[Intel-gfx] [PATCH v2] drm/i915: Assert that the context-switch completion matches our context

[Chris Wilson]

On Mon, Jan 23, 2017 at 10:44:03AM +0000, Tvrtko Ursulin wrote:
> 
> On 23/01/2017 10:30, Chris Wilson wrote:
> >On Mon, Jan 23, 2017 at 10:26:15AM +0000, Tvrtko Ursulin wrote:
> >>
> >>On 23/01/2017 08:50, Chris Wilson wrote:
> >>>When execlists signals the context completion, it also provides the
> >>>context id for the status event. Assert that id matches the one we expect.
> >>>
> >>>v2: The upper dword of the context status is a duplicate of the upper
> >>>dword from elsp submission (i.e. includes the group id as well as the
> >>>context id). Include this check as well.
> >>>
> >>>Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> >>>Cc: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
> >>>---
> >>>drivers/gpu/drm/i915/intel_lrc.c | 6 ++++++
> >>>1 file changed, 6 insertions(+)
> >>>
> >>>diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
> >>>index 432ee495dec2..963b1888d8a0 100644
> >>>--- a/drivers/gpu/drm/i915/intel_lrc.c
> >>>+++ b/drivers/gpu/drm/i915/intel_lrc.c
> >>>@@ -595,6 +595,12 @@ static void intel_lrc_irq_handler(unsigned long data)
> >>>			if (!(status & GEN8_CTX_STATUS_COMPLETED_MASK))
> >>>				continue;
> >>>
> >>>+			/* Check the context/desc id for this event matches */
> >>>+			GEM_BUG_ON(((readl(buf + 2 * idx + 1) & (MAX_CONTEXT_HW_ID-1)) !=
> >>>+				    port[0].request->ctx->hw_id));
> >>>+			GEM_BUG_ON(readl(buf + 2 * idx + 1) !=
> >>>+				   upper_32_bits(port[0].request->ctx->engine[engine->id].lrc_desc));
> >>>+
> >>
> >>Not sure that you need the first line since the lrc_desc includes
> >>the hw_id? Or perhaps you don't need the second one since lrc_desc
> >>is built from hw_id? :)
> >
> >It was defense in depth :)
> >
> >The first line is certain, that make sure the event matches
> >port[0].request.
> >
> >The second line makes sure that everything matches my understanding, and
> >that lrc_desc is correct (or vice versa).
> 
> Too redundant IMO. lrc_desc correctness could/should be asserted
> somewhere else, not in the irq handler. I would keep one of them,
> probably the assert agains lrc_desc because that's the right layer
> for this place.

Preemptive r-b on the second then? Joonas?
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre