Skip to contentSkip to conversation

drm/amdgpu: Update PDEs flush TLB if PDEs is moved

PATCH Superseded Philip Yang 3 participants 3 messages Jun 1, 2022 – Jun 2, 2022

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/amdgpu: Update PDEs flush TLB if PDEs is moved

Update PDEs, PTEs don't need flush TLB after updating mapping, this will remove the unnecessary TLB flush to reduce map to GPUs time.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +4 -3
index 9596c22fded6..8cdfd09fd70d 100644
@@ -755,6 +755,10 @@ int amdgpu_vm_update_pdes(struct amdgpu_device *adev,
 		goto error;
 
 	list_for_each_entry(entry, &vm->relocated, vm_status) {
+		/* vm_flush_needed after updating moved PDEs */
+		if (entry->moved)
+			atomic64_inc(&vm->tlb_seq);
+
 		r = amdgpu_vm_pde_update(&params, entry);
 		if (r)
 			goto error;
@@ -764,9 +768,6 @@ int amdgpu_vm_update_pdes(struct amdgpu_device *adev,
 	if (r)
 		goto error;
 
-	/* vm_flush_needed after updating PDEs */
-	atomic64_inc(&vm->tlb_seq);
-
 	while (!list_empty(&vm->relocated)) {
 		entry = list_first_entry(&vm->relocated,
 					 struct amdgpu_vm_bo_base,
Signature · 1 line
2.35.1
#1 lore raw

Am 2022-06-01 um 19:12 schrieb Philip Yang:

Quoted 2 lines

Update PDEs, PTEs don't need flush TLB after updating mapping, this will remove the unnecessary TLB flush to reduce map to GPUs time.

This description is unclear. I think what this change does is, flush TLBs when existing PDEs are updated (because a PTB or PDB moved). But it avoids unnecessary TLB flushes when new PDBs or PTBs are added to the page table, which commonly happens when memory is mapped for the first time.

Regards,
  Felix

Quoted 32 lines
---
  drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 7 ++++---
  1 file changed, 4 insertions(+), 3 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +4 -3
index 9596c22fded6..8cdfd09fd70d 100644
@@ -755,6 +755,10 @@ int amdgpu_vm_update_pdes(struct amdgpu_device *adev,
  		goto error;
 
  	list_for_each_entry(entry, &vm->relocated, vm_status) {
+		/* vm_flush_needed after updating moved PDEs */
+		if (entry->moved)
+			atomic64_inc(&vm->tlb_seq);
+
  		r = amdgpu_vm_pde_update(&params, entry);
  		if (r)
  			goto error;
@@ -764,9 +768,6 @@ int amdgpu_vm_update_pdes(struct amdgpu_device *adev,
  	if (r)
  		goto error;
 
-	/* vm_flush_needed after updating PDEs */
-	atomic64_inc(&vm->tlb_seq);
-
  	while (!list_empty(&vm->relocated)) {
  		entry = list_first_entry(&vm->relocated,
  					 struct amdgpu_vm_bo_base,
maintainer
#2 lore raw

Am 02.06.22 um 01:19 schrieb Felix Kuehling:

Quoted 5 lines

Am 2022-06-01 um 19:12 schrieb Philip Yang:

Quoted 2 lines

Update PDEs, PTEs don't need flush TLB after updating mapping, this will remove the unnecessary TLB flush to reduce map to GPUs time.

This description is unclear. I think what this change does is, flush TLBs when existing PDEs are updated (because a PTB or PDB moved). But it avoids unnecessary TLB flushes when new PDBs or PTBs are added to the page table, which commonly happens when memory is mapped for the first time.

Yes, agree. Additional to that (I know reviewing my own suggestion) setting a local variable and then incrementing the atomic only once might be better because atomics are barriers on x86.

Regards,
Christian.

Quoted 37 lines

Regards,
  Felix

Quoted 32 lines

---
  drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 7 ++++---

  1 file changed, 4 insertions(+), 3 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +4 -0
index 9596c22fded6..8cdfd09fd70d 100644
@@ -755,6 +755,10 @@ int amdgpu_vm_update_pdes(struct amdgpu_device 
 *adev,
           goto error;
         list_for_each_entry(entry, &vm->relocated, vm_status) {
+        /* vm_flush_needed after updating moved PDEs */
+        if (entry->moved)
+            atomic64_inc(&vm->tlb_seq);
+
           r = amdgpu_vm_pde_update(&params, entry);
           if (r)
               goto error;
@@ -764,9 +768,6 @@ int amdgpu_vm_update_pdes(struct amdgpu_device 
 *adev,
       if (r)
           goto error;

  -    /* vm_flush_needed after updating PDEs */ -    atomic64_inc(&vm->tlb_seq);
-
      while (!list_empty(&vm->relocated)) { entry = list_first_entry(&vm->relocated, struct amdgpu_vm_bo_base,