All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH v3] drm/amdgpu: clear root PT after setting vm->is_npa = true
@ 2026-11-19 18:46 Shahyan Soltani
  2026-11-19 18:55 ` sashiko-bot
  2026-11-20 22:01 ` Felix Kuehling
  0 siblings, 2 replies; 3+ messages in thread
From: Shahyan Soltani @ 2026-11-19 18:46 UTC (permalink / raw)
  To: amd-gfx; +Cc: felix.kuehling, christian.koenig, alexander.deucher, yangp

Previously, the root PT was never cleared after setting vm->is_npa =
true within amdgpu_vm_make_npa(). This resulted in the flag
adev->gmc.no_retry_flags to never be applied for the NPA VMs.

Call amdgpu_vm_pt_clear() after vm->is_npa = true is set and the bo has
been pinned/validated. This allows the root PT of the NPA VM to be
cleared with the correct value.

If the pinning or reservation of the bo failed, the vm->is_npa flag will
no longer remain as true. This is to avoid conditionals outside of
amdgpu_vm_make_npa() that depend on the vm->is_npa flag to determine if
the VM is a correctly initialized NPA VM.

Warn if the VM already has mappings, since clearing the root PT will
break existing mappings.

Remove the redundant second setting of vm->is_npa = true.

v2:
- Add the WARN_ON sanity check regarding an empty VM (Felix)
- Change the return handling to only set vm->is_npa = true upon a
  successful pin (Felix)
- Add the heavy weight TLB flush after the root PT clear (Philip)

v3:
- Remove TLB flush (Felix)
- Move WARN_ON sanity check to start to return early (Christian)
- Reserve fence slots for amdgpu_vm_pt_clear() and amdgpu_bo_pin()
  by replacing amdgpu_bo_reserve with amdgpu_vm_lock_pd()
  (Christian)
- Use goto error handling rather than calling return when applicable

Suggested-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Shahyan Soltani <shahyan.soltani@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 36 ++++++++++++++++++++------
 1 file changed, 28 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 94b1cfe05e3e..d9e3916feea6 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -2820,23 +2820,43 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm)
  */
 int amdgpu_vm_make_npa(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 {
-	int r = amdgpu_vm_make_compute(adev, vm);
+	struct drm_exec exec;
+	int r;
 
+	/* Check if VM mappings are empty, clearing would break existing mapping */
+	if (WARN_ON(!RB_EMPTY_ROOT(&vm->va.rb_root)))
+		return -EINVAL;
+
+	r = amdgpu_vm_make_compute(adev, vm);
 	if (r)
-		return r;
-	vm->is_npa = true;
-	r = amdgpu_bo_reserve(vm->root.bo, false);
-	if (r)
-		return r;
+		goto error;
+
+	drm_exec_init(&exec, 0, 0);
+	drm_exec_until_all_locked(&exec) {
+		r = amdgpu_vm_lock_pd(vm, &exec, 0);
+		drm_exec_retry_on_contention(&exec);
+		if (unlikely(r))
+			goto error_unlock;
+	}
+
 	r = amdgpu_bo_pin(vm->root.bo, AMDGPU_GEM_DOMAIN_VRAM);
-	amdgpu_bo_unreserve(vm->root.bo);
 	if (r)
-		return r;
+		goto error_unlock;
 
 	vm->is_npa = true;
+	r = amdgpu_vm_pt_clear(adev, vm, to_amdgpu_bo_vm(vm->root.bo), false);
+	if (r)
+		goto error_unlock;
+
+	drm_exec_fini(&exec);
 	vm->need_tlb_fence = false;
 
 	return 0;
+
+error_unlock:
+	drm_exec_fini(&exec);
+error:
+	return r;
 }
 
 static int amdgpu_vm_stats_is_zero(struct amdgpu_vm *vm)
-- 
2.56.0


^ permalink raw reply related	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2026-11-20 22:02 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-19 18:46 [PATCH v3] drm/amdgpu: clear root PT after setting vm->is_npa = true Shahyan Soltani
2026-11-19 18:55 ` sashiko-bot
2026-11-20 22:01 ` Felix Kuehling

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.