Re: [Xen-devel] [PATCH] common/vm_event: Prevent guest locking with large max_vcpus

[Razvan Cojocaru <rcojocaru@xxxxxxxxxxxxxxx>]

On 02/08/2017 06:25 PM, Tamas K Lengyel wrote:
> 
> 
> On Wed, Feb 8, 2017 at 2:00 AM, Razvan Cojocaru
> <rcojocaru@xxxxxxxxxxxxxxx <mailto:rcojocaru@xxxxxxxxxxxxxxx>> wrote:
> 
>     It is currently possible for the guest to lock when subscribing
>     to synchronous vm_events if max_vcpus is larger than the
>     number of available ring buffer slots. This patch no longer
>     blocks already paused VCPUs, fixing the issue for this use
>     case.
> 
>     Signed-off-by: Razvan Cojocaru <rcojocaru@xxxxxxxxxxxxxxx
>     <mailto:rcojocaru@xxxxxxxxxxxxxxx>>
>     ---
>      xen/common/vm_event.c | 3 ++-
>      1 file changed, 2 insertions(+), 1 deletion(-)
> 
>     diff --git a/xen/common/vm_event.c b/xen/common/vm_event.c
>     index 82ce8f1..2005a64 100644
>     --- a/xen/common/vm_event.c
>     +++ b/xen/common/vm_event.c
>     @@ -316,7 +316,8 @@ void vm_event_put_request(struct domain *d,
>           * See the comments above wake_blocked() for more information
>           * on how this mechanism works to avoid waiting. */
>          avail_req = vm_event_ring_available(ved);
>     -    if( current->domain == d && avail_req < d->max_vcpus )
>     +    if( current->domain == d && avail_req < d->max_vcpus &&
>     +        !atomic_read( &current->vm_event_pause_count ) )
>              vm_event_mark_and_pause(current, ved);
> 
> 
> Hi Razvan,
> I would also like to have the change made in this patch that unblocks
> the vCPUs as soon as a spot opens up on the ring. Doing just what this
> patch has will not solve the problem if there are asynchronous events used.

Fair enough, I thought that might need more discussion and thus put into
a subsequent patch, but I'll modify that as well, give it a spin and
submit V2.


Thanks,
Razvan

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel