Matthew Wilcox <willy@xxxxxxxxxxxxx> writes:
> On Tue, Jan 01, 2019 at 08:57:53AM +0530, Aneesh Kumar K.V wrote:
>> Matthew Wilcox <willy@xxxxxxxxxxxxx> writes:
>> > +/* Returns the number of bytes in this potentially compound page. */
>> > +static inline unsigned long page_size(struct page *page)
>> > +{
>> > + return (unsigned long)PAGE_SIZE << compound_order(page);
>> > +}
>> > +
>>
>> How about compound_page_size() to make it clear this is for
>> compound_pages? Should we make it work with Tail pages by doing
>> compound_head(page)?
>
> I think that's a terrible idea. Actually, I think the whole way we handle
> compound pages is terrible; we should only ever see head pages. Doing
> page cache lookups should only give us head pages. Calling pfn_to_page()
> should give us the head page. We should only put head pages into SG lists.
> Everywhere you see a struct page should only be a head page.
>
> I know we're far from that today, and there's lots of work to be done
> to get there. But the current state of handling compound pages is awful
> and confusing.
One exception is THP compound pages which can also be mapped as regular
pages in some page tables?
-aneesh
