On 07.05.22 21:23, Andrew Morton wrote:
> On Wed, 4 May 2022 23:44:29 -0700 Minchan Kim <minchan@xxxxxxxxxx> wrote:
>
>> Pages on CMA area could have MIGRATE_ISOLATE as well as MIGRATE_CMA
>> so current is_pinnable_page could miss CMA pages which has MIGRATE_
>> ISOLATE. It ends up putting CMA pages longterm pinning possible on
>> pin_user_pages APIs so CMA allocation fails.
>>
>> The CMA allocation path protects the migration type change race
>> using zone->lock but what GUP path need to know is just whether the
>> page is on CMA area or not rather than exact type. Thus, we don't
>> need zone->lock but just checks the migratype in either of
>> (MIGRATE_ISOLATE and MIGRATE_CMA).
>>
>> Adding the MIGRATE_ISOLATE check in is_pinnable_page could cause
>> rejecting of pinning the page on MIGRATE_ISOLATE pageblock even
>> thouth it's neither CMA nor movable zone if the page is temporarily
>
> "though"
>
>> unmovable. However, the migration failure is general issue, not
>> only come from MIGRATE_ISOLATE and the MIGRATE_ISOLATE is also
>> transient state like other temporal refcount holding of pages.
>>
>> ...
>>
>> --- a/include/linux/mm.h
>> +++ b/include/linux/mm.h
>> @@ -1625,8 +1625,18 @@ static inline bool page_needs_cow_for_dma(struct vm_area_struct *vma,
>> #ifdef CONFIG_MIGRATION
>> static inline bool is_pinnable_page(struct page *page)
>> {
>> - return !(is_zone_movable_page(page) || is_migrate_cma_page(page)) ||
>> - is_zero_pfn(page_to_pfn(page));
>> +#ifdef CONFIG_CMA
>> + /*
>> + * use volatile to use local variable mt instead of
>> + * refetching mt value.
>> + */
>> + volatile int mt = get_pageblock_migratetype(page);
>> +
>> + if (mt == MIGRATE_CMA || mt == MIGRATE_ISOLATE)
>> + return false;
>> +#endif
>
> Open-coded use of `volatile' draws unwelcome attention.
>
> What are we trying to do here? Prevent the compiler from rerunning all
> of get_pageblock_migratetype() (really __get_pfnblock_flags_mask())
> twice? That would be pretty dumb of it?
>
> Would a suitably-commented something like
>
> int __mt = get_pageblock_migratetype(page);
> int mt = __READ_ONCE(__mt);
>
> express this better?
Yes, we want READ_ONCE I think. Apart from that LGTM.
--
Thanks,
David / dhildenb
