[Intel-gfx] [PATCH] drm/i915: enable semaphores on gen6 if dmar is not active

[Daniel Vetter]

On Mon, Apr 02, 2012 at 02:44:14PM -0700, Andrew Lutomirski wrote:
> On Mon, Apr 2, 2012 at 1:52 PM, Daniel Vetter <daniel at ffwll.ch> wrote:
> > On Mon, Apr 02, 2012 at 01:45:39PM -0700, Andrew Lutomirski wrote:
> >> On Mon, Apr 2, 2012 at 11:48 AM, Daniel Vetter <daniel.vetter at ffwll.ch> wrote:
> >> > Inspired by the recent ppgtt regression report, where switching of
> >> > dmar only for the gpu seems to fix things completely, I've looked
> >> > again at the semaphores+vt-d situation.
> >> >
> >> > Contrary to my earlier testing a few months back my system is now
> >> > stable with dmar disabled for the igd, and not only when disabling
> >> > dmar completely.
> >> >
> >> > So I'm rather hopeful that all our recent fixes for snb have changed
> >> > things for code and it's time to try enabling semaphores again. We've
> >> > also had issues with enabling semaphores which are not vt-d related,
> >> > but I guess these are all fixed by the autoreport-disabling and lazy
> >> > request fix. And there's only one way to find out whether there are
> >> > still other issues ...
> >> >
> >> > Signed-Off-by: Daniel Vetter <daniel.vetter at ffwll.ch>
> >> >
> >> > ---
> >> >
> >> > If no further vt-d regressions show up in the 3.4 cycle I plan to
> >> > merge this into -next for 3.5 (in a month or so). Comments about how
> >> > unfeasibly you deem this highly welcome.
> >> > -Daniel
> >> > ---
> >> >  drivers/gpu/drm/i915/i915_gem_execbuffer.c |    8 +++++---
> >> >  1 files changed, 5 insertions(+), 3 deletions(-)
> >> >
> >> > diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
> >> > index 8e0b686..ac52433 100644
> >> > --- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
> >> > +++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
> >> > @@ -849,9 +849,11 @@ intel_enable_semaphores(struct drm_device *dev)
> >> >        if (i915_semaphores >= 0)
> >> >                return i915_semaphores;
> >> >
> >> > -       /* Disable semaphores on SNB */
> >> > -       if (INTEL_INFO(dev)->gen == 6)
> >> > -               return 0;
> >> > +#ifdef CONFIG_INTEL_IOMMU
> >> > +       /* Disable semaphores on SNB if VT-d is on. */
> >> > +       if (INTEL_INFO(dev)->gen == 6 && intel_iommu_gfx_mapped)
> >> > +               return false;
> >>
> >> It might be nice to have a printk here giving instructions for how to work
> >> around this.  For example:
> >>
> >> i915: Disabling semaphores to work around a DMAR issue.  As an
> >> alternative, boot with intel_iommu=igfx_off [or on, igfx_off, or
> >> whatever it's supposed to be].
> >>
> >> The documentation in kernel-parameters.txt is at best unclear to the
> >> uninitiated.
> >
> > If this really does work, I'll look into this. Atm it's still unclear
> > where to exactly put the plain, we need to annoy internal hardware people
> > to analyze this. Once we have enough evidence that the combination of gpu
> > with various features enable plus VT-d just doesn't seem to work reliably.
> >
> > So can you check whether things do indeed work with this patch, atop
> > kernel 3.4-rc1? Iirc you've been the one with the most trouble when
> > semaphores are enabled ...
> 
> I've had a hard time reproducing the problems lately.  The
> unconditional instant hard hang went away a few months ago, I think.
> I'll give it another shot when I get home to that machine.

Well, I've managed to kill my machine shortly after login with semaphores,
rc6 and dmar enabled. It hasn't died in the same configuration after a few
hours of light usage (in my case that seems to be the best killer) with
dmar disabled for just the gpu.

So if you can give your machine some serious beating with the different
options, that's really great.
-Daniel
-- 
Daniel Vetter
Mail: daniel at ffwll.ch
Mobile: +41 (0)79 365 57 48