Re: [PATCH v8 1/5] block: Added folio-ized version of bvec_try_merge_hw_page()

[Matthew Wilcox <willy@infradead.org>]

On Thu, Jul 11, 2024 at 10:37:46AM +0530, Kundan Kumar wrote:
> -bool bvec_try_merge_hw_page(struct request_queue *q, struct bio_vec *bv,
> -		struct page *page, unsigned len, unsigned offset,
> +bool bvec_try_merge_hw_folio(struct request_queue *q, struct bio_vec *bv,
> +		struct folio *folio, size_t len, size_t offset,
>  		bool *same_page)
>  {
> +	struct page *page = folio_page(folio, 0);
>  	unsigned long mask = queue_segment_boundary(q);
>  	phys_addr_t addr1 = bvec_phys(bv);
>  	phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
[...]
> +bool bvec_try_merge_hw_page(struct request_queue *q, struct bio_vec *bv,
> +		struct page *page, unsigned int len, unsigned int offset,
> +		bool *same_page)
> +{
> +	struct folio *folio = page_folio(page);
> +
> +	return bvec_try_merge_hw_folio(q, bv, folio, len,
> +			((size_t)folio_page_idx(folio, page) << PAGE_SHIFT) +
> +			offset, same_page);
> +}

This is the wrong way to do it.  bio_add_folio() does it correctly
by being a wrapper around bio_add_page().

The reason is that in the future, not all pages will belong to folios.
For those pages, page_folio() will return NULL, and this will crash.