Series comparison

-[PULL 0/3] Block patches
+[PULL for-6.1 0/3] Block patches
-The following changes since commit d0ed6a69d399ae193959225cdeaa9382746c91cc:
+The following changes since commit 3521ade3510eb5cefb2e27a101667f25dad89935:
-  Update version for v5.1.0 release (2020-08-11 17:07:03 +0100)
+  Merge remote-tracking branch 'remotes/thuth-gitlab/tags/pull-request-2021-07-29' into staging (2021-07-29 13:17:20 +0100)
 are available in the Git repository at:
-  https://github.com/stefanha/qemu.git tags/block-pull-request
+  https://gitlab.com/stefanha/qemu.git tags/block-pull-request
-for you to fetch changes up to 44277bf914471962c9e88e09c859aae65ae109c4:
+for you to fetch changes up to cc8eecd7f105a1dff5876adeb238a14696061a4a:
-  aio-posix: keep aio_notify_me disabled during polling (2020-08-13 13:34:14 =
+  MAINTAINERS: Added myself as a reviewer for the NVMe Block Driver (2021-07-29 17:17:34 +0100)
 +0100)
 ----------------------------------------------------------------
 Pull request
+The main fix here is for io_uring. Spurious -EAGAIN errors can happen and the
+request needs to be resubmitted.
+The MAINTAINERS changes carry no risk and we might as well include them in QEMU
+.1.
 ----------------------------------------------------------------
-Stefan Hajnoczi (3):
+Fabian Ebner (1):
-  async: rename event_notifier_dummy_cb/poll()
+  block/io_uring: resubmit when result is -EAGAIN
   async: always set ctx->notified in aio_notify()
   aio-posix: keep aio_notify_me disabled during polling
- util/aio-posix.c | 47 +++++++++++++++++++++++++----------------------
+Philippe Mathieu-Daudé (1):
- util/async.c     | 36 +++++++++++++++++++++++-------------
+  MAINTAINERS: Added myself as a reviewer for the NVMe Block Driver
 files changed, 48 insertions(+), 35 deletions(-)
---=20
+Stefano Garzarella (1):
-.26.2
+  MAINTAINERS: add Stefano Garzarella as io_uring reviewer
+ MAINTAINERS      |  2 ++
+ block/io_uring.c | 16 +++++++++++++++-
+files changed, 17 insertions(+), 1 deletion(-)
+--
+.31.1

-[PULL 3/3] aio-posix: keep aio_notify_me disabled during polling
+[PULL for-6.1 1/3] MAINTAINERS: add Stefano Garzarella as io_uring reviewer
-Polling only monitors the ctx->notified field and does not need the
+From: Stefano Garzarella <sgarzare@redhat.com>
 ctx->notifier EventNotifier to be signalled. Keep ctx->aio_notify_me
 disabled while polling to avoid unnecessary EventNotifier syscalls.
-This optimization improves virtio-blk 4KB random read performance by
+I've been working with io_uring for a while so I'd like to help
-%. The following results are with an IOThread and the null-co block
+with reviews.
 driver:
-Test         IOPS   Error
+Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
-Before  244518.62 ± 1.20%
+Message-Id: <20210728131515.131045-1-sgarzare@redhat.com>
 After   290706.11 ± 0.44%
 Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
 Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
 Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
 Message-id: 20200806131802.569478-4-stefanha@redhat.com
 Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
 ---
- util/aio-posix.c | 47 +++++++++++++++++++++++++----------------------
+ MAINTAINERS | 1 +
-file changed, 25 insertions(+), 22 deletions(-)
+file changed, 1 insertion(+)
-diff --git a/util/aio-posix.c b/util/aio-posix.c
+diff --git a/MAINTAINERS b/MAINTAINERS
 index XXXXXXX..XXXXXXX 100644
---- a/util/aio-posix.c
+--- a/MAINTAINERS
-+++ b/util/aio-posix.c
++++ b/MAINTAINERS
-@@ -XXX,XX +XXX,XX @@ static bool remove_idle_poll_handlers(AioContext *ctx, int64_t now)
+@@ -XXX,XX +XXX,XX @@ Linux io_uring
-  *
+ M: Aarushi Mehta <mehta.aaru20@gmail.com>
-  * Polls for a given time.
+ M: Julia Suvorova <jusual@redhat.com>
-  *
+ M: Stefan Hajnoczi <stefanha@redhat.com>
-- * Note that ctx->notify_me must be non-zero so this function can detect
++R: Stefano Garzarella <sgarzare@redhat.com>
-- * aio_notify().
+ L: qemu-block@nongnu.org
-- *
+ S: Maintained
-  * Note that the caller must have incremented ctx->list_lock.
+ F: block/io_uring.c
   *
   * Returns: true if progress was made, false otherwise
@@ -XXX,XX +XXX,XX @@ static bool run_poll_handlers(AioContext *ctx, int64_t max_ns, int64_t *timeout)
      bool progress;
      int64_t start_time, elapsed_time;
 -    assert(ctx->notify_me);
      assert(qemu_lockcnt_count(&ctx->list_lock) > 0);
      trace_run_poll_handlers_begin(ctx, max_ns, *timeout);
@@ -XXX,XX +XXX,XX @@ static bool run_poll_handlers(AioContext *ctx, int64_t max_ns, int64_t *timeout)
   * @timeout: timeout for blocking wait, computed by the caller and updated if
   *    polling succeeds.
   *
 - * ctx->notify_me must be non-zero so this function can detect aio_notify().
 - *
   * Note that the caller must have incremented ctx->list_lock.
   *
   * Returns: true if progress was made, false otherwise
@@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking)
      AioHandlerList ready_list = QLIST_HEAD_INITIALIZER(ready_list);
      int ret = 0;
      bool progress;
 +    bool use_notify_me;
      int64_t timeout;
      int64_t start = 0;
@@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking)
       */
      assert(in_aio_context_home_thread(ctx));
 -    /* aio_notify can avoid the expensive event_notifier_set if
 +    qemu_lockcnt_inc(&ctx->list_lock);
 +
 +    if (ctx->poll_max_ns) {
 +        start = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
 +    }
 +
 +    timeout = blocking ? aio_compute_timeout(ctx) : 0;
 +    progress = try_poll_mode(ctx, &timeout);
 +    assert(!(timeout && progress));
 +
 +    /*
 +     * aio_notify can avoid the expensive event_notifier_set if
       * everything (file descriptors, bottom halves, timers) will
       * be re-evaluated before the next blocking poll().  This is
       * already true when aio_poll is called with blocking == false;
       * if blocking == true, it is only true after poll() returns,
       * so disable the optimization now.
       */
 -    if (blocking) {
 +    use_notify_me = timeout != 0;
 +    if (use_notify_me) {
          atomic_set(&ctx->notify_me, atomic_read(&ctx->notify_me) + 2);
          /*
 -         * Write ctx->notify_me before computing the timeout
 -         * (reading bottom half flags, etc.).  Pairs with
 +         * Write ctx->notify_me before reading ctx->notified.  Pairs with
           * smp_mb in aio_notify().
           */
          smp_mb();
 -    }
 -
 -    qemu_lockcnt_inc(&ctx->list_lock);
 -    if (ctx->poll_max_ns) {
 -        start = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
 +        /* Don't block if aio_notify() was called */
 +        if (atomic_read(&ctx->notified)) {
 +            timeout = 0;
 +        }
      }
 -    timeout = blocking ? aio_compute_timeout(ctx) : 0;
 -    progress = try_poll_mode(ctx, &timeout);
 -    assert(!(timeout && progress));
 -
      /* If polling is allowed, non-blocking aio_poll does not need the
       * system call---a single round of run_poll_handlers_once suffices.
       */
@@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking)
          ret = ctx->fdmon_ops->wait(ctx, &ready_list, timeout);
      }
 -    if (blocking) {
 +    if (use_notify_me) {
          /* Finish the poll before clearing the flag.  */
 -        atomic_store_release(&ctx->notify_me, atomic_read(&ctx->notify_me) - 2);
 -        aio_notify_accept(ctx);
 +        atomic_store_release(&ctx->notify_me,
 +                             atomic_read(&ctx->notify_me) - 2);
      }
 +    aio_notify_accept(ctx);
 +
      /* Adjust polling time */
      if (ctx->poll_max_ns) {
          int64_t block_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - start;
 --
-.26.2
+.31.1

-[PULL 2/3] async: always set ctx->notified in aio_notify()
+[PULL for-6.1 2/3] block/io_uring: resubmit when result is -EAGAIN
-aio_notify() does not set ctx->notified when called with
+From: Fabian Ebner <f.ebner@proxmox.com>
 ctx->aio_notify_me disabled. Therefore aio_notify_me needs to be enabled
 during polling.
-This is suboptimal since expensive event_notifier_set(&ctx->notifier)
+Linux SCSI can throw spurious -EAGAIN in some corner cases in its
-and event_notifier_test_and_clear(&ctx->notifier) calls are required
+completion path, which will end up being the result in the completed
-when ctx->aio_notify_me is enabled.
+io_uring request.
-Change aio_notify() so that aio->notified is always set, regardless of
+Resubmitting such requests should allow block jobs to complete, even
-ctx->aio_notify_me. This will make polling cheaper since
+if such spurious errors are encountered.
 ctx->aio_notify_me can remain disabled. Move the
 event_notifier_test_and_clear() to the fd handler function (which is now
 no longer an empty function so "dummy" has been dropped from its name).
-The next patch takes advantage of this by optimizing polling in
+Co-authored-by: Stefan Hajnoczi <stefanha@gmail.com>
-util/aio-posix.c.
+Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
+Signed-off-by: Fabian Ebner <f.ebner@proxmox.com>
-Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
+Message-id: 20210729091029.65369-1-f.ebner@proxmox.com
 Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
 Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
 Message-id: 20200806131802.569478-3-stefanha@redhat.com
 [Paolo Bonzini pointed out that the smp_wmb() in aio_notify_accept()
 should be smp_wb() but the comment should be smp_wmb() instead of
 smp_wb(). Fixed.
 --Stefan]
 Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
 ---
- util/async.c | 32 +++++++++++++++++++++-----------
+ block/io_uring.c | 16 +++++++++++++++-
-file changed, 21 insertions(+), 11 deletions(-)
+file changed, 15 insertions(+), 1 deletion(-)
-diff --git a/util/async.c b/util/async.c
+diff --git a/block/io_uring.c b/block/io_uring.c
 index XXXXXXX..XXXXXXX 100644
---- a/util/async.c
+--- a/block/io_uring.c
-+++ b/util/async.c
++++ b/block/io_uring.c
-@@ -XXX,XX +XXX,XX @@ LuringState *aio_get_linux_io_uring(AioContext *ctx)
+@@ -XXX,XX +XXX,XX @@ static void luring_process_completions(LuringState *s)
+         total_bytes = ret + luringcb->total_read;
- void aio_notify(AioContext *ctx)
- {
+         if (ret < 0) {
--    /* Write e.g. bh->scheduled before reading ctx->notify_me.  Pairs
+-            if (ret == -EINTR) {
-+    /*
++            /*
-+     * Write e.g. bh->flags before writing ctx->notified.  Pairs with smp_mb in
++             * Only writev/readv/fsync requests on regular files or host block
-+     * aio_notify_accept.
++             * devices are submitted. Therefore -EAGAIN is not expected but it's
-+     */
++             * known to happen sometimes with Linux SCSI. Submit again and hope
-+    smp_wmb();
++             * the request completes successfully.
-+    atomic_set(&ctx->notified, true);
++             *
-+
++             * For more information, see:
-+    /*
++             * https://lore.kernel.org/io-uring/20210727165811.284510-3-axboe@kernel.dk/T/#u
-+     * Write ctx->notified before reading ctx->notify_me.  Pairs
++             *
-      * with smp_mb in aio_ctx_prepare or aio_poll.
++             * If the code is changed to submit other types of requests in the
-      */
++             * future, then this workaround may need to be extended to deal with
-     smp_mb();
++             * genuine -EAGAIN results that should not be resubmitted
-     if (atomic_read(&ctx->notify_me)) {
++             * immediately.
-         event_notifier_set(&ctx->notifier);
++             */
--        atomic_mb_set(&ctx->notified, true);
++            if (ret == -EINTR || ret == -EAGAIN) {
-     }
+                 luring_resubmit(s, luringcb);
- }
+                 continue;
+             }
  void aio_notify_accept(AioContext *ctx)
  {
 -    if (atomic_xchg(&ctx->notified, false)
 -#ifdef WIN32
 -        || true
 -#endif
 -    ) {
 -        event_notifier_test_and_clear(&ctx->notifier);
 -    }
 +    atomic_set(&ctx->notified, false);
 +
 +    /*
 +     * Write ctx->notified before reading e.g. bh->flags.  Pairs with smp_wmb
 +     * in aio_notify.
 +     */
 +    smp_mb();
  }
  static void aio_timerlist_notify(void *opaque, QEMUClockType type)
@@ -XXX,XX +XXX,XX @@ static void aio_timerlist_notify(void *opaque, QEMUClockType type)
      aio_notify(opaque);
  }
 -static void aio_context_notifier_dummy_cb(EventNotifier *e)
 +static void aio_context_notifier_cb(EventNotifier *e)
  {
 +    AioContext *ctx = container_of(e, AioContext, notifier);
 +
 +    event_notifier_test_and_clear(&ctx->notifier);
  }
  /* Returns true if aio_notify() was called (e.g. a BH was scheduled) */
@@ -XXX,XX +XXX,XX @@ AioContext *aio_context_new(Error **errp)
      aio_set_event_notifier(ctx, &ctx->notifier,
                             false,
 -                           aio_context_notifier_dummy_cb,
 +                           aio_context_notifier_cb,
                             aio_context_notifier_poll);
  #ifdef CONFIG_LINUX_AIO
      ctx->linux_aio = NULL;
 --
-.26.2
+.31.1

-[PULL 1/3] async: rename event_notifier_dummy_cb/poll()
+[PULL for-6.1 3/3] MAINTAINERS: Added myself as a reviewer for the NVMe Block Driver
-The event_notifier_*() prefix can be confused with the EventNotifier
+From: Philippe Mathieu-Daudé <philmd@redhat.com>
 APIs that are also called event_notifier_*().
-Rename the functions to aio_context_notifier_*() to make it clear that
+I'm interested in following the activity around the NVMe bdrv.
 they relate to the AioContext::notifier field.
-Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
+Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
-Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
+Message-id: 20210728183340.2018313-1-philmd@redhat.com
 Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
 Message-id: 20200806131802.569478-2-stefanha@redhat.com
 Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
 ---
- util/async.c | 8 ++++----
+ MAINTAINERS | 1 +
-file changed, 4 insertions(+), 4 deletions(-)
+file changed, 1 insertion(+)
-diff --git a/util/async.c b/util/async.c
+diff --git a/MAINTAINERS b/MAINTAINERS
 index XXXXXXX..XXXXXXX 100644
---- a/util/async.c
+--- a/MAINTAINERS
-+++ b/util/async.c
++++ b/MAINTAINERS
-@@ -XXX,XX +XXX,XX @@ static void aio_timerlist_notify(void *opaque, QEMUClockType type)
+@@ -XXX,XX +XXX,XX @@ F: block/null.c
-     aio_notify(opaque);
+ NVMe Block Driver
- }
+ M: Stefan Hajnoczi <stefanha@redhat.com>
+ R: Fam Zheng <fam@euphon.net>
--static void event_notifier_dummy_cb(EventNotifier *e)
++R: Philippe Mathieu-Daudé <philmd@redhat.com>
-+static void aio_context_notifier_dummy_cb(EventNotifier *e)
+ L: qemu-block@nongnu.org
- {
+ S: Supported
- }
+ F: block/nvme*
  /* Returns true if aio_notify() was called (e.g. a BH was scheduled) */
 -static bool event_notifier_poll(void *opaque)
 +static bool aio_context_notifier_poll(void *opaque)
  {
      EventNotifier *e = opaque;
      AioContext *ctx = container_of(e, AioContext, notifier);
@@ -XXX,XX +XXX,XX @@ AioContext *aio_context_new(Error **errp)
      aio_set_event_notifier(ctx, &ctx->notifier,
                             false,
 -                           event_notifier_dummy_cb,
 -                           event_notifier_poll);
 +                           aio_context_notifier_dummy_cb,
 +                           aio_context_notifier_poll);
  #ifdef CONFIG_LINUX_AIO
      ctx->linux_aio = NULL;
  #endif
 --
-.26.2
+.31.1

The event_notifier_*() prefix can be confused with the EventNotifier
APIs that are also called event_notifier_*().

Rename the functions to aio_context_notifier_*() to make it clear that
they relate to the AioContext::notifier field.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-id: 20200806131802.569478-2-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
---
 util/async.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/util/async.c b/util/async.c
index XXXXXXX..XXXXXXX 100644
--- a/util/async.c
+++ b/util/async.c
@@ -XXX,XX +XXX,XX @@ static void aio_timerlist_notify(void *opaque, QEMUClockType type)
     aio_notify(opaque);
 }
 
-static void event_notifier_dummy_cb(EventNotifier *e)
+static void aio_context_notifier_dummy_cb(EventNotifier *e)
 {
 }
 
 /* Returns true if aio_notify() was called (e.g. a BH was scheduled) */
-static bool event_notifier_poll(void *opaque)
+static bool aio_context_notifier_poll(void *opaque)
 {
     EventNotifier *e = opaque;
     AioContext *ctx = container_of(e, AioContext, notifier);
@@ -XXX,XX +XXX,XX @@ AioContext *aio_context_new(Error **errp)
 
     aio_set_event_notifier(ctx, &ctx->notifier,
                            false,
-                           event_notifier_dummy_cb,
-                           event_notifier_poll);
+                           aio_context_notifier_dummy_cb,
+                           aio_context_notifier_poll);
 #ifdef CONFIG_LINUX_AIO
     ctx->linux_aio = NULL;
 #endif
-- 
2.26.2

aio_notify() does not set ctx->notified when called with
ctx->aio_notify_me disabled. Therefore aio_notify_me needs to be enabled
during polling.

This is suboptimal since expensive event_notifier_set(&ctx->notifier)
and event_notifier_test_and_clear(&ctx->notifier) calls are required
when ctx->aio_notify_me is enabled.

Change aio_notify() so that aio->notified is always set, regardless of
ctx->aio_notify_me. This will make polling cheaper since
ctx->aio_notify_me can remain disabled. Move the
event_notifier_test_and_clear() to the fd handler function (which is now
no longer an empty function so "dummy" has been dropped from its name).

The next patch takes advantage of this by optimizing polling in
util/aio-posix.c.

[Paolo Bonzini pointed out that the smp_wmb() in aio_notify_accept()
should be smp_wb() but the comment should be smp_wmb() instead of
smp_wb(). Fixed.
--Stefan]

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
---
 util/async.c | 32 +++++++++++++++++++++-----------
 1 file changed, 21 insertions(+), 11 deletions(-)

diff --git a/util/async.c b/util/async.c
index XXXXXXX..XXXXXXX 100644
--- a/util/async.c
+++ b/util/async.c
@@ -XXX,XX +XXX,XX @@ LuringState *aio_get_linux_io_uring(AioContext *ctx)
 
 void aio_notify(AioContext *ctx)
 {
-    /* Write e.g. bh->scheduled before reading ctx->notify_me.  Pairs
+    /*
+     * Write e.g. bh->flags before writing ctx->notified.  Pairs with smp_mb in
+     * aio_notify_accept.
+     */
+    smp_wmb();
+    atomic_set(&ctx->notified, true);
+
+    /*
+     * Write ctx->notified before reading ctx->notify_me.  Pairs
      * with smp_mb in aio_ctx_prepare or aio_poll.
      */
     smp_mb();
     if (atomic_read(&ctx->notify_me)) {
         event_notifier_set(&ctx->notifier);
-        atomic_mb_set(&ctx->notified, true);
     }
 }
 
 void aio_notify_accept(AioContext *ctx)
 {
-    if (atomic_xchg(&ctx->notified, false)
-#ifdef WIN32
-        || true
-#endif
-    ) {
-        event_notifier_test_and_clear(&ctx->notifier);
-    }
+    atomic_set(&ctx->notified, false);
+
+    /*
+     * Write ctx->notified before reading e.g. bh->flags.  Pairs with smp_wmb
+     * in aio_notify.
+     */
+    smp_mb();
 }
 
 static void aio_timerlist_notify(void *opaque, QEMUClockType type)
@@ -XXX,XX +XXX,XX @@ static void aio_timerlist_notify(void *opaque, QEMUClockType type)
     aio_notify(opaque);
 }
 
-static void aio_context_notifier_dummy_cb(EventNotifier *e)
+static void aio_context_notifier_cb(EventNotifier *e)
 {
+    AioContext *ctx = container_of(e, AioContext, notifier);
+
+    event_notifier_test_and_clear(&ctx->notifier);
 }
 
 /* Returns true if aio_notify() was called (e.g. a BH was scheduled) */
@@ -XXX,XX +XXX,XX @@ AioContext *aio_context_new(Error **errp)
 
     aio_set_event_notifier(ctx, &ctx->notifier,
                            false,
-                           aio_context_notifier_dummy_cb,
+                           aio_context_notifier_cb,
                            aio_context_notifier_poll);
 #ifdef CONFIG_LINUX_AIO
     ctx->linux_aio = NULL;
-- 
2.26.2

Polling only monitors the ctx->notified field and does not need the
ctx->notifier EventNotifier to be signalled. Keep ctx->aio_notify_me
disabled while polling to avoid unnecessary EventNotifier syscalls.

This optimization improves virtio-blk 4KB random read performance by
18%. The following results are with an IOThread and the null-co block
driver:

Test         IOPS   Error
Before  244518.62 ± 1.20%
After   290706.11 ± 0.44%

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-id: 20200806131802.569478-4-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
---
 util/aio-posix.c | 47 +++++++++++++++++++++++++----------------------
 1 file changed, 25 insertions(+), 22 deletions(-)

diff --git a/util/aio-posix.c b/util/aio-posix.c
index XXXXXXX..XXXXXXX 100644
--- a/util/aio-posix.c
+++ b/util/aio-posix.c
@@ -XXX,XX +XXX,XX @@ static bool remove_idle_poll_handlers(AioContext *ctx, int64_t now)
  *
  * Polls for a given time.
  *
- * Note that ctx->notify_me must be non-zero so this function can detect
- * aio_notify().
- *
  * Note that the caller must have incremented ctx->list_lock.
  *
  * Returns: true if progress was made, false otherwise
@@ -XXX,XX +XXX,XX @@ static bool run_poll_handlers(AioContext *ctx, int64_t max_ns, int64_t *timeout)
     bool progress;
     int64_t start_time, elapsed_time;
 
-    assert(ctx->notify_me);
     assert(qemu_lockcnt_count(&ctx->list_lock) > 0);
 
     trace_run_poll_handlers_begin(ctx, max_ns, *timeout);
@@ -XXX,XX +XXX,XX @@ static bool run_poll_handlers(AioContext *ctx, int64_t max_ns, int64_t *timeout)
  * @timeout: timeout for blocking wait, computed by the caller and updated if
  *    polling succeeds.
  *
- * ctx->notify_me must be non-zero so this function can detect aio_notify().
- *
  * Note that the caller must have incremented ctx->list_lock.
  *
  * Returns: true if progress was made, false otherwise
@@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking)
     AioHandlerList ready_list = QLIST_HEAD_INITIALIZER(ready_list);
     int ret = 0;
     bool progress;
+    bool use_notify_me;
     int64_t timeout;
     int64_t start = 0;
 
@@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking)
      */
     assert(in_aio_context_home_thread(ctx));
 
-    /* aio_notify can avoid the expensive event_notifier_set if
+    qemu_lockcnt_inc(&ctx->list_lock);
+
+    if (ctx->poll_max_ns) {
+        start = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
+    }
+
+    timeout = blocking ? aio_compute_timeout(ctx) : 0;
+    progress = try_poll_mode(ctx, &timeout);
+    assert(!(timeout && progress));
+
+    /*
+     * aio_notify can avoid the expensive event_notifier_set if
      * everything (file descriptors, bottom halves, timers) will
      * be re-evaluated before the next blocking poll().  This is
      * already true when aio_poll is called with blocking == false;
      * if blocking == true, it is only true after poll() returns,
      * so disable the optimization now.
      */
-    if (blocking) {
+    use_notify_me = timeout != 0;
+    if (use_notify_me) {
         atomic_set(&ctx->notify_me, atomic_read(&ctx->notify_me) + 2);
         /*
-         * Write ctx->notify_me before computing the timeout
-         * (reading bottom half flags, etc.).  Pairs with
+         * Write ctx->notify_me before reading ctx->notified.  Pairs with
          * smp_mb in aio_notify().
          */
         smp_mb();
-    }
-
-    qemu_lockcnt_inc(&ctx->list_lock);
 
-    if (ctx->poll_max_ns) {
-        start = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
+        /* Don't block if aio_notify() was called */
+        if (atomic_read(&ctx->notified)) {
+            timeout = 0;
+        }
     }
 
-    timeout = blocking ? aio_compute_timeout(ctx) : 0;
-    progress = try_poll_mode(ctx, &timeout);
-    assert(!(timeout && progress));
-
     /* If polling is allowed, non-blocking aio_poll does not need the
      * system call---a single round of run_poll_handlers_once suffices.
      */
@@ -XXX,XX +XXX,XX @@ bool aio_poll(AioContext *ctx, bool blocking)
         ret = ctx->fdmon_ops->wait(ctx, &ready_list, timeout);
     }
 
-    if (blocking) {
+    if (use_notify_me) {
         /* Finish the poll before clearing the flag.  */
-        atomic_store_release(&ctx->notify_me, atomic_read(&ctx->notify_me) - 2);
-        aio_notify_accept(ctx);
+        atomic_store_release(&ctx->notify_me,
+                             atomic_read(&ctx->notify_me) - 2);
     }
 
+    aio_notify_accept(ctx);
+
     /* Adjust polling time */
     if (ctx->poll_max_ns) {
         int64_t block_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - start;
-- 
2.26.2

The following changes since commit 3521ade3510eb5cefb2e27a101667f25dad89935:

Merge remote-tracking branch 'remotes/thuth-gitlab/tags/pull-request-2021-07-29' into staging (2021-07-29 13:17:20 +0100)

are available in the Git repository at:

https://gitlab.com/stefanha/qemu.git tags/block-pull-request

for you to fetch changes up to cc8eecd7f105a1dff5876adeb238a14696061a4a:

MAINTAINERS: Added myself as a reviewer for the NVMe Block Driver (2021-07-29 17:17:34 +0100)

----------------------------------------------------------------
Pull request

The main fix here is for io_uring. Spurious -EAGAIN errors can happen and the
request needs to be resubmitted.

The MAINTAINERS changes carry no risk and we might as well include them in QEMU
6.1.

----------------------------------------------------------------

Fabian Ebner (1):
  block/io_uring: resubmit when result is -EAGAIN

Philippe Mathieu-Daudé (1):
  MAINTAINERS: Added myself as a reviewer for the NVMe Block Driver

Stefano Garzarella (1):
  MAINTAINERS: add Stefano Garzarella as io_uring reviewer

MAINTAINERS      |  2 ++
 block/io_uring.c | 16 +++++++++++++++-
 2 files changed, 17 insertions(+), 1 deletion(-)

-- 
2.31.1

From: Fabian Ebner <f.ebner@proxmox.com>

Linux SCSI can throw spurious -EAGAIN in some corner cases in its
completion path, which will end up being the result in the completed
io_uring request.

Resubmitting such requests should allow block jobs to complete, even
if such spurious errors are encountered.

Co-authored-by: Stefan Hajnoczi <stefanha@gmail.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Fabian Ebner <f.ebner@proxmox.com>
Message-id: 20210729091029.65369-1-f.ebner@proxmox.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
---
 block/io_uring.c | 16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)

diff --git a/block/io_uring.c b/block/io_uring.c
index XXXXXXX..XXXXXXX 100644
--- a/block/io_uring.c
+++ b/block/io_uring.c
@@ -XXX,XX +XXX,XX @@ static void luring_process_completions(LuringState *s)
         total_bytes = ret + luringcb->total_read;
 
         if (ret < 0) {
-            if (ret == -EINTR) {
+            /*
+             * Only writev/readv/fsync requests on regular files or host block
+             * devices are submitted. Therefore -EAGAIN is not expected but it's
+             * known to happen sometimes with Linux SCSI. Submit again and hope
+             * the request completes successfully.
+             *
+             * For more information, see:
+             * https://lore.kernel.org/io-uring/20210727165811.284510-3-axboe@kernel.dk/T/#u
+             *
+             * If the code is changed to submit other types of requests in the
+             * future, then this workaround may need to be extended to deal with
+             * genuine -EAGAIN results that should not be resubmitted
+             * immediately.
+             */
+            if (ret == -EINTR || ret == -EAGAIN) {
                 luring_resubmit(s, luringcb);
                 continue;
             }
-- 
2.31.1