Skip to contentSkip to conversation

drm/amdgpu: Use atomic64_t type for pdd->tlb_seq

PATCH Superseded Philip Yang 3 participants 4 messages Mar 26, 2022 – Mar 28, 2022

Subsystems (3)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 drm/amdgpu: Use atomic64_t type for pdd->tlb_seq

To support multi-thread update page table.

---
 drivers/gpu/drm/amd/amdkfd/kfd_priv.h    | 2 +-
 drivers/gpu/drm/amd/amdkfd/kfd_process.c | 6 +++---
 2 files changed, 4 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_priv.h +1 -1
index 945982a5d688..e1b7e6afa920 100644
@@ -705,7 +705,7 @@ struct kfd_process_device {
 	/* VM context for GPUVM allocations */
 	struct file *drm_file;
 	void *drm_priv;
-	uint64_t tlb_seq;
+	atomic64_t tlb_seq;
 
 	/* GPUVM allocations storage */
 	struct idr alloc_idr;
drivers/gpu/drm/amd/amdkfd/kfd_process.c +3 -3
index ac8123c1ee8f..43ed8ec1f975 100644
@@ -1560,7 +1560,7 @@ int kfd_process_device_init_vm(struct kfd_process_device *pdd,
 		return ret;
 	}
 	pdd->drm_priv = drm_file->private_data;
-	pdd->tlb_seq = 0;
+	atomic64_set(&pdd->tlb_seq, 0);
 
 	ret = kfd_process_device_reserve_ib_mem(pdd);
 	if (ret)
@@ -1954,10 +1954,10 @@ void kfd_flush_tlb(struct kfd_process_device *pdd, enum TLB_FLUSH_TYPE type)
 	uint64_t tlb_seq = amdgpu_vm_tlb_seq(vm);
 	struct kfd_dev *dev = pdd->dev;
 
-	if (pdd->tlb_seq == tlb_seq)
+	if (atomic64_read(&pdd->tlb_seq) == tlb_seq)
 		return;
 
-	pdd->tlb_seq = tlb_seq;
+	atomic64_set(&pdd->tlb_seq, tlb_seq);
 	if (dev->dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS) {
 		/* Nothing to flush until a VMID is assigned, which
 		 * only happens when the first queue is created.
Signature · 1 line
2.35.1
#1 lore raw

PATCH 2/2 drm/amdgpu: Use last_unlocked fence for unlocked update

MMU callback update page table set unlocked flag, add callback tlb_cb to vm->last_unlocked fence, and pass it back to upper layer to wait for page table update done.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +4 -1
index 48f326609976..aac1b625194f 100644
@@ -927,7 +927,10 @@ int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 
 	r = vm->update_funcs->commit(&params, fence);
 
-	if (!unlocked && (!(flags & AMDGPU_PTE_VALID) || params.table_freed)) {
+	if (!(flags & AMDGPU_PTE_VALID) || params.table_freed) {
+		if (unlocked && fence)
+			*fence = dma_fence_get(vm->last_unlocked);
+
 		tlb_cb->vm = vm;
 		if (!fence || !*fence ||
 		    dma_fence_add_callback(*fence, &tlb_cb->cb,
Signature · 1 line
2.35.1
maintainer
#2 lore raw

Re: [PATCH 2/2] drm/amdgpu: Use last_unlocked fence for unlocked update

Am 26.03.22 um 01:56 schrieb Philip Yang:

Quoted 14 lines

MMU callback update page table set unlocked flag, add callback tlb_cb to vm->last_unlocked fence, and pass it back to upper layer to wait for page table update done.

---
  drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 5 ++++-
  1 file changed, 4 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0 -0
index 48f326609976..aac1b625194f 100644

drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, · line 930

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +1 -1
@@ -927,7 +927,10 @@ int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 
  	r = vm->update_funcs->commit(&params, fence);
 
-	if (!unlocked && (!(flags & AMDGPU_PTE_VALID) || params.table_freed)) {
+	if (!(flags & AMDGPU_PTE_VALID) || params.table_freed) {

Dropping the check for the unlocked flag here seems to be ok.

drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, · line 933

Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +3 -0
+		if (unlocked && fence)
+			*fence = dma_fence_get(vm->last_unlocked);
+

But that doesn't seems to make sense, fence should already contain the necessary fence object if any.

Regards,
Christian.

Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0 -0
  		tlb_cb->vm = vm;
  		if (!fence || !*fence ||
  		    dma_fence_add_callback(*fence, &tlb_cb->cb,
#3 lore raw

Re: [PATCH 2/2] drm/amdgpu: Use last_unlocked fence for unlocked update

On 2022-03-28 5:21 a.m., Christian König wrote:
Am 26.03.22 um 01:56 schrieb Philip Yang:
MMU callback update page table set unlocked flag, add callback tlb_cb to vm->last_unlocked fence, and pass it back to upper layer to wait for page table update done.

---
  drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 5 ++++-

  1 file changed, 4 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +0 -0
index 48f326609976..aac1b625194f 100644
@@ -927,7 +927,10 @@ int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,

        r = vm->update_funcs->commit(&params, fence);
  -    if (!unlocked && (!(flags & AMDGPU_PTE_VALID) || params.table_freed)) { +    if (!(flags & AMDGPU_PTE_VALID) || params.table_freed) {

Dropping the check for the unlocked flag here seems to be ok.

+        if (unlocked && fence)
+            *fence = dma_fence_get(vm->last_unlocked); +

But that doesn't seems to make sense, fence should already contain the necessary fence object if any. You are right, seems there is another bug, will send out v2 patch to fix the bug and remove this.
Regards,
Philip

Regards,
Christian.

          tlb_cb->vm = vm;
          if (!fence || !*fence || dma_fence_add_callback(*fence, &tlb_cb->cb,