On 2018å¹´04æ??10æ?¥ 17:00, Christian König wrote: > Am 10.04.2018 um 04:43 schrieb zhoucm1: >> >> >> On 2018å¹´04æ??09æ?¥ 18:19, Christian König wrote: >>> That should purely be handled by preferred/allowed domains. >> Although this flag isn't exported to user space yet, I'm curious that >> how preferred/allowed domains handle no_fallback? >> IIRC, currently, our driver will always add GTT fallback for VRAM bo. > > And that is intentional. Going a step further back I think moving the > fallback handling into amdgpu_bo_do_create() and adding the flag was a > mistake to begin with. > > Going to send patches to revert all this and further clean the stuff up. if you are able to not change the preferred domain when fallback, it's no problem to me. David Zhou > > Regards, > Christian. > >> >> Regards, >> David Zhou >>> >>> Signed-off-by: Christian König <christian.koenig at amd.com> >>> --- >>>  drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 3 +-- >>>  1 file changed, 1 insertion(+), 2 deletions(-) >>> >>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c >>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c >>> index 68af2f878bc9..e1756b68a17b 100644 >>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c >>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c >>> @@ -385,8 +385,7 @@ static int amdgpu_cs_bo_validate(struct >>> amdgpu_cs_parser *p, >>>          amdgpu_bo_in_cpu_visible_vram(bo)) >>>          p->bytes_moved_vis += ctx.bytes_moved; >>>  -   if (unlikely(r == -ENOMEM) && domain != bo->allowed_domains && >>> -       !(bo->flags & AMDGPU_GEM_CREATE_NO_FALLBACK)) { >>> +   if (unlikely(r == -ENOMEM) && domain != bo->allowed_domains) { >>>          domain = bo->allowed_domains; >>>          goto retry; >>>      } >> >