Re: [PATCH] drm/amdgpu: fix TLB flushing during eviction

2022-04-03 Thread Christian König
Am 01.04.22 um 21:47 schrieb Felix Kuehling: On 2022-04-01 04:24, Christian König wrote: Am 31.03.22 um 16:37 schrieb Felix Kuehling: Am 2022-03-31 um 02:27 schrieb Christian König: Am 30.03.22 um 22:51 schrieb philip yang: On 2022-03-30 05:00, Christian König wrote: Testing the valid bit

Re: [PATCH] drm/radeon: change si_default_state table from global to static

2022-04-03 Thread Christian König
Am 02.04.22 um 18:27 schrieb Tom Rix: Smatch reports these issues si_blit_shaders.c:31:11: warning: symbol 'si_default_state' was not declared. Should it be static? si_blit_shaders.c:253:11: warning: symbol 'si_default_size' was not declared. Should it be static? Both symbols are only used

Re: [PATCH 1/1] drm/amdgpu: Flush TLB after mapping for VG20+XGMI

2022-04-03 Thread Christian König
Am 01.04.22 um 21:57 schrieb Philip Yang: For VG20 + XGMI bridge, all mappings PTEs cache in TC, this may have stall invalid PTEs in TC because one cache line has 8 pages. Need always flush_tlb after updating mapping. Signed-off-by: Philip Yang You could drop the extra (), but either way Revi

[Bug][5.18-rc0] Between commits ed4643521e6a and 34af78c4e616, appears warning "WARNING: CPU: 31 PID: 51848 at drivers/dma-buf/dma-fence-array.c:191 dma_fence_array_create+0x101/0x120" and some games

2022-04-03 Thread Mikhail Gavrilov
Hi, Between commits ed4643521e6a and 34af78c4e616 something was broken. I noted that kernel log flooded with warning message "WARNING: CPU: 31 PID: 51848 at drivers/dma-buf/dma-fence-array.c:191 dma_fence_array_create+0x101/0x120" when some games are running: "Resident Evil Village", "Marvel's Aven

Re: [PATCH 5/5] drm/amdgpu: allow specifying a syncobj for VM map operations

2022-04-03 Thread Bas Nieuwenhuizen
Hi Christian, Couple of concerns here: 1) This is missing the option to wait on a syncobj before executing the VA op. 2) There are no mechanisms to disable implicit sync yet. 3) in radv we typically signal multiple syncobj, so it would be nice if we could have that capability here too. Technicall

Re: [Bug][5.18-rc0] Between commits ed4643521e6a and 34af78c4e616, appears warning "WARNING: CPU: 31 PID: 51848 at drivers/dma-buf/dma-fence-array.c:191 dma_fence_array_create+0x101/0x120" and some ga

2022-04-03 Thread Christian König
Hi Mikhail, those are two independent and already known problems. The warning triggered from the sync_file is already fixed in drm-misc-next-fixes, but so far I couldn't figure out why the games suddenly doesn't work any more. There is a bug report for that, but bisecting the changes didn't

Re: [PATCH 5/5] drm/amdgpu: allow specifying a syncobj for VM map operations

2022-04-03 Thread Christian König
Hi Bas, Am 04.04.22 um 03:32 schrieb Bas Nieuwenhuizen: Hi Christian, Couple of concerns here: 1) This is missing the option to wait on a syncobj before executing the VA op. Uff, that was not really planned in any way. Currently all VM operations are scheduled to run behind all CS operatio