[Intel-gfx] [PATCH 5.4.y] drm/i915: Don't use BAR mappings for ring buffers with LLC

[John Harrison]

On 3/17/2023 05:58, Greg KH wrote:
> On Thu, Mar 16, 2023 at 01:58:35PM -0700, John Harrison wrote:
>> On 3/15/2023 10:57, Greg KH wrote:
>>> On Wed, Mar 15, 2023 at 10:07:53AM -0700, John Harrison wrote:
>>>> On 3/15/2023 00:51, Greg KH wrote:
>>>>> On Mon, Mar 13, 2023 at 07:22:11PM -0700, John.C.Harrison at Intel.com wrote:
>>>>>> From: John Harrison <John.C.Harrison at Intel.com>
>>>>>>
>>>>>> Direction from hardware is that ring buffers should never be mapped
>>>>>> via the BAR on systems with LLC. There are too many caching pitfalls
>>>>>> due to the way BAR accesses are routed. So it is safest to just not
>>>>>> use it.
>>>>>>
>>>>>> Signed-off-by: John Harrison <John.C.Harrison at Intel.com>
>>>>>> Fixes: 9d80841ea4c9 ("drm/i915: Allow ringbuffers to be bound anywhere")
>>>>>> Cc: Chris Wilson <chris at chris-wilson.co.uk>
>>>>>> Cc: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
>>>>>> Cc: Jani Nikula <jani.nikula at linux.intel.com>
>>>>>> Cc: Rodrigo Vivi <rodrigo.vivi at intel.com>
>>>>>> Cc: Tvrtko Ursulin <tvrtko.ursulin at linux.intel.com>
>>>>>> Cc: intel-gfx at lists.freedesktop.org
>>>>>> Cc: <stable at vger.kernel.org> # v4.9+
>>>>>> Tested-by: Jouni Högander <jouni.hogander at intel.com>
>>>>>> Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio at intel.com>
>>>>>> Link: https://patchwork.freedesktop.org/patch/msgid/20230216011101.1909009-3-John.C.Harrison@Intel.com
>>>>>> (cherry picked from commit 65c08339db1ada87afd6cfe7db8e60bb4851d919)
>>>>>> Signed-off-by: Jani Nikula <jani.nikula at intel.com>
>>>>>> (cherry picked from commit 85636167e3206c3fbd52254fc432991cc4e90194)
>>>>>> Signed-off-by: John Harrison <John.C.Harrison at Intel.com>
>>>>>> ---
>>>>>>     drivers/gpu/drm/i915/gt/intel_ringbuffer.c | 4 ++--
>>>>>>     1 file changed, 2 insertions(+), 2 deletions(-)
>>>>> Also queued up for 5.10.y, you forgot that one :)
>>>> I'm still working through the backlog of them.
>>>>
>>>> Note that these patches must all be applied as a pair. The 'don't use
>>>> stolen' can be applied in isolation but won't totally fix the problem.
>>>> However, applying 'don't use BAR mappings' without applying the stolen patch
>>>> first will results in problems such as the failure to boot that was recently
>>>> reported and resulted in a revert in one of the trees.
>>> I do not understand, you only submitted 1 patch here, what is the
>>> "pair"?
>> The original patch series was two patches -
>> https://patchwork.freedesktop.org/series/114080/. One to not use stolen
>> memory and the other to not use BAR mappings. If the anti-BAR patch is
>> applied without the anti-stolen patch then the i915 driver will attempt to
>> access stolen memory directly which will fail. So both patches must be
>> applied and in the correct order to fix the problem of cache aliasing when
>> using BAR accesses on LLC systems.
>>
>> As above, I am working my way through the bunch of 'FAILED patch' emails.
>> The what-to-do instructions in those emails explicitly say to send the patch
>> individually in reply to the 'FAILED' message rather than as part of any
>> original series.
> So what commits exactly in Linus's tree should be in these stable
> branches?  Sorry, I still do not understand if we are missing one or if
> we need to revert something.
>
> confused,
>
> greg k-h
As far as I can tell, I have replied to all the "FAILED: patch" emails 
now. There should be a versions of these two patches available for all 
trees (being 4.14, 4.19, 5.4, 5.10 and 5.15):
     690e0ec8e63d drm/i915: Don't use stolen memory for ring buffers 
with LLC
     85636167e320 drm/i915: Don't use BAR mappings for ring buffers with LLC

They should be applied in the order of 'stolen memory' first and 'BAR 
mappings' second.

Thanks,
John.