From: Mukul Joshi <[email protected]> Convert amdgpu_ualink_invalidate_import_mappings() to lock the BO and every client VM root PD in a single drm_exec transaction instead of borrowing the caller's reservation ticket via dma_resv_locking_ctx().
drm_exec owns its ww_acquire_ctx (exec.ticket) for the lifetime of the transaction, so amdgpu_vm_handle_moved() now uses &exec.ticket. Also, drop the explicit dma_resv_reserve_fences call as amdgpu_vm_lock_pd() already reserves 2 + num_fences on the same VM root PD. Signed-off-by: Mukul Joshi <[email protected]> Reviewed-by: Harish Kasiviswanathan <[email protected]> Signed-off-by: Alex Deucher <[email protected]> --- drivers/gpu/drm/amd/amdgpu/amdgpu_ualink.c | 63 +++++++++++----------- 1 file changed, 30 insertions(+), 33 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ualink.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ualink.c index 9d21f08123eed..82a362e44a3fe 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ualink.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ualink.c @@ -1763,68 +1763,65 @@ static void amdgpu_ualink_exp_cleanup_worker(struct work_struct *work) */ static void amdgpu_ualink_invalidate_import_mappings(struct amdgpu_bo *bo) { - struct drm_gem_object *obj = &bo->tbo.base; - struct ww_acquire_ctx *ticket = dma_resv_locking_ctx(obj->resv); struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); struct ttm_operation_ctx ctx = { false, false }; struct ttm_placement placement = {}; struct amdgpu_vm_bo_base *bo_base; + struct drm_exec exec; int r; - amdgpu_bo_reserve(bo, false); + /* + * Lock the BO together with every client VM page directory it is + * mapped into in a single drm_exec transaction. + */ + drm_exec_init(&exec, DRM_EXEC_IGNORE_DUPLICATES, 0); + drm_exec_until_all_locked(&exec) { + r = drm_exec_lock_obj(&exec, &bo->tbo.base); + drm_exec_retry_on_contention(&exec); + if (unlikely(r)) + goto fini; + + for (bo_base = bo->vm_bo; bo_base; bo_base = bo_base->next) { + r = amdgpu_vm_lock_pd(bo_base->vm, &exec, 0); + drm_exec_retry_on_contention(&exec); + if (unlikely(r)) + goto fini; + } + } /* FIXME: This should be after the "if", but needs a fix to make sure * DMABuf imports are initialized in the right VM list. */ amdgpu_vm_bo_invalidate(bo, false); if (!bo->tbo.resource || bo->tbo.resource->mem_type == TTM_PL_SYSTEM) - goto out; + goto fini; r = ttm_bo_validate(&bo->tbo, &placement, &ctx); if (r) { dev_err(adev->dev, "Failed to invalidate NPA DMA-buf import (%d)\n", r); - goto out; + goto fini; } for (bo_base = bo->vm_bo; bo_base; bo_base = bo_base->next) { struct amdgpu_vm *vm = bo_base->vm; - struct dma_resv *resv = vm->root.bo->tbo.base.resv; - - if (ticket) { - /* When we get an error here it means that somebody - * else is holding the VM lock and updating page tables - * So we can just continue here. - */ - r = dma_resv_lock(resv, ticket); - if (r) - continue; - - } else { - /* TODO: This is more problematic and we actually need - * to allow page tables updates without holding the - * lock. - */ - if (!dma_resv_trylock(resv)) - continue; - } - /* Reserve fences for two SDMA page table updates */ - r = dma_resv_reserve_fences(resv, 2); - if (!r) - r = amdgpu_vm_clear_freed(adev, vm, NULL); + /* + * Fences for the two SDMA page table updates were already + * reserved by amdgpu_vm_lock_pd() above (it reserves 2 + + * num_fences on the same VM root PD dma_resv). + */ + r = amdgpu_vm_clear_freed(adev, vm, NULL); if (!r) - r = amdgpu_vm_handle_moved(adev, vm, ticket); + r = amdgpu_vm_handle_moved(adev, vm, &exec.ticket); if (r && r != -EBUSY) dev_err(adev->dev, "Failed to invalidate VM page tables (%d))\n", r); - - dma_resv_unlock(resv); } -out: - amdgpu_bo_unreserve(bo); +fini: + drm_exec_fini(&exec); } static int amdgpu_ualink_map_npa_to_dmabuf(struct amdgpu_device *adev, -- 2.55.0
