
+ AioContext *ctx = iothread_get_aio_context(vs->conf.iothread);
+ for (uint16_t i = 0; i < num_vqs; i++) {
+ s->vq_aio_context[i] = ctx;
+ }
+
+ /* Released in virtio_scsi_dataplane_cleanup() */
+ object_ref(OBJECT(vs->conf.iothread));
+ } else {
+ AioContext *ctx = qemu_get_aio_context();
+ for (unsigned i = 0; i < num_vqs; i++) {
+ s->vq_aio_context[i] = ctx;
+ }
+ }
+}
+
+/* Context: BQL held */
+void virtio_scsi_dataplane_cleanup(VirtIOSCSI *s)
+{
+ VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(s);
+
+ if (vs->conf.iothread_vq_mapping_list) {
+ iothread_vq_mapping_cleanup(vs->conf.iothread_vq_mapping_list);
}
+
+ if (vs->conf.iothread) {
+ object_unref(OBJECT(vs->conf.iothread));
+ }
+
+ g_free(s->vq_aio_context);
+ s->vq_aio_context = NULL;
}
static int virtio_scsi_set_host_notifier(VirtIOSCSI *s, VirtQueue *vq, int n)
@@ -66,31 +110,20 @@ static int virtio_scsi_set_host_notifier(VirtIOSCSI *s, VirtQueue *vq, int n)
}
/* Context: BH in IOThread */
-static void virtio_scsi_dataplane_stop_bh(void *opaque)
+static void virtio_scsi_dataplane_stop_vq_bh(void *opaque)
{
- VirtIOSCSI *s = opaque;
- VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(s);
+ AioContext *ctx = qemu_get_current_aio_context();
+ VirtQueue *vq = opaque;
EventNotifier *host_notifier;
- int i;
- virtio_queue_aio_detach_host_notifier(vs->ctrl_vq, s->ctx);
- host_notifier = virtio_queue_get_host_notifier(vs->ctrl_vq);
+ virtio_queue_aio_detach_host_notifier(vq, ctx);
+ host_notifier = virtio_queue_get_host_notifier(vq);
/*
* Test and clear notifier after disabling event, in case poll callback
* didn't have time to run.
*/
virtio_queue_host_notifier_read(host_notifier);
-
- virtio_queue_aio_detach_host_notifier(vs->event_vq, s->ctx);
- host_notifier = virtio_queue_get_host_notifier(vs->event_vq);
- virtio_queue_host_notifier_read(host_notifier);
-
- for (i = 0; i < vs->conf.num_queues; i++) {
- virtio_queue_aio_detach_host_notifier(vs->cmd_vqs[i], s->ctx);
- host_notifier = virtio_queue_get_host_notifier(vs->cmd_vqs[i]);
- virtio_queue_host_notifier_read(host_notifier);
- }
}
/* Context: BQL held */
@@ -154,11 +187,14 @@ int virtio_scsi_dataplane_start(VirtIODevice *vdev)
smp_wmb(); /* paired with aio_notify_accept() */
if (s->bus.drain_count == 0) {
- virtio_queue_aio_attach_host_notifier(vs->ctrl_vq, s->ctx);
- virtio_queue_aio_attach_host_notifier_no_poll(vs->event_vq, s->ctx);
+ virtio_queue_aio_attach_host_notifier(vs->ctrl_vq,
+ s->vq_aio_context[0]);
+ virtio_queue_aio_attach_host_notifier_no_poll(vs->event_vq,
+ s->vq_aio_context[1]);
for (i = 0; i < vs->conf.num_queues; i++) {
- virtio_queue_aio_attach_host_notifier(vs->cmd_vqs[i], s->ctx);
+ AioContext *ctx = s->vq_aio_context[VIRTIO_SCSI_VQ_NUM_FIXED + i];
+ virtio_queue_aio_attach_host_notifier(vs->cmd_vqs[i], ctx);
}
}
return 0;
@@ -207,7 +243,11 @@ void virtio_scsi_dataplane_stop(VirtIODevice *vdev)
s->dataplane_stopping = true;
if (s->bus.drain_count == 0) {
- aio_wait_bh_oneshot(s->ctx, virtio_scsi_dataplane_stop_bh, s);
+ for (i = 0; i < vs->conf.num_queues + VIRTIO_SCSI_VQ_NUM_FIXED; i++) {
+ VirtQueue *vq = virtio_get_queue(&vs->parent_obj, i);
+ AioContext *ctx = s->vq_aio_context[i];
+ aio_wait_bh_oneshot(ctx, virtio_scsi_dataplane_stop_vq_bh, vq);
+ }
}
blk_drain_all(); /* ensure there are no in-flight requests */
diff --git a/hw/scsi/virtio-scsi.c b/hw/scsi/virtio-scsi.c
index 2045d27289..9f61eb97db 100644
--- a/hw/scsi/virtio-scsi.c
+++ b/hw/scsi/virtio-scsi.c
@@ -27,6 +27,7 @@
#include "hw/qdev-properties.h"
#include "hw/scsi/scsi.h"
#include "scsi/constants.h"
+#include "hw/virtio/iothread-vq-mapping.h"
#include "hw/virtio/virtio-bus.h"
#include "hw/virtio/virtio-access.h"
#include "trace.h"
@@ -318,13 +319,6 @@ static void virtio_scsi_cancel_notify(Notifier *notifier, void *data)
g_free(n);
}
-static inline void virtio_scsi_ctx_check(VirtIOSCSI *s, SCSIDevice *d)
-{
- if (s->dataplane_started && d && blk_is_available(d->conf.blk)) {
- assert(blk_get_aio_context(d->conf.blk) == s->ctx);
- }
-}
-
static void virtio_scsi_do_one_tmf_bh(VirtIOSCSIReq *req)
{
VirtIOSCSI *s = req->dev;
@@ -517,9 +511,11 @@ static void virtio_scsi_flush_defer_tmf_to_aio_context(VirtIOSCSI *s)
assert(!s->dataplane_started);
- if (s->ctx) {
+ for (uint32_t i = 0; i < s->parent_obj.conf.num_queues; i++) {
+ AioContext *ctx = s->vq_aio_context[VIRTIO_SCSI_VQ_NUM_FIXED + i];
+
/* Our BH only runs after previously scheduled BHs */
- aio_wait_bh_oneshot(s->ctx, dummy_bh, NULL);
+ aio_wait_bh_oneshot(ctx, dummy_bh, NULL);
}
}
@@ -575,7 +571,6 @@ static int virtio_scsi_do_tmf(VirtIOSCSI *s, VirtIOSCSIReq *req)
AioContext *ctx;
int ret = 0;
- virtio_scsi_ctx_check(s, d);
/* Here VIRTIO_SCSI_S_OK means "FUNCTION COMPLETE". */
req->resp.tmf.response = VIRTIO_SCSI_S_OK;
@@ -639,6 +634,8 @@ static int virtio_scsi_do_tmf(VirtIOSCSI *s, VirtIOSCSIReq *req)
case VIRTIO_SCSI_T_TMF_ABORT_TASK_SET:
case VIRTIO_SCSI_T_TMF_CLEAR_TASK_SET: {
+ g_autoptr(GHashTable) aio_contexts = g_hash_table_new(NULL, NULL);
+
if (!d) {
goto fail;
}
@@ -648,8 +645,15 @@ static int virtio_scsi_do_tmf(VirtIOSCSI *s, VirtIOSCSIReq *req)
qatomic_inc(&req->remaining);
- ctx = s->ctx ?: qemu_get_aio_context();
- virtio_scsi_defer_tmf_to_aio_context(req, ctx);
+ for (uint32_t i = 0; i < s->parent_obj.conf.num_queues; i++) {
+ ctx = s->vq_aio_context[VIRTIO_SCSI_VQ_NUM_FIXED + i];
+
+ if (!g_hash_table_add(aio_contexts, ctx)) {
+ continue; /* skip previously added AioContext */
+ }
+
+ virtio_scsi_defer_tmf_to_aio_context(req, ctx);
+ }
virtio_scsi_tmf_dec_remaining(req);
ret = -EINPROGRESS;
@@ -770,9 +774,12 @@ static void virtio_scsi_handle_ctrl_vq(VirtIOSCSI *s, VirtQueue *vq)
*/
static bool virtio_scsi_defer_to_dataplane(VirtIOSCSI *s)
{
- if (!s->ctx || s->dataplane_started) {
+ if (s->dataplane_started) {
return false;
}
+ if (s->vq_aio_context[0] == qemu_get_aio_context()) {
+ return false; /* not using IOThreads */
+ }
virtio_device_start_ioeventfd(&s->parent_obj.parent_obj);
return !s->dataplane_fenced;
@@ -946,7 +953,6 @@ static int virtio_scsi_handle_cmd_req_prepare(VirtIOSCSI *s, VirtIOSCSIReq *req)
virtio_scsi_complete_cmd_req(req);
return -ENOENT;
}
- virtio_scsi_ctx_check(s, d);
req->sreq = scsi_req_new(d, req->req.cmd.tag,
virtio_scsi_get_lun(req->req.cmd.lun),
req->req.cmd.cdb, vs->cdb_size, req);
@@ -1218,14 +1224,16 @@ static void virtio_scsi_hotplug(HotplugHandler *hotplug_dev, DeviceState *dev,
{
VirtIODevice *vdev = VIRTIO_DEVICE(hotplug_dev);
VirtIOSCSI *s = VIRTIO_SCSI(vdev);
+ AioContext *ctx = s->vq_aio_context[VIRTIO_SCSI_VQ_NUM_FIXED];
SCSIDevice *sd = SCSI_DEVICE(dev);
- int ret;
- if (s->ctx && !s->dataplane_fenced) {
- ret = blk_set_aio_context(sd->conf.blk, s->ctx, errp);
- if (ret < 0) {
- return;
- }
+ if (ctx != qemu_get_aio_context() && !s->dataplane_fenced) {
+ /*
+ * Try to make the BlockBackend's AioContext match ours. Ignore failure
+ * because I/O will still work although block jobs and other users
+ * might be slower when multiple AioContexts use a BlockBackend.
+ */
+ blk_set_aio_context(sd->conf.blk, ctx, errp);
}
if (virtio_vdev_has_feature(vdev, VIRTIO_SCSI_F_HOTPLUG)) {
@@ -1260,7 +1268,7 @@ static void virtio_scsi_hotunplug(HotplugHandler *hotplug_dev, DeviceState *dev,
qdev_simple_device_unplug_cb(hotplug_dev, dev, errp);
- if (s->ctx) {
+ if (s->vq_aio_context[VIRTIO_SCSI_VQ_NUM_FIXED] != qemu_get_aio_context()) {
/* If other users keep the BlockBackend in the iothread, that's ok */
blk_set_aio_context(sd->conf.blk, qemu_get_aio_context(), NULL);
}
@@ -1294,7 +1302,7 @@ static void virtio_scsi_drained_begin(SCSIBus *bus)
for (uint32_t i = 0; i < total_queues; i++) {
VirtQueue *vq = virtio_get_queue(vdev, i);
- virtio_queue_aio_detach_host_notifier(vq, s->ctx);
+ virtio_queue_aio_detach_host_notifier(vq, s->vq_aio_context[i]);
}
}
@@ -1320,10 +1328,12 @@ static void virtio_scsi_drained_end(SCSIBus *bus)
for (uint32_t i = 0; i < total_queues; i++) {
VirtQueue *vq = virtio_get_queue(vdev, i);
+ AioContext *ctx = s->vq_aio_context[i];
+
if (vq == vs->event_vq) {
- virtio_queue_aio_attach_host_notifier_no_poll(vq, s->ctx);
+ virtio_queue_aio_attach_host_notifier_no_poll(vq, ctx);
} else {
- virtio_queue_aio_attach_host_notifier(vq, s->ctx);
+ virtio_queue_aio_attach_host_notifier(vq, ctx);
}
}
}
@@ -1430,12 +1440,13 @@ void virtio_scsi_common_unrealize(DeviceState *dev)
virtio_cleanup(vdev);
}
+/* main loop */
static void virtio_scsi_device_unrealize(DeviceState *dev)
{
VirtIOSCSI *s = VIRTIO_SCSI(dev);
virtio_scsi_reset_tmf_bh(s);
-
+ virtio_scsi_dataplane_cleanup(s);
qbus_set_hotplug_handler(BUS(&s->bus), NULL);
virtio_scsi_common_unrealize(dev);
qemu_mutex_destroy(&s->tmf_bh_lock);
@@ -1460,6 +1471,8 @@ static const Property virtio_scsi_properties[] = {
VIRTIO_SCSI_F_CHANGE, true),
DEFINE_PROP_LINK("iothread", VirtIOSCSI, parent_obj.conf.iothread,
TYPE_IOTHREAD, IOThread *),
+ DEFINE_PROP_IOTHREAD_VQ_MAPPING_LIST("iothread-vq-mapping", VirtIOSCSI,
+ parent_obj.conf.iothread_vq_mapping_list),
};
static const VMStateDescription vmstate_virtio_scsi = {
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* Re: [PULL 20/22] virtio-scsi: add iothread-vq-mapping parameter
2025-03-11 16:00 ` [PULL 20/22] virtio-scsi: add iothread-vq-mapping parameter Kevin Wolf
@ 2025-03-26 10:43 ` Thomas Huth
2025-04-24 9:39 ` iotest 240 is failing (was: Re: [PULL 20/22] virtio-scsi: add iothread-vq-mapping parameter) Thomas Huth
0 siblings, 1 reply; 28+ messages in thread
From: Thomas Huth @ 2025-03-26 10:43 UTC (permalink / raw)
To: qemu-block, stefanha; +Cc: qemu-devel, Kevin Wolf
On 11/03/2025 17.00, Kevin Wolf wrote:
> From: Stefan Hajnoczi <stefanha@redhat.com>
>
> Allow virtio-scsi virtqueues to be assigned to different IOThreads. This
> makes it possible to take advantage of host multi-queue block layer
> scalability by assigning virtqueues that have affinity with vCPUs to
> different IOThreads that have affinity with host CPUs. The same feature
> was introduced for virtio-blk in the past:
> https://developers.redhat.com/articles/2024/09/05/scaling-virtio-blk-disk-io-iothread-virtqueue-mapping
Hi Stefan!
This patch seems to cause some troubles with iotest 240 in raw (and some
other) mode:
240 fail [11:41:33] [11:41:34] 0.8s (last: 0.8s) output
mismatch (see tests/qemu-iotests/scratch/raw-file-240/240.out.bad)
--- tests/qemu-iotests/240.out
+++ tests/qemu-iotests/scratch/raw-file-240/240.out.bad
@@ -46,11 +46,11 @@
{"execute": "device_add", "arguments": {"bus": "scsi0.0", "drive": "hd0",
"driver": "scsi-hd", "id": "scsi-hd0"}}
{"return": {}}
{"execute": "device_add", "arguments": {"bus": "scsi1.0", "drive": "hd0",
"driver": "scsi-hd", "id": "scsi-hd1"}}
-{"error": {"class": "GenericError", "desc": "Cannot change iothread of
active block backend"}}
+{"return": {}}
{"execute": "device_del", "arguments": {"id": "scsi-hd0"}}
{"return": {}}
{"execute": "device_add", "arguments": {"bus": "scsi1.0", "drive": "hd0",
"driver": "scsi-hd", "id": "scsi-hd1"}}
-{"return": {}}
+{"error": {"class": "GenericError", "desc": "Duplicate device ID 'scsi-hd1'"}}
{"execute": "device_del", "arguments": {"id": "scsi-hd1"}}
{"return": {}}
{"execute": "blockdev-del", "arguments": {"node-name": "hd0"}}
Failures: 240
Failed 1 of 1 iotests
Could you please have a look?
Thanks,
Thomas
^ permalink raw reply [flat|nested] 28+ messages in thread* iotest 240 is failing (was: Re: [PULL 20/22] virtio-scsi: add iothread-vq-mapping parameter)
2025-03-26 10:43 ` Thomas Huth
@ 2025-04-24 9:39 ` Thomas Huth
2025-05-16 7:21 ` iotest 240 is failing Thomas Huth
0 siblings, 1 reply; 28+ messages in thread
From: Thomas Huth @ 2025-04-24 9:39 UTC (permalink / raw)
To: qemu-block, stefanha, Kevin Wolf; +Cc: qemu-devel
On 26/03/2025 11.43, Thomas Huth wrote:
> On 11/03/2025 17.00, Kevin Wolf wrote:
>> From: Stefan Hajnoczi <stefanha@redhat.com>
>>
>> Allow virtio-scsi virtqueues to be assigned to different IOThreads. This
>> makes it possible to take advantage of host multi-queue block layer
>> scalability by assigning virtqueues that have affinity with vCPUs to
>> different IOThreads that have affinity with host CPUs. The same feature
>> was introduced for virtio-blk in the past:
>> https://developers.redhat.com/articles/2024/09/05/scaling-virtio-blk-disk-
>> io-iothread-virtqueue-mapping
>
> Hi Stefan!
>
> This patch seems to cause some troubles with iotest 240 in raw (and some
> other) mode:
>
> 240 fail [11:41:33] [11:41:34] 0.8s (last: 0.8s) output
> mismatch (see tests/qemu-iotests/scratch/raw-file-240/240.out.bad)
> --- tests/qemu-iotests/240.out
> +++ tests/qemu-iotests/scratch/raw-file-240/240.out.bad
> @@ -46,11 +46,11 @@
> {"execute": "device_add", "arguments": {"bus": "scsi0.0", "drive": "hd0",
> "driver": "scsi-hd", "id": "scsi-hd0"}}
> {"return": {}}
> {"execute": "device_add", "arguments": {"bus": "scsi1.0", "drive": "hd0",
> "driver": "scsi-hd", "id": "scsi-hd1"}}
> -{"error": {"class": "GenericError", "desc": "Cannot change iothread of
> active block backend"}}
> +{"return": {}}
> {"execute": "device_del", "arguments": {"id": "scsi-hd0"}}
> {"return": {}}
> {"execute": "device_add", "arguments": {"bus": "scsi1.0", "drive": "hd0",
> "driver": "scsi-hd", "id": "scsi-hd1"}}
> -{"return": {}}
> +{"error": {"class": "GenericError", "desc": "Duplicate device ID 'scsi-hd1'"}}
> {"execute": "device_del", "arguments": {"id": "scsi-hd1"}}
> {"return": {}}
> {"execute": "blockdev-del", "arguments": {"node-name": "hd0"}}
> Failures: 240
> Failed 1 of 1 iotests
>
> Could you please have a look?
Ping!
The problem still persists in QEMU 10.0 :-(
Thomas
^ permalink raw reply [flat|nested] 28+ messages in thread* Re: iotest 240 is failing
2025-04-24 9:39 ` iotest 240 is failing (was: Re: [PULL 20/22] virtio-scsi: add iothread-vq-mapping parameter) Thomas Huth
@ 2025-05-16 7:21 ` Thomas Huth
2025-05-20 14:00 ` Stefan Hajnoczi
0 siblings, 1 reply; 28+ messages in thread
From: Thomas Huth @ 2025-05-16 7:21 UTC (permalink / raw)
To: qemu-block, stefanha, Kevin Wolf; +Cc: qemu-devel, Hanna Czenczek
On 24/04/2025 11.39, Thomas Huth wrote:
> On 26/03/2025 11.43, Thomas Huth wrote:
>> On 11/03/2025 17.00, Kevin Wolf wrote:
>>> From: Stefan Hajnoczi <stefanha@redhat.com>
>>>
>>> Allow virtio-scsi virtqueues to be assigned to different IOThreads. This
>>> makes it possible to take advantage of host multi-queue block layer
>>> scalability by assigning virtqueues that have affinity with vCPUs to
>>> different IOThreads that have affinity with host CPUs. The same feature
>>> was introduced for virtio-blk in the past:
>>> https://developers.redhat.com/articles/2024/09/05/scaling-virtio-blk-
>>> disk- io-iothread-virtqueue-mapping
>>
>> Hi Stefan!
>>
>> This patch seems to cause some troubles with iotest 240 in raw (and some
>> other) mode:
>>
>> 240 fail [11:41:33] [11:41:34] 0.8s (last: 0.8s) output
>> mismatch (see tests/qemu-iotests/scratch/raw-file-240/240.out.bad)
>> --- tests/qemu-iotests/240.out
>> +++ tests/qemu-iotests/scratch/raw-file-240/240.out.bad
>> @@ -46,11 +46,11 @@
>> {"execute": "device_add", "arguments": {"bus": "scsi0.0", "drive":
>> "hd0", "driver": "scsi-hd", "id": "scsi-hd0"}}
>> {"return": {}}
>> {"execute": "device_add", "arguments": {"bus": "scsi1.0", "drive":
>> "hd0", "driver": "scsi-hd", "id": "scsi-hd1"}}
>> -{"error": {"class": "GenericError", "desc": "Cannot change iothread of
>> active block backend"}}
>> +{"return": {}}
>> {"execute": "device_del", "arguments": {"id": "scsi-hd0"}}
>> {"return": {}}
>> {"execute": "device_add", "arguments": {"bus": "scsi1.0", "drive":
>> "hd0", "driver": "scsi-hd", "id": "scsi-hd1"}}
>> -{"return": {}}
>> +{"error": {"class": "GenericError", "desc": "Duplicate device ID 'scsi-
>> hd1'"}}
>> {"execute": "device_del", "arguments": {"id": "scsi-hd1"}}
>> {"return": {}}
>> {"execute": "blockdev-del", "arguments": {"node-name": "hd0"}}
>> Failures: 240
>> Failed 1 of 1 iotests
>>
>> Could you please have a look?
>
> Ping!
>
> The problem still persists in QEMU 10.0 :-(
Ping^2 !
It's still failing. Could anybody please have a look?
Thomas
^ permalink raw reply [flat|nested] 28+ messages in thread* Re: iotest 240 is failing
2025-05-16 7:21 ` iotest 240 is failing Thomas Huth
@ 2025-05-20 14:00 ` Stefan Hajnoczi
0 siblings, 0 replies; 28+ messages in thread
From: Stefan Hajnoczi @ 2025-05-20 14:00 UTC (permalink / raw)
To: Thomas Huth; +Cc: qemu-block, Kevin Wolf, qemu-devel, Hanna Czenczek
[-- Attachment #1: Type: text/plain, Size: 2397 bytes --]
On Fri, May 16, 2025 at 09:21:32AM +0200, Thomas Huth wrote:
> On 24/04/2025 11.39, Thomas Huth wrote:
> > On 26/03/2025 11.43, Thomas Huth wrote:
> > > On 11/03/2025 17.00, Kevin Wolf wrote:
> > > > From: Stefan Hajnoczi <stefanha@redhat.com>
> > > >
> > > > Allow virtio-scsi virtqueues to be assigned to different IOThreads. This
> > > > makes it possible to take advantage of host multi-queue block layer
> > > > scalability by assigning virtqueues that have affinity with vCPUs to
> > > > different IOThreads that have affinity with host CPUs. The same feature
> > > > was introduced for virtio-blk in the past:
> > > > https://developers.redhat.com/articles/2024/09/05/scaling-virtio-blk-
> > > > disk- io-iothread-virtqueue-mapping
> > >
> > > Hi Stefan!
> > >
> > > This patch seems to cause some troubles with iotest 240 in raw (and
> > > some other) mode:
> > >
> > > 240 fail [11:41:33] [11:41:34] 0.8s (last: 0.8s) output
> > > mismatch (see tests/qemu-iotests/scratch/raw-file-240/240.out.bad)
> > > --- tests/qemu-iotests/240.out
> > > +++ tests/qemu-iotests/scratch/raw-file-240/240.out.bad
> > > @@ -46,11 +46,11 @@
> > > {"execute": "device_add", "arguments": {"bus": "scsi0.0", "drive":
> > > "hd0", "driver": "scsi-hd", "id": "scsi-hd0"}}
> > > {"return": {}}
> > > {"execute": "device_add", "arguments": {"bus": "scsi1.0", "drive":
> > > "hd0", "driver": "scsi-hd", "id": "scsi-hd1"}}
> > > -{"error": {"class": "GenericError", "desc": "Cannot change iothread
> > > of active block backend"}}
> > > +{"return": {}}
> > > {"execute": "device_del", "arguments": {"id": "scsi-hd0"}}
> > > {"return": {}}
> > > {"execute": "device_add", "arguments": {"bus": "scsi1.0", "drive":
> > > "hd0", "driver": "scsi-hd", "id": "scsi-hd1"}}
> > > -{"return": {}}
> > > +{"error": {"class": "GenericError", "desc": "Duplicate device ID
> > > 'scsi- hd1'"}}
> > > {"execute": "device_del", "arguments": {"id": "scsi-hd1"}}
> > > {"return": {}}
> > > {"execute": "blockdev-del", "arguments": {"node-name": "hd0"}}
> > > Failures: 240
> > > Failed 1 of 1 iotests
> > >
> > > Could you please have a look?
> >
> > Ping!
> >
> > The problem still persists in QEMU 10.0 :-(
>
> Ping^2 !
>
> It's still failing. Could anybody please have a look?
Hi Thomas,
I will take a look on Thursday.
Stefan
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]
^ permalink raw reply [flat|nested] 28+ messages in thread
* [PULL 21/22] virtio-scsi: handle ctrl virtqueue in main loop
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
` (19 preceding siblings ...)
2025-03-11 16:00 ` [PULL 20/22] virtio-scsi: add iothread-vq-mapping parameter Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-11 16:00 ` [PULL 22/22] virtio-scsi: only expose cmd vqs via iothread-vq-mapping Kevin Wolf
2025-03-12 13:40 ` [PULL 00/22] Block layer patches Stefan Hajnoczi
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
From: Stefan Hajnoczi <stefanha@redhat.com>
Previously the ctrl virtqueue was handled in the AioContext where SCSI
requests are processed. When IOThread Virtqueue Mapping was added things
become more complicated because SCSI requests could run in other
AioContexts.
Simplify by handling the ctrl virtqueue in the main loop where reset
operations can be performed. Note that BHs are still used canceling SCSI
requests in their AioContexts but at least the mean loop activity
doesn't need BHs anymore.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-ID: <20250311132616.1049687-13-stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
include/hw/virtio/virtio-scsi.h | 8 --
hw/scsi/virtio-scsi-dataplane.c | 6 ++
hw/scsi/virtio-scsi.c | 144 ++++++--------------------------
3 files changed, 33 insertions(+), 125 deletions(-)
diff --git a/include/hw/virtio/virtio-scsi.h b/include/hw/virtio/virtio-scsi.h
index 086201efa2..31e852ed6c 100644
--- a/include/hw/virtio/virtio-scsi.h
+++ b/include/hw/virtio/virtio-scsi.h
@@ -90,14 +90,6 @@ struct VirtIOSCSI {
QemuMutex ctrl_lock; /* protects ctrl_vq */
- /*
- * TMFs deferred to main loop BH. These fields are protected by
- * tmf_bh_lock.
- */
- QemuMutex tmf_bh_lock;
- QEMUBH *tmf_bh;
- QTAILQ_HEAD(, VirtIOSCSIReq) tmf_bh_list;
-
/* Fields for dataplane below */
AioContext **vq_aio_context; /* per-virtqueue AioContext pointer */
diff --git a/hw/scsi/virtio-scsi-dataplane.c b/hw/scsi/virtio-scsi-dataplane.c
index 6bb368c8a5..2d37fa6712 100644
--- a/hw/scsi/virtio-scsi-dataplane.c
+++ b/hw/scsi/virtio-scsi-dataplane.c
@@ -73,6 +73,12 @@ void virtio_scsi_dataplane_setup(VirtIOSCSI *s, Error **errp)
s->vq_aio_context[i] = ctx;
}
}
+
+ /*
+ * Always handle the ctrl virtqueue in the main loop thread where device
+ * resets can be performed.
+ */
+ s->vq_aio_context[0] = qemu_get_aio_context();
}
/* Context: BQL held */
diff --git a/hw/scsi/virtio-scsi.c b/hw/scsi/virtio-scsi.c
index 9f61eb97db..d9c32c3245 100644
--- a/hw/scsi/virtio-scsi.c
+++ b/hw/scsi/virtio-scsi.c
@@ -319,115 +319,6 @@ static void virtio_scsi_cancel_notify(Notifier *notifier, void *data)
g_free(n);
}
-static void virtio_scsi_do_one_tmf_bh(VirtIOSCSIReq *req)
-{
- VirtIOSCSI *s = req->dev;
- SCSIDevice *d = virtio_scsi_device_get(s, req->req.tmf.lun);
- BusChild *kid;
- int target;
-
- switch (req->req.tmf.subtype) {
- case VIRTIO_SCSI_T_TMF_LOGICAL_UNIT_RESET:
- if (!d) {
- req->resp.tmf.response = VIRTIO_SCSI_S_BAD_TARGET;
- goto out;
- }
- if (d->lun != virtio_scsi_get_lun(req->req.tmf.lun)) {
- req->resp.tmf.response = VIRTIO_SCSI_S_INCORRECT_LUN;
- goto out;
- }
- qatomic_inc(&s->resetting);
- device_cold_reset(&d->qdev);
- qatomic_dec(&s->resetting);
- break;
-
- case VIRTIO_SCSI_T_TMF_I_T_NEXUS_RESET:
- target = req->req.tmf.lun[1];
- qatomic_inc(&s->resetting);
-
- rcu_read_lock();
- QTAILQ_FOREACH_RCU(kid, &s->bus.qbus.children, sibling) {
- SCSIDevice *d1 = SCSI_DEVICE(kid->child);
- if (d1->channel == 0 && d1->id == target) {
- device_cold_reset(&d1->qdev);
- }
- }
- rcu_read_unlock();
-
- qatomic_dec(&s->resetting);
- break;
-
- default:
- g_assert_not_reached();
- }
-
-out:
- object_unref(OBJECT(d));
- virtio_scsi_complete_req(req, &s->ctrl_lock);
-}
-
-/* Some TMFs must be processed from the main loop thread */
-static void virtio_scsi_do_tmf_bh(void *opaque)
-{
- VirtIOSCSI *s = opaque;
- QTAILQ_HEAD(, VirtIOSCSIReq) reqs = QTAILQ_HEAD_INITIALIZER(reqs);
- VirtIOSCSIReq *req;
- VirtIOSCSIReq *tmp;
-
- GLOBAL_STATE_CODE();
-
- WITH_QEMU_LOCK_GUARD(&s->tmf_bh_lock) {
- QTAILQ_FOREACH_SAFE(req, &s->tmf_bh_list, next, tmp) {
- QTAILQ_REMOVE(&s->tmf_bh_list, req, next);
- QTAILQ_INSERT_TAIL(&reqs, req, next);
- }
-
- qemu_bh_delete(s->tmf_bh);
- s->tmf_bh = NULL;
- }
-
- QTAILQ_FOREACH_SAFE(req, &reqs, next, tmp) {
- QTAILQ_REMOVE(&reqs, req, next);
- virtio_scsi_do_one_tmf_bh(req);
- }
-}
-
-static void virtio_scsi_reset_tmf_bh(VirtIOSCSI *s)
-{
- VirtIOSCSIReq *req;
- VirtIOSCSIReq *tmp;
-
- GLOBAL_STATE_CODE();
-
- /* Called after ioeventfd has been stopped, so tmf_bh_lock is not needed */
- if (s->tmf_bh) {
- qemu_bh_delete(s->tmf_bh);
- s->tmf_bh = NULL;
- }
-
- QTAILQ_FOREACH_SAFE(req, &s->tmf_bh_list, next, tmp) {
- QTAILQ_REMOVE(&s->tmf_bh_list, req, next);
-
- /* SAM-6 6.3.2 Hard reset */
- req->resp.tmf.response = VIRTIO_SCSI_S_TARGET_FAILURE;
- virtio_scsi_complete_req(req, &req->dev->ctrl_lock);
- }
-}
-
-static void virtio_scsi_defer_tmf_to_main_loop(VirtIOSCSIReq *req)
-{
- VirtIOSCSI *s = req->dev;
-
- WITH_QEMU_LOCK_GUARD(&s->tmf_bh_lock) {
- QTAILQ_INSERT_TAIL(&s->tmf_bh_list, req, next);
-
- if (!s->tmf_bh) {
- s->tmf_bh = qemu_bh_new(virtio_scsi_do_tmf_bh, s);
- qemu_bh_schedule(s->tmf_bh);
- }
- }
-}
-
static void virtio_scsi_tmf_cancel_req(VirtIOSCSIReq *tmf, SCSIRequest *r)
{
VirtIOSCSICancelNotifier *notifier;
@@ -627,11 +518,35 @@ static int virtio_scsi_do_tmf(VirtIOSCSI *s, VirtIOSCSIReq *req)
break;
case VIRTIO_SCSI_T_TMF_LOGICAL_UNIT_RESET:
- case VIRTIO_SCSI_T_TMF_I_T_NEXUS_RESET:
- virtio_scsi_defer_tmf_to_main_loop(req);
- ret = -EINPROGRESS;
+ if (!d) {
+ goto fail;
+ }
+ if (d->lun != virtio_scsi_get_lun(req->req.tmf.lun)) {
+ goto incorrect_lun;
+ }
+ qatomic_inc(&s->resetting);
+ device_cold_reset(&d->qdev);
+ qatomic_dec(&s->resetting);
break;
+ case VIRTIO_SCSI_T_TMF_I_T_NEXUS_RESET: {
+ BusChild *kid;
+ int target = req->req.tmf.lun[1];
+ qatomic_inc(&s->resetting);
+
+ rcu_read_lock();
+ QTAILQ_FOREACH_RCU(kid, &s->bus.qbus.children, sibling) {
+ SCSIDevice *d1 = SCSI_DEVICE(kid->child);
+ if (d1->channel == 0 && d1->id == target) {
+ device_cold_reset(&d1->qdev);
+ }
+ }
+ rcu_read_unlock();
+
+ qatomic_dec(&s->resetting);
+ break;
+ }
+
case VIRTIO_SCSI_T_TMF_ABORT_TASK_SET:
case VIRTIO_SCSI_T_TMF_CLEAR_TASK_SET: {
g_autoptr(GHashTable) aio_contexts = g_hash_table_new(NULL, NULL);
@@ -1087,7 +1002,6 @@ static void virtio_scsi_reset(VirtIODevice *vdev)
assert(!s->dataplane_started);
- virtio_scsi_reset_tmf_bh(s);
virtio_scsi_flush_defer_tmf_to_aio_context(s);
qatomic_inc(&s->resetting);
@@ -1402,10 +1316,8 @@ static void virtio_scsi_device_realize(DeviceState *dev, Error **errp)
VirtIOSCSI *s = VIRTIO_SCSI(dev);
Error *err = NULL;
- QTAILQ_INIT(&s->tmf_bh_list);
qemu_mutex_init(&s->ctrl_lock);
qemu_mutex_init(&s->event_lock);
- qemu_mutex_init(&s->tmf_bh_lock);
virtio_scsi_common_realize(dev,
virtio_scsi_handle_ctrl,
@@ -1445,11 +1357,9 @@ static void virtio_scsi_device_unrealize(DeviceState *dev)
{
VirtIOSCSI *s = VIRTIO_SCSI(dev);
- virtio_scsi_reset_tmf_bh(s);
virtio_scsi_dataplane_cleanup(s);
qbus_set_hotplug_handler(BUS(&s->bus), NULL);
virtio_scsi_common_unrealize(dev);
- qemu_mutex_destroy(&s->tmf_bh_lock);
qemu_mutex_destroy(&s->event_lock);
qemu_mutex_destroy(&s->ctrl_lock);
}
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* [PULL 22/22] virtio-scsi: only expose cmd vqs via iothread-vq-mapping
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
` (20 preceding siblings ...)
2025-03-11 16:00 ` [PULL 21/22] virtio-scsi: handle ctrl virtqueue in main loop Kevin Wolf
@ 2025-03-11 16:00 ` Kevin Wolf
2025-03-12 13:40 ` [PULL 00/22] Block layer patches Stefan Hajnoczi
22 siblings, 0 replies; 28+ messages in thread
From: Kevin Wolf @ 2025-03-11 16:00 UTC (permalink / raw)
To: qemu-block; +Cc: kwolf, stefanha, qemu-devel
From: Stefan Hajnoczi <stefanha@redhat.com>
Peter Krempa and Kevin Wolf observed that iothread-vq-mapping is
confusing to use because the control and event virtqueues have a fixed
location before the command virtqueues but need to be treated
differently.
Only expose the command virtqueues via iothread-vq-mapping so that the
command-line parameter is intuitive: it controls where SCSI requests are
processed.
The control virtqueue needs to be hardcoded to the main loop thread for
technical reasons anyway. Kevin also pointed out that it's better to
place the event virtqueue in the main loop thread since its no poll
behavior would prevent polling if assigned to an IOThread.
This change is its own commit to avoid squashing the previous commit.
Suggested-by: Kevin Wolf <kwolf@redhat.com>
Suggested-by: Peter Krempa <pkrempa@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-ID: <20250311132616.1049687-14-stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
hw/scsi/virtio-scsi-dataplane.c | 33 ++++++++++++++++++++-------------
1 file changed, 20 insertions(+), 13 deletions(-)
diff --git a/hw/scsi/virtio-scsi-dataplane.c b/hw/scsi/virtio-scsi-dataplane.c
index 2d37fa6712..95f13fb7c2 100644
--- a/hw/scsi/virtio-scsi-dataplane.c
+++ b/hw/scsi/virtio-scsi-dataplane.c
@@ -28,7 +28,6 @@ void virtio_scsi_dataplane_setup(VirtIOSCSI *s, Error **errp)
VirtIODevice *vdev = VIRTIO_DEVICE(s);
BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
- uint16_t num_vqs = vs->conf.num_queues + VIRTIO_SCSI_VQ_NUM_FIXED;
if (vs->conf.iothread && vs->conf.iothread_vq_mapping_list) {
error_setg(errp,
@@ -50,35 +49,43 @@ void virtio_scsi_dataplane_setup(VirtIOSCSI *s, Error **errp)
}
}
- s->vq_aio_context = g_new(AioContext *, num_vqs);
+ s->vq_aio_context = g_new(AioContext *, vs->conf.num_queues +
+ VIRTIO_SCSI_VQ_NUM_FIXED);
+
+ /*
+ * Handle the ctrl virtqueue in the main loop thread where device resets
+ * can be performed.
+ */
+ s->vq_aio_context[0] = qemu_get_aio_context();
+
+ /*
+ * Handle the event virtqueue in the main loop thread where its no_poll
+ * behavior won't stop IOThread polling.
+ */
+ s->vq_aio_context[1] = qemu_get_aio_context();
if (vs->conf.iothread_vq_mapping_list) {
if (!iothread_vq_mapping_apply(vs->conf.iothread_vq_mapping_list,
- s->vq_aio_context, num_vqs, errp)) {
+ &s->vq_aio_context[VIRTIO_SCSI_VQ_NUM_FIXED],
+ vs->conf.num_queues, errp)) {
g_free(s->vq_aio_context);
s->vq_aio_context = NULL;
return;
}
} else if (vs->conf.iothread) {
AioContext *ctx = iothread_get_aio_context(vs->conf.iothread);
- for (uint16_t i = 0; i < num_vqs; i++) {
- s->vq_aio_context[i] = ctx;
+ for (uint16_t i = 0; i < vs->conf.num_queues; i++) {
+ s->vq_aio_context[VIRTIO_SCSI_VQ_NUM_FIXED + i] = ctx;
}
/* Released in virtio_scsi_dataplane_cleanup() */
object_ref(OBJECT(vs->conf.iothread));
} else {
AioContext *ctx = qemu_get_aio_context();
- for (unsigned i = 0; i < num_vqs; i++) {
- s->vq_aio_context[i] = ctx;
+ for (unsigned i = 0; i < vs->conf.num_queues; i++) {
+ s->vq_aio_context[VIRTIO_SCSI_VQ_NUM_FIXED + i] = ctx;
}
}
-
- /*
- * Always handle the ctrl virtqueue in the main loop thread where device
- * resets can be performed.
- */
- s->vq_aio_context[0] = qemu_get_aio_context();
}
/* Context: BQL held */
--
2.48.1
^ permalink raw reply related [flat|nested] 28+ messages in thread* Re: [PULL 00/22] Block layer patches
2025-03-11 15:59 [PULL 00/22] Block layer patches Kevin Wolf
` (21 preceding siblings ...)
2025-03-11 16:00 ` [PULL 22/22] virtio-scsi: only expose cmd vqs via iothread-vq-mapping Kevin Wolf
@ 2025-03-12 13:40 ` Stefan Hajnoczi
22 siblings, 0 replies; 28+ messages in thread
From: Stefan Hajnoczi @ 2025-03-12 13:40 UTC (permalink / raw)
To: Kevin Wolf; +Cc: qemu-block, stefanha, qemu-devel
On Wed, Mar 12, 2025 at 12:06 AM Kevin Wolf <kwolf@redhat.com> wrote:
>
> The following changes since commit 825b96dbcee23d134b691fc75618b59c5f53da32:
>
> Merge tag 'migration-20250310-pull-request' of https://gitlab.com/farosas/qemu into staging (2025-03-11 09:32:07 +0800)
>
> are available in the Git repository at:
>
> https://repo.or.cz/qemu/kevin.git tags/for-upstream
>
> for you to fetch changes up to a93c04f3cbe690877b3297a9df4767aa811fcd97:
>
> virtio-scsi: only expose cmd vqs via iothread-vq-mapping (2025-03-11 15:49:22 +0100)
>
> ----------------------------------------------------------------
> Block layer patches
>
> - virtio-scsi: add iothread-vq-mapping parameter
> - Improve writethrough performance
> - Fix missing zero init in bdrv_snapshot_goto()
> - Code cleanup and iotests fixes
The following CI failures came up:
1. laio_has_fua() is missing on some host OSes:
https://gitlab.com/qemu-project/qemu/-/jobs/9380928997#L3369
2. io_uring_prep_writev2() without #ifdef HAVE_IO_PREP_PWRITEV2:
https://gitlab.com/qemu-project/qemu/-/jobs/9380928883#L2515
3. qemu-iotests 051:
https://gitlab.com/qemu-project/qemu/-/jobs/9380928853#L4519
Please send a v2 pull request. Thanks!
Stefan
>
> ----------------------------------------------------------------
> Kevin Wolf (8):
> block: Remove unused blk_op_is_blocked()
> block: Zero block driver state before reopening
> file-posix: Support FUA writes
> block/io: Ignore FUA with cache.no-flush=on
> aio: Create AioPolledEvent
> aio-posix: Factor out adjust_polling_time()
> aio-posix: Separate AioPolledEvent per AioHandler
> aio-posix: Adjust polling time also for new handlers
>
> Stefan Hajnoczi (13):
> scsi-disk: drop unused SCSIDiskState->bh field
> dma: use current AioContext for dma_blk_io()
> scsi: track per-SCSIRequest AioContext
> scsi: introduce requests_lock
> virtio-scsi: introduce event and ctrl virtqueue locks
> virtio-scsi: protect events_dropped field
> virtio-scsi: perform TMFs in appropriate AioContexts
> virtio-blk: extract cleanup_iothread_vq_mapping() function
> virtio-blk: tidy up iothread_vq_mapping functions
> virtio: extract iothread-vq-mapping.h API
> virtio-scsi: add iothread-vq-mapping parameter
> virtio-scsi: handle ctrl virtqueue in main loop
> virtio-scsi: only expose cmd vqs via iothread-vq-mapping
>
> Thomas Huth (1):
> iotests: Limit qsd-migrate to working formats
>
> include/block/aio.h | 5 +-
> include/block/raw-aio.h | 8 +-
> include/hw/scsi/scsi.h | 8 +-
> include/hw/virtio/iothread-vq-mapping.h | 45 +++
> include/hw/virtio/virtio-scsi.h | 15 +-
> include/system/block-backend-global-state.h | 1 -
> include/system/dma.h | 3 +-
> util/aio-posix.h | 1 +
> block/block-backend.c | 12 -
> block/file-posix.c | 26 +-
> block/io.c | 4 +
> block/io_uring.c | 13 +-
> block/linux-aio.c | 24 +-
> block/snapshot.c | 1 +
> hw/block/virtio-blk.c | 132 +-------
> hw/ide/core.c | 3 +-
> hw/ide/macio.c | 3 +-
> hw/scsi/scsi-bus.c | 121 +++++--
> hw/scsi/scsi-disk.c | 24 +-
> hw/scsi/virtio-scsi-dataplane.c | 103 ++++--
> hw/scsi/virtio-scsi.c | 502 ++++++++++++++++------------
> hw/virtio/iothread-vq-mapping.c | 131 ++++++++
> system/dma-helpers.c | 8 +-
> util/aio-posix.c | 114 ++++---
> util/async.c | 1 -
> hw/virtio/meson.build | 1 +
> meson.build | 4 +
> tests/qemu-iotests/tests/qsd-migrate | 2 +-
> 28 files changed, 803 insertions(+), 512 deletions(-)
> create mode 100644 include/hw/virtio/iothread-vq-mapping.h
> create mode 100644 hw/virtio/iothread-vq-mapping.c
>
>
^ permalink raw reply [flat|nested] 28+ messages in thread