Skip to contentSkip to conversation

drm/amdgpu/ras: Add debug mask to disable CE logs for uniras

PATCH Superseded Ce Sun 3 participants 3 messages Jul 15

Subsystems (2)

v1 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Ce Sun

author
#0 lore raw

PATCH 1/1 drm/amdgpu/ras: Add debug mask to disable CE logs for uniras

Add debug mask to disable kernel logs of RAS correctable errors, including both ACA and CE error counter kernel messages.

---
 drivers/gpu/drm/amd/ras/ras_mgr/amdgpu_ras_mgr.c | 12 ++++++++++++
 drivers/gpu/drm/amd/ras/ras_mgr/amdgpu_ras_mgr.h |  1 +
 drivers/gpu/drm/amd/ras/rascore/ras.h            |  3 +++
 drivers/gpu/drm/amd/ras/rascore/ras_aca.c        |  7 +++++++
 drivers/gpu/drm/amd/ras/rascore/ras_aca_v1_0.h   |  4 ++++
 drivers/gpu/drm/amd/ras/rascore/ras_core.c       | 10 ++++++++++
 6 files changed, 37 insertions(+)
drivers/gpu/drm/amd/ras/ras_mgr/amdgpu_ras_mgr.c +12 -0
index ea95e0f93d1c..2be008a9da6a 100644
@@ -437,6 +437,7 @@ static int amdgpu_ras_mgr_hw_init(struct amdgpu_ip_block *ip_block)
 	ras_mgr->ras_is_ready = true;
 
 	amdgpu_enable_uniras(adev, true);
+	amdgpu_set_ce_log_state(adev);
 
 	RAS_DEV_INFO(adev, "AMDGPU RAS Is Ready.\n");
 	return 0;
@@ -510,6 +511,17 @@ const struct amdgpu_ip_block_version ras_v1_0_ip_block = {
 	.funcs = &ras_v1_0_ip_funcs,
 };
 
+void amdgpu_set_ce_log_state(struct amdgpu_device *adev)
+{
+	struct amdgpu_ras_mgr *ras_mgr = amdgpu_ras_mgr_get_context(adev);
+	bool disable_ce_log = !!adev->debug_disable_ce_logs;
+
+	if (!ras_mgr || !ras_mgr->ras_core)
+		return;
+
+	ras_core_set_ce_log_status(ras_mgr->ras_core, disable_ce_log);
+}
+
 int amdgpu_enable_uniras(struct amdgpu_device *adev, bool enable)
 {
 	struct amdgpu_ras_mgr *ras_mgr = amdgpu_ras_mgr_get_context(adev);
drivers/gpu/drm/amd/ras/ras_mgr/amdgpu_ras_mgr.h +1 -0
index a20bb8fdce87..efbdb25241a3 100644
@@ -63,6 +63,7 @@ extern const struct amdgpu_ip_block_version ras_v1_0_ip_block;
 struct amdgpu_ras_mgr *amdgpu_ras_mgr_get_context(
 			struct amdgpu_device *adev);
 int amdgpu_enable_uniras(struct amdgpu_device *adev, bool enable);
+void amdgpu_set_ce_log_state(struct amdgpu_device *adev);
 bool amdgpu_uniras_enabled(struct amdgpu_device *adev);
 int amdgpu_ras_mgr_handle_fatal_interrupt(struct amdgpu_device *adev, void *data);
 int amdgpu_ras_mgr_handle_controller_interrupt(struct amdgpu_device *adev, void *data);
drivers/gpu/drm/amd/ras/rascore/ras.h +3 -0
index 878dfdfcb18a..5911ce63f889 100644
@@ -346,6 +346,7 @@ struct ras_core_context {
 	spinlock_t seqno_lock;
 
 	bool ras_core_enabled;
+	bool ras_core_ce_log_disabled;
 
 	u64 ras_fw_features;
 };
@@ -403,4 +404,6 @@ int ras_core_convert_soc_pa_to_cur_nps_pages(struct ras_core_context *ras_core,
 int ras_core_check_address_sanity(struct ras_core_context *ras_core, uint64_t addr);
 
 int ras_core_set_debug_mode(struct ras_core_context *ras_core, bool enable);
+void ras_core_set_ce_log_status(struct ras_core_context *ras_core, bool disable);
+bool ras_core_get_ce_log_status(struct ras_core_context *ras_core);
 #endif
drivers/gpu/drm/amd/ras/rascore/ras_aca.c +7 -0
index 67a35409ff0e..5a46a265e26d 100644
@@ -99,6 +99,8 @@ static void aca_report_ecc_info(struct ras_core_context *ras_core,
 	}
 
 	if (ecc_count.new_ce_count) {
+		if (ras_core_get_ce_log_status(ras_core))
+			return;
 		RAS_DEV_INFO(ras_core->dev,
 		"{%llu} socket: %d, die: %d, %u new correctable hardware errors detected in %s block\n",
 			seq_no, skt, aid, ecc_count.new_ce_count, blk_name(blk));
@@ -114,6 +116,11 @@ static void aca_bank_log(struct ras_core_context *ras_core,
 {
 	int i;
 
+	if(ras_core_get_ce_log_status(ras_core) &&
+	   bank->ecc_type == RAS_ERR_TYPE__CE &&
+	   !ACA_BANK_ERR_IS_DEFFERED(bank))
+		return;
+
 	RAS_DEV_INFO(ras_core->dev,
 		"{%llu}" RAS_HW_ERR "Accelerator Check Architecture events logged\n",
 		bank->seq_no);
drivers/gpu/drm/amd/ras/rascore/ras_aca_v1_0.h +4 -0
index 40e5d94b037f..c42a47492d4a 100644
@@ -67,5 +67,9 @@
 #define mmSMNXCD_XCD0_MCA_SMU 0x40430400	/* SMN XCD XCD0 */
 #define mmSMNAID_AID0_MCA_SMU 0x03b30400	/* SMN AID AID0 */
 
+#define ACA_BANK_ERR_IS_DEFFERED(bank)                                \
+	(ACA_REG_STATUS_POISON((bank)->regs[ACA_REG_IDX__STATUS]) || \
+	 ACA_REG_STATUS_DEFERRED((bank)->regs[ACA_REG_IDX__STATUS]))
+
 extern const struct ras_aca_ip_func ras_aca_func_v1_0;
 #endif
drivers/gpu/drm/amd/ras/rascore/ras_core.c +10 -0
index 08e17a83ad5b..80974b8d6297 100644
@@ -545,6 +545,16 @@ bool ras_core_is_enabled(struct ras_core_context *ras_core)
 	return ras_core->ras_core_enabled;
 }
 
+void ras_core_set_ce_log_status(struct ras_core_context *ras_core, bool disable)
+{
+	ras_core->ras_core_ce_log_disabled = disable;
+}
+
+bool ras_core_get_ce_log_status(struct ras_core_context *ras_core)
+{
+	return ras_core->ras_core_ce_log_disabled;
+}
+
 uint64_t ras_core_get_utc_second_timestamp(struct ras_core_context *ras_core)
 {
 	if (!ras_core)
Signature · 1 line
2.34.1
#1 lore raw

AMD General

Best Regards,
Thomas
-----Original Message-----
From: Sun, Ce(Overlord) <[email protected]>
Sent: Wednesday, July 15, 2026 11:02 AM
To: [email protected]
Cc: Zhang, Hawking <[email protected]>; Chai, Thomas <[email protected]>; Zhou1, Tao <[email protected]>; Yang, Stanley <[email protected]>; Sun, Ce(Overlord) <[email protected]>
Subject: [PATCH] drm/amdgpu/ras: Add debug mask to disable CE logs for uniras

Add debug mask to disable kernel logs of RAS correctable errors, including both ACA and CE error counter kernel messages.

---
 drivers/gpu/drm/amd/ras/ras_mgr/amdgpu_ras_mgr.c | 12 ++++++++++++  drivers/gpu/drm/amd/ras/ras_mgr/amdgpu_ras_mgr.h |  1 +
 drivers/gpu/drm/amd/ras/rascore/ras.h            |  3 +++
 drivers/gpu/drm/amd/ras/rascore/ras_aca.c        |  7 +++++++
 drivers/gpu/drm/amd/ras/rascore/ras_aca_v1_0.h   |  4 ++++
 drivers/gpu/drm/amd/ras/rascore/ras_core.c       | 10 ++++++++++
 6 files changed, 37 insertions(+)
drivers/gpu/drm/amd/ras/ras_mgr/amdgpu_ras_mgr.c +10 -0
index ea95e0f93d1c..2be008a9da6a 100644
@@ -437,6 +437,7 @@ static int amdgpu_ras_mgr_hw_init(struct amdgpu_ip_block *ip_block)
        ras_mgr->ras_is_ready = true;
 
        amdgpu_enable_uniras(adev, true);
+       amdgpu_set_ce_log_state(adev);
 
        RAS_DEV_INFO(adev, "AMDGPU RAS Is Ready.\n");
        return 0;
@@ -510,6 +511,17 @@ const struct amdgpu_ip_block_version ras_v1_0_ip_block = {
        .funcs = &ras_v1_0_ip_funcs,
 };
 
+void amdgpu_set_ce_log_state(struct amdgpu_device *adev) {
+       struct amdgpu_ras_mgr *ras_mgr = amdgpu_ras_mgr_get_context(adev);
+       bool disable_ce_log = !!adev->debug_disable_ce_logs;
+
+       if (!ras_mgr || !ras_mgr->ras_core)
+               return;
+
+       ras_core_set_ce_log_status(ras_mgr->ras_core, disable_ce_log); }
+
 

[Thomas ] Since ras_core_ce_log_disabled is a boolean, consider renaming the function to xxx_disable_ce_log() or xxx_ce_log_disabled() to better reflect its semantics.

 int amdgpu_enable_uniras(struct amdgpu_device *adev, bool enable)  {
        struct amdgpu_ras_mgr *ras_mgr = amdgpu_ras_mgr_get_context(adev); diff --git a/drivers/gpu/drm/amd/ras/ras_mgr/amdgpu_ras_mgr.h b/drivers/gpu/drm/amd/ras/ras_mgr/amdgpu_ras_mgr.h

index a20bb8fdce87..efbdb25241a3 100644

drivers/gpu/drm/amd/ras/ras_mgr/amdgpu_ras_mgr.h +1 -0
@@ -63,6 +63,7 @@ extern const struct amdgpu_ip_block_version ras_v1_0_ip_block;  struct amdgpu_ras_mgr *amdgpu_ras_mgr_get_context(
                        struct amdgpu_device *adev);
 int amdgpu_enable_uniras(struct amdgpu_device *adev, bool enable);
+void amdgpu_set_ce_log_state(struct amdgpu_device *adev);
 bool amdgpu_uniras_enabled(struct amdgpu_device *adev);  int amdgpu_ras_mgr_handle_fatal_interrupt(struct amdgpu_device *adev, void *data);  int amdgpu_ras_mgr_handle_controller_interrupt(struct amdgpu_device *adev, void *data); diff --git a/drivers/gpu/drm/amd/ras/rascore/ras.h b/drivers/gpu/drm/amd/ras/rascore/ras.h

index 878dfdfcb18a..5911ce63f889 100644

drivers/gpu/drm/amd/ras/rascore/ras.h +4 -0
@@ -346,6 +346,7 @@ struct ras_core_context {
        spinlock_t seqno_lock;
 
        bool ras_core_enabled;
+       bool ras_core_ce_log_disabled;
 
        u64 ras_fw_features;
 };
@@ -403,4 +404,6 @@ int ras_core_convert_soc_pa_to_cur_nps_pages(struct ras_core_context *ras_core,  int ras_core_check_address_sanity(struct ras_core_context *ras_core, uint64_t addr);
 
 int ras_core_set_debug_mode(struct ras_core_context *ras_core, bool enable);
+void ras_core_set_ce_log_status(struct ras_core_context *ras_core, bool
+disable); bool ras_core_get_ce_log_status(struct ras_core_context
+*ras_core);
 #endif
drivers/gpu/drm/amd/ras/rascore/ras_aca.c +2 -0
index 67a35409ff0e..5a46a265e26d 100644
@@ -99,6 +99,8 @@ static void aca_report_ecc_info(struct ras_core_context *ras_core,
        }
 
        if (ecc_count.new_ce_count) {
+               if (ras_core_get_ce_log_status(ras_core))
+                       return;
                RAS_DEV_INFO(ras_core->dev,
                "{%llu} socket: %d, die: %d, %u new correctable hardware errors detected in %s block\n",
                        seq_no, skt, aid, ecc_count.new_ce_count, blk_name(blk)); @@ -114,6 +116,11 @@ static void aca_bank_log(struct ras_core_context *ras_core,  {
        int i;

+ if(ras_core_get_ce_log_status(ras_core) &&
+ bank->ecc_type == RAS_ERR_TYPE__CE &&
+ !ACA_BANK_ERR_IS_DEFFERED(bank))

[Thomas] Can we use " bank_ecc -> de_count " as the check condition ?

+ return;
+

        RAS_DEV_INFO(ras_core->dev,
                "{%llu}" RAS_HW_ERR "Accelerator Check Architecture events logged\n",
                bank->seq_no);
drivers/gpu/drm/amd/ras/rascore/ras_aca_v1_0.h +4 -0
index 40e5d94b037f..c42a47492d4a 100644
@@ -67,5 +67,9 @@
 #define mmSMNXCD_XCD0_MCA_SMU 0x40430400       /* SMN XCD XCD0 */
 #define mmSMNAID_AID0_MCA_SMU 0x03b30400       /* SMN AID AID0 */
 
+#define ACA_BANK_ERR_IS_DEFFERED(bank)                                \
+       (ACA_REG_STATUS_POISON((bank)->regs[ACA_REG_IDX__STATUS]) || \
+        ACA_REG_STATUS_DEFERRED((bank)->regs[ACA_REG_IDX__STATUS]))
+
 extern const struct ras_aca_ip_func ras_aca_func_v1_0;  #endif diff --git a/drivers/gpu/drm/amd/ras/rascore/ras_core.c b/drivers/gpu/drm/amd/ras/rascore/ras_core.c

index 08e17a83ad5b..80974b8d6297 100644

drivers/gpu/drm/amd/ras/rascore/ras_core.c +8 -0
@@ -545,6 +545,16 @@ bool ras_core_is_enabled(struct ras_core_context *ras_core)
        return ras_core->ras_core_enabled;
 }
 
+void ras_core_set_ce_log_status(struct ras_core_context *ras_core, bool
+disable) {
+       ras_core->ras_core_ce_log_disabled = disable; }
+
+bool ras_core_get_ce_log_status(struct ras_core_context *ras_core) {
+       return ras_core->ras_core_ce_log_disabled;
+}
+
 
 uint64_t ras_core_get_utc_second_timestamp(struct ras_core_context *ras_core)  {
        if (!ras_core)

--
2.34.1

#2 lore raw
In reply to Ce Sun

AMD General

Quoted 122 lines

-----Original Message-----
From: Sun, Ce(Overlord) <[email protected]>
Sent: Wednesday, July 15, 2026 11:02 AM
To: [email protected]
Cc: Zhang, Hawking <[email protected]>; Chai, Thomas <[email protected]>; Zhou1, Tao <[email protected]>; Yang, Stanley <[email protected]>; Sun, Ce(Overlord) <[email protected]>
Subject: [PATCH] drm/amdgpu/ras: Add debug mask to disable CE logs for uniras

Add debug mask to disable kernel logs of RAS correctable errors, including both ACA and CE error counter kernel messages.

---
 drivers/gpu/drm/amd/ras/ras_mgr/amdgpu_ras_mgr.c | 12 ++++++++++++
drivers/gpu/drm/amd/ras/ras_mgr/amdgpu_ras_mgr.h |  1 +
 drivers/gpu/drm/amd/ras/rascore/ras.h            |  3 +++
 drivers/gpu/drm/amd/ras/rascore/ras_aca.c        |  7 +++++++
 drivers/gpu/drm/amd/ras/rascore/ras_aca_v1_0.h   |  4 ++++
 drivers/gpu/drm/amd/ras/rascore/ras_core.c       | 10 ++++++++++
 6 files changed, 37 insertions(+)
drivers/gpu/drm/amd/ras/ras_mgr/amdgpu_ras_mgr.c +1 -0
index ea95e0f93d1c..2be008a9da6a 100644
@@ -437,6 +437,7 @@ static int amdgpu_ras_mgr_hw_init(struct amdgpu_ip_block
 *ip_block)
      ras_mgr->ras_is_ready = true;
 
      amdgpu_enable_uniras(adev, true);
+     amdgpu_set_ce_log_state(adev);
 
      RAS_DEV_INFO(adev, "AMDGPU RAS Is Ready.\n");
      return 0;
drivers/gpu/drm/amd/ras/ras_mgr/amdgpu_ras_mgr.c +9 -1
@@ -510,6 +511,17 @@ const struct amdgpu_ip_block_version ras_v1_0_ip_block
 = {
      .funcs = &ras_v1_0_ip_funcs,
 };
 
+void amdgpu_set_ce_log_state(struct amdgpu_device *adev) {
+     struct amdgpu_ras_mgr *ras_mgr = amdgpu_ras_mgr_get_context(adev);
+     bool disable_ce_log = !!adev->debug_disable_ce_logs;
+
+     if (!ras_mgr || !ras_mgr->ras_core)
+             return;
+
+     ras_core_set_ce_log_status(ras_mgr->ras_core, disable_ce_log); }
+
 int amdgpu_enable_uniras(struct amdgpu_device *adev, bool enable)  {
      struct amdgpu_ras_mgr *ras_mgr = amdgpu_ras_mgr_get_context(adev); diff
--git a/drivers/gpu/drm/amd/ras/ras_mgr/amdgpu_ras_mgr.h

b/drivers/gpu/drm/amd/ras/ras_mgr/amdgpu_ras_mgr.h index a20bb8fdce87..efbdb25241a3 100644

drivers/gpu/drm/amd/ras/ras_mgr/amdgpu_ras_mgr.h +0 -0
@@ -63,6 +63,7 @@ extern const struct amdgpu_ip_block_version

ras_v1_0_ip_block; struct amdgpu_ras_mgr *amdgpu_ras_mgr_get_context(

                      struct amdgpu_device *adev);

int amdgpu_enable_uniras(struct amdgpu_device *adev, bool enable);
+void amdgpu_set_ce_log_state(struct amdgpu_device *adev);
bool amdgpu_uniras_enabled(struct amdgpu_device *adev); int
amdgpu_ras_mgr_handle_fatal_interrupt(struct amdgpu_device *adev, void *data); int amdgpu_ras_mgr_handle_controller_interrupt(struct amdgpu_device *adev, void *data); diff --git a/drivers/gpu/drm/amd/ras/rascore/ras.h b/drivers/gpu/drm/amd/ras/rascore/ras.h
index 878dfdfcb18a..5911ce63f889 100644

drivers/gpu/drm/amd/ras/rascore/ras.h +1 -0
@@ -346,6 +346,7 @@ struct ras_core_context {
      spinlock_t seqno_lock;
 
      bool ras_core_enabled;
+     bool ras_core_ce_log_disabled;
 
      u64 ras_fw_features;
 };
@@ -403,4 +404,6 @@ int ras_core_convert_soc_pa_to_cur_nps_pages(struct

ras_core_context *ras_core, int ras_core_check_address_sanity(struct ras_core_context *ras_core, uint64_t addr);

int ras_core_set_debug_mode(struct ras_core_context *ras_core, bool enable);

drivers/gpu/drm/amd/ras/rascore/ras.h +3 -0
+void ras_core_set_ce_log_status(struct ras_core_context *ras_core, bool
+disable); bool ras_core_get_ce_log_status(struct ras_core_context
+*ras_core);
 #endif
drivers/gpu/drm/amd/ras/rascore/ras_aca.c +2 -0
index 67a35409ff0e..5a46a265e26d 100644
@@ -99,6 +99,8 @@ static void aca_report_ecc_info(struct ras_core_context
 *ras_core,
      }
 
      if (ecc_count.new_ce_count) {
+             if (ras_core_get_ce_log_status(ras_core))
+                     return;
              RAS_DEV_INFO(ras_core->dev,
              "{%llu} socket: %d, die: %d, %u new correctable hardware errors

detected in %s block\n",

                      seq_no, skt, aid, ecc_count.new_ce_count, blk_name(blk));
drivers/gpu/drm/amd/ras/rascore/ras_aca.c +5 -0
@@ -114,6 +116,11 @@ static void aca_bank_log(struct ras_core_context
 *ras_core,  {
      int i;
 
+     if(ras_core_get_ce_log_status(ras_core) &&
+        bank->ecc_type == RAS_ERR_TYPE__CE &&
+        !ACA_BANK_ERR_IS_DEFFERED(bank))
+             return;
+
      RAS_DEV_INFO(ras_core->dev,
              "{%llu}" RAS_HW_ERR "Accelerator Check Architecture events
 logged\n",
              bank->seq_no);
drivers/gpu/drm/amd/ras/rascore/ras_aca_v1_0.h +0 -0
index 40e5d94b037f..c42a47492d4a 100644

drivers/gpu/drm/amd/ras/rascore/ras_aca_v1_0.h line 73

Quoted 7 lines
drivers/gpu/drm/amd/ras/rascore/ras_aca_v1_0.h +3 -0
@@ -67,5 +67,9 @@
 #define mmSMNXCD_XCD0_MCA_SMU 0x40430400     /* SMN XCD XCD0 */
 #define mmSMNAID_AID0_MCA_SMU 0x03b30400     /* SMN AID AID0 */
 
+#define ACA_BANK_ERR_IS_DEFFERED(bank)                                \
+     (ACA_REG_STATUS_POISON((bank)->regs[ACA_REG_IDX__STATUS]) || \
+      ACA_REG_STATUS_DEFERRED((bank)-
 >regs[ACA_REG_IDX__STATUS]))

[Tao] ACA_BANK_ERR_IS_DEFFERED is defined in v1_0 code, version specific, but it's used in ras_aca.c, common for all versions. I don't think this is a good idea.

Quoted 24 lines
drivers/gpu/drm/amd/ras/rascore/ras_aca_v1_0.h +1 -0
+
 extern const struct ras_aca_ip_func ras_aca_func_v1_0;  #endif diff --git

a/drivers/gpu/drm/amd/ras/rascore/ras_core.c
b/drivers/gpu/drm/amd/ras/rascore/ras_core.c
index 08e17a83ad5b..80974b8d6297 100644

drivers/gpu/drm/amd/ras/rascore/ras_core.c +8 -1
@@ -545,6 +545,16 @@ bool ras_core_is_enabled(struct ras_core_context
 *ras_core)
      return ras_core->ras_core_enabled;
 }
 
+void ras_core_set_ce_log_status(struct ras_core_context *ras_core, bool
+disable) {
+     ras_core->ras_core_ce_log_disabled = disable; }
+
+bool ras_core_get_ce_log_status(struct ras_core_context *ras_core) {
+     return ras_core->ras_core_ce_log_disabled;
+}
+
 uint64_t ras_core_get_utc_second_timestamp(struct ras_core_context *ras_core)  {
      if (!ras_core)
--

2.34.1