x86, CMCI: Add proper detection of end of CMCI storms
A message that is not in the archive (sent off-list or not yet received): [email protected]
Quoted 32 lines · Gong
Commit-ID: 27f6c573e0f77f7d1cc907c1494c99a61e48b7d8 Gitweb: http://git.kernel.org/tip/27f6c573e0f77f7d1cc907c1494c99a61e48b7d8 Author: Chen, Gong <[email protected]> AuthorDate: Thu, 27 Mar 2014 21:24:36 -0400 Committer: Tony Luck <[email protected]> CommitDate: Fri, 28 Mar 2014 13:40:16 -0700x86, CMCI: Add proper detection of end of CMCI storms
When CMCI storm persists for a long time(at least beyond predefined threshold. It's 30 seconds for now), we can watch CMCI storm is detected immediately after it subsides.
...
Dec 10 22:04:29 kernel: CMCI storm detected: switching to poll mode Dec 10 22:04:59 kernel: CMCI storm subsided: switching to interrupt mode Dec 10 22:04:59 kernel: CMCI storm detected: switching to poll mode Dec 10 22:05:29 kernel: CMCI storm subsided: switching to interrupt mode...
The problem is that our logic that determines that the storm has ended is incorrect. We announce the end, re-enable interrupts and realize that the storm is still going on, so we switch back to polling mode. Rinse, repeat.
When a storm happens we disable signaling of errors via CMCI and begin polling machine check banks instead. If we find any logged errors, then we need to set a per-cpu flag so that our per-cpu tests that check whether the storm is ongoing will see that errors are still being logged independently of whether
mce_notify_irq()says that the error has been fully processed.
Quoted 8 lines
cmci_clear()is not the right tool to disable a bank. It disables the interrupt for the bank as desired, but it also clears the bit for this bank in "mce_banks_owned" so we will skip the bank when polling (so we fail to see that the storm continues because we stop looking). Newcmci_storm_disable_banks()just disables the interrupt while allowing polling to continue.Reported-by William Dauchy <[email protected]>
Could you use the following address instead?
Thanks,
Quoted 2 lines
Signed-off-by Chen, Gong <[email protected]>Signed-off-by Tony Luck <[email protected]>
Signature · 1 line
William
Borislav Petkov
maintainerQuoted 2 lines · William Dauchy
Could you use the following address instead?
Reported-by William Dauchy <[email protected]>
It is too late for that now as the patch is in -tip already... Unless Ingo can still amend it, that is.
But, we're working on a real solution for the storm issue and there we'll be asking you to test stuff anyway so we'll make sure to use this mail address then, ok?
:-)
Signature · 5 lines
Regards/Gruss,
Boris.
Sent from a fat crate under my desk. Formatting is fine.
--Quoted 6 lines · Borislav Petkov
It is too late for that now as the patch is in -tip already... Unless Ingo can still amend it, that is.
But, we're working on a real solution for the storm issue and there we'll be asking you to test stuff anyway so we'll make sure to use this mail address then, ok?
ack
Signature · 1 line
William
Quoted 6 lines · Borislav Petkov
Quoted 2 lines · William Dauchy
Could you use the following address instead?
Reported-by William Dauchy <[email protected]>It is too late for that now as the patch is in -tip already... Unless Ingo can still amend it, that is.
There are already patches on top of it, so it's not possible - but even if it was the last one, since it got committed by Tony I cannot rebase or amend it.
Thanks,
Ingo
Borislav Petkov
maintainerQuoted 3 lines · Ingo Molnar
There are already patches on top of it, so it's not possible - but even if it was the last one, since it got committed by Tony I cannot rebase or amend it.
Ah right, you pulled it from him, sure.
Thanks.
Signature · 5 lines
Regards/Gruss,
Boris.
Sent from a fat crate under my desk. Formatting is fine.
--H. Peter Anvin
reviewerQuoted 10 lines · Gong
(quoted diff) +2
-0@@ -614,6 +618,8 @@ void machine_check_poll(enum mcp_flags flags, mce_banks_t *b) if (!(m.status & MCI_STATUS_VAL)) continue; + v = &get_cpu_var(mce_polled_error); + set_bit(0, v); /* * Uncorrected or signalled events are handled by the exception * handler when it is enabled, so don't process those here.
Review static unsigned long mce_adjust_timer_default(unsigned long interval) · line 1293
Quoted 10 lines
(quoted diff) +7
-0@@ -1278,10 +1284,18 @@ static unsigned long mce_adjust_timer_default(unsigned long interval) static unsigned long (*mce_adjust_timer)(unsigned long interval) = mce_adjust_timer_default; +static int cmc_error_seen(void) +{ + unsigned long *v = &__get_cpu_var(mce_polled_error); + + return test_and_clear_bit(0, v); +} +
Please use this_cpu_*() whereever possible instead of __get_cpu_var(). Since this is not actually a bitmask this_cpu_xchg() can be used at the end.
In fact, using set_bit() is completely wasteful.
I'll push this onward since it is a bit late, but please submit a cleanup patch.
-hpa
PATCH 1/2 x86, CMCI: Fix a missed put_cpu_var
In reply to a message not in the archiveThis issue is introduced in commit 27f6c573e0. I forget to execute put_cpu_var operation after get_cpu_var.
--- arch/x86/kernel/cpu/mcheck/mce.c | 1 + 1 file changed, 1 insertion(+)
arch/x86/kernel/cpu/mcheck/mce.c +1 -0
@@ -620,6 +620,7 @@ void machine_check_poll(enum mcp_flags flags, mce_banks_t *b) v = &get_cpu_var (mce_polled_error);set_bit (0 , v); +put_cpu_var (mce_polled_error);/* * Uncorrected or signalled events are handled by the exception * handler when it is enabled, so don't process those here.
Signature · 1 line
1.9.0
PATCH 2/2 x86, MCE: Cleanup macro __get_cpu_var
According to Peter's suggestion, use this_cpu_* instead of __get_cpu_var. BTW, remove bitmask ops to avoid unnecessary overhead.
--- arch/x86/kernel/cpu/mcheck/mce.c | 36 ++++++++++++++++++------------------ 1 file changed, 18 insertions(+), 18 deletions(-)
arch/x86/kernel/cpu/mcheck/mce.c +18 -18
@@ -402,7 +402,7 @@ static u64 mce_rdmsrl(u32 msr)if (offset <0 )return 0 ; -return *(u64 *)((char *)&__get_cpu_var (injectm) + offset); +return *(u64 *)((char *)this_cpu_ptr (&injectm) + offset); }if (rdmsrl_safe (msr, &v)) {
@@ -424,7 +424,7 @@ static void mce_wrmsrl(u32 msr, u64 v)int offset =msr_to_offset (msr);if (offset >=0 ) - *(u64 *)((char *)&__get_cpu_var (injectm) + offset) = v; + *(u64 *)((char *)this_cpu_ptr (&injectm) + offset) = v;return ; }wrmsrl (msr, v);
@@ -480,7 +480,7 @@ static DEFINE_PER_CPU(struct mce_ring, mce_ring);/* Runs with CPU affinity in workqueue */ static int mce_ring_empty (void ) { -struct mce_ring *r = &__get_cpu_var (mce_ring); +struct mce_ring *r =this_cpu_ptr (&mce_ring);return r->start == r->end; }
@@ -492,7 +492,7 @@ static int mce_ring_get(unsigned long *pfn) *pfn =0 ;get_cpu (); - r = &__get_cpu_var (mce_ring); + r =this_cpu_ptr (&mce_ring);if (r->start == r->end) goto out; *pfn = r->ring[r->start];
@@ -506,7 +506,7 @@ out:/* Always runs in MCE context with preempt off */ static int mce_ring_add (unsigned long pfn) { -struct mce_ring *r = &__get_cpu_var (mce_ring); +struct mce_ring *r =this_cpu_ptr (&mce_ring);unsigned next; next = (r->end +1 ) %MCE_RING_SIZE ;
@@ -528,7 +528,7 @@ int mce_available(struct cpuinfo_x86 *c)static void mce_schedule_work (void ) {if (!mce_ring_empty ()) -schedule_work (&__get_cpu_var (mce_work)); +schedule_work (this_cpu_ptr (&mce_work)); } DEFINE_PER_CPU(struct irq_work, mce_irq_work);
@@ -553,7 +553,7 @@ static void mce_report_event(struct pt_regs *regs)return ; } -irq_work_queue (&__get_cpu_var (mce_irq_work)); +irq_work_queue (this_cpu_ptr (&mce_irq_work)); }/*
@@ -619,7 +619,7 @@ void machine_check_poll(enum mcp_flags flags, mce_banks_t *b)continue ; v = &get_cpu_var (mce_polled_error); -set_bit (0 , v); + *v =1 ;put_cpu_var (mce_polled_error);/* * Uncorrected or signalled events are handled by the exception
@@ -1053,7 +1053,7 @@ void do_machine_check(struct pt_regs *regs, long error_code)mce_gather_info (&m, regs); - final = &__get_cpu_var (mces_seen); + final =this_cpu_ptr (&mces_seen); *final = m;memset (valid_banks,0 ,sizeof (valid_banks));
@@ -1287,14 +1287,14 @@ static unsigned long (*mce_adjust_timer)(unsigned long interval) =static int cmc_error_seen (void ) { -unsigned long *v = &__get_cpu_var (mce_polled_error); +unsigned long *v =this_cpu_ptr (&mce_polled_error); -return test_and_clear_bit (0 , v); +return this_cpu_xchg (*v,0 ); }static void mce_timer_fn (unsigned long data) { -struct timer_list *t = &__get_cpu_var (mce_timer); +struct timer_list *t =this_cpu_ptr (&mce_timer);unsigned long iv;int notify;
@@ -1302,7 +1302,7 @@ static void mce_timer_fn(unsigned long data)if (mce_available (__this_cpu_ptr (&cpu_info))) {machine_check_poll (MCP_TIMESTAMP , - &__get_cpu_var (mce_poll_banks)); +this_cpu_ptr (&mce_poll_banks));mce_intel_cmci_poll (); }
@@ -1332,7 +1332,7 @@ static void mce_timer_fn(unsigned long data)*/ void mce_timer_kick (unsigned long interval) { -struct timer_list *t = &__get_cpu_var (mce_timer); +struct timer_list *t =this_cpu_ptr (&mce_timer);unsigned long when = jiffies + interval;unsigned long iv =__this_cpu_read (mce_next_interval);
@@ -1668,7 +1668,7 @@ static void mce_start_timer(unsigned int cpu, struct timer_list *t)static void __mcheck_cpu_init_timer (void ) { -struct timer_list *t = &__get_cpu_var (mce_timer); +struct timer_list *t =this_cpu_ptr (&mce_timer);unsigned int cpu =smp_processor_id ();setup_timer (t, mce_timer_fn, cpu);
@@ -1711,8 +1711,8 @@ void mcheck_cpu_init(struct cpuinfo_x86 *c)__mcheck_cpu_init_generic ();__mcheck_cpu_init_vendor (c);__mcheck_cpu_init_timer (); -INIT_WORK (&__get_cpu_var (mce_work), mce_process_work); -init_irq_work (&__get_cpu_var (mce_irq_work), &mce_irq_work_cb); +INIT_WORK (this_cpu_ptr (&mce_work), mce_process_work); +init_irq_work (this_cpu_ptr (&mce_irq_work), &mce_irq_work_cb); }/*
@@ -1964,7 +1964,7 @@ static struct miscdevice mce_chrdev_device = {static void __mce_disable_bank (void *arg) {int bank = *((int *)arg); -__clear_bit (bank,__get_cpu_var (mce_poll_banks)); +__clear_bit (bank, *this_cpu_ptr (&mce_poll_banks));cmci_disable_bank (bank); }
Signature · 1 line
1.9.0
H. Peter Anvin
reviewerRe: [PATCH 2/2] x86, MCE: Cleanup macro __get_cpu_var
arch/x86/kernel/cpu/mcheck/mce.c void machine_check_poll(enum mcp_flags flags, mce_banks_t *b) · line 625
Quoted 9 lines · Gong
arch/x86/kernel/cpu/mcheck/mce.c +1
-1@@ -619,7 +619,7 @@ void machine_check_poll(enum mcp_flags flags, mce_banks_t *b)continue ; v = &get_cpu_var (mce_polled_error); -set_bit (0 , v); + *v =1 ;put_cpu_var (mce_polled_error);/* * Uncorrected or signalled events are handled by the exception
The amazing thing is that you managed to miss the one place where you could actively elide a pointer.
The above should simply be:
this_cpu_write(mce_polled_error, 1); -hpa
H. Peter Anvin
reviewerRe: [PATCH 2/2] x86, MCE: Cleanup macro __get_cpu_var
In reply to Chen, Gongarch/x86/kernel/cpu/mcheck/mce.c static unsigned long (*mce_adjust_timer)(unsigned long interval) = · line 1294
Quoted 11 lines · Gong
arch/x86/kernel/cpu/mcheck/mce.c +2
-2@@ -1287,14 +1287,14 @@ static unsigned long (*mce_adjust_timer)(unsigned long interval) =static int cmc_error_seen (void ) { -unsigned long *v = &__get_cpu_var (mce_polled_error); +unsigned long *v =this_cpu_ptr (&mce_polled_error); -return test_and_clear_bit (0 , v); +return this_cpu_xchg (*v,0 ); }
Here you produce a pointer and *then* passing it through a this_cpu_ function... this is actively wrong.
It should simply be:
return this_cpu_xchg(mce_polled_error, 0); -hpa
H. Peter Anvin
reviewerRe: [PATCH 2/2] x86, MCE: Cleanup macro __get_cpu_var
In reply to Chen, GongPlease read Documentation/this_cpu_ops.txt for reference for how to use these functions. However, you want to avoid forming a pointer if you can; it is relatively expensive to do so.
-hpa
Re: [PATCH 2/2] x86, MCE: Cleanup macro __get_cpu_var
Quoted 4 lines · H. Peter Anvin
Please read
Documentation/this_cpu_ops.txtfor reference for how to use these functions. However, you want to avoid forming a pointer if you can; it is relatively expensive to do so.
:-). I just found your comment yesterday so that I was eager to finish it. I should do more homework as you told and send it again. Thanks very much for your patience.