Skip to contentSkip to conversation

thread-pool: add missing qemu_bh_cancel in completion function

PATCH Applied Peter Lieven 3 participants 4 messages Mar 16, 2017 – Mar 17, 2017

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 thread-pool: add missing qemu_bh_cancel in completion function

commit 3c80ca15 fixed a deadlock scenarion with nested aio_poll invocations.

However, the rescheduling of the completion BH introcuded unnecessary spinning in the main-loop. On very fast file backends this can even lead to the "WARNING: I/O thread spun for 1000 iterations" message popping up.

Callgrind reports about 3-4% less instructions with this patch running qemu-img bench on a ramdisk based VMDK file.

Fixes 3c80ca158c96ff902a30883a8933e755988948b1
---
 util/thread-pool.c | 7 +++++++
 1 file changed, 7 insertions(+)
util/thread-pool.c +7 -0
index ce6cd30..610646d 100644
@@ -188,6 +188,13 @@ restart:
             aio_context_release(pool->ctx);
             elem->common.cb(elem->common.opaque, elem->ret);
             aio_context_acquire(pool->ctx);
+
+            /* We can safely cancel the completion_bh here regardless of someone
+             * else having scheduled it meanwhile because we reenter the
+             * completion function anyway (goto restart).
+             */
+            qemu_bh_cancel(pool->completion_bh);
+
             qemu_aio_unref(elem);
             goto restart;
         } else {
Signature · 1 line
1.9.1
#1 lore raw
Quoted 25 lines · Peter Lieven

commit 3c80ca15 fixed a deadlock scenarion with nested aio_poll invocations.

However, the rescheduling of the completion BH introcuded unnecessary spinning in the main-loop. On very fast file backends this can even lead to the "WARNING: I/O thread spun for 1000 iterations" message popping up.

Callgrind reports about 3-4% less instructions with this patch running qemu-img bench on a ramdisk based VMDK file.

Fixes 3c80ca158c96ff902a30883a8933e755988948b1
---
 util/thread-pool.c | 7 +++++++
 1 file changed, 7 insertions(+)
util/thread-pool.c +0 -0
index ce6cd30..610646d 100644
@@ -188,6 +188,13 @@ restart:
             aio_context_release(pool->ctx);
             elem->common.cb(elem->common.opaque, elem->ret);
             aio_context_acquire(pool->ctx);

util/thread-pool.c restart: · line 200

Quoted 10 lines
util/thread-pool.c +7 -0
+
+            /* We can safely cancel the completion_bh here regardless of someone
+             * else having scheduled it meanwhile because we reenter the
+             * completion function anyway (goto restart).
+             */
+            qemu_bh_cancel(pool->completion_bh);
+
             qemu_aio_unref(elem);
             goto restart;
         } else {

Right, this is the same thing that block/linux-aio.c does.

Paolo

#2 lore raw
Quoted 35 lines

Am 16.03.2017 um 17:18 schrieb Paolo Bonzini <[email protected]>:

Quoted 29 lines · Peter Lieven

commit 3c80ca15 fixed a deadlock scenarion with nested aio_poll invocations.

However, the rescheduling of the completion BH introcuded unnecessary spinning in the main-loop. On very fast file backends this can even lead to the "WARNING: I/O thread spun for 1000 iterations" message popping up.

Callgrind reports about 3-4% less instructions with this patch running qemu-img bench on a ramdisk based VMDK file.

Fixes 3c80ca158c96ff902a30883a8933e755988948b1

---
util/thread-pool.c | 7 +++++++

1 file changed, 7 insertions(+)
util/thread-pool.c +5 -0
index ce6cd30..610646d 100644
@@ -188,6 +188,13 @@ restart:
            aio_context_release(pool->ctx);
            elem->common.cb(elem->common.opaque, elem->ret);
            aio_context_acquire(pool->ctx);
+
+            /* We can safely cancel the completion_bh here regardless of someone
+             * else having scheduled it meanwhile because we reenter the
+             * completion function anyway (goto restart).
+             */
Quoted 8 lines

util/thread-pool.c restart: · line 200

Quoted 6 lines
util/thread-pool.c +2 -0
+            qemu_bh_cancel(pool->completion_bh);
+
            qemu_aio_unref(elem);
            goto restart;
        } else {

Right, this is the same thing that block/linux-aio.c does.

Okay, so you also think its safe to do this? As far as I have seen you have been working a lot on the aio code recently.

Thanks,
Peter

Applied
#3 lore raw
In reply to Peter Lieven

Am 16.03.2017 um 17:02 hat Peter Lieven geschrieben:

Quoted 6 lines

commit 3c80ca15 fixed a deadlock scenarion with nested aio_poll invocations.

However, the rescheduling of the completion BH introcuded unnecessary spinning in the main-loop. On very fast file backends this can even lead to the "WARNING: I/O thread spun for 1000 iterations" message popping up.

Quoted 6 lines

Callgrind reports about 3-4% less instructions with this patch running qemu-img bench on a ramdisk based VMDK file.

Fixes 3c80ca158c96ff902a30883a8933e755988948b1

Thanks, applied to the block branch.

Kevin