Re: [PATCH] mm: bio_alloc never fails when set GFP_NOIO, GFP_KERNEL

[Jens Axboe <axboe@kernel.dk>]

On 10/25/20 1:09 PM, Andrew Morton wrote:
> On Wed, 21 Oct 2020 11:11:28 +0800 Xianting Tian <tian.xianting@h3c.com> wrote:
> 
>> bio_alloc with __GFP_DIRECT_RECLAIM(which is included in GFP_NOIO,
>> GFP_KERNEL) never fails, as stated in the comments of bio_alloc_bioset.
>>
>> So we can remove multiple unneeded null checks of bio_alloc and simplify
>> the code.
>>
>> We have done it in fs/ext4/readpage.c, fs/ext4/page-io.c, fs/direct-io.c,
>> and so forth.
>>
> 
> (cc Jens)
> 
>> --- a/mm/page_io.c
>> +++ b/mm/page_io.c
>> @@ -30,18 +30,20 @@ static struct bio *get_swap_bio(gfp_t gfp_flags,
>>  				struct page *page, bio_end_io_t end_io)
>>  {
>>  	struct bio *bio;
>> +	struct block_device *bdev;
>>  
>> +	/*
>> +	 * bio_alloc will _always_ be able to allocate a bio if
>> +	 * __GFP_DIRECT_RECLAIM is set, see comments for bio_alloc_bioset().
>> +	 */
>>  	bio = bio_alloc(gfp_flags, 1);
>> -	if (bio) {
>> -		struct block_device *bdev;
>> +	bio->bi_iter.bi_sector = map_swap_page(page, &bdev);
>> +	bio_set_dev(bio, bdev);
>> +	bio->bi_iter.bi_sector <<= PAGE_SHIFT - 9;
>> +	bio->bi_end_io = end_io;
>>  
>> -		bio->bi_iter.bi_sector = map_swap_page(page, &bdev);
>> -		bio_set_dev(bio, bdev);
>> -		bio->bi_iter.bi_sector <<= PAGE_SHIFT - 9;
>> -		bio->bi_end_io = end_io;
>> +	bio_add_page(bio, page, thp_size(page), 0);
>>  
>> -		bio_add_page(bio, page, thp_size(page), 0);
>> -	}
>>  	return bio;
>>  }
>>  
>> @@ -351,19 +353,13 @@ int __swap_writepage(struct page *page, struct writeback_control *wbc,
>>  
>>  	ret = 0;
>>  	bio = get_swap_bio(GFP_NOIO, page, end_write_func);
>> -	if (bio == NULL) {
>> -		set_page_dirty(page);
>> -		unlock_page(page);
>> -		ret = -ENOMEM;
>> -		goto out;
>> -	}
>>  	bio->bi_opf = REQ_OP_WRITE | REQ_SWAP | wbc_to_write_flags(wbc);
>>  	bio_associate_blkg_from_page(bio, page);
>>  	count_swpout_vm_event(page);
>>  	set_page_writeback(page);
>>  	unlock_page(page);
>>  	submit_bio(bio);
>> -out:
>> +
>>  	return ret;
>>  }
>>  
>> @@ -416,11 +412,6 @@ int swap_readpage(struct page *page, bool synchronous)
>>  
>>  	ret = 0;
>>  	bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read);
>> -	if (bio == NULL) {
>> -		unlock_page(page);
>> -		ret = -ENOMEM;
>> -		goto out;
>> -	}
>>  	disk = bio->bi_disk;
>>  	/*
>>  	 * Keep this task valid during swap readpage because the oom killer may
> 
> I'm reluctant to remove these checks - yours is a fairly subtle
> discovery and things might change in the future.  Intentionally or
> otherwise!

On the block/bio side, it all boils down to the mempool backing of the
bio allocations. If __GFP_DIRECT_RECLAIM is set for that, then we'll
always wait on allocations for a bio. That is intentional, by design, to
guarantee forward progress. It used to be __GFP_WAIT based, but I guess
that changed at some point...

-- 
Jens Axboe