The first rcu_read_lock/unlock() is expensive.  Nested calls are cheap.

This optimization increases IOPS from 73k to 162k with a Linux guest
that has 2 virtio-blk,num-queues=1 and 99 virtio-blk,num-queues=32
devices.

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

diff --git a/util/aio-posix.c b/util/aio-posix.c
index a4977f538e..f67f5b34e9 100644
--- a/util/aio-posix.c
+++ b/util/aio-posix.c
@@ -15,6 +15,7 @@
 
 #include "qemu/osdep.h"
 #include "block/block.h"
+#include "qemu/rcu.h"
 #include "qemu/rcu_queue.h"
 #include "qemu/sockets.h"
 #include "qemu/cutils.h"
@@ -514,6 +515,16 @@ static bool run_poll_handlers_once(AioContext *ctx, int64_t *timeout)
     bool progress = false;
     AioHandler *node;
 
+    /*
+     * Optimization: ->io_poll() handlers often contain RCU read critical
+     * sections and we therefore see many rcu_read_lock() -> rcu_read_unlock()
+     * -> rcu_read_lock() -> ... sequences with expensive memory
+     * synchronization primitives.  Make the entire polling loop an RCU
+     * critical section because nested rcu_read_lock()/rcu_read_unlock() calls
+     * are cheap.
+     */
+    RCU_READ_LOCK_GUARD();
+
     QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) {
         if (!node->deleted && node->io_poll &&
             aio_node_check(ctx, node->is_external) &&
-- 
2.24.1

On 18/02/20 19:27, Stefan Hajnoczi wrote:
> The first rcu_read_lock/unlock() is expensive.  Nested calls are cheap.
> 
> This optimization increases IOPS from 73k to 162k with a Linux guest
> that has 2 virtio-blk,num-queues=1 and 99 virtio-blk,num-queues=32
> devices.
> 
> Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
> ---
>  util/aio-posix.c | 11 +++++++++++
>  1 file changed, 11 insertions(+)
> 
> diff --git a/util/aio-posix.c b/util/aio-posix.c
> index a4977f538e..f67f5b34e9 100644
> --- a/util/aio-posix.c
> +++ b/util/aio-posix.c
> @@ -15,6 +15,7 @@
>  
>  #include "qemu/osdep.h"
>  #include "block/block.h"
> +#include "qemu/rcu.h"
>  #include "qemu/rcu_queue.h"
>  #include "qemu/sockets.h"
>  #include "qemu/cutils.h"
> @@ -514,6 +515,16 @@ static bool run_poll_handlers_once(AioContext *ctx, int64_t *timeout)
>      bool progress = false;
>      AioHandler *node;
>  
> +    /*
> +     * Optimization: ->io_poll() handlers often contain RCU read critical
> +     * sections and we therefore see many rcu_read_lock() -> rcu_read_unlock()
> +     * -> rcu_read_lock() -> ... sequences with expensive memory
> +     * synchronization primitives.  Make the entire polling loop an RCU
> +     * critical section because nested rcu_read_lock()/rcu_read_unlock() calls
> +     * are cheap.
> +     */
> +    RCU_READ_LOCK_GUARD();
> +
>      QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) {
>          if (!node->deleted && node->io_poll &&
>              aio_node_check(ctx, node->is_external) &&
> 

Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>

On Tue, Feb 18, 2020 at 06:27:08PM +0000, Stefan Hajnoczi wrote:
> The first rcu_read_lock/unlock() is expensive.  Nested calls are cheap.
> 
> This optimization increases IOPS from 73k to 162k with a Linux guest
> that has 2 virtio-blk,num-queues=1 and 99 virtio-blk,num-queues=32
> devices.
> 
> Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
> ---
>  util/aio-posix.c | 11 +++++++++++
>  1 file changed, 11 insertions(+)

Thanks, applied to my block tree:
https://github.com/stefanha/qemu/commits/block

Stefan