[Intel-gfx] [PATCH 03/11] drm/i915/execlists: Pull submit after dequeue under timeline lock

[Chris Wilson]

Quoting Tvrtko Ursulin (2018-06-04 10:25:46)
> 
> On 01/06/2018 15:07, Chris Wilson wrote:
> > Quoting Tvrtko Ursulin (2018-06-01 15:02:38)
> >>
> >> On 31/05/2018 19:51, Chris Wilson wrote:
> >>> In the next patch, we will begin processing the CSB from inside the
> >>> interrupt handler. This means that updating the execlists->port[] will
> >>
> >> The message that we will be processing CSB from irq handler, here and in
> >> following patch 5/11, doesn't seem to be true. So why is this needed?
> >> Why not just drop some patches from the series?
> > 
> > It will still be called from irq context; as submit_request is called
> > from irq context. Hence we still require irqsafe variants.
> 
> submit_request is already called from irqoff contexts so I don't get it.

Right, but the patch series pulls the CSB processing into
submit_request as well.
 
> > So yes, while the overall direction of the patchset changed slightly to
> > be less enthusiastic about calling from irq context, such calls were not
> > eliminated.
> 
> I have a problem with commit messages where one says we'll be processing 
> CSB directly from hard irq, while another says we'll always use the tasklet.

It's done inside the tasklet callback; the tasklet callback may be
invoked directly, and that may also happen inside an irq.
-Chris