From 772df3df809a536c3e1b14db5a1dd74ae7baa102 Mon Sep 17 00:00:00 2001
From: Tao Zhou <tao.zhou1@amd.com>
Date: Fri, 18 Oct 2024 17:52:59 +0800
Subject: drm/amdgpu: add flag to indicate the type of RAS eeprom record

One UMC MCA address could map to multiply physical address (PA):

AMDGPU_RAS_EEPROM_REC_PA: one record store one PA
AMDGPU_RAS_EEPROM_REC_MCA: one record store one MCA address, PA
is not cared about

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 33 ++++++++++++++++++++++++++-------
 1 file changed, 26 insertions(+), 7 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index 4c9fa24dd972..e7eccc30c692 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -2821,10 +2821,20 @@ static int amdgpu_ras_load_bad_pages(struct amdgpu_device *adev)
 		return -ENOMEM;
 
 	ret = amdgpu_ras_eeprom_read(control, bps, control->ras_num_recs);
-	if (ret)
+	if (ret) {
 		dev_err(adev->dev, "Failed to load EEPROM table records!");
-	else
+	} else {
+		if (control->ras_num_recs > 1 &&
+		    adev->umc.ras && adev->umc.ras->convert_ras_err_addr) {
+			if ((bps[0].address == bps[1].address) &&
+			    (bps[0].mem_channel == bps[1].mem_channel))
+				control->rec_type = AMDGPU_RAS_EEPROM_REC_PA;
+			else
+				control->rec_type = AMDGPU_RAS_EEPROM_REC_MCA;
+		}
+
 		ret = amdgpu_ras_add_bad_pages(adev, bps, control->ras_num_recs);
+	}
 
 	kfree(bps);
 	return ret;
@@ -3205,13 +3215,14 @@ static int amdgpu_ras_page_retirement_thread(void *param)
 int amdgpu_ras_init_badpage_info(struct amdgpu_device *adev)
 {
 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+	struct amdgpu_ras_eeprom_control *control;
 	int ret;
 
 	if (!con || amdgpu_sriov_vf(adev))
 		return 0;
 
-	ret = amdgpu_ras_eeprom_init(&con->eeprom_control);
-
+	control = &con->eeprom_control;
+	ret = amdgpu_ras_eeprom_init(control);
 	if (ret)
 		return ret;
 
@@ -3219,17 +3230,25 @@ int amdgpu_ras_init_badpage_info(struct amdgpu_device *adev)
 	if (amdgpu_ras_is_rma(adev))
 		return -EHWPOISON;
 
-	if (con->eeprom_control.ras_num_recs) {
+	if (!adev->umc.ras || !adev->umc.ras->convert_ras_err_addr)
+		control->rec_type = AMDGPU_RAS_EEPROM_REC_PA;
+
+	/* default status is MCA storage */
+	if (control->ras_num_recs <= 1 &&
+	    adev->umc.ras && adev->umc.ras->convert_ras_err_addr)
+		control->rec_type = AMDGPU_RAS_EEPROM_REC_MCA;
+
+	if (control->ras_num_recs) {
 		ret = amdgpu_ras_load_bad_pages(adev);
 		if (ret)
 			return ret;
 
 		amdgpu_dpm_send_hbm_bad_pages_num(
-			adev, con->eeprom_control.ras_num_recs);
+			adev, control->ras_num_recs);
 
 		if (con->update_channel_flag == true) {
 			amdgpu_dpm_send_hbm_bad_channel_flag(
-				adev, con->eeprom_control.bad_channel_bitmap);
+				adev, control->bad_channel_bitmap);
 			con->update_channel_flag = false;
 		}
 	}
-- 
cgit v1.2.3


From 0eecff79e49f8ce5475e1b4d968f26263587be66 Mon Sep 17 00:00:00 2001
From: Tao Zhou <tao.zhou1@amd.com>
Date: Fri, 18 Oct 2024 14:49:00 +0800
Subject: drm/amdgpu: do RAS MCA2PA conversion in device init phase

NPS mode is introduced, the value of memory physical address (PA)
related to a MCA address varies per nps mode. We need to rely on
MCA address and convert it into PA accroding to nps mode.

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 94 ++++++++++++++++++++++++++++-----
 1 file changed, 82 insertions(+), 12 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index e7eccc30c692..f22242ab2407 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -2717,40 +2717,110 @@ static int amdgpu_ras_realloc_eh_data_space(struct amdgpu_device *adev,
 	return 0;
 }
 
+static int amdgpu_ras_mca2pa(struct amdgpu_device *adev,
+			struct eeprom_table_record *bps,
+			struct ras_err_data *err_data)
+{
+	struct ta_ras_query_address_input addr_in;
+	uint32_t socket = 0;
+	int ret = 0;
+
+	if (adev->smuio.funcs && adev->smuio.funcs->get_socket_id)
+		socket = adev->smuio.funcs->get_socket_id(adev);
+
+	/* reinit err_data */
+	err_data->err_addr_cnt = 0;
+	err_data->err_addr_len = adev->umc.retire_unit;
+
+	memset(&addr_in, 0, sizeof(addr_in));
+	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 (adev->umc.ras && adev->umc.ras->convert_ras_err_addr)
+		ret = adev->umc.ras->convert_ras_err_addr(adev, err_data,
+				&addr_in, NULL, false);
+
+	return ret;
+}
+
 /* it deal with vram only. */
 int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev,
 		struct eeprom_table_record *bps, int pages)
 {
 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
 	struct ras_err_handler_data *data;
+	struct ras_err_data err_data;
+	struct eeprom_table_record *err_rec;
 	int ret = 0;
-	uint32_t i;
+	uint32_t i, j, loop_cnt = 1;
+	bool is_mca_add = true;
 
 	if (!con || !con->eh_data || !bps || pages <= 0)
 		return 0;
 
+	if (!adev->umc.ras || !adev->umc.ras->convert_ras_err_addr) {
+		is_mca_add = false;
+	} else {
+		if ((pages > 1) &&
+		    (bps[0].address == bps[1].address) &&
+		    (bps[0].mem_channel == bps[1].mem_channel))
+			is_mca_add = false;
+	}
+
 	mutex_lock(&con->recovery_lock);
 	data = con->eh_data;
 	if (!data)
 		goto out;
 
-	for (i = 0; i < pages; i++) {
-		if (amdgpu_ras_check_bad_page_unlock(con,
-			bps[i].retired_page << AMDGPU_GPU_PAGE_SHIFT))
-			continue;
-
-		if (!data->space_left &&
-			amdgpu_ras_realloc_eh_data_space(adev, data, 256)) {
+	if (is_mca_add) {
+		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 UMC error address record in mca2pa conversion!\n");
 			ret = -ENOMEM;
 			goto out;
 		}
 
-		amdgpu_ras_reserve_page(adev, bps[i].retired_page);
+		loop_cnt = adev->umc.retire_unit;
+	}
+
+	for (i = 0; i < pages; i++) {
+		if (is_mca_add) {
+			if (amdgpu_ras_mca2pa(adev, &bps[i], &err_data))
+				goto free;
+
+			err_rec = err_data.err_addr;
+		} else {
+			err_rec = &bps[i];
+		}
 
-		memcpy(&data->bps[data->count], &bps[i], sizeof(*data->bps));
-		data->count++;
-		data->space_left--;
+		for (j = 0; j < loop_cnt; j++) {
+			if (amdgpu_ras_check_bad_page_unlock(con,
+				err_rec[j].retired_page << AMDGPU_GPU_PAGE_SHIFT))
+				continue;
+
+			if (!data->space_left &&
+			    amdgpu_ras_realloc_eh_data_space(adev, data, 256)) {
+				ret = -ENOMEM;
+				goto free;
+			}
+
+			amdgpu_ras_reserve_page(adev, err_rec[j].retired_page);
+
+			memcpy(&data->bps[data->count], &(err_rec[j]),
+					sizeof(struct eeprom_table_record));
+			data->count++;
+			data->space_left--;
+		}
 	}
+
+free:
+	if (is_mca_add)
+		kfree(err_data.err_addr);
 out:
 	mutex_unlock(&con->recovery_lock);
 
-- 
cgit v1.2.3


From e1ee2111ca48169a9fdc5075f7863f5d4d591e2f Mon Sep 17 00:00:00 2001
From: Lijo Lazar <lijo.lazar@amd.com>
Date: Thu, 24 Oct 2024 11:01:57 +0530
Subject: drm/amdgpu: Prefer RAS recovery for scheduler hang

Before scheduling a recovery due to scheduler/job hang, check if a RAS
error is detected. If so, choose RAS recovery to handle the situation. A
scheduler/job hang could be the side effect of a RAS error. In such
cases, it is required to go through the RAS error recovery process. A
RAS error recovery process in certains cases also could avoid a full
device device reset.

An error state is maintained in RAS context to detect the block
affected. Fatal Error state uses unused block id. Set the block id when
error is detected. If the interrupt handler detected a poison error,
it's not required to look for a fatal error. Skip fatal error checking
in such cases.

Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 55 +++++++++++++++++++++++++++++++--
 1 file changed, 53 insertions(+), 2 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index f22242ab2407..4df9a8dfe9eb 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -2156,6 +2156,16 @@ void amdgpu_ras_interrupt_fatal_error_handler(struct amdgpu_device *adev)
 	/* Fatal error events are handled on host side */
 	if (amdgpu_sriov_vf(adev))
 		return;
+	/**
+	 * If the current interrupt is caused by a non-fatal RAS error, skip
+	 * check for fatal error. For fatal errors, FED status of all devices
+	 * in XGMI hive gets set when the first device gets fatal error
+	 * interrupt. The error gets propagated to other devices as well, so
+	 * make sure to ack the interrupt regardless of FED status.
+	 */
+	if (!amdgpu_ras_get_fed_status(adev) &&
+	    amdgpu_ras_is_err_state(adev, AMDGPU_RAS_BLOCK__ANY))
+		return;
 
 	if (adev->nbio.ras &&
 	    adev->nbio.ras->handle_ras_controller_intr_no_bifring)
@@ -2185,6 +2195,7 @@ static void amdgpu_ras_interrupt_poison_consumption_handler(struct ras_manager *
 	if (ret)
 		return;
 
+	amdgpu_ras_set_err_poison(adev, block_obj->ras_comm.block);
 	/* both query_poison_status and handle_poison_consumption are optional,
 	 * but at least one of them should be implemented if we need poison
 	 * consumption handler
@@ -4172,16 +4183,56 @@ bool amdgpu_ras_get_fed_status(struct amdgpu_device *adev)
 	if (!ras)
 		return false;
 
-	return atomic_read(&ras->fed);
+	return test_bit(AMDGPU_RAS_BLOCK__LAST, &ras->ras_err_state);
 }
 
 void amdgpu_ras_set_fed(struct amdgpu_device *adev, bool status)
 {
 	struct amdgpu_ras *ras;
 
+	ras = amdgpu_ras_get_context(adev);
+	if (ras) {
+		if (status)
+			set_bit(AMDGPU_RAS_BLOCK__LAST, &ras->ras_err_state);
+		else
+			clear_bit(AMDGPU_RAS_BLOCK__LAST, &ras->ras_err_state);
+	}
+}
+
+void amdgpu_ras_clear_err_state(struct amdgpu_device *adev)
+{
+	struct amdgpu_ras *ras;
+
 	ras = amdgpu_ras_get_context(adev);
 	if (ras)
-		atomic_set(&ras->fed, !!status);
+		ras->ras_err_state = 0;
+}
+
+void amdgpu_ras_set_err_poison(struct amdgpu_device *adev,
+			       enum amdgpu_ras_block block)
+{
+	struct amdgpu_ras *ras;
+
+	ras = amdgpu_ras_get_context(adev);
+	if (ras)
+		set_bit(block, &ras->ras_err_state);
+}
+
+bool amdgpu_ras_is_err_state(struct amdgpu_device *adev, int block)
+{
+	struct amdgpu_ras *ras;
+
+	ras = amdgpu_ras_get_context(adev);
+	if (ras) {
+		if (block == AMDGPU_RAS_BLOCK__ANY)
+			return (ras->ras_err_state != 0);
+		else
+			return test_bit(block, &ras->ras_err_state) ||
+			       test_bit(AMDGPU_RAS_BLOCK__LAST,
+					&ras->ras_err_state);
+	}
+
+	return false;
 }
 
 static struct ras_event_manager *__get_ras_event_mgr(struct amdgpu_device *adev)
-- 
cgit v1.2.3


From c3d4acf0c3bbba4eb24812f12412d642fa3d5378 Mon Sep 17 00:00:00 2001
From: Tao Zhou <tao.zhou1@amd.com>
Date: Fri, 18 Oct 2024 18:58:54 +0800
Subject: drm/amdgpu: store only one RAS bad page record for all pages in one
 row

So eeprom space can be saved, compatible with legacy way.

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 35 +++++++++++++++++++++++++--------
 1 file changed, 27 insertions(+), 8 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index 4df9a8dfe9eb..882a33e134d8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -2849,7 +2849,7 @@ int amdgpu_ras_save_bad_pages(struct amdgpu_device *adev,
 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
 	struct ras_err_handler_data *data;
 	struct amdgpu_ras_eeprom_control *control;
-	int save_count;
+	int save_count, unit_num, bad_page_num, i;
 
 	if (!con || !con->eh_data) {
 		if (new_cnt)
@@ -2861,19 +2861,38 @@ int amdgpu_ras_save_bad_pages(struct amdgpu_device *adev,
 	mutex_lock(&con->recovery_lock);
 	control = &con->eeprom_control;
 	data = con->eh_data;
-	save_count = data->count - control->ras_num_recs;
+	bad_page_num = control->ras_num_recs;
+	/* one record on eeprom stands for all pages in one memory row
+	 * in this mode
+	 */
+	if (control->rec_type == AMDGPU_RAS_EEPROM_REC_MCA)
+		bad_page_num = control->ras_num_recs * adev->umc.retire_unit;
+
+	save_count = data->count - bad_page_num;
 	mutex_unlock(&con->recovery_lock);
 
+	unit_num = save_count / adev->umc.retire_unit;
 	if (new_cnt)
-		*new_cnt = save_count / adev->umc.retire_unit;
+		*new_cnt = unit_num;
 
 	/* only new entries are saved */
 	if (save_count > 0) {
-		if (amdgpu_ras_eeprom_append(control,
-					     &data->bps[control->ras_num_recs],
-					     save_count)) {
-			dev_err(adev->dev, "Failed to save EEPROM table data!");
-			return -EIO;
+		if (control->rec_type == AMDGPU_RAS_EEPROM_REC_PA) {
+			if (amdgpu_ras_eeprom_append(control,
+						     &data->bps[control->ras_num_recs],
+						     save_count)) {
+				dev_err(adev->dev, "Failed to save EEPROM table data!");
+				return -EIO;
+			}
+		} else {
+			for (i = 0; i < unit_num; i++) {
+				if (amdgpu_ras_eeprom_append(control,
+						&data->bps[bad_page_num + i * adev->umc.retire_unit],
+						1)) {
+					dev_err(adev->dev, "Failed to save EEPROM table data!");
+					return -EIO;
+				}
+			}
 		}
 
 		dev_info(adev->dev, "Saved %d pages to EEPROM table.\n", save_count);
-- 
cgit v1.2.3


From 07dd49e1fc42fcb67a1786c2b2f71e0dd0079935 Mon Sep 17 00:00:00 2001
From: Tao Zhou <tao.zhou1@amd.com>
Date: Thu, 24 Oct 2024 15:34:27 +0800
Subject: drm/amdgpu: support to find RAS bad pages via old TA

Old version of RAS TA doesn't support to convert MCA address stored on
eeprom to physical address (PA), support to find all bad pages in one
memory row by PA with old RAS TA. This approach is only suitable for
nps1 mode.

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 28 +++++++++++++++++++++++++---
 1 file changed, 25 insertions(+), 3 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index 882a33e134d8..d41a0664aa27 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -2765,9 +2765,10 @@ int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev,
 	struct ras_err_handler_data *data;
 	struct ras_err_data err_data;
 	struct eeprom_table_record *err_rec;
+	enum amdgpu_memory_partition nps = AMDGPU_NPS1_PARTITION_MODE;
 	int ret = 0;
 	uint32_t i, j, loop_cnt = 1;
-	bool is_mca_add = true;
+	bool is_mca_add = true, find_pages_per_pa = false;
 
 	if (!con || !con->eh_data || !bps || pages <= 0)
 		return 0;
@@ -2797,12 +2798,33 @@ int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev,
 		}
 
 		loop_cnt = adev->umc.retire_unit;
+		if (adev->gmc.gmc_funcs->query_mem_partition_mode)
+			nps = adev->gmc.gmc_funcs->query_mem_partition_mode(adev);
 	}
 
 	for (i = 0; i < pages; i++) {
 		if (is_mca_add) {
-			if (amdgpu_ras_mca2pa(adev, &bps[i], &err_data))
-				goto free;
+			if (!find_pages_per_pa) {
+				if (amdgpu_ras_mca2pa(adev, &bps[i], &err_data)) {
+					if (!i && nps == AMDGPU_NPS1_PARTITION_MODE) {
+						/* may use old RAS TA, use PA to find pages in
+						 * one row
+						 */
+						if (amdgpu_umc_pages_in_a_row(adev, &err_data,
+								bps[i].retired_page << AMDGPU_GPU_PAGE_SHIFT))
+							goto free;
+						else
+							find_pages_per_pa = true;
+					} else {
+						/* unsupported cases */
+						goto free;
+					}
+				}
+			} else {
+				if (amdgpu_umc_pages_in_a_row(adev, &err_data,
+						bps[i].retired_page << AMDGPU_GPU_PAGE_SHIFT))
+					goto free;
+			}
 
 			err_rec = err_data.err_addr;
 		} else {
-- 
cgit v1.2.3


From a8d133e625ceb147a173b6cafc862a9bd4312894 Mon Sep 17 00:00:00 2001
From: Tao Zhou <tao.zhou1@amd.com>
Date: Thu, 31 Oct 2024 15:48:10 +0800
Subject: drm/amdgpu: parse legacy RAS bad page mixed with new data in various
 NPS modes

All legacy RAS bad pages are generated in NPS1 mode, but new bad page
can be generated in any NPS mode, so we can't use retired_page stored
on eeprom directly in non-nps1 mode even for legacy data. We need to
take different actions for different data, new data can be identified
from old data by UMC_CHANNEL_IDX_V2 flag.

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 96 +++++++++++++++++++++++++++------
 1 file changed, 81 insertions(+), 15 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index d41a0664aa27..81713888d073 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -192,7 +192,7 @@ static int amdgpu_reserve_page_direct(struct amdgpu_device *adev, uint64_t addre
 
 	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);
 	}
 
@@ -2728,7 +2728,7 @@ static int amdgpu_ras_realloc_eh_data_space(struct amdgpu_device *adev,
 	return 0;
 }
 
-static int amdgpu_ras_mca2pa(struct amdgpu_device *adev,
+static int amdgpu_ras_mca2pa_by_idx(struct amdgpu_device *adev,
 			struct eeprom_table_record *bps,
 			struct ras_err_data *err_data)
 {
@@ -2757,9 +2757,46 @@ static int amdgpu_ras_mca2pa(struct amdgpu_device *adev,
 	return ret;
 }
 
+static int amdgpu_ras_mca2pa(struct amdgpu_device *adev,
+			struct eeprom_table_record *bps,
+			struct ras_err_data *err_data)
+{
+	struct ta_ras_query_address_input addr_in;
+	uint32_t die_id, socket = 0;
+
+	if (adev->smuio.funcs && adev->smuio.funcs->get_socket_id)
+		socket = adev->smuio.funcs->get_socket_id(adev);
+
+	/* although die id is gotten from PA in nps1 mode, the id is
+	 * fitable for any nps mode
+	 */
+	if (adev->umc.ras && adev->umc.ras->get_die_id_from_pa)
+		die_id = adev->umc.ras->get_die_id_from_pa(adev, bps->address,
+					bps->retired_page << AMDGPU_GPU_PAGE_SHIFT);
+	else
+		return -EINVAL;
+
+	/* reinit err_data */
+	err_data->err_addr_cnt = 0;
+	err_data->err_addr_len = adev->umc.retire_unit;
+
+	memset(&addr_in, 0, sizeof(addr_in));
+	addr_in.ma.err_addr = bps->address;
+	addr_in.ma.ch_inst = bps->mem_channel;
+	addr_in.ma.umc_inst = bps->mcumc_id;
+	addr_in.ma.node_inst = die_id;
+	addr_in.ma.socket_id = socket;
+
+	if (adev->umc.ras && adev->umc.ras->convert_ras_err_addr)
+		return adev->umc.ras->convert_ras_err_addr(adev, err_data,
+					&addr_in, NULL, false);
+	else
+		return  -EINVAL;
+}
+
 /* it deal with vram only. */
 int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev,
-		struct eeprom_table_record *bps, int pages)
+		struct eeprom_table_record *bps, int pages, bool from_rom)
 {
 	struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
 	struct ras_err_handler_data *data;
@@ -2782,12 +2819,7 @@ int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev,
 			is_mca_add = false;
 	}
 
-	mutex_lock(&con->recovery_lock);
-	data = con->eh_data;
-	if (!data)
-		goto out;
-
-	if (is_mca_add) {
+	if (from_rom) {
 		err_data.err_addr =
 			kcalloc(adev->umc.retire_unit,
 				sizeof(struct eeprom_table_record), GFP_KERNEL);
@@ -2797,15 +2829,21 @@ int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev,
 			goto out;
 		}
 
+		err_rec = err_data.err_addr;
 		loop_cnt = adev->umc.retire_unit;
 		if (adev->gmc.gmc_funcs->query_mem_partition_mode)
 			nps = adev->gmc.gmc_funcs->query_mem_partition_mode(adev);
 	}
 
+	mutex_lock(&con->recovery_lock);
+	data = con->eh_data;
+	if (!data)
+		goto free;
+
 	for (i = 0; i < pages; i++) {
 		if (is_mca_add) {
 			if (!find_pages_per_pa) {
-				if (amdgpu_ras_mca2pa(adev, &bps[i], &err_data)) {
+				if (amdgpu_ras_mca2pa_by_idx(adev, &bps[i], &err_data)) {
 					if (!i && nps == AMDGPU_NPS1_PARTITION_MODE) {
 						/* may use old RAS TA, use PA to find pages in
 						 * one row
@@ -2825,10 +2863,38 @@ int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev,
 						bps[i].retired_page << AMDGPU_GPU_PAGE_SHIFT))
 					goto free;
 			}
-
-			err_rec = err_data.err_addr;
 		} else {
-			err_rec = &bps[i];
+			if (from_rom && !find_pages_per_pa) {
+				if (bps[i].retired_page & UMC_CHANNEL_IDX_V2) {
+					/* bad page in any NPS mode in eeprom */
+					if (amdgpu_ras_mca2pa_by_idx(adev, &bps[i], &err_data))
+						goto free;
+				} else {
+					/* legacy bad page in eeprom, generated only in
+					 * NPS1 mode
+					 */
+					if (amdgpu_ras_mca2pa(adev, &bps[i], &err_data)) {
+						/* old RAS TA or ASICs which don't support to
+						 * convert addrss via mca address
+						 */
+						if (!i && nps == AMDGPU_NPS1_PARTITION_MODE) {
+							find_pages_per_pa = true;
+							err_rec = &bps[i];
+							loop_cnt = 1;
+						} else {
+							/* non-nps1 mode, old RAS TA
+							 * can't support it
+							 */
+							goto free;
+						}
+					}
+				}
+
+				if (!find_pages_per_pa)
+					i += (adev->umc.retire_unit - 1);
+			} else {
+				err_rec = &bps[i];
+			}
 		}
 
 		for (j = 0; j < loop_cnt; j++) {
@@ -2852,7 +2918,7 @@ int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev,
 	}
 
 free:
-	if (is_mca_add)
+	if (from_rom)
 		kfree(err_data.err_addr);
 out:
 	mutex_unlock(&con->recovery_lock);
@@ -2955,7 +3021,7 @@ static int amdgpu_ras_load_bad_pages(struct amdgpu_device *adev)
 				control->rec_type = AMDGPU_RAS_EEPROM_REC_MCA;
 		}
 
-		ret = amdgpu_ras_add_bad_pages(adev, bps, control->ras_num_recs);
+		ret = amdgpu_ras_add_bad_pages(adev, bps, control->ras_num_recs, true);
 	}
 
 	kfree(bps);
-- 
cgit v1.2.3


From d08fb6637014c8e80cfcc1aabc38beac857a1f29 Mon Sep 17 00:00:00 2001
From: Tao Zhou <tao.zhou1@amd.com>
Date: Fri, 1 Nov 2024 12:09:49 +0800
Subject: drm/amdgpu: remove is_mca_add for ras_add_bad_pages

Remove unnecessary variable and simplify the logic.

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 16 +++++-----------
 1 file changed, 5 insertions(+), 11 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index 81713888d073..66e80fcc2899 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -2802,23 +2802,16 @@ int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev,
 	struct ras_err_handler_data *data;
 	struct ras_err_data err_data;
 	struct eeprom_table_record *err_rec;
+	struct amdgpu_ras_eeprom_control *control =
+			&adev->psp.ras_context.ras->eeprom_control;
 	enum amdgpu_memory_partition nps = AMDGPU_NPS1_PARTITION_MODE;
 	int ret = 0;
 	uint32_t i, j, loop_cnt = 1;
-	bool is_mca_add = true, find_pages_per_pa = false;
+	bool find_pages_per_pa = false;
 
 	if (!con || !con->eh_data || !bps || pages <= 0)
 		return 0;
 
-	if (!adev->umc.ras || !adev->umc.ras->convert_ras_err_addr) {
-		is_mca_add = false;
-	} else {
-		if ((pages > 1) &&
-		    (bps[0].address == bps[1].address) &&
-		    (bps[0].mem_channel == bps[1].mem_channel))
-			is_mca_add = false;
-	}
-
 	if (from_rom) {
 		err_data.err_addr =
 			kcalloc(adev->umc.retire_unit,
@@ -2841,7 +2834,8 @@ int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev,
 		goto free;
 
 	for (i = 0; i < pages; i++) {
-		if (is_mca_add) {
+		if (from_rom &&
+		    control->rec_type == AMDGPU_RAS_EEPROM_REC_MCA) {
 			if (!find_pages_per_pa) {
 				if (amdgpu_ras_mca2pa_by_idx(adev, &bps[i], &err_data)) {
 					if (!i && nps == AMDGPU_NPS1_PARTITION_MODE) {
-- 
cgit v1.2.3


From 1f06e7f344c4ac66184ee6bb49d83b8c413306ca Mon Sep 17 00:00:00 2001
From: Tao Zhou <tao.zhou1@amd.com>
Date: Thu, 28 Nov 2024 18:30:36 +0800
Subject: drm/amdgpu: split ras_eeprom_init into init and check functions

Init function is for ras table header read and check function is
responsible for the validation of the header. Call them in different
stages.

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index 66e80fcc2899..738a645867ef 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -3015,9 +3015,20 @@ static int amdgpu_ras_load_bad_pages(struct amdgpu_device *adev)
 				control->rec_type = AMDGPU_RAS_EEPROM_REC_MCA;
 		}
 
+		ret = amdgpu_ras_eeprom_check(control);
+		if (ret)
+			goto out;
+
+		/* HW not usable */
+		if (amdgpu_ras_is_rma(adev)) {
+			ret = -EHWPOISON;
+			goto out;
+		}
+
 		ret = amdgpu_ras_add_bad_pages(adev, bps, control->ras_num_recs, true);
 	}
 
+out:
 	kfree(bps);
 	return ret;
 }
@@ -3408,10 +3419,6 @@ int amdgpu_ras_init_badpage_info(struct amdgpu_device *adev)
 	if (ret)
 		return ret;
 
-	/* HW not usable */
-	if (amdgpu_ras_is_rma(adev))
-		return -EHWPOISON;
-
 	if (!adev->umc.ras || !adev->umc.ras->convert_ras_err_addr)
 		control->rec_type = AMDGPU_RAS_EEPROM_REC_PA;
 
-- 
cgit v1.2.3


From ae756cd8536923841b45340f7ee369272c9bf105 Mon Sep 17 00:00:00 2001
From: Tao Zhou <tao.zhou1@amd.com>
Date: Fri, 29 Nov 2024 16:52:41 +0800
Subject: drm/amdgpu: correct the calculation of RAS bad page

After the introduction of NPS RAS, one bad page record on eeprom may be
related to 1 or 16 bad pages, so the bad page record and bad page are
two different concepts, define a new variable to store bad page number.

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 10 ++--------
 1 file changed, 2 insertions(+), 8 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index 738a645867ef..1de934cd5764 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -2943,13 +2943,7 @@ int amdgpu_ras_save_bad_pages(struct amdgpu_device *adev,
 	mutex_lock(&con->recovery_lock);
 	control = &con->eeprom_control;
 	data = con->eh_data;
-	bad_page_num = control->ras_num_recs;
-	/* one record on eeprom stands for all pages in one memory row
-	 * in this mode
-	 */
-	if (control->rec_type == AMDGPU_RAS_EEPROM_REC_MCA)
-		bad_page_num = control->ras_num_recs * adev->umc.retire_unit;
-
+	bad_page_num = control->ras_num_bad_pages;
 	save_count = data->count - bad_page_num;
 	mutex_unlock(&con->recovery_lock);
 
@@ -3433,7 +3427,7 @@ int amdgpu_ras_init_badpage_info(struct amdgpu_device *adev)
 			return ret;
 
 		amdgpu_dpm_send_hbm_bad_pages_num(
-			adev, control->ras_num_recs);
+			adev, control->ras_num_bad_pages);
 
 		if (con->update_channel_flag == true) {
 			amdgpu_dpm_send_hbm_bad_channel_flag(
-- 
cgit v1.2.3


From 9a826c4af8f1be3cf6838a8ea2fc70d03315aaa6 Mon Sep 17 00:00:00 2001
From: Hawking Zhang <Hawking.Zhang@amd.com>
Date: Sun, 18 Aug 2024 15:26:13 +0800
Subject: drm/amdgpu: Enable RAS for psp v13_0_12

Enable RAS Cap check and initialize RAS funcs
for psp v13_0_12

Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 5 +++++
 1 file changed, 5 insertions(+)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index 1de934cd5764..9961ff6c29f6 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -2015,6 +2015,7 @@ static bool amdgpu_ras_aca_is_supported(struct amdgpu_device *adev)
 
 	switch (amdgpu_ip_version(adev, MP0_HWIP, 0)) {
 	case IP_VERSION(13, 0, 6):
+	case IP_VERSION(13, 0, 12):
 	case IP_VERSION(13, 0, 14):
 		ret = true;
 		break;
@@ -3568,6 +3569,7 @@ static bool amdgpu_ras_asic_supported(struct amdgpu_device *adev)
 		switch (amdgpu_ip_version(adev, MP0_HWIP, 0)) {
 		case IP_VERSION(13, 0, 2):
 		case IP_VERSION(13, 0, 6):
+		case IP_VERSION(13, 0, 12):
 		case IP_VERSION(13, 0, 14):
 			return true;
 		default:
@@ -3580,6 +3582,7 @@ static bool amdgpu_ras_asic_supported(struct amdgpu_device *adev)
 		case IP_VERSION(13, 0, 0):
 		case IP_VERSION(13, 0, 6):
 		case IP_VERSION(13, 0, 10):
+		case IP_VERSION(13, 0, 12):
 		case IP_VERSION(13, 0, 14):
 			return true;
 		default:
@@ -3831,6 +3834,7 @@ static void amdgpu_ras_init_reserved_vram_size(struct amdgpu_device *adev)
 	switch (amdgpu_ip_version(adev, MP0_HWIP, 0)) {
 	case IP_VERSION(13, 0, 2):
 	case IP_VERSION(13, 0, 6):
+	case IP_VERSION(13, 0, 12):
 	case IP_VERSION(13, 0, 14):
 		con->reserved_pages_in_bytes = AMDGPU_RAS_RESERVED_VRAM_SIZE;
 		break;
@@ -3907,6 +3911,7 @@ int amdgpu_ras_init(struct amdgpu_device *adev)
 			adev->nbio.ras = &nbio_v4_3_ras;
 		break;
 	case IP_VERSION(7, 9, 0):
+	case IP_VERSION(7, 9, 1):
 		if (!adev->gmc.is_app_apu)
 			adev->nbio.ras = &nbio_v7_9_ras;
 		break;
-- 
cgit v1.2.3


From 2c2b84f193bbbd6e4dad2ccd5b1e600f8433f96f Mon Sep 17 00:00:00 2001
From: Candice Li <candice.li@amd.com>
Date: Wed, 4 Dec 2024 17:47:11 +0800
Subject: drm/amdgpu: Add psp v14_0_3 ras support

Add psp v14_0_3 ras support.

Signed-off-by: Candice Li <candice.li@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 1 +
 1 file changed, 1 insertion(+)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index 9961ff6c29f6..623ae9b38800 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -3571,6 +3571,7 @@ static bool amdgpu_ras_asic_supported(struct amdgpu_device *adev)
 		case IP_VERSION(13, 0, 6):
 		case IP_VERSION(13, 0, 12):
 		case IP_VERSION(13, 0, 14):
+		case IP_VERSION(14, 0, 3):
 			return true;
 		default:
 			return false;
-- 
cgit v1.2.3


From ecd1191e128aacaf7d35f09bbbb28d890a7120ab Mon Sep 17 00:00:00 2001
From: Candice Li <candice.li@amd.com>
Date: Thu, 8 Aug 2024 16:40:41 +0800
Subject: drm/amdgpu: Support nbif v6_3_1 fatal error handling

Add nbif v6_3_1 fatal error handling support.

Signed-off-by: Candice Li <candice.li@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index 623ae9b38800..db081618e85c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -36,6 +36,7 @@
 #include "amdgpu_xgmi.h"
 #include "ivsrcid/nbio/irqsrcs_nbif_7_4.h"
 #include "nbio_v4_3.h"
+#include "nbif_v6_3_1.h"
 #include "nbio_v7_9.h"
 #include "atom.h"
 #include "amdgpu_reset.h"
@@ -3911,6 +3912,17 @@ int amdgpu_ras_init(struct amdgpu_device *adev)
 			 * check DF RAS */
 			adev->nbio.ras = &nbio_v4_3_ras;
 		break;
+	case IP_VERSION(6, 3, 1):
+		if (adev->ras_hw_enabled & (1 << AMDGPU_RAS_BLOCK__DF))
+			/* unlike other generation of nbio ras,
+			 * nbif v6_3_1 only support fatal error interrupt
+			 * to inform software that DF is freezed due to
+			 * system fatal error event. driver should not
+			 * enable nbio ras in such case. Instead,
+			 * check DF RAS
+			 */
+			adev->nbio.ras = &nbif_v6_3_1_ras;
+		break;
 	case IP_VERSION(7, 9, 0):
 	case IP_VERSION(7, 9, 1):
 		if (!adev->gmc.is_app_apu)
-- 
cgit v1.2.3


From d1ebe307b44bbc9a98578c8f8089bb8789c5ecd7 Mon Sep 17 00:00:00 2001
From: Candice Li <candice.li@amd.com>
Date: Mon, 16 Dec 2024 17:20:12 +0800
Subject: drm/amdgpu: Enable psp v14_0_3 RAS support for non-SRIOV
 configurations.

Enable psp v14_0_3 RAS support for non-SRIOV configurations.

Signed-off-by: Candice Li <candice.li@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index db081618e85c..01c947066a2e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -3572,7 +3572,6 @@ static bool amdgpu_ras_asic_supported(struct amdgpu_device *adev)
 		case IP_VERSION(13, 0, 6):
 		case IP_VERSION(13, 0, 12):
 		case IP_VERSION(13, 0, 14):
-		case IP_VERSION(14, 0, 3):
 			return true;
 		default:
 			return false;
@@ -3586,6 +3585,7 @@ static bool amdgpu_ras_asic_supported(struct amdgpu_device *adev)
 		case IP_VERSION(13, 0, 10):
 		case IP_VERSION(13, 0, 12):
 		case IP_VERSION(13, 0, 14):
+		case IP_VERSION(14, 0, 3):
 			return true;
 		default:
 			return false;
-- 
cgit v1.2.3