diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 20 |
1 files changed, 15 insertions, 5 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c index 025625e7e800..eb8bbfc7e6d9 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -2684,12 +2684,22 @@ void amdgpu_sdma_set_buffer_funcs_scheds(struct amdgpu_device *adev, return; } - /* Navi1x's workaround requires us to limit to a single SDMA sched - * for ttm. - */ hub = &adev->vmhub[AMDGPU_GFXHUB(0)]; - adev->mman.num_buffer_funcs_scheds = hub->sdma_invalidation_workaround ? - 1 : n; + + /* + * Allow using multiple SDMA schedulers only on GPUs where + * we are allowed to do concurrent VM flushes. + * This consideration is necessary because all GART windows + * are mapped in VMID 0 (the kernel VMID) so each buffer + * entity would flush VMID 0 concurrently. + * + * Also consider the SDMA invalidation workaround on + * Navi 1x GPUs, which also prevents us from using + * multiple SDMA engines on VMID 0 at the same time. + */ + adev->mman.num_buffer_funcs_scheds = + (adev->vm_manager.concurrent_flush && + !hub->sdma_invalidation_workaround) ? n : 1; } #if defined(CONFIG_DEBUG_FS) |