[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CABYd82biLQ7x1zD0MibXMrMHD9Y-cejitbNo=M2=tPNfUMTePA@mail.gmail.com>
Date: Tue, 1 Feb 2022 10:00:06 -0800
From: Will McVicker <willmcvicker@...gle.com>
To: John Hubbard <jhubbard@...dia.com>
Cc: Andrew Morton <akpm@...ux-foundation.org>,
LKML <linux-kernel@...r.kernel.org>, linux-mm@...ck.org,
Christoph Hellwig <hch@....de>,
Minchan Kim <minchan@...gle.com>,
Matthew Wilcox <willy@...radead.org>,
Christian Borntraeger <borntraeger@...ibm.com>,
Heiko Carstens <hca@...ux.ibm.com>,
Vasily Gorbik <gor@...ux.ibm.com>,
Linus Torvalds <torvalds@...ux-foundation.org>
Subject: Re: [PATCH] Revert mm/gup: small refactoring: simplify try_grab_page()
On Tue, Feb 1, 2022 at 1:29 AM John Hubbard <jhubbard@...dia.com> wrote:
>
> This reverts commit 54d516b1d62ff8f17cee2da06e5e4706a0d00b8a
>
> That commit did a refactoring that effectively combined fast and slow
> gup paths (again). And that was again incorrect, for two reasons:
>
> a) Fast gup and slow gup get reference counts on pages in different ways
> and with different goals: see Linus' writeup in commit cd1adf1b63a1
> ("Revert "mm/gup: remove try_get_page(), call try_get_compound_head()
> directly""), and
>
> b) try_grab_compound_head() also has a specific check for "FOLL_LONGTERM
> && !is_pinned(page)", that assumes that the caller can fall back to slow
> gup. This resulted in new failures, as recently report by Will McVicker
> [1].
>
> But (a) has problems too, even though they may not have been reported
> yet. So just revert this.
>
> [1] https://lore.kernel.org/r/20220131203504.3458775-1-willmcvicker@google.com
>
> Fixes: 54d516b1d62f ("mm/gup: small refactoring: simplify try_grab_page()")
> Cc: Christoph Hellwig <hch@....de>
> Cc: Will McVicker <willmcvicker@...gle.com>
> Cc: Minchan Kim <minchan@...gle.com>
> Cc: Matthew Wilcox <willy@...radead.org>
> Cc: Christian Borntraeger <borntraeger@...ibm.com>
> Cc: Heiko Carstens <hca@...ux.ibm.com>
> Cc: Vasily Gorbik <gor@...ux.ibm.com>
> Cc: Linus Torvalds <torvalds@...ux-foundation.org>
> Signed-off-by: John Hubbard <jhubbard@...dia.com>
> ---
> mm/gup.c | 35 ++++++++++++++++++++++++++++++-----
> 1 file changed, 30 insertions(+), 5 deletions(-)
>
> diff --git a/mm/gup.c b/mm/gup.c
> index f0af462ac1e2..a9d4d724aef7 100644
> --- a/mm/gup.c
> +++ b/mm/gup.c
> @@ -124,8 +124,8 @@ static inline struct page *try_get_compound_head(struct page *page, int refs)
> * considered failure, and furthermore, a likely bug in the caller, so a warning
> * is also emitted.
> */
> -struct page *try_grab_compound_head(struct page *page,
> - int refs, unsigned int flags)
> +__maybe_unused struct page *try_grab_compound_head(struct page *page,
> + int refs, unsigned int flags)
> {
> if (flags & FOLL_GET)
> return try_get_compound_head(page, refs);
> @@ -208,10 +208,35 @@ static void put_compound_head(struct page *page, int refs, unsigned int flags)
> */
> bool __must_check try_grab_page(struct page *page, unsigned int flags)
> {
> - if (!(flags & (FOLL_GET | FOLL_PIN)))
> - return true;
> + WARN_ON_ONCE((flags & (FOLL_GET | FOLL_PIN)) == (FOLL_GET | FOLL_PIN));
>
> - return try_grab_compound_head(page, 1, flags);
> + if (flags & FOLL_GET)
> + return try_get_page(page);
> + else if (flags & FOLL_PIN) {
> + int refs = 1;
> +
> + page = compound_head(page);
> +
> + if (WARN_ON_ONCE(page_ref_count(page) <= 0))
> + return false;
> +
> + if (hpage_pincount_available(page))
> + hpage_pincount_add(page, 1);
> + else
> + refs = GUP_PIN_COUNTING_BIAS;
> +
> + /*
> + * Similar to try_grab_compound_head(): even if using the
> + * hpage_pincount_add/_sub() routines, be sure to
> + * *also* increment the normal page refcount field at least
> + * once, so that the page really is pinned.
> + */
> + page_ref_add(page, refs);
> +
> + mod_node_page_state(page_pgdat(page), NR_FOLL_PIN_ACQUIRED, 1);
> + }
> +
> + return true;
> }
>
> /**
>
> base-commit: 26291c54e111ff6ba87a164d85d4a4e134b7315c
> --
> 2.35.1
>
Thanks John! I verified this works on the Pixel 6 with the 5.15 kernel
for my camera use-case. Free free to include:
Tested-by: Will McVicker <willmcvicker@...gle.com>
Thanks,
Will
Powered by blists - more mailing lists