Re: [PATCH] /proc/kpagecount: return 0 for special pages that are never mapped

[anthony.yznaga@oracle.com]

On 12/04/2018 04:48 PM, Matthew Wilcox wrote:
> On Tue, Dec 04, 2018 at 02:45:26PM -0800, Anthony Yznaga wrote:
>> Certain pages that are never mapped to userspace have a type
>> indicated in the page_type field of their struct pages (e.g. PG_buddy).
>> page_type overlaps with _mapcount so set the count to 0 and avoid
>> calling page_mapcount() for these pages.
>>
>> Signed-off-by: Anthony Yznaga <anthony.yznaga@oracle.com>
>> ---
>>  fs/proc/page.c             | 2 +-
>>  include/linux/page-flags.h | 7 +++++++
>>  2 files changed, 8 insertions(+), 1 deletion(-)
>>
>> diff --git a/fs/proc/page.c b/fs/proc/page.c
>> index 6c517b11acf8..40b05e0d4274 100644
>> --- a/fs/proc/page.c
>> +++ b/fs/proc/page.c
>> @@ -46,7 +46,7 @@ static ssize_t kpagecount_read(struct file *file, char __user *buf,
>>  			ppage = pfn_to_page(pfn);
>>  		else
>>  			ppage = NULL;
>> -		if (!ppage || PageSlab(ppage))
>> +		if (!ppage || PageSlab(ppage) || page_has_type(ppage))
>>  			pcount = 0;
>>  		else
>>  			pcount = page_mapcount(ppage);
>> diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
>> index 50ce1bddaf56..f9a1c50ccefc 100644
>> --- a/include/linux/page-flags.h
>> +++ b/include/linux/page-flags.h
>> @@ -673,10 +673,17 @@ static inline int TestClearPageDoubleMap(struct page *page)
>>  #define PG_balloon	0x00000100
>>  #define PG_kmemcg	0x00000200
>>  #define PG_table	0x00000400
>> +#define PAGE_TYPE_ALL	(PG_buddy | PG_balloon | PG_kmemcg | PG_table)
>>  
>>  #define PageType(page, flag)						\
>>  	((page->page_type & (PAGE_TYPE_BASE | flag)) == PAGE_TYPE_BASE)
>>  
>> +static inline int page_has_type(struct page *page)
>> +{
>> +	return (PageType(page, 0) &&
>> +	       ((page->page_type & PAGE_TYPE_ALL) != PAGE_TYPE_ALL));
>> +}
>> +
>>  #define PAGE_TYPE_OPS(uname, lname)					\
> I think this is a bit complex, and a bit of a pain to update as we add
> new page types.  How about this?
>
> 	return (int)page_type < -128;
>
> (I'm open to appropriate #defines to make this more obvious that it's ~0x7F)


I thought about having this:

#define PAGE_TYPE_END    0xffffff80

static int inline page_has_type(struct page *page)
{
    return page->page_type > PAGE_TYPE_BASE &&
           page->page_type < PAGE_TYPE_END;
}

But I opted for the additional complexity to avoid more false-positives from
possibly corrupted values.  I'm certainly fine with a simple approach, though.