Re: [PATCH V3 1/2] mm: Add get_user_pages_cma_migrate

[Aneesh Kumar K.V]

Alexey Kardashevskiy <aik@xxxxxxxxx> writes:

> On 18/09/2018 21:58, Aneesh Kumar K.V wrote:
>> This helper does a get_user_pages_fast and if it find pages in the CMA area
>> it will try to migrate them before taking page reference. This makes sure that
>> we don't keep non-movable pages (due to page reference count) in the CMA area.
>> Not able to move pages out of CMA area result in CMA allocation failures.
>> 
>> Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@xxxxxxxxxxxxx>
>> ---
>>  include/linux/hugetlb.h |   2 +
>>  include/linux/migrate.h |   3 +
>>  mm/hugetlb.c            |   4 +-
>>  mm/migrate.c            | 132 ++++++++++++++++++++++++++++++++++++++++
>>  4 files changed, 139 insertions(+), 2 deletions(-)
>> 
>> diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
>> index 6b68e345f0ca..1abccb1a1ecc 100644
>> --- a/include/linux/hugetlb.h
>> +++ b/include/linux/hugetlb.h
>> @@ -357,6 +357,8 @@ struct page *alloc_huge_page_nodemask(struct hstate *h, int preferred_nid,
>>  				nodemask_t *nmask);
>>  struct page *alloc_huge_page_vma(struct hstate *h, struct vm_area_struct *vma,
>>  				unsigned long address);
>> +struct page *alloc_migrate_huge_page(struct hstate *h, gfp_t gfp_mask,
>> +				     int nid, nodemask_t *nmask);
>>  int huge_add_to_page_cache(struct page *page, struct address_space *mapping,
>>  			pgoff_t idx);
>>  
>> diff --git a/include/linux/migrate.h b/include/linux/migrate.h
>> index f2b4abbca55e..d82b35afd2eb 100644
>> --- a/include/linux/migrate.h
>> +++ b/include/linux/migrate.h
>> @@ -286,6 +286,9 @@ static inline int migrate_vma(const struct migrate_vma_ops *ops,
>>  }
>>  #endif /* IS_ENABLED(CONFIG_MIGRATE_VMA_HELPER) */
>>  
>> +extern int get_user_pages_cma_migrate(unsigned long start, int nr_pages, int write,
>> +				      struct page **pages);
>> +
>>  #endif /* CONFIG_MIGRATION */
>>  
>>  #endif /* _LINUX_MIGRATE_H */
>> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
>> index 3c21775f196b..1abbfcb84f66 100644
>> --- a/mm/hugetlb.c
>> +++ b/mm/hugetlb.c
>> @@ -1585,8 +1585,8 @@ static struct page *alloc_surplus_huge_page(struct hstate *h, gfp_t gfp_mask,
>>  	return page;
>>  }
>>  
>> -static struct page *alloc_migrate_huge_page(struct hstate *h, gfp_t gfp_mask,
>> -		int nid, nodemask_t *nmask)
>> +struct page *alloc_migrate_huge_page(struct hstate *h, gfp_t gfp_mask,
>> +				     int nid, nodemask_t *nmask)
>>  {
>>  	struct page *page;
>>  
>> diff --git a/mm/migrate.c b/mm/migrate.c
>> index d6a2e89b086a..2f92534ea7a1 100644
>> --- a/mm/migrate.c
>> +++ b/mm/migrate.c
>> @@ -3006,3 +3006,135 @@ int migrate_vma(const struct migrate_vma_ops *ops,
>>  }
>>  EXPORT_SYMBOL(migrate_vma);
>>  #endif /* defined(MIGRATE_VMA_HELPER) */
>> +
>> +static struct page *new_non_cma_page(struct page *page, unsigned long private)
>> +{
>> +	/*
>> +	 * We want to make sure we allocate the new page from the same node
>> +	 * as the source page.
>> +	 */
>> +	int nid = page_to_nid(page);
>> +	gfp_t gfp_mask = GFP_USER | __GFP_THISNODE;
>> +
>> +	if (PageHighMem(page))
>> +		gfp_mask |= __GFP_HIGHMEM;
>> +
>> +	if (PageTransHuge(page)) {
>> +		struct page *thp;
>> +		gfp_t thp_gfpmask = GFP_TRANSHUGE | __GFP_THISNODE;
>> +
>> +		/*
>> +		 * Remove the movable mask so that we don't allocate from
>> +		 * CMA area again.
>> +		 */
>> +		thp_gfpmask &= ~__GFP_MOVABLE;
>> +		thp = __alloc_pages_node(nid, thp_gfpmask, HPAGE_PMD_ORDER);
>
>
> HPAGE_PMD_ORDER is 2MB or 1GB? THP are always that PMD order?

2M or 16M. THP is at PMD level.

>
>
>> +		if (!thp)
>> +			return NULL;
>> +		prep_transhuge_page(thp);
>> +		return thp;
>> +
>> +#ifdef  CONFIG_HUGETLB_PAGE
>> +	} else if (PageHuge(page)) {
>> +
>> +		struct hstate *h = page_hstate(page);
>> +		/*
>> +		 * We don't want to dequeue from the pool because pool pages will
>> +		 * mostly be from the CMA region.
>> +		 */
>> +		return alloc_migrate_huge_page(h, gfp_mask, nid, NULL);
>> +#endif
>> +	}
>> +
>> +	return __alloc_pages_node(nid, gfp_mask, 0);
>> +}
>> +
>> +/**
>> + * get_user_pages_cma_migrate() - pin user pages in memory by migrating pages in CMA region
>> + * @start:	starting user address
>> + * @nr_pages:	number of pages from start to pin
>> + * @write:	whether pages will be written to
>> + * @pages:	array that receives pointers to the pages pinned.
>> + *		Should be at least nr_pages long.
>> + *
>> + * Attempt to pin user pages in memory without taking mm->mmap_sem.
>> + * If not successful, it will fall back to taking the lock and
>> + * calling get_user_pages().
>
>
> I do not see any locking or get_user_pages(), hidden somewhere?
>

The rules are same as get_user_pages_fast, which does that pin attempt
without taking mm->mmap_sem. If it fail get_user_pages_fast will take
the mmap_sem and try to pin the pages. The details are in
get_user_pages_fast. You can look at get_user_pages_unlocked


>> + *
>> + * If the pinned pages are backed by CMA region, we migrate those pages out,
>> + * allocating new pages from non-CMA region. This helps in avoiding keeping
>> + * pages pinned in the CMA region for a long time thereby resulting in
>> + * CMA allocation failures.
>> + *
>> + * Returns number of pages pinned. This may be fewer than the number
>> + * requested. If nr_pages is 0 or negative, returns 0. If no pages
>> + * were pinned, returns -errno.
>> + */
>> +
>> +int get_user_pages_cma_migrate(unsigned long start, int nr_pages, int write,
>> +			       struct page **pages)
>> +{
>> +	int i, ret;
>> +	bool drain_allow = true;
>> +	bool migrate_allow = true;
>> +	LIST_HEAD(cma_page_list);
>> +
>> +get_user_again:
>> +	ret = get_user_pages_fast(start, nr_pages, write, pages);
>> +	if (ret <= 0)
>> +		return ret;
>> +
>> +	for (i = 0; i < ret; ++i) {
>> +		/*
>> +		 * If we get a page from the CMA zone, since we are going to
>> +		 * be pinning these entries, we might as well move them out
>> +		 * of the CMA zone if possible.
>> +		 */
>> +		if (is_migrate_cma_page(pages[i]) && migrate_allow) {
>> +			if (PageHuge(pages[i]))
>> +				isolate_huge_page(pages[i], &cma_page_list);
>> +			else {
>> +				struct page *head = compound_head(pages[i]);
>> +
>> +				if (!PageLRU(head) && drain_allow) {
>> +					lru_add_drain_all();
>> +					drain_allow = false;
>> +				}
>> +
>> +				if (!isolate_lru_page(head)) {
>> +					list_add_tail(&head->lru, &cma_page_list);
>> +					mod_node_page_state(page_pgdat(head),
>> +							    NR_ISOLATED_ANON +
>> +							    page_is_file_cache(head),
>> +							    hpage_nr_pages(head));
>
>
> Above 10 lines I cannot really comment due to my massive ignorance in
> this area, especially about what lru_add_drain_all() and
> mod_node_page_state() :(

That makes sure we move the pages from per cpu lru vec and add them to
the right lru list so that we can isolate the pages correctly.

>
>
>> +				}
>> +			}
>> +		}
>> +	}
>> +	if (!list_empty(&cma_page_list)) {
>> +		/*
>> +		 * drop the above get_user_pages reference.
>> +		 */
>> +		for (i = 0; i < ret; ++i)
>> +			put_page(pages[i]);
>> +
>> +		if (migrate_pages(&cma_page_list, new_non_cma_page,
>> +				  NULL, 0, MIGRATE_SYNC, MR_CONTIG_RANGE)) {
>> +			/*
>> +			 * some of the pages failed migration. Do get_user_pages
>> +			 * without migration.
>> +			 */
>> +			migrate_allow = false;
>
>
> migrate_allow seems useless, simply calling get_user_pages_fast() should
> make the code easier to read imho. And the comment says
> get_user_pages(), where does this guy hide?

I didn't get that suggestion. What we want to do here is try to migrate pages as
long as we find CMA pages in the result of get_user_pages_fast. If we
failed any migration attempt, don't try to migrate again.

>
>> +
>> +			if (!list_empty(&cma_page_list))
>> +				putback_movable_pages(&cma_page_list);
>> +		}
>> +		/*
>> +		 * We did migrate all the pages, Try to get the page references again
>> +		 * migrating any new CMA pages which we failed to isolate earlier.
>> +		 */
>> +		drain_allow = true;
>
> Move this "drain_allow = true" right after "get_user_again:"? 1
 
>
>
>> +		goto get_user_again;
>> +	}
>> +	return ret;
>> +}
>> 
>
> -- 
> Alexey

-aneesh