[Intel-gfx] [PATCH 19/19] drm/i915: only enable HWSTAM interrupts on postinstall on ILK+

[Damien Lespiau]

On Wed, Apr 02, 2014 at 04:28:06PM +0100, Chris Wilson wrote:
> On Wed, Apr 02, 2014 at 11:14:58AM -0300, Paulo Zanoni wrote:
> > 2014-04-02 8:27 GMT-03:00 Chris Wilson <chris at chris-wilson.co.uk>:
> > > On Wed, Apr 02, 2014 at 12:23:51PM +0100, Chris Wilson wrote:
> > >> On Wed, Apr 02, 2014 at 12:21:45PM +0100, Damien Lespiau wrote:
> > >> > On Tue, Apr 01, 2014 at 03:37:27PM -0300, Paulo Zanoni wrote:
> > >> > > From: Paulo Zanoni <paulo.r.zanoni at intel.com>
> > >> > >
> > >> > > We should only enable interrupts at postinstall.
> > >> > >
> > >> > > And now on ILK/SNB/IVB/HSW the irq_preinstall and irq_postinstall
> > >> > > functions leave the hardware in the same state.
> > >> > >
> > >> > > Signed-off-by: Paulo Zanoni <paulo.r.zanoni at intel.com>
> > >> >
> > >> > Orthogonal note to this patch, I'm wondering why we enable any interrupt
> > >> > reporting in the HW status page, I don't see anywhere we read that
> > >> > information back? Any idea?
> > >>
> > >> Back in the SNB days, interrupts broke without this piece of magic.
> > >
> > > To be more precise, iirc, it was a step towards getting coherent
> > > breadcrumb writes into the HWS. As it turns out, there were more steps
> > > required. Considering that it can now probably be dropped again? Any
> > > takers?
> > 
> > I noticed the HWSTAM code looks weird, but I don't really know much
> > about why it's there, so I decided to not add any regressions. I wrote
> > this patch a long time ago, and gave up on upstreaming it, but in case
> > anybody wants, I can send it to the list:
> > http://cgit.freedesktop.org/~pzanoni/linux/commit/?h=c8-wip&id=7ebe245a2c55379ddc7b36f1fb440215c23f1570
> > . Look at the commit message, it may be an interesting starting point
> > if anybody wants to dig on the issue...
> 
> Yup, that is very interesting. That may be the minimal piece of HWSTAM
> magic we need, and be an interim step to removing HWSTAM entirely.

Would you mind resending:
  - a patch with Chris' comment above, that's alone is useful for the
    next guy looking at the code,
  - a patch with the change (Maybe reusing GT_RENDER_USER_INTERRUPT as
    HWSTAM uses the same definitions as the other interrupts registers)?

I'm sure we can get some reviews on the first, hopefully some testing on
the second.

-- 
Damien