[PATCH v2 0/2] AMD MCE fixes

[PATCH v2 0/2] AMD MCE fixes

[Boris Ostrovsky]

Boris,

Here is the updated patch for determining number of regiter banks on
AMD plus a patch removing shared_bank array, as you suggested.

Offline/online testing didn't show any issues.



Boris Ostrovsky (2):
  x86/mce: Replace shared_bank array with is_shared_bank() helper
  x86/mce: Use MCG_CAP MSR to find out number of banks on AMD

 arch/x86/kernel/cpu/mcheck/mce_amd.c | 38 ++++++++++++++++++++++--------------
 1 file changed, 23 insertions(+), 15 deletions(-)

-- 
1.8.1.2

[PATCH v2 1/2] x86/mce: Replace shared_bank array with is_shared_bank() helper

[Boris Ostrovsky]

Use helper function instead of an array to report whether register
bank is shared. Currently only bank 4 (northbridge) is shared.

Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
---
 arch/x86/kernel/cpu/mcheck/mce_amd.c | 17 +++++++++--------
 1 file changed, 9 insertions(+), 8 deletions(-)

diff --git a/arch/x86/kernel/cpu/mcheck/mce_amd.c b/arch/x86/kernel/cpu/mcheck/mce_amd.c
index 1ac581f..654a155 100644
--- a/arch/x86/kernel/cpu/mcheck/mce_amd.c
+++ b/arch/x86/kernel/cpu/mcheck/mce_amd.c
@@ -58,11 +58,6 @@ static const char * const th_names[] = {
 };
 
 static DEFINE_PER_CPU(struct threshold_bank * [NR_BANKS], threshold_banks);
-
-static unsigned char shared_bank[NR_BANKS] = {
-	0, 0, 0, 0, 1
-};
-
 static DEFINE_PER_CPU(unsigned char, bank_map);	/* see which banks are on */
 
 static void amd_threshold_interrupt(void);
@@ -79,6 +74,12 @@ struct thresh_restart {
 	u16			old_limit;
 };
 
+static inline bool is_shared_bank(int bank)
+{
+	/* Bank 4 is for northbridge reporting is thus is shared */
+	return (bank == 4);
+}
+
 static const char * const bank4_names(struct threshold_block *b)
 {
 	switch (b->address) {
@@ -575,7 +576,7 @@ static __cpuinit int threshold_create_bank(unsigned int cpu, unsigned int bank)
 	const char *name = th_names[bank];
 	int err = 0;
 
-	if (shared_bank[bank]) {
+	if (is_shared_bank(bank)) {
 		nb = node_to_amd_nb(amd_get_nb_id(cpu));
 
 		/* threshold descriptor already initialized on this node? */
@@ -609,7 +610,7 @@ static __cpuinit int threshold_create_bank(unsigned int cpu, unsigned int bank)
 
 	per_cpu(threshold_banks, cpu)[bank] = b;
 
-	if (shared_bank[bank]) {
+	if (is_shared_bank(bank)) {
 		atomic_set(&b->cpus, 1);
 
 		/* nb is already initialized, see above */
@@ -691,7 +692,7 @@ static void threshold_remove_bank(unsigned int cpu, int bank)
 	if (!b->blocks)
 		goto free_out;
 
-	if (shared_bank[bank]) {
+	if (is_shared_bank(bank)) {
 		if (!atomic_dec_and_test(&b->cpus)) {
 			__threshold_remove_blocks(b);
 			per_cpu(threshold_banks, cpu)[bank] = NULL;
-- 
1.8.1.2

Re: [PATCH v2 1/2] x86/mce: Replace shared_bank array with is_shared_bank() helper

[Greg KH]

On Thu, Mar 14, 2013 at 05:10:40PM -0400, Boris Ostrovsky wrote:
> Use helper function instead of an array to report whether register
> bank is shared. Currently only bank 4 (northbridge) is shared.
> 
> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
> ---
>  arch/x86/kernel/cpu/mcheck/mce_amd.c | 17 +++++++++--------
>  1 file changed, 9 insertions(+), 8 deletions(-)

<formletter>

This is not the correct way to submit patches for inclusion in the
stable kernel tree.  Please read Documentation/stable_kernel_rules.txt
for how to do this properly.

</formletter>

Re: [PATCH v2 1/2] x86/mce: Replace shared_bank array with is_shared_bank() helper

[Borislav Petkov]

On Thu, Mar 14, 2013 at 03:20:05PM -0700, Greg KH wrote:
> On Thu, Mar 14, 2013 at 05:10:40PM -0400, Boris Ostrovsky wrote:
> > Use helper function instead of an array to report whether register
> > bank is shared. Currently only bank 4 (northbridge) is shared.
> > 
> > Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
> > ---
> >  arch/x86/kernel/cpu/mcheck/mce_amd.c | 17 +++++++++--------
> >  1 file changed, 9 insertions(+), 8 deletions(-)
> 
> <formletter>
> 
> This is not the correct way to submit patches for inclusion in the
> stable kernel tree.  Please read Documentation/stable_kernel_rules.txt
> for how to do this properly.

Yeah,

stable@ was obviously wrongly CCed. This patch is clearly not stable
material.

Thanks.

-- 
Regards/Gruss,
    Boris.

Sent from a fat crate under my desk. Formatting is fine.
--

[PATCH v2 2/2] x86/mce: Use MCG_CAP MSR to find out number of banks on AMD

[Boris Ostrovsky]

Currently number of error reporting register banks is hardcoded to
6 on AMD processors. This may break in virtualized scenarios when
a hypervisor prefers to report fewer banks than what the physical
HW provides.

Since number of supported banks is reported in MSR_IA32_MCG_CAP[7:0]
that's what we should use.

Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
---
 arch/x86/kernel/cpu/mcheck/mce_amd.c | 21 ++++++++++++++-------
 1 file changed, 14 insertions(+), 7 deletions(-)

diff --git a/arch/x86/kernel/cpu/mcheck/mce_amd.c b/arch/x86/kernel/cpu/mcheck/mce_amd.c
index 654a155..13a22e2 100644
--- a/arch/x86/kernel/cpu/mcheck/mce_amd.c
+++ b/arch/x86/kernel/cpu/mcheck/mce_amd.c
@@ -33,7 +33,6 @@
 #include <asm/mce.h>
 #include <asm/msr.h>
 
-#define NR_BANKS          6
 #define NR_BLOCKS         9
 #define THRESHOLD_MAX     0xFFF
 #define INT_TYPE_APIC     0x00020000
@@ -57,7 +56,7 @@ static const char * const th_names[] = {
 	"execution_unit",
 };
 
-static DEFINE_PER_CPU(struct threshold_bank * [NR_BANKS], threshold_banks);
+static DEFINE_PER_CPU(struct threshold_bank **, threshold_banks);
 static DEFINE_PER_CPU(unsigned char, bank_map);	/* see which banks are on */
 
 static void amd_threshold_interrupt(void);
@@ -215,7 +214,7 @@ void mce_amd_feature_init(struct cpuinfo_x86 *c)
 	unsigned int bank, block;
 	int offset = -1;
 
-	for (bank = 0; bank < NR_BANKS; ++bank) {
+	for (bank = 0; bank < mca_cfg.banks; ++bank) {
 		for (block = 0; block < NR_BLOCKS; ++block) {
 			if (block == 0)
 				address = MSR_IA32_MC0_MISC + bank * 4;
@@ -277,7 +276,7 @@ static void amd_threshold_interrupt(void)
 	mce_setup(&m);
 
 	/* assume first bank caused it */
-	for (bank = 0; bank < NR_BANKS; ++bank) {
+	for (bank = 0; bank < mca_cfg.banks; ++bank) {
 		if (!(per_cpu(bank_map, m.cpu) & (1 << bank)))
 			continue;
 		for (block = 0; block < NR_BLOCKS; ++block) {
@@ -468,7 +467,7 @@ static __cpuinit int allocate_threshold_blocks(unsigned int cpu,
 	u32 low, high;
 	int err;
 
-	if ((bank >= NR_BANKS) || (block >= NR_BLOCKS))
+	if ((bank >= mca_cfg.banks) || (block >= NR_BLOCKS))
 		return 0;
 
 	if (rdmsr_safe_on_cpu(cpu, address, &low, &high))
@@ -636,9 +635,16 @@ static __cpuinit int threshold_create_bank(unsigned int cpu, unsigned int bank)
 static __cpuinit int threshold_create_device(unsigned int cpu)
 {
 	unsigned int bank;
+	struct threshold_bank **bp;
 	int err = 0;
 
-	for (bank = 0; bank < NR_BANKS; ++bank) {
+	bp = kzalloc(sizeof(struct threshold_bank *) * mca_cfg.banks,
+		     GFP_KERNEL);
+	if (bp == NULL)
+		return -ENOMEM;
+	per_cpu(threshold_banks, cpu) = bp;
+
+	for (bank = 0; bank < mca_cfg.banks; ++bank) {
 		if (!(per_cpu(bank_map, cpu) & (1 << bank)))
 			continue;
 		err = threshold_create_bank(cpu, bank);
@@ -720,11 +726,12 @@ static void threshold_remove_device(unsigned int cpu)
 {
 	unsigned int bank;
 
-	for (bank = 0; bank < NR_BANKS; ++bank) {
+	for (bank = 0; bank < mca_cfg.banks; ++bank) {
 		if (!(per_cpu(bank_map, cpu) & (1 << bank)))
 			continue;
 		threshold_remove_bank(cpu, bank);
 	}
+	kfree(per_cpu(threshold_banks, cpu));
 }
 
 /* get notified when a cpu comes on/off */
-- 
1.8.1.2

Re: [PATCH v2 0/2] AMD MCE fixes

[Borislav Petkov]

On Thu, Mar 14, 2013 at 05:10:39PM -0400, Boris Ostrovsky wrote:
> 
> Boris,
> 
> Here is the updated patch for determining number of regiter banks on
> AMD plus a patch removing shared_bank array, as you suggested.
> 
> Offline/online testing didn't show any issues.
> 
> 
> 
> Boris Ostrovsky (2):
>   x86/mce: Replace shared_bank array with is_shared_bank() helper
>   x86/mce: Use MCG_CAP MSR to find out number of banks on AMD

Both applied,

thanks.

-- 
Regards/Gruss,
    Boris.

Sent from a fat crate under my desk. Formatting is fine.
--