[Qemu-devel] linux-aio: fix batch submission

[Qemu-devel] linux-aio: fix batch submission

[Ming Lei]

Hi,

The 1st patch fixes batch submission.

The 2nd one fixes -EAGAIN for non-batch case.

The 3rd one increase max event to 256 for supporting the comming
multi virt-queue.

The 4th one is a cleanup.

This patchset is splitted from previous patchset(dataplane: optimization
and multi virtqueue support), as suggested by Stefan.

Thanks,
--
Ming Lei

[Qemu-devel] [PATCH 1/4] linux-aio: fix submit aio as a batch

[Ming Lei]

In the enqueue path, we can't complete request, otherwise
"Co-routine re-entered recursively" may be caused, so this
patch fixes the issue with below ideas:

	- for -EAGAIN or partial completion, retry the submision by
	schedule an BH in following completion cb
	- for part of completion, also update the io queue
	- for other failure, return the failure if in enqueue path,
	otherwise, abort all queued I/O

Signed-off-by: Ming Lei <ming.lei@canonical.com>
---
 block/linux-aio.c |   99 +++++++++++++++++++++++++++++++++++++++++------------
 1 file changed, 77 insertions(+), 22 deletions(-)

diff --git a/block/linux-aio.c b/block/linux-aio.c
index 7ac7e8c..4cdf507 100644
--- a/block/linux-aio.c
+++ b/block/linux-aio.c
@@ -38,11 +38,19 @@ struct qemu_laiocb {
     QLIST_ENTRY(qemu_laiocb) node;
 };
 
+/*
+ * TODO: support to batch I/O from multiple bs in one same
+ * AIO context, one important use case is multi-lun scsi,
+ * so in future the IO queue should be per AIO context.
+ */
 typedef struct {
     struct iocb *iocbs[MAX_QUEUED_IO];
     int plugged;
     unsigned int size;
     unsigned int idx;
+
+    /* handle -EAGAIN and partial completion */
+    QEMUBH *retry;
 } LaioQueue;
 
 struct qemu_laio_state {
@@ -86,6 +94,12 @@ static void qemu_laio_process_completion(struct qemu_laio_state *s,
     qemu_aio_release(laiocb);
 }
 
+static void qemu_laio_start_retry(struct qemu_laio_state *s)
+{
+    if (s->io_q.idx)
+        qemu_bh_schedule(s->io_q.retry);
+}
+
 static void qemu_laio_completion_cb(EventNotifier *e)
 {
     struct qemu_laio_state *s = container_of(e, struct qemu_laio_state, e);
@@ -108,6 +122,7 @@ static void qemu_laio_completion_cb(EventNotifier *e)
             qemu_laio_process_completion(s, laiocb);
         }
     }
+    qemu_laio_start_retry(s);
 }
 
 static void laio_cancel(BlockDriverAIOCB *blockacb)
@@ -127,6 +142,7 @@ static void laio_cancel(BlockDriverAIOCB *blockacb)
     ret = io_cancel(laiocb->ctx->ctx, &laiocb->iocb, &event);
     if (ret == 0) {
         laiocb->ret = -ECANCELED;
+        qemu_laio_start_retry(laiocb->ctx);
         return;
     }
 
@@ -154,45 +170,80 @@ static void ioq_init(LaioQueue *io_q)
     io_q->plugged = 0;
 }
 
-static int ioq_submit(struct qemu_laio_state *s)
+static void abort_queue(struct qemu_laio_state *s)
+{
+    int i;
+    for (i = 0; i < s->io_q.idx; i++) {
+        struct qemu_laiocb *laiocb = container_of(s->io_q.iocbs[i],
+                                                  struct qemu_laiocb,
+                                                  iocb);
+        laiocb->ret = -EIO;
+        qemu_laio_process_completion(s, laiocb);
+    }
+}
+
+static int ioq_submit(struct qemu_laio_state *s, bool enqueue)
 {
     int ret, i = 0;
     int len = s->io_q.idx;
+    int j = 0;
 
-    do {
-        ret = io_submit(s->ctx, len, s->io_q.iocbs);
-    } while (i++ < 3 && ret == -EAGAIN);
+    if (!len) {
+        return 0;
+    }
+
+    ret = io_submit(s->ctx, len, s->io_q.iocbs);
+    if (ret == -EAGAIN) { /* retry in following completion cb */
+        return 0;
+    } else if (ret < 0) {
+        if (enqueue) {
+            return ret;
+        }
 
-    /* empty io queue */
-    s->io_q.idx = 0;
+        /* in non-queue path, all IOs have to be completed */
+        abort_queue(s);
+        ret = len;
+    } else if (ret == 0) {
+        goto out;
+    }
 
-    if (ret < 0) {
-        i = 0;
-    } else {
-        i = ret;
+    for (i = ret; i < len; i++) {
+        s->io_q.iocbs[j++] = s->io_q.iocbs[i];
     }
 
-    for (; i < len; i++) {
-        struct qemu_laiocb *laiocb =
-            container_of(s->io_q.iocbs[i], struct qemu_laiocb, iocb);
+ out:
+    /*
+     * update io queue, for partial completion, retry will be
+     * started automatically in following completion cb.
+     */
+    s->io_q.idx -= ret;
 
-        laiocb->ret = (ret < 0) ? ret : -EIO;
-        qemu_laio_process_completion(s, laiocb);
-    }
     return ret;
 }
 
-static void ioq_enqueue(struct qemu_laio_state *s, struct iocb *iocb)
+static void ioq_submit_retry(void *opaque)
+{
+    struct qemu_laio_state *s = opaque;
+    ioq_submit(s, false);
+}
+
+static int ioq_enqueue(struct qemu_laio_state *s, struct iocb *iocb)
 {
     unsigned int idx = s->io_q.idx;
 
+    if (unlikely(idx == s->io_q.size)) {
+        return -1;
+    }
+
     s->io_q.iocbs[idx++] = iocb;
     s->io_q.idx = idx;
 
-    /* submit immediately if queue is full */
-    if (idx == s->io_q.size) {
-        ioq_submit(s);
+    /* submit immediately if queue depth is above 2/3 */
+    if (idx > s->io_q.size * 2 / 3) {
+        return ioq_submit(s, true);
     }
+
+    return 0;
 }
 
 void laio_io_plug(BlockDriverState *bs, void *aio_ctx)
@@ -214,7 +265,7 @@ int laio_io_unplug(BlockDriverState *bs, void *aio_ctx, bool unplug)
     }
 
     if (s->io_q.idx > 0) {
-        ret = ioq_submit(s);
+        ret = ioq_submit(s, false);
     }
 
     return ret;
@@ -258,7 +309,9 @@ BlockDriverAIOCB *laio_submit(BlockDriverState *bs, void *aio_ctx, int fd,
             goto out_free_aiocb;
         }
     } else {
-        ioq_enqueue(s, iocbs);
+        if (ioq_enqueue(s, iocbs) < 0) {
+            goto out_free_aiocb;
+        }
     }
     return &laiocb->common;
 
@@ -272,6 +325,7 @@ void laio_detach_aio_context(void *s_, AioContext *old_context)
     struct qemu_laio_state *s = s_;
 
     aio_set_event_notifier(old_context, &s->e, NULL);
+    qemu_bh_delete(s->io_q.retry);
 }
 
 void laio_attach_aio_context(void *s_, AioContext *new_context)
@@ -279,6 +333,7 @@ void laio_attach_aio_context(void *s_, AioContext *new_context)
     struct qemu_laio_state *s = s_;
 
     aio_set_event_notifier(new_context, &s->e, qemu_laio_completion_cb);
+    s->io_q.retry = aio_bh_new(new_context, ioq_submit_retry, s);
 }
 
 void *laio_init(void)
-- 
1.7.9.5

[Qemu-devel] [PATCH 2/4] linux-aio: handling -EAGAIN for !s->io_q.plugged case

[Ming Lei]

Previously -EAGAIN is simply ignored for !s->io_q.plugged case,
and sometimes it is easy to cause -EIO to VM, such as NVME device.

This patch handles -EAGAIN by io queue for !s->io_q.plugged case,
and it will be retried in following aio completion cb.

Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Ming Lei <ming.lei@canonical.com>
---
 block/linux-aio.c |   22 +++++++++++++++++++++-
 1 file changed, 21 insertions(+), 1 deletion(-)

diff --git a/block/linux-aio.c b/block/linux-aio.c
index 4cdf507..0e21f76 100644
--- a/block/linux-aio.c
+++ b/block/linux-aio.c
@@ -238,6 +238,11 @@ static int ioq_enqueue(struct qemu_laio_state *s, struct iocb *iocb)
     s->io_q.iocbs[idx++] = iocb;
     s->io_q.idx = idx;
 
+    /* don't submit until next completion for -EAGAIN of non plug case */
+    if (unlikely(!s->io_q.plugged)) {
+        return 0;
+    }
+
     /* submit immediately if queue depth is above 2/3 */
     if (idx > s->io_q.size * 2 / 3) {
         return ioq_submit(s, true);
@@ -305,10 +310,25 @@ BlockDriverAIOCB *laio_submit(BlockDriverState *bs, void *aio_ctx, int fd,
     io_set_eventfd(&laiocb->iocb, event_notifier_get_fd(&s->e));
 
     if (!s->io_q.plugged) {
-        if (io_submit(s->ctx, 1, &iocbs) < 0) {
+        int ret;
+
+        if (!s->io_q.idx) {
+            ret = io_submit(s->ctx, 1, &iocbs);
+        } else {
+            ret = -EAGAIN;
+        }
+        /*
+         * Switch to queue mode until -EAGAIN is handled, we suppose
+         * there is always uncompleted I/O, so try to enqueue it first,
+         * and will be submitted again in following aio completion cb.
+         */
+        if (ret == -EAGAIN) {
+            goto enqueue;
+        } else if (ret < 0) {
             goto out_free_aiocb;
         }
     } else {
+ enqueue:
         if (ioq_enqueue(s, iocbs) < 0) {
             goto out_free_aiocb;
         }
-- 
1.7.9.5

[Qemu-devel] [PATCH 3/4] linux-aio: increase max event to 256

[Ming Lei]

This patch increases max event to 256 for the comming
virtio-blk multi virtqueue support.

Signed-off-by: Ming Lei <ming.lei@canonical.com>
---
 block/linux-aio.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/block/linux-aio.c b/block/linux-aio.c
index 0e21f76..bf94ae9 100644
--- a/block/linux-aio.c
+++ b/block/linux-aio.c
@@ -23,7 +23,7 @@
  *      than this we will get EAGAIN from io_submit which is communicated to
  *      the guest as an I/O error.
  */
-#define MAX_EVENTS 128
+#define MAX_EVENTS 256
 
 #define MAX_QUEUED_IO  128
 
-- 
1.7.9.5

[Qemu-devel] [PATCH 4/4] linux-aio: remove 'node' from 'struct qemu_laiocb'

[Ming Lei]

No one uses the 'node' field any more, so remove it
from 'struct qemu_laiocb', and this can save 16byte
for the struct on 64bit arch.

Signed-off-by: Ming Lei <ming.lei@canonical.com>
---
 block/linux-aio.c |    1 -
 1 file changed, 1 deletion(-)

diff --git a/block/linux-aio.c b/block/linux-aio.c
index bf94ae9..da50ea5 100644
--- a/block/linux-aio.c
+++ b/block/linux-aio.c
@@ -35,7 +35,6 @@ struct qemu_laiocb {
     size_t nbytes;
     QEMUIOVector *qiov;
     bool is_read;
-    QLIST_ENTRY(qemu_laiocb) node;
 };
 
 /*
-- 
1.7.9.5

Re: [Qemu-devel] linux-aio: fix batch submission

[Ming Lei]

Hi Guys,

I appreciate much if anyone can review or just take this patchset, which
is needed by my following patches.

Thanks,

On Thu, Aug 14, 2014 at 5:41 PM, Ming Lei <ming.lei@canonical.com> wrote:
> Hi,
>
> The 1st patch fixes batch submission.
>
> The 2nd one fixes -EAGAIN for non-batch case.
>
> The 3rd one increase max event to 256 for supporting the comming
> multi virt-queue.
>
> The 4th one is a cleanup.
>
> This patchset is splitted from previous patchset(dataplane: optimization
> and multi virtqueue support), as suggested by Stefan.
>
> Thanks,
> --
> Ming Lei
>

Re: [Qemu-devel] [PATCH 3/4] linux-aio: increase max event to 256

[Benoît Canet]

The Thursday 14 Aug 2014 à 17:41:43 (+0800), Ming Lei wrote :
> This patch increases max event to 256 for the comming

s/comming/coming/

See http://en.wiktionary.org/wiki/comming and
https://www.wordnik.com/words/comming

> virtio-blk multi virtqueue support.
> 
> Signed-off-by: Ming Lei <ming.lei@canonical.com>
> ---
>  block/linux-aio.c |    2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/block/linux-aio.c b/block/linux-aio.c
> index 0e21f76..bf94ae9 100644
> --- a/block/linux-aio.c
> +++ b/block/linux-aio.c
> @@ -23,7 +23,7 @@
>   *      than this we will get EAGAIN from io_submit which is communicated to
>   *      the guest as an I/O error.
>   */
> -#define MAX_EVENTS 128
> +#define MAX_EVENTS 256
>  
>  #define MAX_QUEUED_IO  128
>  
> -- 
> 1.7.9.5
> 
> 

Re: [Qemu-devel] [PATCH 1/4] linux-aio: fix submit aio as a batch

[Benoît Canet]

The Thursday 14 Aug 2014 à 17:41:41 (+0800), Ming Lei wrote :
> In the enqueue path, we can't complete request, otherwise
> "Co-routine re-entered recursively" may be caused, so this
> patch fixes the issue with below ideas:

s/with below ideas/with the following ideas/g

> 
> 	- for -EAGAIN or partial completion, retry the submision by

s/submision/submission/

> 	schedule an BH in following completion cb

s/schedule an/sheduling a/

> 	- for part of completion, also update the io queue
> 	- for other failure, return the failure if in enqueue path,
> 	otherwise, abort all queued I/O
> 
> Signed-off-by: Ming Lei <ming.lei@canonical.com>
> ---
>  block/linux-aio.c |   99 +++++++++++++++++++++++++++++++++++++++++------------
>  1 file changed, 77 insertions(+), 22 deletions(-)
> 
> diff --git a/block/linux-aio.c b/block/linux-aio.c
> index 7ac7e8c..4cdf507 100644
> --- a/block/linux-aio.c
> +++ b/block/linux-aio.c
> @@ -38,11 +38,19 @@ struct qemu_laiocb {
>      QLIST_ENTRY(qemu_laiocb) node;
>  };
>  
> +/*
> + * TODO: support to batch I/O from multiple bs in one same
> + * AIO context, one important use case is multi-lun scsi,
> + * so in future the IO queue should be per AIO context.
> + */
>  typedef struct {

In QEMU we typically write the name twice in these kind of declarations:

typedef struct LaioQueue {
... stuff ...
} LaioQueue;

>      struct iocb *iocbs[MAX_QUEUED_IO];
>      int plugged;

Are plugged values either 0 and 1 ?
If so it should be "bool plugged;"

>      unsigned int size;
>      unsigned int idx;

See:

benoit@Laure:~/code/qemu$ git grep "unsigned int"|wc
   2283   14038  154201
benoit@Laure:~/code/qemu$ git grep "uint32"|wc
  12535   63129  810822

Maybe you could use the most popular type.

> +
> +    /* handle -EAGAIN and partial completion */
> +    QEMUBH *retry;
>  } LaioQueue;
>  
>  struct qemu_laio_state {
> @@ -86,6 +94,12 @@ static void qemu_laio_process_completion(struct qemu_laio_state *s,
>      qemu_aio_release(laiocb);
>  }
>  
> +static void qemu_laio_start_retry(struct qemu_laio_state *s)
> +{


> +    if (s->io_q.idx)
> +        qemu_bh_schedule(s->io_q.retry);

In QEMU this test is writen like this:

    if (s->io_q.idx) {
        qemu_bh_schedule(s->io_q.retry);
    }

I suggest you ran ./scripts/checkpatch.pl on your series before submitting it.


> +}
> +
>  static void qemu_laio_completion_cb(EventNotifier *e)
>  {
>      struct qemu_laio_state *s = container_of(e, struct qemu_laio_state, e);
> @@ -108,6 +122,7 @@ static void qemu_laio_completion_cb(EventNotifier *e)
>              qemu_laio_process_completion(s, laiocb);
>          }
>      }
> +    qemu_laio_start_retry(s);
>  }
>  
>  static void laio_cancel(BlockDriverAIOCB *blockacb)
> @@ -127,6 +142,7 @@ static void laio_cancel(BlockDriverAIOCB *blockacb)
>      ret = io_cancel(laiocb->ctx->ctx, &laiocb->iocb, &event);
>      if (ret == 0) {
>          laiocb->ret = -ECANCELED;
> +        qemu_laio_start_retry(laiocb->ctx);
>          return;
>      }
>  
> @@ -154,45 +170,80 @@ static void ioq_init(LaioQueue *io_q)
>      io_q->plugged = 0;
>  }
>  
> -static int ioq_submit(struct qemu_laio_state *s)
> +static void abort_queue(struct qemu_laio_state *s)
> +{
> +    int i;
> +    for (i = 0; i < s->io_q.idx; i++) {
> +        struct qemu_laiocb *laiocb = container_of(s->io_q.iocbs[i],
> +                                                  struct qemu_laiocb,
> +                                                  iocb);
> +        laiocb->ret = -EIO;
> +        qemu_laio_process_completion(s, laiocb);
> +    }
> +}
> +
> +static int ioq_submit(struct qemu_laio_state *s, bool enqueue)
>  {
>      int ret, i = 0;
>      int len = s->io_q.idx;
> +    int j = 0;
>  
> -    do {
> -        ret = io_submit(s->ctx, len, s->io_q.iocbs);
> -    } while (i++ < 3 && ret == -EAGAIN);
> +    if (!len) {
> +        return 0;
> +    }
> +
> +    ret = io_submit(s->ctx, len, s->io_q.iocbs);
> +    if (ret == -EAGAIN) { /* retry in following completion cb */
> +        return 0;
> +    } else if (ret < 0) {
> +        if (enqueue) {
> +            return ret;
> +        }
>  
> -    /* empty io queue */
> -    s->io_q.idx = 0;
> +        /* in non-queue path, all IOs have to be completed */
> +        abort_queue(s);
> +        ret = len;
> +    } else if (ret == 0) {
> +        goto out;
> +    }
>  
> -    if (ret < 0) {
> -        i = 0;
> -    } else {
> -        i = ret;
> +    for (i = ret; i < len; i++) {
> +        s->io_q.iocbs[j++] = s->io_q.iocbs[i];
>      }
>  
> -    for (; i < len; i++) {
> -        struct qemu_laiocb *laiocb =
> -            container_of(s->io_q.iocbs[i], struct qemu_laiocb, iocb);
> + out:
> +    /*
> +     * update io queue, for partial completion, retry will be
> +     * started automatically in following completion cb.
> +     */
> +    s->io_q.idx -= ret;
>  
> -        laiocb->ret = (ret < 0) ? ret : -EIO;
> -        qemu_laio_process_completion(s, laiocb);
> -    }
>      return ret;
>  }
>  
> -static void ioq_enqueue(struct qemu_laio_state *s, struct iocb *iocb)
> +static void ioq_submit_retry(void *opaque)
> +{
> +    struct qemu_laio_state *s = opaque;
> +    ioq_submit(s, false);
> +}
> +
> +static int ioq_enqueue(struct qemu_laio_state *s, struct iocb *iocb)
>  {
>      unsigned int idx = s->io_q.idx;
>  
> +    if (unlikely(idx == s->io_q.size)) {
> +        return -1;
> +    }
> +
>      s->io_q.iocbs[idx++] = iocb;
>      s->io_q.idx = idx;
>  
> -    /* submit immediately if queue is full */
> -    if (idx == s->io_q.size) {
> -        ioq_submit(s);
> +    /* submit immediately if queue depth is above 2/3 */
> +    if (idx > s->io_q.size * 2 / 3) {
> +        return ioq_submit(s, true);
>      }
> +
> +    return 0;
>  }
>  
>  void laio_io_plug(BlockDriverState *bs, void *aio_ctx)
> @@ -214,7 +265,7 @@ int laio_io_unplug(BlockDriverState *bs, void *aio_ctx, bool unplug)
>      }
>  
>      if (s->io_q.idx > 0) {
> -        ret = ioq_submit(s);
> +        ret = ioq_submit(s, false);
>      }
>  
>      return ret;
> @@ -258,7 +309,9 @@ BlockDriverAIOCB *laio_submit(BlockDriverState *bs, void *aio_ctx, int fd,
>              goto out_free_aiocb;
>          }
>      } else {
> -        ioq_enqueue(s, iocbs);
> +        if (ioq_enqueue(s, iocbs) < 0) {
> +            goto out_free_aiocb;
> +        }
>      }
>      return &laiocb->common;
>  
> @@ -272,6 +325,7 @@ void laio_detach_aio_context(void *s_, AioContext *old_context)
>      struct qemu_laio_state *s = s_;
>  
>      aio_set_event_notifier(old_context, &s->e, NULL);
> +    qemu_bh_delete(s->io_q.retry);
>  }
>  
>  void laio_attach_aio_context(void *s_, AioContext *new_context)
> @@ -279,6 +333,7 @@ void laio_attach_aio_context(void *s_, AioContext *new_context)
>      struct qemu_laio_state *s = s_;
>  
>      aio_set_event_notifier(new_context, &s->e, qemu_laio_completion_cb);
> +    s->io_q.retry = aio_bh_new(new_context, ioq_submit_retry, s);
>  }
>  
>  void *laio_init(void)
> -- 
> 1.7.9.5
> 
> 

Re: [Qemu-devel] [PATCH 1/4] linux-aio: fix submit aio as a batch

[Ming Lei]

On Thu, Sep 4, 2014 at 10:59 PM, Benoît Canet <benoit.canet@irqsave.net> wrote:
> The Thursday 14 Aug 2014 à 17:41:41 (+0800), Ming Lei wrote :
>> In the enqueue path, we can't complete request, otherwise
>> "Co-routine re-entered recursively" may be caused, so this
>> patch fixes the issue with below ideas:
>
> s/with below ideas/with the following ideas/g
>
>>
>>       - for -EAGAIN or partial completion, retry the submision by
>
> s/submision/submission/
>
>>       schedule an BH in following completion cb
>
> s/schedule an/sheduling a/
>
>>       - for part of completion, also update the io queue
>>       - for other failure, return the failure if in enqueue path,
>>       otherwise, abort all queued I/O
>>
>> Signed-off-by: Ming Lei <ming.lei@canonical.com>
>> ---
>>  block/linux-aio.c |   99 +++++++++++++++++++++++++++++++++++++++++------------
>>  1 file changed, 77 insertions(+), 22 deletions(-)
>>
>> diff --git a/block/linux-aio.c b/block/linux-aio.c
>> index 7ac7e8c..4cdf507 100644
>> --- a/block/linux-aio.c
>> +++ b/block/linux-aio.c
>> @@ -38,11 +38,19 @@ struct qemu_laiocb {
>>      QLIST_ENTRY(qemu_laiocb) node;
>>  };
>>
>> +/*
>> + * TODO: support to batch I/O from multiple bs in one same
>> + * AIO context, one important use case is multi-lun scsi,
>> + * so in future the IO queue should be per AIO context.
>> + */
>>  typedef struct {
>
> In QEMU we typically write the name twice in these kind of declarations:
>
> typedef struct LaioQueue {
> ... stuff ...
> } LaioQueue;
>
>>      struct iocb *iocbs[MAX_QUEUED_IO];
>>      int plugged;
>
> Are plugged values either 0 and 1 ?
> If so it should be "bool plugged;"

It is actually a reference counter.

>
>>      unsigned int size;
>>      unsigned int idx;
>
> See:
>
> benoit@Laure:~/code/qemu$ git grep "unsigned int"|wc
>    2283   14038  154201
> benoit@Laure:~/code/qemu$ git grep "uint32"|wc
>   12535   63129  810822
>
> Maybe you could use the most popular type.
>
>> +
>> +    /* handle -EAGAIN and partial completion */
>> +    QEMUBH *retry;
>>  } LaioQueue;
>>
>>  struct qemu_laio_state {
>> @@ -86,6 +94,12 @@ static void qemu_laio_process_completion(struct qemu_laio_state *s,
>>      qemu_aio_release(laiocb);
>>  }
>>
>> +static void qemu_laio_start_retry(struct qemu_laio_state *s)
>> +{
>
>
>> +    if (s->io_q.idx)
>> +        qemu_bh_schedule(s->io_q.retry);
>
> In QEMU this test is writen like this:
>
>     if (s->io_q.idx) {
>         qemu_bh_schedule(s->io_q.retry);
>     }
>
> I suggest you ran ./scripts/checkpatch.pl on your series before submitting it.

I don't know why the blow pre-commit hook didn't complain that:

[tom@qemu]$cat .git/hooks/pre-commit
#!/bin/bash
exec git diff --cached | scripts/checkpatch.pl --no-signoff -q -

>
>
>> +}
>> +
>>  static void qemu_laio_completion_cb(EventNotifier *e)
>>  {
>>      struct qemu_laio_state *s = container_of(e, struct qemu_laio_state, e);
>> @@ -108,6 +122,7 @@ static void qemu_laio_completion_cb(EventNotifier *e)
>>              qemu_laio_process_completion(s, laiocb);
>>          }
>>      }
>> +    qemu_laio_start_retry(s);
>>  }
>>
>>  static void laio_cancel(BlockDriverAIOCB *blockacb)
>> @@ -127,6 +142,7 @@ static void laio_cancel(BlockDriverAIOCB *blockacb)
>>      ret = io_cancel(laiocb->ctx->ctx, &laiocb->iocb, &event);
>>      if (ret == 0) {
>>          laiocb->ret = -ECANCELED;
>> +        qemu_laio_start_retry(laiocb->ctx);
>>          return;
>>      }
>>
>> @@ -154,45 +170,80 @@ static void ioq_init(LaioQueue *io_q)
>>      io_q->plugged = 0;
>>  }
>>
>> -static int ioq_submit(struct qemu_laio_state *s)
>> +static void abort_queue(struct qemu_laio_state *s)
>> +{
>> +    int i;
>> +    for (i = 0; i < s->io_q.idx; i++) {
>> +        struct qemu_laiocb *laiocb = container_of(s->io_q.iocbs[i],
>> +                                                  struct qemu_laiocb,
>> +                                                  iocb);
>> +        laiocb->ret = -EIO;
>> +        qemu_laio_process_completion(s, laiocb);
>> +    }
>> +}
>> +
>> +static int ioq_submit(struct qemu_laio_state *s, bool enqueue)
>>  {
>>      int ret, i = 0;
>>      int len = s->io_q.idx;
>> +    int j = 0;
>>
>> -    do {
>> -        ret = io_submit(s->ctx, len, s->io_q.iocbs);
>> -    } while (i++ < 3 && ret == -EAGAIN);
>> +    if (!len) {
>> +        return 0;
>> +    }
>> +
>> +    ret = io_submit(s->ctx, len, s->io_q.iocbs);
>> +    if (ret == -EAGAIN) { /* retry in following completion cb */
>> +        return 0;
>> +    } else if (ret < 0) {
>> +        if (enqueue) {
>> +            return ret;
>> +        }
>>
>> -    /* empty io queue */
>> -    s->io_q.idx = 0;
>> +        /* in non-queue path, all IOs have to be completed */
>> +        abort_queue(s);
>> +        ret = len;
>> +    } else if (ret == 0) {
>> +        goto out;
>> +    }
>>
>> -    if (ret < 0) {
>> -        i = 0;
>> -    } else {
>> -        i = ret;
>> +    for (i = ret; i < len; i++) {
>> +        s->io_q.iocbs[j++] = s->io_q.iocbs[i];
>>      }
>>
>> -    for (; i < len; i++) {
>> -        struct qemu_laiocb *laiocb =
>> -            container_of(s->io_q.iocbs[i], struct qemu_laiocb, iocb);
>> + out:
>> +    /*
>> +     * update io queue, for partial completion, retry will be
>> +     * started automatically in following completion cb.
>> +     */
>> +    s->io_q.idx -= ret;
>>
>> -        laiocb->ret = (ret < 0) ? ret : -EIO;
>> -        qemu_laio_process_completion(s, laiocb);
>> -    }
>>      return ret;
>>  }
>>
>> -static void ioq_enqueue(struct qemu_laio_state *s, struct iocb *iocb)
>> +static void ioq_submit_retry(void *opaque)
>> +{
>> +    struct qemu_laio_state *s = opaque;
>> +    ioq_submit(s, false);
>> +}
>> +
>> +static int ioq_enqueue(struct qemu_laio_state *s, struct iocb *iocb)
>>  {
>>      unsigned int idx = s->io_q.idx;
>>
>> +    if (unlikely(idx == s->io_q.size)) {
>> +        return -1;
>> +    }
>> +
>>      s->io_q.iocbs[idx++] = iocb;
>>      s->io_q.idx = idx;
>>
>> -    /* submit immediately if queue is full */
>> -    if (idx == s->io_q.size) {
>> -        ioq_submit(s);
>> +    /* submit immediately if queue depth is above 2/3 */
>> +    if (idx > s->io_q.size * 2 / 3) {
>> +        return ioq_submit(s, true);
>>      }
>> +
>> +    return 0;
>>  }
>>
>>  void laio_io_plug(BlockDriverState *bs, void *aio_ctx)
>> @@ -214,7 +265,7 @@ int laio_io_unplug(BlockDriverState *bs, void *aio_ctx, bool unplug)
>>      }
>>
>>      if (s->io_q.idx > 0) {
>> -        ret = ioq_submit(s);
>> +        ret = ioq_submit(s, false);
>>      }
>>
>>      return ret;
>> @@ -258,7 +309,9 @@ BlockDriverAIOCB *laio_submit(BlockDriverState *bs, void *aio_ctx, int fd,
>>              goto out_free_aiocb;
>>          }
>>      } else {
>> -        ioq_enqueue(s, iocbs);
>> +        if (ioq_enqueue(s, iocbs) < 0) {
>> +            goto out_free_aiocb;
>> +        }
>>      }
>>      return &laiocb->common;
>>
>> @@ -272,6 +325,7 @@ void laio_detach_aio_context(void *s_, AioContext *old_context)
>>      struct qemu_laio_state *s = s_;
>>
>>      aio_set_event_notifier(old_context, &s->e, NULL);
>> +    qemu_bh_delete(s->io_q.retry);
>>  }
>>
>>  void laio_attach_aio_context(void *s_, AioContext *new_context)
>> @@ -279,6 +333,7 @@ void laio_attach_aio_context(void *s_, AioContext *new_context)
>>      struct qemu_laio_state *s = s_;
>>
>>      aio_set_event_notifier(new_context, &s->e, qemu_laio_completion_cb);
>> +    s->io_q.retry = aio_bh_new(new_context, ioq_submit_retry, s);
>>  }
>>
>>  void *laio_init(void)
>> --
>> 1.7.9.5
>>
>>