1 files changed, 218 insertions, 72 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
index 20436f81856a..c92b8794aa73 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
@@ -21,10 +21,13 @@
  *
  */
 
+#include <linux/sort.h>
 #include "amdgpu.h"
 #include "umc_v6_7.h"
 #define MAX_UMC_POISON_POLLING_TIME_SYNC   20  //ms
 
+#define MAX_UMC_HASH_STRING_SIZE  256
+
 static int amdgpu_umc_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)
@@ -63,6 +66,8 @@ int amdgpu_umc_page_retirement_mca(struct amdgpu_device *adev,
 		goto out_fini_err_data;
 	}
 
+	err_data.err_addr_len = adev->umc.max_ras_err_cnt_per_query;
+
 	/*
 	 * Translate UMC channel address to Physical address
 	 */
@@ -73,7 +78,7 @@ int amdgpu_umc_page_retirement_mca(struct amdgpu_device *adev,
 
 	if (amdgpu_bad_page_threshold != 0) {
 		amdgpu_ras_add_bad_pages(adev, err_data.err_addr,
-						err_data.err_addr_cnt);
+						err_data.err_addr_cnt, false);
 		amdgpu_ras_save_bad_pages(adev, NULL);
 	}
 
@@ -86,7 +91,7 @@ out_fini_err_data:
 	return ret;
 }
 
-static void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev,
+void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev,
 			void *ras_error_status)
 {
 	struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
@@ -118,6 +123,8 @@ static void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev,
 			if(!err_data->err_addr)
 				dev_warn(adev->dev, "Failed to alloc memory for "
 						"umc error address record!\n");
+			else
+				err_data->err_addr_len = adev->umc.max_ras_err_cnt_per_query;
 
 			/* umc query_ras_error_address is also responsible for clearing
 			 * error status
@@ -143,6 +150,8 @@ static void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev,
 			if(!err_data->err_addr)
 				dev_warn(adev->dev, "Failed to alloc memory for "
 						"umc error address record!\n");
+			else
+				err_data->err_addr_len = adev->umc.max_ras_err_cnt_per_query;
 
 			/* umc query_ras_error_address is also responsible for clearing
 			 * error status
@@ -157,10 +166,11 @@ static void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev,
 		if ((amdgpu_bad_page_threshold != 0) &&
 			err_data->err_addr_cnt) {
 			amdgpu_ras_add_bad_pages(adev, err_data->err_addr,
-						err_data->err_addr_cnt);
+						err_data->err_addr_cnt, false);
 			amdgpu_ras_save_bad_pages(adev, &err_count);
 
-			amdgpu_dpm_send_hbm_bad_pages_num(adev, con->eeprom_control.ras_num_recs);
+			amdgpu_dpm_send_hbm_bad_pages_num(adev,
+					con->eeprom_control.ras_num_bad_pages);
 
 			if (con->update_channel_flag == true) {
 				amdgpu_dpm_send_hbm_bad_channel_flag(adev, con->eeprom_control.bad_channel_bitmap);
@@ -170,6 +180,7 @@ static void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev,
 	}
 
 	kfree(err_data->err_addr);
+	err_data->err_addr = NULL;
 
 	mutex_unlock(&con->page_retirement_lock);
 }
@@ -177,7 +188,7 @@ static void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev,
 static int amdgpu_umc_do_page_retirement(struct amdgpu_device *adev,
 		void *ras_error_status,
 		struct amdgpu_iv_entry *entry,
-		bool reset)
+		uint32_t reset)
 {
 	struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
@@ -185,69 +196,18 @@ static int amdgpu_umc_do_page_retirement(struct amdgpu_device *adev,
 	kgd2kfd_set_sram_ecc_flag(adev->kfd.dev);
 	amdgpu_umc_handle_bad_pages(adev, ras_error_status);
 
-	if (err_data->ue_count && reset) {
-		/* use mode-2 reset for poison consumption */
-		if (!entry)
-			con->gpu_reset_flags |= AMDGPU_RAS_GPU_RESET_MODE2_RESET;
+	if ((err_data->ue_count || err_data->de_count) &&
+	    (reset || amdgpu_ras_is_rma(adev))) {
+		con->gpu_reset_flags |= reset;
 		amdgpu_ras_reset_gpu(adev);
 	}
 
 	return AMDGPU_RAS_SUCCESS;
 }
 
-int amdgpu_umc_bad_page_polling_timeout(struct amdgpu_device *adev,
-			bool reset, uint32_t timeout_ms)
-{
-	struct ras_err_data err_data;
-	struct ras_common_if head = {
-		.block = AMDGPU_RAS_BLOCK__UMC,
-	};
-	struct ras_manager *obj = amdgpu_ras_find_obj(adev, &head);
-	uint32_t timeout = timeout_ms;
-
-	memset(&err_data, 0, sizeof(err_data));
-	amdgpu_ras_error_data_init(&err_data);
-
-	do {
-
-		amdgpu_umc_handle_bad_pages(adev, &err_data);
-
-		if (timeout && !err_data.de_count) {
-			msleep(1);
-			timeout--;
-		}
-
-	} while (timeout && !err_data.de_count);
-
-	if (!timeout)
-		dev_warn(adev->dev, "Can't find bad pages\n");
-
-	if (err_data.de_count)
-		dev_info(adev->dev, "%ld new deferred hardware errors detected\n", err_data.de_count);
-
-	if (obj) {
-		obj->err_data.ue_count += err_data.ue_count;
-		obj->err_data.ce_count += err_data.ce_count;
-		obj->err_data.de_count += err_data.de_count;
-	}
-
-	amdgpu_ras_error_data_fini(&err_data);
-
-	kgd2kfd_set_sram_ecc_flag(adev->kfd.dev);
-
-	if (reset) {
-		struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
-
-		/* use mode-2 reset for poison consumption */
-		con->gpu_reset_flags |= AMDGPU_RAS_GPU_RESET_MODE2_RESET;
-		amdgpu_ras_reset_gpu(adev);
-	}
-
-	return 0;
-}
-
-int amdgpu_umc_poison_handler(struct amdgpu_device *adev,
-			enum amdgpu_ras_block block, bool reset)
+int amdgpu_umc_pasid_poison_handler(struct amdgpu_device *adev,
+			enum amdgpu_ras_block block, uint16_t pasid,
+			pasid_notify pasid_fn, void *data, uint32_t reset)
 {
 	int ret = AMDGPU_RAS_SUCCESS;
 
@@ -285,14 +245,13 @@ int amdgpu_umc_poison_handler(struct amdgpu_device *adev,
 
 			amdgpu_ras_error_data_fini(&err_data);
 		} else {
-			if (reset) {
-				amdgpu_umc_bad_page_polling_timeout(adev,
-							reset, MAX_UMC_POISON_POLLING_TIME_SYNC);
-			} else {
-				struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+			struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+			int ret;
 
+			ret = amdgpu_ras_put_poison_req(adev,
+				block, pasid, pasid_fn, data, reset);
+			if (!ret) {
 				atomic_inc(&con->page_retirement_req_cnt);
-
 				wake_up(&con->page_retirement_wq);
 			}
 		}
@@ -307,11 +266,19 @@ int amdgpu_umc_poison_handler(struct amdgpu_device *adev,
 	return ret;
 }
 
+int amdgpu_umc_poison_handler(struct amdgpu_device *adev,
+			enum amdgpu_ras_block block, uint32_t reset)
+{
+	return amdgpu_umc_pasid_poison_handler(adev,
+				block, 0, NULL, NULL, reset);
+}
+
 int amdgpu_umc_process_ras_data_cb(struct amdgpu_device *adev,
 		void *ras_error_status,
 		struct amdgpu_iv_entry *entry)
 {
-	return amdgpu_umc_do_page_retirement(adev, ras_error_status, entry, true);
+	return amdgpu_umc_do_page_retirement(adev, ras_error_status, entry,
+				AMDGPU_RAS_GPU_RESET_MODE1_RESET);
 }
 
 int amdgpu_umc_ras_sw_init(struct amdgpu_device *adev)
@@ -352,6 +319,9 @@ int amdgpu_umc_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *r
 	if (r)
 		return r;
 
+	if (amdgpu_sriov_vf(adev))
+		return r;
+
 	if (amdgpu_ras_is_supported(adev, ras_block->block)) {
 		r = amdgpu_irq_get(adev, &adev->gmc.ecc_irq, 0);
 		if (r)
@@ -388,14 +358,20 @@ int amdgpu_umc_process_ecc_irq(struct amdgpu_device *adev,
 	return 0;
 }
 
-void amdgpu_umc_fill_error_record(struct ras_err_data *err_data,
+int 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];
+	struct eeprom_table_record *err_rec;
+
+	if (!err_data ||
+	    !err_data->err_addr ||
+	    (err_data->err_addr_cnt >= err_data->err_addr_len))
+		return -EINVAL;
+
+	err_rec = &err_data->err_addr[err_data->err_addr_cnt];
 
 	err_rec->address = err_addr;
 	/* page frame address is saved */
@@ -407,6 +383,47 @@ void amdgpu_umc_fill_error_record(struct ras_err_data *err_data,
 	err_rec->mcumc_id = umc_inst;
 
 	err_data->err_addr_cnt++;
+
+	return 0;
+}
+
+static int amdgpu_umc_loop_all_aid(struct amdgpu_device *adev, umc_func func,
+				   void *data)
+{
+	uint32_t umc_node_inst;
+	uint32_t node_inst;
+	uint32_t umc_inst;
+	uint32_t ch_inst;
+	int ret;
+
+	/*
+	 * This loop is done based on the following -
+	 * umc.active mask = mask of active umc instances across all nodes
+	 * umc.umc_inst_num = maximum number of umc instancess per node
+	 * umc.node_inst_num = maximum number of node instances
+	 * Channel instances are not assumed to be harvested.
+	 */
+	dev_dbg(adev->dev, "active umcs :%lx umc_inst per node: %d",
+		adev->umc.active_mask, adev->umc.umc_inst_num);
+	for_each_set_bit(umc_node_inst, &(adev->umc.active_mask),
+			 adev->umc.node_inst_num * adev->umc.umc_inst_num) {
+		node_inst = umc_node_inst / adev->umc.umc_inst_num;
+		umc_inst = umc_node_inst % adev->umc.umc_inst_num;
+		LOOP_UMC_CH_INST(ch_inst) {
+			dev_dbg(adev->dev,
+				"node_inst :%d umc_inst: %d ch_inst: %d",
+				node_inst, umc_inst, ch_inst);
+			ret = func(adev, node_inst, umc_inst, ch_inst, data);
+			if (ret) {
+				dev_err(adev->dev,
+					"Node %d umc %d ch %d func returns %d\n",
+					node_inst, umc_inst, ch_inst, ret);
+				return ret;
+			}
+		}
+	}
+
+	return 0;
 }
 
 int amdgpu_umc_loop_channels(struct amdgpu_device *adev,
@@ -417,6 +434,9 @@ int amdgpu_umc_loop_channels(struct amdgpu_device *adev,
 	uint32_t ch_inst         = 0;
 	int ret = 0;
 
+	if (adev->aid_mask)
+		return amdgpu_umc_loop_all_aid(adev, func, data);
+
 	if (adev->umc.node_inst_num) {
 		LOOP_UMC_EACH_NODE_INST_AND_CH(node_inst, umc_inst, ch_inst) {
 			ret = func(adev, node_inst, umc_inst, ch_inst, data);
@@ -439,3 +459,129 @@ int amdgpu_umc_loop_channels(struct amdgpu_device *adev,
 
 	return 0;
 }
+
+int amdgpu_umc_update_ecc_status(struct amdgpu_device *adev,
+				uint64_t status, uint64_t ipid, uint64_t addr)
+{
+	if (adev->umc.ras->update_ecc_status)
+		return adev->umc.ras->update_ecc_status(adev,
+					status, ipid, addr);
+	return 0;
+}
+
+int amdgpu_umc_logs_ecc_err(struct amdgpu_device *adev,
+		struct radix_tree_root *ecc_tree, struct ras_ecc_err *ecc_err)
+{
+	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct ras_ecc_log_info *ecc_log;
+	int ret;
+
+	ecc_log = &con->umc_ecc_log;
+
+	mutex_lock(&ecc_log->lock);
+	ret = radix_tree_insert(ecc_tree, ecc_err->pa_pfn, ecc_err);
+	if (!ret)
+		radix_tree_tag_set(ecc_tree,
+			ecc_err->pa_pfn, UMC_ECC_NEW_DETECTED_TAG);
+	mutex_unlock(&ecc_log->lock);
+
+	return ret;
+}
+
+int amdgpu_umc_pages_in_a_row(struct amdgpu_device *adev,
+			struct ras_err_data *err_data, uint64_t pa_addr)
+{
+	struct ta_ras_query_address_output addr_out;
+
+	/* reinit err_data */
+	err_data->err_addr_cnt = 0;
+	err_data->err_addr_len = adev->umc.retire_unit;
+
+	addr_out.pa.pa = pa_addr;
+	if (adev->umc.ras && adev->umc.ras->convert_ras_err_addr)
+		return adev->umc.ras->convert_ras_err_addr(adev, err_data, NULL,
+				&addr_out, false);
+	else
+		return -EINVAL;
+}
+
+int amdgpu_umc_lookup_bad_pages_in_a_row(struct amdgpu_device *adev,
+			uint64_t pa_addr, uint64_t *pfns, int len)
+{
+	int i, ret;
+	struct ras_err_data err_data;
+
+	err_data.err_addr = kcalloc(adev->umc.retire_unit,
+				sizeof(struct eeprom_table_record), GFP_KERNEL);
+	if (!err_data.err_addr) {
+		dev_warn(adev->dev, "Failed to alloc memory in bad page lookup!\n");
+		return 0;
+	}
+
+	ret = amdgpu_umc_pages_in_a_row(adev, &err_data, pa_addr);
+	if (ret)
+		goto out;
+
+	for (i = 0; i < adev->umc.retire_unit; i++) {
+		if (i >= len)
+			goto out;
+
+		pfns[i] = err_data.err_addr[i].retired_page;
+	}
+	ret = i;
+	adev->umc.err_addr_cnt = err_data.err_addr_cnt;
+
+out:
+	kfree(err_data.err_addr);
+	return ret;
+}
+
+int amdgpu_umc_mca_to_addr(struct amdgpu_device *adev,
+			uint64_t err_addr, uint32_t ch, uint32_t umc,
+			uint32_t node, uint32_t socket,
+			struct ta_ras_query_address_output *addr_out, bool dump_addr)
+{
+	struct ta_ras_query_address_input addr_in;
+	int ret;
+
+	memset(&addr_in, 0, sizeof(addr_in));
+	addr_in.ma.err_addr = err_addr;
+	addr_in.ma.ch_inst = ch;
+	addr_in.ma.umc_inst = umc;
+	addr_in.ma.node_inst = node;
+	addr_in.ma.socket_id = socket;
+
+	if (adev->umc.ras && adev->umc.ras->convert_ras_err_addr) {
+		ret = adev->umc.ras->convert_ras_err_addr(adev, NULL, &addr_in,
+				addr_out, dump_addr);
+		if (ret)
+			return ret;
+	} else {
+		return 0;
+	}
+
+	return 0;
+}
+
+int amdgpu_umc_pa2mca(struct amdgpu_device *adev,
+		uint64_t pa, uint64_t *mca, enum amdgpu_memory_partition nps)
+{
+	struct ta_ras_query_address_input addr_in;
+	struct ta_ras_query_address_output addr_out;
+	int ret;
+
+	/* nps: the pa belongs to */
+	addr_in.pa.pa = pa | ((uint64_t)nps << 58);
+	addr_in.addr_type = TA_RAS_PA_TO_MCA;
+	ret = psp_ras_query_address(&adev->psp, &addr_in, &addr_out);
+	if (ret) {
+		dev_warn(adev->dev, "Failed to query RAS MCA address for 0x%llx",
+			pa);
+
+		return ret;
+	}
+
+	*mca = addr_out.ma.err_addr;
+
+	return 0;
+}