On Fri, Dec 07, 2012 at 02:11:02PM -0800, Andrew Morton wrote:
>On Fri, 7 Dec 2012 16:48:45 +0800
>Xishi Qiu <qiuxishi@xxxxxxxxxx> wrote:
>
>> On x86 platform, if we use "/sys/devices/system/memory/soft_offline_page" to offline a
>> free page twice, the value of mce_bad_pages will be added twice. So this is an error,
>> since the page was already marked HWPoison, we should skip the page and don't add the
>> value of mce_bad_pages.
>>
>> $ cat /proc/meminfo | grep HardwareCorrupted
>>
>> soft_offline_page()
>> get_any_page()
>> atomic_long_add(1, &mce_bad_pages)
>>
>> ...
>>
>> --- a/mm/memory-failure.c
>> +++ b/mm/memory-failure.c
>> @@ -1582,8 +1582,11 @@ int soft_offline_page(struct page *page, int flags)
>> return ret;
>>
>> done:
>> - atomic_long_add(1, &mce_bad_pages);
>> - SetPageHWPoison(page);
>> /* keep elevated page count for bad page */
>> + if (!PageHWPoison(page)) {
>> + atomic_long_add(1, &mce_bad_pages);
>> + SetPageHWPoison(page);
>> + }
>> +
>> return ret;
>> }
>
>A few things:
>
>- soft_offline_page() already checks for this case:
>
> if (PageHWPoison(page)) {
> unlock_page(page);
> put_page(page);
> pr_info("soft offline: %#lx page already poisoned\n", pfn);
> return -EBUSY;
> }
>
> so why didn't this check work for you?
>
> Presumably because one of the earlier "goto done" branches was
> taken. Which one, any why?
>
> This function is an utter mess. It contains six return points
> randomly intermingled with three "goto done" return points.
>
> This mess is probably the cause of the bug you have observed. Can
> we please fix it up somehow? It *seems* that the design (lol) of
> this function is "for errors, return immediately. For success, goto
> done". In which case "done" should have been called "success". But
> if you just look at the function you'll see that this approach didn't
> work. I suggest it be converted to have two return points - one for
> the success path, one for the failure path. Or something.
>
>- soft_offline_huge_page() is a miniature copy of soft_offline_page()
> and might suffer the same bug.
>
>- A cleaner, shorter and possibly faster implementation is
>
> if (!TestSetPageHWPoison(page))
> atomic_long_add(1, &mce_bad_pages);
>
Hi Andrew,
Since hwpoison bit for free buddy page has already be set in get_any_page,
!TestSetPageHWPoison(page) will not increase mce_bad_pages count even for
the first time.
Regards,
Wanpeng Li
>- We have atomic_long_inc(). Use it?
>
>- Why do we have a variable called "mce_bad_pages"? MCE is an x86
> concept, and this code is in mm/. Lights are flashing, bells are
> ringing and a loudspeaker is blaring "layering violation" at us!
>
>--
>To unsubscribe, send a message with 'unsubscribe linux-mm' in
>the body to majordomo@xxxxxxxxx. For more info on Linux MM,
>see: http://www.linux-mm.org/ .
>Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>