drm/amdgpu: fix fundamental suspend/resume issue

Reinitializing the VM manager during suspend/resume is a very very bad
idea since all the VMs are still active and kicking.

This can lead to random VM faults after resume when new processes
become the same client ID assigned.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

1 files changed, 21 insertions, 1 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 07ff3b1514f1..1bf36c3542c1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -672,6 +672,7 @@ void amdgpu_vm_reset_id(struct amdgpu_device *adev, unsigned vmhub,
 	struct amdgpu_vm_id_manager *id_mgr = &adev->vm_manager.id_mgr[vmhub];
 	struct amdgpu_vm_id *id = &id_mgr->ids[vmid];
 
+	atomic64_set(&id->owner, 0);
 	id->gds_base = 0;
 	id->gds_size = 0;
 	id->gws_base = 0;
@@ -681,6 +682,26 @@ void amdgpu_vm_reset_id(struct amdgpu_device *adev, unsigned vmhub,
 }
 
 /**
+ * amdgpu_vm_reset_all_id - reset VMID to zero
+ *
+ * @adev: amdgpu device structure
+ *
+ * Reset VMID to force flush on next use
+ */
+void amdgpu_vm_reset_all_ids(struct amdgpu_device *adev)
+{
+	unsigned i, j;
+
+	for (i = 0; i < AMDGPU_MAX_VMHUBS; ++i) {
+		struct amdgpu_vm_id_manager *id_mgr =
+			&adev->vm_manager.id_mgr[i];
+
+		for (j = 1; j < id_mgr->num_ids; ++j)
+			amdgpu_vm_reset_id(adev, i, j);
+	}
+}
+
+/**
  * amdgpu_vm_bo_find - find the bo_va for a specific vm & bo
  *
  * @vm: requested vm
@@ -2270,7 +2291,6 @@ void amdgpu_vm_manager_init(struct amdgpu_device *adev)
 	for (i = 0; i < AMDGPU_MAX_RINGS; ++i)
 		adev->vm_manager.seqno[i] = 0;
 
-
 	atomic_set(&adev->vm_manager.vm_pte_next_ring, 0);
 	atomic64_set(&adev->vm_manager.client_counter, 0);
 	spin_lock_init(&adev->vm_manager.prt_lock);