Skip to contentSkip to conversation

drm/amdgpu: support ras on SRIOV

PATCH In review Stanley.Yang 2 participants 3 messages May 18, 2022 – May 19, 2022

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

PATCH 1/1 drm/amdgpu: support ras on SRIOV

support umc/gfx/sdma ras on guest side

Changed from V1:
move sriov judgment in amdgpu_ras_interrupt_fatal_error_handler

Change-Id Ic7dda45d8f8cf2d5f1abc7705abc153d558da8a1
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  4 +++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c    | 42 ++++++++++++++++------
 drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c   |  4 +++
 drivers/gpu/drm/amd/amdgpu/psp_v13_0.c     |  9 +++--
 4 files changed, 45 insertions(+), 14 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +4 -0
index b583026dc893..ba7990d0dc0e 100644
@@ -5218,6 +5218,10 @@ int amdgpu_device_gpu_recover_imp(struct amdgpu_device *adev,
 		r = amdgpu_device_reset_sriov(adev, job ? false : true);
 		if (r)
 			adev->asic_reset_res = r;
+
+		/* Aldebaran supports ras in SRIOV, so need resume ras during reset */
+		if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2))
+			amdgpu_ras_resume(adev);
 	} else {
 		r = amdgpu_do_asic_reset(device_list_handle, &reset_context);
 		if (r && r == -EAGAIN)
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +31 -11

75 lines not on this page · raw message

drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c +4 -0
index 8e221a1ba937..42c1f050542f 100644
@@ -124,6 +124,10 @@ int amdgpu_sdma_process_ras_data_cb(struct amdgpu_device *adev,
 		struct amdgpu_iv_entry *entry)
 {
 	kgd2kfd_set_sram_ecc_flag(adev->kfd.dev);
+
+	if (amdgpu_sriov_vf(adev))
+		return AMDGPU_RAS_SUCCESS;
+
 	amdgpu_ras_reset_gpu(adev);
 
 	return AMDGPU_RAS_SUCCESS;
drivers/gpu/drm/amd/amdgpu/psp_v13_0.c +6 -3
index d6d79e97def9..18014ed0e853 100644
@@ -85,9 +85,12 @@ static int psp_v13_0_init_microcode(struct psp_context *psp)
 		err = psp_init_sos_microcode(psp, chip_name);
 		if (err)
 			return err;
-		err = psp_init_ta_microcode(&adev->psp, chip_name);
-		if (err)
-			return err;
+		/* It's not necessary to load ras ta on Guest side */
+		if (!amdgpu_sriov_vf(adev)) {
+			err = psp_init_ta_microcode(&adev->psp, chip_name);
+			if (err)
+				return err;
+		}
 		break;
 	case IP_VERSION(13, 0, 1):
 	case IP_VERSION(13, 0, 3):
Signature · 1 line
2.17.1
Reviewed-by on v1 1/1
#1 lore raw
Quoted 63 lines

-----Original Message-----
From: Stanley.Yang <[email protected]>
Sent: Wednesday, May 18, 2022 11:44 PM
To: [email protected]; Zhang, Hawking <[email protected]>; Zhou1, Tao <[email protected]>
Cc: Yang, Stanley <[email protected]>
Subject: [PATCH Review 1/1] drm/amdgpu: support ras on SRIOV

support umc/gfx/sdma ras on guest side

Changed from V1:
move sriov judgment in amdgpu_ras_interrupt_fatal_error_handler

Change-Id Ic7dda45d8f8cf2d5f1abc7705abc153d558da8a1
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  4 +++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c    | 42 ++++++++++++++++------
 drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c   |  4 +++
 drivers/gpu/drm/amd/amdgpu/psp_v13_0.c     |  9 +++--
 4 files changed, 45 insertions(+), 14 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +2 -0
index b583026dc893..ba7990d0dc0e 100644
@@ -5218,6 +5218,10 @@ int amdgpu_device_gpu_recover_imp(struct
 amdgpu_device *adev,
 		r = amdgpu_device_reset_sriov(adev, job ? false : true);
 		if (r)
 			adev->asic_reset_res = r;
+
+		/* Aldebaran supports ras in SRIOV, so need resume ras during

reset */

drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +2 -0
+		if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2))
+			amdgpu_ras_resume(adev);
 	} else {
 		r = amdgpu_do_asic_reset(device_list_handle, &reset_context);
 		if (r && r == -EAGAIN)
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +3 -1
index a653cf3b3d13..2b28210c4994 100644
@@ -726,7 +726,9 @@ int amdgpu_ras_feature_enable(struct amdgpu_device
 *adev,
 	/* Do not enable if it is not allowed. */
 	WARN_ON(enable && !amdgpu_ras_is_feature_allowed(adev, head));
 
-	if (!amdgpu_ras_intr_triggered()) {
+	/* Only enable ras feature operation handle on host side */
+	if (!amdgpu_sriov_vf(adev) &&
+		!amdgpu_ras_intr_triggered()) {
 		ret = psp_ras_enable_features(&adev->psp, info, enable);
 		if (ret) {

dev_err(adev->dev, "ras %s %s failed poison:%d
ret:%d\n", @@ -1523,6 +1525,10 @@ static int amdgpu_ras_fs_fini(struct amdgpu_device *adev)
*/
void amdgpu_ras_interrupt_fatal_error_handler(struct amdgpu_device *adev) {

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c static int amdgpu_ras_fs_fini(struct amdgpu_device *adev) · line 1533

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +4 -0
+	/* Fatal error events are handled on host side */
+	if (amdgpu_sriov_vf(adev))
+		return;
+
 	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__PCIE_BIF))
 		return;

[Tao] The two conditions above can be merged, other than that the patch is:

Quoted 119 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +5 -1
@@ -2270,10 +2276,14 @@ static void amdgpu_ras_check_supported(struct
 amdgpu_device *adev)  {
 	adev->ras_hw_enabled = adev->ras_enabled = 0;
 
-	if (amdgpu_sriov_vf(adev) || !adev->is_atom_fw ||
+	if (!adev->is_atom_fw ||
 	    !amdgpu_ras_asic_supported(adev))
 		return;
 
+	if (!(amdgpu_sriov_vf(adev) &&
+		(adev->ip_versions[MP1_HWIP][0] == IP_VERSION(13, 0, 2))))
+		return;
+
 	if (!adev->gmc.xgmi.connected_to_cpu) {
 		if (amdgpu_atomfirmware_mem_ecc_supported(adev)) {
 			dev_info(adev->dev, "MEM ECC is active.\n"); @@ -

2285,15 +2295,21 @@ static void amdgpu_ras_check_supported(struct amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +0 -1
 		if (amdgpu_atomfirmware_sram_ecc_supported(adev)) {
 			dev_info(adev->dev, "SRAM ECC is active.\n");
-			adev->ras_hw_enabled |= ~(1 <<

AMDGPU_RAS_BLOCK__UMC |

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +0 -1
-						    1 <<

AMDGPU_RAS_BLOCK__DF);

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +0 -2
-
-			if (adev->ip_versions[VCN_HWIP][0] == IP_VERSION(2,

6, 0))

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +0 -1
-				adev->ras_hw_enabled |= (1 <<

AMDGPU_RAS_BLOCK__VCN |

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +0 -1
-						1 <<

AMDGPU_RAS_BLOCK__JPEG);

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +0 -2
-			else
-				adev->ras_hw_enabled &= ~(1 <<

AMDGPU_RAS_BLOCK__VCN |

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +0 -1
-						1 <<

AMDGPU_RAS_BLOCK__JPEG);

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +2 -0
+			if (!amdgpu_sriov_vf(adev)) {
+				adev->ras_hw_enabled |= ~(1 <<

AMDGPU_RAS_BLOCK__UMC |

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +1 -0
+							    1 <<

AMDGPU_RAS_BLOCK__DF);

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +2 -0
+
+				if (adev->ip_versions[VCN_HWIP][0] ==

IP_VERSION(2, 6, 0))

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +1 -0
+					adev->ras_hw_enabled |= (1 <<

AMDGPU_RAS_BLOCK__VCN |

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +1 -0
+							1 <<

AMDGPU_RAS_BLOCK__JPEG);

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +2 -0
+				else
+					adev->ras_hw_enabled &= ~(1 <<

AMDGPU_RAS_BLOCK__VCN |

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +1 -0
+							1 <<

AMDGPU_RAS_BLOCK__JPEG);

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +2 -0
+			} else {
+				adev->ras_hw_enabled |= (1 <<

AMDGPU_RAS_BLOCK__PCIE_BIF |

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +1 -0
+								1 <<

AMDGPU_RAS_BLOCK__SDMA |

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +1 -0
+								1 <<

AMDGPU_RAS_BLOCK__GFX);

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +5 -0
+			}
 		} else {
 			dev_info(adev->dev, "SRAM ECC is not presented.\n");
 		}
@@ -2637,6 +2653,10 @@ int amdgpu_ras_late_init(struct amdgpu_device
 *adev)
 	struct amdgpu_ras_block_object *obj;
 	int r;
 
+	/* Guest side doesn't need init ras feature */
+	if (amdgpu_sriov_vf(adev))
+		return 0;
+
 	list_for_each_entry_safe(node, tmp, &adev->ras_list, node) {
 		if (!node->ras_obj) {

dev_warn(adev->dev, "Warning: abnormal ras list
node.\n"); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c
index 8e221a1ba937..42c1f050542f 100644

drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c +4 -0
@@ -124,6 +124,10 @@ int amdgpu_sdma_process_ras_data_cb(struct
 amdgpu_device *adev,
 		struct amdgpu_iv_entry *entry)
 {
 	kgd2kfd_set_sram_ecc_flag(adev->kfd.dev);
+
+	if (amdgpu_sriov_vf(adev))
+		return AMDGPU_RAS_SUCCESS;
+
 	amdgpu_ras_reset_gpu(adev);
 
drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c +0 -0
 	return AMDGPU_RAS_SUCCESS;
drivers/gpu/drm/amd/amdgpu/psp_v13_0.c +6 -3
index d6d79e97def9..18014ed0e853 100644
@@ -85,9 +85,12 @@ static int psp_v13_0_init_microcode(struct psp_context
 *psp)
 		err = psp_init_sos_microcode(psp, chip_name);
 		if (err)
 			return err;
-		err = psp_init_ta_microcode(&adev->psp, chip_name);
-		if (err)
-			return err;
+		/* It's not necessary to load ras ta on Guest side */
+		if (!amdgpu_sriov_vf(adev)) {
+			err = psp_init_ta_microcode(&adev->psp, chip_name);
+			if (err)
+				return err;
+		}
 		break;
 	case IP_VERSION(13, 0, 1):

case IP_VERSION(13, 0, 3):
--
2.17.1

author Reviewed-by Tao Zhou
#2 lore raw

[AMD Official Use Only - General]

[AMD Official Use Only - General]
Thanks tao, will update before submit.

Regards,
Stanley
发件人: Zhou1, Tao <[email protected]>
日期: 星期四, 2022年5月19日 上午10:30
收件人: Yang, Stanley <[email protected]>, [email protected] <[email protected]>, Zhang, Hawking <[email protected]> 抄送: Yang, Stanley <[email protected]>
主题: RE: [PATCH Review 1/1] drm/amdgpu: support ras on SRIOV

Quoted 61 lines

-----Original Message-----
From: Stanley.Yang <[email protected]>
Sent: Wednesday, May 18, 2022 11:44 PM
To: [email protected]; Zhang, Hawking <[email protected]>; Zhou1, Tao <[email protected]>
Cc: Yang, Stanley <[email protected]>
Subject: [PATCH Review 1/1] drm/amdgpu: support ras on SRIOV

support umc/gfx/sdma ras on guest side

Changed from V1:
move sriov judgment in amdgpu_ras_interrupt_fatal_error_handler

Change-Id Ic7dda45d8f8cf2d5f1abc7705abc153d558da8a1
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  4 +++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c    | 42 ++++++++++++++++------
 drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c   |  4 +++
 drivers/gpu/drm/amd/amdgpu/psp_v13_0.c     |  9 +++--
 4 files changed, 45 insertions(+), 14 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +2 -0
index b583026dc893..ba7990d0dc0e 100644
@@ -5218,6 +5218,10 @@ int amdgpu_device_gpu_recover_imp(struct
 amdgpu_device *adev,
               r = amdgpu_device_reset_sriov(adev, job ? false : true);
               if (r)
                       adev->asic_reset_res = r;
+
+             /* Aldebaran supports ras in SRIOV, so need resume ras during

reset */

drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +2 -0
+             if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2))
+                     amdgpu_ras_resume(adev);
       } else {
               r = amdgpu_do_asic_reset(device_list_handle, &reset_context);
               if (r && r == -EAGAIN)
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +3 -1
index a653cf3b3d13..2b28210c4994 100644
@@ -726,7 +726,9 @@ int amdgpu_ras_feature_enable(struct amdgpu_device
 *adev,
       /* Do not enable if it is not allowed. */
       WARN_ON(enable && !amdgpu_ras_is_feature_allowed(adev, head));
 
-     if (!amdgpu_ras_intr_triggered()) {
+     /* Only enable ras feature operation handle on host side */
+     if (!amdgpu_sriov_vf(adev) &&
+             !amdgpu_ras_intr_triggered()) {
               ret = psp_ras_enable_features(&adev->psp, info, enable);
               if (ret) {
                       dev_err(adev->dev, "ras %s %s failed poison:%d

ret:%d\n", @@ -1523,6 +1525,10 @@ static int amdgpu_ras_fs_fini(struct amdgpu_device *adev)
*/
void amdgpu_ras_interrupt_fatal_error_handler(struct amdgpu_device *adev) {

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c static int amdgpu_ras_fs_fini(struct amdgpu_device *adev) · line 1533

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +4 -0
+     /* Fatal error events are handled on host side */
+     if (amdgpu_sriov_vf(adev))
+             return;
+
       if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__PCIE_BIF))
               return;

[Tao] The two conditions above can be merged, other than that the patch is:

Quoted 119 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +5 -1
@@ -2270,10 +2276,14 @@ static void amdgpu_ras_check_supported(struct
 amdgpu_device *adev)  {
       adev->ras_hw_enabled = adev->ras_enabled = 0;
 
-     if (amdgpu_sriov_vf(adev) || !adev->is_atom_fw ||
+     if (!adev->is_atom_fw ||
           !amdgpu_ras_asic_supported(adev))
               return;
 
+     if (!(amdgpu_sriov_vf(adev) &&
+             (adev->ip_versions[MP1_HWIP][0] == IP_VERSION(13, 0, 2))))
+             return;
+
       if (!adev->gmc.xgmi.connected_to_cpu) {
               if (amdgpu_atomfirmware_mem_ecc_supported(adev)) {
                       dev_info(adev->dev, "MEM ECC is active.\n"); @@ -

2285,15 +2295,21 @@ static void amdgpu_ras_check_supported(struct amdgpu_device *adev)

               if (amdgpu_atomfirmware_sram_ecc_supported(adev)) {
                       dev_info(adev->dev, "SRAM ECC is active.\n");

- adev->ras_hw_enabled |= ~(1 <<
AMDGPU_RAS_BLOCK__UMC |
- 1 <<
AMDGPU_RAS_BLOCK__DF);

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +0 -2
-
-                     if (adev->ip_versions[VCN_HWIP][0] == IP_VERSION(2,

6, 0))
- adev->ras_hw_enabled |= (1 <<
AMDGPU_RAS_BLOCK__VCN |
- 1 <<
AMDGPU_RAS_BLOCK__JPEG);

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +0 -2
-                     else
-                             adev->ras_hw_enabled &= ~(1 <<

AMDGPU_RAS_BLOCK__VCN |
- 1 <<
AMDGPU_RAS_BLOCK__JPEG);

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +2 -0
+                     if (!amdgpu_sriov_vf(adev)) {
+                             adev->ras_hw_enabled |= ~(1 <<

AMDGPU_RAS_BLOCK__UMC |
+ 1 <<
AMDGPU_RAS_BLOCK__DF);

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +2 -0
+
+                             if (adev->ip_versions[VCN_HWIP][0] ==

IP_VERSION(2, 6, 0))
+ adev->ras_hw_enabled |= (1 <<
AMDGPU_RAS_BLOCK__VCN |
+ 1 <<
AMDGPU_RAS_BLOCK__JPEG);

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +2 -0
+                             else
+                                     adev->ras_hw_enabled &= ~(1 <<

AMDGPU_RAS_BLOCK__VCN |
+ 1 <<
AMDGPU_RAS_BLOCK__JPEG);

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +2 -0
+                     } else {
+                             adev->ras_hw_enabled |= (1 <<

AMDGPU_RAS_BLOCK__PCIE_BIF |
+ 1 <<
AMDGPU_RAS_BLOCK__SDMA |
+ 1 <<
AMDGPU_RAS_BLOCK__GFX);

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +5 -0
+                     }
               } else {
                       dev_info(adev->dev, "SRAM ECC is not presented.\n");
               }
@@ -2637,6 +2653,10 @@ int amdgpu_ras_late_init(struct amdgpu_device
 *adev)
       struct amdgpu_ras_block_object *obj;
       int r;
 
+     /* Guest side doesn't need init ras feature */
+     if (amdgpu_sriov_vf(adev))
+             return 0;
+
       list_for_each_entry_safe(node, tmp, &adev->ras_list, node) {
               if (!node->ras_obj) {
                       dev_warn(adev->dev, "Warning: abnormal ras list

node.\n"); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c
index 8e221a1ba937..42c1f050542f 100644

drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c +4 -0
@@ -124,6 +124,10 @@ int amdgpu_sdma_process_ras_data_cb(struct
 amdgpu_device *adev,
               struct amdgpu_iv_entry *entry)
 {
       kgd2kfd_set_sram_ecc_flag(adev->kfd.dev);
+
+     if (amdgpu_sriov_vf(adev))
+             return AMDGPU_RAS_SUCCESS;
+
       amdgpu_ras_reset_gpu(adev);
 
       return AMDGPU_RAS_SUCCESS;
drivers/gpu/drm/amd/amdgpu/psp_v13_0.c +6 -3
index d6d79e97def9..18014ed0e853 100644
@@ -85,9 +85,12 @@ static int psp_v13_0_init_microcode(struct psp_context
 *psp)
               err = psp_init_sos_microcode(psp, chip_name);
               if (err)
                       return err;
-             err = psp_init_ta_microcode(&adev->psp, chip_name);
-             if (err)
-                     return err;
+             /* It's not necessary to load ras ta on Guest side */
+             if (!amdgpu_sriov_vf(adev)) {
+                     err = psp_init_ta_microcode(&adev->psp, chip_name);
+                     if (err)
+                             return err;
+             }
               break;
       case IP_VERSION(13, 0, 1):
       case IP_VERSION(13, 0, 3):

--
2.17.1