Message-ID: <20161027062404.ex3w6y54rvrdzdfo@phenom.ffwll.local>
Date:   Thu, 27 Oct 2016 08:24:04 +0200
From:   Daniel Vetter <daniel@...ll.ch>
To:     Lyude <lyude@...hat.com>
Cc:     stable@...r.kernel.org, David Airlie <airlied@...ux.ie>,
        intel-gfx@...ts.freedesktop.org, linux-kernel@...r.kernel.org,
        Hans de Goede <hdegoede@...hat.com>,
        dri-devel@...ts.freedesktop.org,
        Daniel Vetter <daniel.vetter@...el.com>
Subject: Re: [Intel-gfx] [PATCH 0/5] drm/i915/skl: Backport watermark fixes
 for 4.8.y

On Wed, Oct 26, 2016 at 03:36:32PM -0400, Lyude wrote:
> Now that these have finally made it into 4.9, it's time to finally backport
> these fixes. Skylake has been a mess in multi-monitor setups for a while now
> because up until recently we've been updating the watermarks on Skylake just
> like we would for previous generations of Intel GPUs. This means updating
> attributes for each plane, and then only after they've been updated writing
> their new watermark values.
> 
> The problem with this approach is Skylake has double buffered watermark
> registers that are flipped at the same time as the rest of the plane registers.
> This means that the original approach will leave planes active with new
> attributes but without the required watermark programming that would ensure the
> display pipe reads enough data from each plane. As a result, pipes start to
> underrun and the user's displays starts to flicker heavily. Usually in response
> to plugging in new monitors, or moving cursors from one screen to another
> (which triggers a plane and watermark update).
> 
> Additionally, issues were found with the original code for configuring ddb,
> display data buffer, allocations between display planes on Skylake. On Skylake
> all planes have space allocated to them in the ddb, and the hardware requires
> that these allocations never overlap at any point in time. Because ddb
> allocations were not updated alongside plane attributes despite also being
> double buffered registers armed by plane updates, planes were likely to get
> stuck momentarily with ddb allocations that overlapped one another. This would
> also lead to pipe underruns and display flickering.
> 
> The new approach fixes this problem by making sure that on Skylake, attributes
> for display planes are always updated at the same time as the watermarks, and
> pipes are updated in an order that ensures their ddb allocations don't
> overlap at any point between plane updates. This ensures the display pipes are
> always programmed correctly, and dramatically reduces the chance of display
> flickering.
> 
> (note: my e-mail has changed since these patches were upstreamed, and I updated
> the e-mails in these patches to reflect this. if this is wrong I will be happy
> to update and resend the patches).
> 
> Lyude (4):
>   drm/i915/skl: Update plane watermarks atomically during plane updates
>   drm/i915: Move CRTC updating in atomic_commit into it's own hook
>   drm/i915/skl: Update DDB values atomically with wms/plane attrs
>   drm/i915/skl: Don't try to update plane watermarks if they haven't
>     changed
> 
> Paulo Zanoni (1):
>   drm/i915/gen9: only add the planes actually affected by ddb changes

fwiw ack from my side.
-Daniel

> 
>  drivers/gpu/drm/i915/i915_drv.h      |   2 +
>  drivers/gpu/drm/i915/intel_display.c | 189 +++++++++++++++++++-----
>  drivers/gpu/drm/i915/intel_drv.h     |  12 ++
>  drivers/gpu/drm/i915/intel_pm.c      | 271 ++++++++++++++---------------------
>  drivers/gpu/drm/i915/intel_sprite.c  |  14 ++
>  5 files changed, 289 insertions(+), 199 deletions(-)
> 
> -- 
> 2.7.4
> 
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@...ts.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

Powered by blists - more mailing lists