Skip to contentSkip to conversation

drm/amdgpu: Optimze checking ras supported

PATCH In review Stanley.Yang 2 participants 5 messages Jun 13, 2023

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

PATCH 1/2 drm/amdgpu: Optimze checking ras supported

Using "is_app_apu" to identify device in the native APU mode or carveout mode.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c |  8 +++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 34 ++++++++++++++-----------
 3 files changed, 23 insertions(+), 21 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c +1 -1
index 78bacea951a9..352e958b190a 100644
@@ -1653,7 +1653,7 @@ int psp_ras_initialize(struct psp_context *psp)
 
 	if (amdgpu_ras_is_poison_mode_supported(adev))
 		ras_cmd->ras_in_message.init_flags.poison_mode_en = 1;
-	if (!adev->gmc.xgmi.connected_to_cpu)
+	if (!adev->gmc.xgmi.connected_to_cpu && !adev->gmc.is_app_apu)
 		ras_cmd->ras_in_message.init_flags.dgpu_mode = 1;
 	ras_cmd->ras_in_message.init_flags.xcc_mask =
 		adev->gfx.xcc_mask;
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +3 -5
index 7a0924469e4f..56bb0db207b9 100644
@@ -1689,8 +1689,7 @@ static void amdgpu_ras_interrupt_poison_consumption_handler(struct ras_manager *
 		}
 	}
 
-	if (!adev->gmc.xgmi.connected_to_cpu)
-		amdgpu_umc_poison_handler(adev, false);
+	amdgpu_umc_poison_handler(adev, false);
 
 	if (block_obj->hw_ops && block_obj->hw_ops->handle_poison_consumption)
 		poison_stat = block_obj->hw_ops->handle_poison_consumption(adev);
@@ -2458,11 +2457,10 @@ static void amdgpu_ras_check_supported(struct amdgpu_device *adev)
 {
 	adev->ras_hw_enabled = adev->ras_enabled = 0;
 
-	if (!adev->is_atom_fw ||
-	    !amdgpu_ras_asic_supported(adev))
+	if (!amdgpu_ras_asic_supported(adev))
 		return;
 
-	if (!adev->gmc.xgmi.connected_to_cpu) {
+	if (!adev->gmc.xgmi.connected_to_cpu &&	!adev->gmc.is_app_apu) {
 		if (amdgpu_atomfirmware_mem_ecc_supported(adev)) {
 			dev_info(adev->dev, "MEM ECC is active.\n");
 			adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__UMC |
drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +19 -15
index 1edf8e6aeb16..db0d94ca4ffc 100644
@@ -169,27 +169,31 @@ int amdgpu_umc_poison_handler(struct amdgpu_device *adev, bool reset)
 {
 	int ret = AMDGPU_RAS_SUCCESS;
 
-	if (!amdgpu_sriov_vf(adev)) {
-		if (!adev->gmc.xgmi.connected_to_cpu) {
-			struct ras_err_data err_data = {0, 0, 0, NULL};
-			struct ras_common_if head = {
-				.block = AMDGPU_RAS_BLOCK__UMC,
-			};
-			struct ras_manager *obj = amdgpu_ras_find_obj(adev, &head);
-
-			ret = amdgpu_umc_do_page_retirement(adev, &err_data, NULL, reset);
-
-			if (ret == AMDGPU_RAS_SUCCESS && obj) {
-				obj->err_data.ue_count += err_data.ue_count;
-				obj->err_data.ce_count += err_data.ce_count;
-			}
-		} else if (reset) {
+	if (adev->gmc.xgmi.connected_to_cpu ||
+		adev->gmc.is_app_apu) {
+		if (reset) {
 			/* MCA poison handler is only responsible for GPU reset,
 			 * let MCA notifier do page retirement.
 			 */
 			kgd2kfd_set_sram_ecc_flag(adev->kfd.dev);
 			amdgpu_ras_reset_gpu(adev);
 		}
+		return ret;
+	}
+
+	if (!amdgpu_sriov_vf(adev)) {
+		struct ras_err_data err_data = {0, 0, 0, NULL};
+		struct ras_common_if head = {
+			.block = AMDGPU_RAS_BLOCK__UMC,
+		};
+		struct ras_manager *obj = amdgpu_ras_find_obj(adev, &head);
+
+		ret = amdgpu_umc_do_page_retirement(adev, &err_data, NULL, reset);
+
+		if (ret == AMDGPU_RAS_SUCCESS && obj) {
+			obj->err_data.ue_count += err_data.ue_count;
+			obj->err_data.ce_count += err_data.ce_count;
+		}
 	} else {
 		if (adev->virt.ops && adev->virt.ops->ras_poison_handler)
 			adev->virt.ops->ras_poison_handler(adev);
Signature · 1 line
2.17.1
#1 lore raw

PATCH 2/2 drm/amdgpu: Add checking mc_vram_size

Do not compare injection address with mc_vram_size if mc_vram_size is zero.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +2 -1
index 56bb0db207b9..3c041efcf0c4 100644
@@ -494,7 +494,8 @@ static ssize_t amdgpu_ras_debugfs_ctrl_write(struct file *f,
 		ret = amdgpu_ras_feature_enable(adev, &data.head, 1);
 		break;
 	case 2:
-		if ((data.inject.address >= adev->gmc.mc_vram_size) ||
+		if ((data.inject.address >= adev->gmc.mc_vram_size &&
+		    adev->gmc.mc_vram_size) ||
 		    (data.inject.address >= RAS_UMC_INJECT_ADDR_LIMIT)) {
 			dev_warn(adev->dev, "RAS WARN: input address "
 					"0x%llx is invalid.",
Signature · 1 line
2.17.1
Reviewed-by on v1 2/2
#2 lore raw

RE: [PATCH Review 2/2] drm/amdgpu: Add checking mc_vram_size

[AMD Official Use Only - General]

With my concerns fixed, the series is:

Quoted 33 lines

-----Original Message-----
From: Stanley.Yang <[email protected]>
Sent: Tuesday, June 13, 2023 11:53 AM
To: [email protected]; Zhang, Hawking <[email protected]>; Zhou1, Tao <[email protected]>
Cc: Yang, Stanley <[email protected]>
Subject: [PATCH Review 2/2] drm/amdgpu: Add checking mc_vram_size

Do not compare injection address with mc_vram_size if mc_vram_size is zero.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +2 -1
index 56bb0db207b9..3c041efcf0c4 100644
@@ -494,7 +494,8 @@ static ssize_t amdgpu_ras_debugfs_ctrl_write(struct file
 *f,
              ret = amdgpu_ras_feature_enable(adev, &data.head, 1);
              break;
      case 2:
-             if ((data.inject.address >= adev->gmc.mc_vram_size) ||
+             if ((data.inject.address >= adev->gmc.mc_vram_size &&
+                 adev->gmc.mc_vram_size) ||
                  (data.inject.address >= RAS_UMC_INJECT_ADDR_LIMIT)) {
                      dev_warn(adev->dev, "RAS WARN: input address "
                                      "0x%llx is invalid.",

--
2.17.1

#3 lore raw
In reply to Stanley.Yang

[AMD Official Use Only - General]

[Tao] typo in title: Optimze -> Optimize

Quoted 50 lines

-----Original Message-----
From: Stanley.Yang <[email protected]>
Sent: Tuesday, June 13, 2023 11:53 AM
To: [email protected]; Zhang, Hawking <[email protected]>; Zhou1, Tao <[email protected]>
Cc: Yang, Stanley <[email protected]>
Subject: [PATCH Review 1/2] drm/amdgpu: Optimze checking ras supported

Using "is_app_apu" to identify device in the native APU mode or carveout mode.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c |  2 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c |  8 +++---
drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 34 ++++++++++++++-----------
 3 files changed, 23 insertions(+), 21 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c +1 -1
index 78bacea951a9..352e958b190a 100644
@@ -1653,7 +1653,7 @@ int psp_ras_initialize(struct psp_context *psp)
 
      if (amdgpu_ras_is_poison_mode_supported(adev))
              ras_cmd->ras_in_message.init_flags.poison_mode_en = 1;
-     if (!adev->gmc.xgmi.connected_to_cpu)
+     if (!adev->gmc.xgmi.connected_to_cpu && !adev->gmc.is_app_apu)
              ras_cmd->ras_in_message.init_flags.dgpu_mode = 1;
      ras_cmd->ras_in_message.init_flags.xcc_mask =
              adev->gfx.xcc_mask;
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +0 -0
index 7a0924469e4f..56bb0db207b9 100644
@@ -1689,8 +1689,7 @@ static void

amdgpu_ras_interrupt_poison_consumption_handler(struct ras_manager *

              }
      }
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +1 -2
-     if (!adev->gmc.xgmi.connected_to_cpu)
-             amdgpu_umc_poison_handler(adev, false);
+     amdgpu_umc_poison_handler(adev, false);
 
      if (block_obj->hw_ops && block_obj->hw_ops-
Quoted 1 line

handle_poison_consumption)

              poison_stat = block_obj->hw_ops-
Quoted 1 line

handle_poison_consumption(adev);

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c static void amdgpu_ras_check_supported(struct · line 2463

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +2 -3
@@ -2458,11 +2457,10 @@ static void amdgpu_ras_check_supported(struct
 amdgpu_device *adev)  {
      adev->ras_hw_enabled = adev->ras_enabled = 0;
 
-     if (!adev->is_atom_fw ||
-         !amdgpu_ras_asic_supported(adev))
+     if (!amdgpu_ras_asic_supported(adev))
              return;
 
-     if (!adev->gmc.xgmi.connected_to_cpu) {
+     if (!adev->gmc.xgmi.connected_to_cpu && !adev-

[Tao] the tab should be replaced with space.

Quoted 62 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +0 -0
 >gmc.is_app_apu) {
              if (amdgpu_atomfirmware_mem_ecc_supported(adev)) {
                      dev_info(adev->dev, "MEM ECC is active.\n");
                      adev->ras_hw_enabled |= (1 <<

AMDGPU_RAS_BLOCK__UMC | diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index 1edf8e6aeb16..db0d94ca4ffc 100644

drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +19 -15
@@ -169,27 +169,31 @@ int amdgpu_umc_poison_handler(struct
 amdgpu_device *adev, bool reset)  {
      int ret = AMDGPU_RAS_SUCCESS;
 
-     if (!amdgpu_sriov_vf(adev)) {
-             if (!adev->gmc.xgmi.connected_to_cpu) {
-                     struct ras_err_data err_data = {0, 0, 0, NULL};
-                     struct ras_common_if head = {
-                             .block = AMDGPU_RAS_BLOCK__UMC,
-                     };
-                     struct ras_manager *obj = amdgpu_ras_find_obj(adev,
 &head);
-
-                     ret = amdgpu_umc_do_page_retirement(adev,
 &err_data, NULL, reset);
-
-                     if (ret == AMDGPU_RAS_SUCCESS && obj) {
-                             obj->err_data.ue_count += err_data.ue_count;
-                             obj->err_data.ce_count += err_data.ce_count;
-                     }
-             } else if (reset) {
+     if (adev->gmc.xgmi.connected_to_cpu ||
+             adev->gmc.is_app_apu) {
+             if (reset) {
                      /* MCA poison handler is only responsible for GPU reset,
                       * let MCA notifier do page retirement.
                       */
                      kgd2kfd_set_sram_ecc_flag(adev->kfd.dev);
                      amdgpu_ras_reset_gpu(adev);
              }
+             return ret;
+     }
+
+     if (!amdgpu_sriov_vf(adev)) {
+             struct ras_err_data err_data = {0, 0, 0, NULL};
+             struct ras_common_if head = {
+                     .block = AMDGPU_RAS_BLOCK__UMC,
+             };
+             struct ras_manager *obj = amdgpu_ras_find_obj(adev, &head);
+
+             ret = amdgpu_umc_do_page_retirement(adev, &err_data, NULL,
 reset);
+
+             if (ret == AMDGPU_RAS_SUCCESS && obj) {
+                     obj->err_data.ue_count += err_data.ue_count;
+                     obj->err_data.ce_count += err_data.ce_count;
+             }
      } else {
              if (adev->virt.ops && adev->virt.ops->ras_poison_handler)
                      adev->virt.ops->ras_poison_handler(adev);

--
2.17.1

#4 lore raw

[AMD Official Use Only - General]

Quoted 9 lines

-----Original Message-----
From: Zhou1, Tao <[email protected]>
Sent: Tuesday, June 13, 2023 3:08 PM
To: Yang, Stanley <[email protected]>; [email protected]; Zhang, Hawking <[email protected]>
Cc: Yang, Stanley <[email protected]>
Subject: RE: [PATCH Review 1/2] drm/amdgpu: Optimze checking ras supported

Quoted 3 lines

[AMD Official Use Only - General]

[Tao] typo in title: Optimze -> Optimize

[Stanley]: Thanks Tao, will update before submitting.

Regards,
Stanley

Quoted 132 lines
Quoted 10 lines

-----Original Message-----
From: Stanley.Yang <[email protected]>
Sent: Tuesday, June 13, 2023 11:53 AM
To: [email protected]; Zhang, Hawking <[email protected]>; Zhou1, Tao <[email protected]>
Cc: Yang, Stanley <[email protected]>
Subject: [PATCH Review 1/2] drm/amdgpu: Optimze checking ras supported

Using "is_app_apu" to identify device in the native APU mode or carveout

mode.

Quoted 6 lines
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c |  2 +-

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 8 +++--- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 34 ++++++++++++++-----

------

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c static void

Quoted 35 lines
 3 files changed, 23 insertions(+), 21 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c +1 -1
index 78bacea951a9..352e958b190a 100644
@@ -1653,7 +1653,7 @@ int psp_ras_initialize(struct psp_context *psp)
 
      if (amdgpu_ras_is_poison_mode_supported(adev))
              ras_cmd->ras_in_message.init_flags.poison_mode_en = 1;
-     if (!adev->gmc.xgmi.connected_to_cpu)
+     if (!adev->gmc.xgmi.connected_to_cpu && !adev->gmc.is_app_apu)
              ras_cmd->ras_in_message.init_flags.dgpu_mode = 1;
      ras_cmd->ras_in_message.init_flags.xcc_mask =
              adev->gfx.xcc_mask;
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +0 -0
index 7a0924469e4f..56bb0db207b9 100644
@@ -1689,8 +1689,7 @@ static void

amdgpu_ras_interrupt_poison_consumption_handler(struct ras_manager *

              }
      }
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +1 -2
-     if (!adev->gmc.xgmi.connected_to_cpu)
-             amdgpu_umc_poison_handler(adev, false);
+     amdgpu_umc_poison_handler(adev, false);
 
      if (block_obj->hw_ops && block_obj->hw_ops-
Quoted 1 line

handle_poison_consumption)

              poison_stat = block_obj->hw_ops-
Quoted 1 line

handle_poison_consumption(adev);

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +0 -0
@@ -2458,11 +2457,10 @@ static void

amdgpu_ras_check_supported(struct

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c static void amdgpu_ras_check_supported(struct amdgpu_device *adev)

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +2 -3
 amdgpu_device *adev)  {
      adev->ras_hw_enabled = adev->ras_enabled = 0;
 
-     if (!adev->is_atom_fw ||
-         !amdgpu_ras_asic_supported(adev))
+     if (!amdgpu_ras_asic_supported(adev))
              return;
 
-     if (!adev->gmc.xgmi.connected_to_cpu) {
+     if (!adev->gmc.xgmi.connected_to_cpu && !adev-

[Tao] the tab should be replaced with space.

Quoted 64 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +0 -0
 >gmc.is_app_apu) {
              if (amdgpu_atomfirmware_mem_ecc_supported(adev)) {
                      dev_info(adev->dev, "MEM ECC is active.\n");
                      adev->ras_hw_enabled |= (1 <<

AMDGPU_RAS_BLOCK__UMC | diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index 1edf8e6aeb16..db0d94ca4ffc 100644

drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +21 -16
@@ -169,27 +169,31 @@ int amdgpu_umc_poison_handler(struct
 amdgpu_device *adev, bool reset)  {
      int ret = AMDGPU_RAS_SUCCESS;
 
-     if (!amdgpu_sriov_vf(adev)) {
-             if (!adev->gmc.xgmi.connected_to_cpu) {
-                     struct ras_err_data err_data = {0, 0, 0, NULL};
-                     struct ras_common_if head = {
-                             .block = AMDGPU_RAS_BLOCK__UMC,
-                     };
-                     struct ras_manager *obj = amdgpu_ras_find_obj(adev,
 &head);
-
-                     ret = amdgpu_umc_do_page_retirement(adev,
 &err_data, NULL, reset);
-
-                     if (ret == AMDGPU_RAS_SUCCESS && obj) {
-                             obj->err_data.ue_count += err_data.ue_count;
-                             obj->err_data.ce_count += err_data.ce_count;
-                     }
-             } else if (reset) {
+     if (adev->gmc.xgmi.connected_to_cpu ||
+             adev->gmc.is_app_apu) {
+             if (reset) {
                      /* MCA poison handler is only responsible for GPU reset,
                       * let MCA notifier do page retirement.
                       */
                      kgd2kfd_set_sram_ecc_flag(adev->kfd.dev);
                      amdgpu_ras_reset_gpu(adev);
              }
+             return ret;
+     }
+
+     if (!amdgpu_sriov_vf(adev)) {
+             struct ras_err_data err_data = {0, 0, 0, NULL};
+             struct ras_common_if head = {
+                     .block = AMDGPU_RAS_BLOCK__UMC,
+             };
+             struct ras_manager *obj = amdgpu_ras_find_obj(adev,
+ &head);
+
+             ret = amdgpu_umc_do_page_retirement(adev, &err_data,
+ NULL,
 reset);
+
+             if (ret == AMDGPU_RAS_SUCCESS && obj) {
+                     obj->err_data.ue_count += err_data.ue_count;
+                     obj->err_data.ce_count += err_data.ce_count;
+             }
      } else {
              if (adev->virt.ops && adev->virt.ops->ras_poison_handler)
                      adev->virt.ops->ras_poison_handler(adev);
--

2.17.1