On Wed, Jun 13, 2018 at 10:54:09AM -0400, Kent Overstreet wrote: > On Wed, Jun 13, 2018 at 03:56:32PM +0200, Christoph Hellwig wrote: > > On Wed, Jun 13, 2018 at 07:06:41PM +0800, Ming Lei wrote: > > > > before bio_alloc_pages) that can be switched to something that just creates a > > > > single bvec. > > > > > > Yes, multipage bvec shouldn't break any driver or fs. > > > > It probably isn't broken, at least I didn't see assumptions of the same > > number of segments. However the current poking into the bio internals as > > a bad idea for a couple of reasons. First because it requires touching > > bcache for any of these changes, second because it won't get merging of > > pages into a single bio segment for bіos built by bch_bio_map or > > bch_bio_alloc_pages, and third bcache is the last user of > > bio_for_each_chunk_all in your branch, which I'd like to kill off to > > keep the number of iterators down. > > Agreed about bio_for_each_chunk_all(), but I just looked at the patch that > introduces them and it looks to me like there's no need, they should just be > bio_for_each_segment_all(). Now we can't change the vector with bio_for_each_segment_all(), so bio_for_each_chunk_all() has to be used. So looks it makes sense to use bio_add_page() to remove bio_for_each_chunk_all(). Thanks, Ming
