[Intel-gfx] [PATCH v3 0/2] Don't use stolen memory or BAR mappings for ring buffers

Hogander, Jouni jouni.hogander at intel.com
Mon Feb 20 06:50:08 UTC 2023


On Fri, 2023-02-17 at 09:18 -0800, John Harrison wrote:
> On 2/17/2023 00:39, Hogander, Jouni wrote:
> > On Wed, 2023-02-15 at 17:10 -0800, John.C.Harrison at Intel.com wrote:
> > > From: John Harrison <John.C.Harrison at Intel.com>
> > > 
> > > Instruction from hardware arch is that stolen memory and BAR
> > > mappings
> > > are unsafe for use as ring buffers. There can be issues with
> > > cache
> > > aliasing due to the CPU access going to memory via the BAR. So,
> > > don't
> > > do it.
> > Tested these patches for GPU Hang I was debugging. Seem to fix that
> > one
> > as well:
> > 
> > Tested-by: Jouni Högander <jouni.hogander at intel.com>
> Sweet! Out of interest, which platform was that? And how reproducible
> was it? It would be interesting to know if an IGT was actually
> regularly 
> showing the issue and we had just been ignoring it!

It was Alderlake. It wasn't igt test. Opened several browser
instances: 

https://webglsamples.org/aquarium/aquarium.html

Took max. couple of minutes to reproduce. For some reason PSR2
DEEP_SLEEP entry/exit was some kind of trigger for this. Without PSR2
DEEP_SLEEP I couldn't reproduce the issue.

BR,

Jouni Högander


> John.
> 
> > 
> > > v2: Dont use BAR mappings either.
> > > Make conditional on LLC so as not to change platforms that don't
> > > need
> > > to change (Daniele).
> > > Add 'Fixes' tags (Tvrtko).
> > > v3: Fix dumb typo.
> > > 
> > > Signed-off-by: John Harrison <John.C.Harrison at Intel.com>
> > > 
> > > 
> > > John Harrison (2):
> > >    drm/i915: Don't use stolen memory for ring buffers with LLC
> > >    drm/i915: Don't use BAR mappings for ring buffers with LLC
> > > 
> > >   drivers/gpu/drm/i915/gt/intel_ring.c | 6 +++---
> > >   1 file changed, 3 insertions(+), 3 deletions(-)
> > > 
> 



More information about the Intel-gfx mailing list