Hi,
On 2/25/21 1:33 PM, Justin Forbes wrote:
On Thu, Feb 25, 2021 at 5:44 AM Hans de Goede
<hdegoede(a)redhat.com> wrote:
>
> Hi,
>
> On 2/23/21 6:25 PM, Hans de Goede wrote:
>> Hi,
>>
>> I've been working with various reporters to try and get the sometimes severe
>> gfx glitches which people are seeing with kernel >= 5.10.9 resolved, see the
>> following bugs:
>>
>> 1843274 - i915 GPU Hang with kernel 5.7 on Haswell (Acer C720P Chromebook)
>> 1925346 - Screen glitches after updating to Kernel 5.10.10
>> 1925903 - Flickering UI elements, screen instability (Wayland)
>> 1931065 - Frequent i915 hangs
>>
>> I've written 2 patches 2 address this and asked users to test Fedora-kernel
>> scratch-builds with those patches added.
>>
>> The first patch is i915-revert.patch, this reverts 3 troublesome commits
>> which were first added in 5.10.9 these commit change the i915 mitigations
>> for a year old i915 CVE which IIRC was mostly about data leaking from a
>> previous GPU context to the next. These patches enable these mitigations
>> on more GPU models, where they were previously disabled because they were
>> causing issues...
>>
>> Reporters have reported a 100% success-rate in getting rid of the gfx
>> issues with these 3 reverts.
>>
>> The second patch is i915-fixes.patch, this cherry-picks 3 patches from
>> -next instead (these were suggested by i915 upstream) and this also helps
>> for most users, but some users still report some rendering artifacts
>> (but overall things are much better).
>>
>> Upstream has gone silent on the issue of there still being rendering
>> artifacts with the 3 cherry-picked fixes unfortunately.
>>
>> Given that we are getting more and more bug-reports about this I think
>> that we should be do something about this downstream. Upstream seams
>> to be aiming for fixing things, rather then reverting, so going with
>> i915-fixes.patch will stay the closest to what I expect to land in
>> 5.10.y and 5.11.y at some point.
>
> The bug reports for this keep coming in; and it seems that this is
> fully resolved for all reporters except one by the i915-fixes.patch which
> I attached earlier; and for the one reporter who is still seeing some
> rendering glitches things are much improved, so I think he is also
> hitting a different issue.
>
> Here is the full lists of bugs which I'm aware of which all have this
> as root cause:
>
> 1843274 - i915 GPU Hang with kernel 5.7 on Haswell (Acer C720P Chromebook)
> 1922511 - Recent upgrades caused smearing/tearing
> 1925346 - Screen glitches after updating to Kernel 5.10.10
> 1925903 - Flickering UI elements, screen instability (Wayland)
> 1931065 - Frequent i915 hangs
>
https://gitlab.freedesktop.org/drm/intel/-/issues/3099
>
> So unless there are any objections I would really like to move forward
> with adding i915-fixes.patch to the Fedora 5.10.y kernels.
>
The actual patch was not attached,
It was attached to the first email in this thread, I've also attached it
to this email again.
so I could not add it, you are
welcome to do so, or attach the patch and I will do it.
If you can take care of this, then that would be great, thanks.
Please also link bug 1925346 in the bodhi-update for the kernel where
this lands, I've dupped all the others to that bug.
Note 5.10
kernels are managed in dist-git right now and follow the regular
process. 5.11 kernels are in the kernel-ark tree in the fedora-5.11
branch, I can handle an MR against this.
I know, I plan to submit my first 5.11 MR for the platform-profile
backport discussed in another thread soonish.
> Note I'm also working on getting the 3 -next commits:
>
> e627d5923cae ("drm/i915/gt: One more flush for Baytrail clear residuals")
> d30bbd62b1bf ("drm/i915/gt: Flush before changing register state")
> 1914911f4aa0 ("drm/i915/gt: Correct surface base address for renderclear")
>
> which are in i915-fixes.patch added to Greg's stable kernels.
Unfortunately only the first commit made it into Linus tree ATM, so
carrying these 3 downstream for now would be good. I've pinged Chris Wilson
about getting the other 2 send out in a fixes pull-req for 5.12 .
Regards,
Hans