Re: [PATCH] dma-buf: avoid scheduling on fence status query

Re: [PATCH] dma-buf: avoid scheduling on fence status query

[Andres Rodriguez]

CC a few extra lists I missed.

Regards,
Andres

On 2017-04-25 09:36 PM, Andres Rodriguez wrote:
> When a timeout of zero is specified, the caller is only interested in
> the fence status.
>
> In the current implementation, dma_fence_default_wait will always call
> schedule_timeout() at least once for an unsignaled fence. This adds a
> significant overhead to a fence status query.
>
> Avoid this overhead by returning early if a zero timeout is specified.
>
> Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
> ---
>
> This heavily affects the performance of the Source2 engine running on
> radv.
>
> This patch improves dota2(radv) perf on a i7-6700k+RX480 system from
> 72fps->81fps.
>
>  drivers/dma-buf/dma-fence.c | 3 +++
>  1 file changed, 3 insertions(+)
>
> diff --git a/drivers/dma-buf/dma-fence.c b/drivers/dma-buf/dma-fence.c
> index 0918d3f..348e9e2 100644
> --- a/drivers/dma-buf/dma-fence.c
> +++ b/drivers/dma-buf/dma-fence.c
> @@ -380,6 +380,9 @@ dma_fence_default_wait(struct dma_fence *fence, bool intr, signed long timeout)
>  	if (test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &fence->flags))
>  		return ret;
>
> +	if (!timeout)
> +		return 0;
> +
>  	spin_lock_irqsave(fence->lock, flags);
>
>  	if (intr && signal_pending(current)) {
>

Re: [PATCH] dma-buf: avoid scheduling on fence status query

[Christian König]

NAK, I'm wondering how often I have to reject that change. We should 
probably add a comment here.

Even with a zero timeout we still need to enable signaling, otherwise 
some fence will never signal if userspace just polls on them.

If a caller is only interested in the fence status without enabling the 
signaling it should call dma_fence_is_signaled() instead.

Regards,
Christian.

Am 26.04.2017 um 04:50 schrieb Andres Rodriguez:
> CC a few extra lists I missed.
>
> Regards,
> Andres
>
> On 2017-04-25 09:36 PM, Andres Rodriguez wrote:
>> When a timeout of zero is specified, the caller is only interested in
>> the fence status.
>>
>> In the current implementation, dma_fence_default_wait will always call
>> schedule_timeout() at least once for an unsignaled fence. This adds a
>> significant overhead to a fence status query.
>>
>> Avoid this overhead by returning early if a zero timeout is specified.
>>
>> Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
>> ---
>>
>> This heavily affects the performance of the Source2 engine running on
>> radv.
>>
>> This patch improves dota2(radv) perf on a i7-6700k+RX480 system from
>> 72fps->81fps.
>>
>>  drivers/dma-buf/dma-fence.c | 3 +++
>>  1 file changed, 3 insertions(+)
>>
>> diff --git a/drivers/dma-buf/dma-fence.c b/drivers/dma-buf/dma-fence.c
>> index 0918d3f..348e9e2 100644
>> --- a/drivers/dma-buf/dma-fence.c
>> +++ b/drivers/dma-buf/dma-fence.c
>> @@ -380,6 +380,9 @@ dma_fence_default_wait(struct dma_fence *fence, 
>> bool intr, signed long timeout)
>>      if (test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &fence->flags))
>>          return ret;
>>
>> +    if (!timeout)
>> +        return 0;
>> +
>>      spin_lock_irqsave(fence->lock, flags);
>>
>>      if (intr && signal_pending(current)) {
>>

Re: [PATCH] dma-buf: avoid scheduling on fence status query

[Dave Airlie]

On 26 April 2017 at 17:20, Christian König <deathsimple@vodafone.de> wrote:
> NAK, I'm wondering how often I have to reject that change. We should
> probably add a comment here.
>
> Even with a zero timeout we still need to enable signaling, otherwise some
> fence will never signal if userspace just polls on them.
>
> If a caller is only interested in the fence status without enabling the
> signaling it should call dma_fence_is_signaled() instead.

Can we not move the return 0 (with spin unlock) down after we enabling
signalling, but before
we enter the schedule_timeout(1)?

Dave.

Re: [PATCH] dma-buf: avoid scheduling on fence status query

[Christian König]

Am 26.04.2017 um 11:59 schrieb Dave Airlie:
> On 26 April 2017 at 17:20, Christian König <deathsimple@vodafone.de> wrote:
>> NAK, I'm wondering how often I have to reject that change. We should
>> probably add a comment here.
>>
>> Even with a zero timeout we still need to enable signaling, otherwise some
>> fence will never signal if userspace just polls on them.
>>
>> If a caller is only interested in the fence status without enabling the
>> signaling it should call dma_fence_is_signaled() instead.
> Can we not move the return 0 (with spin unlock) down after we enabling
> signalling, but before
> we enter the schedule_timeout(1)?

Yes, that would be an option.

Christian.

>
> Dave.

Re: [PATCH] dma-buf: avoid scheduling on fence status query

[Andres Rodriguez]

On 2017-04-26 06:13 AM, Christian König wrote:
> Am 26.04.2017 um 11:59 schrieb Dave Airlie:
>> On 26 April 2017 at 17:20, Christian König <deathsimple@vodafone.de>
>> wrote:
>>> NAK, I'm wondering how often I have to reject that change. We should
>>> probably add a comment here.
>>>
>>> Even with a zero timeout we still need to enable signaling, otherwise
>>> some
>>> fence will never signal if userspace just polls on them.
>>>
>>> If a caller is only interested in the fence status without enabling the
>>> signaling it should call dma_fence_is_signaled() instead.
>> Can we not move the return 0 (with spin unlock) down after we enabling
>> signalling, but before
>> we enter the schedule_timeout(1)?
>
> Yes, that would be an option.
>

I was actually arguing with Dave about this on IRC yesterday. Seems like 
I owe him a beer now.

-Andres

> Christian.
>
>>
>> Dave.
>
>