Skip to contentSkip to conversation

drm/amdgpu: support ras on SRIOV

PATCH In review Stanley.Yang 2 participants 2 messages May 18, 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

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  4 ++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c    |  4 ++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c    | 23 ++++++++++++++++++----
 drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c   |  4 ++++
 drivers/gpu/drm/amd/amdgpu/psp_v13_0.c     |  9 ++++++---
 5 files changed, 37 insertions(+), 7 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +4 -0
index 85412e1a04be..e832c5bceb63 100644
@@ -5372,6 +5372,10 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
 		r = amdgpu_device_reset_sriov(adev, job ? false : true);
 		if (r)
 			adev->asic_reset_res = r;
+
+		/* Aldebaran support 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_irq.c +4 -0
index 31e07dfc874b..12a1f2389714 100644
@@ -202,6 +202,10 @@ irqreturn_t amdgpu_irq_handler(int irq, void *arg)
 	if (ret == IRQ_HANDLED)
 		pm_runtime_mark_last_busy(dev->dev);
 
+	/* Fatal error events are handled on host side */
+	if (amdgpu_sriov_vf(adev))
+		return ret;
+
 	/* For the hardware that cannot enable bif ring for both ras_controller_irq
          * and ras_err_evnet_athub_irq ih cookies, the driver has to poll status
 	 * register to check whether the interrupt is triggered or not, and properly
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +19 -4
index 2b80a3037481..930fa3837ef9 100644
@@ -721,7 +721,9 @@ 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()) {
+	/* Enable ras feature operator 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",
@@ -2181,10 +2183,14 @@ 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");
@@ -2196,8 +2202,13 @@ 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);
+			if (!amdgpu_sriov_vf(adev))
+				adev->ras_hw_enabled |= ~(1 << AMDGPU_RAS_BLOCK__UMC |
+							    1 << AMDGPU_RAS_BLOCK__DF);
+			else
+				adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__PCIE_BIF |
+								1 << AMDGPU_RAS_BLOCK__SDMA |
+								1 << AMDGPU_RAS_BLOCK__GFX);
 		} else {
 			dev_info(adev->dev, "SRAM ECC is not presented.\n");
 		}
@@ -2532,6 +2543,10 @@ 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");
drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c +4 -0
index 3b5c43575aa3..72bfac9bf9d4 100644
@@ -123,6 +123,10 @@ 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 2c6070b90dcf..165cdc2d7f0b 100644
@@ -75,9 +75,12 @@ 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
#1 lore raw

[AMD Official Use Only - General]

Quoted 32 lines

-----Original Message-----
From: Stanley.Yang <[email protected]>
Sent: Wednesday, May 18, 2022 4:32 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

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  4 ++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c    |  4 ++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c    | 23
++++++++++++++++++----
 drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c   |  4 ++++
 drivers/gpu/drm/amd/amdgpu/psp_v13_0.c     |  9 ++++++---
 5 files changed, 37 insertions(+), 7 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +2 -0
index 85412e1a04be..e832c5bceb63 100644
@@ -5372,6 +5372,10 @@ int amdgpu_device_gpu_recover(struct
 amdgpu_device *adev,
 		r = amdgpu_device_reset_sriov(adev, job ? false : true);
 		if (r)
 			adev->asic_reset_res = r;
+
+		/* Aldebaran support ras in SRIOV, so need resume ras
Quoted 1 line

during reset */

[Tao] support -> supports

Quoted 12 lines
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);

drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +0 -0
 		if (r && r == -EAGAIN)
drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c +0 -0
index 31e07dfc874b..12a1f2389714 100644

drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c irqreturn_t amdgpu_irq_handler(int irq, void *arg) · line 207

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c +3 -0
@@ -202,6 +202,10 @@ irqreturn_t amdgpu_irq_handler(int irq, void *arg)
 	if (ret == IRQ_HANDLED)
 		pm_runtime_mark_last_busy(dev->dev);
 
+	/* Fatal error events are handled on host side */
+	if (amdgpu_sriov_vf(adev))
+		return ret;

[Tao]: can we place the code in amdgpu_ras_interrupt_fatal_error_handler?

Quoted 11 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c +1 -0
+
 	/* For the hardware that cannot enable bif ring for both

ras_controller_irq
* and ras_err_evnet_athub_irq ih cookies, the driver has to poll status
* register to check whether the interrupt is triggered or not, and
properly diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index 2b80a3037481..930fa3837ef9 100644

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c int amdgpu_ras_feature_enable(struct · line 726

Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +1 -1
@@ -721,7 +721,9 @@ 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()) {
+	/* Enable ras feature operator handle on host side */

[Tao] "Only Enable ras ..." is better.

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c int amdgpu_ras_feature_enable(struct amdgpu_device *adev, · line 726

Quoted 2 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +2 -0
+	if (!amdgpu_sriov_vf(adev) &&
+			!amdgpu_ras_intr_triggered()) {

[Tao]: it's better to replace the two Tabs with four spaces.

Quoted 14 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +0 -0
 		ret = psp_ras_enable_features(&adev->psp, info, enable);
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +0 -0
 		if (ret) {
 			dev_err(adev->dev, "ras %s %s failed poison:%d

ret:%d\n", @@ -2181,10 +2183,14 @@ static void amdgpu_ras_check_supported(struct amdgpu_device *adev) {

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +3 -1
 	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] ==
Quoted 1 line

IP_VERSION(13, 0, 2))))

[Tao] replace the tabs with spaces.

Quoted 84 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +2 -0
+		return;
+
 	if (!adev->gmc.xgmi.connected_to_cpu) {
 		if (amdgpu_atomfirmware_mem_ecc_supported(adev)) {
 			dev_info(adev->dev, "MEM ECC is active.\n"); @@ -

2196,8 +2202,13 @@ 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 +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
+			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 +0 -0
 		} else {
 			dev_info(adev->dev, "SRAM ECC is not

presented.\n");

drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +4 -0
 		}
@@ -2532,6 +2543,10 @@ 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 3b5c43575aa3..72bfac9bf9d4 100644

drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.c +4 -0
@@ -123,6 +123,10 @@ 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 2c6070b90dcf..165cdc2d7f0b 100644
@@ -75,9 +75,12 @@ 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;
drivers/gpu/drm/amd/amdgpu/psp_v13_0.c +0 -1
 	case IP_VERSION(13, 0, 1):
 	case IP_VERSION(13, 0, 3):
--

2.17.1