Skip to contentSkip to conversation

drm/amdgpu: Skip some registers config for SRIOV

PATCH In review Liu ChengZhe 2 participants 2 messages Aug 6, 2020

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

PATCH 1/1 drm/amdgpu: Skip some registers config for SRIOV

For VF, registers L2_CNTL, L2_CONTEXT1_IDENTITY_APERTURE L2_PROTECTION_FAULT_CNTL are not accesible, skip the configuration for them in SRIOV mode.

---
 drivers/gpu/drm/amd/amdgpu/gfxhub_v2_1.c | 12 ++++++++++--
 drivers/gpu/drm/amd/amdgpu/mmhub_v2_0.c  | 12 ++++++++++--
 2 files changed, 20 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdgpu/gfxhub_v2_1.c +10 -2
index 1f6112b7fa49..6b96f45fde2a 100644
@@ -330,10 +330,13 @@ int gfxhub_v2_1_gart_enable(struct amdgpu_device *adev)
 	gfxhub_v2_1_init_gart_aperture_regs(adev);
 	gfxhub_v2_1_init_system_aperture_regs(adev);
 	gfxhub_v2_1_init_tlb_regs(adev);
-	gfxhub_v2_1_init_cache_regs(adev);
+	if (!amdgpu_sriov_vf(adev))
+		gfxhub_v2_1_init_cache_regs(adev);
 
 	gfxhub_v2_1_enable_system_domain(adev);
-	gfxhub_v2_1_disable_identity_aperture(adev);
+	if (!amdgpu_sriov_vf(adev))
+		gfxhub_v2_1_disable_identity_aperture(adev);
+
 	gfxhub_v2_1_setup_vmid_config(adev);
 	gfxhub_v2_1_program_invalidation(adev);
 
@@ -372,7 +375,12 @@ void gfxhub_v2_1_gart_disable(struct amdgpu_device *adev)
 void gfxhub_v2_1_set_fault_enable_default(struct amdgpu_device *adev,
 					  bool value)
 {
+	/*These regs are not accessible for VF, PF will program in SRIOV */
+	if (amdgpu_sriov_vf(adev))
+		return;
+
 	u32 tmp;
+
 	tmp = RREG32_SOC15(GC, 0, mmGCVM_L2_PROTECTION_FAULT_CNTL);
 	tmp = REG_SET_FIELD(tmp, GCVM_L2_PROTECTION_FAULT_CNTL,
 			    RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
drivers/gpu/drm/amd/amdgpu/mmhub_v2_0.c +10 -2
index d83912901f73..9cfde9b81600 100644
@@ -321,10 +321,13 @@ int mmhub_v2_0_gart_enable(struct amdgpu_device *adev)
 	mmhub_v2_0_init_gart_aperture_regs(adev);
 	mmhub_v2_0_init_system_aperture_regs(adev);
 	mmhub_v2_0_init_tlb_regs(adev);
-	mmhub_v2_0_init_cache_regs(adev);
+	if (!amdgpu_sriov_vf(adev))
+		mmhub_v2_0_init_cache_regs(adev);
 
 	mmhub_v2_0_enable_system_domain(adev);
-	mmhub_v2_0_disable_identity_aperture(adev);
+	if (!amdgpu_sriov_vf(adev))
+		mmhub_v2_0_disable_identity_aperture(adev);
+
 	mmhub_v2_0_setup_vmid_config(adev);
 	mmhub_v2_0_program_invalidation(adev);
 
@@ -364,7 +367,12 @@ void mmhub_v2_0_gart_disable(struct amdgpu_device *adev)
  */
 void mmhub_v2_0_set_fault_enable_default(struct amdgpu_device *adev, bool value)
 {
+	/*These regs are not accessible for VF, PF will program in SRIOV */
+	if (amdgpu_sriov_vf(adev))
+		return;
+
 	u32 tmp;
+
 	tmp = RREG32_SOC15(MMHUB, 0, mmMMVM_L2_PROTECTION_FAULT_CNTL);
 	tmp = REG_SET_FIELD(tmp, MMVM_L2_PROTECTION_FAULT_CNTL,
 			    RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
Signature · 6 lines
2.25.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#1 lore raw
Quoted 2 lines · Liu ChengZhe

For VF, registers L2_CNTL, L2_CONTEXT1_IDENTITY_APERTURE L2_PROTECTION_FAULT_CNTL are not accesible, skip the

Spelling: "accessible".

I'd rather move the "For VF" to the end, after "accessible", like this. I also believe it should be "to VF" as opposed to "for VF". Both are correct, but mean different things. Maybe like this:

Some registers are not accessible to virtual function setup, so skip their initialization when in VF-SRIOV mode.

Quoted 13 lines

configuration for them in SRIOV mode.

---
 drivers/gpu/drm/amd/amdgpu/gfxhub_v2_1.c | 12 ++++++++++--
 drivers/gpu/drm/amd/amdgpu/mmhub_v2_0.c  | 12 ++++++++++--
 2 files changed, 20 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdgpu/gfxhub_v2_1.c +0 -0
index 1f6112b7fa49..6b96f45fde2a 100644

drivers/gpu/drm/amd/amdgpu/gfxhub_v2_1.c int gfxhub_v2_1_gart_enable(struct amdgpu_device *adev) · line 334

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/gfxhub_v2_1.c +2 -1
@@ -330,10 +330,13 @@ int gfxhub_v2_1_gart_enable(struct amdgpu_device *adev)
 	gfxhub_v2_1_init_gart_aperture_regs(adev);
 	gfxhub_v2_1_init_system_aperture_regs(adev);
 	gfxhub_v2_1_init_tlb_regs(adev);
-	gfxhub_v2_1_init_cache_regs(adev);
+	if (!amdgpu_sriov_vf(adev))
+		gfxhub_v2_1_init_cache_regs(adev);

Yes, that's the literal meaning of the commit description, but it would be cleaner if gfxhub_v2_1_init_cache_regs(adev) did that check instead. (Some might even argue it'd be more object-oriented...)

So then, this code would look like a sequence of statements, unchanged, and each method of initialization would know/check whether it is applicable to "adev". It also makes it more centralized, less code duplication and less code clutter.

drivers/gpu/drm/amd/amdgpu/gfxhub_v2_1.c int gfxhub_v2_1_gart_enable(struct amdgpu_device *adev) · line 339

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/gfxhub_v2_1.c +3 -1
 	gfxhub_v2_1_enable_system_domain(adev);
-	gfxhub_v2_1_disable_identity_aperture(adev);
+	if (!amdgpu_sriov_vf(adev))
+		gfxhub_v2_1_disable_identity_aperture(adev);
+

Ditto.

drivers/gpu/drm/amd/amdgpu/gfxhub_v2_1.c void gfxhub_v2_1_gart_disable(struct amdgpu_device *adev) · line 378

Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/gfxhub_v2_1.c +1 -0
 	gfxhub_v2_1_setup_vmid_config(adev);
 	gfxhub_v2_1_program_invalidation(adev);
 
@@ -372,7 +375,12 @@ void gfxhub_v2_1_gart_disable(struct amdgpu_device *adev)
 void gfxhub_v2_1_set_fault_enable_default(struct amdgpu_device *adev,
 					  bool value)
 {
+	/*These regs are not accessible for VF, PF will program in SRIOV */

Add a space after the asterisk and before the beginning of the sentence. Format the comment in one of the two acceptable Linux Kernel Coding styles. End the sentence with a full stop. It also helps to spell out "registers". Like this perhaps:

	/* These registers are not accessible to VF-SRIOV.
	 * The PF will program them instead.
	 */

drivers/gpu/drm/amd/amdgpu/gfxhub_v2_1.c void gfxhub_v2_1_gart_disable(struct amdgpu_device *adev) · line 382

Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/gfxhub_v2_1.c +3 -0
+	if (amdgpu_sriov_vf(adev))
+		return;
+
 	u32 tmp;

The definition of this automatic variable should precede all other code in this function. The compiler will optimize it away anyway.

Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/gfxhub_v2_1.c +1 -0
+
 	tmp = RREG32_SOC15(GC, 0, mmGCVM_L2_PROTECTION_FAULT_CNTL);
 	tmp = REG_SET_FIELD(tmp, GCVM_L2_PROTECTION_FAULT_CNTL,
 			    RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
drivers/gpu/drm/amd/amdgpu/mmhub_v2_0.c +0 -0
index d83912901f73..9cfde9b81600 100644

drivers/gpu/drm/amd/amdgpu/mmhub_v2_0.c int mmhub_v2_0_gart_enable(struct amdgpu_device *adev) · line 325

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/mmhub_v2_0.c +2 -1
@@ -321,10 +321,13 @@ int mmhub_v2_0_gart_enable(struct amdgpu_device *adev)
 	mmhub_v2_0_init_gart_aperture_regs(adev);
 	mmhub_v2_0_init_system_aperture_regs(adev);
 	mmhub_v2_0_init_tlb_regs(adev);
-	mmhub_v2_0_init_cache_regs(adev);
+	if (!amdgpu_sriov_vf(adev))
+		mmhub_v2_0_init_cache_regs(adev);

Move this check inside said function.

drivers/gpu/drm/amd/amdgpu/mmhub_v2_0.c int mmhub_v2_0_gart_enable(struct amdgpu_device *adev) · line 330

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/mmhub_v2_0.c +3 -1
 	mmhub_v2_0_enable_system_domain(adev);
-	mmhub_v2_0_disable_identity_aperture(adev);
+	if (!amdgpu_sriov_vf(adev))
+		mmhub_v2_0_disable_identity_aperture(adev);
+

Ditto.

drivers/gpu/drm/amd/amdgpu/mmhub_v2_0.c void mmhub_v2_0_gart_disable(struct amdgpu_device *adev) · line 370

Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/mmhub_v2_0.c +1 -0
 	mmhub_v2_0_setup_vmid_config(adev);
 	mmhub_v2_0_program_invalidation(adev);
 
@@ -364,7 +367,12 @@ void mmhub_v2_0_gart_disable(struct amdgpu_device *adev)
  */
 void mmhub_v2_0_set_fault_enable_default(struct amdgpu_device *adev, bool value)
 {
+	/*These regs are not accessible for VF, PF will program in SRIOV */

You can duplicate this comment from the one above.

drivers/gpu/drm/amd/amdgpu/mmhub_v2_0.c void mmhub_v2_0_gart_disable(struct amdgpu_device *adev) · line 374

Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/mmhub_v2_0.c +3 -0
+	if (amdgpu_sriov_vf(adev))
+		return;
+
 	u32 tmp;

Should be the first thing in the function body, as noted above.

Regards,
Luben

drivers/gpu/drm/amd/amdgpu/mmhub_v2_0.c void mmhub_v2_0_gart_disable(struct amdgpu_device *adev) · line 378

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/mmhub_v2_0.c +1 -0
+
 	tmp = RREG32_SOC15(MMHUB, 0, mmMMVM_L2_PROTECTION_FAULT_CNTL);
 	tmp = REG_SET_FIELD(tmp, MMVM_L2_PROTECTION_FAULT_CNTL,
 			    RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);

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