Skip to contentSkip to conversation

EDAC/mce_amd: Do not print a stray newline

PATCH In review Borislav Petkov 4 participants 10 messages Jul 12 – Aug 14

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 EDAC/mce_amd: Do not print a stray newline

From: "Borislav Petkov (AMD)" <[email protected]>

With error records which don't have FRU text, an empty newline is printed like this:

  [28014.094610] mce: [Hardware Error]: Machine check events logged
  [28014.094674] [Hardware Error]: Corrected error, no action required.
  [28014.094687] [Hardware Error]: CPU:11 (19:44:1) MC0_STATUS[-|CE|MiscV|AddrV|-|-|SyndV|CECC|-|-|-]: 0x9c20400001010135
  [28014.094724] [Hardware Error]: Error Addr: 0x000000019ba9a974
  [28014.094732] [Hardware Error]: IPID: 0x001000b000000000, Syndrome: 0x000000081a1f2505
  [28014.094744] [Hardware Error]: Load Store Unit Ext. Error Code: 1
  [28014.094748] [Hardware Error]: cache level: L1, tx: DATA, mem-tx: DRD

Move the \n into the last printk() call where it belongs.

---
 drivers/edac/mce_amd.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
drivers/edac/mce_amd.c +1 -3
index bd252cb3c38e..ec7395cb7a70 100644
@@ -870,12 +870,10 @@ amd_decode_mce(struct notifier_block *nb, unsigned long val, void *data)
 				memcpy(&frutext[0], &err->vendor.amd.synd1, 8);
 				memcpy(&frutext[8], &err->vendor.amd.synd2, 8);
 
-				pr_emerg(HW_ERR "FRU Text: %s", frutext);
+				pr_emerg(HW_ERR "FRU Text: %s\n", frutext);
 			}
 		}
 
-		pr_cont("\n");
-
 		decode_smca_error(m);
 		goto err_code;
 	}
Signature · 1 line
2.53.0
maintainer Reviewed-by on v1 1/1
#1 lore raw
Quoted 11 lines · Borislav Petkov

From: "Borislav Petkov (AMD)" <[email protected]>

With error records which don't have FRU text, an empty newline is printed like this:

  [28014.094610] mce: [Hardware Error]: Machine check events logged
  [28014.094674] [Hardware Error]: Corrected error, no action required.
  [28014.094687] [Hardware Error]: CPU:11 (19:44:1) MC0_STATUS[-|CE|MiscV|AddrV|-|-|SyndV|CECC|-|-|-]: 0x9c20400001010135
  [28014.094724] [Hardware Error]: Error Addr: 0x000000019ba9a974
  [28014.094732] [Hardware Error]: IPID: 0x001000b000000000, Syndrome: 0x000000081a1f2505
Quoted 6 lines
  [28014.094744] [Hardware Error]: Load Store Unit Ext. Error Code: 1
  [28014.094748] [Hardware Error]: cache level: L1, tx: DATA, mem-tx: DRD

Move the \n into the last printk() call where it belongs.

Thanks,
Yazen

maintainer
#2 lore raw
Quoted 12 lines · Yazen Ghannam
Quoted 11 lines · Borislav Petkov

From: "Borislav Petkov (AMD)" <[email protected]>

With error records which don't have FRU text, an empty newline is printed like this:

  [28014.094610] mce: [Hardware Error]: Machine check events logged
  [28014.094674] [Hardware Error]: Corrected error, no action required.
  [28014.094687] [Hardware Error]: CPU:11 (19:44:1) MC0_STATUS[-|CE|MiscV|AddrV|-|-|SyndV|CECC|-|-|-]: 0x9c20400001010135
  [28014.094724] [Hardware Error]: Error Addr: 0x000000019ba9a974
  [28014.094732] [Hardware Error]: IPID: 0x001000b000000000, Syndrome: 0x000000081a1f2505
Quoted 8 lines
Quoted 6 lines
  [28014.094744] [Hardware Error]: Load Store Unit Ext. Error Code: 1
  [28014.094748] [Hardware Error]: cache level: L1, tx: DATA, mem-tx: DRD

Move the \n into the last printk() call where it belongs.

Yeah, Sashiko has a point:

https://sashiko.dev/#/patchset/20260712031648.486110-1-bp%40kernel.org

So we want the error records to be compact and fit in a block so that you can recognize them easier and thus not print each thing with a newline.

But those things are not unconditional so you cannot always know.

I need to think of a slick scheme what to do here.

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
#3 lore raw
Quoted 24 lines · Borislav Petkov
Quoted 12 lines · Yazen Ghannam
Quoted 11 lines · Borislav Petkov

From: "Borislav Petkov (AMD)" <[email protected]>

With error records which don't have FRU text, an empty newline is printed like this:

  [28014.094610] mce: [Hardware Error]: Machine check events logged
  [28014.094674] [Hardware Error]: Corrected error, no action required.
  [28014.094687] [Hardware Error]: CPU:11 (19:44:1) MC0_STATUS[-|CE|MiscV|AddrV|-|-|SyndV|CECC|-|-|-]: 0x9c20400001010135
  [28014.094724] [Hardware Error]: Error Addr: 0x000000019ba9a974
  [28014.094732] [Hardware Error]: IPID: 0x001000b000000000, Syndrome: 0x000000081a1f2505
Quoted 8 lines
Quoted 6 lines
  [28014.094744] [Hardware Error]: Load Store Unit Ext. Error Code: 1
  [28014.094748] [Hardware Error]: cache level: L1, tx: DATA, mem-tx: DRD

Move the \n into the last printk() call where it belongs.

Yeah, Sashiko has a point:

Quoted 9 lines

https://sashiko.dev/#/patchset/20260712031648.486110-1-bp%40kernel.org

So we want the error records to be compact and fit in a block so that you can recognize them easier and thus not print each thing with a newline.

But those things are not unconditional so you cannot always know.

I need to think of a slick scheme what to do here.

Do you mean something other than moving the newline print into an else block like this?

---
 drivers/edac/mce_amd.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)
drivers/edac/mce_amd.c +7 -3
index bd252cb3c38e..60d1a62e1a95 100644
@@ -870,12 +870,16 @@ amd_decode_mce(struct notifier_block *nb, unsigned long val, void *data)
 				memcpy(&frutext[0], &err->vendor.amd.synd1, 8);
 				memcpy(&frutext[8], &err->vendor.amd.synd2, 8);
 
-				pr_emerg(HW_ERR "FRU Text: %s", frutext);
+				pr_emerg(HW_ERR "FRU Text: %s\n", frutext);
 			}
+		} else {
+			/*
+			 * Explicit line termination to handle nested
+			 * conditional prints correctly.
+			 */
+			pr_cont("\n");
 		}
 
-		pr_cont("\n");
-
 		decode_smca_error(m);
 		goto err_code;
 	}
Signature · 6 lines
2.43.0


Regards,
Srivatsa
Microsoft Linux Systems Group
maintainer
#4 lore raw
Quoted 57 lines · Srivatsa S. Bhat
Quoted 24 lines · Borislav Petkov
Quoted 12 lines · Yazen Ghannam
Quoted 11 lines · Borislav Petkov

From: "Borislav Petkov (AMD)" <[email protected]>

With error records which don't have FRU text, an empty newline is printed like this:

  [28014.094610] mce: [Hardware Error]: Machine check events logged
  [28014.094674] [Hardware Error]: Corrected error, no action required.
  [28014.094687] [Hardware Error]: CPU:11 (19:44:1) MC0_STATUS[-|CE|MiscV|AddrV|-|-|SyndV|CECC|-|-|-]: 0x9c20400001010135
  [28014.094724] [Hardware Error]: Error Addr: 0x000000019ba9a974
  [28014.094732] [Hardware Error]: IPID: 0x001000b000000000, Syndrome: 0x000000081a1f2505
Quoted 8 lines
Quoted 6 lines
  [28014.094744] [Hardware Error]: Load Store Unit Ext. Error Code: 1
  [28014.094748] [Hardware Error]: cache level: L1, tx: DATA, mem-tx: DRD

Move the \n into the last printk() call where it belongs.

Yeah, Sashiko has a point:

Quoted 9 lines

https://sashiko.dev/#/patchset/20260712031648.486110-1-bp%40kernel.org

So we want the error records to be compact and fit in a block so that you can recognize them easier and thus not print each thing with a newline.

But those things are not unconditional so you cannot always know.

I need to think of a slick scheme what to do here.

Do you mean something other than moving the newline print into an else block like this?

---
 drivers/edac/mce_amd.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)
drivers/edac/mce_amd.c +4 -1
index bd252cb3c38e..60d1a62e1a95 100644
@@ -870,12 +870,16 @@ amd_decode_mce(struct notifier_block *nb, unsigned long val, void *data)
 				memcpy(&frutext[0], &err->vendor.amd.synd1, 8);
 				memcpy(&frutext[8], &err->vendor.amd.synd2, 8);
 
-				pr_emerg(HW_ERR "FRU Text: %s", frutext);
+				pr_emerg(HW_ERR "FRU Text: %s\n", frutext);
 			}
+		} else {
+			/*
+			 * Explicit line termination to handle nested

drivers/edac/mce_amd.c amd_decode_mce(struct notifier_block *nb, unsigned long val, void *data) · line 885

Quoted 10 lines
drivers/edac/mce_amd.c +3 -2
+			 * conditional prints correctly.
+			 */
+			pr_cont("\n");
 		}
 
-		pr_cont("\n");
-
 		decode_smca_error(m);
 		goto err_code;
 	}

Another option could be to build up each newline string and print it once. The gist is to get rid of pr_cont() usage.

I've attached an old patch for that. Would this work?

Thanks,
Yazen

From 2035e5b829de2834aef8d84de24bafb5f4e5e03b Mon Sep 17 00:00:00 2001
From: Yazen Ghannam <[email protected]>
Date: Thu, 16 Dec 2021 19:06:14 +0000
Subject: [PATCH] [WIP] EDAC/mce_amd: Construct string before printing

---
 drivers/edac/mce_amd.c | 64 +++++++++++++++++++++++++++++-------------
 1 file changed, 45 insertions(+), 19 deletions(-)
drivers/edac/mce_amd.c +45 -19
index bd252cb3c38e..ea6dc7437cae 100644
@@ -744,6 +744,8 @@ static void decode_smca_error(struct mce *m)
 {
 	enum smca_bank_types bank_type = smca_get_bank_type(m->extcpu, m->bank);
 	u8 xec = XEC(m->status, xec_mask);
+	char buf[256];
+	char *p, *end;
 
 	if (bank_type >= N_SMCA_BANK_TYPES)
 		return;
@@ -753,7 +755,13 @@ static void decode_smca_error(struct mce *m)
 		return;
 	}
 
-	pr_emerg(HW_ERR "%s Ext. Error Code: %d", smca_get_long_name(bank_type), xec);
+	memset(buf, 0, sizeof(buf));
+	p = buf;
+	end = buf + sizeof(buf);
+
+	p += scnprintf(p, end - p, "%s Ext. Error Code: %d", smca_get_long_name(bank_type), xec);
+
+	pr_emerg(HW_ERR "%s", buf);
 
 	if ((bank_type == SMCA_UMC || bank_type == SMCA_UMC_V2) &&
 	    xec == 0 && decode_dram_ecc)
@@ -762,26 +770,33 @@ static void decode_smca_error(struct mce *m)
 
 static inline void amd_decode_err_code(u16 ec)
 {
+	char buf[256];
+	char *p, *end;
+
 	if (INT_ERROR(ec)) {
 		pr_emerg(HW_ERR "internal: %s\n", UU_MSG(ec));
 		return;
 	}
 
-	pr_emerg(HW_ERR "cache level: %s", LL_MSG(ec));
+	memset(buf, 0, sizeof(buf));
+	p = buf;
+	end = buf + sizeof(buf);
+
+	p += scnprintf(p, end - p, "cache level: %s", LL_MSG(ec));
 
 	if (BUS_ERROR(ec))
-		pr_cont(", mem/io: %s", II_MSG(ec));
+		p += scnprintf(p, end - p, ", mem/io: %s", II_MSG(ec));
 	else
-		pr_cont(", tx: %s", TT_MSG(ec));
+		p += scnprintf(p, end - p, ", tx: %s", TT_MSG(ec));
 
 	if (MEM_ERROR(ec) || BUS_ERROR(ec)) {
-		pr_cont(", mem-tx: %s", R4_MSG(ec));
+		p += scnprintf(p, end - p, ", mem-tx: %s", R4_MSG(ec));
 
 		if (BUS_ERROR(ec))
-			pr_cont(", part-proc: %s (%s)", PP_MSG(ec), TO_MSG(ec));
+			p += scnprintf(p, end - p, ", part-proc: %s (%s)", PP_MSG(ec), TO_MSG(ec));
 	}
 
-	pr_cont("\n");
+	pr_emerg(HW_ERR "%s", buf);
 }
 
 static const char *decode_error_status(struct mce *m)
@@ -807,6 +822,8 @@ amd_decode_mce(struct notifier_block *nb, unsigned long val, void *data)
 	struct mce_hw_err *err = to_mce_hw_err(m);
 	unsigned int fam = x86_family(m->cpuid);
 	u32 mca_config_lo = 0, dummy;
+	char buf[256];
+	char *p, *end;
 	int ecc;
 
 	if (m->kflags & MCE_HANDLED_CEC)
@@ -814,7 +831,11 @@ amd_decode_mce(struct notifier_block *nb, unsigned long val, void *data)
 
 	pr_emerg(HW_ERR "%s\n", decode_error_status(m));
 
-	pr_emerg(HW_ERR "CPU:%d (%x:%x:%x) MC%d_STATUS[%s|%s|%s|%s|%s",
+	memset(buf, 0, sizeof(buf));
+	p = buf;
+	end = buf + sizeof(buf);
+
+	p += scnprintf(p, end - p, "CPU:%d (%x:%x:%x) MC%d_STATUS[%s|%s|%s|%s|%s",
 		m->extcpu,
 		fam, x86_model(m->cpuid), x86_stepping(m->cpuid),
 		m->bank,
@@ -829,28 +850,29 @@ amd_decode_mce(struct notifier_block *nb, unsigned long val, void *data)
 		rdmsr_safe(MSR_AMD64_SMCA_MCx_CONFIG(m->bank), &mca_config_lo, &dummy);
 
 		if (mca_config_lo & MCI_CONFIG_MCAX)
-			pr_cont("|%s", ((m->status & MCI_STATUS_TCC) ? "TCC" : "-"));
+			p += scnprintf(p, end - p, "|%s", ((m->status & MCI_STATUS_TCC) ? "TCC" : "-"));
 
-		pr_cont("|%s", ((m->status & MCI_STATUS_SYNDV) ? "SyndV" : "-"));
+		p += scnprintf(p, end - p, "|%s", ((m->status & MCI_STATUS_SYNDV) ? "SyndV" : "-"));
 	}
 
 	/* do the two bits[14:13] together */
 	ecc = (m->status >> 45) & 0x3;
 	if (ecc)
-		pr_cont("|%sECC", ((ecc == 2) ? "C" : "U"));
+		p += scnprintf(p, end - p, "|%sECC", ((ecc == 2) ? "C" : "U"));
 
 	if (fam >= 0x15) {
-		pr_cont("|%s", (m->status & MCI_STATUS_DEFERRED ? "Deferred" : "-"));
+		p += scnprintf(p, end - p, "|%s", (m->status & MCI_STATUS_DEFERRED ? "Deferred" : "-"));
 
 		/* F15h, bank4, bit 43 is part of McaStatSubCache. */
 		if (fam != 0x15 || m->bank != 4)
-			pr_cont("|%s", (m->status & MCI_STATUS_POISON ? "Poison" : "-"));
+			p += scnprintf(p, end - p, "|%s", (m->status & MCI_STATUS_POISON ? "Poison" : "-"));
 	}
 
 	if (fam >= 0x17)
-		pr_cont("|%s", (m->status & MCI_STATUS_SCRUB ? "Scrub" : "-"));
+		p += scnprintf(p, end - p, "|%s", (m->status & MCI_STATUS_SCRUB ? "Scrub" : "-"));
 
-	pr_cont("]: 0x%016llx\n", m->status);
+	p += scnprintf(p, end - p, "]: 0x%016llx\n", m->status);
+	pr_emerg(HW_ERR "%s", buf);
 
 	if (m->status & MCI_STATUS_ADDRV)
 		pr_emerg(HW_ERR "Error Addr: 0x%016llx\n", m->addr);
@@ -859,10 +881,14 @@ amd_decode_mce(struct notifier_block *nb, unsigned long val, void *data)
 		pr_emerg(HW_ERR "PPIN: 0x%016llx\n", m->ppin);
 
 	if (boot_cpu_has(X86_FEATURE_SMCA)) {
-		pr_emerg(HW_ERR "IPID: 0x%016llx", m->ipid);
+		memset(buf, 0, sizeof(buf));
+		p = buf;
+
+		p += scnprintf(p, end - p, "IPID: 0x%016llx", m->ipid);
 
 		if (m->status & MCI_STATUS_SYNDV) {
-			pr_cont(", Syndrome: 0x%016llx\n", m->synd);
+			p += scnprintf(p, end - p, ", Syndrome: 0x%016llx", m->synd);
+			pr_emerg(HW_ERR "%s", buf);
 			if (mca_config_lo & MCI_CONFIG_FRUTEXT) {
 				char frutext[17];
 
@@ -872,10 +898,10 @@ amd_decode_mce(struct notifier_block *nb, unsigned long val, void *data)
 
 				pr_emerg(HW_ERR "FRU Text: %s", frutext);
 			}
+		} else {
+			pr_emerg(HW_ERR "%s", buf);
 		}
 
-		pr_cont("\n");
-
 		decode_smca_error(m);
 		goto err_code;
 	}
Signature · 1 line
2.54.0
#5 lore raw

Hi Yazen,

Quoted 62 lines · Yazen Ghannam
Quoted 61 lines · Srivatsa S. Bhat
Quoted 24 lines · Borislav Petkov
Quoted 12 lines · Yazen Ghannam
Quoted 11 lines · Borislav Petkov

From: "Borislav Petkov (AMD)" <[email protected]>

With error records which don't have FRU text, an empty newline is printed like this:

  [28014.094610] mce: [Hardware Error]: Machine check events logged
  [28014.094674] [Hardware Error]: Corrected error, no action required.
  [28014.094687] [Hardware Error]: CPU:11 (19:44:1) MC0_STATUS[-|CE|MiscV|AddrV|-|-|SyndV|CECC|-|-|-]: 0x9c20400001010135
  [28014.094724] [Hardware Error]: Error Addr: 0x000000019ba9a974
  [28014.094732] [Hardware Error]: IPID: 0x001000b000000000, Syndrome: 0x000000081a1f2505
Quoted 8 lines
Quoted 6 lines
  [28014.094744] [Hardware Error]: Load Store Unit Ext. Error Code: 1
  [28014.094748] [Hardware Error]: cache level: L1, tx: DATA, mem-tx: DRD

Move the \n into the last printk() call where it belongs.

Yeah, Sashiko has a point:

Quoted 9 lines

https://sashiko.dev/#/patchset/20260712031648.486110-1-bp%40kernel.org

So we want the error records to be compact and fit in a block so that you can recognize them easier and thus not print each thing with a newline.

But those things are not unconditional so you cannot always know.

I need to think of a slick scheme what to do here.

Do you mean something other than moving the newline print into an else block like this?

---
 drivers/edac/mce_amd.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)
drivers/edac/mce_amd.c +7 -1
index bd252cb3c38e..60d1a62e1a95 100644
@@ -870,12 +870,16 @@ amd_decode_mce(struct notifier_block *nb, unsigned long val, void *data)
 				memcpy(&frutext[0], &err->vendor.amd.synd1, 8);
 				memcpy(&frutext[8], &err->vendor.amd.synd2, 8);
 
-				pr_emerg(HW_ERR "FRU Text: %s", frutext);
+				pr_emerg(HW_ERR "FRU Text: %s\n", frutext);
 			}
+		} else {
+			/*
+			 * Explicit line termination to handle nested
+			 * conditional prints correctly.
+			 */
+			pr_cont("\n");
 		}
 
Quoted 10 lines

drivers/edac/mce_amd.c amd_decode_mce(struct notifier_block *nb, unsigned long val, void *data) · line 907

Quoted 6 lines
drivers/edac/mce_amd.c +0 -2
-		pr_cont("\n");
-
 		decode_smca_error(m);
 		goto err_code;
 	}

Another option could be to build up each newline string and print it once. The gist is to get rid of pr_cont() usage.

I don't think getting rid of pr_cont() was the intention behind the original patch by Boris; he was trying to address a user-visible stray newline in the logs, if I understood correctly.

Quoted 2 lines

I've attached an old patch for that. Would this work?

The approach to build up each string fully before printing it that you demonstrated below does look like a nice improvement in general, but I'm afraid it may not be a good fit for this particular subsystem though, for the following reasons:

1. Given the EDAC subsystem's role in handling/reporting hardware errors, I think the logging preference for this subsystem would be to get the error message out as soon as possible, as the machine is already likely to be in trouble if we are printing out kernel logs at the criticality level of "emergency". So, delaying these prints to construct the full string may not be desirable, as doing so can potentially cripple the ability to get even the error indicators known upto that point printed out (in case of further occurrence of hardware errors).

2. The static buffer sizing for the character string (256) is potentially prone to buffer overflows in the future as and when these print statements evolve (without taking the buffer size into careful consideration for every change to the string).

3. This is a much larger (cleanup) change than what Boris had posted earlier (which was ~2 lines).

Thoughts?

Regards,
Srivatsa
Microsoft Linux Systems Group

Quoted 172 lines

From 2035e5b829de2834aef8d84de24bafb5f4e5e03b Mon Sep 17 00:00:00 2001
From: Yazen Ghannam <[email protected]>
Date: Thu, 16 Dec 2021 19:06:14 +0000
Subject: [PATCH] [WIP] EDAC/mce_amd: Construct string before printing

---
 drivers/edac/mce_amd.c | 64 +++++++++++++++++++++++++++++-------------
 1 file changed, 45 insertions(+), 19 deletions(-)
drivers/edac/mce_amd.c +45 -19
index bd252cb3c38e..ea6dc7437cae 100644
@@ -744,6 +744,8 @@ static void decode_smca_error(struct mce *m)
 {
 	enum smca_bank_types bank_type = smca_get_bank_type(m->extcpu, m->bank);
 	u8 xec = XEC(m->status, xec_mask);
+	char buf[256];
+	char *p, *end;
 
 	if (bank_type >= N_SMCA_BANK_TYPES)
 		return;
@@ -753,7 +755,13 @@ static void decode_smca_error(struct mce *m)
 		return;
 	}
 
-	pr_emerg(HW_ERR "%s Ext. Error Code: %d", smca_get_long_name(bank_type), xec);
+	memset(buf, 0, sizeof(buf));
+	p = buf;
+	end = buf + sizeof(buf);
+
+	p += scnprintf(p, end - p, "%s Ext. Error Code: %d", smca_get_long_name(bank_type), xec);
+
+	pr_emerg(HW_ERR "%s", buf);
 
 	if ((bank_type == SMCA_UMC || bank_type == SMCA_UMC_V2) &&
 	    xec == 0 && decode_dram_ecc)
@@ -762,26 +770,33 @@ static void decode_smca_error(struct mce *m)
 
 static inline void amd_decode_err_code(u16 ec)
 {
+	char buf[256];
+	char *p, *end;
+
 	if (INT_ERROR(ec)) {
 		pr_emerg(HW_ERR "internal: %s\n", UU_MSG(ec));
 		return;
 	}
 
-	pr_emerg(HW_ERR "cache level: %s", LL_MSG(ec));
+	memset(buf, 0, sizeof(buf));
+	p = buf;
+	end = buf + sizeof(buf);
+
+	p += scnprintf(p, end - p, "cache level: %s", LL_MSG(ec));
 
 	if (BUS_ERROR(ec))
-		pr_cont(", mem/io: %s", II_MSG(ec));
+		p += scnprintf(p, end - p, ", mem/io: %s", II_MSG(ec));
 	else
-		pr_cont(", tx: %s", TT_MSG(ec));
+		p += scnprintf(p, end - p, ", tx: %s", TT_MSG(ec));
 
 	if (MEM_ERROR(ec) || BUS_ERROR(ec)) {
-		pr_cont(", mem-tx: %s", R4_MSG(ec));
+		p += scnprintf(p, end - p, ", mem-tx: %s", R4_MSG(ec));
 
 		if (BUS_ERROR(ec))
-			pr_cont(", part-proc: %s (%s)", PP_MSG(ec), TO_MSG(ec));
+			p += scnprintf(p, end - p, ", part-proc: %s (%s)", PP_MSG(ec), TO_MSG(ec));
 	}
 
-	pr_cont("\n");
+	pr_emerg(HW_ERR "%s", buf);
 }
 
 static const char *decode_error_status(struct mce *m)
@@ -807,6 +822,8 @@ amd_decode_mce(struct notifier_block *nb, unsigned long val, void *data)
 	struct mce_hw_err *err = to_mce_hw_err(m);
 	unsigned int fam = x86_family(m->cpuid);
 	u32 mca_config_lo = 0, dummy;
+	char buf[256];
+	char *p, *end;
 	int ecc;
 
 	if (m->kflags & MCE_HANDLED_CEC)
@@ -814,7 +831,11 @@ amd_decode_mce(struct notifier_block *nb, unsigned long val, void *data)
 
 	pr_emerg(HW_ERR "%s\n", decode_error_status(m));
 
-	pr_emerg(HW_ERR "CPU:%d (%x:%x:%x) MC%d_STATUS[%s|%s|%s|%s|%s",
+	memset(buf, 0, sizeof(buf));
+	p = buf;
+	end = buf + sizeof(buf);
+
+	p += scnprintf(p, end - p, "CPU:%d (%x:%x:%x) MC%d_STATUS[%s|%s|%s|%s|%s",
 		m->extcpu,
 		fam, x86_model(m->cpuid), x86_stepping(m->cpuid),
 		m->bank,
@@ -829,28 +850,29 @@ amd_decode_mce(struct notifier_block *nb, unsigned long val, void *data)
 		rdmsr_safe(MSR_AMD64_SMCA_MCx_CONFIG(m->bank), &mca_config_lo, &dummy);
 
 		if (mca_config_lo & MCI_CONFIG_MCAX)
-			pr_cont("|%s", ((m->status & MCI_STATUS_TCC) ? "TCC" : "-"));
+			p += scnprintf(p, end - p, "|%s", ((m->status & MCI_STATUS_TCC) ? "TCC" : "-"));
 
-		pr_cont("|%s", ((m->status & MCI_STATUS_SYNDV) ? "SyndV" : "-"));
+		p += scnprintf(p, end - p, "|%s", ((m->status & MCI_STATUS_SYNDV) ? "SyndV" : "-"));
 	}
 
 	/* do the two bits[14:13] together */
 	ecc = (m->status >> 45) & 0x3;
 	if (ecc)
-		pr_cont("|%sECC", ((ecc == 2) ? "C" : "U"));
+		p += scnprintf(p, end - p, "|%sECC", ((ecc == 2) ? "C" : "U"));
 
 	if (fam >= 0x15) {
-		pr_cont("|%s", (m->status & MCI_STATUS_DEFERRED ? "Deferred" : "-"));
+		p += scnprintf(p, end - p, "|%s", (m->status & MCI_STATUS_DEFERRED ? "Deferred" : "-"));
 
 		/* F15h, bank4, bit 43 is part of McaStatSubCache. */
 		if (fam != 0x15 || m->bank != 4)
-			pr_cont("|%s", (m->status & MCI_STATUS_POISON ? "Poison" : "-"));
+			p += scnprintf(p, end - p, "|%s", (m->status & MCI_STATUS_POISON ? "Poison" : "-"));
 	}
 
 	if (fam >= 0x17)
-		pr_cont("|%s", (m->status & MCI_STATUS_SCRUB ? "Scrub" : "-"));
+		p += scnprintf(p, end - p, "|%s", (m->status & MCI_STATUS_SCRUB ? "Scrub" : "-"));
 
-	pr_cont("]: 0x%016llx\n", m->status);
+	p += scnprintf(p, end - p, "]: 0x%016llx\n", m->status);
+	pr_emerg(HW_ERR "%s", buf);
 
 	if (m->status & MCI_STATUS_ADDRV)
 		pr_emerg(HW_ERR "Error Addr: 0x%016llx\n", m->addr);
@@ -859,10 +881,14 @@ amd_decode_mce(struct notifier_block *nb, unsigned long val, void *data)
 		pr_emerg(HW_ERR "PPIN: 0x%016llx\n", m->ppin);
 
 	if (boot_cpu_has(X86_FEATURE_SMCA)) {
-		pr_emerg(HW_ERR "IPID: 0x%016llx", m->ipid);
+		memset(buf, 0, sizeof(buf));
+		p = buf;
+
+		p += scnprintf(p, end - p, "IPID: 0x%016llx", m->ipid);
 
 		if (m->status & MCI_STATUS_SYNDV) {
-			pr_cont(", Syndrome: 0x%016llx\n", m->synd);
+			p += scnprintf(p, end - p, ", Syndrome: 0x%016llx", m->synd);
+			pr_emerg(HW_ERR "%s", buf);
 			if (mca_config_lo & MCI_CONFIG_FRUTEXT) {
 				char frutext[17];
 
@@ -872,10 +898,10 @@ amd_decode_mce(struct notifier_block *nb, unsigned long val, void *data)
 
 				pr_emerg(HW_ERR "FRU Text: %s", frutext);
 			}
+		} else {
+			pr_emerg(HW_ERR "%s", buf);
 		}
 
-		pr_cont("\n");
-
 		decode_smca_error(m);
 		goto err_code;
 	}
Signature · 1 line
2.54.0
maintainer
#6 lore raw

On Mon, Jul 20, 2026 at 10:43:10AM +0530, Srivatsa S. Bhat wrote:

[...]

Quoted 50 lines
Quoted 39 lines
Quoted 39 lines
Quoted 12 lines

Yeah, Sashiko has a point:

https://sashiko.dev/#/patchset/20260712031648.486110-1-bp%40kernel.org

So we want the error records to be compact and fit in a block so that you can recognize them easier and thus not print each thing with a newline.

But those things are not unconditional so you cannot always know.

I need to think of a slick scheme what to do here.

Do you mean something other than moving the newline print into an else block like this?

---
 drivers/edac/mce_amd.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)
drivers/edac/mce_amd.c +7 -1
index bd252cb3c38e..60d1a62e1a95 100644
@@ -870,12 +870,16 @@ amd_decode_mce(struct notifier_block *nb, unsigned long val, void *data)
 				memcpy(&frutext[0], &err->vendor.amd.synd1, 8);
 				memcpy(&frutext[8], &err->vendor.amd.synd2, 8);
 
-				pr_emerg(HW_ERR "FRU Text: %s", frutext);
+				pr_emerg(HW_ERR "FRU Text: %s\n", frutext);
 			}
+		} else {
+			/*
+			 * Explicit line termination to handle nested
+			 * conditional prints correctly.
+			 */
+			pr_cont("\n");
 		}
 
Quoted 10 lines

drivers/edac/mce_amd.c amd_decode_mce(struct notifier_block *nb, unsigned long val, void *data) · line 907

Quoted 6 lines
drivers/edac/mce_amd.c +0 -2
-		pr_cont("\n");
-
 		decode_smca_error(m);
 		goto err_code;
 	}

Another option could be to build up each newline string and print it once. The gist is to get rid of pr_cont() usage.

Quoted 3 lines

I don't think getting rid of pr_cont() was the intention behind the original patch by Boris; he was trying to address a user-visible stray newline in the logs, if I understood correctly.

Correct. Though the issue may be resolved with other rework.

Quoted 9 lines
Quoted 3 lines

I've attached an old patch for that. Would this work?

The approach to build up each string fully before printing it that you demonstrated below does look like a nice improvement in general, but I'm afraid it may not be a good fit for this particular subsystem though, for the following reasons:

Quoted 9 lines

1. Given the EDAC subsystem's role in handling/reporting hardware errors, I think the logging preference for this subsystem would be to get the error message out as soon as possible, as the machine is already likely to be in trouble if we are printing out kernel logs at the criticality level of "emergency". So, delaying these prints to construct the full string may not be desirable, as doing so can potentially cripple the ability to get even the error indicators known upto that point printed out (in case of further occurrence of hardware errors).

These EDAC messages are a convenience to the user and are not indicative of the severity.

In fact, this is another low priority item I think should be addressed.

The "printk" log level could be linked to the error severity. For example, "info" for corrected errors and "warning" for uncorrected errors. It is not rational to treat all decoded errors with "emergency" log level.

The x86 EDAC modules listen on the MCE notifier chain. If there is time for EDAC to decode, then the event is not an emergency.

Quoted 5 lines

2. The static buffer sizing for the character string (256) is potentially prone to buffer overflows in the future as and when these print statements evolve (without taking the buffer size into careful consideration for every change to the string).

Yes, that's the goal of patch review.

Quoted 4 lines

3. This is a much larger (cleanup) change than what Boris had posted earlier (which was ~2 lines).

Yes, sometimes simple issues lead to larger cleanup.

Thanks,
Yazen

maintainer
#7 lore raw
In reply to Yazen Ghannam

drivers/edac/mce_amd.c static void decode_smca_error(struct mce *m) · line 764

Quoted 8 lines · Yazen Ghannam
drivers/edac/mce_amd.c +7 -1
-	pr_emerg(HW_ERR "%s Ext. Error Code: %d", smca_get_long_name(bank_type), xec);
+	memset(buf, 0, sizeof(buf));
+	p = buf;
+	end = buf + sizeof(buf);
+
+	p += scnprintf(p, end - p, "%s Ext. Error Code: %d", smca_get_long_name(bank_type), xec);
+
+	pr_emerg(HW_ERR "%s", buf);

Yes, I think this is the right idea, albeit with a couple of improvements:

1. We should define our own vararg log() function - static, used only in this file, which gets a format string and params and vscnprintf()s into our local, heap-allocated string array of 256 initially, we can grow it later if needed

2. Function will be a fire-and-forget and should not care about ptr advancement, buffer size, yadda yadda - log() will do that

3. When string buffer is full, log() will complain ofc.

4. When logging is done, we will have log_print() which goes over the whole string array and splits it into, say, 100-ish chars and breaks the lines then, during printing. We would use a space " " so that we can mark where we can break the string

5. No locking yet - we can do that later, if really needed

This way we'll be able to handle all kinds of error record formats but still keep them compact enough so that they can go out in a couple of printk() calls.

The other advantage when having a buffer like that is, if you look at kdump memory image, you can find the buffer in memory and actually dump its contents
- one more way to get to the error.

Thoughts?

Lemme know if you have cycles to do this. If not, I can take a stab at it myself.

Thx.

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
maintainer
#8 lore raw
Quoted 20 lines · Borislav Petkov

drivers/edac/mce_amd.c static void decode_smca_error(struct mce *m) · line 764

Quoted 8 lines · Yazen Ghannam
drivers/edac/mce_amd.c +7 -1
-	pr_emerg(HW_ERR "%s Ext. Error Code: %d", smca_get_long_name(bank_type), xec);
+	memset(buf, 0, sizeof(buf));
+	p = buf;
+	end = buf + sizeof(buf);
+
+	p += scnprintf(p, end - p, "%s Ext. Error Code: %d", smca_get_long_name(bank_type), xec);
+
+	pr_emerg(HW_ERR "%s", buf);

Yes, I think this is the right idea, albeit with a couple of improvements:

1. We should define our own vararg log() function - static, used only in this file, which gets a format string and params and vscnprintf()s into our local, heap-allocated string array of 256 initially, we can grow it later if needed

2. Function will be a fire-and-forget and should not care about ptr advancement, buffer size, yadda yadda - log() will do that

Quoted 6 lines

3. When string buffer is full, log() will complain ofc.

4. When logging is done, we will have log_print() which goes over the whole string array and splits it into, say, 100-ish chars and breaks the lines then, during printing. We would use a space " " so that we can mark where we can break the string

The delimiter would need more thought. There are a few strings with spaces like "Ext. Error Code", etc. Though maybe you mean we can use extra spaces like double-space between strings?

Quoted 7 lines

5. No locking yet - we can do that later, if really needed

This way we'll be able to handle all kinds of error record formats but still keep them compact enough so that they can go out in a couple of printk() calls.

Quoted 9 lines

The other advantage when having a buffer like that is, if you look at kdump memory image, you can find the buffer in memory and actually dump its contents
- one more way to get to the error.

Thoughts?

Lemme know if you have cycles to do this. If not, I can take a stab at it myself.

I like the idea but I'm not familiar with the concepts. It'll take me some time to figure it out, so I don't have cycles to do it now.

Thanks,
Yazen

maintainer
#9 lore raw
Quoted 3 lines · Yazen Ghannam

The delimiter would need more thought. There are a few strings with spaces like "Ext. Error Code", etc. Though maybe you mean we can use extra spaces like double-space between strings?

That's a good point. I probably should use an explicit delimiter like ';' or so.

Quoted 2 lines

I like the idea but I'm not familiar with the concepts. It'll take me some time to figure it out, so I don't have cycles to do it now.

Ok, lemme take a stab it.

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette