1 files changed, 86 insertions, 11 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index c8b4dd3ea5c3..2a6cf7963dde 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -150,6 +150,8 @@ static void amdgpu_ras_critical_region_fini(struct amdgpu_device *adev);
 
 #ifdef CONFIG_X86_MCE_AMD
 static void amdgpu_register_bad_pages_mca_notifier(struct amdgpu_device *adev);
+static void
+amdgpu_unregister_bad_pages_mca_notifier(struct amdgpu_device *adev);
 struct mce_notifier_adev_list {
 	struct amdgpu_device *devs[MAX_GPU_INSTANCE];
 	int num_gpu;
@@ -1661,7 +1663,7 @@ static int amdgpu_uniras_error_inject(struct amdgpu_device *adev,
 	inject_req.address = info->address;
 	inject_req.error_type = info->head.type;
 	inject_req.instance_mask = info->instance_mask;
-	inject_req.value = info->value;
+	inject_req.method = info->value;
 
 	return amdgpu_ras_mgr_handle_ras_cmd(adev, RAS_CMD__INJECT_ERROR,
 			&inject_req, sizeof(inject_req), &rsp, sizeof(rsp));
@@ -2921,8 +2923,12 @@ static void amdgpu_ras_do_recovery(struct work_struct *work)
 		type = amdgpu_ras_get_fatal_error_event(adev);
 		list_for_each_entry(remote_adev,
 				device_list_handle, gmc.xgmi.head) {
-			amdgpu_ras_query_err_status(remote_adev);
-			amdgpu_ras_log_on_err_counter(remote_adev, type);
+			if (amdgpu_uniras_enabled(remote_adev)) {
+				amdgpu_ras_mgr_update_ras_ecc(remote_adev);
+			} else {
+				amdgpu_ras_query_err_status(remote_adev);
+				amdgpu_ras_log_on_err_counter(remote_adev, type);
+			}
 		}
 
 	}
@@ -3010,8 +3016,13 @@ static int amdgpu_ras_mca2pa_by_idx(struct amdgpu_device *adev,
 	addr_in.ma.err_addr = bps->address;
 	addr_in.ma.socket_id = socket;
 	addr_in.ma.ch_inst = bps->mem_channel;
-	/* tell RAS TA the node instance is not used */
-	addr_in.ma.node_inst = TA_RAS_INV_NODE;
+	if (!amdgpu_ras_smu_eeprom_supported(adev)) {
+		/* tell RAS TA the node instance is not used */
+		addr_in.ma.node_inst = TA_RAS_INV_NODE;
+	} else {
+		addr_in.ma.umc_inst = bps->mcumc_id;
+		addr_in.ma.node_inst = bps->cu;
+	}
 
 	if (adev->umc.ras && adev->umc.ras->convert_ras_err_addr)
 		ret = adev->umc.ras->convert_ras_err_addr(adev, err_data,
@@ -3154,8 +3165,16 @@ static int __amdgpu_ras_convert_rec_from_rom(struct amdgpu_device *adev,
 	int i = 0;
 	enum amdgpu_memory_partition save_nps;
 
-	save_nps = (bps->retired_page >> UMC_NPS_SHIFT) & UMC_NPS_MASK;
-	bps->retired_page &= ~(UMC_NPS_MASK << UMC_NPS_SHIFT);
+	if (!amdgpu_ras_smu_eeprom_supported(adev)) {
+		save_nps = (bps->retired_page >> UMC_NPS_SHIFT) & UMC_NPS_MASK;
+		bps->retired_page &= ~(UMC_NPS_MASK << UMC_NPS_SHIFT);
+	} else {
+		/* if pmfw manages eeprom, save_nps is not stored on eeprom,
+		 * we should always convert mca address into physical address,
+		 * make save_nps different from nps
+		 */
+		save_nps = nps + 1;
+	}
 
 	if (save_nps == nps) {
 		if (amdgpu_umc_pages_in_a_row(adev, err_data,
@@ -3221,7 +3240,8 @@ int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev,
 
 	if (from_rom) {
 		/* there is no pa recs in V3, so skip pa recs processing */
-		if (control->tbl_hdr.version < RAS_TABLE_VER_V3) {
+		if ((control->tbl_hdr.version < RAS_TABLE_VER_V3) &&
+		    !amdgpu_ras_smu_eeprom_supported(adev)) {
 			for (i = 0; i < pages; i++) {
 				if (control->ras_num_recs - i >= adev->umc.retire_unit) {
 					if ((bps[i].address == bps[i + 1].address) &&
@@ -3291,7 +3311,13 @@ int amdgpu_ras_save_bad_pages(struct amdgpu_device *adev,
 	mutex_lock(&con->recovery_lock);
 	control = &con->eeprom_control;
 	data = con->eh_data;
-	unit_num = data->count / adev->umc.retire_unit - control->ras_num_recs;
+	if (amdgpu_ras_smu_eeprom_supported(adev))
+		unit_num = control->ras_num_recs -
+			control->ras_num_recs_old;
+	else
+		unit_num = data->count / adev->umc.retire_unit -
+			control->ras_num_recs;
+
 	save_count = con->bad_page_num - control->ras_num_bad_pages;
 	mutex_unlock(&con->recovery_lock);
 
@@ -3352,7 +3378,8 @@ static int amdgpu_ras_load_bad_pages(struct amdgpu_device *adev)
 			/*In V3, there is no pa recs, and some cases(when address==0) may be parsed
 			as pa recs, so add verion check to avoid it.
 			*/
-			if (control->tbl_hdr.version < RAS_TABLE_VER_V3) {
+			if ((control->tbl_hdr.version < RAS_TABLE_VER_V3) &&
+			    !amdgpu_ras_smu_eeprom_supported(adev)) {
 				for (i = 0; i < control->ras_num_recs; i++) {
 					if ((control->ras_num_recs - i) >= adev->umc.retire_unit) {
 						if ((bps[i].address == bps[i + 1].address) &&
@@ -3767,6 +3794,8 @@ int amdgpu_ras_init_badpage_info(struct amdgpu_device *adev)
 		return 0;
 
 	control = &con->eeprom_control;
+	con->ras_smu_drv = amdgpu_dpm_get_ras_smu_driver(adev);
+
 	ret = amdgpu_ras_eeprom_init(control);
 	control->is_eeprom_valid = !ret;
 
@@ -3927,7 +3956,9 @@ static int amdgpu_ras_recovery_fini(struct amdgpu_device *adev)
 	mutex_unlock(&con->recovery_lock);
 
 	amdgpu_ras_critical_region_init(adev);
-
+#ifdef CONFIG_X86_MCE_AMD
+	amdgpu_unregister_bad_pages_mca_notifier(adev);
+#endif
 	return 0;
 }
 /* recovery end */
@@ -4961,6 +4992,28 @@ static void amdgpu_register_bad_pages_mca_notifier(struct amdgpu_device *adev)
 		notifier_registered = true;
 	}
 }
+static void amdgpu_unregister_bad_pages_mca_notifier(struct amdgpu_device *adev)
+{
+	int i, j;
+
+	if (!notifier_registered && !mce_adev_list.num_gpu)
+		return;
+	for (i = 0, j = 0; i < mce_adev_list.num_gpu; i++) {
+		if (mce_adev_list.devs[i] == adev)
+			mce_adev_list.devs[i] = NULL;
+		if (!mce_adev_list.devs[i])
+			++j;
+	}
+
+	if (j == mce_adev_list.num_gpu) {
+		mce_adev_list.num_gpu = 0;
+		/* Unregister x86 notifier with MCE subsystem. */
+		if (notifier_registered) {
+			mce_unregister_decode_chain(&amdgpu_bad_page_nb);
+			notifier_registered = false;
+		}
+	}
+}
 #endif
 
 struct amdgpu_ras *amdgpu_ras_get_context(struct amdgpu_device *adev)
@@ -5671,3 +5724,25 @@ bool amdgpu_ras_check_critical_address(struct amdgpu_device *adev, uint64_t addr
 
 	return ret;
 }
+
+void amdgpu_ras_pre_reset(struct amdgpu_device *adev,
+					  struct list_head *device_list)
+{
+	struct amdgpu_device *tmp_adev = NULL;
+
+	list_for_each_entry(tmp_adev, device_list, reset_list) {
+		if (amdgpu_uniras_enabled(tmp_adev))
+			amdgpu_ras_mgr_pre_reset(tmp_adev);
+	}
+}
+
+void amdgpu_ras_post_reset(struct amdgpu_device *adev,
+					  struct list_head *device_list)
+{
+	struct amdgpu_device *tmp_adev = NULL;
+
+	list_for_each_entry(tmp_adev, device_list, reset_list) {
+		if (amdgpu_uniras_enabled(tmp_adev))
+			amdgpu_ras_mgr_post_reset(tmp_adev);
+	}
+}