Skip to contentSkip to conversation

drm/amdgpu: Set VM_L2_CNTL.PDE_FAULT_CLASSIFICATION to 0

PATCH In review Zhao, Yong 4 participants 9 messages Feb 25, 2019 – Feb 27, 2019 7 messages not in the archive

Subsystems (2)
lore
Share & tools
lore

Zhao, Yong

author
#0 lore raw

PATCH 1/1 drm/amdgpu: Set VM_L2_CNTL.PDE_FAULT_CLASSIFICATION to 0

This is recommended by HW designers. Previously when it was set to 1, the PDE walk error in VM fault will be treated as PERMISSION_OR_INVALID_PAGE_FAULT rather than usually expected OTHER_FAULT. As a result, the retry control in VM_CONTEXT*_CNTL will change accordingly.

The above behavior is kind of abnormal. Furthermore, the PDE_FAULT_CLASSIFICATION == 1 feature was targeted for very old ASICs and it never made it way to production. Therefore, we should set it to 0.

---
 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 f5edddf3b29d..c10ed568ca6c 100644
@@ -143,7 +143,7 @@ static void gfxhub_v1_0_init_cache_regs(struct amdgpu_device *adev)
 	/* XXX for emulation, Refer to closed source code.*/
 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, L2_PDE0_CACHE_TAG_GENERATION_MODE,
 			    0);
-	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 1);
+	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, IDENTITY_MODE_FRAGMENT_SIZE, 0);
 	WREG32_SOC15(GC, 0, mmVM_L2_CNTL, tmp);
drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +1 -1
index d0d966d6080a..2a039946a549 100644
@@ -163,7 +163,7 @@ static void mmhub_v1_0_init_cache_regs(struct amdgpu_device *adev)
 	/* XXX for emulation, Refer to closed source code.*/
 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, L2_PDE0_CACHE_TAG_GENERATION_MODE,
 			    0);
-	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 1);
+	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, IDENTITY_MODE_FRAGMENT_SIZE, 0);
 	WREG32_SOC15(MMHUB, 0, mmVM_L2_CNTL, tmp);
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): [email protected]

Acked-by
#2 lore raw
Quoted 11 lines · Yong

This is recommended by HW designers. Previously when it was set to 1, the PDE walk error in VM fault will be treated as PERMISSION_OR_INVALID_PAGE_FAULT rather than usually expected OTHER_FAULT. As a result, the retry control in VM_CONTEXT*_CNTL will change accordingly.

The above behavior is kind of abnormal. Furthermore, the PDE_FAULT_CLASSIFICATION == 1 feature was targeted for very old ASICs and it never made it way to production. Therefore, we should set it to 0.

Quoted 34 lines
---
 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 f5edddf3b29d..c10ed568ca6c 100644
@@ -143,7 +143,7 @@ static void gfxhub_v1_0_init_cache_regs(struct amdgpu_device *adev)
        /* XXX for emulation, Refer to closed source code.*/
        tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, L2_PDE0_CACHE_TAG_GENERATION_MODE,
                            0);
-       tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 1);
+       tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
        tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
        tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, IDENTITY_MODE_FRAGMENT_SIZE, 0);
        WREG32_SOC15(GC, 0, mmVM_L2_CNTL, tmp);
drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +1 -2
index d0d966d6080a..2a039946a549 100644
@@ -163,7 +163,7 @@ static void mmhub_v1_0_init_cache_regs(struct amdgpu_device *adev)
        /* XXX for emulation, Refer to closed source code.*/
        tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, L2_PDE0_CACHE_TAG_GENERATION_MODE,
                            0);
-       tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 1);
+       tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
        tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
        tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, IDENTITY_MODE_FRAGMENT_SIZE, 0);
        WREG32_SOC15(MMHUB, 0, mmVM_L2_CNTL, tmp);
--

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

A message that is not in the archive (sent off-list or not yet received): CADnq5_N9cvoGrHKBe22XWU1KPxFkAFhLkeAwPfeUdpYHfV2cMQ-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org

Acked-by Alex Deucher
#4 lore raw

Don't we want PDE faults to be treated the same way as page faults? Or am I misinterpreting the commit message?

Thanks,
John
Original Message
From: Alex Deucher
Sent: Monday, February 25, 2019 21:53
To: Zhao, Yong
Cc: [email protected]
Subject: Re: [PATCH] drm/amdgpu: Set VM_L2_CNTL.PDE_FAULT_CLASSIFICATION to 0

Quoted 11 lines · Yong

This is recommended by HW designers. Previously when it was set to 1, the PDE walk error in VM fault will be treated as PERMISSION_OR_INVALID_PAGE_FAULT rather than usually expected OTHER_FAULT. As a result, the retry control in VM_CONTEXT*_CNTL will change accordingly.

The above behavior is kind of abnormal. Furthermore, the PDE_FAULT_CLASSIFICATION == 1 feature was targeted for very old ASICs and it never made it way to production. Therefore, we should set it to 0.

Quoted 34 lines
---
 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 f5edddf3b29d..c10ed568ca6c 100644
@@ -143,7 +143,7 @@ static void gfxhub_v1_0_init_cache_regs(struct amdgpu_device *adev)
        /* XXX for emulation, Refer to closed source code.*/
        tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, L2_PDE0_CACHE_TAG_GENERATION_MODE,
                            0);
-       tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 1);
+       tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
        tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
        tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, IDENTITY_MODE_FRAGMENT_SIZE, 0);
        WREG32_SOC15(GC, 0, mmVM_L2_CNTL, tmp);
drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +1 -2
index d0d966d6080a..2a039946a549 100644
@@ -163,7 +163,7 @@ static void mmhub_v1_0_init_cache_regs(struct amdgpu_device *adev)
        /* XXX for emulation, Refer to closed source code.*/
        tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, L2_PDE0_CACHE_TAG_GENERATION_MODE,
                            0);
-       tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 1);
+       tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
        tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
        tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, IDENTITY_MODE_FRAGMENT_SIZE, 0);
        WREG32_SOC15(MMHUB, 0, mmVM_L2_CNTL, tmp);
--

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
_______________________________________________
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): [email protected]

Acked-by Alex Deucher
#6 lore raw

Or is the idea that we should never see a PDE fault unless something goes wrong, and that we would set up an entry corresponding to an unmapped subtree as an invalid PTE for a very large page rather than an invalid PDE?

Thanks,
John
Original Message
From: Bridgman, John
Sent: Monday, February 25, 2019 22:46
To: Alex Deucher; Zhao, Yong
Cc: amd-gfx list
Subject: Re: [PATCH] drm/amdgpu: Set VM_L2_CNTL.PDE_FAULT_CLASSIFICATION to 0

Don't we want PDE faults to be treated the same way as page faults? Or am I misinterpreting the commit message?

Thanks,
John
Original Message
From: Alex Deucher
Sent: Monday, February 25, 2019 21:53
To: Zhao, Yong
Cc: [email protected]
Subject: Re: [PATCH] drm/amdgpu: Set VM_L2_CNTL.PDE_FAULT_CLASSIFICATION to 0

Quoted 11 lines · Yong

This is recommended by HW designers. Previously when it was set to 1, the PDE walk error in VM fault will be treated as PERMISSION_OR_INVALID_PAGE_FAULT rather than usually expected OTHER_FAULT. As a result, the retry control in VM_CONTEXT*_CNTL will change accordingly.

The above behavior is kind of abnormal. Furthermore, the PDE_FAULT_CLASSIFICATION == 1 feature was targeted for very old ASICs and it never made it way to production. Therefore, we should set it to 0.

Quoted 34 lines
---
 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 f5edddf3b29d..c10ed568ca6c 100644
@@ -143,7 +143,7 @@ static void gfxhub_v1_0_init_cache_regs(struct amdgpu_device *adev)
        /* XXX for emulation, Refer to closed source code.*/
        tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, L2_PDE0_CACHE_TAG_GENERATION_MODE,
                            0);
-       tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 1);
+       tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
        tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
        tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, IDENTITY_MODE_FRAGMENT_SIZE, 0);
        WREG32_SOC15(GC, 0, mmVM_L2_CNTL, tmp);
drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +1 -2
index d0d966d6080a..2a039946a549 100644
@@ -163,7 +163,7 @@ static void mmhub_v1_0_init_cache_regs(struct amdgpu_device *adev)
        /* XXX for emulation, Refer to closed source code.*/
        tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, L2_PDE0_CACHE_TAG_GENERATION_MODE,
                            0);
-       tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 1);
+       tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
        tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
        tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, IDENTITY_MODE_FRAGMENT_SIZE, 0);
        WREG32_SOC15(MMHUB, 0, mmVM_L2_CNTL, tmp);
--

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
_______________________________________________
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): [email protected]

#8 lore raw
Quoted 1 line

Don't we want PDE faults to be treated the same way as page faults?

Yeah, I think that this was actually the original intention. Problem is that this is actually the wrong approach for this.

Quoted 1 line

Or is the idea that we should never see a PDE fault unless something goes wrong

Exactly, yes. What you really want to do is to set the P bit for PDEs which can be allocated dynamically.

This way the walker stops at the current processed entry and sends your a standard page fault for a leave page.

Regards,
Christian.

Am 26.02.19 um 04:50 schrieb Bridgman, John:

Quoted 71 lines

Or is the idea that we should never see a PDE fault unless something goes wrong, and that we would set up an entry corresponding to an unmapped subtree as an invalid PTE for a very large page rather than an invalid PDE?

Thanks,
John
Original Message
From: Bridgman, John
Sent: Monday, February 25, 2019 22:46
To: Alex Deucher; Zhao, Yong
Cc: amd-gfx list
Subject: Re: [PATCH] drm/amdgpu: Set VM_L2_CNTL.PDE_FAULT_CLASSIFICATION to 0

Don't we want PDE faults to be treated the same way as page faults? Or am I misinterpreting the commit message?

Thanks,
John
Original Message
From: Alex Deucher
Sent: Monday, February 25, 2019 21:53
To: Zhao, Yong
Cc: [email protected]
Subject: Re: [PATCH] drm/amdgpu: Set VM_L2_CNTL.PDE_FAULT_CLASSIFICATION to 0

Quoted 10 lines · Yong

This is recommended by HW designers. Previously when it was set to 1, the PDE walk error in VM fault will be treated as PERMISSION_OR_INVALID_PAGE_FAULT rather than usually expected OTHER_FAULT. As a result, the retry control in VM_CONTEXT*_CNTL will change accordingly.

The above behavior is kind of abnormal. Furthermore, the PDE_FAULT_CLASSIFICATION == 1 feature was targeted for very old ASICs and it never made it way to production. Therefore, we should set it to 0.

Quoted 34 lines
---
  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 f5edddf3b29d..c10ed568ca6c 100644
@@ -143,7 +143,7 @@ static void gfxhub_v1_0_init_cache_regs(struct amdgpu_device *adev)
         /* XXX for emulation, Refer to closed source code.*/
         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, L2_PDE0_CACHE_TAG_GENERATION_MODE,
                             0);
-       tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 1);
+       tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, IDENTITY_MODE_FRAGMENT_SIZE, 0);
         WREG32_SOC15(GC, 0, mmVM_L2_CNTL, tmp);
drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +1 -2
index d0d966d6080a..2a039946a549 100644
@@ -163,7 +163,7 @@ static void mmhub_v1_0_init_cache_regs(struct amdgpu_device *adev)
         /* XXX for emulation, Refer to closed source code.*/
         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, L2_PDE0_CACHE_TAG_GENERATION_MODE,
                             0);
-       tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 1);
+       tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, IDENTITY_MODE_FRAGMENT_SIZE, 0);
         WREG32_SOC15(MMHUB, 0, mmVM_L2_CNTL, tmp);
--

2.17.1

Quoted 12 lines
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
_______________________________________________
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
In reply to a message not in the archive

Am 26.02.19 um 00:03 schrieb Zhao, Yong:

Quoted 10 lines

This is recommended by HW designers. Previously when it was set to 1, the PDE walk error in VM fault will be treated as PERMISSION_OR_INVALID_PAGE_FAULT rather than usually expected OTHER_FAULT. As a result, the retry control in VM_CONTEXT*_CNTL will change accordingly.

The above behavior is kind of abnormal. Furthermore, the PDE_FAULT_CLASSIFICATION == 1 feature was targeted for very old ASICs and it never made it way to production. Therefore, we should set it to 0.

Quoted 23 lines
---
  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 f5edddf3b29d..c10ed568ca6c 100644
@@ -143,7 +143,7 @@ static void gfxhub_v1_0_init_cache_regs(struct amdgpu_device *adev)
  	/* XXX for emulation, Refer to closed source code.*/
  	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, L2_PDE0_CACHE_TAG_GENERATION_MODE,
  			    0);
-	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 1);
+	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
  	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
  	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, IDENTITY_MODE_FRAGMENT_SIZE, 0);
  	WREG32_SOC15(GC, 0, mmVM_L2_CNTL, tmp);
drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +0 -0
index d0d966d6080a..2a039946a549 100644

drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c static void mmhub_v1_0_init_cache_regs(struct amdgpu_device *adev) · line 169

Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +1 -1
@@ -163,7 +163,7 @@ static void mmhub_v1_0_init_cache_regs(struct amdgpu_device *adev)
  	/* XXX for emulation, Refer to closed source code.*/
  	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, L2_PDE0_CACHE_TAG_GENERATION_MODE,
  			    0);
-	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 1);
+	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
  	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
  	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, IDENTITY_MODE_FRAGMENT_SIZE, 0);
  	WREG32_SOC15(MMHUB, 0, mmVM_L2_CNTL, tmp);

_______________________________________________
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): 0f3d8a85-e01b-625d-b591-205f7f28af8e-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org

author
#11 lore raw

So can I submit it only with the ACKs?

Yong

Quoted 50 lines · Christian König

Am 26.02.19 um 00:03 schrieb Zhao, Yong:

Quoted 7 lines

This is recommended by HW designers. Previously when it was set to 1, the PDE walk error in VM fault will be treated as PERMISSION_OR_INVALID_PAGE_FAULT rather than usually expected OTHER_FAULT.
As a result, the retry control in VM_CONTEXT*_CNTL will change accordingly.

Quoted 6 lines

The above behavior is kind of abnormal. Furthermore, the PDE_FAULT_CLASSIFICATION == 1 feature was targeted for very old ASICs and it never made it way to production. Therefore, we should set it to 0.

Quoted 33 lines

---
  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 f5edddf3b29d..c10ed568ca6c 100644
@@ -143,7 +143,7 @@ static void gfxhub_v1_0_init_cache_regs(struct 
 amdgpu_device *adev)
       /* XXX for emulation, Refer to closed source code.*/
       tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, 
 L2_PDE0_CACHE_TAG_GENERATION_MODE,
                   0);
-    tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 1);
+    tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
       tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, 

CONTEXT1_IDENTITY_ACCESS_MODE, 1);
      tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,
IDENTITY_MODE_FRAGMENT_SIZE, 0);
      WREG32_SOC15(GC, 0, mmVM_L2_CNTL, tmp);

drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +0 -0
index d0d966d6080a..2a039946a549 100644
@@ -163,7 +163,7 @@ static void mmhub_v1_0_init_cache_regs(struct 
 amdgpu_device *adev)
       /* XXX for emulation, Refer to closed source code.*/
       tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, 
 L2_PDE0_CACHE_TAG_GENERATION_MODE,
Quoted 9 lines
Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +1 -1
                   0);
-    tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 1);
+    tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
       tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, 

CONTEXT1_IDENTITY_ACCESS_MODE, 1);
      tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,
IDENTITY_MODE_FRAGMENT_SIZE, 0);
      WREG32_SOC15(MMHUB, 0, mmVM_L2_CNTL, tmp);

_______________________________________________
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): d107229a-062b-8535-87c8-17110d2223ea-5C7GfCeVMHo@public.gmane.org

#13 lore raw

yes, go ahead.

Alex

Quoted 65 lines · Yong

So can I submit it only with the ACKs?

Yong

Quoted 50 lines · Christian König

Am 26.02.19 um 00:03 schrieb Zhao, Yong:

Quoted 7 lines

This is recommended by HW designers. Previously when it was set to 1, the PDE walk error in VM fault will be treated as PERMISSION_OR_INVALID_PAGE_FAULT rather than usually expected OTHER_FAULT.
As a result, the retry control in VM_CONTEXT*_CNTL will change accordingly.

Quoted 6 lines

The above behavior is kind of abnormal. Furthermore, the PDE_FAULT_CLASSIFICATION == 1 feature was targeted for very old ASICs and it never made it way to production. Therefore, we should set it to 0.

Quoted 33 lines
---
  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 f5edddf3b29d..c10ed568ca6c 100644
@@ -143,7 +143,7 @@ static void gfxhub_v1_0_init_cache_regs(struct
 amdgpu_device *adev)
      /* XXX for emulation, Refer to closed source code.*/
      tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,
 L2_PDE0_CACHE_TAG_GENERATION_MODE,
                  0);
-    tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 1);
+    tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
      tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,

CONTEXT1_IDENTITY_ACCESS_MODE, 1);
tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,

IDENTITY_MODE_FRAGMENT_SIZE, 0);
      WREG32_SOC15(GC, 0, mmVM_L2_CNTL, tmp);
drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +0 -0
index d0d966d6080a..2a039946a549 100644
@@ -163,7 +163,7 @@ static void mmhub_v1_0_init_cache_regs(struct
 amdgpu_device *adev)
      /* XXX for emulation, Refer to closed source code.*/
      tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,
 L2_PDE0_CACHE_TAG_GENERATION_MODE,
Quoted 9 lines
Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +1 -1
                  0);
-    tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 1);
+    tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
      tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,

CONTEXT1_IDENTITY_ACCESS_MODE, 1);
tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,

IDENTITY_MODE_FRAGMENT_SIZE, 0);
      WREG32_SOC15(MMHUB, 0, mmVM_L2_CNTL, tmp);
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

A message that is not in the archive (sent off-list or not yet received): CADnq5_O55w2btKnNK7TWvgRmOuE3wC0rGXUXg+u06SdJeOSZ4Q-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org

author
#15 lore raw

Pushed. Thanks.

Yong
________________________________
From: Alex Deucher <[email protected]>
Sent: Tuesday, February 26, 2019 10:53 PM
To: Zhao, Yong
Cc: Koenig, Christian; [email protected]
Subject: Re: [PATCH] drm/amdgpu: Set VM_L2_CNTL.PDE_FAULT_CLASSIFICATION to 0

yes, go ahead.

Alex

Quoted 65 lines · Yong

So can I submit it only with the ACKs?

Yong

Quoted 50 lines · Christian König

Am 26.02.19 um 00:03 schrieb Zhao, Yong:

Quoted 7 lines

This is recommended by HW designers. Previously when it was set to 1, the PDE walk error in VM fault will be treated as PERMISSION_OR_INVALID_PAGE_FAULT rather than usually expected OTHER_FAULT.
As a result, the retry control in VM_CONTEXT*_CNTL will change accordingly.

Quoted 6 lines

The above behavior is kind of abnormal. Furthermore, the PDE_FAULT_CLASSIFICATION == 1 feature was targeted for very old ASICs and it never made it way to production. Therefore, we should set it to 0.

Quoted 33 lines
---
  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 f5edddf3b29d..c10ed568ca6c 100644
@@ -143,7 +143,7 @@ static void gfxhub_v1_0_init_cache_regs(struct
 amdgpu_device *adev)
      /* XXX for emulation, Refer to closed source code.*/
      tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,
 L2_PDE0_CACHE_TAG_GENERATION_MODE,
                  0);
-    tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 1);
+    tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
      tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,

CONTEXT1_IDENTITY_ACCESS_MODE, 1);
tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,

IDENTITY_MODE_FRAGMENT_SIZE, 0);
      WREG32_SOC15(GC, 0, mmVM_L2_CNTL, tmp);
drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +0 -0
index d0d966d6080a..2a039946a549 100644
@@ -163,7 +163,7 @@ static void mmhub_v1_0_init_cache_regs(struct
 amdgpu_device *adev)
      /* XXX for emulation, Refer to closed source code.*/
      tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,
 L2_PDE0_CACHE_TAG_GENERATION_MODE,
Quoted 9 lines
Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +1 -1
                  0);
-    tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 1);
+    tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
      tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,

CONTEXT1_IDENTITY_ACCESS_MODE, 1);
tmp = REG_SET_FIELD(tmp, VM_L2_CNTL,

IDENTITY_MODE_FRAGMENT_SIZE, 0);
      WREG32_SOC15(MMHUB, 0, mmVM_L2_CNTL, tmp);
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