[Intel-gfx] [PATCH 1/1] drm/i915: re-disable RC6p on Sandy Bridge

[Rodrigo Vivi]

On Tue, Dec 20, 2022 at 12:32:05PM +0200, Ville Syrjälä wrote:
> On Mon, Dec 19, 2022 at 06:29:27PM +0100, Sasa Dragic wrote:
> > RC6p on Sandy Bridge got re-enabled over time, causing visual glitches
> > and GPU hangs.
> > 
> > Disabled originally in commit 1c8ecf80fdee ("drm/i915: do not enable
> > RC6p on Sandy Bridge").
> 
> re cover letter:
> > It was originally disabled in commit 1c8ecf80fdee ("drm/i915: do not
> > enable RC6p on Sandy Bridge"), and subsequently re-enabled by
> > 13c5a577b342 ("drm/i915/gt: Select the deepest available parking mode
> > for rc6"), which seems to focus only on Ivy Bridge.
> 
> That only kicks in while parked (ie. fully idle from
> software POV). I think the real bad commit is
> fb6db0f5bf1d ("drm/i915: Remove unsafe i915.enable_rc6")
> which seems to affects which rc6 level is selected while
> unparked.
> 
> We should mention both of those commits here:
> Fixes: fb6db0f5bf1d ("drm/i915: Remove unsafe i915.enable_rc6")
> Fixes: 13c5a577b342 ("drm/i915/gt: Select the deepest available parking mode for rc6")
> 
> Also we want
> Cc: stable at vger.kernel.org
> 
> We can add those while pushing, so no need to resend for that.

agreed.

Reviewed-by: Rodrigo Vivi <rodrigo.vivi at intel.com>

> 
> > 
> > Signed-off-by: Sasa Dragic <sasa.dragic at gmail.com>
> > ---
> >  drivers/gpu/drm/i915/i915_pci.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/gpu/drm/i915/i915_pci.c b/drivers/gpu/drm/i915/i915_pci.c
> > index 668e9da52584..69377564028a 100644
> > --- a/drivers/gpu/drm/i915/i915_pci.c
> > +++ b/drivers/gpu/drm/i915/i915_pci.c
> > @@ -423,7 +423,8 @@ static const struct intel_device_info ilk_m_info = {
> >  	.has_coherent_ggtt = true, \
> >  	.has_llc = 1, \
> >  	.has_rc6 = 1, \
> > -	.has_rc6p = 1, \
> > +	/* snb does support rc6p, but enabling it causes various issues */ \
> > +	.has_rc6p = 0, \
> 
> The one downside of doing it this way is that we also lose
> the debugfs/sysfs files so we can't monitor rc6+/++
> residency anymore to make sure they are not entered.
> 
> I think ideally we'd split this into two parts: which rc6
> states the hw actually has vs. which rc6 states we actually
> want to use. But at least for the time being I think this
> simple should be sufficient, and should be easy to backport
> to stable releases.
> 
> >  	.has_rps = true, \
> >  	.dma_mask_size = 40, \
> >  	.__runtime.ppgtt_type = INTEL_PPGTT_ALIASING, \
> > -- 
> > 2.37.2
> 
> -- 
> Ville Syrjälä
> Intel