Re: [Intel-gfx] drm/i915 4.5/4.6 stable backport request for CHV

[Greg KH <greg@kroah.com>]

On Wed, Jun 22, 2016 at 03:55:03PM +0200, Daniel Vetter wrote:
> On Mon, Jun 6, 2016 at 11:32 AM, Ville Syrj�l�
> <ville.syrjala@linux.intel.com> wrote:
> > On Sat, Jun 04, 2016 at 02:06:58PM -0700, Greg KH wrote:
> >> On Fri, May 27, 2016 at 11:30:30AM +0300, ville.syrjala@linux.intel.com wrote:
> >> > From: Ville Syrj�l� <ville.syrjala@linux.intel.com>
> >> >
> >> > Several nasty i915 regressions affecting CHV slipped through
> >> > to 4.5 and 4.6.
> >> >
> >> > The first fix we want in 4.5 and 4.6 is
> >> > commit caed361d83b2 ("drm/i915: Fix watermarks for VLV/CHV")
> >> > It won't cherry-pick cleanly to either one, so I've included conflict
> >> > free versions for both. This one fixes display FIFO underruns that can
> >> > lead to the screen totally blanking out.
> >>
> >> Now applied, thanks.
> >>
> >> > The other one I'd like to have in 4.6 is
> >> > commit 9f6151c90390 ("drm/i915: Pass the correct crtc state to .update_plane()")
> >> > which avoids a totally corrupted display in some cases.
> >>
> >> Now applied.
> >>
> >> > And the third on is a bit more annoying. The regression is caused by
> >> > commit 9dbaab56ac09 ("drm/i915: Exit cherryview_irq_handler() after one pass")
> >> > which I though we had prevented from getting out on its own, but turns
> >> > out I was wrong. It basically makes the GPU unusable, so we do need to
> >> > fix it somehow. The simple solution would be to revert it in 4.6 only.
> >> > The more complicated solution is to backport the proper fix, which more
> >> > or less requires the following set of commits [1], which is maybe a bit
> >> > too much for stable. I could try to trim it a bit perhaps, but then we
> >> > start to enter the territory of untested code which I don't particularly
> >> > like. Let me know what you think.
> >> >
> >> > [1]
> >> >  1e1cace942ef ("drm/i915: Eliminate loop from VLV irq handler")
> >> >  a5e485a95c9c ("drm/i915: Clear VLV_IER around irq processing")
> >> >  4a0a0202b023 ("drm/i915: Clear VLV_MASTER_IER around irq processing")
> >> >  7ce4d1f2730f ("drm/i915: Clear VLV_IIR after PIPESTAT")
> >> >  34c7b8a7b8b5 ("drm/i915: Set up VLV_MASTER_IER consistently")
> >> >  e5328c43d46e ("drm/i915: Use GEN8_MASTER_IRQ_CONTROL consistently")
> >> >  71b8b41d5b35 ("drm/i915: Move DPINVGTT setup to vlv_display_irq_reset()")
> >> >  6b7eafc1b43d ("drm/i915: Warn if irq_mask isn't ~0 during vlv/cvh display irq postinstall")
> >> >  9ab981f22bef ("drm/i915: Use GEN5_IRQ_INIT() in vlv_display_irq_postinstall()")
> >> >  d6c698035892 ("drm/i915: Clear display interrupt before enabling when turning on the power well")
> >> >  8bb613068a63 ("drm/i915: Move vlv/chv display irq code to a more logical place")
> >> >  9918271efc7a ("drm/i915: Skip display irq setup if display irqs aren't flagged as enabled")
> >> >  ad22d10654ea ("drm/i915: Fix up vlv/chv display irq setup")
> >> >  93de68f94081 ("drm/i915: Remove "VLV magic" from irq setup")
> >>
> >> I think reverting that one patch for 4.6 makes more sense than adding
> >> all of these patches.  I'll do that if you want me to.
> >
> > That works for me. Thanks.
> 
> Has this happened? I just got a ping on irc that braswell in 4.6 is
> still entirely toasted :(

Sorry for the delay, revert is now queued up.

greg k-h