[PATCH 00/20] RAS support for amdgpu

[PATCH 00/20] RAS support for amdgpu

[Alex Deucher]

This patch set adds initial RAS (Reliability, Availability, Serviceability)
support to amdgpu on supported boards.  Features include SRAM and VRAM ECC,
bad page tracking, and error containment.

Eric Huang (2):
  drm/amdkfd: add RAS capabilities in topology for Vega20 (v2)
  drm/amdkfd: add RAS ECC event support (v2)

Feifei Xu (1):
  drm/amdgpu: enable ras on gfx9 (v2)

xinhui pan (17):
  drm/amdgpu: add ta ras fw info (v2)
  drm/amdgpu: export ta fw info
  drm/amdgpu: add module parameters for ras
  drm/amdgpu: add ta_ras_if.h
  drm/amdgpu: add psp ras callback func and macro
  drm/amdgpu: add psp ras subsystem infrastructure (v2)
  drm/amdgpu: add psp v11 ras callback
  drm/amdgpu: add psp cmd submit timeout
  drm/amdgpu: add amdgpu_ras.c to support ras (v2)
  drm/amdgpu: add debugfs ctrl node
  drm/amdgpu: reserve bad pages during recovery
  drm/amdgpu: enable ras on sdma4
  drm/amdgpu: enable ras on gmc9
  drm/amdgpu: Add a new flag to AMDGPU_CTX_OP_QUERY_STATE2
  drm/amdgpu: add ioctl query for enabled ras features (v2)
  drm/amdgpu: skip gpu reset when ras error occured
  drm/amdgpu: add human readable debugfs control support (v2)

 drivers/gpu/drm/amd/amdgpu/Makefile        |    2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu.h        |    2 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h |    1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c    |   17 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h    |    2 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |   14 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c    |   17 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h    |    3 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h    |    2 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c    |   31 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c    |  220 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h    |   32 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c    | 1438 ++++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h    |  229 ++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h   |    4 +
 drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c      |  174 +++
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c      |  277 ++++
 drivers/gpu/drm/amd/amdgpu/psp_v11_0.c     |   57 +
 drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c     |  186 ++-
 drivers/gpu/drm/amd/amdgpu/ta_ras_if.h     |  108 ++
 drivers/gpu/drm/amd/amdkfd/kfd_device.c    |   11 +
 drivers/gpu/drm/amd/amdkfd/kfd_events.c    |   18 +-
 drivers/gpu/drm/amd/amdkfd/kfd_priv.h      |    3 +
 drivers/gpu/drm/amd/amdkfd/kfd_topology.c  |   16 +
 drivers/gpu/drm/amd/amdkfd/kfd_topology.h  |    4 +
 include/uapi/drm/amdgpu_drm.h              |   35 +
 include/uapi/linux/kfd_ioctl.h             |   12 +-
 27 files changed, 2910 insertions(+), 5 deletions(-)
 create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
 create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
 create mode 100644 drivers/gpu/drm/amd/amdgpu/ta_ras_if.h

-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 01/20] drm/amdgpu: add ta ras fw info (v2)

[Alex Deucher]

From: xinhui pan <xinhui.pan@amd.com>

Add ras fw part, xgmi and ras fw are combined together in ta binary.
Reading the data from the info is not implemented yet.

v2: squash in "drm/amdgpu: fix NULL pointer when ta is missing"

Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h | 4 ++++
 drivers/gpu/drm/amd/amdgpu/psp_v11_0.c  | 7 +++++++
 2 files changed, 11 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h
index 2ef98cc755d6..49c3942e469c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h
@@ -150,9 +150,13 @@ struct psp_context
 
 	/* xgmi ta firmware and buffer */
 	const struct firmware		*ta_fw;
+	uint32_t			ta_fw_version;
 	uint32_t			ta_xgmi_ucode_version;
 	uint32_t			ta_xgmi_ucode_size;
 	uint8_t				*ta_xgmi_start_addr;
+	uint32_t			ta_ras_ucode_version;
+	uint32_t			ta_ras_ucode_size;
+	uint8_t				*ta_ras_start_addr;
 	struct psp_xgmi_context		xgmi_context;
 };
 
diff --git a/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c b/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c
index 2c584cc9375f..8fb267684787 100644
--- a/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c
@@ -116,6 +116,13 @@ static int psp_v11_0_init_microcode(struct psp_context *psp)
 		adev->psp.ta_xgmi_ucode_size = le32_to_cpu(ta_hdr->ta_xgmi_size_bytes);
 		adev->psp.ta_xgmi_start_addr = (uint8_t *)ta_hdr +
 			le32_to_cpu(ta_hdr->header.ucode_array_offset_bytes);
+
+		adev->psp.ta_fw_version = le32_to_cpu(ta_hdr->header.ucode_version);
+
+		adev->psp.ta_ras_ucode_version = le32_to_cpu(ta_hdr->ta_ras_ucode_version);
+		adev->psp.ta_ras_ucode_size = le32_to_cpu(ta_hdr->ta_ras_size_bytes);
+		adev->psp.ta_ras_start_addr = (uint8_t *)adev->psp.ta_xgmi_start_addr +
+			le32_to_cpu(ta_hdr->ta_ras_offset_bytes);
 	}
 
 	return 0;
-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 02/20] drm/amdgpu: export ta fw info

[Alex Deucher]

From: xinhui pan <xinhui.pan@amd.com>

Output the ta fw, aka xgmi/ras, via debugfs.

Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 21 +++++++++++++++++++++
 include/uapi/drm/amdgpu_drm.h           |  1 +
 2 files changed, 22 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index b65e18101108..4d0ccca891cb 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -295,6 +295,17 @@ static int amdgpu_firmware_info(struct drm_amdgpu_info_firmware *fw_info,
 		fw_info->ver = adev->pm.fw_version;
 		fw_info->feature = 0;
 		break;
+	case AMDGPU_INFO_FW_TA:
+		if (query_fw->index > 1)
+			return -EINVAL;
+		if (query_fw->index == 0) {
+			fw_info->ver = adev->psp.ta_fw_version;
+			fw_info->feature = adev->psp.ta_xgmi_ucode_version;
+		} else {
+			fw_info->ver = adev->psp.ta_fw_version;
+			fw_info->feature = adev->psp.ta_ras_ucode_version;
+		}
+		break;
 	case AMDGPU_INFO_FW_SDMA:
 		if (query_fw->index >= adev->sdma.num_instances)
 			return -EINVAL;
@@ -1327,6 +1338,16 @@ static int amdgpu_debugfs_firmware_info(struct seq_file *m, void *data)
 	seq_printf(m, "ASD feature version: %u, firmware version: 0x%08x\n",
 		   fw_info.feature, fw_info.ver);
 
+	query_fw.fw_type = AMDGPU_INFO_FW_TA;
+	for (i = 0; i < 2; i++) {
+		query_fw.index = i;
+		ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
+		if (ret)
+			continue;
+		seq_printf(m, "TA %s feature version: %u, firmware version: 0x%08x\n",
+				i ? "RAS" : "XGMI", fw_info.feature, fw_info.ver);
+	}
+
 	/* SMC */
 	query_fw.fw_type = AMDGPU_INFO_FW_SMC;
 	ret = amdgpu_firmware_info(&fw_info, &query_fw, adev);
diff --git a/include/uapi/drm/amdgpu_drm.h b/include/uapi/drm/amdgpu_drm.h
index 4a53f6cfa034..e5275d4481f5 100644
--- a/include/uapi/drm/amdgpu_drm.h
+++ b/include/uapi/drm/amdgpu_drm.h
@@ -680,6 +680,7 @@ struct drm_amdgpu_cs_chunk_data {
 	#define AMDGPU_INFO_FW_GFX_RLC_RESTORE_LIST_SRM_MEM 0x11
 	/* Subquery id: Query DMCU firmware version */
 	#define AMDGPU_INFO_FW_DMCU		0x12
+	#define AMDGPU_INFO_FW_TA		0x13
 /* number of bytes moved for TTM migration */
 #define AMDGPU_INFO_NUM_BYTES_MOVED		0x0f
 /* the used VRAM size */
-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 03/20] drm/amdgpu: add module parameters for ras

[Alex Deucher]

From: xinhui pan <xinhui.pan@amd.com>

Allow RAS feature enable/disable via boot parameter.

Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h     |  2 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 17 +++++++++++++++++
 2 files changed, 19 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 374e1d2f2456..369548e2f0f6 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -157,6 +157,8 @@ extern int amdgpu_emu_mode;
 extern uint amdgpu_smu_memory_pool_size;
 extern uint amdgpu_dc_feature_mask;
 extern struct amdgpu_mgpu_info mgpu_info;
+extern int amdgpu_ras_enable;
+extern uint amdgpu_ras_mask;
 
 #ifdef CONFIG_DRM_AMDGPU_SI
 extern int amdgpu_si_support;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
index d35c5a273e84..b4ac80d20f75 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
@@ -137,6 +137,8 @@ uint amdgpu_dc_feature_mask = 0;
 struct amdgpu_mgpu_info mgpu_info = {
 	.mutex = __MUTEX_INITIALIZER(mgpu_info.mutex),
 };
+int amdgpu_ras_enable = -1;
+uint amdgpu_ras_mask = 0xffffffff;
 
 /**
  * DOC: vramlimit (int)
@@ -495,6 +497,21 @@ module_param_named(gpu_recovery, amdgpu_gpu_recovery, int, 0444);
 MODULE_PARM_DESC(emu_mode, "Emulation mode, (1 = enable, 0 = disable)");
 module_param_named(emu_mode, amdgpu_emu_mode, int, 0444);
 
+/**
+ * DOC: amdgpu_ras_enable (int)
+ * Enable RAS features on the GPU (0 = disable, 1 = enable, -1 = auto (default))
+ */
+MODULE_PARM_DESC(amdgpu_ras_enable, "Enable RAS features on the GPU (0 = disable, 1 = enable, -1 = auto (default))");
+module_param_named(ras_enable, amdgpu_ras_enable, int, 0444);
+
+/**
+ * DOC: amdgpu_ras_mask (uint)
+ * Mask of RAS features to enable (default 0xffffffff), only valid when ras_enable == 1
+ * See the flags in drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
+ */
+MODULE_PARM_DESC(amdgpu_ras_mask, "Mask of RAS features to enable (default 0xffffffff), only valid when ras_enable == 1");
+module_param_named(ras_mask, amdgpu_ras_mask, uint, 0444);
+
 /**
  * DOC: si_support (int)
  * Set SI support driver. This parameter works after set config CONFIG_DRM_AMDGPU_SI. For SI asic, when radeon driver is enabled,
-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 04/20] drm/amdgpu: add ta_ras_if.h

[Alex Deucher]

From: xinhui pan <xinhui.pan@amd.com>

Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/ta_ras_if.h | 108 +++++++++++++++++++++++++
 1 file changed, 108 insertions(+)
 create mode 100644 drivers/gpu/drm/amd/amdgpu/ta_ras_if.h

diff --git a/drivers/gpu/drm/amd/amdgpu/ta_ras_if.h b/drivers/gpu/drm/amd/amdgpu/ta_ras_if.h
new file mode 100644
index 000000000000..0b4e7b55595a
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/ta_ras_if.h
@@ -0,0 +1,108 @@
+/****************************************************************************\
+* 
+*  File Name      ta_ras_if.h
+*  Project        AMD PSP SW IP Module
+*
+*  Description    Interface to the RAS Trusted Application
+*
+*  Copyright 2019 Advanced Micro Devices, Inc.
+*
+* Permission is hereby granted, free of charge, to any person obtaining a copy of this software 
+* and associated documentation files (the "Software"), to deal in the Software without restriction,
+* including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense,
+* and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so,
+* subject to the following conditions:
+*
+* The above copyright notice and this permission notice shall be included in all copies or substantial
+* portions of the Software.
+*
+* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+* OTHER DEALINGS IN THE SOFTWARE.
+*/
+#ifndef _TA_RAS_IF_H
+#define _TA_RAS_IF_H
+
+/* Responses have bit 31 set */
+#define RSP_ID_MASK (1U << 31)
+#define RSP_ID(cmdId) (((uint32_t)(cmdId)) | RSP_ID_MASK)
+
+#define TA_NUM_BLOCK_MAX		14
+
+enum ras_command {
+	TA_RAS_COMMAND__ENABLE_FEATURES = 0,
+	TA_RAS_COMMAND__DISABLE_FEATURES,
+	TA_RAS_COMMAND__TRIGGER_ERROR,
+};
+
+enum ta_ras_status {
+	TA_RAS_STATUS__SUCCESS				= 0x00,
+	TA_RAS_STATUS__RESET_NEEDED			= 0x01,
+	TA_RAS_STATUS__ERROR_INVALID_PARAMETER		= 0x02,
+	TA_RAS_STATUS__ERROR_RAS_NOT_AVAILABLE		= 0x03,
+	TA_RAS_STATUS__ERROR_RAS_DUPLICATE_CMD		= 0x04,
+	TA_RAS_STATUS__ERROR_INJECTION_FAILED		= 0x05
+};
+
+enum ta_ras_block {
+	TA_RAS_BLOCK__UMC = 0,
+	TA_RAS_BLOCK__SDMA,
+	TA_RAS_BLOCK__GFX,
+	TA_RAS_BLOCK__MMHUB,
+	TA_RAS_BLOCK__ATHUB,
+	TA_RAS_BLOCK__PCIE_BIF,
+	TA_RAS_BLOCK__HDP,
+	TA_RAS_BLOCK__XGMI_WAFL,
+	TA_RAS_BLOCK__DF,
+	TA_RAS_BLOCK__SMN,
+	TA_RAS_BLOCK__SEM,
+	TA_RAS_BLOCK__MP0,
+	TA_RAS_BLOCK__MP1,
+	TA_RAS_BLOCK__FUSE = (TA_NUM_BLOCK_MAX - 1),
+};
+
+enum ta_ras_error_type {
+	TA_RAS_ERROR__NONE				= 0,
+	TA_RAS_ERROR__PARITY				= 1,
+	TA_RAS_ERROR__SINGLE_CORRECTABLE		= 2,
+	TA_RAS_ERROR__MULTI_UNCORRECTABLE		= 4,
+	TA_RAS_ERROR__POISON				= 8
+};
+
+struct ta_ras_enable_features_input {
+	enum ta_ras_block       block_id;
+	enum ta_ras_error_type  error_type;
+};
+
+struct ta_ras_disable_features_input {
+	enum ta_ras_block       block_id;
+	enum ta_ras_error_type  error_type;
+};
+
+struct ta_ras_trigger_error_input {
+	enum ta_ras_block		block_id;
+	enum ta_ras_error_type		inject_error_type;
+	uint32_t			sub_block_index;
+	uint64_t			address;
+	uint64_t			value;
+};
+
+union ta_ras_cmd_input {
+	struct ta_ras_enable_features_input	enable_features;
+	struct ta_ras_disable_features_input	disable_features;
+	struct ta_ras_trigger_error_input	trigger_error;
+};
+
+struct ta_ras_shared_memory {
+	uint32_t			cmd_id;
+	uint32_t			resp_id;
+	enum ta_ras_status		ras_status;
+	uint32_t			reserved;
+	union ta_ras_cmd_input		ras_in_message;
+};
+
+#endif // TL_RAS_IF_H_
-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 05/20] drm/amdgpu: add psp ras callback func and macro

[Alex Deucher]

From: xinhui pan <xinhui.pan@amd.com>

Define the driver side interface for ras ta.

Acked-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h
index 49c3942e469c..3e6fcc9cdef0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h
@@ -28,6 +28,7 @@
 #include "amdgpu.h"
 #include "psp_gfx_if.h"
 #include "ta_xgmi_if.h"
+#include "ta_ras_if.h"
 
 #define PSP_FENCE_BUFFER_SIZE	0x1000
 #define PSP_CMD_BUFFER_SIZE	0x1000
@@ -88,6 +89,9 @@ struct psp_funcs
 	int (*xgmi_set_topology_info)(struct psp_context *psp, int number_devices,
 				      struct psp_xgmi_topology_info *topology);
 	bool (*support_vmr_ring)(struct psp_context *psp);
+	int (*ras_trigger_error)(struct psp_context *psp,
+			struct ta_ras_trigger_error_input *info);
+	int (*ras_cure_posion)(struct psp_context *psp, uint64_t *mode_ptr);
 };
 
 struct psp_xgmi_context {
@@ -211,6 +215,13 @@ struct psp_xgmi_topology_info {
 
 #define amdgpu_psp_check_fw_loading_status(adev, i) (adev)->firmware.funcs->check_fw_loading_status((adev), (i))
 
+#define psp_ras_trigger_error(psp, info) \
+	((psp)->funcs->ras_trigger_error ? \
+	(psp)->funcs->ras_trigger_error((psp), (info)) : -EINVAL)
+#define psp_ras_cure_posion(psp, addr) \
+	((psp)->funcs->ras_cure_posion ? \
+	(psp)->funcs->ras_cure_posion(psp, (addr)) : -EINVAL)
+
 extern const struct amd_ip_funcs psp_ip_funcs;
 
 extern const struct amdgpu_ip_block_version psp_v3_1_ip_block;
-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 06/20] drm/amdgpu: add psp ras subsystem infrastructure (v2)

[Alex Deucher]

From: xinhui pan <xinhui.pan@amd.com>

Add ras fw loading, init, terminate.
Add ras cmd submit helper.
Add ras feature enable/disable common function.

v2: squash in unused variable warning fix

Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c | 214 ++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h |  16 ++
 2 files changed, 230 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c
index 3091488cd8cc..c1b862016245 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c
@@ -466,6 +466,206 @@ static int psp_xgmi_initialize(struct psp_context *psp)
 	return ret;
 }
 
+// ras begin
+static void psp_prep_ras_ta_load_cmd_buf(struct psp_gfx_cmd_resp *cmd,
+		uint64_t ras_ta_mc, uint64_t ras_mc_shared,
+		uint32_t ras_ta_size, uint32_t shared_size)
+{
+	cmd->cmd_id = GFX_CMD_ID_LOAD_TA;
+	cmd->cmd.cmd_load_ta.app_phy_addr_lo = lower_32_bits(ras_ta_mc);
+	cmd->cmd.cmd_load_ta.app_phy_addr_hi = upper_32_bits(ras_ta_mc);
+	cmd->cmd.cmd_load_ta.app_len = ras_ta_size;
+
+	cmd->cmd.cmd_load_ta.cmd_buf_phy_addr_lo = lower_32_bits(ras_mc_shared);
+	cmd->cmd.cmd_load_ta.cmd_buf_phy_addr_hi = upper_32_bits(ras_mc_shared);
+	cmd->cmd.cmd_load_ta.cmd_buf_len = shared_size;
+}
+
+static int psp_ras_init_shared_buf(struct psp_context *psp)
+{
+	int ret;
+
+	/*
+	 * Allocate 16k memory aligned to 4k from Frame Buffer (local
+	 * physical) for ras ta <-> Driver
+	 */
+	ret = amdgpu_bo_create_kernel(psp->adev, PSP_RAS_SHARED_MEM_SIZE,
+			PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM,
+			&psp->ras.ras_shared_bo,
+			&psp->ras.ras_shared_mc_addr,
+			&psp->ras.ras_shared_buf);
+
+	return ret;
+}
+
+static int psp_ras_load(struct psp_context *psp)
+{
+	int ret;
+	struct psp_gfx_cmd_resp *cmd;
+
+	/*
+	 * TODO: bypass the loading in sriov for now
+	 */
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	cmd = kzalloc(sizeof(struct psp_gfx_cmd_resp), GFP_KERNEL);
+	if (!cmd)
+		return -ENOMEM;
+
+	memset(psp->fw_pri_buf, 0, PSP_1_MEG);
+	memcpy(psp->fw_pri_buf, psp->ta_ras_start_addr, psp->ta_ras_ucode_size);
+
+	psp_prep_ras_ta_load_cmd_buf(cmd, psp->fw_pri_mc_addr,
+			psp->ras.ras_shared_mc_addr,
+			psp->ta_ras_ucode_size, PSP_RAS_SHARED_MEM_SIZE);
+
+	ret = psp_cmd_submit_buf(psp, NULL, cmd,
+			psp->fence_buf_mc_addr);
+
+	if (!ret) {
+		psp->ras.ras_initialized = 1;
+		psp->ras.session_id = cmd->resp.session_id;
+	}
+
+	kfree(cmd);
+
+	return ret;
+}
+
+static void psp_prep_ras_ta_unload_cmd_buf(struct psp_gfx_cmd_resp *cmd,
+						uint32_t ras_session_id)
+{
+	cmd->cmd_id = GFX_CMD_ID_UNLOAD_TA;
+	cmd->cmd.cmd_unload_ta.session_id = ras_session_id;
+}
+
+static int psp_ras_unload(struct psp_context *psp)
+{
+	int ret;
+	struct psp_gfx_cmd_resp *cmd;
+
+	/*
+	 * TODO: bypass the unloading in sriov for now
+	 */
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	cmd = kzalloc(sizeof(struct psp_gfx_cmd_resp), GFP_KERNEL);
+	if (!cmd)
+		return -ENOMEM;
+
+	psp_prep_ras_ta_unload_cmd_buf(cmd, psp->ras.session_id);
+
+	ret = psp_cmd_submit_buf(psp, NULL, cmd,
+			psp->fence_buf_mc_addr);
+
+	kfree(cmd);
+
+	return ret;
+}
+
+static void psp_prep_ras_ta_invoke_cmd_buf(struct psp_gfx_cmd_resp *cmd,
+		uint32_t ta_cmd_id,
+		uint32_t ras_session_id)
+{
+	cmd->cmd_id = GFX_CMD_ID_INVOKE_CMD;
+	cmd->cmd.cmd_invoke_cmd.session_id = ras_session_id;
+	cmd->cmd.cmd_invoke_cmd.ta_cmd_id = ta_cmd_id;
+	/* Note: cmd_invoke_cmd.buf is not used for now */
+}
+
+int psp_ras_invoke(struct psp_context *psp, uint32_t ta_cmd_id)
+{
+	int ret;
+	struct psp_gfx_cmd_resp *cmd;
+
+	/*
+	 * TODO: bypass the loading in sriov for now
+	 */
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	cmd = kzalloc(sizeof(struct psp_gfx_cmd_resp), GFP_KERNEL);
+	if (!cmd)
+		return -ENOMEM;
+
+	psp_prep_ras_ta_invoke_cmd_buf(cmd, ta_cmd_id,
+			psp->ras.session_id);
+
+	ret = psp_cmd_submit_buf(psp, NULL, cmd,
+			psp->fence_buf_mc_addr);
+
+	kfree(cmd);
+
+	return ret;
+}
+
+int psp_ras_enable_features(struct psp_context *psp,
+		union ta_ras_cmd_input *info, bool enable)
+{
+	struct ta_ras_shared_memory *ras_cmd;
+	int ret;
+
+	if (!psp->ras.ras_initialized)
+		return -EINVAL;
+
+	ras_cmd = (struct ta_ras_shared_memory *)psp->ras.ras_shared_buf;
+	memset(ras_cmd, 0, sizeof(struct ta_ras_shared_memory));
+
+	if (enable)
+		ras_cmd->cmd_id = TA_RAS_COMMAND__ENABLE_FEATURES;
+	else
+		ras_cmd->cmd_id = TA_RAS_COMMAND__DISABLE_FEATURES;
+
+	ras_cmd->ras_in_message = *info;
+
+	ret = psp_ras_invoke(psp, ras_cmd->cmd_id);
+	if (ret)
+		return -EINVAL;
+
+	return ras_cmd->ras_status;
+}
+
+static int psp_ras_terminate(struct psp_context *psp)
+{
+	int ret;
+
+	if (!psp->ras.ras_initialized)
+		return 0;
+
+	ret = psp_ras_unload(psp);
+	if (ret)
+		return ret;
+
+	psp->ras.ras_initialized = 0;
+
+	/* free ras shared memory */
+	amdgpu_bo_free_kernel(&psp->ras.ras_shared_bo,
+			&psp->ras.ras_shared_mc_addr,
+			&psp->ras.ras_shared_buf);
+
+	return 0;
+}
+
+static int psp_ras_initialize(struct psp_context *psp)
+{
+	int ret;
+
+	if (!psp->ras.ras_initialized) {
+		ret = psp_ras_init_shared_buf(psp);
+		if (ret)
+			return ret;
+	}
+
+	ret = psp_ras_load(psp);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+// ras end
+
 static int psp_hw_start(struct psp_context *psp)
 {
 	struct amdgpu_device *adev = psp->adev;
@@ -502,6 +702,12 @@ static int psp_hw_start(struct psp_context *psp)
 			dev_err(psp->adev->dev,
 				"XGMI: Failed to initialize XGMI session\n");
 	}
+
+	ret = psp_ras_initialize(psp);
+	if (ret)
+		dev_err(psp->adev->dev,
+				"RAS: Failed to initialize RAS\n");
+
 	return 0;
 }
 
@@ -753,6 +959,8 @@ static int psp_hw_fini(void *handle)
 	    psp->xgmi_context.initialized == 1)
                 psp_xgmi_terminate(psp);
 
+	psp_ras_terminate(psp);
+
 	psp_ring_destroy(psp, PSP_RING_TYPE__KM);
 
 	amdgpu_bo_free_kernel(&psp->tmr_bo, &psp->tmr_mc_addr, &psp->tmr_buf);
@@ -786,6 +994,12 @@ static int psp_suspend(void *handle)
 		}
 	}
 
+	ret = psp_ras_terminate(psp);
+	if (ret) {
+		DRM_ERROR("Failed to terminate ras ta\n");
+		return ret;
+	}
+
 	ret = psp_ring_stop(psp, PSP_RING_TYPE__KM);
 	if (ret) {
 		DRM_ERROR("PSP ring stop failed\n");
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h
index 3e6fcc9cdef0..9c067e760a5f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h
@@ -34,6 +34,7 @@
 #define PSP_CMD_BUFFER_SIZE	0x1000
 #define PSP_ASD_SHARED_MEM_SIZE 0x4000
 #define PSP_XGMI_SHARED_MEM_SIZE 0x4000
+#define PSP_RAS_SHARED_MEM_SIZE 0x4000
 #define PSP_1_MEG		0x100000
 #define PSP_TMR_SIZE	0x400000
 
@@ -102,6 +103,15 @@ struct psp_xgmi_context {
 	void                            *xgmi_shared_buf;
 };
 
+struct psp_ras_context {
+	/*ras fw*/
+	bool			ras_initialized;
+	uint32_t		session_id;
+	struct amdgpu_bo	*ras_shared_bo;
+	uint64_t		ras_shared_mc_addr;
+	void			*ras_shared_buf;
+};
+
 struct psp_context
 {
 	struct amdgpu_device            *adev;
@@ -162,6 +172,7 @@ struct psp_context
 	uint32_t			ta_ras_ucode_size;
 	uint8_t				*ta_ras_start_addr;
 	struct psp_xgmi_context		xgmi_context;
+	struct psp_ras_context		ras;
 };
 
 struct amdgpu_psp_funcs {
@@ -232,6 +243,11 @@ extern const struct amdgpu_ip_block_version psp_v10_0_ip_block;
 
 int psp_gpu_reset(struct amdgpu_device *adev);
 int psp_xgmi_invoke(struct psp_context *psp, uint32_t ta_cmd_id);
+
+int psp_ras_invoke(struct psp_context *psp, uint32_t ta_cmd_id);
+int psp_ras_enable_features(struct psp_context *psp,
+		union ta_ras_cmd_input *info, bool enable);
+
 extern const struct amdgpu_ip_block_version psp_v11_0_ip_block;
 
 #endif
-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 07/20] drm/amdgpu: add psp v11 ras callback

[Alex Deucher]

From: xinhui pan <xinhui.pan@amd.com>

Add trigger_error and cure_posion.

Acked-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/psp_v11_0.c | 50 ++++++++++++++++++++++++++
 1 file changed, 50 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c b/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c
index 8fb267684787..2b3429d90690 100644
--- a/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c
@@ -722,6 +722,54 @@ static int psp_v11_0_xgmi_get_node_id(struct psp_context *psp, uint64_t *node_id
 	return 0;
 }
 
+static int psp_v11_0_ras_trigger_error(struct psp_context *psp,
+		struct ta_ras_trigger_error_input *info)
+{
+	struct ta_ras_shared_memory *ras_cmd;
+	int ret;
+
+	if (!psp->ras.ras_initialized)
+		return -EINVAL;
+
+	ras_cmd = (struct ta_ras_shared_memory *)psp->ras.ras_shared_buf;
+	memset(ras_cmd, 0, sizeof(struct ta_ras_shared_memory));
+
+	ras_cmd->cmd_id = TA_RAS_COMMAND__TRIGGER_ERROR;
+	ras_cmd->ras_in_message.trigger_error = *info;
+
+	ret = psp_ras_invoke(psp, ras_cmd->cmd_id);
+	if (ret)
+		return -EINVAL;
+
+	return ras_cmd->ras_status;
+}
+
+static int psp_v11_0_ras_cure_posion(struct psp_context *psp, uint64_t *mode_ptr)
+{
+#if 0
+	// not support yet.
+	struct ta_ras_shared_memory *ras_cmd;
+	int ret;
+
+	if (!psp->ras.ras_initialized)
+		return -EINVAL;
+
+	ras_cmd = (struct ta_ras_shared_memory *)psp->ras.ras_shared_buf;
+	memset(ras_cmd, 0, sizeof(struct ta_ras_shared_memory));
+
+	ras_cmd->cmd_id = TA_RAS_COMMAND__CURE_POISON;
+	ras_cmd->ras_in_message.cure_poison.mode_ptr = mode_ptr;
+
+	ret = psp_ras_invoke(psp, ras_cmd->cmd_id);
+	if (ret)
+		return -EINVAL;
+
+	return ras_cmd->ras_status;
+#else
+	return -EINVAL;
+#endif
+}
+
 static const struct psp_funcs psp_v11_0_funcs = {
 	.init_microcode = psp_v11_0_init_microcode,
 	.bootloader_load_sysdrv = psp_v11_0_bootloader_load_sysdrv,
@@ -738,6 +786,8 @@ static const struct psp_funcs psp_v11_0_funcs = {
 	.xgmi_get_hive_id = psp_v11_0_xgmi_get_hive_id,
 	.xgmi_get_node_id = psp_v11_0_xgmi_get_node_id,
 	.support_vmr_ring = psp_v11_0_support_vmr_ring,
+	.ras_trigger_error = psp_v11_0_ras_trigger_error,
+	.ras_cure_posion = psp_v11_0_ras_cure_posion,
 };
 
 void psp_v11_0_set_psp_funcs(struct psp_context *psp)
-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 08/20] drm/amdgpu: add psp cmd submit timeout

[Alex Deucher]

From: xinhui pan <xinhui.pan@amd.com>

Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c
index c1b862016245..7e3e1d588d74 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c
@@ -120,6 +120,7 @@ psp_cmd_submit_buf(struct psp_context *psp,
 {
 	int ret;
 	int index;
+	int timeout = 2000;
 
 	memset(psp->cmd_buf_mem, 0, PSP_CMD_BUFFER_SIZE);
 
@@ -133,8 +134,11 @@ psp_cmd_submit_buf(struct psp_context *psp,
 		return ret;
 	}
 
-	while (*((unsigned int *)psp->fence_buf) != index)
+	while (*((unsigned int *)psp->fence_buf) != index) {
+		if (--timeout == 0)
+			return -EINVAL;
 		msleep(1);
+	}
 
 	/* In some cases, psp response status is not 0 even there is no
 	 * problem while the command is submitted. Some version of PSP FW
-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 09/20] drm/amdgpu: add amdgpu_ras.c to support ras (v2)

[Alex Deucher]

From: xinhui pan <xinhui.pan@amd.com>

add obj management.
add feature control.
add debugfs infrastructure.
add sysfs infrastructure.
add IH infrastructure.
add recovery infrastructure.

It is a framework. Other IPs need call amdgpu_ras_xxx function instead of
psp_ras_xxx functions.

v2: squash in warning fixes

Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/Makefile        |    2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |    9 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h    |    1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c    | 1247 ++++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h    |  217 ++++
 5 files changed, 1475 insertions(+), 1 deletion(-)
 create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
 create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h

diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile b/drivers/gpu/drm/amd/amdgpu/Makefile
index 851001ced5e8..6039944abb71 100644
--- a/drivers/gpu/drm/amd/amdgpu/Makefile
+++ b/drivers/gpu/drm/amd/amdgpu/Makefile
@@ -53,7 +53,7 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \
 	amdgpu_ucode.o amdgpu_bo_list.o amdgpu_ctx.o amdgpu_sync.o \
 	amdgpu_gtt_mgr.o amdgpu_vram_mgr.o amdgpu_virt.o amdgpu_atomfirmware.o \
 	amdgpu_vf_error.o amdgpu_sched.o amdgpu_debugfs.o amdgpu_ids.o \
-	amdgpu_gmc.o amdgpu_xgmi.o amdgpu_csa.o
+	amdgpu_gmc.o amdgpu_xgmi.o amdgpu_csa.o amdgpu_ras.o
 
 # add asic specific block
 amdgpu-$(CONFIG_DRM_AMDGPU_CIK)+= cik.o cik_ih.o kv_smc.o kv_dpm.o \
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 05cd5c460a1d..d6f130bdb985 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -60,6 +60,7 @@
 #include "amdgpu_pm.h"
 
 #include "amdgpu_xgmi.h"
+#include "amdgpu_ras.h"
 
 MODULE_FIRMWARE("amdgpu/vega10_gpu_info.bin");
 MODULE_FIRMWARE("amdgpu/vega12_gpu_info.bin");
@@ -1638,6 +1639,10 @@ static int amdgpu_device_ip_init(struct amdgpu_device *adev)
 {
 	int i, r;
 
+	r = amdgpu_ras_init(adev);
+	if (r)
+		return r;
+
 	for (i = 0; i < adev->num_ip_blocks; i++) {
 		if (!adev->ip_blocks[i].status.valid)
 			continue;
@@ -1869,6 +1874,8 @@ static int amdgpu_device_ip_fini(struct amdgpu_device *adev)
 {
 	int i, r;
 
+	amdgpu_ras_pre_fini(adev);
+
 	if (adev->gmc.xgmi.num_physical_nodes > 1)
 		amdgpu_xgmi_remove_device(adev);
 
@@ -1937,6 +1944,8 @@ static int amdgpu_device_ip_fini(struct amdgpu_device *adev)
 		adev->ip_blocks[i].status.late_initialized = false;
 	}
 
+	amdgpu_ras_fini(adev);
+
 	if (amdgpu_sriov_vf(adev))
 		if (amdgpu_virt_release_full_gpu(adev, false))
 			DRM_ERROR("failed to release exclusive mode on fini\n");
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h
index 9c067e760a5f..cde113f07c96 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h
@@ -110,6 +110,7 @@ struct psp_ras_context {
 	struct amdgpu_bo	*ras_shared_bo;
 	uint64_t		ras_shared_mc_addr;
 	void			*ras_shared_buf;
+	struct amdgpu_ras	*ras;
 };
 
 struct psp_context
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
new file mode 100644
index 000000000000..d7c7090fade9
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -0,0 +1,1247 @@
+/*
+ * Copyright 2018 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ *
+ */
+#include <linux/debugfs.h>
+#include <linux/list.h>
+#include <linux/module.h>
+#include "amdgpu.h"
+#include "amdgpu_ras.h"
+
+struct ras_ih_data {
+	/* interrupt bottom half */
+	struct work_struct ih_work;
+	int inuse;
+	/* IP callback */
+	ras_ih_cb cb;
+	/* full of entries */
+	unsigned char *ring;
+	unsigned int ring_size;
+	unsigned int element_size;
+	unsigned int aligned_element_size;
+	unsigned int rptr;
+	unsigned int wptr;
+};
+
+struct ras_fs_data {
+	char sysfs_name[32];
+	char debugfs_name[32];
+};
+
+struct ras_err_data {
+	unsigned long ue_count;
+	unsigned long ce_count;
+};
+
+struct ras_err_handler_data {
+	/* point to bad pages array */
+	struct {
+		unsigned long bp;
+		struct amdgpu_bo *bo;
+	} *bps;
+	/* the count of entries */
+	int count;
+	/* the space can place new entries */
+	int space_left;
+	/* last reserved entry's index + 1 */
+	int last_reserved;
+};
+
+struct ras_manager {
+	struct ras_common_if head;
+	/* reference count */
+	int use;
+	/* ras block link */
+	struct list_head node;
+	/* the device */
+	struct amdgpu_device *adev;
+	/* debugfs */
+	struct dentry *ent;
+	/* sysfs */
+	struct device_attribute sysfs_attr;
+	int attr_inuse;
+
+	/* fs node name */
+	struct ras_fs_data fs_data;
+
+	/* IH data */
+	struct ras_ih_data ih_data;
+
+	struct ras_err_data err_data;
+};
+
+const char *ras_error_string[] = {
+	"none",
+	"parity",
+	"single_correctable",
+	"multi_uncorrectable",
+	"poison",
+};
+
+const char *ras_block_string[] = {
+	"umc",
+	"sdma",
+	"gfx",
+	"mmhub",
+	"athub",
+	"pcie_bif",
+	"hdp",
+	"xgmi_wafl",
+	"df",
+	"smn",
+	"sem",
+	"mp0",
+	"mp1",
+	"fuse",
+};
+
+#define ras_err_str(i) (ras_error_string[ffs(i)])
+#define ras_block_str(i) (ras_block_string[i])
+
+static void amdgpu_ras_self_test(struct amdgpu_device *adev)
+{
+	/* TODO */
+}
+
+static ssize_t amdgpu_ras_debugfs_read(struct file *f, char __user *buf,
+					size_t size, loff_t *pos)
+{
+	struct ras_manager *obj = (struct ras_manager *)file_inode(f)->i_private;
+	struct ras_query_if info = {
+		.head = obj->head,
+	};
+	ssize_t s;
+	char val[128];
+
+	if (amdgpu_ras_error_query(obj->adev, &info))
+		return -EINVAL;
+
+	s = snprintf(val, sizeof(val), "%s: %lu\n%s: %lu\n",
+			"ue", info.ue_count,
+			"ce", info.ce_count);
+	if (*pos >= s)
+		return 0;
+
+	s -= *pos;
+	s = min_t(u64, s, size);
+
+
+	if (copy_to_user(buf, &val[*pos], s))
+		return -EINVAL;
+
+	*pos += s;
+
+	return s;
+}
+
+static ssize_t amdgpu_ras_debugfs_write(struct file *f, const char __user *buf,
+		size_t size, loff_t *pos)
+{
+	struct ras_manager *obj = (struct ras_manager *)file_inode(f)->i_private;
+	struct ras_inject_if info = {
+		.head = obj->head,
+	};
+	ssize_t s = min_t(u64, 64, size);
+	char val[64];
+	char *str = val;
+	memset(val, 0, sizeof(val));
+
+	if (*pos)
+		return -EINVAL;
+
+	if (copy_from_user(str, buf, s))
+		return -EINVAL;
+
+	/* only care ue/ce for now. */
+	if (memcmp(str, "ue", 2) == 0) {
+		info.head.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
+		str += 2;
+	} else if (memcmp(str, "ce", 2) == 0) {
+		info.head.type = AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE;
+		str += 2;
+	}
+
+	if (sscanf(str, "0x%llx 0x%llx", &info.address, &info.value) != 2) {
+		if (sscanf(str, "%llu %llu", &info.address, &info.value) != 2)
+			return -EINVAL;
+	}
+
+	*pos = s;
+
+	if (amdgpu_ras_error_inject(obj->adev, &info))
+		return -EINVAL;
+
+	return size;
+}
+
+static const struct file_operations amdgpu_ras_debugfs_ops = {
+	.owner = THIS_MODULE,
+	.read = amdgpu_ras_debugfs_read,
+	.write = amdgpu_ras_debugfs_write,
+	.llseek = default_llseek
+};
+
+static ssize_t amdgpu_ras_sysfs_read(struct device *dev,
+		struct device_attribute *attr, char *buf)
+{
+	struct ras_manager *obj = container_of(attr, struct ras_manager, sysfs_attr);
+	struct ras_query_if info = {
+		.head = obj->head,
+	};
+
+	if (amdgpu_ras_error_query(obj->adev, &info))
+		return -EINVAL;
+
+	return snprintf(buf, PAGE_SIZE, "%s: %lu\n%s: %lu\n",
+			"ue", info.ue_count,
+			"ce", info.ce_count);
+}
+
+/* obj begin */
+
+#define get_obj(obj) do { (obj)->use++; } while (0)
+#define alive_obj(obj) ((obj)->use)
+
+static inline void put_obj(struct ras_manager *obj)
+{
+	if (obj && --obj->use == 0)
+		list_del(&obj->node);
+	if (obj && obj->use < 0) {
+		 DRM_ERROR("RAS ERROR: Unbalance obj(%s) use\n", obj->head.name);
+	}
+}
+
+/* make one obj and return it. */
+static struct ras_manager *amdgpu_ras_create_obj(struct amdgpu_device *adev,
+		struct ras_common_if *head)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct ras_manager *obj;
+
+	if (!con)
+		return NULL;
+
+	if (head->block >= AMDGPU_RAS_BLOCK_COUNT)
+		return NULL;
+
+	obj = &con->objs[head->block];
+	/* already exist. return obj? */
+	if (alive_obj(obj))
+		return NULL;
+
+	obj->head = *head;
+	obj->adev = adev;
+	list_add(&obj->node, &con->head);
+	get_obj(obj);
+
+	return obj;
+}
+
+/* return an obj equal to head, or the first when head is NULL */
+static struct ras_manager *amdgpu_ras_find_obj(struct amdgpu_device *adev,
+		struct ras_common_if *head)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct ras_manager *obj;
+	int i;
+
+	if (!con)
+		return NULL;
+
+	if (head) {
+		if (head->block >= AMDGPU_RAS_BLOCK_COUNT)
+			return NULL;
+
+		obj = &con->objs[head->block];
+
+		if (alive_obj(obj)) {
+			WARN_ON(head->block != obj->head.block);
+			return obj;
+		}
+	} else {
+		for (i = 0; i < AMDGPU_RAS_BLOCK_COUNT; i++) {
+			obj = &con->objs[i];
+			if (alive_obj(obj)) {
+				WARN_ON(i != obj->head.block);
+				return obj;
+			}
+		}
+	}
+
+	return NULL;
+}
+/* obj end */
+
+/* feature ctl begin */
+static int amdgpu_ras_is_feature_allowed(struct amdgpu_device *adev,
+		struct ras_common_if *head)
+{
+	return amdgpu_ras_enable && (amdgpu_ras_mask & BIT(head->block));
+}
+
+static int amdgpu_ras_is_feature_enabled(struct amdgpu_device *adev,
+		struct ras_common_if *head)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+
+	return con->features & BIT(head->block);
+}
+
+/*
+ * if obj is not created, then create one.
+ * set feature enable flag.
+ */
+static int __amdgpu_ras_feature_enable(struct amdgpu_device *adev,
+		struct ras_common_if *head, int enable)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct ras_manager *obj = amdgpu_ras_find_obj(adev, head);
+
+	if (!amdgpu_ras_is_feature_allowed(adev, head))
+		return 0;
+	if (!(!!enable ^ !!amdgpu_ras_is_feature_enabled(adev, head)))
+		return 0;
+
+	if (enable) {
+		if (!obj) {
+			obj = amdgpu_ras_create_obj(adev, head);
+			if (!obj)
+				return -EINVAL;
+		} else {
+			/* In case we create obj somewhere else */
+			get_obj(obj);
+		}
+		con->features |= BIT(head->block);
+	} else {
+		if (obj && amdgpu_ras_is_feature_enabled(adev, head)) {
+			con->features &= ~BIT(head->block);
+			put_obj(obj);
+		}
+	}
+
+	return 0;
+}
+
+/* wrapper of psp_ras_enable_features */
+int amdgpu_ras_feature_enable(struct amdgpu_device *adev,
+		struct ras_common_if *head, bool enable)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	union ta_ras_cmd_input info;
+	int ret;
+
+	if (!con)
+		return -EINVAL;
+
+	if (!enable) {
+		info.disable_features = (struct ta_ras_disable_features_input) {
+			.block_id =  head->block,
+			.error_type = head->type,
+		};
+	} else {
+		info.enable_features = (struct ta_ras_enable_features_input) {
+			.block_id =  head->block,
+			.error_type = head->type,
+		};
+	}
+
+	/* Do not enable if it is not allowed. */
+	WARN_ON(enable && !amdgpu_ras_is_feature_allowed(adev, head));
+	/* Are we alerady in that state we are going to set? */
+	if (!(!!enable ^ !!amdgpu_ras_is_feature_enabled(adev, head)))
+		return 0;
+
+	ret = psp_ras_enable_features(&adev->psp, &info, enable);
+	if (ret) {
+		DRM_ERROR("RAS ERROR: %s %s feature failed ret %d\n",
+				enable ? "enable":"disable",
+				ras_block_str(head->block),
+				ret);
+		return -EINVAL;
+	}
+
+	/* setup the obj */
+	__amdgpu_ras_feature_enable(adev, head, enable);
+
+	return 0;
+}
+
+static int amdgpu_ras_disable_all_features(struct amdgpu_device *adev,
+		bool bypass)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct ras_manager *obj, *tmp;
+
+	list_for_each_entry_safe(obj, tmp, &con->head, node) {
+		/* bypass psp.
+		 * aka just release the obj and corresponding flags
+		 */
+		if (bypass) {
+			if (__amdgpu_ras_feature_enable(adev, &obj->head, 0))
+				break;
+		} else {
+			if (amdgpu_ras_feature_enable(adev, &obj->head, 0))
+				break;
+		}
+	};
+
+	return con->features;
+}
+
+static int amdgpu_ras_enable_all_features(struct amdgpu_device *adev,
+		bool bypass)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	int ras_block_count = AMDGPU_RAS_BLOCK_COUNT;
+	int i;
+
+	for (i = 0; i < ras_block_count; i++) {
+		struct ras_common_if head = {
+			.block = i,
+			.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE,
+			.sub_block_index = 0,
+		};
+		strcpy(head.name, ras_block_str(i));
+		if (bypass) {
+			/*
+			 * bypass psp. vbios enable ras for us.
+			 * so just create the obj
+			 */
+			if (__amdgpu_ras_feature_enable(adev, &head, 1))
+				break;
+		} else {
+			if (amdgpu_ras_feature_enable(adev, &head, 1))
+				break;
+		}
+	};
+
+	return con->features;
+}
+/* feature ctl end */
+
+/* query/inject/cure begin */
+int amdgpu_ras_error_query(struct amdgpu_device *adev,
+		struct ras_query_if *info)
+{
+	struct ras_manager *obj = amdgpu_ras_find_obj(adev, &info->head);
+
+	if (!obj)
+		return -EINVAL;
+	/* TODO might read the register to read the count */
+
+	info->ue_count = obj->err_data.ue_count;
+	info->ce_count = obj->err_data.ce_count;
+
+	return 0;
+}
+
+/* wrapper of psp_ras_trigger_error */
+int amdgpu_ras_error_inject(struct amdgpu_device *adev,
+		struct ras_inject_if *info)
+{
+	struct ras_manager *obj = amdgpu_ras_find_obj(adev, &info->head);
+	struct ta_ras_trigger_error_input block_info = {
+		.block_id = info->head.block,
+		.inject_error_type = info->head.type,
+		.sub_block_index = info->head.sub_block_index,
+		.address = info->address,
+		.value = info->value,
+	};
+	int ret = 0;
+
+	if (!obj)
+		return -EINVAL;
+
+	ret = psp_ras_trigger_error(&adev->psp, &block_info);
+	if (ret)
+		DRM_ERROR("RAS ERROR: inject %s error failed ret %d\n",
+				ras_block_str(info->head.block),
+				ret);
+
+	return ret;
+}
+
+int amdgpu_ras_error_cure(struct amdgpu_device *adev,
+		struct ras_cure_if *info)
+{
+	/* psp fw has no cure interface for now. */
+	return 0;
+}
+
+/* get the total error counts on all IPs */
+int amdgpu_ras_query_error_count(struct amdgpu_device *adev,
+		bool is_ce)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct ras_manager *obj;
+	struct ras_err_data data = {0, 0};
+
+	if (!con)
+		return -EINVAL;
+
+	list_for_each_entry(obj, &con->head, node) {
+		struct ras_query_if info = {
+			.head = obj->head,
+		};
+
+		if (amdgpu_ras_error_query(adev, &info))
+			return -EINVAL;
+
+		data.ce_count += info.ce_count;
+		data.ue_count += info.ue_count;
+	}
+
+	return is_ce ? data.ce_count : data.ue_count;
+}
+/* query/inject/cure end */
+
+
+/* sysfs begin */
+
+static ssize_t amdgpu_ras_sysfs_features_read(struct device *dev,
+		struct device_attribute *attr, char *buf)
+{
+	struct amdgpu_ras *con =
+		container_of(attr, struct amdgpu_ras, features_attr);
+	struct drm_device *ddev = dev_get_drvdata(dev);
+	struct amdgpu_device *adev = ddev->dev_private;
+	struct ras_common_if head;
+	int ras_block_count = AMDGPU_RAS_BLOCK_COUNT;
+	int i;
+	ssize_t s;
+	struct ras_manager *obj;
+
+	s = scnprintf(buf, PAGE_SIZE, "feature mask: 0x%x\n", con->features);
+
+	for (i = 0; i < ras_block_count; i++) {
+		head.block = i;
+
+		if (amdgpu_ras_is_feature_enabled(adev, &head)) {
+			obj = amdgpu_ras_find_obj(adev, &head);
+			s += scnprintf(&buf[s], PAGE_SIZE - s,
+					"%s: %s\n",
+					ras_block_str(i),
+					ras_err_str(obj->head.type));
+		} else
+			s += scnprintf(&buf[s], PAGE_SIZE - s,
+					"%s: disabled\n",
+					ras_block_str(i));
+	}
+
+	return s;
+}
+
+static int amdgpu_ras_sysfs_create_feature_node(struct amdgpu_device *adev)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct attribute *attrs[] = {
+		&con->features_attr.attr,
+		NULL
+	};
+	struct attribute_group group = {
+		.name = "ras",
+		.attrs = attrs,
+	};
+
+	con->features_attr = (struct device_attribute) {
+		.attr = {
+			.name = "features",
+			.mode = S_IRUGO,
+		},
+			.show = amdgpu_ras_sysfs_features_read,
+	};
+
+	return sysfs_create_group(&adev->dev->kobj, &group);
+}
+
+static int amdgpu_ras_sysfs_remove_feature_node(struct amdgpu_device *adev)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct attribute *attrs[] = {
+		&con->features_attr.attr,
+		NULL
+	};
+	struct attribute_group group = {
+		.name = "ras",
+		.attrs = attrs,
+	};
+
+	sysfs_remove_group(&adev->dev->kobj, &group);
+
+	return 0;
+}
+
+int amdgpu_ras_sysfs_create(struct amdgpu_device *adev,
+		struct ras_fs_if *head)
+{
+	struct ras_manager *obj = amdgpu_ras_find_obj(adev, &head->head);
+
+	if (!obj || obj->attr_inuse)
+		return -EINVAL;
+
+	get_obj(obj);
+
+	memcpy(obj->fs_data.sysfs_name,
+			head->sysfs_name,
+			sizeof(obj->fs_data.sysfs_name));
+
+	obj->sysfs_attr = (struct device_attribute){
+		.attr = {
+			.name = obj->fs_data.sysfs_name,
+			.mode = S_IRUGO,
+		},
+			.show = amdgpu_ras_sysfs_read,
+	};
+
+	if (sysfs_add_file_to_group(&adev->dev->kobj,
+				&obj->sysfs_attr.attr,
+				"ras")) {
+		put_obj(obj);
+		return -EINVAL;
+	}
+
+	obj->attr_inuse = 1;
+
+	return 0;
+}
+
+int amdgpu_ras_sysfs_remove(struct amdgpu_device *adev,
+		struct ras_common_if *head)
+{
+	struct ras_manager *obj = amdgpu_ras_find_obj(adev, head);
+
+	if (!obj || !obj->attr_inuse)
+		return -EINVAL;
+
+	sysfs_remove_file_from_group(&adev->dev->kobj,
+				&obj->sysfs_attr.attr,
+				"ras");
+	obj->attr_inuse = 0;
+	put_obj(obj);
+
+	return 0;
+}
+
+static int amdgpu_ras_sysfs_remove_all(struct amdgpu_device *adev)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct ras_manager *obj, *tmp;
+
+	list_for_each_entry_safe(obj, tmp, &con->head, node) {
+		amdgpu_ras_sysfs_remove(adev, &obj->head);
+	}
+
+	amdgpu_ras_sysfs_remove_feature_node(adev);
+
+	return 0;
+}
+/* sysfs end */
+
+/* debugfs begin */
+int amdgpu_ras_debugfs_create(struct amdgpu_device *adev,
+		struct ras_fs_if *head)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct ras_manager *obj = amdgpu_ras_find_obj(adev, &head->head);
+	struct dentry *ent;
+
+	if (!obj || obj->ent)
+		return -EINVAL;
+
+	get_obj(obj);
+
+	memcpy(obj->fs_data.debugfs_name,
+			head->debugfs_name,
+			sizeof(obj->fs_data.debugfs_name));
+
+	ent = debugfs_create_file(obj->fs_data.debugfs_name,
+			S_IWUGO | S_IRUGO, con->dir,
+			obj, &amdgpu_ras_debugfs_ops);
+
+	if (IS_ERR(ent))
+		return -EINVAL;
+
+	obj->ent = ent;
+
+	return 0;
+}
+
+int amdgpu_ras_debugfs_remove(struct amdgpu_device *adev,
+		struct ras_common_if *head)
+{
+	struct ras_manager *obj = amdgpu_ras_find_obj(adev, head);
+
+	if (!obj || !obj->ent)
+		return 0;
+
+	debugfs_remove(obj->ent);
+	obj->ent = NULL;
+	put_obj(obj);
+
+	return 0;
+}
+
+static int amdgpu_ras_debugfs_remove_all(struct amdgpu_device *adev)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct ras_manager *obj, *tmp;
+
+	list_for_each_entry_safe(obj, tmp, &con->head, node) {
+		amdgpu_ras_debugfs_remove(adev, &obj->head);
+	}
+
+	debugfs_remove(con->dir);
+	con->dir = NULL;
+
+	return 0;
+}
+/* debugfs end */
+
+/* ras fs */
+
+static int amdgpu_ras_fs_init(struct amdgpu_device *adev)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct drm_minor *minor = adev->ddev->primary;
+	struct dentry *root = minor->debugfs_root, *dir;
+
+	dir = debugfs_create_dir("ras", root);
+	if (IS_ERR(dir))
+		return -EINVAL;
+
+	con->dir = dir;
+
+	amdgpu_ras_sysfs_create_feature_node(adev);
+
+	return 0;
+}
+
+static int amdgpu_ras_fs_fini(struct amdgpu_device *adev)
+{
+	amdgpu_ras_debugfs_remove_all(adev);
+	amdgpu_ras_sysfs_remove_all(adev);
+	return 0;
+}
+/* ras fs end */
+
+/* ih begin */
+static void amdgpu_ras_interrupt_handler(struct ras_manager *obj)
+{
+	struct ras_ih_data *data = &obj->ih_data;
+	struct amdgpu_iv_entry entry;
+	int ret;
+
+	while (data->rptr != data->wptr) {
+		rmb();
+		memcpy(&entry, &data->ring[data->rptr],
+				data->element_size);
+
+		wmb();
+		data->rptr = (data->aligned_element_size +
+				data->rptr) % data->ring_size;
+
+		/* Let IP handle its data, maybe we need get the output
+		 * from the callback to udpate the error type/count, etc
+		 */
+		if (data->cb) {
+			ret = data->cb(obj->adev, &entry);
+			/* ue will trigger an interrupt, and in that case
+			 * we need do a reset to recovery the whole system.
+			 * But leave IP do that recovery, here we just dispatch
+			 * the error.
+			 */
+			if (ret == AMDGPU_RAS_UE) {
+				obj->err_data.ue_count++;
+			}
+			/* Might need get ce count by register, but not all IP
+			 * saves ce count, some IP just use one bit or two bits
+			 * to indicate ce happened.
+			 */
+		}
+	}
+}
+
+static void amdgpu_ras_interrupt_process_handler(struct work_struct *work)
+{
+	struct ras_ih_data *data =
+		container_of(work, struct ras_ih_data, ih_work);
+	struct ras_manager *obj =
+		container_of(data, struct ras_manager, ih_data);
+
+	amdgpu_ras_interrupt_handler(obj);
+}
+
+int amdgpu_ras_interrupt_dispatch(struct amdgpu_device *adev,
+		struct ras_dispatch_if *info)
+{
+	struct ras_manager *obj = amdgpu_ras_find_obj(adev, &info->head);
+	struct ras_ih_data *data = &obj->ih_data;
+
+	if (!obj)
+		return -EINVAL;
+
+	if (data->inuse == 0)
+		return 0;
+
+	/* Might be overflow... */
+	memcpy(&data->ring[data->wptr], info->entry,
+			data->element_size);
+
+	wmb();
+	data->wptr = (data->aligned_element_size +
+			data->wptr) % data->ring_size;
+
+	schedule_work(&data->ih_work);
+
+	return 0;
+}
+
+int amdgpu_ras_interrupt_remove_handler(struct amdgpu_device *adev,
+		struct ras_ih_if *info)
+{
+	struct ras_manager *obj = amdgpu_ras_find_obj(adev, &info->head);
+	struct ras_ih_data *data;
+
+	if (!obj)
+		return -EINVAL;
+
+	data = &obj->ih_data;
+	if (data->inuse == 0)
+		return 0;
+
+	cancel_work_sync(&data->ih_work);
+
+	kfree(data->ring);
+	memset(data, 0, sizeof(*data));
+	put_obj(obj);
+
+	return 0;
+}
+
+int amdgpu_ras_interrupt_add_handler(struct amdgpu_device *adev,
+		struct ras_ih_if *info)
+{
+	struct ras_manager *obj = amdgpu_ras_find_obj(adev, &info->head);
+	struct ras_ih_data *data;
+
+	if (!obj) {
+		/* in case we registe the IH before enable ras feature */
+		obj = amdgpu_ras_create_obj(adev, &info->head);
+		if (!obj)
+			return -EINVAL;
+	} else
+		get_obj(obj);
+
+	data = &obj->ih_data;
+	/* add the callback.etc */
+	*data = (struct ras_ih_data) {
+		.inuse = 0,
+		.cb = info->cb,
+		.element_size = sizeof(struct amdgpu_iv_entry),
+		.rptr = 0,
+		.wptr = 0,
+	};
+
+	INIT_WORK(&data->ih_work, amdgpu_ras_interrupt_process_handler);
+
+	data->aligned_element_size = ALIGN(data->element_size, 8);
+	/* the ring can store 64 iv entries. */
+	data->ring_size = 64 * data->aligned_element_size;
+	data->ring = kmalloc(data->ring_size, GFP_KERNEL);
+	if (!data->ring) {
+		put_obj(obj);
+		return -ENOMEM;
+	}
+
+	/* IH is ready */
+	data->inuse = 1;
+
+	return 0;
+}
+
+static int amdgpu_ras_interrupt_remove_all(struct amdgpu_device *adev)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct ras_manager *obj, *tmp;
+
+	list_for_each_entry_safe(obj, tmp, &con->head, node) {
+		struct ras_ih_if info = {
+			.head = obj->head,
+		};
+		amdgpu_ras_interrupt_remove_handler(adev, &info);
+	}
+
+	return 0;
+}
+/* ih end */
+
+/* recovery begin */
+static void amdgpu_ras_do_recovery(struct work_struct *work)
+{
+	struct amdgpu_ras *ras =
+		container_of(work, struct amdgpu_ras, recovery_work);
+
+	amdgpu_device_gpu_recover(ras->adev, 0);
+	atomic_set(&ras->in_recovery, 0);
+}
+
+static int amdgpu_ras_release_vram(struct amdgpu_device *adev,
+		struct amdgpu_bo **bo_ptr)
+{
+	/* no need to free it actually. */
+	amdgpu_bo_free_kernel(bo_ptr, NULL, NULL);
+	return 0;
+}
+
+/* reserve vram with size@offset */
+static int amdgpu_ras_reserve_vram(struct amdgpu_device *adev,
+		uint64_t offset, uint64_t size,
+		struct amdgpu_bo **bo_ptr)
+{
+	struct ttm_operation_ctx ctx = { false, false };
+	struct amdgpu_bo_param bp;
+	int r = 0;
+	int i;
+	struct amdgpu_bo *bo;
+
+	if (bo_ptr)
+		*bo_ptr = NULL;
+	memset(&bp, 0, sizeof(bp));
+	bp.size = size;
+	bp.byte_align = PAGE_SIZE;
+	bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
+	bp.flags = AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS |
+		AMDGPU_GEM_CREATE_NO_CPU_ACCESS;
+	bp.type = ttm_bo_type_kernel;
+	bp.resv = NULL;
+
+	r = amdgpu_bo_create(adev, &bp, &bo);
+	if (r)
+		return -EINVAL;
+
+	r = amdgpu_bo_reserve(bo, false);
+	if (r)
+		goto error_reserve;
+
+	offset = ALIGN(offset, PAGE_SIZE);
+	for (i = 0; i < bo->placement.num_placement; ++i) {
+		bo->placements[i].fpfn = offset >> PAGE_SHIFT;
+		bo->placements[i].lpfn = (offset + size) >> PAGE_SHIFT;
+	}
+
+	ttm_bo_mem_put(&bo->tbo, &bo->tbo.mem);
+	r = ttm_bo_mem_space(&bo->tbo, &bo->placement, &bo->tbo.mem, &ctx);
+	if (r)
+		goto error_pin;
+
+	r = amdgpu_bo_pin_restricted(bo,
+			AMDGPU_GEM_DOMAIN_VRAM,
+			offset,
+			offset + size);
+	if (r)
+		goto error_pin;
+
+	if (bo_ptr)
+		*bo_ptr = bo;
+
+	amdgpu_bo_unreserve(bo);
+	return r;
+
+error_pin:
+	amdgpu_bo_unreserve(bo);
+error_reserve:
+	amdgpu_bo_unref(&bo);
+	return r;
+}
+
+/* alloc/realloc bps array */
+static int amdgpu_ras_realloc_eh_data_space(struct amdgpu_device *adev,
+		struct ras_err_handler_data *data, int pages)
+{
+	unsigned int old_space = data->count + data->space_left;
+	unsigned int new_space = old_space + pages;
+	unsigned int align_space = ALIGN(new_space, 1024);
+	void *tmp = kmalloc(align_space * sizeof(*data->bps), GFP_KERNEL);
+
+	if (!tmp)
+		return -ENOMEM;
+
+	if (data->bps) {
+		memcpy(tmp, data->bps,
+				data->count * sizeof(*data->bps));
+		kfree(data->bps);
+	}
+
+	data->bps = tmp;
+	data->space_left += align_space - old_space;
+	return 0;
+}
+
+/* it deal with vram only. */
+int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev,
+		unsigned long *bps, int pages)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct ras_err_handler_data *data = con->eh_data;
+	int i = pages;
+	int ret = 0;
+
+	if (!con || !data || !bps || pages <= 0)
+		return 0;
+
+	mutex_lock(&con->recovery_lock);
+	if (!data)
+		goto out;
+
+	if (data->space_left <= pages)
+		if (amdgpu_ras_realloc_eh_data_space(adev, data, pages)) {
+			ret = -ENOMEM;
+			goto out;
+		}
+
+	while (i--)
+		data->bps[data->count++].bp = bps[i];
+
+	data->space_left -= pages;
+out:
+	mutex_unlock(&con->recovery_lock);
+
+	return ret;
+}
+
+/* called in gpu recovery/init */
+int amdgpu_ras_reserve_bad_pages(struct amdgpu_device *adev)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct ras_err_handler_data *data = con->eh_data;
+	uint64_t bp;
+	struct amdgpu_bo *bo;
+	int i;
+
+	if (!con || !data)
+		return 0;
+
+	mutex_lock(&con->recovery_lock);
+	/* reserve vram at driver post stage. */
+	for (i = data->last_reserved; i < data->count; i++) {
+		bp = data->bps[i].bp;
+
+		if (amdgpu_ras_reserve_vram(adev, bp << PAGE_SHIFT,
+					PAGE_SIZE, &bo))
+			DRM_ERROR("RAS ERROR: reserve vram %llx fail\n", bp);
+
+		data->bps[i].bo = bo;
+		data->last_reserved = i + 1;
+	}
+	mutex_unlock(&con->recovery_lock);
+	return 0;
+}
+
+/* called when driver unload */
+static int amdgpu_ras_release_bad_pages(struct amdgpu_device *adev)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct ras_err_handler_data *data = con->eh_data;
+	struct amdgpu_bo *bo;
+	int i;
+
+	if (!con || !data)
+		return 0;
+
+	mutex_lock(&con->recovery_lock);
+	for (i = data->last_reserved - 1; i >= 0; i--) {
+		bo = data->bps[i].bo;
+
+		amdgpu_ras_release_vram(adev, &bo);
+
+		data->bps[i].bo = bo;
+		data->last_reserved = i;
+	}
+	mutex_unlock(&con->recovery_lock);
+	return 0;
+}
+
+static int amdgpu_ras_save_bad_pages(struct amdgpu_device *adev)
+{
+	/* TODO
+	 * write the array to eeprom when SMU disabled.
+	 */
+	return 0;
+}
+
+static int amdgpu_ras_load_bad_pages(struct amdgpu_device *adev)
+{
+	/* TODO
+	 * read the array to eeprom when SMU disabled.
+	 */
+	return 0;
+}
+
+static int amdgpu_ras_recovery_init(struct amdgpu_device *adev)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct ras_err_handler_data **data = &con->eh_data;
+
+	*data = kmalloc(sizeof(**data),
+			GFP_KERNEL|__GFP_ZERO);
+	if (!*data)
+		return -ENOMEM;
+
+	mutex_init(&con->recovery_lock);
+	INIT_WORK(&con->recovery_work, amdgpu_ras_do_recovery);
+	atomic_set(&con->in_recovery, 0);
+	con->adev = adev;
+
+	amdgpu_ras_load_bad_pages(adev);
+	amdgpu_ras_reserve_bad_pages(adev);
+
+	return 0;
+}
+
+static int amdgpu_ras_recovery_fini(struct amdgpu_device *adev)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct ras_err_handler_data *data = con->eh_data;
+
+	cancel_work_sync(&con->recovery_work);
+	amdgpu_ras_save_bad_pages(adev);
+	amdgpu_ras_release_bad_pages(adev);
+
+	mutex_lock(&con->recovery_lock);
+	con->eh_data = NULL;
+	kfree(data->bps);
+	kfree(data);
+	mutex_unlock(&con->recovery_lock);
+
+	return 0;
+}
+/* recovery end */
+
+struct ras_DID_capability {
+	u16 did;
+	u8 rid;
+	u32 capability;
+};
+
+static const struct ras_DID_capability supported_DID_array[] = {
+	{0x66a0, 0x00, AMDGPU_RAS_BLOCK_MASK},
+	{0x66a0, 0x02, AMDGPU_RAS_BLOCK_MASK},
+	{0x66a1, 0x00, AMDGPU_RAS_BLOCK_MASK},
+	{0x66a1, 0x01, AMDGPU_RAS_BLOCK_MASK},
+	{0x66a1, 0x04, AMDGPU_RAS_BLOCK_MASK},
+	{0x66a3, 0x00, AMDGPU_RAS_BLOCK_MASK},
+	{0x66a7, 0x00, AMDGPU_RAS_BLOCK_MASK},
+};
+
+static uint32_t amdgpu_ras_check_supported(struct amdgpu_device *adev)
+{
+	/* TODO need check vbios table */
+	int i;
+	int did = adev->pdev->device;
+	int rid = adev->pdev->revision;
+
+	for (i = 0; i < ARRAY_SIZE(supported_DID_array); i++) {
+		if (did == supported_DID_array[i].did &&
+				rid == supported_DID_array[i].rid) {
+			return supported_DID_array[i].capability;
+		}
+	}
+	return 0;
+}
+
+int amdgpu_ras_init(struct amdgpu_device *adev)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	uint32_t supported = amdgpu_ras_check_supported(adev);
+
+	if (con || supported == 0)
+		return 0;
+
+	con = kmalloc(sizeof(struct amdgpu_ras) +
+			sizeof(struct ras_manager) * AMDGPU_RAS_BLOCK_COUNT,
+			GFP_KERNEL|__GFP_ZERO);
+	if (!con)
+		return -ENOMEM;
+
+	con->objs = (struct ras_manager *)(con + 1);
+
+	amdgpu_ras_set_context(adev, con);
+
+	con->supported = supported;
+	con->features = 0;
+	INIT_LIST_HEAD(&con->head);
+
+	if (amdgpu_ras_recovery_init(adev))
+		goto recovery_out;
+
+	amdgpu_ras_mask &= AMDGPU_RAS_BLOCK_MASK;
+
+	amdgpu_ras_enable_all_features(adev, 1);
+
+	if (amdgpu_ras_fs_init(adev))
+		goto fs_out;
+
+	amdgpu_ras_self_test(adev);
+	return 0;
+fs_out:
+	amdgpu_ras_recovery_fini(adev);
+recovery_out:
+	amdgpu_ras_set_context(adev, NULL);
+	kfree(con);
+
+	return -EINVAL;
+}
+
+/* do some fini work before IP fini as dependence */
+int amdgpu_ras_pre_fini(struct amdgpu_device *adev)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+
+	if (!con)
+		return 0;
+
+	/* Need disable ras on all IPs here before ip [hw/sw]fini */
+	amdgpu_ras_disable_all_features(adev, 0);
+	amdgpu_ras_recovery_fini(adev);
+	return 0;
+}
+
+int amdgpu_ras_fini(struct amdgpu_device *adev)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+
+	if (!con)
+		return 0;
+
+	amdgpu_ras_fs_fini(adev);
+	amdgpu_ras_interrupt_remove_all(adev);
+
+	WARN(con->features, "Feature mask is not cleared");
+
+	if (con->features)
+		amdgpu_ras_disable_all_features(adev, 1);
+
+	amdgpu_ras_set_context(adev, NULL);
+	kfree(con);
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
new file mode 100644
index 000000000000..d6d4bbb92a42
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
@@ -0,0 +1,217 @@
+/*
+ * Copyright 2018 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ *
+ */
+#ifndef _AMDGPU_RAS_H
+#define _AMDGPU_RAS_H
+
+#include <linux/debugfs.h>
+#include <linux/list.h>
+#include "amdgpu.h"
+#include "amdgpu_psp.h"
+#include "ta_ras_if.h"
+
+enum amdgpu_ras_block {
+	AMDGPU_RAS_BLOCK__UMC = 0,
+	AMDGPU_RAS_BLOCK__SDMA,
+	AMDGPU_RAS_BLOCK__GFX,
+	AMDGPU_RAS_BLOCK__MMHUB,
+	AMDGPU_RAS_BLOCK__ATHUB,
+	AMDGPU_RAS_BLOCK__PCIE_BIF,
+	AMDGPU_RAS_BLOCK__HDP,
+	AMDGPU_RAS_BLOCK__XGMI_WAFL,
+	AMDGPU_RAS_BLOCK__DF,
+	AMDGPU_RAS_BLOCK__SMN,
+	AMDGPU_RAS_BLOCK__SEM,
+	AMDGPU_RAS_BLOCK__MP0,
+	AMDGPU_RAS_BLOCK__MP1,
+	AMDGPU_RAS_BLOCK__FUSE,
+
+	AMDGPU_RAS_BLOCK__LAST
+};
+
+#define AMDGPU_RAS_BLOCK_COUNT	AMDGPU_RAS_BLOCK__LAST
+#define AMDGPU_RAS_BLOCK_MASK	((1ULL << AMDGPU_RAS_BLOCK_COUNT) - 1)
+
+enum amdgpu_ras_error_type {
+	AMDGPU_RAS_ERROR__NONE							= 0,
+	AMDGPU_RAS_ERROR__PARITY						= 1,
+	AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE					= 2,
+	AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE					= 4,
+	AMDGPU_RAS_ERROR__POISON						= 8,
+};
+
+enum amdgpu_ras_ret {
+	AMDGPU_RAS_SUCCESS = 0,
+	AMDGPU_RAS_FAIL,
+	AMDGPU_RAS_UE,
+	AMDGPU_RAS_CE,
+	AMDGPU_RAS_PT,
+};
+
+struct ras_common_if {
+	enum amdgpu_ras_block block;
+	enum amdgpu_ras_error_type type;
+	uint32_t sub_block_index;
+	/* block name */
+	char name[32];
+};
+
+typedef int (*ras_ih_cb)(struct amdgpu_device *adev,
+		struct amdgpu_iv_entry *entry);
+
+struct amdgpu_ras {
+	/* ras infrastructure */
+	uint32_t supported;
+	uint32_t features;
+	struct list_head head;
+	/* debugfs */
+	struct dentry *dir;
+	/* sysfs */
+	struct device_attribute features_attr;
+	/* block array */
+	struct ras_manager *objs;
+
+	/* gpu recovery */
+	struct work_struct recovery_work;
+	atomic_t in_recovery;
+	struct amdgpu_device *adev;
+	/* error handler data */
+	struct ras_err_handler_data *eh_data;
+	struct mutex recovery_lock;
+};
+
+/* interfaces for IP */
+
+struct ras_fs_if {
+	struct ras_common_if head;
+	char sysfs_name[32];
+	char debugfs_name[32];
+};
+
+struct ras_query_if {
+	struct ras_common_if head;
+	unsigned long ue_count;
+	unsigned long ce_count;
+};
+
+struct ras_inject_if {
+	struct ras_common_if head;
+	uint64_t address;
+	uint64_t value;
+};
+
+struct ras_cure_if {
+	struct ras_common_if head;
+	uint64_t address;
+};
+
+struct ras_ih_if {
+	struct ras_common_if head;
+	ras_ih_cb cb;
+};
+
+struct ras_dispatch_if {
+	struct ras_common_if head;
+	struct amdgpu_iv_entry *entry;
+};
+
+/* work flow
+ * vbios
+ * 1: ras feature enable (enabled by default)
+ * psp
+ * 2: ras framework init (in ip_init)
+ * IP
+ * 3: IH add
+ * 4: debugfs/sysfs create
+ * 5: query/inject
+ * 6: debugfs/sysfs remove
+ * 7: IH remove
+ * 8: feature disable
+ */
+
+#define amdgpu_ras_get_context(adev)		((adev)->psp.ras.ras)
+#define amdgpu_ras_set_context(adev, ras_con)	((adev)->psp.ras.ras = (ras_con))
+
+/* check if ras is supported on block, say, sdma, gfx */
+static inline int amdgpu_ras_is_supported(struct amdgpu_device *adev,
+		unsigned int block)
+{
+	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
+
+	return ras && (ras->supported & (1 << block));
+}
+
+int amdgpu_ras_query_error_count(struct amdgpu_device *adev,
+		bool is_ce);
+
+/* error handling functions */
+int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev,
+		unsigned long *bps, int pages);
+
+int amdgpu_ras_reserve_bad_pages(struct amdgpu_device *adev);
+
+static inline int amdgpu_ras_reset_gpu(struct amdgpu_device *adev,
+		bool is_baco)
+{
+	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
+
+	if (atomic_cmpxchg(&ras->in_recovery, 0, 1) == 0)
+		schedule_work(&ras->recovery_work);
+	return 0;
+}
+
+/* called in ip_init and ip_fini */
+int amdgpu_ras_init(struct amdgpu_device *adev);
+int amdgpu_ras_fini(struct amdgpu_device *adev);
+int amdgpu_ras_pre_fini(struct amdgpu_device *adev);
+
+int amdgpu_ras_feature_enable(struct amdgpu_device *adev,
+		struct ras_common_if *head, bool enable);
+
+int amdgpu_ras_sysfs_create(struct amdgpu_device *adev,
+		struct ras_fs_if *head);
+
+int amdgpu_ras_sysfs_remove(struct amdgpu_device *adev,
+		struct ras_common_if *head);
+
+int amdgpu_ras_debugfs_create(struct amdgpu_device *adev,
+		struct ras_fs_if *head);
+
+int amdgpu_ras_debugfs_remove(struct amdgpu_device *adev,
+		struct ras_common_if *head);
+
+int amdgpu_ras_error_query(struct amdgpu_device *adev,
+		struct ras_query_if *info);
+
+int amdgpu_ras_error_inject(struct amdgpu_device *adev,
+		struct ras_inject_if *info);
+
+int amdgpu_ras_interrupt_add_handler(struct amdgpu_device *adev,
+		struct ras_ih_if *info);
+
+int amdgpu_ras_interrupt_remove_handler(struct amdgpu_device *adev,
+		struct ras_ih_if *info);
+
+int amdgpu_ras_interrupt_dispatch(struct amdgpu_device *adev,
+		struct ras_dispatch_if *info);
+#endif
-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 10/20] drm/amdgpu: add debugfs ctrl node

[Alex Deucher]

From: xinhui pan <xinhui.pan@amd.com>

allow userspace enable/disable ras

Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 122 ++++++++++++++++++++++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h |   9 ++
 2 files changed, 121 insertions(+), 10 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index d7c7090fade9..ca9c7d1ede2f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -200,6 +200,90 @@ static const struct file_operations amdgpu_ras_debugfs_ops = {
 	.llseek = default_llseek
 };
 
+/*
+ * DOC: ras debugfs control interface
+ *
+ * It accepts struct ras_debug_if who has two members.
+ *
+ * First member: ras_debug_if::head or ras_debug_if::inject.
+ * It is used to indicate which IP block will be under control.
+ * Its contents are not human readable, IOW, write it by your programs.
+ *
+ * head has four members, they are block, type, sub_block_index, name.
+ * block: which IP will be under control.
+ * type: what kind of error will be enabled/disabled/injected.
+ * sub_block_index: some IPs have subcomponets. say, GFX, sDMA.
+ * name: the name of IP.
+ *
+ * inject has two more members than head, they are address, value.
+ * As their names indicate, inject operation will write the
+ * value to the address.
+ *
+ * Second member: struct ras_debug_if::op.
+ * It has three kinds of operations.
+ *  0: disable RAS on the block. Take ::head as its data.
+ *  1: enable RAS on the block. Take ::head as its data.
+ *  2: inject errors on the block. Take ::inject as its data.
+ *
+ * How to check the result?
+ *
+ * For disable/enable, please check ras features at
+ * /sys/class/drm/card[0/1/2...]/device/ras/features
+ *
+ * For inject, please check corresponding err count at
+ * /sys/class/drm/card[0/1/2...]/device/ras/[gfx/sdma/...]_err_count
+ *
+ * NOTE: operation is only allowed on blocks which are supported.
+ * Please check ras mask at /sys/module/amdgpu/parameters/ras_mask
+ */
+static ssize_t amdgpu_ras_debugfs_ctrl_write(struct file *f, const char __user *buf,
+		size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private;
+	struct ras_debug_if data;
+	int ret = 0;
+
+	if (size < sizeof(data))
+		return -EINVAL;
+
+	memset(&data, 0, sizeof(data));
+
+	if (*pos)
+		return -EINVAL;
+
+	if (copy_from_user(&data, buf, sizeof(data)))
+		return -EINVAL;
+
+	*pos = size;
+
+	if (!amdgpu_ras_is_supported(adev, data.head.block))
+		return -EINVAL;
+
+	switch (data.op) {
+	case 0:
+		ret = amdgpu_ras_feature_enable(adev, &data.head, 0);
+		break;
+	case 1:
+		ret = amdgpu_ras_feature_enable(adev, &data.head, 1);
+		break;
+	case 2:
+		ret = amdgpu_ras_error_inject(adev, &data.inject);
+		break;
+	};
+
+	if (ret)
+		return -EINVAL;
+
+	return size;
+}
+
+static const struct file_operations amdgpu_ras_debugfs_ctrl_ops = {
+	.owner = THIS_MODULE,
+	.read = NULL,
+	.write = amdgpu_ras_debugfs_ctrl_write,
+	.llseek = default_llseek
+};
+
 static ssize_t amdgpu_ras_sysfs_read(struct device *dev,
 		struct device_attribute *attr, char *buf)
 {
@@ -657,6 +741,31 @@ static int amdgpu_ras_sysfs_remove_all(struct amdgpu_device *adev)
 /* sysfs end */
 
 /* debugfs begin */
+static int amdgpu_ras_debugfs_create_ctrl_node(struct amdgpu_device *adev)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct drm_minor *minor = adev->ddev->primary;
+	struct dentry *root = minor->debugfs_root, *dir;
+	struct dentry *ent;
+
+	dir = debugfs_create_dir("ras", root);
+	if (IS_ERR(dir))
+		return -EINVAL;
+
+	con->dir = dir;
+
+	ent = debugfs_create_file("ras_ctrl",
+			S_IWUGO | S_IRUGO, con->dir,
+			adev, &amdgpu_ras_debugfs_ctrl_ops);
+	if (IS_ERR(ent)) {
+		debugfs_remove(con->dir);
+		return -EINVAL;
+	}
+
+	con->ent = ent;
+	return 0;
+}
+
 int amdgpu_ras_debugfs_create(struct amdgpu_device *adev,
 		struct ras_fs_if *head)
 {
@@ -709,8 +818,10 @@ static int amdgpu_ras_debugfs_remove_all(struct amdgpu_device *adev)
 		amdgpu_ras_debugfs_remove(adev, &obj->head);
 	}
 
+	debugfs_remove(con->ent);
 	debugfs_remove(con->dir);
 	con->dir = NULL;
+	con->ent = NULL;
 
 	return 0;
 }
@@ -720,17 +831,8 @@ static int amdgpu_ras_debugfs_remove_all(struct amdgpu_device *adev)
 
 static int amdgpu_ras_fs_init(struct amdgpu_device *adev)
 {
-	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
-	struct drm_minor *minor = adev->ddev->primary;
-	struct dentry *root = minor->debugfs_root, *dir;
-
-	dir = debugfs_create_dir("ras", root);
-	if (IS_ERR(dir))
-		return -EINVAL;
-
-	con->dir = dir;
-
 	amdgpu_ras_sysfs_create_feature_node(adev);
+	amdgpu_ras_debugfs_create_ctrl_node(adev);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
index d6d4bbb92a42..599dcd01aec7 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
@@ -86,6 +86,8 @@ struct amdgpu_ras {
 	struct list_head head;
 	/* debugfs */
 	struct dentry *dir;
+	/* debugfs ctrl */
+	struct dentry *ent;
 	/* sysfs */
 	struct device_attribute features_attr;
 	/* block array */
@@ -135,6 +137,13 @@ struct ras_dispatch_if {
 	struct amdgpu_iv_entry *entry;
 };
 
+struct ras_debug_if {
+	union {
+		struct ras_common_if head;
+		struct ras_inject_if inject;
+	};
+	int op;
+};
 /* work flow
  * vbios
  * 1: ras feature enable (enabled by default)
-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 11/20] drm/amdgpu: reserve bad pages during recovery

[Alex Deucher]

From: xinhui pan <xinhui.pan@amd.com>

Mark vram pages with errors as bad and prevent the driver
from using them.

Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index d6f130bdb985..a7876fafc558 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -3396,6 +3396,11 @@ static int amdgpu_do_asic_reset(struct amdgpu_hive_info *hive,
 						break;
 				}
 			}
+
+			list_for_each_entry(tmp_adev, device_list_handle,
+					gmc.xgmi.head) {
+				amdgpu_ras_reserve_bad_pages(tmp_adev);
+			}
 		}
 	}
 
-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 12/20] drm/amdgpu: enable ras on sdma4

[Alex Deucher]

From: xinhui pan <xinhui.pan@amd.com>

register IH, enable ras features on sdma.
create sysfs debugfs file for sdma.

Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Signed-off-by: Feifei Xu <Feifei.Xu@amd.com>
Signed-off-by: Eric Huang <JinhuiEric.Huang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h |   4 +
 drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c   | 184 ++++++++++++++++++++++-
 2 files changed, 187 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h
index 16b1a6ae5ba6..c17af30e758d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h
@@ -30,6 +30,8 @@
 enum amdgpu_sdma_irq {
 	AMDGPU_SDMA_IRQ_TRAP0 = 0,
 	AMDGPU_SDMA_IRQ_TRAP1,
+	AMDGPU_SDMA_IRQ_ECC0,
+	AMDGPU_SDMA_IRQ_ECC1,
 
 	AMDGPU_SDMA_IRQ_LAST
 };
@@ -49,9 +51,11 @@ struct amdgpu_sdma {
 	struct amdgpu_sdma_instance instance[AMDGPU_MAX_SDMA_INSTANCES];
 	struct amdgpu_irq_src	trap_irq;
 	struct amdgpu_irq_src	illegal_inst_irq;
+	struct amdgpu_irq_src	ecc_irq;
 	int			num_instances;
 	uint32_t                    srbm_soft_reset;
 	bool			has_page_queue;
+	struct ras_common_if	*ras_if;
 };
 
 /*
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c
index 127b85983e8f..058b9daec514 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c
@@ -41,6 +41,8 @@
 #include "ivsrcid/sdma0/irqsrcs_sdma0_4_0.h"
 #include "ivsrcid/sdma1/irqsrcs_sdma1_4_0.h"
 
+#include "amdgpu_ras.h"
+
 MODULE_FIRMWARE("amdgpu/vega10_sdma.bin");
 MODULE_FIRMWARE("amdgpu/vega10_sdma1.bin");
 MODULE_FIRMWARE("amdgpu/vega12_sdma.bin");
@@ -1493,6 +1495,83 @@ static int sdma_v4_0_early_init(void *handle)
 	return 0;
 }
 
+static int sdma_v4_0_process_ras_data_cb(struct amdgpu_device *adev,
+		struct amdgpu_iv_entry *entry);
+
+static int sdma_v4_0_late_init(void *handle)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	struct ras_common_if **ras_if = &adev->sdma.ras_if;
+	struct ras_ih_if ih_info = {
+		.cb = sdma_v4_0_process_ras_data_cb,
+	};
+	struct ras_fs_if fs_info = {
+		.sysfs_name = "sdma_err_count",
+		.debugfs_name = "sdma_err_inject",
+	};
+	struct ras_common_if ras_block = {
+		.block = AMDGPU_RAS_BLOCK__SDMA,
+		.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE,
+		.sub_block_index = 0,
+		.name = "sdma",
+	};
+	int r;
+
+	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__SDMA)) {
+		amdgpu_ras_feature_enable(adev, &ras_block, 0);
+		return 0;
+	}
+
+	*ras_if = kmalloc(sizeof(**ras_if), GFP_KERNEL);
+	if (!*ras_if)
+		return -ENOMEM;
+
+	**ras_if = ras_block;
+
+	r = amdgpu_ras_feature_enable(adev, *ras_if, 1);
+	if (r)
+		goto feature;
+
+	ih_info.head = **ras_if;
+	fs_info.head = **ras_if;
+
+	r = amdgpu_ras_interrupt_add_handler(adev, &ih_info);
+	if (r)
+		goto interrupt;
+
+	r = amdgpu_ras_debugfs_create(adev, &fs_info);
+	if (r)
+		goto debugfs;
+
+	r = amdgpu_ras_sysfs_create(adev, &fs_info);
+	if (r)
+		goto sysfs;
+
+	r = amdgpu_irq_get(adev, &adev->sdma.ecc_irq, AMDGPU_SDMA_IRQ_ECC0);
+	if (r)
+		goto irq;
+
+	r = amdgpu_irq_get(adev, &adev->sdma.ecc_irq, AMDGPU_SDMA_IRQ_ECC1);
+	if (r) {
+		amdgpu_irq_put(adev, &adev->sdma.ecc_irq, AMDGPU_SDMA_IRQ_ECC0);
+		goto irq;
+	}
+
+	return 0;
+irq:
+	amdgpu_ras_sysfs_remove(adev, *ras_if);
+sysfs:
+	amdgpu_ras_debugfs_remove(adev, *ras_if);
+debugfs:
+	amdgpu_ras_interrupt_remove_handler(adev, &ih_info);
+interrupt:
+	amdgpu_ras_feature_enable(adev, *ras_if, 0);
+feature:
+	kfree(*ras_if);
+	*ras_if = NULL;
+	return -EINVAL;
+}
+
 static int sdma_v4_0_sw_init(void *handle)
 {
 	struct amdgpu_ring *ring;
@@ -1511,6 +1590,18 @@ static int sdma_v4_0_sw_init(void *handle)
 	if (r)
 		return r;
 
+	/* SDMA SRAM ECC event */
+	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_SDMA0, SDMA0_4_0__SRCID__SDMA_SRAM_ECC,
+			&adev->sdma.ecc_irq);
+	if (r)
+		return r;
+
+	/* SDMA SRAM ECC event */
+	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_SDMA1, SDMA1_4_0__SRCID__SDMA_SRAM_ECC,
+			&adev->sdma.ecc_irq);
+	if (r)
+		return r;
+
 	for (i = 0; i < adev->sdma.num_instances; i++) {
 		ring = &adev->sdma.instance[i].ring;
 		ring->ring_obj = NULL;
@@ -1561,6 +1652,22 @@ static int sdma_v4_0_sw_fini(void *handle)
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 	int i;
 
+	if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__SDMA) &&
+			adev->sdma.ras_if) {
+		struct ras_common_if *ras_if = adev->sdma.ras_if;
+		struct ras_ih_if ih_info = {
+			.head = *ras_if,
+		};
+
+		/*remove fs first*/
+		amdgpu_ras_debugfs_remove(adev, ras_if);
+		amdgpu_ras_sysfs_remove(adev, ras_if);
+		/*remove the IH*/
+		amdgpu_ras_interrupt_remove_handler(adev, &ih_info);
+		amdgpu_ras_feature_enable(adev, ras_if, 0);
+		kfree(ras_if);
+	}
+
 	for (i = 0; i < adev->sdma.num_instances; i++) {
 		amdgpu_ring_fini(&adev->sdma.instance[i].ring);
 		if (adev->sdma.has_page_queue)
@@ -1598,6 +1705,9 @@ static int sdma_v4_0_hw_fini(void *handle)
 	if (amdgpu_sriov_vf(adev))
 		return 0;
 
+	amdgpu_irq_put(adev, &adev->sdma.ecc_irq, AMDGPU_SDMA_IRQ_ECC0);
+	amdgpu_irq_put(adev, &adev->sdma.ecc_irq, AMDGPU_SDMA_IRQ_ECC1);
+
 	sdma_v4_0_ctx_switch_enable(adev, false);
 	sdma_v4_0_enable(adev, false);
 
@@ -1714,6 +1824,50 @@ static int sdma_v4_0_process_trap_irq(struct amdgpu_device *adev,
 	return 0;
 }
 
+static int sdma_v4_0_process_ras_data_cb(struct amdgpu_device *adev,
+		struct amdgpu_iv_entry *entry)
+{
+	uint32_t instance, err_source;
+
+	switch (entry->client_id) {
+	case SOC15_IH_CLIENTID_SDMA0:
+		instance = 0;
+		break;
+	case SOC15_IH_CLIENTID_SDMA1:
+		instance = 1;
+		break;
+	default:
+		return 0;
+	}
+
+	switch (entry->src_id) {
+	case SDMA0_4_0__SRCID__SDMA_SRAM_ECC:
+		err_source = 0;
+		break;
+	case SDMA0_4_0__SRCID__SDMA_ECC:
+		err_source = 1;
+		break;
+	default:
+		return 0;
+	}
+
+	amdgpu_ras_reset_gpu(adev, 0);
+
+	return AMDGPU_RAS_UE;
+}
+
+static int sdma_v4_0_process_ecc_irq(struct amdgpu_device *adev,
+				      struct amdgpu_irq_src *source,
+				      struct amdgpu_iv_entry *entry)
+{
+	struct ras_dispatch_if ih_data = {
+		.head = *adev->sdma.ras_if,
+		.entry = entry,
+	};
+	amdgpu_ras_interrupt_dispatch(adev, &ih_data);
+	return 0;
+}
+
 static int sdma_v4_0_process_illegal_inst_irq(struct amdgpu_device *adev,
 					      struct amdgpu_irq_src *source,
 					      struct amdgpu_iv_entry *entry)
@@ -1741,6 +1895,25 @@ static int sdma_v4_0_process_illegal_inst_irq(struct amdgpu_device *adev,
 	return 0;
 }
 
+static int sdma_v4_0_set_ecc_irq_state(struct amdgpu_device *adev,
+					struct amdgpu_irq_src *source,
+					unsigned type,
+					enum amdgpu_interrupt_state state)
+{
+	u32 sdma_edc_config;
+
+	u32 reg_offset = (type == AMDGPU_SDMA_IRQ_ECC0) ?
+		sdma_v4_0_get_reg_offset(adev, 0, mmSDMA0_EDC_CONFIG) :
+		sdma_v4_0_get_reg_offset(adev, 1, mmSDMA0_EDC_CONFIG);
+
+	sdma_edc_config = RREG32(reg_offset);
+	sdma_edc_config = REG_SET_FIELD(sdma_edc_config, SDMA0_EDC_CONFIG, ECC_INT_ENABLE,
+		       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
+	WREG32(reg_offset, sdma_edc_config);
+
+	return 0;
+}
+
 static void sdma_v4_0_update_medium_grain_clock_gating(
 		struct amdgpu_device *adev,
 		bool enable)
@@ -1906,7 +2079,7 @@ static void sdma_v4_0_get_clockgating_state(void *handle, u32 *flags)
 const struct amd_ip_funcs sdma_v4_0_ip_funcs = {
 	.name = "sdma_v4_0",
 	.early_init = sdma_v4_0_early_init,
-	.late_init = NULL,
+	.late_init = sdma_v4_0_late_init,
 	.sw_init = sdma_v4_0_sw_init,
 	.sw_fini = sdma_v4_0_sw_fini,
 	.hw_init = sdma_v4_0_hw_init,
@@ -2008,11 +2181,20 @@ static const struct amdgpu_irq_src_funcs sdma_v4_0_illegal_inst_irq_funcs = {
 	.process = sdma_v4_0_process_illegal_inst_irq,
 };
 
+static const struct amdgpu_irq_src_funcs sdma_v4_0_ecc_irq_funcs = {
+	.set = sdma_v4_0_set_ecc_irq_state,
+	.process = sdma_v4_0_process_ecc_irq,
+};
+
+
+
 static void sdma_v4_0_set_irq_funcs(struct amdgpu_device *adev)
 {
 	adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST;
 	adev->sdma.trap_irq.funcs = &sdma_v4_0_trap_irq_funcs;
 	adev->sdma.illegal_inst_irq.funcs = &sdma_v4_0_illegal_inst_irq_funcs;
+	adev->sdma.ecc_irq.num_types = AMDGPU_SDMA_IRQ_LAST;
+	adev->sdma.ecc_irq.funcs = &sdma_v4_0_ecc_irq_funcs;
 }
 
 /**
-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 13/20] drm/amdgpu: enable ras on gfx9 (v2)

[Alex Deucher]

From: Feifei Xu <Feifei.Xu@amd.com>

Register ecc interrupts and ecc interrupt handler on gfx9.
Add ras support on gfx9

v2: squash in warning fix

Signed-off-by: Feifei Xu <Feifei.Xu@amd.com>
Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h |   3 +
 drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c   | 173 ++++++++++++++++++++++++
 2 files changed, 176 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
index f790e15bcd08..09fc53af3d35 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
@@ -258,6 +258,9 @@ struct amdgpu_gfx {
 	/* pipe reservation */
 	struct mutex			pipe_reserve_mutex;
 	DECLARE_BITMAP			(pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
+
+	/*ras */
+	struct ras_common_if		*ras_if;
 };
 
 #define amdgpu_gfx_get_gpu_clock_counter(adev) (adev)->gfx.funcs->get_gpu_clock_counter((adev))
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
index 5533f6e4f4a4..88c45f990f05 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
@@ -40,6 +40,8 @@
 
 #include "ivsrcid/gfx/irqsrcs_gfx_9_0.h"
 
+#include "amdgpu_ras.h"
+
 #define GFX9_NUM_GFX_RINGS     1
 #define GFX9_MEC_HPD_SIZE 4096
 #define RLCG_UCODE_LOADING_START_ADDRESS 0x00002000L
@@ -1638,6 +1640,18 @@ static int gfx_v9_0_sw_init(void *handle)
 	if (r)
 		return r;
 
+	/* ECC error */
+	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_ECC_ERROR,
+			      &adev->gfx.cp_ecc_error_irq);
+	if (r)
+		return r;
+
+	/* FUE error */
+	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_FUE_ERROR,
+			      &adev->gfx.cp_ecc_error_irq);
+	if (r)
+		return r;
+
 	adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
 
 	gfx_v9_0_scratch_init(adev);
@@ -1730,6 +1744,20 @@ static int gfx_v9_0_sw_fini(void *handle)
 	int i;
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
+	if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX) &&
+			adev->gfx.ras_if) {
+		struct ras_common_if *ras_if = adev->gfx.ras_if;
+		struct ras_ih_if ih_info = {
+			.head = *ras_if,
+		};
+
+		amdgpu_ras_debugfs_remove(adev, ras_if);
+		amdgpu_ras_sysfs_remove(adev, ras_if);
+		amdgpu_ras_interrupt_remove_handler(adev,  &ih_info);
+		amdgpu_ras_feature_enable(adev, ras_if, 0);
+		kfree(ras_if);
+	}
+
 	amdgpu_bo_free_kernel(&adev->gds.oa_gfx_bo, NULL, NULL);
 	amdgpu_bo_free_kernel(&adev->gds.gws_gfx_bo, NULL, NULL);
 	amdgpu_bo_free_kernel(&adev->gds.gds_gfx_bo, NULL, NULL);
@@ -3304,6 +3332,7 @@ static int gfx_v9_0_hw_fini(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
+	amdgpu_irq_put(adev, &adev->gfx.cp_ecc_error_irq, 0);
 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
 
@@ -3493,6 +3522,77 @@ static int gfx_v9_0_early_init(void *handle)
 	return 0;
 }
 
+static int gfx_v9_0_process_ras_data_cb(struct amdgpu_device *adev,
+		struct amdgpu_iv_entry *entry);
+
+static int gfx_v9_0_ecc_late_init(void *handle)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	struct ras_common_if **ras_if = &adev->gfx.ras_if;
+	struct ras_ih_if ih_info = {
+		.cb = gfx_v9_0_process_ras_data_cb,
+	};
+	struct ras_fs_if fs_info = {
+		.sysfs_name = "gfx_err_count",
+		.debugfs_name = "gfx_err_inject",
+	};
+	struct ras_common_if ras_block = {
+		.block = AMDGPU_RAS_BLOCK__GFX,
+		.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE,
+		.sub_block_index = 0,
+		.name = "gfx",
+	};
+	int r;
+
+	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX)) {
+		amdgpu_ras_feature_enable(adev, &ras_block, 0);
+		return 0;
+	}
+
+	*ras_if = kmalloc(sizeof(**ras_if), GFP_KERNEL);
+	if (!*ras_if)
+		return -ENOMEM;
+
+	**ras_if = ras_block;
+
+	r = amdgpu_ras_feature_enable(adev, *ras_if, 1);
+	if (r)
+		goto feature;
+
+	ih_info.head = **ras_if;
+	fs_info.head = **ras_if;
+
+	r = amdgpu_ras_interrupt_add_handler(adev, &ih_info);
+	if (r)
+		goto interrupt;
+
+	r = amdgpu_ras_debugfs_create(adev, &fs_info);
+	if (r)
+		goto debugfs;
+
+	r = amdgpu_ras_sysfs_create(adev, &fs_info);
+	if (r)
+		goto sysfs;
+
+	r = amdgpu_irq_get(adev, &adev->gfx.cp_ecc_error_irq, 0);
+	if (r)
+		goto irq;
+
+	return 0;
+irq:
+	amdgpu_ras_sysfs_remove(adev, *ras_if);
+sysfs:
+	amdgpu_ras_debugfs_remove(adev, *ras_if);
+debugfs:
+	amdgpu_ras_interrupt_remove_handler(adev, &ih_info);
+interrupt:
+	amdgpu_ras_feature_enable(adev, *ras_if, 0);
+feature:
+	kfree(*ras_if);
+	*ras_if = NULL;
+	return -EINVAL;
+}
+
 static int gfx_v9_0_late_init(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
@@ -3506,6 +3606,10 @@ static int gfx_v9_0_late_init(void *handle)
 	if (r)
 		return r;
 
+	r = gfx_v9_0_ecc_late_init(handle);
+	if (r)
+		return r;
+
 	return 0;
 }
 
@@ -4542,6 +4646,45 @@ static int gfx_v9_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
 	return 0;
 }
 
+#define ENABLE_ECC_ON_ME_PIPE(me, pipe)				\
+	WREG32_FIELD15(GC, 0, CP_ME##me##_PIPE##pipe##_INT_CNTL,\
+			CP_ECC_ERROR_INT_ENABLE, 1)
+
+#define DISABLE_ECC_ON_ME_PIPE(me, pipe)			\
+	WREG32_FIELD15(GC, 0, CP_ME##me##_PIPE##pipe##_INT_CNTL,\
+			CP_ECC_ERROR_INT_ENABLE, 0)
+
+static int gfx_v9_0_set_cp_ecc_error_state(struct amdgpu_device *adev,
+					      struct amdgpu_irq_src *source,
+					      unsigned type,
+					      enum amdgpu_interrupt_state state)
+{
+	switch (state) {
+	case AMDGPU_IRQ_STATE_DISABLE:
+		WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
+				CP_ECC_ERROR_INT_ENABLE, 0);
+		DISABLE_ECC_ON_ME_PIPE(1, 0);
+		DISABLE_ECC_ON_ME_PIPE(1, 1);
+		DISABLE_ECC_ON_ME_PIPE(1, 2);
+		DISABLE_ECC_ON_ME_PIPE(1, 3);
+		break;
+
+	case AMDGPU_IRQ_STATE_ENABLE:
+		WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
+				CP_ECC_ERROR_INT_ENABLE, 1);
+		ENABLE_ECC_ON_ME_PIPE(1, 0);
+		ENABLE_ECC_ON_ME_PIPE(1, 1);
+		ENABLE_ECC_ON_ME_PIPE(1, 2);
+		ENABLE_ECC_ON_ME_PIPE(1, 3);
+		break;
+	default:
+		break;
+	}
+
+	return 0;
+}
+
+
 static int gfx_v9_0_set_eop_interrupt_state(struct amdgpu_device *adev,
 					    struct amdgpu_irq_src *src,
 					    unsigned type,
@@ -4658,6 +4801,27 @@ static int gfx_v9_0_priv_inst_irq(struct amdgpu_device *adev,
 	return 0;
 }
 
+static int gfx_v9_0_process_ras_data_cb(struct amdgpu_device *adev,
+		struct amdgpu_iv_entry *entry)
+{
+	/* TODO ue will trigger an interrupt. */
+	amdgpu_ras_reset_gpu(adev, 0);
+	return AMDGPU_RAS_UE;
+}
+
+static int gfx_v9_0_cp_ecc_error_irq(struct amdgpu_device *adev,
+				  struct amdgpu_irq_src *source,
+				  struct amdgpu_iv_entry *entry)
+{
+	struct ras_dispatch_if ih_data = {
+		.head = *adev->gfx.ras_if,
+		.entry = entry,
+	};
+	DRM_ERROR("CP ECC ERROR IRQ\n");
+	amdgpu_ras_interrupt_dispatch(adev, &ih_data);
+	return 0;
+}
+
 static const struct amd_ip_funcs gfx_v9_0_ip_funcs = {
 	.name = "gfx_v9_0",
 	.early_init = gfx_v9_0_early_init,
@@ -4819,6 +4983,12 @@ static const struct amdgpu_irq_src_funcs gfx_v9_0_priv_inst_irq_funcs = {
 	.process = gfx_v9_0_priv_inst_irq,
 };
 
+static const struct amdgpu_irq_src_funcs gfx_v9_0_cp_ecc_error_irq_funcs = {
+	.set = gfx_v9_0_set_cp_ecc_error_state,
+	.process = gfx_v9_0_cp_ecc_error_irq,
+};
+
+
 static void gfx_v9_0_set_irq_funcs(struct amdgpu_device *adev)
 {
 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
@@ -4829,6 +4999,9 @@ static void gfx_v9_0_set_irq_funcs(struct amdgpu_device *adev)
 
 	adev->gfx.priv_inst_irq.num_types = 1;
 	adev->gfx.priv_inst_irq.funcs = &gfx_v9_0_priv_inst_irq_funcs;
+
+	adev->gfx.cp_ecc_error_irq.num_types = 2; /*C5 ECC error and C9 FUE error*/
+	adev->gfx.cp_ecc_error_irq.funcs = &gfx_v9_0_cp_ecc_error_irq_funcs;
 }
 
 static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev)
-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 14/20] drm/amdgpu: enable ras on gmc9

[Alex Deucher]

From: xinhui pan <xinhui.pan@amd.com>

Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h |   2 +
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c   | 276 ++++++++++++++++++++++++
 2 files changed, 278 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
index d6c10b4d68c0..6ce45664ff87 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
@@ -144,6 +144,8 @@ struct amdgpu_gmc {
 	const struct amdgpu_gmc_funcs	*gmc_funcs;
 
 	struct amdgpu_xgmi xgmi;
+	struct amdgpu_irq_src	ecc_irq;
+	struct ras_common_if    *ras_if;
 };
 
 #define amdgpu_gmc_flush_gpu_tlb(adev, vmid, type) (adev)->gmc.gmc_funcs->flush_gpu_tlb((adev), (vmid), (type))
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
index d91274037c9f..1a164fe2dd9a 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
@@ -47,6 +47,8 @@
 
 #include "ivsrcid/vmc/irqsrcs_vmc_1_0.h"
 
+#include "amdgpu_ras.h"
+
 /* add these here since we already include dce12 headers and these are for DCN */
 #define mmHUBP0_DCSURF_PRI_VIEWPORT_DIMENSION                                                          0x055d
 #define mmHUBP0_DCSURF_PRI_VIEWPORT_DIMENSION_BASE_IDX                                                 2
@@ -199,6 +201,175 @@ static const uint32_t ecc_umcch_eccctrl_addrs[] = {
 	UMCCH_ECCCTRL_ADDR15,
 };
 
+static const uint32_t ecc_umc_mcumc_ctrl_addrs[] = {
+	(0x000143c0 + 0x00000000),
+	(0x000143c0 + 0x00000800),
+	(0x000143c0 + 0x00001000),
+	(0x000143c0 + 0x00001800),
+	(0x000543c0 + 0x00000000),
+	(0x000543c0 + 0x00000800),
+	(0x000543c0 + 0x00001000),
+	(0x000543c0 + 0x00001800),
+	(0x000943c0 + 0x00000000),
+	(0x000943c0 + 0x00000800),
+	(0x000943c0 + 0x00001000),
+	(0x000943c0 + 0x00001800),
+	(0x000d43c0 + 0x00000000),
+	(0x000d43c0 + 0x00000800),
+	(0x000d43c0 + 0x00001000),
+	(0x000d43c0 + 0x00001800),
+	(0x001143c0 + 0x00000000),
+	(0x001143c0 + 0x00000800),
+	(0x001143c0 + 0x00001000),
+	(0x001143c0 + 0x00001800),
+	(0x001543c0 + 0x00000000),
+	(0x001543c0 + 0x00000800),
+	(0x001543c0 + 0x00001000),
+	(0x001543c0 + 0x00001800),
+	(0x001943c0 + 0x00000000),
+	(0x001943c0 + 0x00000800),
+	(0x001943c0 + 0x00001000),
+	(0x001943c0 + 0x00001800),
+	(0x001d43c0 + 0x00000000),
+	(0x001d43c0 + 0x00000800),
+	(0x001d43c0 + 0x00001000),
+	(0x001d43c0 + 0x00001800),
+};
+
+static const uint32_t ecc_umc_mcumc_ctrl_mask_addrs[] = {
+	(0x000143e0 + 0x00000000),
+	(0x000143e0 + 0x00000800),
+	(0x000143e0 + 0x00001000),
+	(0x000143e0 + 0x00001800),
+	(0x000543e0 + 0x00000000),
+	(0x000543e0 + 0x00000800),
+	(0x000543e0 + 0x00001000),
+	(0x000543e0 + 0x00001800),
+	(0x000943e0 + 0x00000000),
+	(0x000943e0 + 0x00000800),
+	(0x000943e0 + 0x00001000),
+	(0x000943e0 + 0x00001800),
+	(0x000d43e0 + 0x00000000),
+	(0x000d43e0 + 0x00000800),
+	(0x000d43e0 + 0x00001000),
+	(0x000d43e0 + 0x00001800),
+	(0x001143e0 + 0x00000000),
+	(0x001143e0 + 0x00000800),
+	(0x001143e0 + 0x00001000),
+	(0x001143e0 + 0x00001800),
+	(0x001543e0 + 0x00000000),
+	(0x001543e0 + 0x00000800),
+	(0x001543e0 + 0x00001000),
+	(0x001543e0 + 0x00001800),
+	(0x001943e0 + 0x00000000),
+	(0x001943e0 + 0x00000800),
+	(0x001943e0 + 0x00001000),
+	(0x001943e0 + 0x00001800),
+	(0x001d43e0 + 0x00000000),
+	(0x001d43e0 + 0x00000800),
+	(0x001d43e0 + 0x00001000),
+	(0x001d43e0 + 0x00001800),
+};
+
+static const uint32_t ecc_umc_mcumc_status_addrs[] = {
+	(0x000143c2 + 0x00000000),
+	(0x000143c2 + 0x00000800),
+	(0x000143c2 + 0x00001000),
+	(0x000143c2 + 0x00001800),
+	(0x000543c2 + 0x00000000),
+	(0x000543c2 + 0x00000800),
+	(0x000543c2 + 0x00001000),
+	(0x000543c2 + 0x00001800),
+	(0x000943c2 + 0x00000000),
+	(0x000943c2 + 0x00000800),
+	(0x000943c2 + 0x00001000),
+	(0x000943c2 + 0x00001800),
+	(0x000d43c2 + 0x00000000),
+	(0x000d43c2 + 0x00000800),
+	(0x000d43c2 + 0x00001000),
+	(0x000d43c2 + 0x00001800),
+	(0x001143c2 + 0x00000000),
+	(0x001143c2 + 0x00000800),
+	(0x001143c2 + 0x00001000),
+	(0x001143c2 + 0x00001800),
+	(0x001543c2 + 0x00000000),
+	(0x001543c2 + 0x00000800),
+	(0x001543c2 + 0x00001000),
+	(0x001543c2 + 0x00001800),
+	(0x001943c2 + 0x00000000),
+	(0x001943c2 + 0x00000800),
+	(0x001943c2 + 0x00001000),
+	(0x001943c2 + 0x00001800),
+	(0x001d43c2 + 0x00000000),
+	(0x001d43c2 + 0x00000800),
+	(0x001d43c2 + 0x00001000),
+	(0x001d43c2 + 0x00001800),
+};
+
+static int gmc_v9_0_ecc_interrupt_state(struct amdgpu_device *adev,
+		struct amdgpu_irq_src *src,
+		unsigned type,
+		enum amdgpu_interrupt_state state)
+{
+	u32 bits, i, tmp, reg;
+
+	bits = 0x7f;
+
+	switch (state) {
+	case AMDGPU_IRQ_STATE_DISABLE:
+		for (i = 0; i < ARRAY_SIZE(ecc_umc_mcumc_ctrl_addrs); i++) {
+			reg = ecc_umc_mcumc_ctrl_addrs[i];
+			tmp = RREG32(reg);
+			tmp &= ~bits;
+			WREG32(reg, tmp);
+		}
+		for (i = 0; i < ARRAY_SIZE(ecc_umc_mcumc_ctrl_mask_addrs); i++) {
+			reg = ecc_umc_mcumc_ctrl_mask_addrs[i];
+			tmp = RREG32(reg);
+			tmp &= ~bits;
+			WREG32(reg, tmp);
+		}
+		break;
+	case AMDGPU_IRQ_STATE_ENABLE:
+		for (i = 0; i < ARRAY_SIZE(ecc_umc_mcumc_ctrl_addrs); i++) {
+			reg = ecc_umc_mcumc_ctrl_addrs[i];
+			tmp = RREG32(reg);
+			tmp |= bits;
+			WREG32(reg, tmp);
+		}
+		for (i = 0; i < ARRAY_SIZE(ecc_umc_mcumc_ctrl_mask_addrs); i++) {
+			reg = ecc_umc_mcumc_ctrl_mask_addrs[i];
+			tmp = RREG32(reg);
+			tmp |= bits;
+			WREG32(reg, tmp);
+		}
+		break;
+	default:
+		break;
+	}
+
+	return 0;
+}
+
+static int gmc_v9_0_process_ras_data_cb(struct amdgpu_device *adev,
+		struct amdgpu_iv_entry *entry)
+{
+	amdgpu_ras_reset_gpu(adev, 0);
+	return AMDGPU_RAS_UE;
+}
+
+static int gmc_v9_0_process_ecc_irq(struct amdgpu_device *adev,
+		struct amdgpu_irq_src *source,
+		struct amdgpu_iv_entry *entry)
+{
+	struct ras_dispatch_if ih_data = {
+		.head = *adev->gmc.ras_if,
+		.entry = entry,
+	};
+	amdgpu_ras_interrupt_dispatch(adev, &ih_data);
+	return 0;
+}
+
 static int gmc_v9_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
 					struct amdgpu_irq_src *src,
 					unsigned type,
@@ -350,10 +521,19 @@ static const struct amdgpu_irq_src_funcs gmc_v9_0_irq_funcs = {
 	.process = gmc_v9_0_process_interrupt,
 };
 
+
+static const struct amdgpu_irq_src_funcs gmc_v9_0_ecc_funcs = {
+	.set = gmc_v9_0_ecc_interrupt_state,
+	.process = gmc_v9_0_process_ecc_irq,
+};
+
 static void gmc_v9_0_set_irq_funcs(struct amdgpu_device *adev)
 {
 	adev->gmc.vm_fault.num_types = 1;
 	adev->gmc.vm_fault.funcs = &gmc_v9_0_irq_funcs;
+
+	adev->gmc.ecc_irq.num_types = 1;
+	adev->gmc.ecc_irq.funcs = &gmc_v9_0_ecc_funcs;
 }
 
 static uint32_t gmc_v9_0_get_invalidate_req(unsigned int vmid,
@@ -723,6 +903,75 @@ static int gmc_v9_0_allocate_vm_inv_eng(struct amdgpu_device *adev)
 	return 0;
 }
 
+static int gmc_v9_0_ecc_late_init(void *handle)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	struct ras_common_if **ras_if = &adev->gmc.ras_if;
+	struct ras_ih_if ih_info = {
+		.cb = gmc_v9_0_process_ras_data_cb,
+	};
+	struct ras_fs_if fs_info = {
+		.sysfs_name = "umc_err_count",
+		.debugfs_name = "umc_err_inject",
+	};
+	struct ras_common_if ras_block = {
+		.block = AMDGPU_RAS_BLOCK__UMC,
+		.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE,
+		.sub_block_index = 0,
+		.name = "umc",
+	};
+	int r;
+
+	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__UMC)) {
+		amdgpu_ras_feature_enable(adev, &ras_block, 0);
+		return 0;
+	}
+
+	*ras_if = kmalloc(sizeof(**ras_if), GFP_KERNEL);
+	if (!*ras_if)
+		return -ENOMEM;
+
+	**ras_if = ras_block;
+
+	r = amdgpu_ras_feature_enable(adev, *ras_if, 1);
+	if (r)
+		goto feature;
+
+	ih_info.head = **ras_if;
+	fs_info.head = **ras_if;
+
+	r = amdgpu_ras_interrupt_add_handler(adev, &ih_info);
+	if (r)
+		goto interrupt;
+
+	r = amdgpu_ras_debugfs_create(adev, &fs_info);
+	if (r)
+		goto debugfs;
+
+	r = amdgpu_ras_sysfs_create(adev, &fs_info);
+	if (r)
+		goto sysfs;
+
+	r = amdgpu_irq_get(adev, &adev->gmc.ecc_irq, 0);
+	if (r)
+		goto irq;
+
+	return 0;
+irq:
+	amdgpu_ras_sysfs_remove(adev, *ras_if);
+sysfs:
+	amdgpu_ras_debugfs_remove(adev, *ras_if);
+debugfs:
+	amdgpu_ras_interrupt_remove_handler(adev, &ih_info);
+interrupt:
+	amdgpu_ras_feature_enable(adev, *ras_if, 0);
+feature:
+	kfree(*ras_if);
+	*ras_if = NULL;
+	return -EINVAL;
+}
+
+
 static int gmc_v9_0_late_init(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
@@ -748,6 +997,10 @@ static int gmc_v9_0_late_init(void *handle)
 		}
 	}
 
+	r = gmc_v9_0_ecc_late_init(handle);
+	if (r)
+		return r;
+
 	return amdgpu_irq_get(adev, &adev->gmc.vm_fault, 0);
 }
 
@@ -959,6 +1212,12 @@ static int gmc_v9_0_sw_init(void *handle)
 	if (r)
 		return r;
 
+	/* interrupt sent to DF. */
+	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_DF, 0,
+			&adev->gmc.ecc_irq);
+	if (r)
+		return r;
+
 	/* Set the internal MC address mask
 	 * This is the max address of the GPU's
 	 * internal address space.
@@ -1024,6 +1283,22 @@ static int gmc_v9_0_sw_fini(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
+	if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__UMC) &&
+			adev->gmc.ras_if) {
+		struct ras_common_if *ras_if = adev->gmc.ras_if;
+		struct ras_ih_if ih_info = {
+			.head = *ras_if,
+		};
+
+		/*remove fs first*/
+		amdgpu_ras_debugfs_remove(adev, ras_if);
+		amdgpu_ras_sysfs_remove(adev, ras_if);
+		/*remove the IH*/
+		amdgpu_ras_interrupt_remove_handler(adev, &ih_info);
+		amdgpu_ras_feature_enable(adev, ras_if, 0);
+		kfree(ras_if);
+	}
+
 	amdgpu_gem_force_release(adev);
 	amdgpu_vm_manager_fini(adev);
 
@@ -1170,6 +1445,7 @@ static int gmc_v9_0_hw_fini(void *handle)
 		return 0;
 	}
 
+	amdgpu_irq_put(adev, &adev->gmc.ecc_irq, 0);
 	amdgpu_irq_put(adev, &adev->gmc.vm_fault, 0);
 	gmc_v9_0_gart_disable(adev);
 
-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 15/20] drm/amdgpu: Add a new flag to AMDGPU_CTX_OP_QUERY_STATE2

[Alex Deucher]

From: xinhui pan <xinhui.pan@amd.com>

Add AMDGPU_CTX_QUERY2_FLAGS_RAS_CE/UE which indicate if any error happened
between previous query and this query.

Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 17 +++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h |  2 ++
 include/uapi/drm/amdgpu_drm.h           |  3 +++
 3 files changed, 22 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index 7b526593eb77..736ed1d67ec2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -26,6 +26,7 @@
 #include <drm/drm_auth.h>
 #include "amdgpu.h"
 #include "amdgpu_sched.h"
+#include "amdgpu_ras.h"
 
 #define to_amdgpu_ctx_entity(e)	\
 	container_of((e), struct amdgpu_ctx_entity, entity)
@@ -344,6 +345,7 @@ static int amdgpu_ctx_query2(struct amdgpu_device *adev,
 {
 	struct amdgpu_ctx *ctx;
 	struct amdgpu_ctx_mgr *mgr;
+	uint32_t ras_counter;
 
 	if (!fpriv)
 		return -EINVAL;
@@ -368,6 +370,21 @@ static int amdgpu_ctx_query2(struct amdgpu_device *adev,
 	if (atomic_read(&ctx->guilty))
 		out->state.flags |= AMDGPU_CTX_QUERY2_FLAGS_GUILTY;
 
+	/*query ue count*/
+	ras_counter = amdgpu_ras_query_error_count(adev, false);
+	/*ras counter is monotonic increasing*/
+	if (ras_counter != ctx->ras_counter_ue) {
+		out->state.flags |= AMDGPU_CTX_QUERY2_FLAGS_RAS_UE;
+		ctx->ras_counter_ue = ras_counter;
+	}
+
+	/*query ce count*/
+	ras_counter = amdgpu_ras_query_error_count(adev, true);
+	if (ras_counter != ctx->ras_counter_ce) {
+		out->state.flags |= AMDGPU_CTX_QUERY2_FLAGS_RAS_CE;
+		ctx->ras_counter_ce = ras_counter;
+	}
+
 	mutex_unlock(&mgr->lock);
 	return 0;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h
index b3b012c0a7da..8e561daa64cb 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h
@@ -49,6 +49,8 @@ struct amdgpu_ctx {
 	enum drm_sched_priority		override_priority;
 	struct mutex			lock;
 	atomic_t			guilty;
+	uint32_t			ras_counter_ce;
+	uint32_t			ras_counter_ue;
 };
 
 struct amdgpu_ctx_mgr {
diff --git a/include/uapi/drm/amdgpu_drm.h b/include/uapi/drm/amdgpu_drm.h
index e5275d4481f5..722598b25f37 100644
--- a/include/uapi/drm/amdgpu_drm.h
+++ b/include/uapi/drm/amdgpu_drm.h
@@ -210,6 +210,9 @@ union drm_amdgpu_bo_list {
 #define AMDGPU_CTX_QUERY2_FLAGS_VRAMLOST (1<<1)
 /* indicate some job from this context once cause gpu hang */
 #define AMDGPU_CTX_QUERY2_FLAGS_GUILTY   (1<<2)
+/* indicate some errors are detected by RAS */
+#define AMDGPU_CTX_QUERY2_FLAGS_RAS_CE   (1<<3)
+#define AMDGPU_CTX_QUERY2_FLAGS_RAS_UE   (1<<4)
 
 /* Context priority level */
 #define AMDGPU_CTX_PRIORITY_UNSET       -2048
-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 16/20] drm/amdgpu: add ioctl query for enabled ras features (v2)

[Alex Deucher]

From: xinhui pan <xinhui.pan@amd.com>

Add a query for userspace to check which RAS features
are enabled.

v2: squash in warning fix

Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 10 ++++++++
 include/uapi/drm/amdgpu_drm.h           | 31 +++++++++++++++++++++++++
 2 files changed, 41 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index 4d0ccca891cb..df40f35db89c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -39,6 +39,7 @@
 #include "amdgpu_amdkfd.h"
 #include "amdgpu_gem.h"
 #include "amdgpu_display.h"
+#include "amdgpu_ras.h"
 
 static void amdgpu_unregister_gpu_instance(struct amdgpu_device *adev)
 {
@@ -919,6 +920,15 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 	case AMDGPU_INFO_VRAM_LOST_COUNTER:
 		ui32 = atomic_read(&adev->vram_lost_counter);
 		return copy_to_user(out, &ui32, min(size, 4u)) ? -EFAULT : 0;
+	case AMDGPU_INFO_RAS_ENABLED_FEATURES: {
+		struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
+
+		if (!ras)
+			return -EINVAL;
+		return copy_to_user(out, &ras->features,
+				min_t(u32, size, sizeof(ras->features))) ?
+			-EFAULT : 0;
+	}
 	default:
 		DRM_DEBUG_KMS("Invalid request %d\n", info->query);
 		return -EINVAL;
diff --git a/include/uapi/drm/amdgpu_drm.h b/include/uapi/drm/amdgpu_drm.h
index 722598b25f37..e3a97da4add9 100644
--- a/include/uapi/drm/amdgpu_drm.h
+++ b/include/uapi/drm/amdgpu_drm.h
@@ -737,6 +737,37 @@ struct drm_amdgpu_cs_chunk_data {
 /* Number of VRAM page faults on CPU access. */
 #define AMDGPU_INFO_NUM_VRAM_CPU_PAGE_FAULTS	0x1E
 #define AMDGPU_INFO_VRAM_LOST_COUNTER		0x1F
+/* query ras mask of enabled features*/
+#define AMDGPU_INFO_RAS_ENABLED_FEATURES	0x20
+
+/* RAS MASK: UMC (VRAM) */
+#define AMDGPU_INFO_RAS_ENABLED_UMC			(1 << 0)
+/* RAS MASK: SDMA */
+#define AMDGPU_INFO_RAS_ENABLED_SDMA			(1 << 1)
+/* RAS MASK: GFX */
+#define AMDGPU_INFO_RAS_ENABLED_GFX			(1 << 2)
+/* RAS MASK: MMHUB */
+#define AMDGPU_INFO_RAS_ENABLED_MMHUB			(1 << 3)
+/* RAS MASK: ATHUB */
+#define AMDGPU_INFO_RAS_ENABLED_ATHUB			(1 << 4)
+/* RAS MASK: PCIE */
+#define AMDGPU_INFO_RAS_ENABLED_PCIE			(1 << 5)
+/* RAS MASK: HDP */
+#define AMDGPU_INFO_RAS_ENABLED_HDP			(1 << 6)
+/* RAS MASK: XGMI */
+#define AMDGPU_INFO_RAS_ENABLED_XGMI			(1 << 7)
+/* RAS MASK: DF */
+#define AMDGPU_INFO_RAS_ENABLED_DF			(1 << 8)
+/* RAS MASK: SMN */
+#define AMDGPU_INFO_RAS_ENABLED_SMN			(1 << 9)
+/* RAS MASK: SEM */
+#define AMDGPU_INFO_RAS_ENABLED_SEM			(1 << 10)
+/* RAS MASK: MP0 */
+#define AMDGPU_INFO_RAS_ENABLED_MP0			(1 << 11)
+/* RAS MASK: MP1 */
+#define AMDGPU_INFO_RAS_ENABLED_MP1			(1 << 12)
+/* RAS MASK: FUSE */
+#define AMDGPU_INFO_RAS_ENABLED_FUSE			(1 << 13)
 
 #define AMDGPU_INFO_MMR_SE_INDEX_SHIFT	0
 #define AMDGPU_INFO_MMR_SE_INDEX_MASK	0xff
-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 17/20] drm/amdgpu: skip gpu reset when ras error occured

[Alex Deucher]

From: xinhui pan <xinhui.pan@amd.com>

gpu reset is not stable on vega20 A1.

Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
index 599dcd01aec7..02cb9a13ddc5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
@@ -182,10 +182,13 @@ int amdgpu_ras_reserve_bad_pages(struct amdgpu_device *adev);
 static inline int amdgpu_ras_reset_gpu(struct amdgpu_device *adev,
 		bool is_baco)
 {
+	/* remove me when gpu reset works on vega20 A1. */
+#if 0
 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
 
 	if (atomic_cmpxchg(&ras->in_recovery, 0, 1) == 0)
 		schedule_work(&ras->recovery_work);
+#endif
 	return 0;
 }
 
-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 18/20] drm/amdgpu: add human readable debugfs control support (v2)

[Alex Deucher]

From: xinhui pan <xinhui.pan@amd.com>

Currently, the debugfs control node can't parse bash-like commands.
Now add such support for any tester that uses scripts.

v2: squash in fixes for input validation

Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 115 +++++++++++++++++++++---
 1 file changed, 102 insertions(+), 13 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index ca9c7d1ede2f..1df6b03a3680 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -200,14 +200,87 @@ static const struct file_operations amdgpu_ras_debugfs_ops = {
 	.llseek = default_llseek
 };
 
+static int amdgpu_ras_find_block_id_by_name(const char *name, int *block_id)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(ras_block_string); i++) {
+		*block_id = i;
+		if (strcmp(name, ras_block_str(i)) == 0)
+			return 0;
+	}
+	return -EINVAL;
+}
+
+static int amdgpu_ras_debugfs_ctrl_parse_data(struct file *f,
+		const char __user *buf, size_t size,
+		loff_t *pos, struct ras_debug_if *data)
+{
+	ssize_t s = min_t(u64, 64, size);
+	char str[65];
+	char block_name[33];
+	char err[9] = "ue";
+	int op = -1;
+	int block_id;
+	u64 address, value;
+
+	if (*pos)
+		return -EINVAL;
+	*pos = size;
+
+	memset(str, 0, sizeof(str));
+	memset(data, 0, sizeof(*data));
+
+	if (copy_from_user(str, buf, s))
+		return -EINVAL;
+
+	if (sscanf(str, "disable %32s", block_name) == 1)
+		op = 0;
+	else if (sscanf(str, "enable %32s %8s", block_name, err) == 2)
+		op = 1;
+	else if (sscanf(str, "inject %32s %8s", block_name, err) == 2)
+		op = 2;
+	else if (sscanf(str, "%32s", block_name) == 1)
+		/* ascii string, but commands are not matched. */
+		return -EINVAL;
+
+	if (op != -1) {
+		if (amdgpu_ras_find_block_id_by_name(block_name, &block_id))
+			return -EINVAL;
+
+		data->head.block = block_id;
+		data->head.type = memcmp("ue", err, 2) == 0 ?
+			AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE :
+			AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE;
+		data->op = op;
+
+		if (op == 2) {
+			if (sscanf(str, "%*s %*s %*s %llu %llu",
+						&address, &value) != 2)
+				if (sscanf(str, "%*s %*s %*s 0x%llx 0x%llx",
+							&address, &value) != 2)
+					return -EINVAL;
+			data->inject.address = address;
+			data->inject.value = value;
+		}
+	} else {
+		if (size < sizeof(data))
+			return -EINVAL;
+
+		if (copy_from_user(data, buf, sizeof(*data)))
+			return -EINVAL;
+	}
+
+	return 0;
+}
 /*
  * DOC: ras debugfs control interface
  *
  * It accepts struct ras_debug_if who has two members.
  *
  * First member: ras_debug_if::head or ras_debug_if::inject.
- * It is used to indicate which IP block will be under control.
- * Its contents are not human readable, IOW, write it by your programs.
+ *
+ * head is used to indicate which IP block will be under control.
  *
  * head has four members, they are block, type, sub_block_index, name.
  * block: which IP will be under control.
@@ -225,6 +298,28 @@ static const struct file_operations amdgpu_ras_debugfs_ops = {
  *  1: enable RAS on the block. Take ::head as its data.
  *  2: inject errors on the block. Take ::inject as its data.
  *
+ * How to use the interface?
+ * programs:
+ * copy the struct ras_debug_if in your codes and initialize it.
+ * write the struct to the control node.
+ *
+ * bash:
+ * echo op block [error [address value]] > .../ras/ras_ctrl
+ *	op: disable, enable, inject
+ *		disable: only block is needed
+ *		enable: block and error are needed
+ *		inject: error, address, value are needed
+ *	block: umc, smda, gfx, .........
+ *		see ras_block_string[] for details
+ *	error: ue, ce
+ *		ue: multi_uncorrectable
+ *		ce: single_correctable
+ *
+ * here are some examples for bash commands,
+ *	echo inject umc ue 0x0 0x0 > /sys/kernel/debug/dri/0/ras/ras_ctrl
+ *	echo inject umc ce 0 0 > /sys/kernel/debug/dri/0/ras/ras_ctrl
+ *	echo disable umc > /sys/kernel/debug/dri/0/ras/ras_ctrl
+ *
  * How to check the result?
  *
  * For disable/enable, please check ras features at
@@ -243,19 +338,10 @@ static ssize_t amdgpu_ras_debugfs_ctrl_write(struct file *f, const char __user *
 	struct ras_debug_if data;
 	int ret = 0;
 
-	if (size < sizeof(data))
-		return -EINVAL;
-
-	memset(&data, 0, sizeof(data));
-
-	if (*pos)
-		return -EINVAL;
-
-	if (copy_from_user(&data, buf, sizeof(data)))
+	ret = amdgpu_ras_debugfs_ctrl_parse_data(f, buf, size, pos, &data);
+	if (ret)
 		return -EINVAL;
 
-	*pos = size;
-
 	if (!amdgpu_ras_is_supported(adev, data.head.block))
 		return -EINVAL;
 
@@ -269,6 +355,9 @@ static ssize_t amdgpu_ras_debugfs_ctrl_write(struct file *f, const char __user *
 	case 2:
 		ret = amdgpu_ras_error_inject(adev, &data.inject);
 		break;
+	default:
+		ret = -EINVAL;
+		break;
 	};
 
 	if (ret)
-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 19/20] drm/amdkfd: add RAS capabilities in topology for Vega20 (v2)

[Alex Deucher]

From: Eric Huang <JinhuiEric.Huang@amd.com>

It is to collaborate with HSA_CAPABILITY in libhsakmt.

v2: squash in NULL pointer check

Signed-off-by: Eric Huang <JinhuiEric.Huang@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 16 ++++++++++++++++
 drivers/gpu/drm/amd/amdkfd/kfd_topology.h |  4 ++++
 2 files changed, 20 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
index 09da91644f9f..2cb09e088dce 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
@@ -37,6 +37,7 @@
 #include "kfd_device_queue_manager.h"
 #include "kfd_iommu.h"
 #include "amdgpu_amdkfd.h"
+#include "amdgpu_ras.h"
 
 /* topology_device_list - Master list of all topology devices */
 static struct list_head topology_device_list;
@@ -1197,6 +1198,7 @@ int kfd_topology_add_device(struct kfd_dev *gpu)
 	void *crat_image = NULL;
 	size_t image_size = 0;
 	int proximity_domain;
+	struct amdgpu_ras *ctx;
 
 	INIT_LIST_HEAD(&temp_topology_device_list);
 
@@ -1328,6 +1330,20 @@ int kfd_topology_add_device(struct kfd_dev *gpu)
 		dev->node_props.capability |= HSA_CAP_ATS_PRESENT;
 	}
 
+	ctx = amdgpu_ras_get_context((struct amdgpu_device *)(dev->gpu->kgd));
+	if (ctx) {
+		/* kfd only concerns sram ecc on GFX/SDMA and HBM ecc on UMC */
+		dev->node_props.capability |=
+			(((ctx->features & BIT(AMDGPU_RAS_BLOCK__SDMA)) != 0) ||
+			 ((ctx->features & BIT(AMDGPU_RAS_BLOCK__GFX)) != 0)) ?
+			HSA_CAP_SRAM_EDCSUPPORTED : 0;
+		dev->node_props.capability |= ((ctx->features & BIT(AMDGPU_RAS_BLOCK__UMC)) != 0) ?
+			HSA_CAP_MEM_EDCSUPPORTED : 0;
+
+		dev->node_props.capability |= (ctx->features != 0) ?
+			HSA_CAP_RASEVENTNOTIFY : 0;
+	}
+
 	kfd_debug_print_topology();
 
 	if (!res)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.h b/drivers/gpu/drm/amd/amdkfd/kfd_topology.h
index 92a19be07344..84710cfd23c2 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.h
@@ -48,6 +48,10 @@
 #define HSA_CAP_DOORBELL_TYPE_2_0		0x2
 #define HSA_CAP_AQL_QUEUE_DOUBLE_MAP		0x00004000
 
+#define HSA_CAP_SRAM_EDCSUPPORTED		0x00080000
+#define HSA_CAP_MEM_EDCSUPPORTED		0x00100000
+#define HSA_CAP_RASEVENTNOTIFY			0x00200000
+
 struct kfd_node_properties {
 	uint64_t hive_id;
 	uint32_t cpu_cores_count;
-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 20/20] drm/amdkfd: add RAS ECC event support (v2)

[Alex Deucher]

From: Eric Huang <JinhuiEric.Huang@amd.com>

RAS ECC event will combine with GPU reset event, due to
ECC interrupts are caused by uncorrectable error that triggers
GPU reset.

v2: Fix misleading-indentation warning

Signed-off-by: Eric Huang <JinhuiEric.Huang@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h |  1 +
 drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c      |  1 +
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c      |  1 +
 drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c     |  2 ++
 drivers/gpu/drm/amd/amdkfd/kfd_device.c    | 11 +++++++++++
 drivers/gpu/drm/amd/amdkfd/kfd_events.c    | 18 +++++++++++++++++-
 drivers/gpu/drm/amd/amdkfd/kfd_priv.h      |  3 +++
 include/uapi/linux/kfd_ioctl.h             | 12 +++++++++++-
 8 files changed, 47 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
index 0e1711a75b68..e6a503760b62 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
@@ -229,5 +229,6 @@ int kgd2kfd_quiesce_mm(struct mm_struct *mm);
 int kgd2kfd_resume_mm(struct mm_struct *mm);
 int kgd2kfd_schedule_evict_and_restore_process(struct mm_struct *mm,
 					       struct dma_fence *fence);
+void kgd2kfd_set_sram_ecc_flag(struct kfd_dev *kfd);
 
 #endif /* AMDGPU_AMDKFD_H_INCLUDED */
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
index 88c45f990f05..6bb71f6ee18e 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
@@ -4805,6 +4805,7 @@ static int gfx_v9_0_process_ras_data_cb(struct amdgpu_device *adev,
 		struct amdgpu_iv_entry *entry)
 {
 	/* TODO ue will trigger an interrupt. */
+	kgd2kfd_set_sram_ecc_flag(adev->kfd.dev);
 	amdgpu_ras_reset_gpu(adev, 0);
 	return AMDGPU_RAS_UE;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
index 1a164fe2dd9a..84904bd680df 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
@@ -354,6 +354,7 @@ static int gmc_v9_0_ecc_interrupt_state(struct amdgpu_device *adev,
 static int gmc_v9_0_process_ras_data_cb(struct amdgpu_device *adev,
 		struct amdgpu_iv_entry *entry)
 {
+	kgd2kfd_set_sram_ecc_flag(adev->kfd.dev);
 	amdgpu_ras_reset_gpu(adev, 0);
 	return AMDGPU_RAS_UE;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c
index 058b9daec514..f7a6fafd70ae 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c
@@ -1851,6 +1851,8 @@ static int sdma_v4_0_process_ras_data_cb(struct amdgpu_device *adev,
 		return 0;
 	}
 
+	kgd2kfd_set_sram_ecc_flag(adev->kfd.dev);
+
 	amdgpu_ras_reset_gpu(adev, 0);
 
 	return AMDGPU_RAS_UE;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 8be9677c0c07..b3cdbf79f47b 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -466,6 +466,8 @@ struct kfd_dev *kgd2kfd_probe(struct kgd_dev *kgd,
 	memset(&kfd->doorbell_available_index, 0,
 		sizeof(kfd->doorbell_available_index));
 
+	atomic_set(&kfd->sram_ecc_flag, 0);
+
 	return kfd;
 }
 
@@ -661,6 +663,9 @@ int kgd2kfd_post_reset(struct kfd_dev *kfd)
 		return ret;
 	count = atomic_dec_return(&kfd_locked);
 	WARN_ONCE(count != 0, "KFD reset ref. error");
+
+	atomic_set(&kfd->sram_ecc_flag, 0);
+
 	return 0;
 }
 
@@ -1024,6 +1029,12 @@ int kfd_gtt_sa_free(struct kfd_dev *kfd, struct kfd_mem_obj *mem_obj)
 	return 0;
 }
 
+void kgd2kfd_set_sram_ecc_flag(struct kfd_dev *kfd)
+{
+	if (kfd)
+		atomic_inc(&kfd->sram_ecc_flag);
+}
+
 #if defined(CONFIG_DEBUG_FS)
 
 /* This function will send a package to HIQ to hang the HWS
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_events.c
index e9f0e0a1b41c..6e1d41c5bf86 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_events.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_events.c
@@ -1011,25 +1011,41 @@ void kfd_signal_vm_fault_event(struct kfd_dev *dev, unsigned int pasid,
 void kfd_signal_reset_event(struct kfd_dev *dev)
 {
 	struct kfd_hsa_hw_exception_data hw_exception_data;
+	struct kfd_hsa_memory_exception_data memory_exception_data;
 	struct kfd_process *p;
 	struct kfd_event *ev;
 	unsigned int temp;
 	uint32_t id, idx;
+	int reset_cause = atomic_read(&dev->sram_ecc_flag) ?
+			KFD_HW_EXCEPTION_ECC :
+			KFD_HW_EXCEPTION_GPU_HANG;
 
 	/* Whole gpu reset caused by GPU hang and memory is lost */
 	memset(&hw_exception_data, 0, sizeof(hw_exception_data));
 	hw_exception_data.gpu_id = dev->id;
 	hw_exception_data.memory_lost = 1;
+	hw_exception_data.reset_cause = reset_cause;
+
+	memset(&memory_exception_data, 0, sizeof(memory_exception_data));
+	memory_exception_data.ErrorType = KFD_MEM_ERR_SRAM_ECC;
+	memory_exception_data.gpu_id = dev->id;
+	memory_exception_data.failure.imprecise = true;
 
 	idx = srcu_read_lock(&kfd_processes_srcu);
 	hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) {
 		mutex_lock(&p->event_mutex);
 		id = KFD_FIRST_NONSIGNAL_EVENT_ID;
-		idr_for_each_entry_continue(&p->event_idr, ev, id)
+		idr_for_each_entry_continue(&p->event_idr, ev, id) {
 			if (ev->type == KFD_EVENT_TYPE_HW_EXCEPTION) {
 				ev->hw_exception_data = hw_exception_data;
 				set_event(ev);
 			}
+			if (ev->type == KFD_EVENT_TYPE_MEMORY &&
+			    reset_cause == KFD_HW_EXCEPTION_ECC) {
+				ev->memory_exception_data = memory_exception_data;
+				set_event(ev);
+			}
+		}
 		mutex_unlock(&p->event_mutex);
 	}
 	srcu_read_unlock(&kfd_processes_srcu, idx);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index 0eeee3c6d6dc..9e0230965675 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -276,6 +276,9 @@ struct kfd_dev {
 	uint64_t hive_id;
 
 	bool pci_atomic_requested;
+
+	/* SRAM ECC flag */
+	atomic_t sram_ecc_flag;
 };
 
 enum kfd_mempool {
diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h
index e622fd1fbd46..dc067ed0b72d 100644
--- a/include/uapi/linux/kfd_ioctl.h
+++ b/include/uapi/linux/kfd_ioctl.h
@@ -211,6 +211,11 @@ struct kfd_ioctl_dbg_wave_control_args {
 #define KFD_HW_EXCEPTION_GPU_HANG	0
 #define KFD_HW_EXCEPTION_ECC		1
 
+/* For kfd_hsa_memory_exception_data.ErrorType */
+#define KFD_MEM_ERR_NO_RAS		0
+#define KFD_MEM_ERR_SRAM_ECC		1
+#define KFD_MEM_ERR_POISON_CONSUMED	2
+#define KFD_MEM_ERR_GPU_HANG		3
 
 struct kfd_ioctl_create_event_args {
 	__u64 event_page_offset;	/* from KFD */
@@ -250,7 +255,12 @@ struct kfd_hsa_memory_exception_data {
 	struct kfd_memory_exception_failure failure;
 	__u64 va;
 	__u32 gpu_id;
-	__u32 pad;
+	__u32 ErrorType; /* 0 = no RAS error,
+			  * 1 = ECC_SRAM,
+			  * 2 = Link_SYNFLOOD (poison),
+			  * 3 = GPU hang (not attributable to a specific cause),
+			  * other values reserved
+			  */
 };
 
 /* hw exception data */
-- 
2.20.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx