1 files changed, 171 insertions, 31 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/umc_v8_10.c b/drivers/gpu/drm/amd/amdgpu/umc_v8_10.c
index da394bc06bba..fb55e8cb9967 100644
--- a/drivers/gpu/drm/amd/amdgpu/umc_v8_10.c
+++ b/drivers/gpu/drm/amd/amdgpu/umc_v8_10.c
@@ -209,6 +209,45 @@ static int umc_v8_10_swizzle_mode_na_to_pa(struct amdgpu_device *adev,
 	return 0;
 }
 
+static void umc_v8_10_convert_error_address(struct amdgpu_device *adev,
+					    struct ras_err_data *err_data, uint64_t err_addr,
+					    uint32_t ch_inst, uint32_t umc_inst,
+					    uint32_t node_inst, uint64_t mc_umc_status)
+{
+	uint64_t na_err_addr_base;
+	uint64_t na_err_addr, retired_page_addr;
+	uint32_t channel_index, addr_lsb, col = 0;
+	int ret = 0;
+
+	channel_index =
+		adev->umc.channel_idx_tbl[node_inst * adev->umc.umc_inst_num *
+					adev->umc.channel_inst_num +
+					umc_inst * adev->umc.channel_inst_num +
+					ch_inst];
+
+	/* the lowest lsb bits should be ignored */
+	addr_lsb = REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, AddrLsb);
+	err_addr &= ~((0x1ULL << addr_lsb) - 1);
+	na_err_addr_base = err_addr & ~(0x3ULL << UMC_V8_10_NA_C5_BIT);
+
+	/* loop for all possibilities of [C6 C5] in normal address. */
+	for (col = 0; col < UMC_V8_10_NA_COL_2BITS_POWER_OF_2_NUM; col++) {
+		na_err_addr = na_err_addr_base | (col << UMC_V8_10_NA_C5_BIT);
+
+		/* Mapping normal error address to retired soc physical address. */
+		ret = umc_v8_10_swizzle_mode_na_to_pa(adev, channel_index,
+						na_err_addr, &retired_page_addr);
+		if (ret) {
+			dev_err(adev->dev, "Failed to map pa from umc na.\n");
+			break;
+		}
+		dev_info(adev->dev, "Error Address(PA): 0x%llx\n",
+			retired_page_addr);
+		amdgpu_umc_fill_error_record(err_data, na_err_addr,
+				retired_page_addr, channel_index, umc_inst);
+	}
+}
+
 static void umc_v8_10_query_error_address(struct amdgpu_device *adev,
 					 struct ras_err_data *err_data,
 					 uint32_t umc_reg_offset,
@@ -218,10 +257,7 @@ static void umc_v8_10_query_error_address(struct amdgpu_device *adev,
 {
 	uint64_t mc_umc_status_addr;
 	uint64_t mc_umc_status, err_addr;
-	uint64_t mc_umc_addrt0, na_err_addr_base;
-	uint64_t na_err_addr, retired_page_addr;
-	uint32_t channel_index, addr_lsb, col = 0;
-	int ret = 0;
+	uint64_t mc_umc_addrt0;
 
 	mc_umc_status_addr =
 		SOC15_REG_OFFSET(UMC, 0, regMCA_UMC_UMC0_MCUMC_STATUST0);
@@ -236,12 +272,6 @@ static void umc_v8_10_query_error_address(struct amdgpu_device *adev,
 		return;
 	}
 
-	channel_index =
-		adev->umc.channel_idx_tbl[node_inst * adev->umc.umc_inst_num *
-					adev->umc.channel_inst_num +
-					umc_inst * adev->umc.channel_inst_num +
-					ch_inst];
-
 	/* calculate error address if ue 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, AddrV) == 1 &&
@@ -251,27 +281,8 @@ static void umc_v8_10_query_error_address(struct amdgpu_device *adev,
 		err_addr = RREG64_PCIE((mc_umc_addrt0 + umc_reg_offset) * 4);
 		err_addr = REG_GET_FIELD(err_addr, MCA_UMC_UMC0_MCUMC_ADDRT0, ErrorAddr);
 
-		/* the lowest lsb bits should be ignored */
-		addr_lsb = REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, AddrLsb);
-		err_addr &= ~((0x1ULL << addr_lsb) - 1);
-		na_err_addr_base = err_addr & ~(0x3ULL << UMC_V8_10_NA_C5_BIT);
-
-		/* loop for all possibilities of [C6 C5] in normal address. */
-		for (col = 0; col < UMC_V8_10_NA_COL_2BITS_POWER_OF_2_NUM; col++) {
-			na_err_addr = na_err_addr_base | (col << UMC_V8_10_NA_C5_BIT);
-
-			/* Mapping normal error address to retired soc physical address. */
-			ret = umc_v8_10_swizzle_mode_na_to_pa(adev, channel_index,
-							na_err_addr, &retired_page_addr);
-			if (ret) {
-				dev_err(adev->dev, "Failed to map pa from umc na.\n");
-				break;
-			}
-			dev_info(adev->dev, "Error Address(PA): 0x%llx\n",
-				retired_page_addr);
-			amdgpu_umc_fill_error_record(err_data, na_err_addr,
-					retired_page_addr, channel_index, umc_inst);
-		}
+		umc_v8_10_convert_error_address(adev, err_data, err_addr,
+					ch_inst, umc_inst, node_inst, mc_umc_status);
 	}
 
 	/* clear umc status */
@@ -349,6 +360,133 @@ static bool umc_v8_10_query_ras_poison_mode(struct amdgpu_device *adev)
 	return true;
 }
 
+static void umc_v8_10_ecc_info_query_correctable_error_count(struct amdgpu_device *adev,
+				      uint32_t node_inst, uint32_t umc_inst, uint32_t ch_inst,
+				      unsigned long *error_count)
+{
+	uint64_t mc_umc_status;
+	uint32_t eccinfo_table_idx;
+	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
+
+	eccinfo_table_idx = node_inst * adev->umc.umc_inst_num *
+				  adev->umc.channel_inst_num +
+				  umc_inst * adev->umc.channel_inst_num +
+				  ch_inst;
+
+	/* check the MCUMC_STATUS */
+	mc_umc_status = ras->umc_ecc.ecc[eccinfo_table_idx].mca_umc_status;
+	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, CECC) == 1) {
+		*error_count += 1;
+	}
+}
+
+static void umc_v8_10_ecc_info_query_uncorrectable_error_count(struct amdgpu_device *adev,
+				      uint32_t node_inst, uint32_t umc_inst, uint32_t ch_inst,
+				      unsigned long *error_count)
+{
+	uint64_t mc_umc_status;
+	uint32_t eccinfo_table_idx;
+	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
+
+	eccinfo_table_idx = node_inst * adev->umc.umc_inst_num *
+				  adev->umc.channel_inst_num +
+				  umc_inst * adev->umc.channel_inst_num +
+				  ch_inst;
+
+	/* check the MCUMC_STATUS */
+	mc_umc_status = ras->umc_ecc.ecc[eccinfo_table_idx].mca_umc_status;
+	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, Deferred) == 1 ||
+	    REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1 ||
+	    REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, PCC) == 1 ||
+	    REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UC) == 1 ||
+	    REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, TCC) == 1)) {
+		*error_count += 1;
+	}
+}
+
+static void umc_v8_10_ecc_info_query_ras_error_count(struct amdgpu_device *adev,
+					void *ras_error_status)
+{
+	struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
+
+	uint32_t node_inst       = 0;
+	uint32_t umc_inst        = 0;
+	uint32_t ch_inst         = 0;
+
+	/* TODO: driver needs to toggle DF Cstate to ensure
+	 * safe access of UMC registers. Will add the protection
+	 */
+	LOOP_UMC_EACH_NODE_INST_AND_CH(node_inst, umc_inst, ch_inst) {
+		umc_v8_10_ecc_info_query_correctable_error_count(adev,
+							node_inst, umc_inst, ch_inst,
+							&(err_data->ce_count));
+		umc_v8_10_ecc_info_query_uncorrectable_error_count(adev,
+							node_inst, umc_inst, ch_inst,
+							&(err_data->ue_count));
+	}
+}
+
+static void umc_v8_10_ecc_info_query_error_address(struct amdgpu_device *adev,
+					struct ras_err_data *err_data,
+					uint32_t ch_inst,
+					uint32_t umc_inst,
+					uint32_t node_inst)
+{
+	uint32_t eccinfo_table_idx;
+	uint64_t mc_umc_status, err_addr;
+
+	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
+
+	eccinfo_table_idx = node_inst * adev->umc.umc_inst_num *
+				  adev->umc.channel_inst_num +
+				  umc_inst * adev->umc.channel_inst_num +
+				  ch_inst;
+
+	mc_umc_status = ras->umc_ecc.ecc[eccinfo_table_idx].mca_umc_status;
+
+	if (mc_umc_status == 0)
+		return;
+
+	if (!err_data->err_addr)
+		return;
+
+	/* calculate error address if ue 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, AddrV) == 1 &&
+	    (REG_GET_FIELD(mc_umc_status, MCA_UMC_UMC0_MCUMC_STATUST0, UECC) == 1)) {
+
+		err_addr = ras->umc_ecc.ecc[eccinfo_table_idx].mca_umc_addr;
+		err_addr = REG_GET_FIELD(err_addr, MCA_UMC_UMC0_MCUMC_ADDRT0, ErrorAddr);
+
+		umc_v8_10_convert_error_address(adev, err_data, err_addr,
+					ch_inst, umc_inst, node_inst, mc_umc_status);
+	}
+}
+
+static void umc_v8_10_ecc_info_query_ras_error_address(struct amdgpu_device *adev,
+					void *ras_error_status)
+{
+	struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
+
+	uint32_t node_inst       = 0;
+	uint32_t umc_inst        = 0;
+	uint32_t ch_inst         = 0;
+
+	/* TODO: driver needs to toggle DF Cstate to ensure
+	 * safe access of UMC resgisters. Will add the protection
+	 * when firmware interface is ready
+	 */
+	LOOP_UMC_EACH_NODE_INST_AND_CH(node_inst, umc_inst, ch_inst) {
+		umc_v8_10_ecc_info_query_error_address(adev,
+						err_data,
+						ch_inst,
+						umc_inst,
+						node_inst);
+	}
+}
+
 const struct amdgpu_ras_block_hw_ops umc_v8_10_ras_hw_ops = {
 	.query_ras_error_count = umc_v8_10_query_ras_error_count,
 	.query_ras_error_address = umc_v8_10_query_ras_error_address,
@@ -360,4 +498,6 @@ struct amdgpu_umc_ras umc_v8_10_ras = {
 	},
 	.err_cnt_init = umc_v8_10_err_cnt_init,
 	.query_ras_poison_mode = umc_v8_10_query_ras_poison_mode,
+	.ecc_info_query_ras_error_count = umc_v8_10_ecc_info_query_ras_error_count,
+	.ecc_info_query_ras_error_address = umc_v8_10_ecc_info_query_ras_error_address,
 };