2012-05-28 15:21:40

by Aneesh Kumar K.V

Subject: [PATCH] mm/hugetlb: Use compound page head in migrate_huge_page

From: "Aneesh Kumar K.V" <[email protected]>

The change was introduced by "hugetlb: simplify migrate_huge_page() "

We should use compound page head instead of tail pages in
migrate_huge_page().

Signed-off-by: Aneesh Kumar K.V <[email protected]>
---
mm/memory-failure.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)

This is an important bug fix. If we want we can fold it with the not
yet merged upstream patch mentioned above in linux-next. The stack
trace for the crash is

[ 75.337421] BUG: unable to handle kernel NULL pointer dereference at 0000000000000080
[ 75.338386] IP: [<ffffffff816b3f0f>] __mutex_lock_common+0xa1/0x350
[ 75.338386] PGD 1d700067 PUD 1d7dd067 PMD 0
[ 75.338386] Oops: 0002 [#1] SMP
[ 75.338386] CPU 1
[ 75.338386] Modules linked in:
...
...

[ 75.338386] Call Trace:
[ 75.338386] [<ffffffff810ffc04>] ? try_to_unmap_file+0x38/0x51c
[ 75.338386] [<ffffffff810ffc04>] ? try_to_unmap_file+0x38/0x51c
[ 75.338386] [<ffffffff813b5f8b>] ? vsnprintf+0x83/0x421
[ 75.338386] [<ffffffff816b427d>] mutex_lock_nested+0x2a/0x31
[ 75.338386] [<ffffffff8110999b>] ? alloc_huge_page_node+0x1d/0x55
[ 75.338386] [<ffffffff810ffc04>] try_to_unmap_file+0x38/0x51c
[ 75.338386] [<ffffffff8110999b>] ? alloc_huge_page_node+0x1d/0x55
[ 75.338386] [<ffffffff810a06b9>] ? arch_local_irq_save+0x9/0xc
[ 75.338386] [<ffffffff816b5e3b>] ? _raw_spin_unlock+0x23/0x27
[ 75.338386] [<ffffffff81100839>] try_to_unmap+0x25/0x3c
[ 75.338386] [<ffffffff810641c2>] ? console_unlock+0x210/0x238
[ 75.338386] [<ffffffff811141e3>] migrate_huge_page+0x8d/0x178


diff --git a/mm/memory-failure.c b/mm/memory-failure.c
index 4a45098..53a1495 100644
--- a/mm/memory-failure.c
+++ b/mm/memory-failure.c
@@ -1428,8 +1428,8 @@ static int soft_offline_huge_page(struct page *page, int flags)
}

/* Keep page count to indicate a given hugepage is isolated. */
- ret = migrate_huge_page(page, new_page, MPOL_MF_MOVE_ALL, 0, true);
- put_page(page);
+ ret = migrate_huge_page(hpage, new_page, MPOL_MF_MOVE_ALL, 0, true);
+ put_page(hpage);
if (ret) {
pr_info("soft offline: %#lx: migration failed %d, type %lx\n",
pfn, ret, page->flags);
--
1.7.10

2012-05-28 19:13:30

by Michal Hocko

Subject: Re: [PATCH] mm/hugetlb: Use compound page head in migrate_huge_page

On Mon 28-05-12 20:51:30, Aneesh Kumar K.V wrote:
> From: "Aneesh Kumar K.V" <[email protected]>
>
> The change was introduced by "hugetlb: simplify migrate_huge_page() "
>
> We should use compound page head instead of tail pages in
> migrate_huge_page().
>
> Signed-off-by: Aneesh Kumar K.V <[email protected]>
> ---
> mm/memory-failure.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> This is an important bug fix. If we want we can fold it with the not
> yet merged upstream patch mentioned above in linux-next. The stack
> trace for the crash is
>
> [ 75.337421] BUG: unable to handle kernel NULL pointer dereference at 0000000000000080
> [ 75.338386] IP: [<ffffffff816b3f0f>] __mutex_lock_common+0xa1/0x350
> [ 75.338386] PGD 1d700067 PUD 1d7dd067 PMD 0
> [ 75.338386] Oops: 0002 [#1] SMP
> [ 75.338386] CPU 1
> [ 75.338386] Modules linked in:
> ...
> ...
>
> [ 75.338386] Call Trace:
> [ 75.338386] [<ffffffff810ffc04>] ? try_to_unmap_file+0x38/0x51c
> [ 75.338386] [<ffffffff810ffc04>] ? try_to_unmap_file+0x38/0x51c
> [ 75.338386] [<ffffffff813b5f8b>] ? vsnprintf+0x83/0x421
> [ 75.338386] [<ffffffff816b427d>] mutex_lock_nested+0x2a/0x31
> [ 75.338386] [<ffffffff8110999b>] ? alloc_huge_page_node+0x1d/0x55
> [ 75.338386] [<ffffffff810ffc04>] try_to_unmap_file+0x38/0x51c
> [ 75.338386] [<ffffffff8110999b>] ? alloc_huge_page_node+0x1d/0x55
> [ 75.338386] [<ffffffff810a06b9>] ? arch_local_irq_save+0x9/0xc
> [ 75.338386] [<ffffffff816b5e3b>] ? _raw_spin_unlock+0x23/0x27
> [ 75.338386] [<ffffffff81100839>] try_to_unmap+0x25/0x3c
> [ 75.338386] [<ffffffff810641c2>] ? console_unlock+0x210/0x238
> [ 75.338386] [<ffffffff811141e3>] migrate_huge_page+0x8d/0x178

This should be part of the changelog.

>
>
> diff --git a/mm/memory-failure.c b/mm/memory-failure.c
> index 4a45098..53a1495 100644
> --- a/mm/memory-failure.c
> +++ b/mm/memory-failure.c
> @@ -1428,8 +1428,8 @@ static int soft_offline_huge_page(struct page *page, int flags)
> }
>
> /* Keep page count to indicate a given hugepage is isolated. */
> - ret = migrate_huge_page(page, new_page, MPOL_MF_MOVE_ALL, 0, true);
> - put_page(page);
> + ret = migrate_huge_page(hpage, new_page, MPOL_MF_MOVE_ALL, 0, true);
> + put_page(hpage);
> if (ret) {
> pr_info("soft offline: %#lx: migration failed %d, type %lx\n",
> pfn, ret, page->flags);

I guess you want hpage->flags here.

--
Michal Hocko
SUSE Labs
SUSE LINUX s.r.o.
Lihovarska 1060/12
190 00 Praha 9
Czech Republic

2012-05-29 03:18:15

by Aneesh Kumar K.V

Subject: Re: [PATCH] mm/hugetlb: Use compound page head in migrate_huge_page

Michal Hocko <[email protected]> writes:

> On Mon 28-05-12 20:51:30, Aneesh Kumar K.V wrote:
>> From: "Aneesh Kumar K.V" <[email protected]>
>>
>> The change was introduced by "hugetlb: simplify migrate_huge_page() "
>>
>> We should use compound page head instead of tail pages in
>> migrate_huge_page().
>>
>> Signed-off-by: Aneesh Kumar K.V <[email protected]>
>> ---
>> mm/memory-failure.c | 4 ++--
>> 1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> This is an important bug fix. If we want we can fold it with the not
>> yet merged upstream patch mentioned above in linux-next. The stack
>> trace for the crash is
>>
>> [ 75.337421] BUG: unable to handle kernel NULL pointer dereference at 0000000000000080
>> [ 75.338386] IP: [<ffffffff816b3f0f>] __mutex_lock_common+0xa1/0x350
>> [ 75.338386] PGD 1d700067 PUD 1d7dd067 PMD 0
>> [ 75.338386] Oops: 0002 [#1] SMP
>> [ 75.338386] CPU 1
>> [ 75.338386] Modules linked in:
>> ...
>> ...
>>
>> [ 75.338386] Call Trace:
>> [ 75.338386] [<ffffffff810ffc04>] ? try_to_unmap_file+0x38/0x51c
>> [ 75.338386] [<ffffffff810ffc04>] ? try_to_unmap_file+0x38/0x51c
>> [ 75.338386] [<ffffffff813b5f8b>] ? vsnprintf+0x83/0x421
>> [ 75.338386] [<ffffffff816b427d>] mutex_lock_nested+0x2a/0x31
>> [ 75.338386] [<ffffffff8110999b>] ? alloc_huge_page_node+0x1d/0x55
>> [ 75.338386] [<ffffffff810ffc04>] try_to_unmap_file+0x38/0x51c
>> [ 75.338386] [<ffffffff8110999b>] ? alloc_huge_page_node+0x1d/0x55
>> [ 75.338386] [<ffffffff810a06b9>] ? arch_local_irq_save+0x9/0xc
>> [ 75.338386] [<ffffffff816b5e3b>] ? _raw_spin_unlock+0x23/0x27
>> [ 75.338386] [<ffffffff81100839>] try_to_unmap+0x25/0x3c
>> [ 75.338386] [<ffffffff810641c2>] ? console_unlock+0x210/0x238
>> [ 75.338386] [<ffffffff811141e3>] migrate_huge_page+0x8d/0x178
>
> This should be part of the changelog.

I was expecting the patch to be folded back to the existing patch in
-mm. That is the reason I added stack trace in the notes section so that
if we decided to keep it as a separate patch we can pull the stack trace
and add it to commit message.


>
>>
>>
>> diff --git a/mm/memory-failure.c b/mm/memory-failure.c
>> index 4a45098..53a1495 100644
>> --- a/mm/memory-failure.c
>> +++ b/mm/memory-failure.c
>> @@ -1428,8 +1428,8 @@ static int soft_offline_huge_page(struct page *page, int flags)
>> }
>>
>> /* Keep page count to indicate a given hugepage is isolated. */
>> - ret = migrate_huge_page(page, new_page, MPOL_MF_MOVE_ALL, 0, true);
>> - put_page(page);
>> + ret = migrate_huge_page(hpage, new_page, MPOL_MF_MOVE_ALL, 0, true);
>> + put_page(hpage);
>> if (ret) {
>> pr_info("soft offline: %#lx: migration failed %d, type %lx\n",
>> pfn, ret, page->flags);
>
> I guess you want hpage->flags here.

Existing code pass the flag details of the page passed as the
argument. I didn't want to change that.

-aneesh

2012-05-29 09:58:32

by Michal Hocko

Subject: Re: [PATCH] mm/hugetlb: Use compound page head in migrate_huge_page

On Tue 29-05-12 08:48:00, Aneesh Kumar K.V wrote:
> Michal Hocko <[email protected]> writes:
>
> > On Mon 28-05-12 20:51:30, Aneesh Kumar K.V wrote:
> >> From: "Aneesh Kumar K.V" <[email protected]>
> >>
> >> The change was introduced by "hugetlb: simplify migrate_huge_page() "
> >>
> >> We should use compound page head instead of tail pages in
> >> migrate_huge_page().
> >>
> >> Signed-off-by: Aneesh Kumar K.V <[email protected]>
> >> ---
> >> mm/memory-failure.c | 4 ++--
> >> 1 file changed, 2 insertions(+), 2 deletions(-)
> >>
> >> This is an important bug fix. If we want we can fold it with the not
> >> yet merged upstream patch mentioned above in linux-next. The stack
> >> trace for the crash is
> >>
> >> [ 75.337421] BUG: unable to handle kernel NULL pointer dereference at 0000000000000080
> >> [ 75.338386] IP: [<ffffffff816b3f0f>] __mutex_lock_common+0xa1/0x350
> >> [ 75.338386] PGD 1d700067 PUD 1d7dd067 PMD 0
> >> [ 75.338386] Oops: 0002 [#1] SMP
> >> [ 75.338386] CPU 1
> >> [ 75.338386] Modules linked in:
> >> ...
> >> ...
> >>
> >> [ 75.338386] Call Trace:
> >> [ 75.338386] [<ffffffff810ffc04>] ? try_to_unmap_file+0x38/0x51c
> >> [ 75.338386] [<ffffffff810ffc04>] ? try_to_unmap_file+0x38/0x51c
> >> [ 75.338386] [<ffffffff813b5f8b>] ? vsnprintf+0x83/0x421
> >> [ 75.338386] [<ffffffff816b427d>] mutex_lock_nested+0x2a/0x31
> >> [ 75.338386] [<ffffffff8110999b>] ? alloc_huge_page_node+0x1d/0x55
> >> [ 75.338386] [<ffffffff810ffc04>] try_to_unmap_file+0x38/0x51c
> >> [ 75.338386] [<ffffffff8110999b>] ? alloc_huge_page_node+0x1d/0x55
> >> [ 75.338386] [<ffffffff810a06b9>] ? arch_local_irq_save+0x9/0xc
> >> [ 75.338386] [<ffffffff816b5e3b>] ? _raw_spin_unlock+0x23/0x27
> >> [ 75.338386] [<ffffffff81100839>] try_to_unmap+0x25/0x3c
> >> [ 75.338386] [<ffffffff810641c2>] ? console_unlock+0x210/0x238
> >> [ 75.338386] [<ffffffff811141e3>] migrate_huge_page+0x8d/0x178
> >
> > This should be part of the changelog.
>
> I was expecting the patch to be folded back to the existing patch in
> -mm. That is the reason I added stack trace in the notes section so that
> if we decided to keep it as a separate patch we can pull the stack trace
> and add it to commit message.

OK.
--
Michal Hocko
SUSE Labs
SUSE LINUX s.r.o.
Lihovarska 1060/12
190 00 Praha 9
Czech Republic