Updates to EDAC mce_amd_inj
Subsystems (3)
cover Updates to EDAC mce_amd_inj
This is basically a V3 of the final 3 patches of an earlier patchset to update injection interfaces in mce_amd_inj.
Here's a link to V2 of the original series:
http://marc.info/?l=linux-edac&m=143327677901235&w=2
Patches 1-6 of above are currently in 'for-next' of bp.git
Spinning this as a separate series as they needed to be rebased on top of CONFIG_X86_HT removal patch which made it into tip.
Patch 1: Store number of nodes as a static global in amd.c
No functional change.
Patch 2: Provide accessor function to obtain the number of nodes per processor Patch 3: Modify injection mechanism for bank 4 errors. Since they are
typically logged or reported only on NBC, we make sure that we inject on the correct core here.
Since the earlier patches are split across tip and bp.git, this series is based on top of tip with EDAC patches merged from 'for-next' of bp.git
Changes wrt V2 of the original series:
- Rebase on top of CONFIG_X86_HT removal patch
- Remove unnecessary amd_set_num_nodes() function (from patch 7 of original series) and simplify code. Current changes are in Patch 1
- And reword the commit message.
- checkpatch with --strict threw couple of checks on Patch 3, fixed them
- Update copyright info in mce_amd_inj.c
Aravind Gopalakrishnan (3):
x86, amd: Store number of nodes in a static global variable x86, amd: Provide accessor for number of nodes edac, mce_amd_inj: Inject errors on NBC for bank 4 errors
arch/x86/include/asm/processor.h | 1 + arch/x86/kernel/cpu/amd.c | 23 ++++++++++++---- drivers/edac/mce_amd_inj.c | 57 +++++++++++++++++++++++++++++++++++++++- 3 files changed, 75 insertions(+), 6 deletions(-)
Signature · 1 line
2.4.0
PATCH 1/3 x86, amd: Store number of nodes in a static global variable
Moving this out of function local scope as we want to allow EDAC to be able to view this topology attribute.
A follow-up patch introduces an accessor function for this variable that will be used by EDAC's mce_amd_inj module.
--- arch/x86/kernel/cpu/amd.c | 17 ++++++++++++----- 1 file changed, 12 insertions(+), 5 deletions(-)
arch/x86/kernel/cpu/amd.c +12 -5
@@ -19,6 +19,13 @@#include "cpu.h" +/* +* nodes_per_processor: Specifies number of nodes per socket +* Refer Fam15h Models 00-0fh BKDG- +* CPUID Fn8000_001E_ECX Node Identifiers[10:8] +*/ +static u32 nodes_per_processor =1 ; +static inline int rdmsrl_amd_safe (unsigned msr,unsigned long long *p) {u32 gprs[8 ] = {0 };
@@ -291,7 +298,7 @@ static int nearby_node(int apicid)#ifdef CONFIG_SMP static void amd_get_topology (struct cpuinfo_x86 *c) { -u32 nodes, cores_per_cu =1 ; +u32 cores_per_cu =1 ;u8 node_id;int cpu =smp_processor_id ();
@@ -300,7 +307,7 @@ static void amd_get_topology(struct cpuinfo_x86 *c)u32 eax, ebx, ecx, edx;cpuid (0x8000001e , &eax, &ebx, &ecx, &edx); - nodes = ((ecx >>8 ) &7 ) +1 ; + nodes_per_processor = ((ecx >>8 ) &7 ) +1 ; node_id = ecx &7 ;/* get compute unit information */
@@ -311,18 +318,18 @@ static void amd_get_topology(struct cpuinfo_x86 *c)u64 value;rdmsrl (MSR_FAM10H_NODE_ID , value); - nodes = ((value >>3 ) &7 ) +1 ; + nodes_per_processor = ((value >>3 ) &7 ) +1 ; node_id = value &7 ; }else return ;/* fixup multi-node processor information */ -if (nodes >1 ) { +if (nodes_per_processor >1 ) {u32 cores_per_node;u32 cus_per_node;set_cpu_cap (c,X86_FEATURE_AMD_DCM ); - cores_per_node = c->x86_max_cores / nodes; + cores_per_node = c->x86_max_cores / nodes_per_processor; cus_per_node = cores_per_node / cores_per_cu;/* store NodeID, use llc_shared_map to store sibling info */
Signature · 1 line
2.4.0
Borislav Petkov
maintainerRe: [PATCH 1/3] x86, amd: Store number of nodes in a static global variable
Quoted 15 lines · Aravind Gopalakrishnan
Moving this out of function local scope as we want to allow EDAC to be able to view this topology attribute.
A follow-up patch introduces an accessor function for this variable that will be used by EDAC's
mce_amd_injmodule.Cc Thomas Gleixner <[email protected]>Cc Ingo Molnar <[email protected]>Cc "H. Peter Anvin" <[email protected]>Cc Borislav Petkov <[email protected]>Cc Jacob Shin <[email protected]>Cc Dave Hansen <[email protected]>Cc Andy Lutomirski <[email protected]>Cc Paolo Bonzini <[email protected]>Signed-off-by Aravind Gopalakrishnan <[email protected]>
Quoted 3 lines
--- arch/x86/kernel/cpu/amd.c | 17 ++++++++++++----- 1 file changed, 12 insertions(+), 5 deletions(-)
I did some cleanup and merged it with the next patch. It seems simple enough and splitting it was not such a good idea, after all. And called it nodes_per_socket to avoid confusion.
---
From: Aravind Gopalakrishnan <[email protected]>
Date: Tue, 9 Jun 2015 11:45:15 -0500
Subject: [PATCH] x86/cpu/amd: Give access to the number of nodes in a physical
package
Stash the number of nodes in a physical processor package locally and add an accessor to be called by interested parties. The first user is the MCE injection module which uses it to find the node base core in a package for injecting a certain type of errors.
--- arch/x86/include/asm/processor.h | 1 + arch/x86/kernel/cpu/amd.c | 23 ++++++++++++++++++----- 2 files changed, 19 insertions(+), 5 deletions(-)
arch/x86/include/asm/processor.h +1 -0
@@ -946,6 +946,7 @@ static inline int mpx_disable_management(struct task_struct *tsk)#endif /* CONFIG_X86_INTEL_MPX */ extern u16 amd_get_nb_id (int cpu); +extern u32 amd_get_nodes_per_socket (void );static inline uint32_t hypervisor_cpuid_base (const char *sig,uint32_t leaves) {
arch/x86/kernel/cpu/amd.c +18 -5
@@ -19,6 +19,13 @@#include "cpu.h" +/* +* nodes_per_socket: Stores the number of nodes per socket. +* Refer to Fam15h Models 00-0fh BKDG - CPUID Fn8000_001E_ECX +* Node Identifiers[10:8] +*/ +static u32 nodes_per_socket =1 ; +static inline int rdmsrl_amd_safe (unsigned msr,unsigned long long *p) {u32 gprs[8 ] = {0 };
@@ -291,7 +298,7 @@ static int nearby_node(int apicid)#ifdef CONFIG_X86_HT static void amd_get_topology (struct cpuinfo_x86 *c) { -u32 nodes, cores_per_cu =1 ; +u32 cores_per_cu =1 ;u8 node_id;int cpu =smp_processor_id ();
@@ -300,7 +307,7 @@ static void amd_get_topology(struct cpuinfo_x86 *c)u32 eax, ebx, ecx, edx;cpuid (0x8000001e , &eax, &ebx, &ecx, &edx); - nodes = ((ecx >>8 ) &7 ) +1 ; + nodes_per_socket = ((ecx >>8 ) &7 ) +1 ; node_id = ecx &7 ;/* get compute unit information */
@@ -311,18 +318,18 @@ static void amd_get_topology(struct cpuinfo_x86 *c)u64 value;rdmsrl (MSR_FAM10H_NODE_ID , value); - nodes = ((value >>3 ) &7 ) +1 ; + nodes_per_socket = ((value >>3 ) &7 ) +1 ; node_id = value &7 ; }else return ;/* fixup multi-node processor information */ -if (nodes >1 ) { +if (nodes_per_socket >1 ) {u32 cores_per_node;u32 cus_per_node;set_cpu_cap (c,X86_FEATURE_AMD_DCM ); - cores_per_node = c->x86_max_cores / nodes; + cores_per_node = c->x86_max_cores / nodes_per_socket; cus_per_node = cores_per_node / cores_per_cu;/* store NodeID, use llc_shared_map to store sibling info */
@@ -366,6 +373,12 @@ u16 amd_get_nb_id(int cpu) }EXPORT_SYMBOL_GPL (amd_get_nb_id); +u32 amd_get_nodes_per_socket (void ) +{ +return nodes_per_socket; +} +EXPORT_SYMBOL_GPL (amd_get_nodes_per_socket); +static void srat_detect_node (struct cpuinfo_x86 *c) {#ifdef CONFIG_NUMA
Signature · 8 lines
2.3.5
--
Regards/Gruss,
Boris.
ECO tip #101: Trim your mails when you reply.
--tip-bot for Aravind Gopalakrishnan
merge bot[tip:x86/cpu] x86/cpu/amd: Give access to the number of nodes in a physical package
In reply to Aravind Gopalakrishnanx86/cpu/amd: Give access to the number of nodes in a physical package
Commit-ID: cc2749e4095cbbcb35518fb2db5e926b85c3f25f Gitweb: http://git.kernel.org/tip/cc2749e4095cbbcb35518fb2db5e926b85c3f25f Author: Aravind Gopalakrishnan <[email protected]> AuthorDate: Mon, 15 Jun 2015 10:28:15 +0200 Committer: Ingo Molnar <[email protected]> CommitDate: Thu, 18 Jun 2015 11:16:06 +0200
x86/cpu/amd: Give access to the number of nodes in a physical package
Stash the number of nodes in a physical processor package locally and add an accessor to be called by interested parties. The first user is the MCE injection module which uses it to find the node base core in a package for injecting a certain type of errors.
--- arch/x86/include/asm/processor.h | 1 + arch/x86/kernel/cpu/amd.c | 23 ++++++++++++++++++----- 2 files changed, 19 insertions(+), 5 deletions(-)
arch/x86/include/asm/processor.h +1 -0
@@ -946,6 +946,7 @@ static inline int mpx_disable_management(struct task_struct *tsk)#endif /* CONFIG_X86_INTEL_MPX */ extern u16 amd_get_nb_id (int cpu); +extern u32 amd_get_nodes_per_socket (void );static inline uint32_t hypervisor_cpuid_base (const char *sig,uint32_t leaves) {
arch/x86/kernel/cpu/amd.c +18 -5
@@ -19,6 +19,13 @@#include "cpu.h" +/* +* nodes_per_socket: Stores the number of nodes per socket. +* Refer to Fam15h Models 00-0fh BKDG - CPUID Fn8000_001E_ECX +* Node Identifiers[10:8] +*/ +static u32 nodes_per_socket =1 ; +static inline int rdmsrl_amd_safe (unsigned msr,unsigned long long *p) {u32 gprs[8 ] = {0 };
@@ -291,7 +298,7 @@ static int nearby_node(int apicid)#ifdef CONFIG_X86_HT static void amd_get_topology (struct cpuinfo_x86 *c) { -u32 nodes, cores_per_cu =1 ; +u32 cores_per_cu =1 ;u8 node_id;int cpu =smp_processor_id ();
@@ -300,7 +307,7 @@ static void amd_get_topology(struct cpuinfo_x86 *c)u32 eax, ebx, ecx, edx;cpuid (0x8000001e , &eax, &ebx, &ecx, &edx); - nodes = ((ecx >>8 ) &7 ) +1 ; + nodes_per_socket = ((ecx >>8 ) &7 ) +1 ; node_id = ecx &7 ;/* get compute unit information */
@@ -311,18 +318,18 @@ static void amd_get_topology(struct cpuinfo_x86 *c)u64 value;rdmsrl (MSR_FAM10H_NODE_ID , value); - nodes = ((value >>3 ) &7 ) +1 ; + nodes_per_socket = ((value >>3 ) &7 ) +1 ; node_id = value &7 ; }else return ;/* fixup multi-node processor information */ -if (nodes >1 ) { +if (nodes_per_socket >1 ) {u32 cores_per_node;u32 cus_per_node;set_cpu_cap (c,X86_FEATURE_AMD_DCM ); - cores_per_node = c->x86_max_cores / nodes; + cores_per_node = c->x86_max_cores / nodes_per_socket; cus_per_node = cores_per_node / cores_per_cu;/* store NodeID, use llc_shared_map to store sibling info */
@@ -366,6 +373,12 @@ u16 amd_get_nb_id(int cpu) }EXPORT_SYMBOL_GPL (amd_get_nb_id); +u32 amd_get_nodes_per_socket (void ) +{ +return nodes_per_socket; +} +EXPORT_SYMBOL_GPL (amd_get_nodes_per_socket); +static void srat_detect_node (struct cpuinfo_x86 *c) {#ifdef CONFIG_NUMA
PATCH 2/3 x86, amd: Provide accessor for number of nodes
In reply to Aravind GopalakrishnanAdd an accessor function amd_get_nodes_cnt() which returns the number of nodes per socket.
In a subsequent patch, we will use this info in EDAC mce_amd_inj module.
--- arch/x86/include/asm/processor.h | 1 + arch/x86/kernel/cpu/amd.c | 6 ++++++ 2 files changed, 7 insertions(+)
arch/x86/include/asm/processor.h +1 -0
@@ -820,6 +820,7 @@ static inline int mpx_disable_management(struct task_struct *tsk)#endif /* CONFIG_X86_INTEL_MPX */ extern u16 amd_get_nb_id (int cpu); +extern u32 amd_get_nodes_cnt (void );static inline uint32_t hypervisor_cpuid_base (const char *sig,uint32_t leaves) {
arch/x86/kernel/cpu/amd.c +6 -0
@@ -373,6 +373,12 @@ u16 amd_get_nb_id(int cpu) }EXPORT_SYMBOL_GPL (amd_get_nb_id); +u32 amd_get_nodes_cnt (void ) +{ +return nodes_per_processor; +} +EXPORT_SYMBOL_GPL (amd_get_nodes_cnt); +static void srat_detect_node (struct cpuinfo_x86 *c) {#ifdef CONFIG_NUMA
Signature · 1 line
2.4.0
PATCH 3/3 edac, mce_amd_inj: Inject errors on NBC for bank 4 errors
In reply to Aravind GopalakrishnanFor bank 4 errors, MCE is logged and reported only on node base cores. Refer D18F3x44[NbMcaToMstCpuEn] field in Fam10h and later BKDGs.
This patch ensures that we inject the error on the node base core for bank 4 errors. Otherwise, triggering #MC or apic interrupts on a non node base core would not have any effect on the system. (i.e), we would not see any relevant output on kernel logs for the error we just injected.
Update copyrights info while at it.
--- drivers/edac/mce_amd_inj.c | 57 +++++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 56 insertions(+), 1 deletion(-)
drivers/edac/mce_amd_inj.c +56 -1
@@ -6,7 +6,7 @@* This file may be distributed under the terms of the GNU General Public * License version 2. * -* Copyright (c) 2010-14: Borislav Petkov <[email protected]> +* Copyright (c) 2010-15: Borislav Petkov <[email protected]> * Advanced Micro Devices Inc. */
@@ -17,10 +17,13 @@#include <linux/cpu.h> #include <linux/string.h> #include <linux/uaccess.h> +#include <linux/pci.h> #include <asm/mce.h> #include <asm/irq_vectors.h> +#include <asm/amd_nb.h> #include "mce_amd.h" +#include "amd64_edac.h" /* * Collect all the MCi_XXX settings
@@ -200,6 +203,44 @@ static void trigger_thr_int(void *info) asmvolatile ("int %0" ::"i" (THRESHOLD_APIC_VECTOR )); } +static u32 amd_get_nbc_for_node (int node_id) +{ +struct cpuinfo_x86 *c = &boot_cpu_data; +u32 cores_per_node; + + cores_per_node = c->x86_max_cores /amd_get_nodes_cnt (); + +return cores_per_node * node_id; +} + +static void toggle_nb_mca_mst_cpu (u16 nid) +{ +struct pci_dev *F3 =node_to_amd_nb (nid)->misc; +u32 val; +int err; + +if (!F3 ) +return ; + + err =pci_read_config_dword (F3 ,NBCFG , &val); +if (err) { +pr_err ("%s: Error reading F%dx%03x.\n" , __func__, +PCI_FUNC (F3 ->devfn), +NBCFG ); +return ; + } + +if (!(val &BIT (27 ))) { +pr_err ("%s: BIOS not setting D18F3x44[NbMcaToMstCpuEn]. Doing that here\n" , __func__); + val |=BIT (27 ); + err =pci_write_config_dword (F3 ,NBCFG , val); +if (err) +pr_err ("%s: Error writing F%dx%03x.\n" , __func__, +PCI_FUNC (F3 ->devfn), +NBCFG ); + } +} +static void do_inject (void ) {u64 mcg_status =0 ;
@@ -235,6 +276,20 @@ static void do_inject(void)if (!(i_mce.status &MCI_STATUS_PCC )) mcg_status |=MCG_STATUS_RIPV ; +/* +* For multi node cpus, logging and reporting of bank == 4 errors +* happen only on the node base core. Refer D18F3x44[NbMcaToMstCpuEn] +* for Fam10h and later BKDGs +*/ +if (static_cpu_has (X86_FEATURE_AMD_DCM ) && b ==4 ) { +/* +* BIOS sets D18F3x44[NbMcaToMstCpuEn] by default. +* But make sure of it here just in case.. +*/ +toggle_nb_mca_mst_cpu (amd_get_nb_id (cpu)); + cpu =amd_get_nbc_for_node (amd_get_nb_id (cpu)); + } +toggle_hw_mce_inject (cpu, true); wrmsr_on_cpu(cpu, MSR_IA32_MCG_STATUS,
Signature · 1 line
2.4.0
Re: [PATCH 3/3] edac, mce_amd_inj: Inject errors on NBC for bank 4 errors
Quoted 10 lines · Aravind Gopalakrishnan
For bank 4 errors, MCE is logged and reported only on node base cores. Refer D18F3x44[NbMcaToMstCpuEn] field in Fam10h and later BKDGs.
This patch ensures that we inject the error on the node base core for bank 4 errors. Otherwise, triggering #MC or apic interrupts on a non node base core would not have any effect on the system. (i.e), we would not see any relevant output on kernel logs for the error we just injected.
Quoted 6 lines
Update copyrights info while at it.
Signed-off-by Aravind Gopalakrishnan <[email protected]>--- drivers/edac/mce_amd_inj.c | 57 +++++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 56 insertions(+), 1 deletion(-)
Applied, did some cleanups too, see below.
It builds, now to do some testing :-)
---
From: Aravind Gopalakrishnan <[email protected]>
Date: Tue, 9 Jun 2015 11:45:17 -0500
Subject: [PATCH] EDAC, mce_amd_inj: Inject errors on NBC for bank 4 errors
Bank 4 MCEs are logged and reported only on the node base core (NBC) in a socket. Refer to the D18F3x44[NbMcaToMstCpuEn] field in Fam10h and later BKDGs. The node base core (NBC) is the lowest numbered core in the node.
This patch ensures that we inject the error on the NBC for bank 4 errors. Otherwise, triggering #MC or APIC interrupts on a core which is not the NBC would not have any effect on the system, i.e we would not see any relevant output on kernel logs for the error we just injected.
Update copyrights info while at it.
[ Massage commit message, save an indentation level in toggle_nb_mca_mst_cpu,
reflow comments. ]
--- drivers/edac/mce_amd_inj.c | 59 ++++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 57 insertions(+), 2 deletions(-)
drivers/edac/mce_amd_inj.c +57 -2
@@ -6,7 +6,7 @@* This file may be distributed under the terms of the GNU General Public * License version 2. * -* Copyright (c) 2010-14: Borislav Petkov <[email protected]> +* Copyright (c) 2010-15: Borislav Petkov <[email protected]> * Advanced Micro Devices Inc. */
@@ -17,10 +17,13 @@#include <linux/cpu.h> #include <linux/string.h> #include <linux/uaccess.h> +#include <linux/pci.h> #include <asm/mce.h> #include <asm/irq_vectors.h> +#include <asm/amd_nb.h> #include "mce_amd.h" +#include "amd64_edac.h" /* * Collect all the MCi_XXX settings
@@ -200,6 +203,45 @@ static void trigger_thr_int(void *info) asmvolatile ("int %0" ::"i" (THRESHOLD_APIC_VECTOR )); } +static u32 get_nbc_for_node (int node_id) +{ +struct cpuinfo_x86 *c = &boot_cpu_data; +u32 cores_per_node; + + cores_per_node = c->x86_max_cores /amd_get_nodes_per_socket (); + +return cores_per_node * node_id; +} + +static void toggle_nb_mca_mst_cpu (u16 nid) +{ +struct pci_dev *F3 =node_to_amd_nb (nid)->misc; +u32 val; +int err; + +if (!F3 ) +return ; + + err =pci_read_config_dword (F3 ,NBCFG , &val); +if (err) { +pr_err ("%s: Error reading F%dx%03x.\n" , + __func__,PCI_FUNC (F3 ->devfn),NBCFG ); +return ; + } + +if (val &BIT (27 )) +return ; + +pr_err ("%s: Set D18F3x44[NbMcaToMstCpuEn] which BIOS hasn't done.\n" , + __func__); + + val |=BIT (27 ); + err =pci_write_config_dword (F3 ,NBCFG , val); +if (err) +pr_err ("%s: Error writing F%dx%03x.\n" , + __func__,PCI_FUNC (F3 ->devfn),NBCFG ); +} +static void do_inject (void ) {u64 mcg_status =0 ;
@@ -220,7 +262,6 @@ static void do_inject(void)* b. unset MCx_STATUS[UC] * As deferred errors are _not_ UC */ - i_mce.status |=MCI_STATUS_DEFERRED ; i_mce.status &= ~MCI_STATUS_UC ; }
@@ -235,6 +276,20 @@ static void do_inject(void)if (!(i_mce.status &MCI_STATUS_PCC )) mcg_status |=MCG_STATUS_RIPV ; +/* +* For multi node CPUs, logging and reporting of bank 4 errors happens +* only on the node base core. Refer to D18F3x44[NbMcaToMstCpuEn] for +* Fam10h and later BKDGs. +*/ +if (static_cpu_has (X86_FEATURE_AMD_DCM ) && b ==4 ) { +/* +* BIOS sets D18F3x44[NbMcaToMstCpuEn] by default. But make sure +* of it here just in case. +*/ +toggle_nb_mca_mst_cpu (amd_get_nb_id (cpu)); + cpu =get_nbc_for_node (amd_get_nb_id (cpu)); + } +toggle_hw_mce_inject (cpu, true); wrmsr_on_cpu(cpu, MSR_IA32_MCG_STATUS,
Signature · 8 lines
2.3.5
--
Regards/Gruss,
Boris.
ECO tip #101: Trim your mails when you reply.
--