From 95024c714b83d267036564be998328762c47fbda Mon Sep 17 00:00:00 2001
From: Tao Zhou <tao.zhou1@amd.com>
Date: Fri, 18 Oct 2024 14:43:04 +0800
Subject: [PATCH 01/16] drm/amdgpu: add TA_RAS_INV_NODE value

We can set UMC node instance to invalid state if we use global channel
index, and RAS TA can choose UMC address conversion approach by checking
node_inst value.

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/ta_ras_if.h | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/ta_ras_if.h b/drivers/gpu/drm/amd/amdgpu/ta_ras_if.h
index 21b71a427b1f..64891f099366 100644
--- a/drivers/gpu/drm/amd/amdgpu/ta_ras_if.h
+++ b/drivers/gpu/drm/amd/amdgpu/ta_ras_if.h
@@ -30,6 +30,9 @@
 #define RSP_ID_MASK (1U << 31)
 #define RSP_ID(cmdId) (((uint32_t)(cmdId)) | RSP_ID_MASK)
 
+/* invalid node instance value */
+#define TA_RAS_INV_NODE 0xffff
+
 /* RAS related enumerations */
 /**********************************************************/
 enum ras_command {
-- 
2.51.0


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: [PATCH 02/16] 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 +++++++++++++++----
 .../gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.h    | 14 ++++++++
 2 files changed, 40 insertions(+), 7 deletions(-)

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;
 		}
 	}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.h
index b9ebda577797..d3a6f7205a2f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.h
@@ -43,6 +43,19 @@ enum amdgpu_ras_eeprom_err_type {
 	AMDGPU_RAS_EEPROM_ERR_COUNT,
 };
 
+/*
+ * one UMC MCA address could map to multiply physical address (PA),
+ * such as 1:16, we use eeprom_table_record.address to store MCA
+ * address and use eeprom_table_record.retired_page to save PA.
+ *
+ * AMDGPU_RAS_EEPROM_REC_PA: one record store one PA
+ * AMDGPU_RAS_EEPROM_REC_MCA: one record store one MCA address
+ */
+enum amdgpu_ras_eeprom_rec_type {
+	AMDGPU_RAS_EEPROM_REC_PA,
+	AMDGPU_RAS_EEPROM_REC_MCA,
+};
+
 struct amdgpu_ras_eeprom_table_header {
 	uint32_t header;
 	uint32_t version;
@@ -102,6 +115,7 @@ struct amdgpu_ras_eeprom_control {
 	/* Record channel info which occurred bad pages
 	 */
 	u32 bad_channel_bitmap;
+	enum amdgpu_ras_eeprom_rec_type rec_type;
 };
 
 /*
-- 
2.51.0


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: [PATCH 03/16] 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(-)

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);
 
-- 
2.51.0


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: [PATCH 04/16] 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/aldebaran.c        |  2 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c    | 15 ++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c       | 55 ++++++++++++++++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h       | 11 +++-
 .../gpu/drm/amd/amdkfd/kfd_int_process_v9.c   |  2 +
 5 files changed, 78 insertions(+), 7 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/aldebaran.c b/drivers/gpu/drm/amd/amdgpu/aldebaran.c
index f44de9d4b6a1..e13fbd974141 100644
--- a/drivers/gpu/drm/amd/amdgpu/aldebaran.c
+++ b/drivers/gpu/drm/amd/amdgpu/aldebaran.c
@@ -334,6 +334,8 @@ aldebaran_mode2_restore_hwcontext(struct amdgpu_reset_control *reset_ctl,
 				AMDGPU_INIT_LEVEL_RESET_RECOVERY);
 		dev_info(tmp_adev->dev,
 			 "GPU reset succeeded, trying to resume\n");
+		/*TBD: Ideally should clear only GFX, SDMA blocks*/
+		amdgpu_ras_clear_err_state(tmp_adev);
 		r = aldebaran_mode2_restore_ip(tmp_adev);
 		if (r)
 			goto end;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index d272d95dd5b2..97d3e5f29638 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -5181,7 +5181,7 @@ static int amdgpu_device_reset_sriov(struct amdgpu_device *adev,
 	if (r)
 		return r;
 
-	amdgpu_ras_set_fed(adev, false);
+	amdgpu_ras_clear_err_state(adev);
 	amdgpu_irq_gpu_reset_resume_helper(adev);
 
 	/* some sw clean up VF needs to do before recover */
@@ -5484,7 +5484,7 @@ int amdgpu_device_reinit_after_reset(struct amdgpu_reset_context *reset_context)
 		amdgpu_set_init_level(tmp_adev, init_level);
 		if (full_reset) {
 			/* post card */
-			amdgpu_ras_set_fed(tmp_adev, false);
+			amdgpu_ras_clear_err_state(tmp_adev);
 			r = amdgpu_device_asic_init(tmp_adev);
 			if (r) {
 				dev_warn(tmp_adev->dev, "asic atom init failed!");
@@ -5817,6 +5817,17 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
 	bool audio_suspended = false;
 	int retry_limit = AMDGPU_MAX_RETRY_LIMIT;
 
+	/*
+	 * If it reaches here because of hang/timeout and a RAS error is
+	 * detected at the same time, let RAS recovery take care of it.
+	 */
+	if (amdgpu_ras_is_err_state(adev, AMDGPU_RAS_BLOCK__ANY) &&
+	    reset_context->src != AMDGPU_RESET_SRC_RAS) {
+		dev_dbg(adev->dev,
+			"Gpu recovery from source: %d yielding to RAS error recovery handling",
+			reset_context->src);
+		return 0;
+	}
 	/*
 	 * Special case: RAS triggered and full reset isn't supported
 	 */
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)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
index 6db772ecfee4..b13debcf48ee 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
@@ -99,7 +99,8 @@ enum amdgpu_ras_block {
 	AMDGPU_RAS_BLOCK__IH,
 	AMDGPU_RAS_BLOCK__MPIO,
 
-	AMDGPU_RAS_BLOCK__LAST
+	AMDGPU_RAS_BLOCK__LAST,
+	AMDGPU_RAS_BLOCK__ANY = -1
 };
 
 enum amdgpu_ras_mca_block {
@@ -558,8 +559,8 @@ struct amdgpu_ras {
 	struct ras_ecc_log_info  umc_ecc_log;
 	struct delayed_work page_retirement_dwork;
 
-	/* Fatal error detected flag */
-	atomic_t fed;
+	/* ras errors detected */
+	unsigned long ras_err_state;
 
 	/* RAS event manager */
 	struct ras_event_manager __event_mgr;
@@ -952,6 +953,10 @@ ssize_t amdgpu_ras_aca_sysfs_read(struct device *dev, struct device_attribute *a
 
 void amdgpu_ras_set_fed(struct amdgpu_device *adev, bool status);
 bool amdgpu_ras_get_fed_status(struct amdgpu_device *adev);
+void amdgpu_ras_set_err_poison(struct amdgpu_device *adev,
+			       enum amdgpu_ras_block block);
+void amdgpu_ras_clear_err_state(struct amdgpu_device *adev);
+bool amdgpu_ras_is_err_state(struct amdgpu_device *adev, int block);
 
 u64 amdgpu_ras_acquire_event_id(struct amdgpu_device *adev, enum ras_event_type type);
 int amdgpu_ras_mark_ras_event_caller(struct amdgpu_device *adev, enum ras_event_type type,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
index d46a13156ee9..0cb5c582ce7d 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
@@ -184,6 +184,7 @@ static void event_interrupt_poison_consumption_v9(struct kfd_node *dev,
 		} else {
 			reset = AMDGPU_RAS_GPU_RESET_MODE2_RESET;
 		}
+		amdgpu_ras_set_err_poison(dev->adev, AMDGPU_RAS_BLOCK__GFX);
 		break;
 	case SOC15_IH_CLIENTID_VMC:
 	case SOC15_IH_CLIENTID_VMC1:
@@ -213,6 +214,7 @@ static void event_interrupt_poison_consumption_v9(struct kfd_node *dev,
 		} else {
 			reset = AMDGPU_RAS_GPU_RESET_MODE2_RESET;
 		}
+		amdgpu_ras_set_err_poison(dev->adev, AMDGPU_RAS_BLOCK__SDMA);
 		break;
 	default:
 		dev_warn(dev->adev->dev,
-- 
2.51.0


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: [PATCH 05/16] 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(-)

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);
-- 
2.51.0


From 19d4b27aedc73d2f5785bdef7c30fe49c16606e7 Mon Sep 17 00:00:00 2001
From: Tao Zhou <tao.zhou1@amd.com>
Date: Thu, 24 Oct 2024 18:51:13 +0800
Subject: [PATCH 06/16] drm/amdgpu: retire RAS bad pages in different NPS modes

There are some changes in format of memory normalized address per
NPS mode, need to adjust bit mapping according 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/umc_v12_0.c | 64 +++++++++++++++++---------
 drivers/gpu/drm/amd/amdgpu/umc_v12_0.h | 11 +++++
 2 files changed, 52 insertions(+), 23 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c b/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c
index ce60fd6675ce..17ef9a6743f5 100644
--- a/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c
@@ -179,10 +179,13 @@ static int umc_v12_0_convert_error_address(struct amdgpu_device *adev,
 					struct ta_ras_query_address_output *addr_out,
 					bool dump_addr)
 {
-	uint32_t col, row, bank, channel_index, umc_inst = 0;
-	uint64_t soc_pa, retired_page, column, err_addr;
+	uint32_t col, col_lower, row, row_lower, bank;
+	uint32_t channel_index, umc_inst = 0;
+	uint32_t i, loop_bits[UMC_V12_0_RETIRE_LOOP_BITS];
+	uint64_t soc_pa, column, err_addr;
 	struct ta_ras_query_address_output addr_out_tmp;
 	struct ta_ras_query_address_output *paddr_out;
+	enum amdgpu_memory_partition nps = AMDGPU_NPS1_PARTITION_MODE;
 	int ret = 0;
 
 	if (!addr_out)
@@ -199,7 +202,7 @@ static int umc_v12_0_convert_error_address(struct amdgpu_device *adev,
 			dev_warn(adev->dev, "Failed to query RAS physical address for 0x%llx",
 				err_addr);
 
-			return ret;
+			goto out;
 		}
 
 		bank = paddr_out->pa.bank;
@@ -208,42 +211,57 @@ static int umc_v12_0_convert_error_address(struct amdgpu_device *adev,
 		umc_inst = addr_in->ma.umc_inst;
 	}
 
-	soc_pa = paddr_out->pa.pa;
+	loop_bits[0] = UMC_V12_0_PA_C2_BIT;
+	loop_bits[1] = UMC_V12_0_PA_C3_BIT;
+	loop_bits[2] = UMC_V12_0_PA_C4_BIT;
+	loop_bits[3] = UMC_V12_0_PA_R13_BIT;
 
-	if (!err_data && !dump_addr)
-		return ret;
+	if (adev->gmc.gmc_funcs->query_mem_partition_mode)
+		nps = adev->gmc.gmc_funcs->query_mem_partition_mode(adev);
+
+	/* other nps modes are taken as nps1 */
+	if (nps == AMDGPU_NPS4_PARTITION_MODE) {
+		loop_bits[0] = UMC_V12_0_PA_CH4_BIT;
+		loop_bits[1] = UMC_V12_0_PA_CH5_BIT;
+		loop_bits[2] = UMC_V12_0_PA_B0_BIT;
+		loop_bits[3] = UMC_V12_0_PA_R11_BIT;
+	}
 
-	col = (err_addr >> 1) & 0x1fULL;
-	/* clear [C3 C2] in soc physical address */
-	soc_pa &= ~(0x3ULL << UMC_V12_0_PA_C2_BIT);
-	/* clear [C4] in soc physical address */
-	soc_pa &= ~(0x1ULL << UMC_V12_0_PA_C4_BIT);
-	/* clear [R13] in soc physical address */
-	soc_pa &= ~(0x1ULL << UMC_V12_0_PA_R13_BIT);
+	soc_pa = paddr_out->pa.pa;
+	/* clear loop bits in soc physical address */
+	for (i = 0; i < UMC_V12_0_RETIRE_LOOP_BITS; i++)
+		soc_pa &= ~BIT_ULL(loop_bits[i]);
 
 	paddr_out->pa.pa = soc_pa;
+	/* get column bit 0 and 1 in mca address */
+	col_lower = (err_addr >> 1) & 0x3ULL;
+	/* MA_R13_BIT will be handled later */
+	row_lower = (err_addr >> UMC_V12_0_MA_R0_BIT) & 0x1fffULL;
+
+	if (!err_data && !dump_addr)
+		goto out;
 
-	/* loop for all possibilities of [R13 C4 C3 C2] */
+	/* loop for all possibilities of retired bits */
 	for (column = 0; column < UMC_V12_0_BAD_PAGE_NUM_PER_CHANNEL; column++) {
-		retired_page = soc_pa | ((column & 0x3) << UMC_V12_0_PA_C2_BIT);
-		retired_page |= (((column & 0x4) >> 2) << UMC_V12_0_PA_C4_BIT);
-		retired_page |= (((column & 0x8) >> 3) << UMC_V12_0_PA_R13_BIT);
+		soc_pa = paddr_out->pa.pa;
+		for (i = 0; i < UMC_V12_0_RETIRE_LOOP_BITS; i++)
+			soc_pa |= (((column >> i) & 0x1ULL) << loop_bits[i]);
 
-		/* include column bit 0 and 1 */
-		col &= 0x3;
-		col |= (column << 2);
-		row = (retired_page >> UMC_V12_0_PA_R0_BIT) & 0x3fffULL;
+		col = ((column & 0x7) << 2) | col_lower;
+		/* add row bit 13 */
+		row = ((column >> 3) << 13) | row_lower;
 
 		if (dump_addr)
 			dev_info(adev->dev,
 				"Error Address(PA):0x%-10llx Row:0x%-4x Col:0x%-2x Bank:0x%x Channel:0x%x\n",
-				retired_page, row, col, bank, channel_index);
+				soc_pa, row, col, bank, channel_index);
 
 		if (err_data)
 			amdgpu_umc_fill_error_record(err_data, err_addr,
-				retired_page, channel_index, umc_inst);
+				soc_pa, channel_index, umc_inst);
 	}
 
+out:
 	return ret;
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/umc_v12_0.h b/drivers/gpu/drm/amd/amdgpu/umc_v12_0.h
index f0074abb5381..9298018d938f 100644
--- a/drivers/gpu/drm/amd/amdgpu/umc_v12_0.h
+++ b/drivers/gpu/drm/amd/amdgpu/umc_v12_0.h
@@ -55,13 +55,24 @@
 #define UMC_V12_0_NA_MAP_PA_NUM        8
 /* R13 bit shift should be considered, double the number */
 #define UMC_V12_0_BAD_PAGE_NUM_PER_CHANNEL (UMC_V12_0_NA_MAP_PA_NUM * 2)
+/* C2, C3, C4, R13, four bits in MCA address are looped in retirement */
+#define UMC_V12_0_RETIRE_LOOP_BITS 4
 
 /* column bits in SOC physical address */
 #define UMC_V12_0_PA_C2_BIT 15
+#define UMC_V12_0_PA_C3_BIT 16
 #define UMC_V12_0_PA_C4_BIT 21
 /* row bits in SOC physical address */
 #define UMC_V12_0_PA_R0_BIT 22
+#define UMC_V12_0_PA_R11_BIT 33
 #define UMC_V12_0_PA_R13_BIT 35
+/* channel bit in SOC physical address */
+#define UMC_V12_0_PA_CH4_BIT 12
+#define UMC_V12_0_PA_CH5_BIT 13
+/* bank bit in SOC physical address */
+#define UMC_V12_0_PA_B0_BIT 19
+/* row bits in MCA address */
+#define UMC_V12_0_MA_R0_BIT 10
 
 #define MCA_UMC_HWID_V12_0     0x96
 #define MCA_UMC_MCATYPE_V12_0  0x0
-- 
2.51.0


From b02ef407729d9eb26fbc74ca229e28dbe971e1c1 Mon Sep 17 00:00:00 2001
From: Tao Zhou <tao.zhou1@amd.com>
Date: Thu, 24 Oct 2024 14:46:19 +0800
Subject: [PATCH 07/16] drm/amdgpu: add function to find all memory pages in
 one physical row

And the function can be reused across amdgpu driver.

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_umc.c | 38 +++++++++++++++----------
 drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h |  2 ++
 2 files changed, 25 insertions(+), 15 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
index 64e1a3406c99..984af815fb38 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
@@ -445,11 +445,27 @@ int amdgpu_umc_logs_ecc_err(struct amdgpu_device *adev,
 	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)
 {
-	uint32_t i, ret = 0, pos = 0;
-	struct ta_ras_query_address_output addr_out;
+	int i, ret;
 	struct ras_err_data err_data;
 
 	err_data.err_addr = kcalloc(adev->umc.retire_unit,
@@ -459,25 +475,17 @@ int amdgpu_umc_lookup_bad_pages_in_a_row(struct amdgpu_device *adev,
 		return 0;
 	}
 
-	addr_out.pa.pa = pa_addr;
-
-	if (adev->umc.ras && adev->umc.ras->convert_ras_err_addr) {
-		ret = adev->umc.ras->convert_ras_err_addr(adev, &err_data, NULL,
-				&addr_out, false);
-		if (ret)
-			goto out;
-	} else {
+	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 (pos >= len)
+		if (i >= len)
 			goto out;
 
-		pfns[pos] = err_data.err_addr[pos].retired_page;
-		pos++;
+		pfns[i] = err_data.err_addr[i].retired_page;
 	}
-	ret = pos;
+	ret = i;
 
 out:
 	kfree(err_data.err_addr);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h
index f45408a6ff03..ce1e4fb385b5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h
@@ -139,6 +139,8 @@ int amdgpu_umc_logs_ecc_err(struct amdgpu_device *adev,
 
 void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev,
 			void *ras_error_status);
+int amdgpu_umc_pages_in_a_row(struct amdgpu_device *adev,
+			struct ras_err_data *err_data, uint64_t pa_addr);
 int amdgpu_umc_lookup_bad_pages_in_a_row(struct amdgpu_device *adev,
 			uint64_t pa_addr, uint64_t *pfns, int len);
 int amdgpu_umc_mca_to_addr(struct amdgpu_device *adev,
-- 
2.51.0


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: [PATCH 08/16] 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(-)

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 {
-- 
2.51.0


From 71a0e9630027f77d7646c5b750593c9ecfaa27d3 Mon Sep 17 00:00:00 2001
From: Tao Zhou <tao.zhou1@amd.com>
Date: Tue, 29 Oct 2024 19:46:44 +0800
Subject: [PATCH 09/16] drm/amdgpu: save UMC global channel index to eeprom

Save the global channel index returned by RAS TA to eeprom.
We can get memory physical address by MCA address and channel index.

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.h |  2 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c |  7 ++-----
 drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h |  2 +-
 drivers/gpu/drm/amd/amdgpu/umc_v12_0.c  | 13 ++++++++-----
 4 files changed, 13 insertions(+), 11 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
index b13debcf48ee..a6473c7cdeb6 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
@@ -483,6 +483,8 @@ struct ras_ecc_err {
 	uint64_t ipid;
 	uint64_t addr;
 	uint64_t pa_pfn;
+	/* save global channel index across all UMC instances */
+	uint32_t channel_idx;
 	struct ras_err_pages err_pages;
 };
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
index 984af815fb38..4107b78d9dda 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
@@ -495,10 +495,9 @@ out:
 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,
-			uint64_t *addr, bool dump_addr)
+			struct ta_ras_query_address_output *addr_out, bool dump_addr)
 {
 	struct ta_ras_query_address_input addr_in;
-	struct ta_ras_query_address_output addr_out;
 	int ret;
 
 	memset(&addr_in, 0, sizeof(addr_in));
@@ -510,14 +509,12 @@ int amdgpu_umc_mca_to_addr(struct amdgpu_device *adev,
 
 	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);
+				addr_out, dump_addr);
 		if (ret)
 			return ret;
 	} else {
 		return 0;
 	}
 
-	*addr = addr_out.pa.pa;
-
 	return 0;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h
index ce1e4fb385b5..2f71194d5da8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h
@@ -146,5 +146,5 @@ int amdgpu_umc_lookup_bad_pages_in_a_row(struct amdgpu_device *adev,
 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,
-			uint64_t *addr, bool dump_addr);
+			struct ta_ras_query_address_output *addr_out, bool dump_addr);
 #endif
diff --git a/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c b/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c
index 17ef9a6743f5..cce93b4ffb58 100644
--- a/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c
@@ -180,7 +180,7 @@ static int umc_v12_0_convert_error_address(struct amdgpu_device *adev,
 					bool dump_addr)
 {
 	uint32_t col, col_lower, row, row_lower, bank;
-	uint32_t channel_index, umc_inst = 0;
+	uint32_t channel_index = 0, umc_inst = 0;
 	uint32_t i, loop_bits[UMC_V12_0_RETIRE_LOOP_BITS];
 	uint64_t soc_pa, column, err_addr;
 	struct ta_ras_query_address_output addr_out_tmp;
@@ -193,7 +193,7 @@ static int umc_v12_0_convert_error_address(struct amdgpu_device *adev,
 	else
 		paddr_out = addr_out;
 
-	err_addr = bank = channel_index = 0;
+	err_addr = bank = 0;
 	if (addr_in) {
 		err_addr = addr_in->ma.err_addr;
 		addr_in->addr_type = TA_RAS_MCA_TO_PA;
@@ -206,7 +206,6 @@ static int umc_v12_0_convert_error_address(struct amdgpu_device *adev,
 		}
 
 		bank = paddr_out->pa.bank;
-		channel_index = paddr_out->pa.channel_idx;
 		/* no need to care about umc inst if addr_in is NULL */
 		umc_inst = addr_in->ma.umc_inst;
 	}
@@ -228,6 +227,7 @@ static int umc_v12_0_convert_error_address(struct amdgpu_device *adev,
 	}
 
 	soc_pa = paddr_out->pa.pa;
+	channel_index = paddr_out->pa.channel_idx;
 	/* clear loop bits in soc physical address */
 	for (i = 0; i < UMC_V12_0_RETIRE_LOOP_BITS; i++)
 		soc_pa &= ~BIT_ULL(loop_bits[i]);
@@ -466,6 +466,7 @@ static int umc_v12_0_update_ecc_status(struct amdgpu_device *adev,
 	uint64_t page_pfn[UMC_V12_0_BAD_PAGE_NUM_PER_CHANNEL];
 	uint64_t err_addr, pa_addr = 0;
 	struct ras_ecc_err *ecc_err;
+	struct ta_ras_query_address_output addr_out;
 	int count, ret, i;
 
 	hwid = REG_GET_FIELD(ipid, MCMP1_IPIDT0, HardwareID);
@@ -495,7 +496,7 @@ static int umc_v12_0_update_ecc_status(struct amdgpu_device *adev,
 	ret = amdgpu_umc_mca_to_addr(adev,
 			err_addr, MCA_IPID_2_UMC_CH(ipid),
 			MCA_IPID_2_UMC_INST(ipid), MCA_IPID_2_DIE_ID(ipid),
-			MCA_IPID_2_SOCKET_ID(ipid), &pa_addr, true);
+			MCA_IPID_2_SOCKET_ID(ipid), &addr_out, true);
 	if (ret)
 		return ret;
 
@@ -503,10 +504,12 @@ static int umc_v12_0_update_ecc_status(struct amdgpu_device *adev,
 	if (!ecc_err)
 		return -ENOMEM;
 
+	pa_addr = addr_out.pa.pa;
 	ecc_err->status = status;
 	ecc_err->ipid = ipid;
 	ecc_err->addr = addr;
 	ecc_err->pa_pfn = pa_addr >> AMDGPU_GPU_PAGE_SHIFT;
+	ecc_err->channel_idx = addr_out.pa.channel_idx;
 
 	/* If converted pa_pfn is 0, use pa C4 pfn. */
 	if (!ecc_err->pa_pfn)
@@ -577,7 +580,7 @@ static int umc_v12_0_fill_error_record(struct amdgpu_device *adev,
 		ret = amdgpu_umc_fill_error_record(err_data,
 				ecc_err->addr,
 				page_pfn[i] << AMDGPU_GPU_PAGE_SHIFT,
-				MCA_IPID_2_UMC_CH(ecc_err->ipid),
+				ecc_err->channel_idx,
 				MCA_IPID_2_UMC_INST(ecc_err->ipid));
 		if (ret)
 			break;
-- 
2.51.0


From 2206daa1f9182a08aea92213dc2bcb387173dab1 Mon Sep 17 00:00:00 2001
From: Tao Zhou <tao.zhou1@amd.com>
Date: Wed, 30 Oct 2024 14:17:49 +0800
Subject: [PATCH 10/16] drm/amdgpu: add a flag to indicate UMC channel index
 version

v1 (legacy way): store channel index within a UMC instance in eeprom
v2: store global channel index in eeprom

V2: only save the flag on eeprom, clear it after saving.

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_eeprom.c | 11 ++++++++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h        | 16 ++++++++++++++++
 2 files changed, 26 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c
index f28f6b4ba765..f4a9e15389ae 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c
@@ -841,7 +841,7 @@ int amdgpu_ras_eeprom_append(struct amdgpu_ras_eeprom_control *control,
 			     const u32 num)
 {
 	struct amdgpu_device *adev = to_amdgpu_device(control);
-	int res;
+	int res, i;
 
 	if (!__is_ras_eeprom_supported(adev))
 		return 0;
@@ -855,6 +855,10 @@ int amdgpu_ras_eeprom_append(struct amdgpu_ras_eeprom_control *control,
 		return -EINVAL;
 	}
 
+	/* set the new channel index flag */
+	for (i = 0; i < num; i++)
+		record[i].retired_page |= UMC_CHANNEL_IDX_V2;
+
 	mutex_lock(&control->ras_tbl_mutex);
 
 	res = amdgpu_ras_eeprom_append_table(control, record, num);
@@ -864,6 +868,11 @@ int amdgpu_ras_eeprom_append(struct amdgpu_ras_eeprom_control *control,
 		amdgpu_ras_debugfs_set_ret_size(control);
 
 	mutex_unlock(&control->ras_tbl_mutex);
+
+	/* clear channel index flag, the flag is only saved on eeprom */
+	for (i = 0; i < num; i++)
+		record[i].retired_page &= ~UMC_CHANNEL_IDX_V2;
+
 	return res;
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h
index 2f71194d5da8..f97c45b4eeb8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h
@@ -54,6 +54,22 @@
 
 /* Page retirement tag */
 #define UMC_ECC_NEW_DETECTED_TAG       0x1
+/*
+ * a flag to indicate v2 of channel index stored in eeprom
+ *
+ * v1 (legacy way): store channel index within a umc instance in eeprom
+ *    range in UMC v12: 0 ~ 7
+ * v2: store global channel index in eeprom
+ *    range in UMC v12: 0 ~ 127
+ *
+ * NOTE: it's better to store it in eeprom_table_record.mem_channel,
+ * but there is only 8 bits in mem_channel, and the channel number may
+ * increase in the future, we decide to save it in
+ * eeprom_table_record.retired_page. retired_page is useless in v2,
+ * we depend on eeprom_table_record.address instead of retired_page in v2.
+ * Only 48 bits are saved on eeprom, use bit 47 here.
+ */
+#define UMC_CHANNEL_IDX_V2	BIT_ULL(47)
 
 typedef int (*umc_func)(struct amdgpu_device *adev, uint32_t node_inst,
 			uint32_t umc_inst, uint32_t ch_inst, void *data);
-- 
2.51.0


From fcb600b07894913a9e73d8c1a1ce86a36324e4a7 Mon Sep 17 00:00:00 2001
From: Tao Zhou <tao.zhou1@amd.com>
Date: Wed, 30 Oct 2024 16:42:42 +0800
Subject: [PATCH 11/16] drm/amdgpu: add interface to get die id from memory
 address

And implement it for UMC v12_0. The die id is calculated from IPID
register in bad page retirement flow, but we don't store it on eeprom
and it can be also gotten from physical address.

v2: get PA_C4 and PA_R13 from MCA address since they may be cleared in
retired page.

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_umc.h |  2 ++
 drivers/gpu/drm/amd/amdgpu/umc_v12_0.c  | 26 +++++++++++++++++++++++++
 2 files changed, 28 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h
index f97c45b4eeb8..a4a7e61817aa 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h
@@ -91,6 +91,8 @@ struct amdgpu_umc_ras {
 			struct ta_ras_query_address_input *addr_in,
 			struct ta_ras_query_address_output *addr_out,
 			bool dump_addr);
+	uint32_t (*get_die_id_from_pa)(struct amdgpu_device *adev,
+			uint64_t mca_addr, uint64_t retired_page);
 };
 
 struct amdgpu_umc_funcs {
diff --git a/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c b/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c
index cce93b4ffb58..30ee4cb9aaab 100644
--- a/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c
@@ -619,6 +619,31 @@ static void umc_v12_0_query_ras_ecc_err_addr(struct amdgpu_device *adev,
 	mutex_unlock(&con->umc_ecc_log.lock);
 }
 
+static uint32_t umc_v12_0_get_die_id(struct amdgpu_device *adev,
+		uint64_t mca_addr, uint64_t retired_page)
+{
+	uint32_t die = 0;
+
+	/* we only calculate die id for nps1 mode right now */
+	die += ((((retired_page >> 12) & 0x1ULL)^
+	    ((retired_page >> 20) & 0x1ULL) ^
+	    ((retired_page >> 27) & 0x1ULL) ^
+	    ((retired_page >> 34) & 0x1ULL) ^
+	    ((retired_page >> 41) & 0x1ULL)) << 0);
+
+	/* the original PA_C4 and PA_R13 may be cleared in retired_page, so
+	 * get them from mca_addr.
+	 */
+	die += ((((retired_page >> 13) & 0x1ULL) ^
+	    ((mca_addr >> 5) & 0x1ULL) ^
+	    ((retired_page >> 28) & 0x1ULL) ^
+	    ((mca_addr >> 23) & 0x1ULL) ^
+	    ((retired_page >> 42) & 0x1ULL)) << 1);
+	die &= 3;
+
+	return die;
+}
+
 struct amdgpu_umc_ras umc_v12_0_ras = {
 	.ras_block = {
 		.hw_ops = &umc_v12_0_ras_hw_ops,
@@ -630,5 +655,6 @@ struct amdgpu_umc_ras umc_v12_0_ras = {
 	.check_ecc_err_status = umc_v12_0_check_ecc_err_status,
 	.update_ecc_status = umc_v12_0_update_ecc_status,
 	.convert_ras_err_addr = umc_v12_0_convert_error_address,
+	.get_die_id_from_pa = umc_v12_0_get_die_id,
 };
 
-- 
2.51.0


From a3300782d5375e280ba7040f323d01960bfe3396 Mon Sep 17 00:00:00 2001
From: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Date: Tue, 19 Nov 2024 21:23:18 +0100
Subject: [PATCH 12/16] drm/amd/pm: Fix an error handling path in
 vega10_enable_se_edc_force_stall_config()

In case of error after a amdgpu_gfx_rlc_enter_safe_mode() call, it is not
balanced by a corresponding amdgpu_gfx_rlc_exit_safe_mode() call.

Add the missing call.

Fixes: 9b7b8154cdb8 ("drm/amd/powerplay: added didt support for vega10")
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/pm/powerplay/hwmgr/vega10_powertune.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/pm/powerplay/hwmgr/vega10_powertune.c b/drivers/gpu/drm/amd/pm/powerplay/hwmgr/vega10_powertune.c
index 3007b054c873..776d58ea63ae 100644
--- a/drivers/gpu/drm/amd/pm/powerplay/hwmgr/vega10_powertune.c
+++ b/drivers/gpu/drm/amd/pm/powerplay/hwmgr/vega10_powertune.c
@@ -1120,13 +1120,14 @@ static int vega10_enable_se_edc_force_stall_config(struct pp_hwmgr *hwmgr)
 	result = vega10_program_didt_config_registers(hwmgr, SEEDCForceStallPatternConfig_Vega10, VEGA10_CONFIGREG_DIDT);
 	result |= vega10_program_didt_config_registers(hwmgr, SEEDCCtrlForceStallConfig_Vega10, VEGA10_CONFIGREG_DIDT);
 	if (0 != result)
-		return result;
+		goto exit_safe_mode;
 
 	vega10_didt_set_mask(hwmgr, false);
 
+exit_safe_mode:
 	amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
 
-	return 0;
+	return result;
 }
 
 static int vega10_disable_se_edc_force_stall_config(struct pp_hwmgr *hwmgr)
-- 
2.51.0


From 8aaf166703751ffd9a9fbc4d8b996f538b278983 Mon Sep 17 00:00:00 2001
From: Boyuan Zhang <boyuan.zhang@amd.com>
Date: Sun, 29 Sep 2024 15:17:29 -0400
Subject: [PATCH 13/16] drm/amd/pm: power up or down vcn by instance
MIME-Version: 1.0
Content-Type: text/plain; charset=utf8
Content-Transfer-Encoding: 8bit

For smu ip with multiple vcn instances (smu 11/13/14), remove all the
for loop in dpm_set_vcn_enable() functions. And use the instance
argument to power up/down vcn for the given instance only, instead
of powering up/down for all vcn instances.

v2: remove all duplicated functions in v1.

remove for-loop from each ip, and temporarily move to dpm_set_vcn_enable,
in order to keep the exact same logic as before, until further separation
in the next patch.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c     |  9 +++--
 .../amd/pm/swsmu/smu11/sienna_cichlid_ppt.c   | 20 +++++------
 .../gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c    | 16 ++++-----
 .../gpu/drm/amd/pm/swsmu/smu14/smu_v14_0.c    | 35 ++++++++-----------
 4 files changed, 35 insertions(+), 45 deletions(-)

diff --git a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
index 6d22c7783898..76dda38ae57f 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
@@ -242,6 +242,7 @@ static int smu_dpm_set_vcn_enable(struct smu_context *smu,
 {
 	struct smu_power_context *smu_power = &smu->smu_power;
 	struct smu_power_gate *power_gate = &smu_power->power_gate;
+	struct amdgpu_device *adev = smu->adev;
 	int ret = 0;
 
 	/*
@@ -256,9 +257,11 @@ static int smu_dpm_set_vcn_enable(struct smu_context *smu,
 	if (atomic_read(&power_gate->vcn_gated) ^ enable)
 		return 0;
 
-	ret = smu->ppt_funcs->dpm_set_vcn_enable(smu, enable, 0xff);
-	if (!ret)
-		atomic_set(&power_gate->vcn_gated, !enable);
+	for (int i = 0; i < adev->vcn.num_vcn_inst; i++) {
+		ret = smu->ppt_funcs->dpm_set_vcn_enable(smu, enable, i);
+		if (ret)
+			return ret;
+	}
 
 	return ret;
 }
diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu11/sienna_cichlid_ppt.c b/drivers/gpu/drm/amd/pm/swsmu/smu11/sienna_cichlid_ppt.c
index 286777ada1df..19a25fdc2f5b 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/smu11/sienna_cichlid_ppt.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/smu11/sienna_cichlid_ppt.c
@@ -1157,19 +1157,15 @@ static int sienna_cichlid_dpm_set_vcn_enable(struct smu_context *smu,
 					      int inst)
 {
 	struct amdgpu_device *adev = smu->adev;
-	int i, ret = 0;
+	int ret = 0;
 
-	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
-		if (adev->vcn.harvest_config & (1 << i))
-			continue;
-		/* vcn dpm on is a prerequisite for vcn power gate messages */
-		if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_MM_DPM_PG_BIT)) {
-			ret = smu_cmn_send_smc_msg_with_param(smu, enable ?
-							      SMU_MSG_PowerUpVcn : SMU_MSG_PowerDownVcn,
-							      0x10000 * i, NULL);
-			if (ret)
-				return ret;
-		}
+	if (adev->vcn.harvest_config & (1 << inst))
+		return ret;
+	/* vcn dpm on is a prerequisite for vcn power gate messages */
+	if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_MM_DPM_PG_BIT)) {
+		ret = smu_cmn_send_smc_msg_with_param(smu, enable ?
+						      SMU_MSG_PowerUpVcn : SMU_MSG_PowerDownVcn,
+						      0x10000 * inst, NULL);
 	}
 
 	return ret;
diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c
index 2bfea740dace..bb506d15d787 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c
@@ -2108,18 +2108,14 @@ int smu_v13_0_set_vcn_enable(struct smu_context *smu,
 			      int inst)
 {
 	struct amdgpu_device *adev = smu->adev;
-	int i, ret = 0;
+	int ret = 0;
 
-	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
-		if (adev->vcn.harvest_config & (1 << i))
-			continue;
+	if (adev->vcn.harvest_config & (1 << inst))
+		return ret;
 
-		ret = smu_cmn_send_smc_msg_with_param(smu, enable ?
-						      SMU_MSG_PowerUpVcn : SMU_MSG_PowerDownVcn,
-						      i << 16U, NULL);
-		if (ret)
-			return ret;
-	}
+	ret = smu_cmn_send_smc_msg_with_param(smu, enable ?
+					      SMU_MSG_PowerUpVcn : SMU_MSG_PowerDownVcn,
+					      inst << 16U, NULL);
 
 	return ret;
 }
diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu14/smu_v14_0.c b/drivers/gpu/drm/amd/pm/swsmu/smu14/smu_v14_0.c
index a87040cb2f2e..4d083f7f772e 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/smu14/smu_v14_0.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/smu14/smu_v14_0.c
@@ -1511,29 +1511,24 @@ int smu_v14_0_set_vcn_enable(struct smu_context *smu,
 			      int inst)
 {
 	struct amdgpu_device *adev = smu->adev;
-	int i, ret = 0;
+	int ret = 0;
 
-	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
-		if (adev->vcn.harvest_config & (1 << i))
-			continue;
+	if (adev->vcn.harvest_config & (1 << inst))
+		return ret;
 
-		if (smu->is_apu) {
-			if (i == 0)
-				ret = smu_cmn_send_smc_msg_with_param(smu, enable ?
-								      SMU_MSG_PowerUpVcn0 : SMU_MSG_PowerDownVcn0,
-								      i << 16U, NULL);
-			else if (i == 1)
-				ret = smu_cmn_send_smc_msg_with_param(smu, enable ?
-								      SMU_MSG_PowerUpVcn1 : SMU_MSG_PowerDownVcn1,
-								      i << 16U, NULL);
-		} else {
+	if (smu->is_apu) {
+		if (inst == 0)
 			ret = smu_cmn_send_smc_msg_with_param(smu, enable ?
-							      SMU_MSG_PowerUpVcn : SMU_MSG_PowerDownVcn,
-							      i << 16U, NULL);
-		}
-
-		if (ret)
-			return ret;
+							      SMU_MSG_PowerUpVcn0 : SMU_MSG_PowerDownVcn0,
+							      inst << 16U, NULL);
+		else if (inst == 1)
+			ret = smu_cmn_send_smc_msg_with_param(smu, enable ?
+							      SMU_MSG_PowerUpVcn1 : SMU_MSG_PowerDownVcn1,
+							      inst << 16U, NULL);
+	} else {
+		ret = smu_cmn_send_smc_msg_with_param(smu, enable ?
+						      SMU_MSG_PowerUpVcn : SMU_MSG_PowerDownVcn,
+						      inst << 16U, NULL);
 	}
 
 	return ret;
-- 
2.51.0


From 15df736afe009856a40baf93df3fd10f4a32a88b Mon Sep 17 00:00:00 2001
From: Boyuan Zhang <boyuan.zhang@amd.com>
Date: Wed, 2 Oct 2024 21:59:33 -0400
Subject: [PATCH 14/16] drm/amd/pm: add inst to smu_dpm_set_vcn_enable
MIME-Version: 1.0
Content-Type: text/plain; charset=utf8
Content-Transfer-Encoding: 8bit

First, add an instance parameter to smu_dpm_set_vcn_enable() function,
and calling dpm_set_vcn_enable() with this given instance.

Second, modify vcn_gated to be an array, to track the gating status
for each vcn instance separately.

With these 2 changes, smu_dpm_set_vcn_enable() will check and set the
gating status for the given vcn instance ONLY.

v2: remove duplicated functions.

remove for-loop in dpm_set_vcn_enable(), and temporarily move it to
to smu_dpm_set_power_gate(), in order to keep the exact same logic as
before, until further separation in next patch.

v3: add instance number in error message.

v4: declaring i at the top of the function.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c     | 75 ++++++++++++-------
 drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h |  2 +-
 2 files changed, 47 insertions(+), 30 deletions(-)

diff --git a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
index 76dda38ae57f..2fd7909dac18 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
@@ -238,11 +238,11 @@ static bool is_vcn_enabled(struct amdgpu_device *adev)
 }
 
 static int smu_dpm_set_vcn_enable(struct smu_context *smu,
-				  bool enable)
+				   bool enable,
+				   int inst)
 {
 	struct smu_power_context *smu_power = &smu->smu_power;
 	struct smu_power_gate *power_gate = &smu_power->power_gate;
-	struct amdgpu_device *adev = smu->adev;
 	int ret = 0;
 
 	/*
@@ -254,14 +254,12 @@ static int smu_dpm_set_vcn_enable(struct smu_context *smu,
 	if (!smu->ppt_funcs->dpm_set_vcn_enable)
 		return 0;
 
-	if (atomic_read(&power_gate->vcn_gated) ^ enable)
+	if (atomic_read(&power_gate->vcn_gated[inst]) ^ enable)
 		return 0;
 
-	for (int i = 0; i < adev->vcn.num_vcn_inst; i++) {
-		ret = smu->ppt_funcs->dpm_set_vcn_enable(smu, enable, i);
-		if (ret)
-			return ret;
-	}
+	ret = smu->ppt_funcs->dpm_set_vcn_enable(smu, enable, inst);
+	if (!ret)
+		atomic_set(&power_gate->vcn_gated[inst], !enable);
 
 	return ret;
 }
@@ -363,7 +361,8 @@ static int smu_dpm_set_power_gate(void *handle,
 				  bool gate)
 {
 	struct smu_context *smu = handle;
-	int ret = 0;
+	struct amdgpu_device *adev = smu->adev;
+	int i, ret = 0;
 
 	if (!smu->pm_enabled || !smu->adev->pm.dpm_enabled) {
 		dev_WARN(smu->adev->dev,
@@ -379,10 +378,12 @@ static int smu_dpm_set_power_gate(void *handle,
 	 */
 	case AMD_IP_BLOCK_TYPE_UVD:
 	case AMD_IP_BLOCK_TYPE_VCN:
-		ret = smu_dpm_set_vcn_enable(smu, !gate);
-		if (ret)
-			dev_err(smu->adev->dev, "Failed to power %s VCN!\n",
-				gate ? "gate" : "ungate");
+		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
+			ret = smu_dpm_set_vcn_enable(smu, !gate, i);
+			if (ret)
+				dev_err(smu->adev->dev, "Failed to power %s VCN instance %d!\n",
+					gate ? "gate" : "ungate", i);
+		}
 		break;
 	case AMD_IP_BLOCK_TYPE_GFX:
 		ret = smu_gfx_off_control(smu, gate);
@@ -785,21 +786,25 @@ static int smu_set_default_dpm_table(struct smu_context *smu)
 	struct amdgpu_device *adev = smu->adev;
 	struct smu_power_context *smu_power = &smu->smu_power;
 	struct smu_power_gate *power_gate = &smu_power->power_gate;
-	int vcn_gate, jpeg_gate;
+	int vcn_gate[AMDGPU_MAX_VCN_INSTANCES], jpeg_gate, i;
 	int ret = 0;
 
 	if (!smu->ppt_funcs->set_default_dpm_table)
 		return 0;
 
-	if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
-		vcn_gate = atomic_read(&power_gate->vcn_gated);
+	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
+		for (i = 0; i < adev->vcn.num_vcn_inst; i++)
+			vcn_gate[i] = atomic_read(&power_gate->vcn_gated[i]);
+	}
 	if (adev->pg_flags & AMD_PG_SUPPORT_JPEG)
 		jpeg_gate = atomic_read(&power_gate->jpeg_gated);
 
 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
-		ret = smu_dpm_set_vcn_enable(smu, true);
-		if (ret)
-			return ret;
+		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
+			ret = smu_dpm_set_vcn_enable(smu, true, i);
+			if (ret)
+				return ret;
+		}
 	}
 
 	if (adev->pg_flags & AMD_PG_SUPPORT_JPEG) {
@@ -816,8 +821,10 @@ static int smu_set_default_dpm_table(struct smu_context *smu)
 	if (adev->pg_flags & AMD_PG_SUPPORT_JPEG)
 		smu_dpm_set_jpeg_enable(smu, !jpeg_gate);
 err_out:
-	if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
-		smu_dpm_set_vcn_enable(smu, !vcn_gate);
+	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
+		for (i = 0; i < adev->vcn.num_vcn_inst; i++)
+			smu_dpm_set_vcn_enable(smu, !vcn_gate[i], i);
+	}
 
 	return ret;
 }
@@ -1271,7 +1278,7 @@ static int smu_sw_init(struct amdgpu_ip_block *ip_block)
 {
 	struct amdgpu_device *adev = ip_block->adev;
 	struct smu_context *smu = adev->powerplay.pp_handle;
-	int ret;
+	int i, ret;
 
 	smu->pool_size = adev->pm.smu_prv_buffer_size;
 	smu->smu_feature.feature_num = SMU_FEATURE_MAX;
@@ -1283,7 +1290,8 @@ static int smu_sw_init(struct amdgpu_ip_block *ip_block)
 	atomic64_set(&smu->throttle_int_counter, 0);
 	smu->watermarks_bitmap = 0;
 
-	atomic_set(&smu->smu_power.power_gate.vcn_gated, 1);
+	for (i = 0; i < adev->vcn.num_vcn_inst; i++)
+		atomic_set(&smu->smu_power.power_gate.vcn_gated[i], 1);
 	atomic_set(&smu->smu_power.power_gate.jpeg_gated, 1);
 	atomic_set(&smu->smu_power.power_gate.vpe_gated, 1);
 	atomic_set(&smu->smu_power.power_gate.umsch_mm_gated, 1);
@@ -1813,7 +1821,7 @@ static int smu_start_smc_engine(struct smu_context *smu)
 
 static int smu_hw_init(struct amdgpu_ip_block *ip_block)
 {
-	int ret;
+	int i, ret;
 	struct amdgpu_device *adev = ip_block->adev;
 	struct smu_context *smu = adev->powerplay.pp_handle;
 
@@ -1839,7 +1847,8 @@ static int smu_hw_init(struct amdgpu_ip_block *ip_block)
 		ret = smu_set_gfx_imu_enable(smu);
 		if (ret)
 			return ret;
-		smu_dpm_set_vcn_enable(smu, true);
+		for (i = 0; i < adev->vcn.num_vcn_inst; i++)
+			smu_dpm_set_vcn_enable(smu, true, i);
 		smu_dpm_set_jpeg_enable(smu, true);
 		smu_dpm_set_vpe_enable(smu, true);
 		smu_dpm_set_umsch_mm_enable(smu, true);
@@ -2037,12 +2046,13 @@ static int smu_hw_fini(struct amdgpu_ip_block *ip_block)
 {
 	struct amdgpu_device *adev = ip_block->adev;
 	struct smu_context *smu = adev->powerplay.pp_handle;
-	int ret;
+	int i, ret;
 
 	if (amdgpu_sriov_vf(adev) && !amdgpu_sriov_is_pp_one_vf(adev))
 		return 0;
 
-	smu_dpm_set_vcn_enable(smu, false);
+	for (i = 0; i < adev->vcn.num_vcn_inst; i++)
+		smu_dpm_set_vcn_enable(smu, false, i);
 	smu_dpm_set_jpeg_enable(smu, false);
 	smu_dpm_set_vpe_enable(smu, false);
 	smu_dpm_set_umsch_mm_enable(smu, false);
@@ -2982,9 +2992,10 @@ static int smu_read_sensor(void *handle,
 			   int *size_arg)
 {
 	struct smu_context *smu = handle;
+	struct amdgpu_device *adev = smu->adev;
 	struct smu_umd_pstate_table *pstate_table =
 				&smu->pstate_table;
-	int ret = 0;
+	int i, ret = 0;
 	uint32_t *size, size_val;
 
 	if (!smu->pm_enabled || !smu->adev->pm.dpm_enabled)
@@ -3030,7 +3041,13 @@ static int smu_read_sensor(void *handle,
 		*size = 4;
 		break;
 	case AMDGPU_PP_SENSOR_VCN_POWER_STATE:
-		*(uint32_t *)data = atomic_read(&smu->smu_power.power_gate.vcn_gated) ? 0 : 1;
+		*(uint32_t *)data = 0;
+		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
+			if (!atomic_read(&smu->smu_power.power_gate.vcn_gated[i])) {
+				*(uint32_t *)data = 1;
+				break;
+			}
+		}
 		*size = 4;
 		break;
 	case AMDGPU_PP_SENSOR_MIN_FAN_RPM:
diff --git a/drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h b/drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h
index c73f84cb5111..3630593bce61 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h
+++ b/drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h
@@ -399,7 +399,7 @@ struct smu_dpm_context {
 struct smu_power_gate {
 	bool uvd_gated;
 	bool vce_gated;
-	atomic_t vcn_gated;
+	atomic_t vcn_gated[AMDGPU_MAX_VCN_INSTANCES];
 	atomic_t jpeg_gated;
 	atomic_t vpe_gated;
 	atomic_t umsch_mm_gated;
-- 
2.51.0


From 697cb5cc2549c5aa8da541ec5d815500edc35f88 Mon Sep 17 00:00:00 2001
From: Boyuan Zhang <boyuan.zhang@amd.com>
Date: Wed, 2 Oct 2024 23:25:45 -0400
Subject: [PATCH 15/16] drm/amd/pm: add inst to set_powergating_by_smu
MIME-Version: 1.0
Content-Type: text/plain; charset=utf8
Content-Transfer-Encoding: 8bit

Add an instance parameter to set_powergating_by_smu() function, and
re-write all amd_pm functions accordingly. Then use the instance to
call smu_dpm_set_vcn_enable().

v2: remove duplicated functions.

remove for-loop in smu_dpm_set_power_gate(), and temporarily move it to
to amdgpu_dpm_set_powergating_by_smu(), in order to keep the exact same
logic as before, until further separation in next patch.

v3: add instance number in error message.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/include/kgd_pp_interface.h   |  4 +++-
 drivers/gpu/drm/amd/pm/amdgpu_dpm.c              | 10 ++++++++--
 drivers/gpu/drm/amd/pm/legacy-dpm/kv_dpm.c       |  4 +++-
 drivers/gpu/drm/amd/pm/powerplay/amd_powerplay.c |  4 +++-
 drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c        | 16 +++++++---------
 5 files changed, 24 insertions(+), 14 deletions(-)

diff --git a/drivers/gpu/drm/amd/include/kgd_pp_interface.h b/drivers/gpu/drm/amd/include/kgd_pp_interface.h
index d7acdd42d80f..9189dcb65188 100644
--- a/drivers/gpu/drm/amd/include/kgd_pp_interface.h
+++ b/drivers/gpu/drm/amd/include/kgd_pp_interface.h
@@ -421,7 +421,9 @@ struct amd_pm_funcs {
 	int (*load_firmware)(void *handle);
 	int (*wait_for_fw_loading_complete)(void *handle);
 	int (*set_powergating_by_smu)(void *handle,
-				uint32_t block_type, bool gate);
+				uint32_t block_type,
+				bool gate,
+				int inst);
 	int (*set_clockgating_by_smu)(void *handle, uint32_t msg_id);
 	int (*set_power_limit)(void *handle, uint32_t n);
 	int (*get_power_limit)(void *handle, uint32_t *limit,
diff --git a/drivers/gpu/drm/amd/pm/amdgpu_dpm.c b/drivers/gpu/drm/amd/pm/amdgpu_dpm.c
index 9e7a652d119b..78e3488b6a35 100644
--- a/drivers/gpu/drm/amd/pm/amdgpu_dpm.c
+++ b/drivers/gpu/drm/amd/pm/amdgpu_dpm.c
@@ -88,7 +88,6 @@ int amdgpu_dpm_set_powergating_by_smu(struct amdgpu_device *adev, uint32_t block
 	case AMD_IP_BLOCK_TYPE_UVD:
 	case AMD_IP_BLOCK_TYPE_VCE:
 	case AMD_IP_BLOCK_TYPE_GFX:
-	case AMD_IP_BLOCK_TYPE_VCN:
 	case AMD_IP_BLOCK_TYPE_SDMA:
 	case AMD_IP_BLOCK_TYPE_JPEG:
 	case AMD_IP_BLOCK_TYPE_GMC:
@@ -96,7 +95,14 @@ int amdgpu_dpm_set_powergating_by_smu(struct amdgpu_device *adev, uint32_t block
 	case AMD_IP_BLOCK_TYPE_VPE:
 		if (pp_funcs && pp_funcs->set_powergating_by_smu)
 			ret = (pp_funcs->set_powergating_by_smu(
-				(adev)->powerplay.pp_handle, block_type, gate));
+				(adev)->powerplay.pp_handle, block_type, gate, 0));
+		break;
+	case AMD_IP_BLOCK_TYPE_VCN:
+		if (pp_funcs && pp_funcs->set_powergating_by_smu) {
+			for (int i = 0; i < adev->vcn.num_vcn_inst; i++)
+				ret = (pp_funcs->set_powergating_by_smu(
+					(adev)->powerplay.pp_handle, block_type, gate, i));
+		}
 		break;
 	default:
 		break;
diff --git a/drivers/gpu/drm/amd/pm/legacy-dpm/kv_dpm.c b/drivers/gpu/drm/amd/pm/legacy-dpm/kv_dpm.c
index 8908646ad620..f0f81ecd9ad6 100644
--- a/drivers/gpu/drm/amd/pm/legacy-dpm/kv_dpm.c
+++ b/drivers/gpu/drm/amd/pm/legacy-dpm/kv_dpm.c
@@ -3276,7 +3276,9 @@ static int kv_dpm_read_sensor(void *handle, int idx,
 }
 
 static int kv_set_powergating_by_smu(void *handle,
-				uint32_t block_type, bool gate)
+				uint32_t block_type,
+				bool gate,
+				int inst)
 {
 	switch (block_type) {
 	case AMD_IP_BLOCK_TYPE_UVD:
diff --git a/drivers/gpu/drm/amd/pm/powerplay/amd_powerplay.c b/drivers/gpu/drm/amd/pm/powerplay/amd_powerplay.c
index 26624a716fc6..90500b419d60 100644
--- a/drivers/gpu/drm/amd/pm/powerplay/amd_powerplay.c
+++ b/drivers/gpu/drm/amd/pm/powerplay/amd_powerplay.c
@@ -1227,7 +1227,9 @@ static void pp_dpm_powergate_sdma(void *handle, bool gate)
 }
 
 static int pp_set_powergating_by_smu(void *handle,
-				uint32_t block_type, bool gate)
+				uint32_t block_type,
+				bool gate,
+				int inst)
 {
 	int ret = 0;
 
diff --git a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
index 2fd7909dac18..dc372d39cb8d 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
@@ -358,11 +358,11 @@ static int smu_set_mall_enable(struct smu_context *smu)
  */
 static int smu_dpm_set_power_gate(void *handle,
 				  uint32_t block_type,
-				  bool gate)
+				  bool gate,
+				  int inst)
 {
 	struct smu_context *smu = handle;
-	struct amdgpu_device *adev = smu->adev;
-	int i, ret = 0;
+	int ret = 0;
 
 	if (!smu->pm_enabled || !smu->adev->pm.dpm_enabled) {
 		dev_WARN(smu->adev->dev,
@@ -378,12 +378,10 @@ static int smu_dpm_set_power_gate(void *handle,
 	 */
 	case AMD_IP_BLOCK_TYPE_UVD:
 	case AMD_IP_BLOCK_TYPE_VCN:
-		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
-			ret = smu_dpm_set_vcn_enable(smu, !gate, i);
-			if (ret)
-				dev_err(smu->adev->dev, "Failed to power %s VCN instance %d!\n",
-					gate ? "gate" : "ungate", i);
-		}
+		ret = smu_dpm_set_vcn_enable(smu, !gate, inst);
+		if (ret)
+			dev_err(smu->adev->dev, "Failed to power %s VCN instance %d!\n",
+				gate ? "gate" : "ungate", inst);
 		break;
 	case AMD_IP_BLOCK_TYPE_GFX:
 		ret = smu_gfx_off_control(smu, gate);
-- 
2.51.0


From ff69bba05f085cd6d4277c27ac7600160167b384 Mon Sep 17 00:00:00 2001
From: Boyuan Zhang <boyuan.zhang@amd.com>
Date: Wed, 2 Oct 2024 23:52:01 -0400
Subject: [PATCH 16/16] drm/amd/pm: add inst to dpm_set_powergating_by_smu
MIME-Version: 1.0
Content-Type: text/plain; charset=utf8
Content-Transfer-Encoding: 8bit

Add an instance parameter to amdgpu_dpm_set_powergating_by_smu() function,
and use the instance to call set_powergating_by_smu().

v2: remove duplicated functions.

remove for-loop in amdgpu_dpm_set_powergating_by_smu(), and temporarily
move it to amdgpu_dpm_enable_vcn(), in order to keep the exact same logic
as before, until further separation in next patch.

v3: drop SI logic in amdgpu_dpm_enable_vcn().

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_acp.c    | 14 ++++----
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c    |  4 +--
 drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c      |  2 +-
 drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c    |  2 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c     |  4 +--
 drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c      |  6 ++--
 drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c      |  4 +--
 drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c      |  4 +--
 drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c      |  4 +--
 drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c      |  4 +--
 drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c    |  4 +--
 drivers/gpu/drm/amd/amdgpu/vcn_v4_0_5.c    |  4 +--
 drivers/gpu/drm/amd/amdgpu/vcn_v5_0_0.c    |  4 +--
 drivers/gpu/drm/amd/pm/amdgpu_dpm.c        | 37 +++++++++++++++-------
 drivers/gpu/drm/amd/pm/inc/amdgpu_dpm.h    |  3 +-
 16 files changed, 59 insertions(+), 43 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_acp.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_acp.c
index ec5e0dcf8613..769200cda626 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_acp.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_acp.c
@@ -140,7 +140,7 @@ static int acp_poweroff(struct generic_pm_domain *genpd)
 	 * 2. power off the acp tiles
 	 * 3. check and enter ulv state
 	 */
-	amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, true);
+	amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, true, 0);
 	return 0;
 }
 
@@ -157,7 +157,7 @@ static int acp_poweron(struct generic_pm_domain *genpd)
 	 * 2. turn on acp clock
 	 * 3. power on acp tiles
 	 */
-	amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, false);
+	amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, false, 0);
 	return 0;
 }
 
@@ -236,7 +236,7 @@ static int acp_hw_init(struct amdgpu_ip_block *ip_block)
 			    ip_block->version->major, ip_block->version->minor);
 	/* -ENODEV means board uses AZ rather than ACP */
 	if (r == -ENODEV) {
-		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, true);
+		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, true, 0);
 		return 0;
 	} else if (r) {
 		return r;
@@ -508,7 +508,7 @@ static int acp_hw_fini(struct amdgpu_ip_block *ip_block)
 
 	/* return early if no ACP */
 	if (!adev->acp.acp_genpd) {
-		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, false);
+		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, false, 0);
 		return 0;
 	}
 
@@ -565,7 +565,7 @@ static int acp_suspend(struct amdgpu_ip_block *ip_block)
 
 	/* power up on suspend */
 	if (!adev->acp.acp_cell)
-		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, false);
+		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, false, 0);
 	return 0;
 }
 
@@ -575,7 +575,7 @@ static int acp_resume(struct amdgpu_ip_block *ip_block)
 
 	/* power down again on resume */
 	if (!adev->acp.acp_cell)
-		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, true);
+		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, true, 0);
 	return 0;
 }
 
@@ -596,7 +596,7 @@ static int acp_set_powergating_state(void *handle,
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 	bool enable = (state == AMD_PG_STATE_GATE);
 
-	amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, enable);
+	amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_ACP, enable, 0);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 97d3e5f29638..e41182645f94 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -3478,7 +3478,7 @@ static void amdgpu_device_delay_enable_gfx_off(struct work_struct *work)
 	WARN_ON_ONCE(adev->gfx.gfx_off_state);
 	WARN_ON_ONCE(adev->gfx.gfx_off_req_count);
 
-	if (!amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, true))
+	if (!amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, true, 0))
 		adev->gfx.gfx_off_state = true;
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
index 69a6b6dba0a5..e54f42e3797e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
@@ -806,7 +806,7 @@ void amdgpu_gfx_off_ctrl(struct amdgpu_device *adev, bool enable)
 			/* If going to s2idle, no need to wait */
 			if (adev->in_s0ix) {
 				if (!amdgpu_dpm_set_powergating_by_smu(adev,
-						AMD_IP_BLOCK_TYPE_GFX, true))
+						AMD_IP_BLOCK_TYPE_GFX, true, 0))
 					adev->gfx.gfx_off_state = true;
 			} else {
 				schedule_delayed_work(&adev->gfx.gfx_off_delay_work,
@@ -818,7 +818,7 @@ void amdgpu_gfx_off_ctrl(struct amdgpu_device *adev, bool enable)
 			cancel_delayed_work_sync(&adev->gfx.gfx_off_delay_work);
 
 			if (adev->gfx.gfx_off_state &&
-			    !amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, false)) {
+			    !amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, false, 0)) {
 				adev->gfx.gfx_off_state = false;
 
 				if (adev->gfx.funcs->init_spm_golden) {
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
index 7d45d4afa44e..5894755b151a 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
@@ -5319,7 +5319,7 @@ static void gfx_v8_0_enable_gfx_static_mg_power_gating(struct amdgpu_device *ade
 	    (adev->asic_type == CHIP_POLARIS12) ||
 	    (adev->asic_type == CHIP_VEGAM))
 		/* Send msg to SMU via Powerplay */
-		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, enable);
+		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, enable, 0);
 
 	WREG32_FIELD(RLC_PG_CNTL, STATIC_PER_CU_PG_ENABLE, enable ? 1 : 0);
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c b/drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c
index e9a6f33ca710..243eabda0607 100644
--- a/drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c
@@ -356,7 +356,7 @@ static void mmhub_v1_0_update_power_gating(struct amdgpu_device *adev,
 	if (adev->pg_flags & AMD_PG_SUPPORT_MMHUB)
 		amdgpu_dpm_set_powergating_by_smu(adev,
 						  AMD_IP_BLOCK_TYPE_GMC,
-						  enable);
+						  enable, 0);
 }
 
 static int mmhub_v1_0_gart_enable(struct amdgpu_device *adev)
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c
index c1f98f6cf20d..3f5959557727 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c
@@ -1956,7 +1956,7 @@ static int sdma_v4_0_hw_init(struct amdgpu_ip_block *ip_block)
 	struct amdgpu_device *adev = ip_block->adev;
 
 	if (adev->flags & AMD_IS_APU)
-		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_SDMA, false);
+		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_SDMA, false, 0);
 
 	if (!amdgpu_sriov_vf(adev))
 		sdma_v4_0_init_golden_registers(adev);
@@ -1983,7 +1983,7 @@ static int sdma_v4_0_hw_fini(struct amdgpu_ip_block *ip_block)
 	sdma_v4_0_enable(adev, false);
 
 	if (adev->flags & AMD_IS_APU)
-		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_SDMA, true);
+		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_SDMA, true, 0);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c b/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c
index 10e99c926fb8..511d76e188f2 100644
--- a/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c
@@ -303,7 +303,7 @@ static int vcn_v1_0_suspend(struct amdgpu_ip_block *ip_block)
 	idle_work_unexecuted = cancel_delayed_work_sync(&adev->vcn.idle_work);
 	if (idle_work_unexecuted) {
 		if (adev->pm.dpm_enabled)
-			amdgpu_dpm_enable_uvd(adev, false);
+			amdgpu_dpm_enable_vcn(adev, false);
 	}
 
 	r = vcn_v1_0_hw_fini(ip_block);
@@ -1856,7 +1856,7 @@ static void vcn_v1_0_idle_work_handler(struct work_struct *work)
 	if (fences == 0) {
 		amdgpu_gfx_off_ctrl(adev, true);
 		if (adev->pm.dpm_enabled)
-			amdgpu_dpm_enable_uvd(adev, false);
+			amdgpu_dpm_enable_vcn(adev, false);
 		else
 			amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
 			       AMD_PG_STATE_GATE);
@@ -1886,7 +1886,7 @@ void vcn_v1_0_set_pg_for_begin_use(struct amdgpu_ring *ring, bool set_clocks)
 	if (set_clocks) {
 		amdgpu_gfx_off_ctrl(adev, false);
 		if (adev->pm.dpm_enabled)
-			amdgpu_dpm_enable_uvd(adev, true);
+			amdgpu_dpm_enable_vcn(adev, true);
 		else
 			amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
 			       AMD_PG_STATE_UNGATE);
diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c b/drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c
index e0322cbca3ec..697822abf3fc 100644
--- a/drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c
@@ -978,7 +978,7 @@ static int vcn_v2_0_start(struct amdgpu_device *adev)
 	int i, j, r;
 
 	if (adev->pm.dpm_enabled)
-		amdgpu_dpm_enable_uvd(adev, true);
+		amdgpu_dpm_enable_vcn(adev, true);
 
 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
 		return vcn_v2_0_start_dpg_mode(adev, adev->vcn.indirect_sram);
@@ -1235,7 +1235,7 @@ static int vcn_v2_0_stop(struct amdgpu_device *adev)
 
 power_off:
 	if (adev->pm.dpm_enabled)
-		amdgpu_dpm_enable_uvd(adev, false);
+		amdgpu_dpm_enable_vcn(adev, false);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c b/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c
index 6aa08281d094..0afbcf72cd51 100644
--- a/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c
+++ b/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c
@@ -1013,7 +1013,7 @@ static int vcn_v2_5_start(struct amdgpu_device *adev)
 	int i, j, k, r;
 
 	if (adev->pm.dpm_enabled)
-		amdgpu_dpm_enable_uvd(adev, true);
+		amdgpu_dpm_enable_vcn(adev, true);
 
 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
 		if (adev->vcn.harvest_config & (1 << i))
@@ -1486,7 +1486,7 @@ static int vcn_v2_5_stop(struct amdgpu_device *adev)
 	}
 
 	if (adev->pm.dpm_enabled)
-		amdgpu_dpm_enable_uvd(adev, false);
+		amdgpu_dpm_enable_vcn(adev, false);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c b/drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c
index 6732ad7f16f5..b28aad37d9ed 100644
--- a/drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c
@@ -1142,7 +1142,7 @@ static int vcn_v3_0_start(struct amdgpu_device *adev)
 	int i, j, k, r;
 
 	if (adev->pm.dpm_enabled)
-		amdgpu_dpm_enable_uvd(adev, true);
+		amdgpu_dpm_enable_vcn(adev, true);
 
 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
 		if (adev->vcn.harvest_config & (1 << i))
@@ -1633,7 +1633,7 @@ static int vcn_v3_0_stop(struct amdgpu_device *adev)
 	}
 
 	if (adev->pm.dpm_enabled)
-		amdgpu_dpm_enable_uvd(adev, false);
+		amdgpu_dpm_enable_vcn(adev, false);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c
index fcc8511e91ee..5c6060374ac4 100644
--- a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c
@@ -1098,7 +1098,7 @@ static int vcn_v4_0_start(struct amdgpu_device *adev)
 	int i, j, k, r;
 
 	if (adev->pm.dpm_enabled)
-		amdgpu_dpm_enable_uvd(adev, true);
+		amdgpu_dpm_enable_vcn(adev, true);
 
 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
 		if (adev->vcn.harvest_config & (1 << i))
@@ -1624,7 +1624,7 @@ static int vcn_v4_0_stop(struct amdgpu_device *adev)
 	}
 
 	if (adev->pm.dpm_enabled)
-		amdgpu_dpm_enable_uvd(adev, false);
+		amdgpu_dpm_enable_vcn(adev, false);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c
index 3f69b9b2bcd0..037c85f80c64 100644
--- a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c
+++ b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c
@@ -1122,7 +1122,7 @@ static int vcn_v4_0_3_start(struct amdgpu_device *adev)
 	uint32_t tmp;
 
 	if (adev->pm.dpm_enabled)
-		amdgpu_dpm_enable_uvd(adev, true);
+		amdgpu_dpm_enable_vcn(adev, true);
 
 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
@@ -1396,7 +1396,7 @@ static int vcn_v4_0_3_stop(struct amdgpu_device *adev)
 	}
 Done:
 	if (adev->pm.dpm_enabled)
-		amdgpu_dpm_enable_uvd(adev, false);
+		amdgpu_dpm_enable_vcn(adev, false);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_5.c b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_5.c
index 71961fb3f7ff..398191a48446 100644
--- a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_5.c
+++ b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_5.c
@@ -1001,7 +1001,7 @@ static int vcn_v4_0_5_start(struct amdgpu_device *adev)
 	int i, j, k, r;
 
 	if (adev->pm.dpm_enabled)
-		amdgpu_dpm_enable_uvd(adev, true);
+		amdgpu_dpm_enable_vcn(adev, true);
 
 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
 		if (adev->vcn.harvest_config & (1 << i))
@@ -1278,7 +1278,7 @@ static int vcn_v4_0_5_stop(struct amdgpu_device *adev)
 	}
 
 	if (adev->pm.dpm_enabled)
-		amdgpu_dpm_enable_uvd(adev, false);
+		amdgpu_dpm_enable_vcn(adev, false);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v5_0_0.c b/drivers/gpu/drm/amd/amdgpu/vcn_v5_0_0.c
index bd3d2bbdc16b..fa274d2f5fd2 100644
--- a/drivers/gpu/drm/amd/amdgpu/vcn_v5_0_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/vcn_v5_0_0.c
@@ -772,7 +772,7 @@ static int vcn_v5_0_0_start(struct amdgpu_device *adev)
 	int i, j, k, r;
 
 	if (adev->pm.dpm_enabled)
-		amdgpu_dpm_enable_uvd(adev, true);
+		amdgpu_dpm_enable_vcn(adev, true);
 
 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
 		if (adev->vcn.harvest_config & (1 << i))
@@ -1019,7 +1019,7 @@ static int vcn_v5_0_0_stop(struct amdgpu_device *adev)
 	}
 
 	if (adev->pm.dpm_enabled)
-		amdgpu_dpm_enable_uvd(adev, false);
+		amdgpu_dpm_enable_vcn(adev, false);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/amd/pm/amdgpu_dpm.c b/drivers/gpu/drm/amd/pm/amdgpu_dpm.c
index 78e3488b6a35..0724f8dda470 100644
--- a/drivers/gpu/drm/amd/pm/amdgpu_dpm.c
+++ b/drivers/gpu/drm/amd/pm/amdgpu_dpm.c
@@ -70,13 +70,18 @@ int amdgpu_dpm_get_mclk(struct amdgpu_device *adev, bool low)
 	return ret;
 }
 
-int amdgpu_dpm_set_powergating_by_smu(struct amdgpu_device *adev, uint32_t block_type, bool gate)
+int amdgpu_dpm_set_powergating_by_smu(struct amdgpu_device *adev,
+				       uint32_t block_type,
+				       bool gate,
+				       int inst)
 {
 	int ret = 0;
 	const struct amd_pm_funcs *pp_funcs = adev->powerplay.pp_funcs;
 	enum ip_power_state pwr_state = gate ? POWER_STATE_OFF : POWER_STATE_ON;
+	bool is_vcn = (block_type == AMD_IP_BLOCK_TYPE_UVD || block_type == AMD_IP_BLOCK_TYPE_VCN);
 
-	if (atomic_read(&adev->pm.pwr_state[block_type]) == pwr_state) {
+	if (atomic_read(&adev->pm.pwr_state[block_type]) == pwr_state &&
+			(!is_vcn || adev->vcn.num_vcn_inst == 1)) {
 		dev_dbg(adev->dev, "IP block%d already in the target %s state!",
 				block_type, gate ? "gate" : "ungate");
 		return 0;
@@ -98,11 +103,9 @@ int amdgpu_dpm_set_powergating_by_smu(struct amdgpu_device *adev, uint32_t block
 				(adev)->powerplay.pp_handle, block_type, gate, 0));
 		break;
 	case AMD_IP_BLOCK_TYPE_VCN:
-		if (pp_funcs && pp_funcs->set_powergating_by_smu) {
-			for (int i = 0; i < adev->vcn.num_vcn_inst; i++)
-				ret = (pp_funcs->set_powergating_by_smu(
-					(adev)->powerplay.pp_handle, block_type, gate, i));
-		}
+		if (pp_funcs && pp_funcs->set_powergating_by_smu)
+			ret = (pp_funcs->set_powergating_by_smu(
+				(adev)->powerplay.pp_handle, block_type, gate, inst));
 		break;
 	default:
 		break;
@@ -572,12 +575,24 @@ void amdgpu_dpm_enable_uvd(struct amdgpu_device *adev, bool enable)
 		return;
 	}
 
-	ret = amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_UVD, !enable);
+	ret = amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_UVD, !enable, 0);
 	if (ret)
 		DRM_ERROR("Dpm %s uvd failed, ret = %d. \n",
 			  enable ? "enable" : "disable", ret);
 }
 
+void amdgpu_dpm_enable_vcn(struct amdgpu_device *adev, bool enable)
+{
+	int i, ret = 0;
+
+	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
+		ret = amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_VCN, !enable, i);
+		if (ret)
+			DRM_ERROR("Dpm %s uvd failed, ret = %d. \n",
+				  enable ? "enable" : "disable", ret);
+	}
+}
+
 void amdgpu_dpm_enable_vce(struct amdgpu_device *adev, bool enable)
 {
 	int ret = 0;
@@ -597,7 +612,7 @@ void amdgpu_dpm_enable_vce(struct amdgpu_device *adev, bool enable)
 		return;
 	}
 
-	ret = amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_VCE, !enable);
+	ret = amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_VCE, !enable, 0);
 	if (ret)
 		DRM_ERROR("Dpm %s vce failed, ret = %d. \n",
 			  enable ? "enable" : "disable", ret);
@@ -607,7 +622,7 @@ void amdgpu_dpm_enable_jpeg(struct amdgpu_device *adev, bool enable)
 {
 	int ret = 0;
 
-	ret = amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_JPEG, !enable);
+	ret = amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_JPEG, !enable, 0);
 	if (ret)
 		DRM_ERROR("Dpm %s jpeg failed, ret = %d. \n",
 			  enable ? "enable" : "disable", ret);
@@ -617,7 +632,7 @@ void amdgpu_dpm_enable_vpe(struct amdgpu_device *adev, bool enable)
 {
 	int ret = 0;
 
-	ret = amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_VPE, !enable);
+	ret = amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_VPE, !enable, 0);
 	if (ret)
 		DRM_ERROR("Dpm %s vpe failed, ret = %d.\n",
 			  enable ? "enable" : "disable", ret);
diff --git a/drivers/gpu/drm/amd/pm/inc/amdgpu_dpm.h b/drivers/gpu/drm/amd/pm/inc/amdgpu_dpm.h
index b64aea7cd188..e87d67484f0e 100644
--- a/drivers/gpu/drm/amd/pm/inc/amdgpu_dpm.h
+++ b/drivers/gpu/drm/amd/pm/inc/amdgpu_dpm.h
@@ -397,7 +397,7 @@ int amdgpu_dpm_get_apu_thermal_limit(struct amdgpu_device *adev, uint32_t *limit
 int amdgpu_dpm_set_apu_thermal_limit(struct amdgpu_device *adev, uint32_t limit);
 
 int amdgpu_dpm_set_powergating_by_smu(struct amdgpu_device *adev,
-				      uint32_t block_type, bool gate);
+				      uint32_t block_type, bool gate, int inst);
 
 extern int amdgpu_dpm_get_sclk(struct amdgpu_device *adev, bool low);
 
@@ -446,6 +446,7 @@ void amdgpu_pm_acpi_event_handler(struct amdgpu_device *adev);
 
 void amdgpu_dpm_compute_clocks(struct amdgpu_device *adev);
 void amdgpu_dpm_enable_uvd(struct amdgpu_device *adev, bool enable);
+void amdgpu_dpm_enable_vcn(struct amdgpu_device *adev, bool enable);
 void amdgpu_dpm_enable_vce(struct amdgpu_device *adev, bool enable);
 void amdgpu_dpm_enable_jpeg(struct amdgpu_device *adev, bool enable);
 void amdgpu_dpm_enable_vpe(struct amdgpu_device *adev, bool enable);
-- 
2.51.0