Skip to contentSkip to conversation

x86/mce: retrieve poison range from hardware

PATCH Superseded Jane Chu 3 participants 7 messages Jul 30, 2022 – Aug 1, 2022

Subsystems (4)

v5 of 8 Superseded Latest is v8 →

lore
Share & tools
lore

Jane Chu

author
#0 lore raw

PATCH v5 1/1 x86/mce: retrieve poison range from hardware

With Commit 7917f9cdb503 ("acpi/nfit: rely on mce->misc to determine poison granularity") that changed nfit_handle_mce() callback to report badrange according to 1ULL << MCI_MISC_ADDR_LSB(mce->misc), it's been discovered that the mce->misc LSB field is 0x1000 bytes, hence injecting 2 back-to-back poisons and the driver ends up logging 8 badblocks, because 0x1000 bytes is 8 512-byte.

Dan Williams noticed that apei_mce_report_mem_error() hardcode the LSB field to PAGE_SHIFT instead of consulting the input struct cper_sec_mem_err record. So change to rely on hardware whenever support is available.

---
 arch/x86/kernel/cpu/mce/apei.c | 14 +++++++++++++-
 1 file changed, 13 insertions(+), 1 deletion(-)
arch/x86/kernel/cpu/mce/apei.c +13 -1
index 717192915f28..2c7ea0ba9dd7 100644
@@ -29,15 +29,27 @@
 void apei_mce_report_mem_error(int severity, struct cper_sec_mem_err *mem_err)
 {
 	struct mce m;
+	int lsb = PAGE_SHIFT;
 
 	if (!(mem_err->validation_bits & CPER_MEM_VALID_PA))
 		return;
 
+	/*
+	 * Even if the ->validation_bits are set for address mask,
+	 * to be extra safe, check and reject an error radius '0',
+	 * and fallback to the default page size.
+	 */
+	if (mem_err->validation_bits & CPER_MEM_VALID_PA_MASK) {
+		lsb = __ffs64(mem_err->physical_addr_mask);
+		if (lsb == 1)
+			lsb = PAGE_SHIFT;
+	}
+
 	mce_setup(&m);
 	m.bank = -1;
 	/* Fake a memory read error with unknown channel */
 	m.status = MCI_STATUS_VAL | MCI_STATUS_EN | MCI_STATUS_ADDRV | MCI_STATUS_MISCV | 0x9f;
-	m.misc = (MCI_MISC_ADDR_PHYS << 6) | PAGE_SHIFT;
+	m.misc = (MCI_MISC_ADDR_PHYS << 6) | lsb;
 
 	if (severity >= GHES_SEV_RECOVERABLE)
 		m.status |= MCI_STATUS_UC;
Signature · 1 line
2.18.4
maintainer
#1 lore raw

arch/x86/kernel/cpu/mce/apei.c line 32

Quoted 2 lines
	struct mce m;
arch/x86/kernel/cpu/mce/apei.c +1 -0
+	int lsb = PAGE_SHIFT;

Some maintainers like to order local declaration lines from longest to shortest

arch/x86/kernel/cpu/mce/apei.c line 46

Quoted 10 lines
arch/x86/kernel/cpu/mce/apei.c +10 -0
+     /*
+      * Even if the ->validation_bits are set for address mask,
+      * to be extra safe, check and reject an error radius '0',
+      * and fallback to the default page size.
+      */
+     if (mem_err->validation_bits & CPER_MEM_VALID_PA_MASK) {
+             lsb = __ffs64(mem_err->physical_addr_mask);
+             if (lsb == 1)
+                     lsb = PAGE_SHIFT;
+     }

The comment above __ffs64() says:

* The result is not defined if no bits are set, so check that @word
* is non-zero before calling this.

So if the intent is "extra safe" should check for that:

        if (mem_err->validation_bits & CPER_MEM_VALID_PA_MASK &&
            mem_err->physical_addr_mask) {
                lsb = __ffs64(mem_err->physical_addr_mask);
                if (lsb == 1)
                        lsb = PAGE_SHIFT;
        }

-Tony

author
#2 lore raw
Quoted 22 lines · Tony

arch/x86/kernel/cpu/mce/apei.c line 32

Quoted 2 lines
	struct mce m;
arch/x86/kernel/cpu/mce/apei.c +1 -0
+	int lsb = PAGE_SHIFT;

Some maintainers like to order local declaration lines from longest to shortest

arch/x86/kernel/cpu/mce/apei.c line 46

Quoted 10 lines
arch/x86/kernel/cpu/mce/apei.c +10 -0
+     /*
+      * Even if the ->validation_bits are set for address mask,
+      * to be extra safe, check and reject an error radius '0',
+      * and fallback to the default page size.
+      */
+     if (mem_err->validation_bits & CPER_MEM_VALID_PA_MASK) {
+             lsb = __ffs64(mem_err->physical_addr_mask);
+             if (lsb == 1)
+                     lsb = PAGE_SHIFT;
+     }

The comment above __ffs64() says:

* The result is not defined if no bits are set, so check that @word
* is non-zero before calling this.

Quoted 8 lines

So if the intent is "extra safe" should check for that:

         if (mem_err->validation_bits & CPER_MEM_VALID_PA_MASK &&
             mem_err->physical_addr_mask) {
                 lsb = __ffs64(mem_err->physical_addr_mask);
                 if (lsb == 1)
                         lsb = PAGE_SHIFT;
         }

Indeed, thanks a lot!

-jane

Quoted 5 lines

-Tony

#3 lore raw
In reply to Jane Chu
Quoted 31 lines · Jane Chu

With Commit 7917f9cdb503 ("acpi/nfit: rely on mce->misc to determine poison granularity") that changed nfit_handle_mce() callback to report badrange according to 1ULL << MCI_MISC_ADDR_LSB(mce->misc), it's been discovered that the mce->misc LSB field is 0x1000 bytes, hence injecting 2 back-to-back poisons and the driver ends up logging 8 badblocks, because 0x1000 bytes is 8 512-byte.

Dan Williams noticed that apei_mce_report_mem_error() hardcode the LSB field to PAGE_SHIFT instead of consulting the input struct cper_sec_mem_err record. So change to rely on hardware whenever support is available.

---
 arch/x86/kernel/cpu/mce/apei.c | 14 +++++++++++++-
 1 file changed, 13 insertions(+), 1 deletion(-)
arch/x86/kernel/cpu/mce/apei.c +1 -0
index 717192915f28..2c7ea0ba9dd7 100644
@@ -29,15 +29,27 @@
 void apei_mce_report_mem_error(int severity, struct cper_sec_mem_err *mem_err)
 {
 	struct mce m;
+	int lsb = PAGE_SHIFT;
 
 	if (!(mem_err->validation_bits & CPER_MEM_VALID_PA))

arch/x86/kernel/cpu/mce/apei.c line 44

Quoted 10 lines
arch/x86/kernel/cpu/mce/apei.c +8 -0
 		return;
 
+	/*
+	 * Even if the ->validation_bits are set for address mask,
+	 * to be extra safe, check and reject an error radius '0',
+	 * and fallback to the default page size.
+	 */
+	if (mem_err->validation_bits & CPER_MEM_VALID_PA_MASK) {
+		lsb = __ffs64(mem_err->physical_addr_mask);
+		if (lsb == 1)

This was the reason I recommended hweight64 and min_not_zero() as hweight64 does not have the undefined behavior. However, an even better option is to just do:

    find_first_bit(&mem_err->physical_addr_mask, PAGE_SHIFT)

...as that trims the result to the PAGE_SHIFT max and handles the zero case.

author
#4 lore raw
Quoted 47 lines · Dan Williams
Quoted 31 lines · Jane Chu

With Commit 7917f9cdb503 ("acpi/nfit: rely on mce->misc to determine poison granularity") that changed nfit_handle_mce() callback to report badrange according to 1ULL << MCI_MISC_ADDR_LSB(mce->misc), it's been discovered that the mce->misc LSB field is 0x1000 bytes, hence injecting 2 back-to-back poisons and the driver ends up logging 8 badblocks, because 0x1000 bytes is 8 512-byte.

Dan Williams noticed that apei_mce_report_mem_error() hardcode the LSB field to PAGE_SHIFT instead of consulting the input struct cper_sec_mem_err record. So change to rely on hardware whenever support is available.

---
  arch/x86/kernel/cpu/mce/apei.c | 14 +++++++++++++-
  1 file changed, 13 insertions(+), 1 deletion(-)
arch/x86/kernel/cpu/mce/apei.c +1 -0
index 717192915f28..2c7ea0ba9dd7 100644
@@ -29,15 +29,27 @@
  void apei_mce_report_mem_error(int severity, struct cper_sec_mem_err *mem_err)
  {
  	struct mce m;
+	int lsb = PAGE_SHIFT;
 
  	if (!(mem_err->validation_bits & CPER_MEM_VALID_PA))

arch/x86/kernel/cpu/mce/apei.c line 44

Quoted 10 lines
arch/x86/kernel/cpu/mce/apei.c +8 -0
  		return;
 
+	/*
+	 * Even if the ->validation_bits are set for address mask,
+	 * to be extra safe, check and reject an error radius '0',
+	 * and fallback to the default page size.
+	 */
+	if (mem_err->validation_bits & CPER_MEM_VALID_PA_MASK) {
+		lsb = __ffs64(mem_err->physical_addr_mask);
+		if (lsb == 1)

This was the reason I recommended hweight64 and min_not_zero() as hweight64 does not have the undefined behavior. However, an even better option is to just do:

Quoted 4 lines
     find_first_bit(&mem_err->physical_addr_mask, PAGE_SHIFT)

...as that trims the result to the PAGE_SHIFT max and handles the zero case.

Thanks Dan! However it looks like find_first_bit() could call into __ffs(x) which has the same limitation as __ffs64(x), as Tony pointed out.

I'll post v6 shortly.

thanks,
-jane

#5 lore raw
Quoted 48 lines · Jane Chu
Quoted 47 lines · Dan Williams
Quoted 31 lines · Jane Chu

With Commit 7917f9cdb503 ("acpi/nfit: rely on mce->misc to determine poison granularity") that changed nfit_handle_mce() callback to report badrange according to 1ULL << MCI_MISC_ADDR_LSB(mce->misc), it's been discovered that the mce->misc LSB field is 0x1000 bytes, hence injecting 2 back-to-back poisons and the driver ends up logging 8 badblocks, because 0x1000 bytes is 8 512-byte.

Dan Williams noticed that apei_mce_report_mem_error() hardcode the LSB field to PAGE_SHIFT instead of consulting the input struct cper_sec_mem_err record. So change to rely on hardware whenever support is available.

---
  arch/x86/kernel/cpu/mce/apei.c | 14 +++++++++++++-
  1 file changed, 13 insertions(+), 1 deletion(-)
arch/x86/kernel/cpu/mce/apei.c +1 -0
index 717192915f28..2c7ea0ba9dd7 100644
@@ -29,15 +29,27 @@
  void apei_mce_report_mem_error(int severity, struct cper_sec_mem_err *mem_err)
  {
  	struct mce m;
+	int lsb = PAGE_SHIFT;
 
  	if (!(mem_err->validation_bits & CPER_MEM_VALID_PA))

arch/x86/kernel/cpu/mce/apei.c line 44

Quoted 10 lines
arch/x86/kernel/cpu/mce/apei.c +8 -0
  		return;
 
+	/*
+	 * Even if the ->validation_bits are set for address mask,
+	 * to be extra safe, check and reject an error radius '0',
+	 * and fallback to the default page size.
+	 */
+	if (mem_err->validation_bits & CPER_MEM_VALID_PA_MASK) {
+		lsb = __ffs64(mem_err->physical_addr_mask);
+		if (lsb == 1)

This was the reason I recommended hweight64 and min_not_zero() as hweight64 does not have the undefined behavior. However, an even better option is to just do:

Quoted 7 lines
Quoted 4 lines
     find_first_bit(&mem_err->physical_addr_mask, PAGE_SHIFT)

...as that trims the result to the PAGE_SHIFT max and handles the zero case.

Thanks Dan! However it looks like find_first_bit() could call into __ffs(x) which has the same limitation as __ffs64(x), as Tony pointed out.

Not quite, no. __ffs() behavior is *undefined* if the input is zero. find_first_bit() is *defined* and returns @size is the input is zero. Which is the behavior this wants to default to PAGE_SHIFT in the absence of any smaller granularity information.

author
#6 lore raw
Quoted 54 lines · Dan Williams
Quoted 53 lines · Jane Chu
Quoted 51 lines · Dan Williams
Quoted 31 lines · Jane Chu

With Commit 7917f9cdb503 ("acpi/nfit: rely on mce->misc to determine poison granularity") that changed nfit_handle_mce() callback to report badrange according to 1ULL << MCI_MISC_ADDR_LSB(mce->misc), it's been discovered that the mce->misc LSB field is 0x1000 bytes, hence injecting 2 back-to-back poisons and the driver ends up logging 8 badblocks, because 0x1000 bytes is 8 512-byte.

Dan Williams noticed that apei_mce_report_mem_error() hardcode the LSB field to PAGE_SHIFT instead of consulting the input struct cper_sec_mem_err record. So change to rely on hardware whenever support is available.

---
   arch/x86/kernel/cpu/mce/apei.c | 14 +++++++++++++-
   1 file changed, 13 insertions(+), 1 deletion(-)
arch/x86/kernel/cpu/mce/apei.c +1 -0
index 717192915f28..2c7ea0ba9dd7 100644
@@ -29,15 +29,27 @@
   void apei_mce_report_mem_error(int severity, struct cper_sec_mem_err *mem_err)
   {
   	struct mce m;
+	int lsb = PAGE_SHIFT;
 
   	if (!(mem_err->validation_bits & CPER_MEM_VALID_PA))

arch/x86/kernel/cpu/mce/apei.c line 44

Quoted 10 lines
arch/x86/kernel/cpu/mce/apei.c +8 -0
   		return;
 
+	/*
+	 * Even if the ->validation_bits are set for address mask,
+	 * to be extra safe, check and reject an error radius '0',
+	 * and fallback to the default page size.
+	 */
+	if (mem_err->validation_bits & CPER_MEM_VALID_PA_MASK) {
+		lsb = __ffs64(mem_err->physical_addr_mask);
+		if (lsb == 1)

This was the reason I recommended hweight64 and min_not_zero() as hweight64 does not have the undefined behavior. However, an even better option is to just do:

      find_first_bit(&mem_err->physical_addr_mask, PAGE_SHIFT)

...as that trims the result to the PAGE_SHIFT max and handles the zero case.

Quoted 8 lines
Quoted 2 lines

Thanks Dan! However it looks like find_first_bit() could call into __ffs(x) which has the same limitation as __ffs64(x), as Tony pointed out.

Not quite, no. __ffs() behavior is *undefined* if the input is zero. find_first_bit() is *defined* and returns @size is the input is zero. Which is the behavior this wants to default to PAGE_SHIFT in the absence of any smaller granularity information.

You're right, because of this line -

    return val ? __ffs(val) : size;

Thanks!
-jane