[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Wed, 9 Sep 2020 18:04:02 +0100
From: Matthew Wilcox <willy@...radead.org>
To: Christoph Hellwig <hch@...radead.org>
Cc: Ming Mao <maoming.maoming@...wei.com>,
linux-kernel@...r.kernel.org, kvm@...r.kernel.org,
linux-mm@...ck.org, alex.williamson@...hat.com,
akpm@...ux-foundation.org, cohuck@...hat.com,
jianjay.zhou@...wei.com, weidong.huang@...wei.com,
peterx@...hat.com, aarcange@...hat.com, wangyunjian@...wei.com,
jhubbard@...dia.com
Subject: Re: [PATCH V4 1/2] vfio dma_map/unmap: optimized for hugetlbfs pages
On Wed, Sep 09, 2020 at 09:01:14AM +0100, Christoph Hellwig wrote:
> (1) a FOLL_HUGEPAGE flag for the pin_user_pages API family that returns
> a single struct page for any kind of huge page, which would also
> benefit all kinds of other users rather than adding these kinds of
> hacks to vfio.
> (2) add a bvec version of the API that returns a variable size
> "extent"
>
> I had started on (2) a while ago, and here is branch with my code (which
> is broken and fails test, but might be a start):
>
> http://git.infradead.org/users/hch/misc.git/shortlog/refs/heads/gup-bvec
>
> But for now I wonder if (1) is the better start, which could still be
> reused to (2) later.
(1) doesn't work. Suppose you have a 16kB MAP_PRIVATE of a file which happens
to have a THP in the page cache. When you write a byte, that base page gets
copied and your bvec for that 16kB needs to look something like:
{ page A, offset 0, length 4096 },
{ page B, offset 0, length 4096 },
{ page A, offset 8192, length 8192 },
You can come up with other scenarios, like partial mappings of THPs with
an adjacent VMA of a different THP, but I think we just need to make
(2) work.
Powered by blists - more mailing lists