Skip to contentSkip to conversation

drm/amdgpu: re-enable retry faults

PATCH In review Christian König 4 participants 7 messages Mar 12, 2019 – Mar 13, 2019 4 messages not in the archive

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/3 drm/amdgpu: re-enable retry faults

Now that we have re-reoute faults to the other IH
ring we can enable retries again.

---
 drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c | 2 +-
 drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c  | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +1 -1
index c10ed568ca6c..7bb5359d0bbd 100644
@@ -236,7 +236,7 @@ static void gfxhub_v1_0_setup_vmid_config(struct amdgpu_device *adev)
 				    block_size);
 		/* Send no-retry XNACK on fault to suppress VM fault storm. */
 		tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
-				    RETRY_PERMISSION_OR_INVALID_PAGE_FAULT, 0);
+				    RETRY_PERMISSION_OR_INVALID_PAGE_FAULT, 1);
 		WREG32_SOC15_OFFSET(GC, 0, mmVM_CONTEXT1_CNTL, i, tmp);
 		WREG32_SOC15_OFFSET(GC, 0, mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32, i*2, 0);
 		WREG32_SOC15_OFFSET(GC, 0, mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32, i*2, 0);
drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +1 -1
index 2a039946a549..1741056e6af6 100644
@@ -255,7 +255,7 @@ static void mmhub_v1_0_setup_vmid_config(struct amdgpu_device *adev)
 				    block_size);
 		/* Send no-retry XNACK on fault to suppress VM fault storm. */
 		tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
-				    RETRY_PERMISSION_OR_INVALID_PAGE_FAULT, 0);
+				    RETRY_PERMISSION_OR_INVALID_PAGE_FAULT, 1);
 		WREG32_SOC15_OFFSET(MMHUB, 0, mmVM_CONTEXT1_CNTL, i, tmp);
 		WREG32_SOC15_OFFSET(MMHUB, 0, mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32, i*2, 0);
 		WREG32_SOC15_OFFSET(MMHUB, 0, mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32, i*2, 0);
Signature · 6 lines
2.17.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): 20190312123749.6249-1-christian.koenig-5C7GfCeVMHo@public.gmane.org

#2 lore raw

PATCH 2/3 drm/amdgpu: free up the first paging queue

We need the first paging queue to handle page faults.

---
 drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c | 20 ++++++++++++--------
 1 file changed, 12 insertions(+), 8 deletions(-)
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +12 -8
index 3ac5abe937f4..bed18e7bbc36 100644
@@ -2266,7 +2266,7 @@ static void sdma_v4_0_set_buffer_funcs(struct amdgpu_device *adev)
 {
 	adev->mman.buffer_funcs = &sdma_v4_0_buffer_funcs;
 	if (adev->sdma.has_page_queue)
-		adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].page;
+		adev->mman.buffer_funcs_ring = &adev->sdma.instance[1].page;
 	else
 		adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
 }
@@ -2285,15 +2285,19 @@ static void sdma_v4_0_set_vm_pte_funcs(struct amdgpu_device *adev)
 	unsigned i;
 
 	adev->vm_manager.vm_pte_funcs = &sdma_v4_0_vm_pte_funcs;
-	for (i = 0; i < adev->sdma.num_instances; i++) {
-		if (adev->sdma.has_page_queue)
-			sched = &adev->sdma.instance[i].page.sched;
-		else
-			sched = &adev->sdma.instance[i].ring.sched;
-		adev->vm_manager.vm_pte_rqs[i] =
+	if (adev->sdma.has_page_queue) {
+		sched = &adev->sdma.instance[1].page.sched;
+		adev->vm_manager.vm_pte_rqs[0] =
 			&sched->sched_rq[DRM_SCHED_PRIORITY_KERNEL];
+		adev->vm_manager.vm_pte_num_rqs = 1;
+	} else {
+		for (i = 0; i < adev->sdma.num_instances; i++) {
+			sched = &adev->sdma.instance[i].ring.sched;
+			adev->vm_manager.vm_pte_rqs[i] =
+				&sched->sched_rq[DRM_SCHED_PRIORITY_KERNEL];
+		}
+		adev->vm_manager.vm_pte_num_rqs = adev->sdma.num_instances;
 	}
-	adev->vm_manager.vm_pte_num_rqs = adev->sdma.num_instances;
 }
 
 const struct amdgpu_ip_block_version sdma_v4_0_ip_block = {
Signature · 6 lines
2.17.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): 20190312123749.6249-2-christian.koenig-5C7GfCeVMHo@public.gmane.org

#4 lore raw

RE: [PATCH 2/3] drm/amdgpu: free up the first paging queue

I think this would break Raven, which only has one SDMA engine.

Regards,
Felix

-----Original Message-----
From: amd-gfx <[email protected]> On Behalf Of Christian König
Sent: Tuesday, March 12, 2019 8:38 AM
To: [email protected]
Subject: [PATCH 2/3] drm/amdgpu: free up the first paging queue

We need the first paging queue to handle page faults.

---
 drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c | 20 ++++++++++++--------
 1 file changed, 12 insertions(+), 8 deletions(-)
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +1 -1
index 3ac5abe937f4..bed18e7bbc36 100644
@@ -2266,7 +2266,7 @@ static void sdma_v4_0_set_buffer_funcs(struct amdgpu_device *adev)  {
 	adev->mman.buffer_funcs = &sdma_v4_0_buffer_funcs;
 	if (adev->sdma.has_page_queue)
-		adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].page;
+		adev->mman.buffer_funcs_ring = &adev->sdma.instance[1].page;
 	else
 		adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;  } @@ -2285,15 +2285,19 @@ static void sdma_v4_0_set_vm_pte_funcs(struct amdgpu_device *adev)
 	unsigned i;
 

adev->vm_manager.vm_pte_funcs = &sdma_v4_0_vm_pte_funcs;
- for (i = 0; i < adev->sdma.num_instances; i++) {
- if (adev->sdma.has_page_queue)
- sched = &adev->sdma.instance[i].page.sched;
- else
- sched = &adev->sdma.instance[i].ring.sched;
- adev->vm_manager.vm_pte_rqs[i] =
+ if (adev->sdma.has_page_queue) {
+ sched = &adev->sdma.instance[1].page.sched;
+ adev->vm_manager.vm_pte_rqs[0] =
&sched->sched_rq[DRM_SCHED_PRIORITY_KERNEL];
+ adev->vm_manager.vm_pte_num_rqs = 1;
+ } else {
+ for (i = 0; i < adev->sdma.num_instances; i++) {
+ sched = &adev->sdma.instance[i].ring.sched;
+ adev->vm_manager.vm_pte_rqs[i] =
+ &sched->sched_rq[DRM_SCHED_PRIORITY_KERNEL];
+ }
+ adev->vm_manager.vm_pte_num_rqs = adev->sdma.num_instances;
}
- adev->vm_manager.vm_pte_num_rqs = adev->sdma.num_instances;
}

const struct amdgpu_ip_block_version sdma_v4_0_ip_block = {
--
2.17.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): BYAPR12MB317644D82ADEF3CA55530A2B92490-ZGDeBxoHBPks/z/[email protected]

#6 lore raw

Re: [PATCH 2/3] drm/amdgpu: free up the first paging queue

I don't think Raven has a paging queue in the first place.

Alex
________________________________
From: amd-gfx <amd-gfx-bounces-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org> on behalf of Kuehling, Felix <[email protected]>
Sent: Tuesday, March 12, 2019 11:29 AM
To: Christian König; [email protected]
Subject: RE: [PATCH 2/3] drm/amdgpu: free up the first paging queue

I think this would break Raven, which only has one SDMA engine.

Regards,
Felix

-----Original Message-----
From: amd-gfx <amd-gfx-bounces-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org> On Behalf Of Christian König
Sent: Tuesday, March 12, 2019 8:38 AM
To: [email protected]
Subject: [PATCH 2/3] drm/amdgpu: free up the first paging queue

We need the first paging queue to handle page faults.

---
 drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c | 20 ++++++++++++--------
 1 file changed, 12 insertions(+), 8 deletions(-)
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +1 -1
index 3ac5abe937f4..bed18e7bbc36 100644
@@ -2266,7 +2266,7 @@ static void sdma_v4_0_set_buffer_funcs(struct amdgpu_device *adev)  {
         adev->mman.buffer_funcs = &sdma_v4_0_buffer_funcs;
         if (adev->sdma.has_page_queue)
-               adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].page;
+               adev->mman.buffer_funcs_ring = &adev->sdma.instance[1].page;
         else
                 adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;  } @@ -2285,15 +2285,19 @@ static void sdma_v4_0_set_vm_pte_funcs(struct amdgpu_device *adev)
         unsigned i;
 
         adev->vm_manager.vm_pte_funcs = &sdma_v4_0_vm_pte_funcs;

- for (i = 0; i < adev->sdma.num_instances; i++) {
- if (adev->sdma.has_page_queue)
- sched = &adev->sdma.instance[i].page.sched;
- else
- sched = &adev->sdma.instance[i].ring.sched;
- adev->vm_manager.vm_pte_rqs[i] =
+ if (adev->sdma.has_page_queue) {
+ sched = &adev->sdma.instance[1].page.sched;
+ adev->vm_manager.vm_pte_rqs[0] =

                         &sched->sched_rq[DRM_SCHED_PRIORITY_KERNEL];

+ adev->vm_manager.vm_pte_num_rqs = 1;
+ } else {
+ for (i = 0; i < adev->sdma.num_instances; i++) {
+ sched = &adev->sdma.instance[i].ring.sched;
+ adev->vm_manager.vm_pte_rqs[i] =
+ &sched->sched_rq[DRM_SCHED_PRIORITY_KERNEL];
+ }

+               adev->vm_manager.vm_pte_num_rqs = adev->sdma.num_instances;
         }

- adev->vm_manager.vm_pte_num_rqs = adev->sdma.num_instances;
}

const struct amdgpu_ip_block_version sdma_v4_0_ip_block = {
--
2.17.1

_______________________________________________
amd-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
amd-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): BN6PR12MB18095A3704551015894223D1F7490-/b2+HYfkarSEx6ez0IUAagdYzm3356FpvxpqHgZTriW3zl9H0oFU5g@public.gmane.org

#8 lore raw

Re: [PATCH 2/3] drm/amdgpu: free up the first paging queue

To be on the safe side I've adjusted the code to work with any number of SDMA instances.

Christian.

Am 12.03.19 um 16:33 schrieb Deucher, Alexander:

Quoted 70 lines

I don't think Raven has a paging queue in the first place.

Alex
------------------------------------------------------------------------
*From:* amd-gfx <amd-gfx-bounces-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org> on behalf of Kuehling, Felix <[email protected]> *Sent:* Tuesday, March 12, 2019 11:29 AM
*To:* Christian König; [email protected]
*Subject:* RE: [PATCH 2/3] drm/amdgpu: free up the first paging queue
I think this would break Raven, which only has one SDMA engine.

Regards,
  Felix

-----Original Message-----
From: amd-gfx <amd-gfx-bounces-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org> On Behalf Of Christian König
Sent: Tuesday, March 12, 2019 8:38 AM
To: [email protected]
Subject: [PATCH 2/3] drm/amdgpu: free up the first paging queue

We need the first paging queue to handle page faults.

---
 drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c | 20 ++++++++++++--------

 1 file changed, 12 insertions(+), 8 deletions(-)
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +1 -1
index 3ac5abe937f4..bed18e7bbc36 100644
@@ -2266,7 +2266,7 @@ static void sdma_v4_0_set_buffer_funcs(struct 
 amdgpu_device *adev)  {
          adev->mman.buffer_funcs = &sdma_v4_0_buffer_funcs;
          if (adev->sdma.has_page_queue)
-               adev->mman.buffer_funcs_ring = 
 &adev->sdma.instance[0].page;
+               adev->mman.buffer_funcs_ring = 
 &adev->sdma.instance[1].page;

         else
                 adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;  } @@ -2285,15 +2285,19 @@ static void sdma_v4_0_set_vm_pte_funcs(struct amdgpu_device *adev) unsigned i;

         adev->vm_manager.vm_pte_funcs = &sdma_v4_0_vm_pte_funcs; -       for (i = 0; i < adev->sdma.num_instances; i++) { -               if (adev->sdma.has_page_queue)
-                       sched = &adev->sdma.instance[i].page.sched; -               else
-                       sched = &adev->sdma.instance[i].ring.sched; -               adev->vm_manager.vm_pte_rqs[i] =
+       if (adev->sdma.has_page_queue) {
+               sched = &adev->sdma.instance[1].page.sched; +               adev->vm_manager.vm_pte_rqs[0] = &sched->sched_rq[DRM_SCHED_PRIORITY_KERNEL];
+               adev->vm_manager.vm_pte_num_rqs = 1;
+       } else {
+               for (i = 0; i < adev->sdma.num_instances; i++) { +                       sched = &adev->sdma.instance[i].ring.sched; +                       adev->vm_manager.vm_pte_rqs[i] =
+ &sched->sched_rq[DRM_SCHED_PRIORITY_KERNEL];
+               }
+               adev->vm_manager.vm_pte_num_rqs = adev->sdma.num_instances;
         }
-       adev->vm_manager.vm_pte_num_rqs = adev->sdma.num_instances; }

 const struct amdgpu_ip_block_version sdma_v4_0_ip_block = { --
2.17.1

Quoted 8 lines

_______________________________________________
amd-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
amd-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

#9 lore raw

Re: [PATCH 2/3] drm/amdgpu: free up the first paging queue

In reply to a message not in the archive
Quoted 13 lines · Christian König

We need the first paging queue to handle page faults.

---
 drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c | 20 ++++++++++++--------
 1 file changed, 12 insertions(+), 8 deletions(-)
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +0 -0
index 3ac5abe937f4..bed18e7bbc36 100644

drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c static void sdma_v4_0_set_buffer_funcs(struct amdgpu_device *adev) · line 2269

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +1 -1
@@ -2266,7 +2266,7 @@ static void sdma_v4_0_set_buffer_funcs(struct amdgpu_device *adev)
 {
        adev->mman.buffer_funcs = &sdma_v4_0_buffer_funcs;
        if (adev->sdma.has_page_queue)
-               adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].page;
+               adev->mman.buffer_funcs_ring = &adev->sdma.instance[1].page;

Maybe add something like:
if (adev->sdma.num_instances > 1)

    adev->mman.buffer_funcs_ring = &adev->sdma.instance[1].page;
else
    adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].page;
Quoted 33 lines
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +11 -8
        else
                adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
 }
@@ -2285,15 +2285,19 @@ static void sdma_v4_0_set_vm_pte_funcs(struct amdgpu_device *adev)
        unsigned i;
 
        adev->vm_manager.vm_pte_funcs = &sdma_v4_0_vm_pte_funcs;
-       for (i = 0; i < adev->sdma.num_instances; i++) {
-               if (adev->sdma.has_page_queue)
-                       sched = &adev->sdma.instance[i].page.sched;
-               else
-                       sched = &adev->sdma.instance[i].ring.sched;
-               adev->vm_manager.vm_pte_rqs[i] =
+       if (adev->sdma.has_page_queue) {
+               sched = &adev->sdma.instance[1].page.sched;
+               adev->vm_manager.vm_pte_rqs[0] =
                        &sched->sched_rq[DRM_SCHED_PRIORITY_KERNEL];
+               adev->vm_manager.vm_pte_num_rqs = 1;
+       } else {
+               for (i = 0; i < adev->sdma.num_instances; i++) {
+                       sched = &adev->sdma.instance[i].ring.sched;
+                       adev->vm_manager.vm_pte_rqs[i] =
+                               &sched->sched_rq[DRM_SCHED_PRIORITY_KERNEL];
+               }
+               adev->vm_manager.vm_pte_num_rqs = adev->sdma.num_instances;
        }
-       adev->vm_manager.vm_pte_num_rqs = adev->sdma.num_instances;
 }
 
 const struct amdgpu_ip_block_version sdma_v4_0_ip_block = {
--

2.17.1

Quoted 4 lines

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

#10 lore raw

PATCH 3/3 drm/amdgpu: use more entries for the first paging queue

In reply to a message not in the archive

To aid recoverable page faults.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c +2 -0
index 335a0edf114b..8f5026c123ef 100644
@@ -248,6 +248,8 @@ int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring,
 	 */
 	if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
 		sched_hw_submission = max(sched_hw_submission, 256);
+	else if (ring == &adev->sdma.instance[0].page)
+		sched_hw_submission = 256;
 
 	if (ring->adev == NULL) {
 		if (adev->num_rings >= AMDGPU_MAX_RINGS)
Signature · 6 lines
2.17.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx