[PATCH] kmemcheck: add scheduling point to kmemleak_scan

[PATCH] kmemcheck: add scheduling point to kmemleak_scan

[Yisheng Xie]

kmemleak_scan will scan struct page for each node and it can be really
large and resulting in a soft lockup. We have seen a soft lockup when do
scan while compile kernel:

 [  220.561051] watchdog: BUG: soft lockup - CPU#53 stuck for 22s! [bash:10287]
 [...]
 [  220.753837] Call Trace:
 [  220.756296]  kmemleak_scan+0x21a/0x4c0
 [  220.760034]  kmemleak_write+0x312/0x350
 [  220.763866]  ? do_wp_page+0x147/0x4c0
 [  220.767521]  full_proxy_write+0x5a/0xa0
 [  220.771351]  __vfs_write+0x33/0x150
 [  220.774833]  ? __inode_security_revalidate+0x4c/0x60
 [  220.779782]  ? selinux_file_permission+0xda/0x130
 [  220.784479]  ? _cond_resched+0x15/0x30
 [  220.788221]  vfs_write+0xad/0x1a0
 [  220.791529]  SyS_write+0x52/0xc0
 [  220.794758]  do_syscall_64+0x61/0x1a0
 [  220.798411]  entry_SYSCALL64_slow_path+0x25/0x25

Fix this by adding cond_resched every 1024 pages.

Signed-off-by: Yisheng Xie <xieyisheng1@huawei.com>
---
 mm/kmemleak.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/mm/kmemleak.c b/mm/kmemleak.c
index e4738d5..e9f2e86 100644
--- a/mm/kmemleak.c
+++ b/mm/kmemleak.c
@@ -1523,6 +1523,8 @@ static void kmemleak_scan(void)
 			if (page_count(page) == 0)
 				continue;
 			scan_block(page, page + 1, NULL);
+			if (!(pfn % 1024))
+				cond_resched();
 		}
 	}
 	put_online_mems();
-- 
1.7.12.4

Re: [PATCH] kmemcheck: add scheduling point to kmemleak_scan

[Catalin Marinas]

Please fix the subject as the tool is called "kmemleak" rather than
"kmemcheck".

On Fri, Nov 17, 2017 at 03:03:56PM +0800, Yisheng Xie wrote:
> diff --git a/mm/kmemleak.c b/mm/kmemleak.c
> index e4738d5..e9f2e86 100644
> --- a/mm/kmemleak.c
> +++ b/mm/kmemleak.c
> @@ -1523,6 +1523,8 @@ static void kmemleak_scan(void)
>  			if (page_count(page) == 0)
>  				continue;
>  			scan_block(page, page + 1, NULL);
> +			if (!(pfn % 1024))
> +				cond_resched();

For consistency with the other places where we call cond_resched() in
kmemleak, I would use MAX_SCAN_SIZE. Something like

			if (!(pfn % (MAX_SCAN_SIZE / sizeof(page))))
				cont_resched();

Thanks.

-- 
Catalin

Re: [PATCH] kmemcheck: add scheduling point to kmemleak_scan

[Yisheng Xie]

Hi Catalin,

On 2017/11/18 2:27, Catalin Marinas wrote:
> Please fix the subject as the tool is called "kmemleak" rather than
> "kmemcheck".

Yeah, this really is a terrible typo.

> 
> On Fri, Nov 17, 2017 at 03:03:56PM +0800, Yisheng Xie wrote:
>> diff --git a/mm/kmemleak.c b/mm/kmemleak.c
>> index e4738d5..e9f2e86 100644
>> --- a/mm/kmemleak.c
>> +++ b/mm/kmemleak.c
>> @@ -1523,6 +1523,8 @@ static void kmemleak_scan(void)
>>  			if (page_count(page) == 0)
>>  				continue;
>>  			scan_block(page, page + 1, NULL);
>> +			if (!(pfn % 1024))
>> +				cond_resched();
> 
> For consistency with the other places where we call cond_resched() in
> kmemleak, I would use MAX_SCAN_SIZE. Something like
> 
> 			if (!(pfn % (MAX_SCAN_SIZE / sizeof(page))))
> 				cont_resched();

Yes, this will keep it consistency with the other places.

I will take both of these suggestion in next version.

Thanks
Yisheng Xie
> 
> Thanks.
>