Skip to contentSkip to conversation

drm/amdgpu: add umc_fill_error_record to make code more simple

PATCH In review Tao Zhou 2 participants 5 messages Jan 25, 2022 – Jan 26, 2022

Subsystems (2)
lore
Share & tools
lore

Tao Zhou

author
#0 lore raw

PATCH 1/4 drm/amdgpu: add umc_fill_error_record to make code more simple

Create common amdgpu_umc_fill_error_record function for all versions of UMC and clean up related codes.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 23 ++++------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 21 +++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h |  5 ++++
 drivers/gpu/drm/amd/amdgpu/umc_v6_1.c   | 18 +++----------
 drivers/gpu/drm/amd/amdgpu/umc_v6_7.c   | 36 +++++--------------------
 drivers/gpu/drm/amd/amdgpu/umc_v8_7.c   | 36 +++++--------------------
 6 files changed, 46 insertions(+), 93 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +5 -18
index d4e07d0acb66..e6324995fc54 100644
@@ -158,14 +158,9 @@ static int amdgpu_reserve_page_direct(struct amdgpu_device *adev, uint64_t addre
 	}
 
 	memset(&err_rec, 0x0, sizeof(struct eeprom_table_record));
-
-	err_rec.address = address;
-	err_rec.retired_page = address >> AMDGPU_GPU_PAGE_SHIFT;
-	err_rec.ts = (uint64_t)ktime_get_real_seconds();
-	err_rec.err_type = AMDGPU_RAS_EEPROM_ERR_NON_RECOVERABLE;
-
 	err_data.err_addr = &err_rec;
-	err_data.err_addr_cnt = 1;
+	amdgpu_umc_fill_error_record(&err_data, address,
+			(address >> AMDGPU_GPU_PAGE_SHIFT), 0, 0);
 
 	if (amdgpu_bad_page_threshold != 0) {
 		amdgpu_ras_add_bad_pages(adev, err_data.err_addr,
@@ -2660,8 +2655,6 @@ static int amdgpu_bad_page_notifier(struct notifier_block *nb,
 	dev_info(adev->dev, "Uncorrectable error detected in UMC inst: %d, chan_idx: %d",
 			     umc_inst, ch_inst);
 
-	memset(&err_rec, 0x0, sizeof(struct eeprom_table_record));
-
 	/*
 	 * Translate UMC channel address to Physical address
 	 */
@@ -2673,16 +2666,10 @@ static int amdgpu_bad_page_notifier(struct notifier_block *nb,
 			ADDR_OF_256B_BLOCK(channel_index) |
 			OFFSET_IN_256B_BLOCK(m->addr);
 
-	err_rec.address = m->addr;
-	err_rec.retired_page = retired_page >> AMDGPU_GPU_PAGE_SHIFT;
-	err_rec.ts = (uint64_t)ktime_get_real_seconds();
-	err_rec.err_type = AMDGPU_RAS_EEPROM_ERR_NON_RECOVERABLE;
-	err_rec.cu = 0;
-	err_rec.mem_channel = channel_index;
-	err_rec.mcumc_id = umc_inst;
-
+	memset(&err_rec, 0x0, sizeof(struct eeprom_table_record));
 	err_data.err_addr = &err_rec;
-	err_data.err_addr_cnt = 1;
+	amdgpu_umc_fill_error_record(&err_data, m->addr,
+			retired_page, channel_index, umc_inst);
 
 	if (amdgpu_bad_page_threshold != 0) {
 		amdgpu_ras_add_bad_pages(adev, err_data.err_addr,
drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +21 -0
index b4c68c09e071..ff7805beda38 100644
@@ -218,3 +218,24 @@ int amdgpu_umc_process_ecc_irq(struct amdgpu_device *adev,
 	amdgpu_ras_interrupt_dispatch(adev, &ih_data);
 	return 0;
 }
+
+void amdgpu_umc_fill_error_record(struct ras_err_data *err_data,
+		uint64_t err_addr,
+		uint64_t retired_page,
+		uint32_t channel_index,
+		uint32_t umc_inst)
+{
+	struct eeprom_table_record *err_rec =
+		&err_data->err_addr[err_data->err_addr_cnt];
+
+	err_rec->address = err_addr;
+	/* page frame address is saved */
+	err_rec->retired_page = retired_page >> AMDGPU_GPU_PAGE_SHIFT;
+	err_rec->ts = (uint64_t)ktime_get_real_seconds();
+	err_rec->err_type = AMDGPU_RAS_EEPROM_ERR_NON_RECOVERABLE;
+	err_rec->cu = 0;
+	err_rec->mem_channel = channel_index;
+	err_rec->mcumc_id = umc_inst;
+
+	err_data->err_addr_cnt++;
+}
drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h +5 -0
index 195740a6d97d..4db0526d0be4 100644
@@ -80,4 +80,9 @@ int amdgpu_umc_poison_handler(struct amdgpu_device *adev,
 int amdgpu_umc_process_ecc_irq(struct amdgpu_device *adev,
 		struct amdgpu_irq_src *source,
 		struct amdgpu_iv_entry *entry);
+void amdgpu_umc_fill_error_record(struct ras_err_data *err_data,
+		uint64_t err_addr,
+		uint64_t retired_page,
+		uint32_t channel_index,
+		uint32_t umc_inst);
 #endif
drivers/gpu/drm/amd/amdgpu/umc_v6_1.c +3 -15
index 4776301972d4..939cb203f7ad 100644
@@ -300,7 +300,6 @@ static void umc_v6_1_query_error_address(struct amdgpu_device *adev,
 {
 	uint32_t lsb, mc_umc_status_addr;
 	uint64_t mc_umc_status, err_addr, retired_page, mc_umc_addrt0;
-	struct eeprom_table_record *err_rec;
 	uint32_t channel_index = adev->umc.channel_idx_tbl[umc_inst * adev->umc.channel_inst_num + ch_inst];
 
 	if (adev->asic_type == CHIP_ARCTURUS) {
@@ -328,8 +327,6 @@ static void umc_v6_1_query_error_address(struct amdgpu_device *adev,
 		return;
 	}
 
-	err_rec = &err_data->err_addr[err_data->err_addr_cnt];
-
 	/* calculate error address if ue/ce error is detected */
 	if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
 	    (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 ||
@@ -348,18 +345,9 @@ static void umc_v6_1_query_error_address(struct amdgpu_device *adev,
 
 		/* we only save ue error information currently, ce is skipped */
 		if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC)
-				== 1) {
-			err_rec->address = err_addr;
-			/* page frame address is saved */
-			err_rec->retired_page = retired_page >> AMDGPU_GPU_PAGE_SHIFT;
-			err_rec->ts = (uint64_t)ktime_get_real_seconds();
-			err_rec->err_type = AMDGPU_RAS_EEPROM_ERR_NON_RECOVERABLE;
-			err_rec->cu = 0;
-			err_rec->mem_channel = channel_index;
-			err_rec->mcumc_id = umc_inst;
-
-			err_data->err_addr_cnt++;
-		}
+				== 1)
+			amdgpu_umc_fill_error_record(err_data, err_addr,
+					retired_page, channel_index, umc_inst);
 	}
 
 	/* clear umc status */
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +6 -30
index f5a1ba7db75a..300dee9ec6b4 100644
@@ -120,7 +120,6 @@ static void umc_v6_7_ecc_info_query_error_address(struct amdgpu_device *adev,
 					 uint32_t umc_inst)
 {
 	uint64_t mc_umc_status, err_addr, retired_page;
-	struct eeprom_table_record *err_rec;
 	uint32_t channel_index;
 	uint32_t eccinfo_table_idx;
 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
@@ -137,8 +136,6 @@ static void umc_v6_7_ecc_info_query_error_address(struct amdgpu_device *adev,
 	if (!err_data->err_addr)
 		return;
 
-	err_rec = &err_data->err_addr[err_data->err_addr_cnt];
-
 	/* calculate error address if ue/ce error is detected */
 	if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
 	    (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 ||
@@ -154,18 +151,9 @@ static void umc_v6_7_ecc_info_query_error_address(struct amdgpu_device *adev,
 
 		/* we only save ue error information currently, ce is skipped */
 		if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC)
-				== 1) {
-			err_rec->address = err_addr;
-			/* page frame address is saved */
-			err_rec->retired_page = retired_page >> AMDGPU_GPU_PAGE_SHIFT;
-			err_rec->ts = (uint64_t)ktime_get_real_seconds();
-			err_rec->err_type = AMDGPU_RAS_EEPROM_ERR_NON_RECOVERABLE;
-			err_rec->cu = 0;
-			err_rec->mem_channel = channel_index;
-			err_rec->mcumc_id = umc_inst;
-
-			err_data->err_addr_cnt++;
-		}
+				== 1)
+			amdgpu_umc_fill_error_record(err_data, err_addr,
+					retired_page, channel_index, umc_inst);
 	}
 }
 
@@ -345,7 +333,6 @@ static void umc_v6_7_query_error_address(struct amdgpu_device *adev,
 {
 	uint32_t mc_umc_status_addr;
 	uint64_t mc_umc_status, err_addr, retired_page, mc_umc_addrt0;
-	struct eeprom_table_record *err_rec;
 	uint32_t channel_index;
 
 	mc_umc_status_addr =
@@ -364,8 +351,6 @@ static void umc_v6_7_query_error_address(struct amdgpu_device *adev,
 		return;
 	}
 
-	err_rec = &err_data->err_addr[err_data->err_addr_cnt];
-
 	channel_index =
 		adev->umc.channel_idx_tbl[umc_inst * adev->umc.channel_inst_num + ch_inst];
 
@@ -384,18 +369,9 @@ static void umc_v6_7_query_error_address(struct amdgpu_device *adev,
 
 		/* we only save ue error information currently, ce is skipped */
 		if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC)
-				== 1) {
-			err_rec->address = err_addr;
-			/* page frame address is saved */
-			err_rec->retired_page = retired_page >> AMDGPU_GPU_PAGE_SHIFT;
-			err_rec->ts = (uint64_t)ktime_get_real_seconds();
-			err_rec->err_type = AMDGPU_RAS_EEPROM_ERR_NON_RECOVERABLE;
-			err_rec->cu = 0;
-			err_rec->mem_channel = channel_index;
-			err_rec->mcumc_id = umc_inst;
-
-			err_data->err_addr_cnt++;
-		}
+				== 1)
+			amdgpu_umc_fill_error_record(err_data, err_addr,
+					retired_page, channel_index, umc_inst);
 	}
 
 	/* clear umc status */
drivers/gpu/drm/amd/amdgpu/umc_v8_7.c +6 -30
index d70417196662..de85a998ef99 100644
@@ -121,7 +121,6 @@ static void umc_v8_7_ecc_info_query_error_address(struct amdgpu_device *adev,
 					uint32_t umc_inst)
 {
 	uint64_t mc_umc_status, err_addr, retired_page;
-	struct eeprom_table_record *err_rec;
 	uint32_t channel_index;
 	uint32_t eccinfo_table_idx;
 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
@@ -138,8 +137,6 @@ static void umc_v8_7_ecc_info_query_error_address(struct amdgpu_device *adev,
 	if (!err_data->err_addr)
 		return;
 
-	err_rec = &err_data->err_addr[err_data->err_addr_cnt];
-
 	/* calculate error address if ue/ce error is detected */
 	if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
 	    (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 ||
@@ -155,18 +152,9 @@ static void umc_v8_7_ecc_info_query_error_address(struct amdgpu_device *adev,
 
 		/* we only save ue error information currently, ce is skipped */
 		if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC)
-				== 1) {
-			err_rec->address = err_addr;
-			/* page frame address is saved */
-			err_rec->retired_page = retired_page >> AMDGPU_GPU_PAGE_SHIFT;
-			err_rec->ts = (uint64_t)ktime_get_real_seconds();
-			err_rec->err_type = AMDGPU_RAS_EEPROM_ERR_NON_RECOVERABLE;
-			err_rec->cu = 0;
-			err_rec->mem_channel = channel_index;
-			err_rec->mcumc_id = umc_inst;
-
-			err_data->err_addr_cnt++;
-		}
+				== 1)
+			amdgpu_umc_fill_error_record(err_data, err_addr,
+					retired_page, channel_index, umc_inst);
 	}
 }
 
@@ -344,7 +332,6 @@ static void umc_v8_7_query_error_address(struct amdgpu_device *adev,
 {
 	uint32_t lsb, mc_umc_status_addr;
 	uint64_t mc_umc_status, err_addr, retired_page, mc_umc_addrt0;
-	struct eeprom_table_record *err_rec;
 	uint32_t channel_index = adev->umc.channel_idx_tbl[umc_inst * adev->umc.channel_inst_num + ch_inst];
 
 	mc_umc_status_addr =
@@ -363,8 +350,6 @@ static void umc_v8_7_query_error_address(struct amdgpu_device *adev,
 		return;
 	}
 
-	err_rec = &err_data->err_addr[err_data->err_addr_cnt];
-
 	/* calculate error address if ue/ce error is detected */
 	if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, Val) == 1 &&
 	    (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 ||
@@ -383,18 +368,9 @@ static void umc_v8_7_query_error_address(struct amdgpu_device *adev,
 
 		/* we only save ue error information currently, ce is skipped */
 		if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC)
-				== 1) {
-			err_rec->address = err_addr;
-			/* page frame address is saved */
-			err_rec->retired_page = retired_page >> AMDGPU_GPU_PAGE_SHIFT;
-			err_rec->ts = (uint64_t)ktime_get_real_seconds();
-			err_rec->err_type = AMDGPU_RAS_EEPROM_ERR_NON_RECOVERABLE;
-			err_rec->cu = 0;
-			err_rec->mem_channel = channel_index;
-			err_rec->mcumc_id = umc_inst;
-
-			err_data->err_addr_cnt++;
-		}
+				== 1)
+			amdgpu_umc_fill_error_record(err_data, err_addr,
+					retired_page, channel_index, umc_inst);
 	}
 
 	/* clear umc status */
Signature · 1 line
2.17.1
author
#1 lore raw

PATCH 2/4 drm/amdgpu: increase bad page number for umc ras query

One piece of umc normalizing address can be mapped to 16 pieces of physical address in each umc channel on ALDEBARAN.

---
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 3 ++-
 drivers/gpu/drm/amd/amdgpu/umc_v6_7.h | 4 ++++
 2 files changed, 6 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +2 -1
index 600ff658ab1b..4595027a8c63 100644
@@ -1213,7 +1213,8 @@ static void gmc_v9_0_set_umc_funcs(struct amdgpu_device *adev)
 		adev->umc.ras = &umc_v6_1_ras;
 		break;
 	case IP_VERSION(6, 7, 0):
-		adev->umc.max_ras_err_cnt_per_query = UMC_V6_7_TOTAL_CHANNEL_NUM;
+		adev->umc.max_ras_err_cnt_per_query =
+			UMC_V6_7_TOTAL_CHANNEL_NUM * UMC_V6_7_BAD_PAGE_NUM_PER_CHANNEL;
 		adev->umc.channel_inst_num = UMC_V6_7_CHANNEL_INSTANCE_NUM;
 		adev->umc.umc_inst_num = UMC_V6_7_UMC_INSTANCE_NUM;
 		adev->umc.channel_offs = UMC_V6_7_PER_CHANNEL_OFFSET;
drivers/gpu/drm/amd/amdgpu/umc_v6_7.h +4 -0
index 1f2edf625370..9adebcf98582 100644
@@ -41,6 +41,10 @@
 #define UMC_V6_7_CHANNEL_INSTANCE_NUM		8
 /* total channel instances in one umc block */
 #define UMC_V6_7_TOTAL_CHANNEL_NUM	(UMC_V6_7_CHANNEL_INSTANCE_NUM * UMC_V6_7_UMC_INSTANCE_NUM)
+/* one piece of normalizing address is mapped to 8 pieces of physical address */
+#define UMC_V6_7_NA_MAP_PA_NUM	8
+/* R14 bit shift should be considered, double the number */
+#define UMC_V6_7_BAD_PAGE_NUM_PER_CHANNEL	(UMC_V6_7_NA_MAP_PA_NUM * 2)
 /* UMC regiser per channel offset */
 #define UMC_V6_7_PER_CHANNEL_OFFSET		0x400
 extern struct amdgpu_umc_ras umc_v6_7_ras;
Signature · 1 line
2.17.1
author
#2 lore raw

PATCH 3/4 drm/amdgpu: update algorithm of umc address conversion

In reply to Tao Zhou

On ALDEBARAN, we need to traverse all column bits higher than BIT11(C4C3C2) in a row, the shift of R14 bit should be also taken into account. Retire all pages we find.

---
 drivers/gpu/drm/amd/amdgpu/umc_v6_7.c | 41 +++++++++++++++++++++------
 drivers/gpu/drm/amd/amdgpu/umc_v6_7.h |  4 +++
 2 files changed, 37 insertions(+), 8 deletions(-)
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +33 -8
index 300dee9ec6b4..1ecba7b5df1c 100644
@@ -119,7 +119,7 @@ static void umc_v6_7_ecc_info_query_error_address(struct amdgpu_device *adev,
 					 uint32_t ch_inst,
 					 uint32_t umc_inst)
 {
-	uint64_t mc_umc_status, err_addr, retired_page;
+	uint64_t mc_umc_status, err_addr, soc_pa, retired_page, column;
 	uint32_t channel_index;
 	uint32_t eccinfo_table_idx;
 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
@@ -145,15 +145,27 @@ static void umc_v6_7_ecc_info_query_error_address(struct amdgpu_device *adev,
 		err_addr = REG_GET_FIELD(err_addr, MCA_UMC_UMC0_MCUMC_ADDRT0, ErrorAddr);
 
 		/* translate umc channel address to soc pa, 3 parts are included */
-		retired_page = ADDR_OF_8KB_BLOCK(err_addr) |
+		soc_pa = ADDR_OF_8KB_BLOCK(err_addr) |
 				ADDR_OF_256B_BLOCK(channel_index) |
 				OFFSET_IN_256B_BLOCK(err_addr);
+		/* clear [C4 C3 C2] in soc physical address */
+		soc_pa &= ~(0x7ULL << UMC_V6_7_PA_C2_BIT);
 
 		/* we only save ue error information currently, ce is skipped */
 		if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC)
-				== 1)
-			amdgpu_umc_fill_error_record(err_data, err_addr,
+				== 1) {
+			/* loop for all possibilities of [C4 C3 C2] */
+			for (column = 0; column < UMC_V6_7_NA_MAP_PA_NUM; column++) {
+				retired_page = soc_pa | (column << UMC_V6_7_PA_C2_BIT);
+				amdgpu_umc_fill_error_record(err_data, err_addr,
 					retired_page, channel_index, umc_inst);
+
+				/* shift R14 bit */
+				retired_page ^= (0x1ULL << UMC_V6_7_PA_R14_BIT);
+				amdgpu_umc_fill_error_record(err_data, err_addr,
+					retired_page, channel_index, umc_inst);
+			}
+		}
 	}
 }
 
@@ -332,8 +344,9 @@ static void umc_v6_7_query_error_address(struct amdgpu_device *adev,
 					 uint32_t umc_inst)
 {
 	uint32_t mc_umc_status_addr;
-	uint64_t mc_umc_status, err_addr, retired_page, mc_umc_addrt0;
 	uint32_t channel_index;
+	uint64_t mc_umc_status, mc_umc_addrt0;
+	uint64_t err_addr, soc_pa, retired_page, column;
 
 	mc_umc_status_addr =
 		SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_STATUST0);
@@ -363,15 +376,27 @@ static void umc_v6_7_query_error_address(struct amdgpu_device *adev,
 		err_addr = REG_GET_FIELD(err_addr, MCA_UMC_UMC0_MCUMC_ADDRT0, ErrorAddr);
 
 		/* translate umc channel address to soc pa, 3 parts are included */
-		retired_page = ADDR_OF_8KB_BLOCK(err_addr) |
+		soc_pa = ADDR_OF_8KB_BLOCK(err_addr) |
 				ADDR_OF_256B_BLOCK(channel_index) |
 				OFFSET_IN_256B_BLOCK(err_addr);
+		/* clear [C4 C3 C2] in soc physical address */
+		soc_pa &= ~(0x7ULL << UMC_V6_7_PA_C2_BIT);
 
 		/* we only save ue error information currently, ce is skipped */
 		if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC)
-				== 1)
-			amdgpu_umc_fill_error_record(err_data, err_addr,
+				== 1) {
+			/* loop for all possibilities of [C4 C3 C2] */
+			for (column = 0; column < UMC_V6_7_NA_MAP_PA_NUM; column++) {
+				retired_page = soc_pa | (column << UMC_V6_7_PA_C2_BIT);
+				amdgpu_umc_fill_error_record(err_data, err_addr,
+					retired_page, channel_index, umc_inst);
+
+				/* shift R14 bit */
+				retired_page ^= (0x1ULL << UMC_V6_7_PA_R14_BIT);
+				amdgpu_umc_fill_error_record(err_data, err_addr,
 					retired_page, channel_index, umc_inst);
+			}
+		}
 	}
 
 	/* clear umc status */
drivers/gpu/drm/amd/amdgpu/umc_v6_7.h +4 -0
index 9adebcf98582..b67677867b45 100644
@@ -45,6 +45,10 @@
 #define UMC_V6_7_NA_MAP_PA_NUM	8
 /* R14 bit shift should be considered, double the number */
 #define UMC_V6_7_BAD_PAGE_NUM_PER_CHANNEL	(UMC_V6_7_NA_MAP_PA_NUM * 2)
+/* The C2 bit in SOC physical address */
+#define UMC_V6_7_PA_C2_BIT	17
+/* The R14 bit in SOC physical address */
+#define UMC_V6_7_PA_R14_BIT	34
 /* UMC regiser per channel offset */
 #define UMC_V6_7_PER_CHANNEL_OFFSET		0x400
 extern struct amdgpu_umc_ras umc_v6_7_ras;
Signature · 1 line
2.17.1
#3 lore raw

Re: [PATCH 3/4] drm/amdgpu: update algorithm of umc address conversion

Quoted 85 lines · Tao Zhou

On ALDEBARAN, we need to traverse all column bits higher than BIT11(C4C3C2) in a row, the shift of R14 bit should be also taken into account. Retire all pages we find.

---
  drivers/gpu/drm/amd/amdgpu/umc_v6_7.c | 41 +++++++++++++++++++++------
  drivers/gpu/drm/amd/amdgpu/umc_v6_7.h |  4 +++
  2 files changed, 37 insertions(+), 8 deletions(-)
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +26 -8
index 300dee9ec6b4..1ecba7b5df1c 100644
@@ -119,7 +119,7 @@ static void umc_v6_7_ecc_info_query_error_address(struct amdgpu_device *adev,
  					 uint32_t ch_inst,
  					 uint32_t umc_inst)
  {
-	uint64_t mc_umc_status, err_addr, retired_page;
+	uint64_t mc_umc_status, err_addr, soc_pa, retired_page, column;
  	uint32_t channel_index;
  	uint32_t eccinfo_table_idx;
  	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
@@ -145,15 +145,27 @@ static void umc_v6_7_ecc_info_query_error_address(struct amdgpu_device *adev,
  		err_addr = REG_GET_FIELD(err_addr, MCA_UMC_UMC0_MCUMC_ADDRT0, ErrorAddr);
 
  		/* translate umc channel address to soc pa, 3 parts are included */
-		retired_page = ADDR_OF_8KB_BLOCK(err_addr) |
+		soc_pa = ADDR_OF_8KB_BLOCK(err_addr) |
  				ADDR_OF_256B_BLOCK(channel_index) |
  				OFFSET_IN_256B_BLOCK(err_addr);
+		/* clear [C4 C3 C2] in soc physical address */
+		soc_pa &= ~(0x7ULL << UMC_V6_7_PA_C2_BIT);
 
  		/* we only save ue error information currently, ce is skipped */
  		if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC)
-				== 1)
-			amdgpu_umc_fill_error_record(err_data, err_addr,
+				== 1) {
+			/* loop for all possibilities of [C4 C3 C2] */
+			for (column = 0; column < UMC_V6_7_NA_MAP_PA_NUM; column++) {
+				retired_page = soc_pa | (column << UMC_V6_7_PA_C2_BIT);
+				amdgpu_umc_fill_error_record(err_data, err_addr,
  					retired_page, channel_index, umc_inst);
+
+				/* shift R14 bit */
+				retired_page ^= (0x1ULL << UMC_V6_7_PA_R14_BIT);
+				amdgpu_umc_fill_error_record(err_data, err_addr,
+					retired_page, channel_index, umc_inst);
+			}
+		}
  	}
  }
 
@@ -332,8 +344,9 @@ static void umc_v6_7_query_error_address(struct amdgpu_device *adev,
  					 uint32_t umc_inst)
  {
  	uint32_t mc_umc_status_addr;
-	uint64_t mc_umc_status, err_addr, retired_page, mc_umc_addrt0;
  	uint32_t channel_index;
+	uint64_t mc_umc_status, mc_umc_addrt0;
+	uint64_t err_addr, soc_pa, retired_page, column;
 
  	mc_umc_status_addr =
  		SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_STATUST0);
@@ -363,15 +376,27 @@ static void umc_v6_7_query_error_address(struct amdgpu_device *adev,
  		err_addr = REG_GET_FIELD(err_addr, MCA_UMC_UMC0_MCUMC_ADDRT0, ErrorAddr);
 
  		/* translate umc channel address to soc pa, 3 parts are included */
-		retired_page = ADDR_OF_8KB_BLOCK(err_addr) |
+		soc_pa = ADDR_OF_8KB_BLOCK(err_addr) |
  				ADDR_OF_256B_BLOCK(channel_index) |
  				OFFSET_IN_256B_BLOCK(err_addr);
+		/* clear [C4 C3 C2] in soc physical address */
+		soc_pa &= ~(0x7ULL << UMC_V6_7_PA_C2_BIT);
 
  		/* we only save ue error information currently, ce is skipped */
  		if (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC)
-				== 1)
-			amdgpu_umc_fill_error_record(err_data, err_addr,
+				== 1) {
+			/* loop for all possibilities of [C4 C3 C2] */
+			for (column = 0; column < UMC_V6_7_NA_MAP_PA_NUM; column++) {
+				retired_page = soc_pa | (column << UMC_V6_7_PA_C2_BIT);
+				amdgpu_umc_fill_error_record(err_data, err_addr,

drivers/gpu/drm/amd/amdgpu/umc_v6_7.c static void umc_v6_7_query_error_address(struct amdgpu_device *adev, · line 401

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +7 -0
+					retired_page, channel_index, umc_inst);
+
+				/* shift R14 bit */
+				retired_page ^= (0x1ULL << UMC_V6_7_PA_R14_BIT);
+				amdgpu_umc_fill_error_record(err_data, err_addr,
  					retired_page, channel_index, umc_inst);
+			}
+		}
  	}
 

Better to maintain the page decode logic in a single function.

umc_v6_7_save_bad_page_info(err_addr, channel)

Thanks,
Lijo

Quoted 17 lines
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +0 -0
  	/* clear umc status */
drivers/gpu/drm/amd/amdgpu/umc_v6_7.h +4 -0
index 9adebcf98582..b67677867b45 100644
@@ -45,6 +45,10 @@
  #define UMC_V6_7_NA_MAP_PA_NUM	8
  /* R14 bit shift should be considered, double the number */
  #define UMC_V6_7_BAD_PAGE_NUM_PER_CHANNEL	(UMC_V6_7_NA_MAP_PA_NUM * 2)
+/* The C2 bit in SOC physical address */
+#define UMC_V6_7_PA_C2_BIT	17
+/* The R14 bit in SOC physical address */
+#define UMC_V6_7_PA_R14_BIT	34
  /* UMC regiser per channel offset */
  #define UMC_V6_7_PER_CHANNEL_OFFSET		0x400
  extern struct amdgpu_umc_ras umc_v6_7_ras;
author
#4 lore raw

PATCH 4/4 drm/amdgpu: get hash bit for CH4 in umc channel index

In reply to Tao Zhou

On ALDEBARAN, the umc channel bits are not original values, they are hashed.

---
 drivers/gpu/drm/amd/amdgpu/umc_v6_7.c |  8 ++++++++
 drivers/gpu/drm/amd/amdgpu/umc_v6_7.h | 15 +++++++++++++++
 2 files changed, 23 insertions(+)
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +8 -0
index 1ecba7b5df1c..47452b61b615 100644
@@ -148,6 +148,10 @@ static void umc_v6_7_ecc_info_query_error_address(struct amdgpu_device *adev,
 		soc_pa = ADDR_OF_8KB_BLOCK(err_addr) |
 				ADDR_OF_256B_BLOCK(channel_index) |
 				OFFSET_IN_256B_BLOCK(err_addr);
+
+		/* The umc channel bits are not original values, they are hashed */
+		SET_CHANNEL_HASH(channel_index, soc_pa);
+
 		/* clear [C4 C3 C2] in soc physical address */
 		soc_pa &= ~(0x7ULL << UMC_V6_7_PA_C2_BIT);
 
@@ -379,6 +383,10 @@ static void umc_v6_7_query_error_address(struct amdgpu_device *adev,
 		soc_pa = ADDR_OF_8KB_BLOCK(err_addr) |
 				ADDR_OF_256B_BLOCK(channel_index) |
 				OFFSET_IN_256B_BLOCK(err_addr);
+
+		/* The umc channel bits are not original values, they are hashed */
+		SET_CHANNEL_HASH(channel_index, soc_pa);
+
 		/* clear [C4 C3 C2] in soc physical address */
 		soc_pa &= ~(0x7ULL << UMC_V6_7_PA_C2_BIT);
 
drivers/gpu/drm/amd/amdgpu/umc_v6_7.h +15 -0
index b67677867b45..e50f7692bd03 100644
@@ -45,12 +45,27 @@
 #define UMC_V6_7_NA_MAP_PA_NUM	8
 /* R14 bit shift should be considered, double the number */
 #define UMC_V6_7_BAD_PAGE_NUM_PER_CHANNEL	(UMC_V6_7_NA_MAP_PA_NUM * 2)
+/* The CH4 bit in SOC physical address */
+#define UMC_V6_7_PA_CH4_BIT	12
 /* The C2 bit in SOC physical address */
 #define UMC_V6_7_PA_C2_BIT	17
 /* The R14 bit in SOC physical address */
 #define UMC_V6_7_PA_R14_BIT	34
 /* UMC regiser per channel offset */
 #define UMC_V6_7_PER_CHANNEL_OFFSET		0x400
+
+/* XOR bit 20, 25, 34 of PA into CH4 bit (bit 12 of PA),
+ * hash bit is only effective when related setting is eanbled
+ */
+#define CHANNEL_HASH(channel_idx, pa) (((channel_idx) >> 4) ^ \
+			(((pa)  >> 20) & 0x1ULL & adev->df.hash_status.hash_64k) ^ \
+			(((pa)  >> 25) & 0x1ULL & adev->df.hash_status.hash_2m) ^ \
+			(((pa)  >> 34) & 0x1ULL & adev->df.hash_status.hash_1g))
+#define SET_CHANNEL_HASH(channel_idx, pa) do { \
+		(pa) &= ~(0x1ULL << UMC_V6_7_PA_CH4_BIT); \
+		(pa) |= (CHANNEL_HASH(channel_idx, pa) << UMC_V6_7_PA_CH4_BIT); \
+	} while (0)
+
 extern struct amdgpu_umc_ras umc_v6_7_ras;
 extern const uint32_t
 	umc_v6_7_channel_idx_tbl_second[UMC_V6_7_UMC_INSTANCE_NUM][UMC_V6_7_CHANNEL_INSTANCE_NUM];
Signature · 1 line
2.17.1