On 13 Jan 2020, at 12:40, Matthew Wilcox wrote:
> On Mon, Jan 13, 2020 at 04:42:10PM +0000, Chris Mason wrote:
>
> I did do a couple of helpers for lists for iomap before deciding the
> whole thing was too painful. I didn't look at btrfs until just now,
> but, um ...
>
> int extent_readpages(struct address_space *mapping, struct list_head
> *pages,
> unsigned nr_pages)
> ..
> struct page *pagepool[16];
> ..
> while (!list_empty(pages)) {
> ..
> list_del(&page->lru);
> if (add_to_page_cache_lru(page, mapping,
> page->index,
> ..
> pagepool[nr++] = page;
>
> you're basically doing exactly what i'm proposing to be the new
> interface!
> OK, you get one extra page per batch ;-P
This is true, I didn't explain that part well ;) Depending on
compression etc we might end up poking the xarray inside the actual IO
functions, but the main difference is that btrfs is building a single
bio. You're moving the plug so you'll merge into single bio, but I'd
rather build 2MB bios than merge them.
I guess it doesn't feel like enough of a win to justify the churn. If
we find a way to do much larger pagevecs, I think this makes more sense.
-chris
