[PATCH 07/91] drm/i915/execlists: Reorder execlists register enabling

Chris Wilson chris at chris-wilson.co.uk
Thu Jan 5 13:53:08 UTC 2017


On Thu, Jan 05, 2017 at 01:44:07PM +0000, Tvrtko Ursulin wrote:
> 
> On 05/01/2017 10:34, Chris Wilson wrote:
> >Empirically we restart more successfully if we call lrc_init_hws()
> >(which contains a posting read) last. (The failure mode that was
> >observed was that breadcrumb writes into the HWS from the recovered
> >request went astray leading to the context-switch maintaining forward
> 
> Recovered request is the first after the skipped one, when
> resetting? Maybe say explicitly in the commit that this is about
> reset?

The first request to be executed after a reset is the current hung
request - we tell the hw to start from its breadcrumb, skipping the
replay of the batch. It is a convenient way of triggering the interrupt
if anything is watching, and avoids the complication of trying to
advance the seqno manually (especially with the goal of avoiding
struct_mutex here).

"restart more successfully after reset"
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre


More information about the Intel-gfx-trybot mailing list