It is valid for an aio_set_fd_handler to happen concurrently with
aio_poll.  In that case, poll_disable_cnt can change under the heels
of aio_poll, and the assertion on poll_disable_cnt can fail in
run_poll_handlers.

Therefore, this patch simply checks the counter on every polling
iteration.  There are no particular needs for ordering, since the
polling loop is terminated anyway by aio_notify at the end of
aio_set_fd_handler.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 util/aio-posix.c | 26 +++++++++++++++-----------
 1 file changed, 15 insertions(+), 11 deletions(-)

diff --git a/util/aio-posix.c b/util/aio-posix.c
index 131ba6b4a8..5c29380575 100644
--- a/util/aio-posix.c
+++ b/util/aio-posix.c
@@ -211,6 +211,7 @@ void aio_set_fd_handler(AioContext *ctx,
     AioHandler *node;
     bool is_new = false;
     bool deleted = false;
+    int poll_disable_change;
 
     qemu_lockcnt_lock(&ctx->list_lock);
 
@@ -244,11 +245,9 @@ void aio_set_fd_handler(AioContext *ctx,
             QLIST_REMOVE(node, node);
             deleted = true;
         }
-
-        if (!node->io_poll) {
-            ctx->poll_disable_cnt--;
-        }
+        poll_disable_change = -!node->io_poll;
     } else {
+        poll_disable_change = !io_poll - (node && !node->io_poll);
         if (node == NULL) {
             /* Alloc and insert if it's not already there */
             node = g_new0(AioHandler, 1);
@@ -257,10 +256,6 @@ void aio_set_fd_handler(AioContext *ctx,
 
             g_source_add_poll(&ctx->source, &node->pfd);
             is_new = true;
-
-            ctx->poll_disable_cnt += !io_poll;
-        } else {
-            ctx->poll_disable_cnt += !io_poll - !node->io_poll;
         }
 
         /* Update handler with latest information */
@@ -274,6 +269,15 @@ void aio_set_fd_handler(AioContext *ctx,
         node->pfd.events |= (io_write ? G_IO_OUT | G_IO_ERR : 0);
     }
 
+    /* No need to order poll_disable_cnt writes against other updates;
+     * the counter is only used to avoid wasting time and latency on
+     * iterated polling when the system call will be ultimately necessary.
+     * Changing handlers is a rare event, and a little wasted polling until
+     * the aio_notify below is not an issue.
+     */
+    atomic_set(&ctx->poll_disable_cnt,
+               atomic_read(&ctx->poll_disable_cnt) + poll_disable_change);
+
     aio_epoll_update(ctx, node, is_new);
     qemu_lockcnt_unlock(&ctx->list_lock);
     aio_notify(ctx);
@@ -525,7 +529,6 @@ static bool run_poll_handlers(AioContext *ctx, int64_t max_ns)
 
     assert(ctx->notify_me);
     assert(qemu_lockcnt_count(&ctx->list_lock) > 0);
-    assert(ctx->poll_disable_cnt == 0);
 
     trace_run_poll_handlers_begin(ctx, max_ns);
 
@@ -533,7 +536,8 @@ static bool run_poll_handlers(AioContext *ctx, int64_t max_ns)
 
     do {
         progress = run_poll_handlers_once(ctx);
-    } while (!progress && qemu_clock_get_ns(QEMU_CLOCK_REALTIME) < end_time);
+    } while (!progress && qemu_clock_get_ns(QEMU_CLOCK_REALTIME) < end_time
+             && !atomic_read(&ctx->poll_disable_cnt));
 
     trace_run_poll_handlers_end(ctx, progress);
 
@@ -552,7 +556,7 @@ static bool run_poll_handlers(AioContext *ctx, int64_t max_ns)
  */
 static bool try_poll_mode(AioContext *ctx, bool blocking)
 {
-    if (blocking && ctx->poll_max_ns && ctx->poll_disable_cnt == 0) {
+    if (blocking && ctx->poll_max_ns && !atomic_read(&ctx->poll_disable_cnt)) {
         /* See qemu_soonest_timeout() uint64_t hack */
         int64_t max_ns = MIN((uint64_t)aio_compute_timeout(ctx),
                              (uint64_t)ctx->poll_ns);
-- 
2.17.1

On Wed, 09/12 19:10, Paolo Bonzini wrote:
> It is valid for an aio_set_fd_handler to happen concurrently with
> aio_poll.  In that case, poll_disable_cnt can change under the heels
> of aio_poll, and the assertion on poll_disable_cnt can fail in
> run_poll_handlers.
> 
> Therefore, this patch simply checks the counter on every polling
> iteration.  There are no particular needs for ordering, since the
> polling loop is terminated anyway by aio_notify at the end of
> aio_set_fd_handler.
> 
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
>  util/aio-posix.c | 26 +++++++++++++++-----------
>  1 file changed, 15 insertions(+), 11 deletions(-)
> 
> diff --git a/util/aio-posix.c b/util/aio-posix.c
> index 131ba6b4a8..5c29380575 100644
> --- a/util/aio-posix.c
> +++ b/util/aio-posix.c
> @@ -211,6 +211,7 @@ void aio_set_fd_handler(AioContext *ctx,
>      AioHandler *node;
>      bool is_new = false;
>      bool deleted = false;
> +    int poll_disable_change;
>  
>      qemu_lockcnt_lock(&ctx->list_lock);
>  
> @@ -244,11 +245,9 @@ void aio_set_fd_handler(AioContext *ctx,
>              QLIST_REMOVE(node, node);
>              deleted = true;
>          }
> -
> -        if (!node->io_poll) {
> -            ctx->poll_disable_cnt--;
> -        }
> +        poll_disable_change = -!node->io_poll;
>      } else {
> +        poll_disable_change = !io_poll - (node && !node->io_poll);
>          if (node == NULL) {
>              /* Alloc and insert if it's not already there */
>              node = g_new0(AioHandler, 1);
> @@ -257,10 +256,6 @@ void aio_set_fd_handler(AioContext *ctx,
>  
>              g_source_add_poll(&ctx->source, &node->pfd);
>              is_new = true;
> -
> -            ctx->poll_disable_cnt += !io_poll;
> -        } else {
> -            ctx->poll_disable_cnt += !io_poll - !node->io_poll;
>          }
>  
>          /* Update handler with latest information */
> @@ -274,6 +269,15 @@ void aio_set_fd_handler(AioContext *ctx,
>          node->pfd.events |= (io_write ? G_IO_OUT | G_IO_ERR : 0);
>      }
>  
> +    /* No need to order poll_disable_cnt writes against other updates;
> +     * the counter is only used to avoid wasting time and latency on
> +     * iterated polling when the system call will be ultimately necessary.
> +     * Changing handlers is a rare event, and a little wasted polling until
> +     * the aio_notify below is not an issue.
> +     */
> +    atomic_set(&ctx->poll_disable_cnt,
> +               atomic_read(&ctx->poll_disable_cnt) + poll_disable_change);

Why not atomic_add?

> +
>      aio_epoll_update(ctx, node, is_new);
>      qemu_lockcnt_unlock(&ctx->list_lock);
>      aio_notify(ctx);
> @@ -525,7 +529,6 @@ static bool run_poll_handlers(AioContext *ctx, int64_t max_ns)
>  
>      assert(ctx->notify_me);
>      assert(qemu_lockcnt_count(&ctx->list_lock) > 0);
> -    assert(ctx->poll_disable_cnt == 0);
>  
>      trace_run_poll_handlers_begin(ctx, max_ns);
>  
> @@ -533,7 +536,8 @@ static bool run_poll_handlers(AioContext *ctx, int64_t max_ns)
>  
>      do {
>          progress = run_poll_handlers_once(ctx);
> -    } while (!progress && qemu_clock_get_ns(QEMU_CLOCK_REALTIME) < end_time);
> +    } while (!progress && qemu_clock_get_ns(QEMU_CLOCK_REALTIME) < end_time
> +             && !atomic_read(&ctx->poll_disable_cnt));
>  
>      trace_run_poll_handlers_end(ctx, progress);
>  
> @@ -552,7 +556,7 @@ static bool run_poll_handlers(AioContext *ctx, int64_t max_ns)
>   */
>  static bool try_poll_mode(AioContext *ctx, bool blocking)
>  {
> -    if (blocking && ctx->poll_max_ns && ctx->poll_disable_cnt == 0) {
> +    if (blocking && ctx->poll_max_ns && !atomic_read(&ctx->poll_disable_cnt)) {
>          /* See qemu_soonest_timeout() uint64_t hack */
>          int64_t max_ns = MIN((uint64_t)aio_compute_timeout(ctx),
>                               (uint64_t)ctx->poll_ns);
> -- 
> 2.17.1
> 
> 

Reviewed-by: Fam Zheng <famz@redhat.com>

On 13/09/2018 08:56, Fam Zheng wrote:
>> +    /* No need to order poll_disable_cnt writes against other updates;
>> +     * the counter is only used to avoid wasting time and latency on
>> +     * iterated polling when the system call will be ultimately necessary.
>> +     * Changing handlers is a rare event, and a little wasted polling until
>> +     * the aio_notify below is not an issue.
>> +     */
>> +    atomic_set(&ctx->poll_disable_cnt,
>> +               atomic_read(&ctx->poll_disable_cnt) + poll_disable_change);
>
> Why not atomic_add?

This is not lockless, it's protected by list_lock, so there's no race
condition involved.  I'm just mimicking what is done for other similar
cases, for example involving seqlocks.

The alternative would be to add a full set of
atomic_{add,sub,...}_relaxed atomics.

Paolo

On Thu, 09/13 10:29, Paolo Bonzini wrote:
> On 13/09/2018 08:56, Fam Zheng wrote:
> >> +    /* No need to order poll_disable_cnt writes against other updates;
> >> +     * the counter is only used to avoid wasting time and latency on
> >> +     * iterated polling when the system call will be ultimately necessary.
> >> +     * Changing handlers is a rare event, and a little wasted polling until
> >> +     * the aio_notify below is not an issue.
> >> +     */
> >> +    atomic_set(&ctx->poll_disable_cnt,
> >> +               atomic_read(&ctx->poll_disable_cnt) + poll_disable_change);
> >
> > Why not atomic_add?
> 
> This is not lockless, it's protected by list_lock, so there's no race
> condition involved.  I'm just mimicking what is done for other similar
> cases, for example involving seqlocks.

Yeah, no doubt about the correctness. Just curious about the preference since
atomic_add is less typing and one less atomic_* op.

Fam

> 
> The alternative would be to add a full set of
> atomic_{add,sub,...}_relaxed atomics.
> 
> Paolo