x86, cacheinfo, amd: L3 Cache Index Disable fixes
Subsystems (2)
cover x86, cacheinfo, amd: L3 Cache Index Disable fixes
From: Borislav Petkov <[email protected]>
Changelog:
-v3: Add smp WBINVD functionality in a less uglier way :)
-v2: Split smp functionality into a different module
===
these are a bunch of fixes which correct the disabling of L3 cache indexes. This is needed, for example, in the case when excessive correctable L3 error rates are observed. In a later step, this functionality will be connected to the mcheck mechanism and used to evaluate occurring L3 tag and data array errors and then disable the respective failing L3 cache regions in order to defer system failure before a sysop can intervene.
Those patches are also good -stable candidates.
Please apply,
thanks.
PATCH 1/5 x86, lib: Add wbinvd smp helpers
From: Borislav Petkov <[email protected]>
Add wbinvd_on_cpu and wbinvd_on_all_cpus stubs for executing wbinvd on a particular CPU.
--- arch/x86/include/asm/smp.h | 5 +++++ arch/x86/lib/Makefile | 1 + arch/x86/lib/smp.c | 19 +++++++++++++++++++ 3 files changed, 25 insertions(+), 0 deletions(-) create mode 100644 arch/x86/lib/smp.c
arch/x86/include/asm/smp.h +5 -0
@@ -135,6 +135,8 @@ int native_cpu_disable(void);void native_cpu_die (unsigned int cpu);void native_play_dead (void );void play_dead_common (void ); +void wbinvd_on_cpu (int cpu); +int wbinvd_on_all_cpus (void );void native_send_call_func_ipi (const struct cpumask *mask);void native_send_call_func_single_ipi (int cpu);
@@ -147,6 +149,9 @@ static inline int num_booting_cpus(void) {return cpumask_weight (cpu_callout_mask); } +#else /* !CONFIG_SMP */ +#define wbinvd_on_cpu (cpu) wbinvd() +#define wbinvd_on_all_cpus () wbinvd()#endif /* CONFIG_SMP */ extern unsigned disabled_cpus__cpuinitdata ;
arch/x86/lib/Makefile +1 -0
@@ -15,6 +15,7 @@ $(obj)/inat.o: $(obj)/inat-tables.c clean-files := inat-tables.c obj-$(CONFIG_SMP) += msr-smp.o +obj-$(CONFIG_SMP) += smp.o lib-y := delay.o lib-y += thunk_$(BITS).o
arch/x86/lib/smp.c new file +19 -0
@@ -0,0 +1,19 @@ +#include <linux/smp.h> +#include <linux/module.h> + +static void __wbinvd (void *dummy) +{ +wbinvd (); +} + +void wbinvd_on_cpu (int cpu) +{ +smp_call_function_single (cpu, __wbinvd,NULL ,1 ); +} +EXPORT_SYMBOL (wbinvd_on_cpu); + +int wbinvd_on_all_cpus (void ) +{ +return on_each_cpu (__wbinvd,NULL ,1 ); +} +EXPORT_SYMBOL (wbinvd_on_all_cpus);
Signature · 1 line
1.6.6
[tip:x86/cpu] x86, lib: Add wbinvd smp helpers
Commit-ID: a7b480e7f30b3813353ec009f10f2ac7a6669f3b Gitweb: http://git.kernel.org/tip/a7b480e7f30b3813353ec009f10f2ac7a6669f3b Author: Borislav Petkov <[email protected]> AuthorDate: Fri, 22 Jan 2010 16:01:03 +0100 Committer: H. Peter Anvin <[email protected]> CommitDate: Fri, 22 Jan 2010 16:05:42 -0800
x86, lib: Add wbinvd smp helpers
Add wbinvd_on_cpu and wbinvd_on_all_cpus stubs for executing wbinvd on a particular CPU.
[ hpa: renamed lib/smp.c to lib/cache-smp.c ]
[ hpa: wbinvd_on_all_cpus() returns int, but wbinvd() returns
void. Thus, the former cannot be a macro for the latter, replace with an inline function. ]
LKML-Reference: <[email protected]>
--- arch/x86/include/asm/smp.h | 9 +++++++++ arch/x86/lib/Makefile | 2 +- arch/x86/lib/cache-smp.c | 19 +++++++++++++++++++ 3 files changed, 29 insertions(+), 1 deletions(-)
arch/x86/include/asm/smp.h +9 -0
@@ -135,6 +135,8 @@ int native_cpu_disable(void);void native_cpu_die (unsigned int cpu);void native_play_dead (void );void play_dead_common (void ); +void wbinvd_on_cpu (int cpu); +int wbinvd_on_all_cpus (void );void native_send_call_func_ipi (const struct cpumask *mask);void native_send_call_func_single_ipi (int cpu);
@@ -147,6 +149,13 @@ static inline int num_booting_cpus(void) {return cpumask_weight (cpu_callout_mask); } +#else /* !CONFIG_SMP */ +#define wbinvd_on_cpu (cpu) wbinvd() +static inline int wbinvd_on_all_cpus (void ) +{ +wbinvd (); +return 0 ; +}#endif /* CONFIG_SMP */ extern unsigned disabled_cpus__cpuinitdata ;
arch/x86/lib/Makefile +1 -1
@@ -14,7 +14,7 @@ $(obj)/inat.o: $(obj)/inat-tables.c clean-files := inat-tables.c -obj-$(CONFIG_SMP) += msr-smp.o +obj-$(CONFIG_SMP) += msr-smp.o cache-smp.o lib-y := delay.o lib-y += thunk_$(BITS).o
arch/x86/lib/cache-smp.c new file +19 -0
@@ -0,0 +1,19 @@ +#include <linux/smp.h> +#include <linux/module.h> + +static void __wbinvd (void *dummy) +{ +wbinvd (); +} + +void wbinvd_on_cpu (int cpu) +{ +smp_call_function_single (cpu, __wbinvd,NULL ,1 ); +} +EXPORT_SYMBOL (wbinvd_on_cpu); + +int wbinvd_on_all_cpus (void ) +{ +return on_each_cpu (__wbinvd,NULL ,1 ); +} +EXPORT_SYMBOL (wbinvd_on_all_cpus);
PATCH 2/5 intel-agp: Switch to wbinvd_on_all_cpus
In reply to Borislav PetkovFrom: Borislav Petkov <[email protected]>
Simplify if-statement while at it.
--- drivers/char/agp/intel-agp.c | 14 +++----------- 1 files changed, 3 insertions(+), 11 deletions(-)
drivers/char/agp/intel-agp.c +3 -11
@@ -815,12 +815,6 @@ static void intel_i830_setup_flush(void)intel_i830_fini_flush (); } -static void -do_wbinvd (void *null) -{ -wbinvd (); -} -/* The chipset_flush interface needs to get data that has already been * flushed out of the CPU all the way out to main memory, because the GPU * doesn't snoop those buffers.
@@ -837,12 +831,10 @@ static void intel_i830_chipset_flush(struct agp_bridge_data *bridge)memset (pg,0 ,1024 ); -if (cpu_has_clflush) { +if (cpu_has_clflush)clflush_cache_range (pg,1024 ); - }else { -if (on_each_cpu (do_wbinvd,NULL ,1 ) !=0 ) -printk (KERN_ERR "Timed out waiting for cache flush.\n" ); - } +else if (wbinvd_on_all_cpus () !=0 ) +printk (KERN_ERR "Timed out waiting for cache flush.\n" ); }/* The intel i830 automatically initializes the agp aperture during POST.
Signature · 1 line
1.6.6
[tip:x86/cpu] intel-agp: Switch to wbinvd_on_all_cpus
Commit-ID: 48a719c238bcbb72d6da79de9c5b3b93ab472107 Gitweb: http://git.kernel.org/tip/48a719c238bcbb72d6da79de9c5b3b93ab472107 Author: Borislav Petkov <[email protected]> AuthorDate: Fri, 22 Jan 2010 16:01:04 +0100 Committer: H. Peter Anvin <[email protected]> CommitDate: Fri, 22 Jan 2010 16:06:30 -0800
intel-agp: Switch to wbinvd_on_all_cpus
Simplify if-statement while at it.
[ hpa: we need to #include <asm/smp.h> ]
LKML-Reference: <[email protected]>
--- drivers/char/agp/intel-agp.c | 15 ++++----------- 1 files changed, 4 insertions(+), 11 deletions(-)
drivers/char/agp/intel-agp.c +4 -11
@@ -8,6 +8,7 @@#include <linux/kernel.h> #include <linux/pagemap.h> #include <linux/agp_backend.h> +#include <asm/smp.h> #include "agp.h" /*
@@ -815,12 +816,6 @@ static void intel_i830_setup_flush(void)intel_i830_fini_flush (); } -static void -do_wbinvd (void *null) -{ -wbinvd (); -} -/* The chipset_flush interface needs to get data that has already been * flushed out of the CPU all the way out to main memory, because the GPU * doesn't snoop those buffers.
@@ -837,12 +832,10 @@ static void intel_i830_chipset_flush(struct agp_bridge_data *bridge)memset (pg,0 ,1024 ); -if (cpu_has_clflush) { +if (cpu_has_clflush)clflush_cache_range (pg,1024 ); - }else { -if (on_each_cpu (do_wbinvd,NULL ,1 ) !=0 ) -printk (KERN_ERR "Timed out waiting for cache flush.\n" ); - } +else if (wbinvd_on_all_cpus () !=0 ) +printk (KERN_ERR "Timed out waiting for cache flush.\n" ); }/* The intel i830 automatically initializes the agp aperture during POST.
PATCH 3/5 x86, cacheinfo: Fix disabling of L3 cache indices
In reply to Borislav PetkovFrom: Borislav Petkov <[email protected]>
* Correct the masks used for writing the cache index disable indices.
* Do not turn off L3 scrubber - it is not necessary.
* Make sure wbinvd is executed on the same node where the L3 is.
* Check for out-of-bounds values written to the registers.
* Make show_cache_disable hex values unambiguous
* Check for Erratum #388
--- arch/x86/kernel/cpu/intel_cacheinfo.c | 34 ++++++++++++++++++++------------ 1 files changed, 21 insertions(+), 13 deletions(-)
arch/x86/kernel/cpu/intel_cacheinfo.c +21 -13
@@ -18,6 +18,7 @@#include <asm/processor.h> #include <linux/smp.h> #include <asm/k8.h> +#include <asm/smp.h> #define LVL_1_INST 1#define LVL_1_DATA 2
@@ -299,8 +300,10 @@ amd_check_l3_disable(int index, struct _cpuid4_info_regs *this_leaf)if (boot_cpu_data.x86 ==0x11 )return ; -/* see erratum #382 */ -if ((boot_cpu_data.x86 ==0x10 ) && (boot_cpu_data.x86_model <0x8 )) +/* see errata #382 and #388 */ +if ((boot_cpu_data.x86 ==0x10 ) && + ((boot_cpu_data.x86_model <0x9 ) || + (boot_cpu_data.x86_mask <0x1 )))return ; this_leaf->can_disable =1 ;
@@ -726,12 +729,12 @@ static ssize_t show_cache_disable(struct _cpuid4_info *this_leaf, char *buf,return -EINVAL ;pci_read_config_dword (dev,0x1BC + index *4 , ®); -return sprintf (buf,"%x\n" , reg); +return sprintf (buf,"0x%08x\n" , reg); }#define SHOW_CACHE_DISABLE (index) \ static ssize_t \ -show_cache_disable_##index(struct _cpuid4_info *this_leaf, char *buf) \ +show_cache_disable_##index(struct _cpuid4_info *this_leaf, char *buf) \ { \ return show_cache_disable(this_leaf, buf, index); \ }
@@ -745,7 +748,9 @@ static ssize_t store_cache_disable(struct _cpuid4_info *this_leaf,int node =cpu_to_node (cpu);struct pci_dev *dev =node_to_k8_nb_misc (node);unsigned long val =0 ; -unsigned int scrubber =0 ; + +#define SUBCACHE_MASK (3UL << 20) +#define SUBCACHE_INDEX 0xfffif (!this_leaf->can_disable)return -EINVAL ;
@@ -759,21 +764,24 @@ static ssize_t store_cache_disable(struct _cpuid4_info *this_leaf,if (strict_strtoul (buf,10 , &val) <0 )return -EINVAL ; - val |=0xc0000000 ; - -pci_read_config_dword (dev,0x58 , &scrubber); - scrubber &= ~0x1f000000 ; -pci_write_config_dword (dev,0x58 , scrubber); +/* do not allow writes outside of allowed bits */ +if (val & ~(SUBCACHE_MASK |SUBCACHE_INDEX )) +return -EINVAL ; -pci_write_config_dword (dev,0x1BC + index *4 , val & ~0x40000000 ); -wbinvd (); + val |=BIT (30 );pci_write_config_dword (dev,0x1BC + index *4 , val); +/* +* We need to WBINVD on a core on the node containing the L3 cache which +* indices we disable therefore a simple wbinvd() is not sufficient. +*/ +wbinvd_on_cpu (cpu); +pci_write_config_dword (dev,0x1BC + index *4 , val |BIT (31 ));return count; }#define STORE_CACHE_DISABLE (index) \ static ssize_t \ -store_cache_disable_##index(struct _cpuid4_info *this_leaf, \ +store_cache_disable_##index(struct _cpuid4_info *this_leaf, \ const char *buf, size_t count) \ { \ return store_cache_disable(this_leaf, buf, count, index); \
Signature · 1 line
1.6.6
[tip:x86/cpu] x86, cacheinfo: Fix disabling of L3 cache indices
Commit-ID: dcf39daf3d6d97f8741e82f0b9fb7554704ed2d1 Gitweb: http://git.kernel.org/tip/dcf39daf3d6d97f8741e82f0b9fb7554704ed2d1 Author: Borislav Petkov <[email protected]> AuthorDate: Fri, 22 Jan 2010 16:01:05 +0100 Committer: H. Peter Anvin <[email protected]> CommitDate: Fri, 22 Jan 2010 16:06:31 -0800
x86, cacheinfo: Fix disabling of L3 cache indices
* Correct the masks used for writing the cache index disable indices.
* Do not turn off L3 scrubber - it is not necessary.
* Make sure wbinvd is executed on the same node where the L3 is.
* Check for out-of-bounds values written to the registers.
* Make show_cache_disable hex values unambiguous
* Check for Erratum #388
LKML-Reference: <[email protected]>
--- arch/x86/kernel/cpu/intel_cacheinfo.c | 34 ++++++++++++++++++++------------ 1 files changed, 21 insertions(+), 13 deletions(-)
arch/x86/kernel/cpu/intel_cacheinfo.c +21 -13
@@ -18,6 +18,7 @@#include <asm/processor.h> #include <linux/smp.h> #include <asm/k8.h> +#include <asm/smp.h> #define LVL_1_INST 1#define LVL_1_DATA 2
@@ -299,8 +300,10 @@ amd_check_l3_disable(int index, struct _cpuid4_info_regs *this_leaf)if (boot_cpu_data.x86 ==0x11 )return ; -/* see erratum #382 */ -if ((boot_cpu_data.x86 ==0x10 ) && (boot_cpu_data.x86_model <0x8 )) +/* see errata #382 and #388 */ +if ((boot_cpu_data.x86 ==0x10 ) && + ((boot_cpu_data.x86_model <0x9 ) || + (boot_cpu_data.x86_mask <0x1 )))return ; this_leaf->can_disable =1 ;
@@ -726,12 +729,12 @@ static ssize_t show_cache_disable(struct _cpuid4_info *this_leaf, char *buf,return -EINVAL ;pci_read_config_dword (dev,0x1BC + index *4 , ®); -return sprintf (buf,"%x\n" , reg); +return sprintf (buf,"0x%08x\n" , reg); }#define SHOW_CACHE_DISABLE (index) \ static ssize_t \ -show_cache_disable_##index(struct _cpuid4_info *this_leaf, char *buf) \ +show_cache_disable_##index(struct _cpuid4_info *this_leaf, char *buf) \ { \ return show_cache_disable(this_leaf, buf, index); \ }
@@ -745,7 +748,9 @@ static ssize_t store_cache_disable(struct _cpuid4_info *this_leaf,int node =cpu_to_node (cpu);struct pci_dev *dev =node_to_k8_nb_misc (node);unsigned long val =0 ; -unsigned int scrubber =0 ; + +#define SUBCACHE_MASK (3UL << 20) +#define SUBCACHE_INDEX 0xfffif (!this_leaf->can_disable)return -EINVAL ;
@@ -759,21 +764,24 @@ static ssize_t store_cache_disable(struct _cpuid4_info *this_leaf,if (strict_strtoul (buf,10 , &val) <0 )return -EINVAL ; - val |=0xc0000000 ; - -pci_read_config_dword (dev,0x58 , &scrubber); - scrubber &= ~0x1f000000 ; -pci_write_config_dword (dev,0x58 , scrubber); +/* do not allow writes outside of allowed bits */ +if (val & ~(SUBCACHE_MASK |SUBCACHE_INDEX )) +return -EINVAL ; -pci_write_config_dword (dev,0x1BC + index *4 , val & ~0x40000000 ); -wbinvd (); + val |=BIT (30 );pci_write_config_dword (dev,0x1BC + index *4 , val); +/* +* We need to WBINVD on a core on the node containing the L3 cache which +* indices we disable therefore a simple wbinvd() is not sufficient. +*/ +wbinvd_on_cpu (cpu); +pci_write_config_dword (dev,0x1BC + index *4 , val |BIT (31 ));return count; }#define STORE_CACHE_DISABLE (index) \ static ssize_t \ -store_cache_disable_##index(struct _cpuid4_info *this_leaf, \ +store_cache_disable_##index(struct _cpuid4_info *this_leaf, \ const char *buf, size_t count) \ { \ return store_cache_disable(this_leaf, buf, count, index); \
PATCH 4/5 x86, cacheinfo: Add cache index disable sysfs attrs only to L3 caches
In reply to Borislav PetkovFrom: Borislav Petkov <[email protected]>
The cache_disable_[01] attribute in
/sys/devices/system/cpu/cpu?/cache/index[0-3]/
is enabled on all cache levels although only L3 supports it. Add it only to the cache level that actually supports it.
--- arch/x86/kernel/cpu/intel_cacheinfo.c | 35 ++++++++++++++++++++++++-------- 1 files changed, 26 insertions(+), 9 deletions(-)
arch/x86/kernel/cpu/intel_cacheinfo.c +26 -9
@@ -814,16 +814,24 @@ static struct _cache_attr cache_disable_0 = __ATTR(cache_disable_0, 0644,static struct _cache_attr cache_disable_1 =__ATTR (cache_disable_1,0644 , show_cache_disable_1, store_cache_disable_1); +#define DEFAULT_SYSFS_CACHE_ATTRS \ + &type.attr, \ + &level.attr, \ + &coherency_line_size.attr, \ + &physical_line_partition.attr, \ + &ways_of_associativity.attr, \ + &number_of_sets.attr, \ + &size.attr, \ + &shared_cpu_map.attr, \ + &shared_cpu_list.attr +static struct attribute *default_attrs[] = { - &type.attr, - &level.attr, - &coherency_line_size.attr, - &physical_line_partition.attr, - &ways_of_associativity.attr, - &number_of_sets.attr, - &size.attr, - &shared_cpu_map.attr, - &shared_cpu_list.attr, +DEFAULT_SYSFS_CACHE_ATTRS , +NULL +}; + +static struct attribute *default_l3_attrs[] = { +DEFAULT_SYSFS_CACHE_ATTRS , &cache_disable_0.attr, &cache_disable_1.attr,NULL
@@ -916,6 +924,7 @@ static int __cpuinit cache_add_dev(struct sys_device * sys_dev)unsigned int cpu = sys_dev->id;unsigned long i, j;struct _index_kobject *this_object; +struct _cpuid4_info *this_leaf;int retval; retval =cpuid4_cache_sysfs_init (cpu);
@@ -934,6 +943,14 @@ static int __cpuinit cache_add_dev(struct sys_device * sys_dev) this_object =INDEX_KOBJECT_PTR (cpu, i); this_object->cpu = cpu; this_object->index = i; + + this_leaf =CPUID4_INFO_IDX (cpu, i); + +if (this_leaf->can_disable) + ktype_cache.default_attrs = default_l3_attrs; +else + ktype_cache.default_attrs = default_attrs; + retval = kobject_init_and_add(&(this_object->kobj), &ktype_cache, per_cpu(ici_cache_kobject, cpu),
Signature · 1 line
1.6.6
[tip:x86/cpu] x86, cacheinfo: Add cache index disable sysfs attrs only to L3 caches
Commit-ID: 897de50e08937663912c86fb12ad7f708af2386c Gitweb: http://git.kernel.org/tip/897de50e08937663912c86fb12ad7f708af2386c Author: Borislav Petkov <[email protected]> AuthorDate: Fri, 22 Jan 2010 16:01:06 +0100 Committer: H. Peter Anvin <[email protected]> CommitDate: Fri, 22 Jan 2010 16:06:31 -0800
x86, cacheinfo: Add cache index disable sysfs attrs only to L3 caches
The cache_disable_[01] attribute in
/sys/devices/system/cpu/cpu?/cache/index[0-3]/
is enabled on all cache levels although only L3 supports it. Add it only to the cache level that actually supports it.
LKML-Reference: <[email protected]>
--- arch/x86/kernel/cpu/intel_cacheinfo.c | 35 ++++++++++++++++++++++++-------- 1 files changed, 26 insertions(+), 9 deletions(-)
arch/x86/kernel/cpu/intel_cacheinfo.c +26 -9
@@ -814,16 +814,24 @@ static struct _cache_attr cache_disable_0 = __ATTR(cache_disable_0, 0644,static struct _cache_attr cache_disable_1 =__ATTR (cache_disable_1,0644 , show_cache_disable_1, store_cache_disable_1); +#define DEFAULT_SYSFS_CACHE_ATTRS \ + &type.attr, \ + &level.attr, \ + &coherency_line_size.attr, \ + &physical_line_partition.attr, \ + &ways_of_associativity.attr, \ + &number_of_sets.attr, \ + &size.attr, \ + &shared_cpu_map.attr, \ + &shared_cpu_list.attr +static struct attribute *default_attrs[] = { - &type.attr, - &level.attr, - &coherency_line_size.attr, - &physical_line_partition.attr, - &ways_of_associativity.attr, - &number_of_sets.attr, - &size.attr, - &shared_cpu_map.attr, - &shared_cpu_list.attr, +DEFAULT_SYSFS_CACHE_ATTRS , +NULL +}; + +static struct attribute *default_l3_attrs[] = { +DEFAULT_SYSFS_CACHE_ATTRS , &cache_disable_0.attr, &cache_disable_1.attr,NULL
@@ -916,6 +924,7 @@ static int __cpuinit cache_add_dev(struct sys_device * sys_dev)unsigned int cpu = sys_dev->id;unsigned long i, j;struct _index_kobject *this_object; +struct _cpuid4_info *this_leaf;int retval; retval =cpuid4_cache_sysfs_init (cpu);
@@ -934,6 +943,14 @@ static int __cpuinit cache_add_dev(struct sys_device * sys_dev) this_object =INDEX_KOBJECT_PTR (cpu, i); this_object->cpu = cpu; this_object->index = i; + + this_leaf =CPUID4_INFO_IDX (cpu, i); + +if (this_leaf->can_disable) + ktype_cache.default_attrs = default_l3_attrs; +else + ktype_cache.default_attrs = default_attrs; + retval = kobject_init_and_add(&(this_object->kobj), &ktype_cache, per_cpu(ici_cache_kobject, cpu),
PATCH 5/5 x86, cacheinfo: Calculate L3 indices
In reply to Borislav PetkovFrom: Borislav Petkov <[email protected]>
We need to know the valid L3 indices interval when disabling them over /sysfs. Do that when the core is brought online and add boundary checks to the sysfs .store attribute.
--- arch/x86/kernel/cpu/intel_cacheinfo.c | 35 +++++++++++++++++++++++++++++--- 1 files changed, 31 insertions(+), 4 deletions(-)
arch/x86/kernel/cpu/intel_cacheinfo.c +31 -4
@@ -151,7 +151,8 @@ struct _cpuid4_info {union _cpuid4_leaf_ebx ebx;union _cpuid4_leaf_ecx ecx;unsigned long size; -unsigned long can_disable; +bool can_disable; +unsigned int l3_indices;DECLARE_BITMAP (shared_cpu_map,NR_CPUS ); };
@@ -161,7 +162,8 @@ struct _cpuid4_info_regs {union _cpuid4_leaf_ebx ebx;union _cpuid4_leaf_ecx ecx;unsigned long size; -unsigned long can_disable; +bool can_disable; +unsigned int l3_indices; };unsigned short num_cache_leaves;
@@ -291,6 +293,29 @@ amd_cpuid4(int leaf, union _cpuid4_leaf_eax *eax, (ebx->split.ways_of_associativity +1 ) -1 ; } +static unsigned int __cpuinit amd_calc_l3_indices (void ) +{ +/* +* We're called over smp_call_function_single() and therefore +* are on the correct cpu. +*/ +int cpu =smp_processor_id (); +int node =cpu_to_node (cpu); +struct pci_dev *dev =node_to_k8_nb_misc (node); +unsigned int sc0, sc1, sc2, sc3; +u32 val; + +pci_read_config_dword (dev,0x1C4 , &val); + +/* calculate subcache sizes */ + sc0 = !(val &BIT (0 )); + sc1 = !(val &BIT (4 )); + sc2 = !(val &BIT (8 )) + !(val &BIT (9 )); + sc3 = !(val &BIT (12 )) + !(val &BIT (13 )); + +return (max (max (max (sc0, sc1), sc2), sc3) <<10 ) -1 ; +} +static void __cpuinit amd_check_l3_disable (int index,struct _cpuid4_info_regs *this_leaf) {
@@ -306,7 +331,8 @@ amd_check_l3_disable(int index, struct _cpuid4_info_regs *this_leaf) (boot_cpu_data.x86_mask < 0x1)))return ; - this_leaf->can_disable =1 ; + this_leaf->can_disable = true; + this_leaf->l3_indices =amd_calc_l3_indices (); } static int
@@ -765,7 +791,8 @@ static ssize_t store_cache_disable(struct _cpuid4_info *this_leaf,return -EINVAL ;/* do not allow writes outside of allowed bits */ -if (val & ~(SUBCACHE_MASK |SUBCACHE_INDEX )) +if ((val & ~(SUBCACHE_MASK |SUBCACHE_INDEX )) || + ((val &SUBCACHE_INDEX ) > this_leaf->l3_indices))return -EINVAL ; val |=BIT (30 );
Signature · 1 line
1.6.6
[tip:x86/cpu] x86, cacheinfo: Calculate L3 indices
Commit-ID: 048a8774ca43488d78605031f11cc206d7a2682a Gitweb: http://git.kernel.org/tip/048a8774ca43488d78605031f11cc206d7a2682a Author: Borislav Petkov <[email protected]> AuthorDate: Fri, 22 Jan 2010 16:01:07 +0100 Committer: H. Peter Anvin <[email protected]> CommitDate: Fri, 22 Jan 2010 16:06:31 -0800
x86, cacheinfo: Calculate L3 indices
We need to know the valid L3 indices interval when disabling them over /sysfs. Do that when the core is brought online and add boundary checks to the sysfs .store attribute.
LKML-Reference: <[email protected]>
--- arch/x86/kernel/cpu/intel_cacheinfo.c | 35 +++++++++++++++++++++++++++++--- 1 files changed, 31 insertions(+), 4 deletions(-)
arch/x86/kernel/cpu/intel_cacheinfo.c +31 -4
@@ -151,7 +151,8 @@ struct _cpuid4_info {union _cpuid4_leaf_ebx ebx;union _cpuid4_leaf_ecx ecx;unsigned long size; -unsigned long can_disable; +bool can_disable; +unsigned int l3_indices;DECLARE_BITMAP (shared_cpu_map,NR_CPUS ); };
@@ -161,7 +162,8 @@ struct _cpuid4_info_regs {union _cpuid4_leaf_ebx ebx;union _cpuid4_leaf_ecx ecx;unsigned long size; -unsigned long can_disable; +bool can_disable; +unsigned int l3_indices; };unsigned short num_cache_leaves;
@@ -291,6 +293,29 @@ amd_cpuid4(int leaf, union _cpuid4_leaf_eax *eax, (ebx->split.ways_of_associativity +1 ) -1 ; } +static unsigned int __cpuinit amd_calc_l3_indices (void ) +{ +/* +* We're called over smp_call_function_single() and therefore +* are on the correct cpu. +*/ +int cpu =smp_processor_id (); +int node =cpu_to_node (cpu); +struct pci_dev *dev =node_to_k8_nb_misc (node); +unsigned int sc0, sc1, sc2, sc3; +u32 val; + +pci_read_config_dword (dev,0x1C4 , &val); + +/* calculate subcache sizes */ + sc0 = !(val &BIT (0 )); + sc1 = !(val &BIT (4 )); + sc2 = !(val &BIT (8 )) + !(val &BIT (9 )); + sc3 = !(val &BIT (12 )) + !(val &BIT (13 )); + +return (max (max (max (sc0, sc1), sc2), sc3) <<10 ) -1 ; +} +static void __cpuinit amd_check_l3_disable (int index,struct _cpuid4_info_regs *this_leaf) {
@@ -306,7 +331,8 @@ amd_check_l3_disable(int index, struct _cpuid4_info_regs *this_leaf) (boot_cpu_data.x86_mask < 0x1)))return ; - this_leaf->can_disable =1 ; + this_leaf->can_disable = true; + this_leaf->l3_indices =amd_calc_l3_indices (); } static int
@@ -765,7 +791,8 @@ static ssize_t store_cache_disable(struct _cpuid4_info *this_leaf,return -EINVAL ;/* do not allow writes outside of allowed bits */ -if (val & ~(SUBCACHE_MASK |SUBCACHE_INDEX )) +if ((val & ~(SUBCACHE_MASK |SUBCACHE_INDEX )) || + ((val &SUBCACHE_INDEX ) > this_leaf->l3_indices))return -EINVAL ; val |=BIT (30 );
H. Peter Anvin
reviewerQuoted 9 lines · Borislav Petkov
these are a bunch of fixes which correct the disabling of L3 cache indexes. This is needed, for example, in the case when excessive correctable L3 error rates are observed. In a later step, this functionality will be connected to the mcheck mechanism and used to evaluate occurring L3 tag and data array errors and then disable the respective failing L3 cache regions in order to defer system failure before a sysop can intervene.
Those patches are also good -stable candidates.
Hmmm... I'm not sure I see a strong justification for a late -rc push into Linus/stable push for for these... I think you would have to explicitly make the case if you want them to be considered as such.
-hpa
Signature · 2 lines
H. Peter Anvin, Intel Open Source Technology Center I work for Intel. I don't speak on their behalf.
Quoted 9 lines · H. Peter Anvin
Quoted 8 lines · Borislav Petkov
these are a bunch of fixes which correct the disabling of L3 cache indexes. This is needed, for example, in the case when excessive correctable L3 error rates are observed. In a later step, this functionality will be connected to the mcheck mechanism and used to evaluate occurring L3 tag and data array errors and then disable the respective failing L3 cache regions in order to defer system failure before a sysop can intervene.
Quoted 5 lines
Quoted 1 line
Those patches are also good -stable candidates.
Hmmm... I'm not sure I see a strong justification for a late -rc push into Linus/stable push for for these... I think you would have to explicitly make the case if you want them to be considered as such.
Well, on the one hand, they fix real bugs in the L3 cache index disable code and since they're bugfixes, they are eligible late -rc candidates.
On the other hand, however and more importantly, the machines which have that feature are not selling yet so postponing the patches for the next merge window is still ok. I'll backport them then to .32 for the distro kernels and .33 and I think we are going to be fine this way.
So queueing them for .34 is still fine with me, thanks.
Signature · 6 lines
Regards/Gruss, ボリス. -- Advanced Micro Devices, Inc. Operating Systems Research Center
H. Peter Anvin
reviewerQuoted 10 lines · Borislav Petkov
Quoted 6 lines
Quoted 2 lines
Those patches are also good -stable candidates.
Hmmm... I'm not sure I see a strong justification for a late -rc push into Linus/stable push for for these... I think you would have to explicitly make the case if you want them to be considered as such.
Well, on the one hand, they fix real bugs in the L3 cache index disable code and since they're bugfixes, they are eligible late -rc candidates.
Bugfixes are *early* -rc candidates. Regression fixes are *late* -rc candidates, at least that seems to be the policy Linus currently implements. -stable seems to use slightly less strict criteria (the whole point is that -final needs to be a stabilization point, backported fixes/drivers can then come onto a stable base) which is why you seem some patches which are "straight to .1".
Quoted 6 lines
On the other hand, however and more importantly, the machines which have that feature are not selling yet so postponing the patches for the next merge window is still ok. I'll backport them then to .32 for the distro kernels and .33 and I think we are going to be fine this way.
So queueing them for .34 is still fine with me, thanks.
OK. You can check with -stable if they want to take the backport post-.33.
-hpa
Signature · 2 lines
H. Peter Anvin, Intel Open Source Technology Center I work for Intel. I don't speak on their behalf.
Quoted 15 lines · H. Peter Anvin
Quoted 9 lines · Borislav Petkov
Quoted 6 lines
Quoted 2 lines
Those patches are also good -stable candidates.
Hmmm... I'm not sure I see a strong justification for a late -rc push into Linus/stable push for for these... I think you would have to explicitly make the case if you want them to be considered as such.
Well, on the one hand, they fix real bugs in the L3 cache index disable code and since they're bugfixes, they are eligible late -rc candidates.
Bugfixes are *early* -rc candidates. Regression fixes are *late* -rc candidates, at least that seems to be the policy Linus currently implements.
(quoted diff) +0
-2-stable seems to use slightly less strict criteria (the whole point is that -final needs to be a stabilization point, backported fixes/drivers can then
Quoted 2 lines
come onto a stable base) which is why you seem some patches which are "straight to .1".
Yes.
Ingo
Quoted 14 lines · Ingo Molnar
Quoted 11 lines · H. Peter Anvin
Quoted 9 lines · Borislav Petkov
Quoted 6 lines
Quoted 2 lines
Those patches are also good -stable candidates.
Hmmm... I'm not sure I see a strong justification for a late -rc push into Linus/stable push for for these... I think you would have to explicitly make the case if you want them to be considered as such.
Well, on the one hand, they fix real bugs in the L3 cache index disable code and since they're bugfixes, they are eligible late -rc candidates.
Quoted 8 lines
Quoted 6 lines
Bugfixes are *early* -rc candidates. Regression fixes are *late* -rc candidates, at least that seems to be the policy Linus currently implements.
(quoted diff) +0
-2-stable seems to use slightly less strict criteria (the whole point is that -final needs to be a stabilization point, backported fixes/drivers can thencome onto a stable base) which is why you seem some patches which are "straight to .1".
Yes.
Ok, thanks for the clarification - my only trouble was that the current code is b0rked as is and those fixes are needed. However, backporting them at a later point seems much more riskfree and I will do so later.
Thanks.
Signature · 2 lines
Regards/Gruss,
Boris.Quoted 24 lines · Borislav Petkov
Quoted 14 lines · Ingo Molnar
Quoted 11 lines · H. Peter Anvin
Quoted 9 lines · Borislav Petkov
Quoted 6 lines
Quoted 2 lines
Those patches are also good -stable candidates.
Hmmm... I'm not sure I see a strong justification for a late -rc push into Linus/stable push for for these... I think you would have to explicitly make the case if you want them to be considered as such.
Well, on the one hand, they fix real bugs in the L3 cache index disable code and since they're bugfixes, they are eligible late -rc candidates.
Quoted 8 lines
Quoted 6 lines
Bugfixes are *early* -rc candidates. Regression fixes are *late* -rc candidates, at least that seems to be the policy Linus currently implements.
(quoted diff) +0
-2-stable seems to use slightly less strict criteria (the whole point is that -final needs to be a stabilization point, backported fixes/drivers can thencome onto a stable base) which is why you seem some patches which are "straight to .1".
Yes.
Quoted 5 lines
Ok, thanks for the clarification - my only trouble was that the current code is b0rked as is and those fixes are needed. However, backporting them at a later point seems much more riskfree and I will do so later.
Thanks.
Well, if there's a crasher in there, then a minimal fix to address just that is preferred for .33 - and that can be tagged for -stable immediately.
Anything more complex (these handful of patches) should go via the usual route of .34-rc1 and then -stable if it's problem-free.
Ingo
Quoted 27 lines · Ingo Molnar
Quoted 24 lines · Borislav Petkov
Quoted 22 lines · Ingo Molnar
Quoted 15 lines · H. Peter Anvin
Quoted 9 lines · Borislav Petkov
Quoted 6 lines
Quoted 2 lines
Those patches are also good -stable candidates.
Hmmm... I'm not sure I see a strong justification for a late -rc push into Linus/stable push for for these... I think you would have to explicitly make the case if you want them to be considered as such.
Well, on the one hand, they fix real bugs in the L3 cache index disable code and since they're bugfixes, they are eligible late -rc candidates.
Bugfixes are *early* -rc candidates. Regression fixes are *late* -rc candidates, at least that seems to be the policy Linus currently implements.
(quoted diff) +0
-2-stable seems to use slightly less strict criteria (the whole point is that -final needs to be a stabilization point, backported fixes/drivers can thenQuoted 2 lines
come onto a stable base) which is why you seem some patches which are "straight to .1".
Yes.
Quoted 8 lines
Quoted 5 lines
Ok, thanks for the clarification - my only trouble was that the current code is b0rked as is and those fixes are needed. However, backporting them at a later point seems much more riskfree and I will do so later.
Thanks.
Well, if there's a crasher in there, then a minimal fix to address just that is preferred for .33 - and that can be tagged for -stable immediately.
As I said earlier, I don't believe we have machines in the wild to practically support the feature yet so we should be OK without a fix. Let me verify that next week and get back to you with a minimal fix in case we have any affected configurations.
Thanks.
Signature · 2 lines
Regards/Gruss,
Boris.Hi Ingo,
Quoted 7 lines · Borislav Petkov
Quoted 2 lines
Well, if there's a crasher in there, then a minimal fix to address just that is preferred for .33 - and that can be tagged for -stable immediately.
As I said earlier, I don't believe we have machines in the wild to practically support the feature yet so we should be OK without a fix. Let me verify that next week and get back to you with a minimal fix in case we have any affected configurations.
I could verify that we don't need a minimal fix for .33 for the reason above so the 5 fixes for .34 merge window should suffice completely.
Thanks.
Signature · 6 lines
Regards/Gruss, Boris. -- Advanced Micro Devices, Inc. Operating Systems Research Center
Quoted 6 lines · Borislav Petkov
Hi Ingo,
Quoted 3 lines · Borislav Petkov
Quoted 2 lines
Well, if there's a crasher in there, then a minimal fix to address just that is preferred for .33 - and that can be tagged for -stable immediately.
Quoted 7 lines
Quoted 4 lines
As I said earlier, I don't believe we have machines in the wild to practically support the feature yet so we should be OK without a fix. Let me verify that next week and get back to you with a minimal fix in case we have any affected configurations.
I could verify that we don't need a minimal fix for .33 for the reason above so the 5 fixes for .34 merge window should suffice completely.
Great - thanks for confirming that.
Ingo