Re: [PATCH][2.6-mm] Avoid flushing AIO workqueue on cancel/exit

[Andrew Morton]

Suparna Bhattacharya <suparna@xxxxxxxxxx> wrote:
>
> When streaming AIO requests are in progress on multiple
>  io context's, flushing the AIO workqueue on i/o cancellation
>  or process exit could potentially end up waiting for a 
>  long time as fresh requests from other active ioctx's keep 
>  getting queued up.

But flush_workqueue() will ignore any newly-added work requests:

 * This function will sample each workqueue's current insert_sequence number and
 * will sleep until the head sequence is greater than or equal to that.  This
 * means that we sleep until all works which were queued on entry have been
 * handled, but we are not livelocked by new incoming ones.

Now, flush_workqueue() is potentially inefficient on SMP because it flushes
each CPU's workqueue sequentially.  But we can fix that in
flush_workqueue() by converting it to a two-pass approach:

a) gather each CPU's insert_sequence number into a local array[NR_CPUS]

b) wait until each CPU's remove_sequence number exceeds the previously-gathered
   insert_sequence number.


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/