[PATCH 1/2] drm/amdgpu: re-validate per VM BOs if required v2

Christian König ckoenig.leichtzumerken at gmail.com
Thu Apr 5 12:19:03 UTC 2018


If a per VM BO ends up in a allowed domain it never moves back into the
prefered domain.

v2: move the extra handling into amdgpu_vm_bo_update when we exit the
    state machine. Make memory type handling generic.

Signed-off-by: Christian König <christian.koenig at amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 24474294c92a..a2f8b57838ff 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -1555,7 +1555,20 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev,
 	}
 
 	spin_lock(&vm->status_lock);
-	list_del_init(&bo_va->base.vm_status);
+	if (bo && bo->tbo.resv == vm->root.base.bo->tbo.resv) {
+		unsigned mem_type = bo->tbo.mem.mem_type;
+
+		/* If the BO is not in its preferred location add it back to
+		 * the evicted list so that it gets validated again on the
+		 * next command submission.
+		 */
+		if (!(bo->preferred_domains & amdgpu_mem_type_to_domain(mem_type)))
+			list_add_tail(&bo_va->base.vm_status, &vm->evicted);
+		else
+			list_del_init(&bo_va->base.vm_status);
+	} else {
+		list_del_init(&bo_va->base.vm_status);
+	}
 	spin_unlock(&vm->status_lock);
 
 	list_splice_init(&bo_va->invalids, &bo_va->valids);
-- 
2.14.1



More information about the amd-gfx mailing list