[Intel-gfx] [PATCH 2/3] drm/i915: handle interrupts from the OA unit

[Dixit, Ashutosh]

On Mon, 13 Apr 2020 08:48:21 -0700, Umesh Nerlige Ramappa wrote:
>
> diff --git a/drivers/gpu/drm/i915/gt/intel_gt_irq.c b/drivers/gpu/drm/i915/gt/intel_gt_irq.c
> index 0cc7dd54f4f9..61eee9fb8872 100644
> --- a/drivers/gpu/drm/i915/gt/intel_gt_irq.c
> +++ b/drivers/gpu/drm/i915/gt/intel_gt_irq.c
> @@ -94,6 +94,18 @@ gen11_gt_engine_identity(struct intel_gt *gt,
>	return ident;
>  }
>
> +static void notify_perfmon_buffer_half_full(struct drm_i915_private *i915)
> +{
> +	atomic64_inc(&i915->perf.exclusive_stream->half_full_count);
> +	wake_up_all(&i915->perf.exclusive_stream->poll_wq);
> +}
> +

I was expecting this function to be almost the same as the timer
oa_poll_check_timer_cb(), something like, maybe with minor variations:

static void notify_perfmon_buffer_half_full(struct drm_i915_private *i915)
{
        struct i915_perf_stream *stream = i915->perf.exclusive_stream;

	if (oa_buffer_check_unlocked(stream)) {
                stream->pollin = true;
                wake_up(&stream->poll_wq);
        }
}

And consequently I was expecting to see zero changes to functions such as
oa_buffer_check_unlocked() and i915_perf_poll_locked() since everything
else is driven off stream->pollin as it is in case the timer callback.

So my question is why is notify_perfmon_buffer_half_full() not as I've
written above and what purpose are these new members half_full_count and
half_full_count_last serving? If it is to save a few cycles to adjust the
tail in oa_buffer_check_unlocked() (and I am not even sure of that) for an
interrupt which fires when half the buffer is full imo it is not worth it.