[Intel-gfx] [PATCH 1/2] drm/i915/execlists: Pull tasklet interrupt-bh local to direct submission

[Tvrtko Ursulin]

On 24/03/2020 12:07, Chris Wilson wrote:
> We dropped calling process_csb prior to handling direct submission in
> order to avoid the nesting of spinlocks and lift process_csb() and the
> majority of the tasklet out of irq-off. However, we do want to avoid
> ksoftirqd latency in the fast path, so try and pull the interrupt-bh
> local to direct submission if we can acquire the tasklet's lock.
> 
> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> Cc: Tvrtko Ursulin <tvrtko.ursulin at linux.intel.com>
> ---
>   drivers/gpu/drm/i915/gt/intel_lrc.c | 6 ++++++
>   1 file changed, 6 insertions(+)
> 
> diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c b/drivers/gpu/drm/i915/gt/intel_lrc.c
> index 210f60e14ef4..82dee2141b46 100644
> --- a/drivers/gpu/drm/i915/gt/intel_lrc.c
> +++ b/drivers/gpu/drm/i915/gt/intel_lrc.c
> @@ -2891,6 +2891,12 @@ static void __submit_queue_imm(struct intel_engine_cs *engine)
>   	if (reset_in_progress(execlists))
>   		return; /* defer until we restart the engine following reset */
>   
> +	/* Hopefully we clear execlists->pending[] to let us through */
> +	if (execlists->pending[0] && tasklet_trylock(&execlists->tasklet))

Does access to pending needs a READ_ONCE?

  {
> +		process_csb(engine);
> +		tasklet_unlock(&execlists->tasklet);
> +	}
> +
>   	__execlists_submission_tasklet(engine);
>   }
>   
> 

__execlists_submission_tasklet does check with READ_ONCE.

I think locking is fine, given how normal flow is tasklet -> irqsave 
engine lock, and here we have the reverse, but a trylock.

Regards,

Tvrtko