| Message ID | 20220211202728.6146-6-alyssa.rosenzweig@collabora.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | drm/panfrost: Initial Valhall support | expand |
On 11/02/2022 20:27, alyssa.rosenzweig@collabora.com wrote: > From: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com> > > TTRX_3485 requires the infamous "dummy job" workaround. I have this > workaround implemented in a local branch, but I have not yet hit a case > that requires it so I cannot test whether the implementation is correct. > In the mean time, add the quirk bit so we can document which platforms > may need it in the future. This one is hideous ;) Although to me this isn't the 'infamous' one as it's not the earliest example of a dummy job. However... I believe as Panfrost currently stands this is probably not very possible to hit. It requires a job to be stopped (soft or hard) at a critical point during submission - which at the moment Panfrost basically never does (the exception is if you close the fd immediately while a job is in progress). And of course the timing has to be 'just right' to hit the bug. That said I think we should probably add pre-emption support sometime at which point this could become an issue. > Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com> Reviewed-by: Steven Price <steven.price@arm.com> > --- > drivers/gpu/drm/panfrost/panfrost_issues.h | 3 +++ > 1 file changed, 3 insertions(+) > > diff --git a/drivers/gpu/drm/panfrost/panfrost_issues.h b/drivers/gpu/drm/panfrost/panfrost_issues.h > index 058f6a4c8435..b8865fc9efce 100644 > --- a/drivers/gpu/drm/panfrost/panfrost_issues.h > +++ b/drivers/gpu/drm/panfrost/panfrost_issues.h > @@ -132,6 +132,9 @@ enum panfrost_hw_issue { > * to hang */ > HW_ISSUE_TTRX_3076, > > + /* Must issue a dummy job before starting real work to prevent hangs */ > + HW_ISSUE_TTRX_3485, > + > HW_ISSUE_END > }; >
> > TTRX_3485 requires the infamous "dummy job" workaround. I have this > > workaround implemented in a local branch, but I have not yet hit a case > > that requires it so I cannot test whether the implementation is correct. > > In the mean time, add the quirk bit so we can document which platforms > > may need it in the future. > > This one is hideous ;) Although to me this isn't the 'infamous' one as > it's not the earliest example of a dummy job. Terrifying. I guess we narrowly avoided the 'replay' workaround which was far worse than this one... > However... I believe as Panfrost currently stands this is probably not > very possible to hit. It requires a job to be stopped (soft or hard) at > a critical point during submission - which at the moment Panfrost > basically never does (the exception is if you close the fd immediately > while a job is in progress). And of course the timing has to be 'just > right' to hit the bug. OK, that's good to know. Still "should" be fixed but that definitely lowers the priority of it. Frankly the multithreading bugs we have on the CPU side would hang the machine sooner...
diff --git a/drivers/gpu/drm/panfrost/panfrost_issues.h b/drivers/gpu/drm/panfrost/panfrost_issues.h index 058f6a4c8435..b8865fc9efce 100644 --- a/drivers/gpu/drm/panfrost/panfrost_issues.h +++ b/drivers/gpu/drm/panfrost/panfrost_issues.h @@ -132,6 +132,9 @@ enum panfrost_hw_issue { * to hang */ HW_ISSUE_TTRX_3076, + /* Must issue a dummy job before starting real work to prevent hangs */ + HW_ISSUE_TTRX_3485, + HW_ISSUE_END };