On Mon, Apr 30, 2012 at 7:29 AM, Konstantin Khlebnikov
<[email protected]> wrote:
> This patch adds line "HWPoinson: <size> kB" into /proc/pid/smaps if
> CONFIG_MEMORY_FAILURE=y and some HWPoison pages were found.
> This may be useful for searching applications which use a broken memory.
I dislike "maybe useful" claim. If we don't know exact motivation of a feature,
we can't maintain them especially when a bugfix can't avoid ABI change.
Please write down exact use case.
KOSAKI Motohiro wrote:
> On Mon, Apr 30, 2012 at 7:29 AM, Konstantin Khlebnikov
> <[email protected]> wrote:
>> This patch adds line "HWPoinson:<size> kB" into /proc/pid/smaps if
>> CONFIG_MEMORY_FAILURE=y and some HWPoison pages were found.
>> This may be useful for searching applications which use a broken memory.
>
> I dislike "maybe useful" claim. If we don't know exact motivation of a feature,
> we can't maintain them especially when a bugfix can't avoid ABI change.
>
> Please write down exact use case.
I don't know how to exactly use this hw-poison stuff, but smaps suppose to
export state of ptes in vma. It seems to rational to show also hw-poisoned ptes,
since kernel has this feature and pte can be in hw-poisoned state.
and now everyone can easily find them:
# sudo grep HWPoison /proc/*/smaps
On Tue, May 1, 2012 at 2:05 PM, Konstantin Khlebnikov
<[email protected]> wrote:
> KOSAKI Motohiro wrote:
>>
>> On Mon, Apr 30, 2012 at 7:29 AM, Konstantin Khlebnikov
>> <[email protected]> ?wrote:
>>>
>>> This patch adds line "HWPoinson:<size> ?kB" into /proc/pid/smaps if
>>> CONFIG_MEMORY_FAILURE=y and some HWPoison pages were found.
>>> This may be useful for searching applications which use a broken memory.
>>
>>
>> I dislike "maybe useful" claim. If we don't know exact motivation of a
>> feature,
>> we can't maintain them especially when a bugfix can't avoid ABI change.
>>
>> Please write down exact use case.
>
> I don't know how to exactly use this hw-poison stuff, but smaps suppose to
> export state of ptes in vma. It seems to rational to show also hw-poisoned
> ptes,
> since kernel has this feature and pte can be in hw-poisoned state.
>
> and now everyone can easily find them:
> # sudo grep HWPoison /proc/*/smaps
First, I don't think "we can expose it" is good reason. Second, hw-poisoned mean
such process is going to be killed at next page touch. But I can't
imagine anyone can
use its information because it's racy against process kill. I think
admin should use mce log.
So, until we find a good use case, I don't ack this.
On Tue, 1 May 2012 14:14:57 -0400
KOSAKI Motohiro <[email protected]> wrote:
> On Tue, May 1, 2012 at 2:05 PM, Konstantin Khlebnikov
> <[email protected]> wrote:
> > KOSAKI Motohiro wrote:
> >>
> >> On Mon, Apr 30, 2012 at 7:29 AM, Konstantin Khlebnikov
> >> <[email protected]> __wrote:
> >>>
> >>> This patch adds line "HWPoinson:<size> __kB" into /proc/pid/smaps if
> >>> CONFIG_MEMORY_FAILURE=y and some HWPoison pages were found.
> >>> This may be useful for searching applications which use a broken memory.
> >>
> >>
> >> I dislike "maybe useful" claim. If we don't know exact motivation of a
> >> feature,
> >> we can't maintain them especially when a bugfix can't avoid ABI change.
> >>
> >> Please write down exact use case.
> >
> > I don't know how to exactly use this hw-poison stuff, but smaps suppose to
> > export state of ptes in vma. It seems to rational to show also hw-poisoned
> > ptes,
> > since kernel has this feature and pte can be in hw-poisoned state.
> >
> > and now everyone can easily find them:
> > # sudo grep HWPoison /proc/*/smaps
>
> First, I don't think "we can expose it" is good reason. Second, hw-poisoned mean
> such process is going to be killed at next page touch. But I can't
> imagine anyone can
> use its information because it's racy against process kill. I think
> admin should use mce log.
>
> So, until we find a good use case, I don't ack this.
Yes, I think I'll drop this patch for now. If we can later produce a
good reason for expanding the kernel API in this fashion then please
resend.
From: Konstantin Khlebnikov <[email protected]>
Subject: proc/smaps: show amount of hwpoison pages
Add the line "HWPoinson: <size> kB" into /proc/pid/smaps if
CONFIG_MEMORY_FAILURE=y and some HWPoison pages were found. This may be
useful for searching applications which use a broken memory.
Signed-off-by: Konstantin Khlebnikov <[email protected]>
Acked-by: Andi Kleen <[email protected]>
Cc: Pavel Emelyanov <[email protected]>
Cc: Alexey Dobriyan <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
---
fs/proc/task_mmu.c | 10 ++++++++++
1 file changed, 10 insertions(+)
diff -puN fs/proc/task_mmu.c~proc-smaps-show-amount-of-hwpoison-pages fs/proc/task_mmu.c
--- a/fs/proc/task_mmu.c~proc-smaps-show-amount-of-hwpoison-pages
+++ a/fs/proc/task_mmu.c
@@ -394,6 +394,7 @@ struct mem_size_stats {
unsigned long anonymous_thp;
unsigned long swap;
unsigned long nonlinear;
+ unsigned long hwpoison;
u64 pss;
};
@@ -416,6 +417,8 @@ static void smaps_pte_entry(pte_t ptent,
mss->swap += ptent_size;
else if (is_migration_entry(swpent))
page = migration_entry_to_page(swpent);
+ else if (is_hwpoison_entry(swpent))
+ mss->hwpoison += ptent_size;
} else if (pte_file(ptent)) {
if (pte_to_pgoff(ptent) != pgoff)
mss->nonlinear += ptent_size;
@@ -430,6 +433,9 @@ static void smaps_pte_entry(pte_t ptent,
if (page->index != pgoff)
mss->nonlinear += ptent_size;
+ if (PageHWPoison(page))
+ mss->hwpoison += ptent_size;
+
mss->resident += ptent_size;
/* Accumulate the size in pages that have been accessed. */
if (pte_young(ptent) || PageReferenced(page))
@@ -535,6 +541,10 @@ static int show_smap(struct seq_file *m,
seq_printf(m, "Nonlinear: %8lu kB\n",
mss.nonlinear >> 10);
+ if (IS_ENABLED(CONFIG_MEMORY_FAILURE) && mss.hwpoison)
+ seq_printf(m, "HWPoison: %8lu kB\n",
+ mss.hwpoison >> 10);
+
if (m->count < m->size) /* vma is copied successfully */
m->version = (vma != get_gate_vma(task->mm))
? vma->vm_start : 0;
_