00:49 <lumag> aka_[m], so, I too a glance. On a510 we should not use preempt at all, so nr_rings should be 1

00:51 <lumag> A506 should support preemption, so it is slightly more interesting

01:00 <lumag> Regarding a510. Does it work if you force nr_rings to 1? Also could you please try forcing prio = 0 at msm_submitqueue_create() ?

01:02 <lumag> if you can capture devcoredumps for your platforms, I can try taking a glance (enable devcoredump, make adreno hang/crash/whatever, then capture dmesg and /sys/class/devcoredump/*/data (it is useful to copy it quick, then dump/etc)

01:04 <lumag> Also I see that for A506/A510 (and all other a5xx) we should ping several regs on suspend/resume, which we do not do.

07:28 <aka_[m]> lumag: after forcing num rings to 1 gpu works fine, obv we have artifacts and shit, but that's common for all.

07:28 <aka_[m]> For a506 I will give it a try next week.

10:23 <lumag> aka_[m], please send a patch limiting num_rings to 1 on a510

10:41 <aka_[m]> lumag: would that fit to patch a5xx_ucode_check_version to return has_whereami to false if gpu_is_a510?

10:42 <lumag> No. Pleas change a5xx_gpu_init() to check for a510 and pass 1 rather than 4 to adreno_gpu_init.

10:43 <lumag> It should be enough to make it work. Everything else is a bug

10:50 <lumag> aka_[m], I'd rather unsigned int nr_rings = 4; if (adreno_is_a510(...)) nr_rings = 1;

10:51 <aka_[m]> ok i get it... (full message at <https://matrix.org/_matrix/media/v3/download/matrix.org/VwBpisNGciRDiTSKcGlllnMm>)

10:58 <lumag> aka_[m], yes

10:58 <aka_[m]> that should count as fix or not?

10:58 <lumag> Add an empty line after the variable declaration as a courtesy for kernel developers and that's good

10:58 <lumag> Yes.

10:58 <aka_[m]> fix for adding a510 support i believe?

10:58 <lumag> Yes

10:59 <aka_[m]> ok will send next week.

10:59 <lumag> :-)