Revisit MCE handling for UE Errors
v4 cover Revisit MCE handling for UE Errors
This patch series is designed to hook up memory_failure on UE errors, this is specially helpful for user_mode UE errors.
The first two patches cleanup bits, remove dead code. I could not find any users of get_mce_fault_addr(). The second one improves printing of physical address
The third patch walks kernel/user mode page tables in real mode to extract the effective address of the instruction that caused the UE error and the effective address it was trying to access (for load/store). The fourth patch hooks up the pfn for instruction UE errors (ierror).
The fifth patch hooks up memory_failure to the MCE patch.
TODO:
Log the address in NVRAM, so that we can recover from bad pages at boot and keep the blacklist persistent.
Changelog v4:
- Add a #define for MAX_MCE_DEPTH instead of hard coding the value
- Refactor addr_to_pfn to deduce the right parameters
Changelog v3:
- Check for recursive MCE invocations (suggested by Nick)
Changelog v2:
- Remove pr_warn from real mode to a more delayed context, where its OK to warn.
- Add a trivial patch to align prints of physical and effective address
Changelog v1:
- Address review comments from Nick and Mahesh (initialization of pfn and more comments on failure when addr_to_pfn() or anaylse_instr() fail)
- Hookup ierrors to the framework as well (comments from Mahesh)
Balbir Singh (5):
powerpc/mce.c: Remove unused function get_mce_fault_addr() powerpc/mce: align the print of physical address better powerpc/mce: Hookup derror (load/store) UE errors powerpc/mce: Hookup ierror (instruction) UE errors powerpc/mce: hookup memory_failure for UE errors
arch/powerpc/include/asm/exception-64s.h | 5 ++ arch/powerpc/include/asm/mce.h | 4 +- arch/powerpc/kernel/exceptions-64s.S | 2 +- arch/powerpc/kernel/mce.c | 117 +++++++++++++++++++------------ arch/powerpc/kernel/mce_power.c | 109 +++++++++++++++++++++++++--- 5 files changed, 181 insertions(+), 56 deletions(-)
Signature · 1 line
2.9.5
PATCH v4 1/5 powerpc/mce.c: Remove unused function get_mce_fault_addr()
There are no users of get_mce_fault_addr()
--- arch/powerpc/include/asm/mce.h | 2 -- arch/powerpc/kernel/mce.c | 39 --------------------------------------- 2 files changed, 41 deletions(-)
arch/powerpc/include/asm/mce.h +0 -2
@@ -210,6 +210,4 @@ extern void release_mce_event(void);extern void machine_check_queue_event (void );extern void machine_check_print_event_info (struct machine_check_event *evt,bool user_mode); -extern uint64_t get_mce_fault_addr (struct machine_check_event *evt); -#endif /* __ASM_PPC64_MCE_H__ */
arch/powerpc/kernel/mce.c +0 -39
@@ -411,45 +411,6 @@ void machine_check_print_event_info(struct machine_check_event *evt, }EXPORT_SYMBOL_GPL (machine_check_print_event_info); -uint64_t get_mce_fault_addr (struct machine_check_event *evt) -{ -switch (evt->error_type) { -case MCE_ERROR_TYPE_UE : -if (evt->u.ue_error.effective_address_provided) -return evt->u.ue_error.effective_address; -break ; -case MCE_ERROR_TYPE_SLB : -if (evt->u.slb_error.effective_address_provided) -return evt->u.slb_error.effective_address; -break ; -case MCE_ERROR_TYPE_ERAT : -if (evt->u.erat_error.effective_address_provided) -return evt->u.erat_error.effective_address; -break ; -case MCE_ERROR_TYPE_TLB : -if (evt->u.tlb_error.effective_address_provided) -return evt->u.tlb_error.effective_address; -break ; -case MCE_ERROR_TYPE_USER : -if (evt->u.user_error.effective_address_provided) -return evt->u.user_error.effective_address; -break ; -case MCE_ERROR_TYPE_RA : -if (evt->u.ra_error.effective_address_provided) -return evt->u.ra_error.effective_address; -break ; -case MCE_ERROR_TYPE_LINK : -if (evt->u.link_error.effective_address_provided) -return evt->u.link_error.effective_address; -break ; -default : -case MCE_ERROR_TYPE_UNKNOWN : -break ; - } -return 0 ; -} -EXPORT_SYMBOL (get_mce_fault_addr); -/* * This function is called in real mode. Strictly no printk's please. *
Signature · 1 line
2.9.5
Michael Ellerman
reviewerRe: [PATCH v4 1/5] powerpc/mce.c: Remove unused function get_mce_fault_addr()
Quoted 3 lines · Balbir Singh
There are no users of
get_mce_fault_addr()Fixes b63a0ff ("powerpc/powernv: Machine check exception handling.")
That fixes line is wrong, get_mce_fault_addr() was used in that commit.
The last usage was removed in:
1363875bdb63 ("powerpc/64s: fix handling of non-synchronous machine checks")
So that's what the fixes tag should point at if anything, I'll update it here.
cheers
Re: [PATCH v4 1/5] powerpc/mce.c: Remove unused function get_mce_fault_addr()
Quoted 6 lines · Michael Ellerman
Quoted 3 lines · Balbir Singh
There are no users of
get_mce_fault_addr()Fixes b63a0ff ("powerpc/powernv: Machine check exception handling.")
Quoted 8 lines
That fixes line is wrong,
get_mce_fault_addr()was used in that commit.The last usage was removed in:
1363875bdb63("powerpc/64s: fix handling of non-synchronous machine checks")So that's what the fixes tag should point at if anything, I'll update it here.
Sorry, my git search was really awe-full! Thanks for fixing it.
Balbir Singh
Re: [v4, 1/5] powerpc/mce.c: Remove unused function get_mce_fault_addr()
In reply to Balbir SinghQuoted 6 lines · Balbir Singh
There are no users of
get_mce_fault_addr()Fixes b63a0ff ("powerpc/powernv: Machine check exception handling.")Signed-off-by Balbir Singh <[email protected]>Reviewed-by Nicholas Piggin <[email protected]>
Series applied to powerpc next, thanks.
https://git.kernel.org/powerpc/c/73e341eb6bea01fde706d10d7edba9
cheers
PATCH v4 2/5 powerpc/mce: align the print of physical address better
In reply to Balbir SinghUse the same alignment as Effective address and rename phyiscal address to Page Frame Number
--- arch/powerpc/kernel/mce.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/powerpc/kernel/mce.c +1 -1
@@ -340,7 +340,7 @@ void machine_check_print_event_info(struct machine_check_event *evt,printk ("%s Effective address: %016llx\n" , level, evt->u.ue_error.effective_address);if (evt->u.ue_error.physical_address_provided) -printk ("%s Physical address: %016llx\n" , +printk ("%s Physical address: %016llx\n" , level, evt->u.ue_error.physical_address);break ;case MCE_ERROR_TYPE_SLB :
Signature · 1 line
2.9.5
Michael Ellerman
reviewerRe: [PATCH v4 2/5] powerpc/mce: align the print of physical address better
Quoted 2 lines · Balbir Singh
Use the same alignment as Effective address
and rename phyiscal address to Page Frame Number
You didn't do the 2nd part AFAICS?
Will fix it up.
cheers
Quoted 15 lines
arch/powerpc/kernel/mce.c +1
-1@@ -340,7 +340,7 @@ void machine_check_print_event_info(struct machine_check_event *evt,printk ("%s Effective address: %016llx\n" , level, evt->u.ue_error.effective_address);if (evt->u.ue_error.physical_address_provided) -printk ("%s Physical address: %016llx\n" , +printk ("%s Physical address: %016llx\n" , level, evt->u.ue_error.physical_address);break ;case MCE_ERROR_TYPE_SLB :Signature · 1 line
2.9.5
Re: [PATCH v4 2/5] powerpc/mce: align the print of physical address better
Quoted 8 lines · Michael Ellerman
Quoted 2 lines · Balbir Singh
Use the same alignment as Effective address
and rename phyiscal address to Page Frame NumberYou didn't do the 2nd part AFAICS?
Will fix it up.
Sorry the changelog is buggy :( Nick asked me to keep the physical address. I should have removed the changes when I reverted the changes
Balbir
PATCH v4 3/5 powerpc/mce: Hookup derror (load/store) UE errors
In reply to Balbir SinghExtract physical_address for UE errors by walking the page tables for the mm and address at the NIP, to extract the instruction. Then use the instruction to find the effective address via analyse_instr().
We might have page table walking races, but we expect them to be rare, the physical address extraction is best effort. The idea is to then hook up this infrastructure to memory failure eventually.
--- arch/powerpc/include/asm/exception-64s.h | 5 ++ arch/powerpc/include/asm/mce.h | 2 +- arch/powerpc/kernel/exceptions-64s.S | 2 +- arch/powerpc/kernel/mce.c | 6 ++- arch/powerpc/kernel/mce_power.c | 87 ++++++++++++++++++++++++++++++-- 5 files changed, 94 insertions(+), 8 deletions(-)
arch/powerpc/include/asm/exception-64s.h +5 -0
@@ -55,6 +55,11 @@#endif /* +* maximum recursive depth of MCE exceptions +*/ +#define MAX_MCE_DEPTH 4 + +/* * EX_LR is only used in EXSLB and where it does not overlap with EX_DAR * EX_CCR similarly with DSISR, but being 4 byte registers there is a hole * in the save area so it's not necessary to overlap them. Could be used
arch/powerpc/include/asm/mce.h +1 -1
@@ -204,7 +204,7 @@ struct mce_error_info {extern void save_mce_event (struct pt_regs *regs,long handled,struct mce_error_info *mce_err,uint64_t nip, -uint64_t addr); +uint64_t addr,uint64_t phys_addr);extern int get_mce_event (struct machine_check_event *mce,bool release);extern void release_mce_event (void );extern void machine_check_queue_event (void );
arch/powerpc/kernel/exceptions-64s.S +1 -1
@@ -232,7 +232,7 @@ BEGIN_FTR_SECTION addi r10,r10,1 /* increment paca->in_mce */ sth r10,PACA_IN_MCE(r13) /* Limit nested MCE to level 4 to avoid stack overflow */ - cmpwi r10,4 + cmpwi r10,MAX_MCE_DEPTH bgt 2f /* Check if we hit limit of 4 */ std r11,GPR1(r1) /* Save r1 on the stack. */ std r11,0(r1) /* make stack chain pointer */
arch/powerpc/kernel/mce.c +5 -1
@@ -82,7 +82,7 @@ static void mce_set_error_info(struct machine_check_event *mce,*/ void save_mce_event (struct pt_regs *regs,long handled,struct mce_error_info *mce_err, -uint64_t nip,uint64_t addr) +uint64_t nip,uint64_t addr,uint64_t phys_addr) {int index =__this_cpu_inc_return (mce_nest_count) -1 ;struct machine_check_event *mce =this_cpu_ptr (&mce_event[index]);
@@ -140,6 +140,10 @@ void save_mce_event(struct pt_regs *regs, long handled, }else if (mce->error_type ==MCE_ERROR_TYPE_UE ) { mce->u.ue_error.effective_address_provided = true; mce->u.ue_error.effective_address = addr; +if (phys_addr !=ULONG_MAX ) { + mce->u.ue_error.physical_address_provided = true; + mce->u.ue_error.physical_address = phys_addr; + } }return ; }
arch/powerpc/kernel/mce_power.c +82 -5
127 lines not on this page · raw message
Signature · 1 line
2.9.5
Michael Ellerman
reviewerRe: [PATCH v4 3/5] powerpc/mce: Hookup derror (load/store) UE errors
Quoted 7 lines · Balbir Singh
arch/powerpc/kernel/mce_power.c +0
-0@@ -27,6 +27,36 @@#include <asm/mmu.h> #include <asm/mce.h>
arch/powerpc/kernel/mce_power.c line 38
Quoted 10 lines
arch/powerpc/kernel/mce_power.c +9
-0#include <asm/machdep.h> +#include <asm/pgtable.h> +#include <asm/pte-walk.h> +#include <asm/sstep.h> +#include <asm/exception-64s.h> + +/* +* Convert an address related to an mm to a PFN. NOTE: we are in real +* mode, we could potentially race with page table updates. +*/
That will blow some serious chunks if we're on pseries.
It seems it can't be triggered on pseries at the moment, but only because of the feature section around the call to machine_check_handle_early().
We should maybe move the powernv only code in here inside an #ifdef POWERNV, just for documentation purposes. But we can do that later.
Quoted 11 lines
arch/powerpc/kernel/mce_power.c +11
-0+static unsigned long addr_to_pfn (struct pt_regs *regs,unsigned long addr) +{ +pte_t *ptep; +unsigned long flags; +struct mm_struct *mm; + +if (user_mode (regs)) + mm = current->mm; +else + mm = &init_mm; +
arch/powerpc/kernel/mce_power.c line 59
Quoted 10 lines
arch/powerpc/kernel/mce_power.c +10
-0+local_irq_save (flags); +if (mm == current->mm) + ptep =find_current_mm_pte (mm->pgd, addr,NULL ,NULL ); +else + ptep =find_init_mm_pte (addr,NULL ); +local_irq_restore (flags); +if (!ptep ||pte_special (*ptep)) +return ULONG_MAX ; +return pte_pfn (*ptep); +}
cheers
Re: [PATCH v4 3/5] powerpc/mce: Hookup derror (load/store) UE errors
Quoted 22 lines · Michael Ellerman
Quoted 7 lines · Balbir Singh
arch/powerpc/kernel/mce_power.c +0
-0@@ -27,6 +27,36 @@#include <asm/mmu.h> #include <asm/mce.h> arch/powerpc/kernel/mce_power.c line 38
Quoted 10 lines
arch/powerpc/kernel/mce_power.c +9
-0#include <asm/machdep.h> +#include <asm/pgtable.h> +#include <asm/pte-walk.h> +#include <asm/sstep.h> +#include <asm/exception-64s.h> + +/* +* Convert an address related to an mm to a PFN. NOTE: we are in real +* mode, we could potentially race with page table updates. +*/ That will blow some serious chunks if we're on pseries.
Quoted 7 lines
It seems it can't be triggered on pseries at the moment, but only because of the feature section around the call to
machine_check_handle_early().We should maybe move the powernv only code in here inside an #ifdef POWERNV, just for documentation purposes. But we can do that later.
Will do that as a follow up thanks! I think the entire mce_power.c and mce.c can move to pseries, will validate with Mahesh and Nick as well.
Balbir
PATCH v4 4/5 powerpc/mce: Hookup ierror (instruction) UE errors
In reply to Balbir SinghHookup instruction errors (UE) for memory offling via memory_failure() in a manner similar to load/store errors (derror). Since we have access to the NIP, the conversion is a one step process in this case.
--- arch/powerpc/kernel/mce_power.c | 22 +++++++++++++++++++--- 1 file changed, 19 insertions(+), 3 deletions(-)
arch/powerpc/kernel/mce_power.c +19 -3
@@ -488,7 +488,8 @@ static int mce_find_instr_ea_and_pfn(struct pt_regs *regs, uint64_t *addr,static int mce_handle_ierror (struct pt_regs *regs,const struct mce_ierror_table table[], -struct mce_error_info *mce_err,uint64_t *addr) +struct mce_error_info *mce_err,uint64_t *addr, +uint64_t *phys_addr) {uint64_t srr1 = regs->msr;int handled =0 ;
@@ -540,8 +541,22 @@ static int mce_handle_ierror(struct pt_regs *regs, } mce_err->severity = table[i].severity; mce_err->initiator = table[i].initiator; -if (table[i].nip_valid) +if (table[i].nip_valid) { *addr = regs->nip; +if (mce_err->severity ==MCE_SEV_ERROR_SYNC && + table[i].error_type ==MCE_ERROR_TYPE_UE ) { +unsigned long pfn; + +if (get_paca ()->in_mce <MAX_MCE_DEPTH ) { + pfn =addr_to_pfn (regs, regs->nip); +if (pfn !=ULONG_MAX ) { + *phys_addr = + (pfn <<PAGE_SHIFT ); + handled =1 ; + } + } + } + }return handled; }
@@ -676,7 +691,8 @@ static long mce_handle_error(struct pt_regs *regs, handled =mce_handle_derror (regs, dtable, &mce_err, &addr, &phys_addr); else - handled =mce_handle_ierror (regs, itable, &mce_err, &addr); + handled =mce_handle_ierror (regs, itable, &mce_err, &addr, + &phys_addr);if (!handled && mce_err.error_type ==MCE_ERROR_TYPE_UE ) handled =mce_handle_ue_error (regs);
Signature · 1 line
2.9.5
PATCH v4 5/5 powerpc/mce: hookup memory_failure for UE errors
In reply to Balbir SinghIf we are in user space and hit a UE error, we now have the basic infrastructure to walk the page tables and find out the effective address that was accessed, since the DAR is not valid.
We use a work_queue content to hookup the bad pfn, any other context causes problems, since memory_failure itself can call into schedule() via lru_drain_ bits.
We could probably poison the struct page to avoid a race between detection and taking corrective action.
--- arch/powerpc/kernel/mce.c | 70 +++++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 67 insertions(+), 3 deletions(-)
arch/powerpc/kernel/mce.c +67 -3
@@ -39,11 +39,21 @@ static DEFINE_PER_CPU(struct machine_check_event[MAX_MC_EVT], mce_event); static DEFINE_PER_CPU(int, mce_queue_count); static DEFINE_PER_CPU(struct machine_check_event[MAX_MC_EVT], mce_event_queue); +/* Queue for delayed MCE UE events. */ +static DEFINE_PER_CPU(int, mce_ue_count); +static DEFINE_PER_CPU(struct machine_check_event[MAX_MC_EVT], + mce_ue_event_queue); +static void machine_check_process_queued_event (struct irq_work *work); +void machine_check_ue_event (struct machine_check_event *evt); +static void machine_process_ue_event (struct work_struct *work); +static struct irq_work mce_event_process_work = { .func = machine_check_process_queued_event, }; +DECLARE_WORK (mce_ue_event_work, machine_process_ue_event); +static void mce_set_error_info (struct machine_check_event *mce,struct mce_error_info *mce_err) {
@@ -143,6 +153,7 @@ void save_mce_event(struct pt_regs *regs, long handled,if (phys_addr !=ULONG_MAX ) { mce->u.ue_error.physical_address_provided = true; mce->u.ue_error.physical_address = phys_addr; +machine_check_ue_event (mce); } }return ;
@@ -197,6 +208,26 @@ void release_mce_event(void)get_mce_event (NULL , true); } + +/* +* Queue up the MCE event which then can be handled later. +*/ +void machine_check_ue_event (struct machine_check_event *evt) +{ +int index; + + index =__this_cpu_inc_return (mce_ue_count) -1 ; +/* If queue is full, just return for now. */ +if (index >=MAX_MC_EVT ) { +__this_cpu_dec (mce_ue_count); +return ; + } +memcpy (this_cpu_ptr (&mce_ue_event_queue[index]), evt,sizeof (*evt)); + +/* Queue work to process this event later. */ +schedule_work (&mce_ue_event_work); +} +/* * Queue up the MCE event which then can be handled later. */
@@ -219,7 +250,39 @@ void machine_check_queue_event(void)/* Queue irq work to process this event later. */ irq_work_queue (&mce_event_process_work); } - +/* +* process pending MCE event from the mce event queue. This function will be +* called during syscall exit. +*/ +static void machine_process_ue_event (struct work_struct *work) +{ +int index; +struct machine_check_event *evt; + +while (__this_cpu_read (mce_ue_count) >0 ) { + index =__this_cpu_read (mce_ue_count) -1 ; + evt =this_cpu_ptr (&mce_ue_event_queue[index]); +#ifdef CONFIG_MEMORY_FAILURE +/* +* This should probably queued elsewhere, but +* oh! well +*/ +if (evt->error_type ==MCE_ERROR_TYPE_UE ) { +if (evt->u.ue_error.physical_address_provided) { +unsigned long pfn; + + pfn = evt->u.ue_error.physical_address >> +PAGE_SHIFT ; +memory_failure (pfn,SIGBUS ,0 ); + }else +pr_warn ("Failed to identify bad address from " +"where the uncorrectable error (UE) " +"was generated\n" ); + } +#endif +__this_cpu_dec (mce_ue_count); + } +}/* * process pending MCE event from the mce event queue. This function will be * called during syscall exit.
@@ -227,6 +290,7 @@ void machine_check_queue_event(void)static void machine_check_process_queued_event (struct irq_work *work) {int index; +struct machine_check_event *evt;add_taint (TAINT_MACHINE_CHECK ,LOCKDEP_NOW_UNRELIABLE );
@@ -236,8 +300,8 @@ static void machine_check_process_queued_event(struct irq_work *work)*/ while (__this_cpu_read (mce_queue_count) >0 ) { index =__this_cpu_read (mce_queue_count) -1 ; -machine_check_print_event_info ( -this_cpu_ptr (&mce_event_queue[index]), false); + evt =this_cpu_ptr (&mce_event_queue[index]); +machine_check_print_event_info (evt, false);__this_cpu_dec (mce_queue_count); } }
Signature · 1 line
2.9.5
Michael Ellerman
reviewerRe: [PATCH v4 5/5] powerpc/mce: hookup memory_failure for UE errors
Quoted 9 lines · Balbir Singh
If we are in user space and hit a UE error, we now have the basic infrastructure to walk the page tables and find out the effective address that was accessed, since the DAR is not valid.
We use a
work_queuecontent to hookup the bad pfn, any other context causes problems, sincememory_failureitself can call intoschedule()vialru_drain_bits.
Quoted 8 lines
We could probably poison the struct page to avoid a race between detection and taking corrective action.
Signed-off-by Balbir Singh <[email protected]>Reviewed-by Nicholas Piggin <[email protected]>--- arch/powerpc/kernel/mce.c | 70 +++++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 67 insertions(+), 3 deletions(-)
I'm not sure why this is in mce.c but the rest was in mce_power.c ?
cheers
Re: [PATCH v4 5/5] powerpc/mce: hookup memory_failure for UE errors
Quoted 14 lines · Michael Ellerman
Quoted 12 lines · Balbir Singh
If we are in user space and hit a UE error, we now have the basic infrastructure to walk the page tables and find out the effective address that was accessed, since the DAR is not valid.
We use a
work_queuecontent to hookup the bad pfn, any other context causes problems, sincememory_failureitself can call intoschedule()vialru_drain_bits.We could probably poison the struct page to avoid a race between detection and taking corrective action.
Quoted 7 lines
Quoted 5 lines
Signed-off-by Balbir Singh <[email protected]>Reviewed-by Nicholas Piggin <[email protected]>--- arch/powerpc/kernel/mce.c | 70 +++++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 67 insertions(+), 3 deletions(-)I'm not sure why this is in
mce.cbut the rest was inmce_power.c?
The way the code is organized is that save_mce_event is implemented here and called from mce_power.c. save_mce_event() does the processing of the event.
Balbir Singh.
Michael Ellerman
reviewerRe: [PATCH v4 5/5] powerpc/mce: hookup memory_failure for UE errors
Quoted 21 lines · Balbir Singh
Quoted 20 lines · Michael Ellerman
Quoted 17 lines · Balbir Singh
If we are in user space and hit a UE error, we now have the basic infrastructure to walk the page tables and find out the effective address that was accessed, since the DAR is not valid.
We use a
work_queuecontent to hookup the bad pfn, any other context causes problems, sincememory_failureitself can call intoschedule()vialru_drain_bits.We could probably poison the struct page to avoid a race between detection and taking corrective action.
Signed-off-by Balbir Singh <[email protected]>Reviewed-by Nicholas Piggin <[email protected]>--- arch/powerpc/kernel/mce.c | 70 +++++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 67 insertions(+), 3 deletions(-)
Quoted 5 lines
Quoted 1 line
I'm not sure why this is in
mce.cbut the rest was inmce_power.c?The way the code is organized is that
save_mce_eventis implemented here and called frommce_power.c.save_mce_event()does the processing of the event.
save_mce_event() is only called from mce_power.c :/
I'd be happy if it was mce_real_mode.c and mce.c, that would be a good distinction to make.
Though even then, most of the code is Book3s64 specific I think so maybe it shouldn't be in just mce.c.
Anyway I'll merge this and we can improve things later.
cheers