Skip to contentSkip to conversation

drm/amdgpu: fix re-program vm invalidate eng address range for gfxhub on resume

PATCH In review Huang Rui 4 participants 12 messages May 17, 2017 7 messages not in the archive

Subsystems (2)
lore
Share & tools
lore

Huang Rui

author
#0 lore raw

PATCH 1/2 drm/amdgpu: fix re-program vm invalidate eng address range for gfxhub on resume

---
 drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +1 -1
index 005075f..41313514 100644
@@ -368,7 +368,7 @@ static int gfxhub_v1_0_suspend(void *handle)
 
 static int gfxhub_v1_0_resume(void *handle)
 {
-	return 0;
+	return gfxhub_v1_0_hw_init(handle);
 }
 
 static bool gfxhub_v1_0_is_idle(void *handle)
Signature · 6 lines
2.7.4

_______________________________________________
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): 1495006729-17310-1-git-send-email-ray.huang-5C7GfCeVMHo@public.gmane.org

author
#2 lore raw

PATCH 2/2 drm/amdgpu: fix re-program vm invalidate eng address range for mmhub on resume

---
 drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +1 -1
index dbfe48d..8979bc2 100644
@@ -386,7 +386,7 @@ static int mmhub_v1_0_suspend(void *handle)
 
 static int mmhub_v1_0_resume(void *handle)
 {
-	return 0;
+	return mmhub_v1_0_hw_init(handle);
 }
 
 static bool mmhub_v1_0_is_idle(void *handle)
Signature · 6 lines
2.7.4

_______________________________________________
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): 1495006729-17310-2-git-send-email-ray.huang-5C7GfCeVMHo@public.gmane.org

Reviewed-by Ken Wang
#4 lore raw

Re: [PATCH 2/2] drm/amdgpu: fix re-program vm invalidate eng address range for mmhub on resume

________________________________
From: Huang Rui <[email protected]>
Sent: Wednesday, May 17, 2017 3:38:49 PM
To: [email protected]; Deucher, Alexander; Koenig, Christian
Cc: Zhou, David(ChunMing); Wang, Ken; Huan, Alvin; Huang, Ray
Subject: [PATCH 2/2] drm/amdgpu: fix re-program vm invalidate eng address range for mmhub on resume

---
 drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c +1 -1
index dbfe48d..8979bc2 100644
@@ -386,7 +386,7 @@ static int mmhub_v1_0_suspend(void *handle)
 
 static int mmhub_v1_0_resume(void *handle)
 {
-       return 0;
+       return mmhub_v1_0_hw_init(handle);
 }
 
 static bool mmhub_v1_0_is_idle(void *handle)

--
2.7.4

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

#5 lore raw
In reply to a message not in the archive

By this change, I suggest to remove mmhub/gfxhub_v1_0_ip_funcs and their IP block, unify them to gmc ip block, this way we cannot lost setting when resume back.

Regards,
David Zhou

Quoted 10 lines · Huang Rui
---
  drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +0 -0
index 005075f..41313514 100644

drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c static int gfxhub_v1_0_suspend(void *handle) · line 374

Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +1 -1
@@ -368,7 +368,7 @@ static int gfxhub_v1_0_suspend(void *handle)
 
  static int gfxhub_v1_0_resume(void *handle)
  {
-	return 0;
+	return gfxhub_v1_0_hw_init(handle);
  }
 
  static bool gfxhub_v1_0_is_idle(void *handle)

_______________________________________________
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]

author
#7 lore raw
Quoted 4 lines · David(ChunMing)

By this change, I suggest to remove mmhub/gfxhub_v1_0_ip_funcs and their IP block, unify them to gmc ip block, this way we cannot lost setting when resume back.

From hw side, wo won't have real gmc since this chip, mmhub and gfxhub(gc) instead of it. Maybe we would better to align with hw desgin.

Thanks,
Rui

Quoted 16 lines

Regards,
David Zhou

drivers/gpu/drm/amd/

Quoted 6 lines · Huang Rui
---
  drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/amd/ renamed from drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +0 -0

amdgpu/gfxhub_v1_0.c

Quoted 5 lines

index 005075f..41313514 100644

drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +0 -0
@@ -368,7 +368,7 @@ static int gfxhub_v1_0_suspend(void *handle)
 
Quoted 8 lines

drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c static int gfxhub_v1_0_suspend(void *handle) · line 374

Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +1 -1
  static int gfxhub_v1_0_resume(void *handle)
  {
-     return 0;
+     return gfxhub_v1_0_hw_init(handle);
  }
 
  static bool gfxhub_v1_0_is_idle(void *handle)

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

#8 lore raw
Quoted 7 lines · Huang Rui
Quoted 4 lines · David(ChunMing)

By this change, I suggest to remove mmhub/gfxhub_v1_0_ip_funcs and their IP block, unify them to gmc ip block, this way we cannot lost setting when resume back.

From hw side, wo won't have real gmc since this chip, mmhub and gfxhub(gc)
instead of it. Maybe we would better to align with hw desgin.

I don't see any advance, as you said, we still have gmc block in soc15, why not unify mmhub/gfxhub calls to gmc block? We can keep mmhub/gfxhub_xxx.c file, but ip_funciton isn't necessary.

Regards,
David Zhou

Quoted 26 lines

Thanks,
Rui

Quoted 22 lines

Regards,
David Zhou

drivers/gpu/drm/amd/

Quoted 6 lines · Huang Rui
---
   drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c | 2 +-
   1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/amd/ renamed from drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +0 -0

amdgpu/gfxhub_v1_0.c

Quoted 11 lines

index 005075f..41313514 100644

drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +1 -1
@@ -368,7 +368,7 @@ static int gfxhub_v1_0_suspend(void *handle)
 
   static int gfxhub_v1_0_resume(void *handle)
   {
-     return 0;
+     return gfxhub_v1_0_hw_init(handle);
   }
 
Quoted 5 lines
Quoted 1 line

drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c static int gfxhub_v1_0_suspend(void *handle) · line 374

Quoted 1 line
drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +0 -0
   static bool gfxhub_v1_0_is_idle(void *handle)

_______________________________________________
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]

#10 lore raw

see here, I already find dis-advance:
enum amd_ip_block_type {
AMD_IP_BLOCK_TYPE_COMMON,
AMD_IP_BLOCK_TYPE_GMC,
AMD_IP_BLOCK_TYPE_IH,
AMD_IP_BLOCK_TYPE_SMC,
AMD_IP_BLOCK_TYPE_PSP,
AMD_IP_BLOCK_TYPE_DCE,
AMD_IP_BLOCK_TYPE_GFX,
AMD_IP_BLOCK_TYPE_SDMA,
AMD_IP_BLOCK_TYPE_UVD,
AMD_IP_BLOCK_TYPE_VCE,
AMD_IP_BLOCK_TYPE_ACP,
AMD_IP_BLOCK_TYPE_GFXHUB,
AMD_IP_BLOCK_TYPE_MMHUB
};
resume will follow this sequence.
but initial sequence is :

              amdgpu_ip_block_add(adev, &vega10_common_ip_block);
                amdgpu_ip_block_add(adev, &gfxhub_v1_0_ip_block);
                amdgpu_ip_block_add(adev, &mmhub_v1_0_ip_block);
                amdgpu_ip_block_add(adev, &gmc_v9_0_ip_block);
                amdgpu_ip_block_add(adev, &vega10_ih_ip_block);
                if (amdgpu_fw_load_type == 2 || amdgpu_fw_load_type == -1)
                        amdgpu_ip_block_add(adev, &psp_v3_1_ip_block);
                if (!amdgpu_sriov_vf(adev))
                        amdgpu_ip_block_add(adev, &amdgpu_pp_ip_block);
                if (adev->enable_virtual_display || amdgpu_sriov_vf(adev))
                        amdgpu_ip_block_add(adev, &dce_virtual_ip_block);

#if defined(CONFIG_DRM_AMD_DC)
else if (amdgpu_device_has_dc_support(adev))
amdgpu_ip_block_add(adev, &dm_ip_block);
#else
# warning "Enable CONFIG_DRM_AMD_DC for display support on SOC15."

#endif
                amdgpu_ip_block_add(adev, &gfx_v9_0_ip_block);
                amdgpu_ip_block_add(adev, &sdma_v4_0_ip_block);
                amdgpu_ip_block_add(adev, &uvd_v7_0_ip_block);
                amdgpu_ip_block_add(adev, &vce_v4_0_ip_block);

They are different. I remember I asked you if they are same, don't know why you answer 'yes'.

With s3 problem still in there, please do this improvement asap.

Regards,
David Zhou

Quoted 46 lines · On 2017年05月17日 16:11, zhoucm1 wrote:
Quoted 8 lines · Huang Rui
Quoted 4 lines · David(ChunMing)

By this change, I suggest to remove mmhub/gfxhub_v1_0_ip_funcs and their IP block, unify them to gmc ip block, this way we cannot lost setting when resume back.

From hw side, wo won't have real gmc since this chip, mmhub and
gfxhub(gc)
instead of it. Maybe we would better to align with hw desgin.

I don't see any advance, as you said, we still have gmc block in soc15, why not unify mmhub/gfxhub calls to gmc block? We can keep mmhub/gfxhub_xxx.c file, but ip_funciton isn't necessary.

Regards,
David Zhou

Quoted 15 lines

Thanks,
Rui

Quoted 11 lines

Regards,
David Zhou

drivers/gpu/drm/amd/

Quoted 6 lines · Huang Rui
---
   drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c | 2 +-
   1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/amd/ renamed from drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +0 -0

amdgpu/gfxhub_v1_0.c

Quoted 14 lines
Quoted 10 lines

drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c static int gfxhub_v1_0_suspend(void *handle) · line 372

Quoted 10 lines

index 005075f..41313514 100644

drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +1 -1
@@ -368,7 +368,7 @@ static int gfxhub_v1_0_suspend(void *handle)
     static int gfxhub_v1_0_resume(void *handle)
   {
-     return 0;
+     return gfxhub_v1_0_hw_init(handle);
   }
     static bool gfxhub_v1_0_is_idle(void *handle)

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

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

#11 lore raw
In reply to a message not in the archive

Am 17.05.2017 um 10:11 schrieb zhoucm1:

Quoted 8 lines · Huang Rui
Quoted 5 lines · David(ChunMing)
Quoted 4 lines

By this change, I suggest to remove mmhub/gfxhub_v1_0_ip_funcs and their IP block, unify them to gmc ip block, this way we cannot lost setting when resume back.

Quoted 5 lines
Quoted 3 lines

From hw side, wo won't have real gmc since this chip, mmhub and
gfxhub(gc)
instead of it. Maybe we would better to align with hw desgin.

I don't see any advance, as you said, we still have gmc block in soc15, why not unify mmhub/gfxhub calls to gmc block? We can keep mmhub/gfxhub_xxx.c file, but ip_funciton isn't necessary.

Well they are two hardware block, but are mostly identical programmed (the MMHUB has a few more bits for guaranteed bandwith, but we ignore those at the moment).

So it doesn't make much sense having two separate code instances to handle them.

One major problem still remaining is that our generated register headers are crap for this. You can't for example include both headers at the same time.

Regards,
Christian.

Quoted 33 lines

Regards,
David Zhou

Quoted 29 lines

Thanks,
Rui

Quoted 11 lines

Regards,
David Zhou

drivers/gpu/drm/amd/

Quoted 6 lines · Huang Rui
---
   drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c | 2 +-
   1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/amd/ renamed from drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +0 -0

amdgpu/gfxhub_v1_0.c

Quoted 10 lines

drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c static int gfxhub_v1_0_suspend(void *handle) · line 372

Quoted 10 lines

index 005075f..41313514 100644

drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +1 -1
@@ -368,7 +368,7 @@ static int gfxhub_v1_0_suspend(void *handle)
     static int gfxhub_v1_0_resume(void *handle)
   {
-     return 0;
+     return gfxhub_v1_0_hw_init(handle);
   }
     static bool gfxhub_v1_0_is_idle(void *handle)

_______________________________________________
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): 806d8b04-e983-ee6b-234f-9cd9fde4a3f5-5C7GfCeVMHo@public.gmane.org

#13 lore raw
Quoted 12 lines · Christian König

Am 17.05.2017 um 10:11 schrieb zhoucm1:

Quoted 11 lines · Huang Rui
Quoted 8 lines · David(ChunMing)
Quoted 4 lines

By this change, I suggest to remove mmhub/gfxhub_v1_0_ip_funcs and their IP block, unify them to gmc ip block, this way we cannot lost setting when resume back.

From hw side, wo won't have real gmc since this chip, mmhub and
gfxhub(gc)
instead of it. Maybe we would better to align with hw desgin.

Quoted 8 lines
Quoted 2 lines

I don't see any advance, as you said, we still have gmc block in soc15, why not unify mmhub/gfxhub calls to gmc block? We can keep mmhub/gfxhub_xxx.c file, but ip_funciton isn't necessary.

Well they are two hardware block, but are mostly identical programmed (the MMHUB has a few more bits for guaranteed bandwith, but we ignore those at the moment).

So it doesn't make much sense having two separate code instances to handle them.

One major problem still remaining is that our generated register headers are crap for this. You can't for example include both headers at the same time.

As I said before, we can keep these two seperate files, just remove ip_functions, but call them from gmc file. like gmc_v9_init()

{
    gfxhub_init();
    mmhub_init();
}

Regards,
David Zhou

Quoted 38 lines

Regards,
Christian.

Quoted 33 lines

Regards,
David Zhou

Quoted 29 lines

Thanks,
Rui

Quoted 11 lines

Regards,
David Zhou

drivers/gpu/drm/amd/

Quoted 6 lines · Huang Rui
---
   drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c | 2 +-
   1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/amd/ renamed from drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +0 -0

amdgpu/gfxhub_v1_0.c

Quoted 10 lines

drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c static int gfxhub_v1_0_suspend(void *handle) · line 372

Quoted 10 lines

index 005075f..41313514 100644

drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +1 -1
@@ -368,7 +368,7 @@ static int gfxhub_v1_0_suspend(void *handle)
     static int gfxhub_v1_0_resume(void *handle)
   {
-     return 0;
+     return gfxhub_v1_0_hw_init(handle);
   }
     static bool gfxhub_v1_0_is_idle(void *handle)

_______________________________________________
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]

#15 lore raw

Am 17.05.2017 um 10:22 schrieb zhoucm1:

Quoted 22 lines · Christian König
Quoted 19 lines

Am 17.05.2017 um 10:11 schrieb zhoucm1:

Quoted 8 lines · Huang Rui
Quoted 5 lines · David(ChunMing)
Quoted 4 lines

By this change, I suggest to remove mmhub/gfxhub_v1_0_ip_funcs and their IP block, unify them to gmc ip block, this way we cannot lost setting when resume back.

Quoted 5 lines
Quoted 3 lines

From hw side, wo won't have real gmc since this chip, mmhub and
gfxhub(gc)
instead of it. Maybe we would better to align with hw desgin.

I don't see any advance, as you said, we still have gmc block in soc15, why not unify mmhub/gfxhub calls to gmc block? We can keep mmhub/gfxhub_xxx.c file, but ip_funciton isn't necessary.

Well they are two hardware block, but are mostly identical programmed (the MMHUB has a few more bits for guaranteed bandwith, but we ignore those at the moment).

So it doesn't make much sense having two separate code instances to handle them.

Quoted 7 lines
Quoted 1 line

One major problem still remaining is that our generated register headers are crap for this. You can't for example include both headers at the same time.

As I said before, we can keep these two seperate files, just remove ip_functions, but call them from gmc file. like gmc_v9_init()

{
    gfxhub_init();
    mmhub_init();
}

The init order isn't so critical here. My concern is rather that we have the duplicated code.

But I agree that we sooner or later should clean up the init order on load/resume to be the same.

Regards,
Christian.

Quoted 42 lines

Regards,
David Zhou

Quoted 38 lines

Regards,
Christian.

Quoted 33 lines

Regards,
David Zhou

Quoted 29 lines

Thanks,
Rui

Quoted 11 lines

Regards,
David Zhou

drivers/gpu/drm/amd/

Quoted 6 lines · Huang Rui
---
   drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c | 2 +-
   1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/amd/ renamed from drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +0 -0

amdgpu/gfxhub_v1_0.c

Quoted 10 lines

drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c static int gfxhub_v1_0_suspend(void *handle) · line 372

Quoted 10 lines

index 005075f..41313514 100644

drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +1 -1
@@ -368,7 +368,7 @@ static int gfxhub_v1_0_suspend(void *handle)
     static int gfxhub_v1_0_resume(void *handle)
   {
-     return 0;
+     return gfxhub_v1_0_hw_init(handle);
   }
     static bool gfxhub_v1_0_is_idle(void *handle)

_______________________________________________
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

#16 lore raw
In reply to Huang Rui

Am 17.05.2017 um 09:55 schrieb Huang Rui:

Quoted 7 lines · David(ChunMing)
Quoted 4 lines

By this change, I suggest to remove mmhub/gfxhub_v1_0_ip_funcs and their IP block, unify them to gmc ip block, this way we cannot lost setting when resume back.

From hw side, wo won't have real gmc since this chip, mmhub and gfxhub(gc)
instead of it. Maybe we would better to align with hw desgin.

Yeah, I had patches for doing exactly this but then decided dropping it.

What we should do is to have any one implementation of the code handling both hubs and then have both as separate IP block in the list.

Anyway for now both patches are Reviewed-by: Christian König <[email protected]>.

I assume that finally fixes S3 on Vega10?

Regards,
Christian.

Quoted 20 lines

Thanks,
Rui

Quoted 16 lines

Regards,
David Zhou

drivers/gpu/drm/amd/

Quoted 6 lines · Huang Rui
---
   drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c | 2 +-
   1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/amd/ renamed from drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +0 -0

amdgpu/gfxhub_v1_0.c

Quoted 5 lines

index 005075f..41313514 100644

drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +0 -0
@@ -368,7 +368,7 @@ static int gfxhub_v1_0_suspend(void *handle)
 
Quoted 7 lines
Quoted 7 lines

drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c static int gfxhub_v1_0_suspend(void *handle) · line 374

Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c +1 -1
   static int gfxhub_v1_0_resume(void *handle)
   {
-     return 0;
+     return gfxhub_v1_0_hw_init(handle);
   }
 
   static bool gfxhub_v1_0_is_idle(void *handle)

_______________________________________________
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): 73fa42d9-1994-83ed-db9f-7c601f8a4598-5C7GfCeVMHo@public.gmane.org

author
#18 lore raw
Quoted 11 lines · Christian

Am 17.05.2017 um 09:55 schrieb Huang Rui:

Quoted 7 lines · David(ChunMing)
Quoted 4 lines

By this change, I suggest to remove mmhub/gfxhub_v1_0_ip_funcs and their IP block, unify them to gmc ip block, this way we cannot lost setting when resume back.

From hw side, wo won't have real gmc since this chip, mmhub and gfxhub(gc)
instead of it. Maybe we would better to align with hw desgin.

Yeah, I had patches for doing exactly this but then decided dropping it.

Quoted 8 lines

What we should do is to have any one implementation of the code handling both hubs and then have both as separate IP block in the list.

Anyway for now both patches are Reviewed-by: Christian König <[email protected]>.

I assume that finally fixes S3 on Vega10?

No, but these fixes have improvement for the result. We can pass a round of gfx/cp/sdma test.

But if we try to test a round again, it will encounter VM fault. Will tell you the detailed in another thread.

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