From: Matthew Wilcox <willy@infradead.org>
To: Jesper Dangaard Brouer <jbrouer@redhat.com>
Cc: Jesper Dangaard Brouer <hawk@kernel.org>,
Ilias Apalodimas <ilias.apalodimas@linaro.org>,
brouer@redhat.com, netdev@vger.kernel.org, linux-mm@kvack.org,
Shakeel Butt <shakeelb@google.com>
Subject: Re: [PATCH v2 05/24] page_pool: Start using netmem in allocation path.
Date: Fri, 6 Jan 2023 15:36:02 +0000 [thread overview]
Message-ID: <Y7g/4vLdpzGOxAKv@casper.infradead.org> (raw)
In-Reply-To: <aa334df4-e362-a6d6-87bf-fd6be16023ec@redhat.com>
On Fri, Jan 06, 2023 at 02:59:30PM +0100, Jesper Dangaard Brouer wrote:
> On 05/01/2023 22.46, Matthew Wilcox (Oracle) wrote:
> > Convert __page_pool_alloc_page_order() and __page_pool_alloc_pages_slow()
> > to use netmem internally. This removes a couple of calls
> > to compound_head() that are hidden inside put_page().
> > Convert trace_page_pool_state_hold(), page_pool_dma_map() and
> > page_pool_set_pp_info() to take a netmem argument.
> >
> > Saves 83 bytes of text in __page_pool_alloc_page_order() and 98 in
> > __page_pool_alloc_pages_slow() for a total of 181 bytes.
> >
> > Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
> > ---
> > include/trace/events/page_pool.h | 14 +++++------
> > net/core/page_pool.c | 42 +++++++++++++++++---------------
> > 2 files changed, 29 insertions(+), 27 deletions(-)
>
> Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
>
> Question below.
>
> > diff --git a/net/core/page_pool.c b/net/core/page_pool.c
> > index 437241aba5a7..4e985502c569 100644
> > --- a/net/core/page_pool.c
> > +++ b/net/core/page_pool.c
> [...]
> > @@ -421,7 +422,8 @@ static struct page *__page_pool_alloc_pages_slow(struct page_pool *pool,
> > page = NULL;
> > }
> > - /* When page just alloc'ed is should/must have refcnt 1. */
> > + /* When page just allocated it should have refcnt 1 (but may have
> > + * speculative references) */
> > return page;
>
> What does it mean page may have speculative references ?
>
> And do I/we need to worry about that for page_pool?
An excellent question. There are two code paths (known to me) which
take speculative references on a page, and there may well be more. One
is in GUP and the other is in the page cache. Both take the form of:
rcu_read_lock();
again:
look-up-page
try-get-page-ref
check-lookup
if lookup-failed drop-page-ref; goto again;
rcu_read_unlock();
If a page _has been_ in the page tables, then GUP can find it. If a
page _has been_ in the page cache, then filemap can find it. Because
there's no RCU grace period between freeing and reallocating a page, it
actually means that any page can see its refcount temporarily raised.
Usually the biggest problem is consumers assuming that they will be the
last code to call put_page() / folio_put(), and can do their cleanup
at that time (when the last caller of folio_put() may be GUP or filemap
which knows nothing of what you're using the page for).
I didn't notice any problems with temporarily elevated refcounts while
doing the netmem conversion, and it's something I'm fairly sensitive to,
so I think you got it all right and there is no need to be concerned.
Hope that's helpful!
next prev parent reply other threads:[~2023-01-06 15:36 UTC|newest]
Thread overview: 84+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-01-05 21:46 [PATCH v2 00/24] Split netmem from struct page Matthew Wilcox (Oracle)
2023-01-05 21:46 ` [PATCH v2 01/24] netmem: Create new type Matthew Wilcox (Oracle)
2023-01-06 13:07 ` Jesper Dangaard Brouer
2023-01-09 17:20 ` Ilias Apalodimas
2023-01-05 21:46 ` [PATCH v2 02/24] netmem: Add utility functions Matthew Wilcox (Oracle)
2023-01-06 2:24 ` kernel test robot
2023-01-06 20:35 ` Matthew Wilcox
2023-01-06 13:35 ` Jesper Dangaard Brouer
2023-01-05 21:46 ` [PATCH v2 03/24] page_pool: Add netmem_set_dma_addr() and netmem_get_dma_addr() Matthew Wilcox (Oracle)
2023-01-06 13:43 ` Jesper Dangaard Brouer
2023-01-09 17:30 ` Ilias Apalodimas
2023-01-10 9:17 ` Ilias Apalodimas
2023-01-10 18:16 ` Matthew Wilcox
2023-01-10 18:15 ` Matthew Wilcox
2023-01-05 21:46 ` [PATCH v2 04/24] page_pool: Convert page_pool_release_page() to page_pool_release_netmem() Matthew Wilcox (Oracle)
2023-01-06 13:46 ` Jesper Dangaard Brouer
2023-01-10 9:28 ` Ilias Apalodimas
2023-01-10 18:47 ` Matthew Wilcox
2023-01-11 13:56 ` Ilias Apalodimas
2023-01-05 21:46 ` [PATCH v2 05/24] page_pool: Start using netmem in allocation path Matthew Wilcox (Oracle)
2023-01-06 2:34 ` kernel test robot
2023-01-06 13:59 ` Jesper Dangaard Brouer
2023-01-06 15:36 ` Matthew Wilcox [this message]
2023-01-10 9:30 ` Ilias Apalodimas
2023-01-05 21:46 ` [PATCH v2 06/24] page_pool: Convert page_pool_return_page() to page_pool_return_netmem() Matthew Wilcox (Oracle)
2023-01-06 14:10 ` Jesper Dangaard Brouer
2023-01-10 9:39 ` Ilias Apalodimas
2023-01-05 21:46 ` [PATCH v2 07/24] page_pool: Convert __page_pool_put_page() to __page_pool_put_netmem() Matthew Wilcox (Oracle)
2023-01-06 14:14 ` Jesper Dangaard Brouer
2023-01-10 9:47 ` Ilias Apalodimas
2023-01-05 21:46 ` [PATCH v2 08/24] page_pool: Convert pp_alloc_cache to contain netmem Matthew Wilcox (Oracle)
2023-01-06 14:18 ` Jesper Dangaard Brouer
2023-01-10 9:58 ` Ilias Apalodimas
2023-01-05 21:46 ` [PATCH v2 09/24] page_pool: Convert page_pool_defrag_page() to page_pool_defrag_netmem() Matthew Wilcox (Oracle)
2023-01-06 14:29 ` Jesper Dangaard Brouer
2023-01-10 10:27 ` Ilias Apalodimas
2023-01-05 21:46 ` [PATCH v2 10/24] page_pool: Convert page_pool_put_defragged_page() to netmem Matthew Wilcox (Oracle)
2023-01-06 14:32 ` Jesper Dangaard Brouer
2023-01-10 10:36 ` Ilias Apalodimas
2023-01-05 21:46 ` [PATCH v2 11/24] page_pool: Convert page_pool_empty_ring() to use netmem Matthew Wilcox (Oracle)
2023-01-06 15:22 ` Jesper Dangaard Brouer
2023-01-10 10:38 ` Ilias Apalodimas
2023-01-05 21:46 ` [PATCH v2 12/24] page_pool: Convert page_pool_alloc_pages() to page_pool_alloc_netmem() Matthew Wilcox (Oracle)
2023-01-06 15:27 ` Jesper Dangaard Brouer
2023-01-10 10:45 ` Ilias Apalodimas
2023-01-05 21:46 ` [PATCH v2 13/24] page_pool: Convert page_pool_dma_sync_for_device() to take a netmem Matthew Wilcox (Oracle)
2023-01-06 15:28 ` Jesper Dangaard Brouer
2023-01-10 10:47 ` Ilias Apalodimas
2023-01-05 21:46 ` [PATCH v2 14/24] page_pool: Convert page_pool_recycle_in_cache() to netmem Matthew Wilcox (Oracle)
2023-01-06 15:29 ` Jesper Dangaard Brouer
2023-01-10 10:48 ` Ilias Apalodimas
2023-01-05 21:46 ` [PATCH v2 15/24] page_pool: Remove page_pool_defrag_page() Matthew Wilcox (Oracle)
2023-01-06 15:29 ` Jesper Dangaard Brouer
2023-01-10 9:47 ` Ilias Apalodimas
2023-01-10 22:00 ` Matthew Wilcox
2023-01-11 13:58 ` Ilias Apalodimas
2023-01-05 21:46 ` [PATCH v2 16/24] page_pool: Use netmem in page_pool_drain_frag() Matthew Wilcox (Oracle)
2023-01-06 15:30 ` Jesper Dangaard Brouer
2023-01-10 11:00 ` Ilias Apalodimas
2023-01-05 21:46 ` [PATCH v2 17/24] page_pool: Convert page_pool_return_skb_page() to use netmem Matthew Wilcox (Oracle)
2023-01-06 15:49 ` Jesper Dangaard Brouer
2023-01-06 16:53 ` Matthew Wilcox
2023-01-06 20:16 ` Jesper Dangaard Brouer
2023-01-09 18:36 ` Matthew Wilcox
2023-01-10 10:04 ` Jesper Dangaard Brouer
2023-01-05 21:46 ` [PATCH v2 18/24] page_pool: Convert frag_page to frag_nmem Matthew Wilcox (Oracle)
2023-01-06 15:51 ` Jesper Dangaard Brouer
2023-01-10 11:36 ` Ilias Apalodimas
2023-01-05 21:46 ` [PATCH v2 19/24] xdp: Convert to netmem Matthew Wilcox (Oracle)
2023-01-06 15:53 ` Jesper Dangaard Brouer
2023-01-10 11:50 ` Ilias Apalodimas
2023-01-05 21:46 ` [PATCH v2 20/24] mm: Remove page pool members from struct page Matthew Wilcox (Oracle)
2023-01-06 15:56 ` Jesper Dangaard Brouer
2023-01-10 11:51 ` Ilias Apalodimas
2023-01-05 21:46 ` [PATCH v2 21/24] page_pool: Pass a netmem to init_callback() Matthew Wilcox (Oracle)
2023-01-06 16:02 ` Jesper Dangaard Brouer
2023-01-10 11:32 ` Ilias Apalodimas
2023-01-05 21:46 ` [PATCH v2 22/24] net: Add support for netmem in skb_frag Matthew Wilcox (Oracle)
2023-01-05 21:46 ` [PATCH v2 23/24] mvneta: Convert to netmem Matthew Wilcox (Oracle)
2023-01-05 21:46 ` [PATCH v2 24/24] mlx5: " Matthew Wilcox (Oracle)
2023-01-06 16:31 ` Jesper Dangaard Brouer
2023-01-09 11:46 ` Tariq Toukan
2023-01-09 12:27 ` Tariq Toukan
2023-01-06 1:20 ` [PATCH v2 00/24] Split netmem from struct page Jesse Brandeburg
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=Y7g/4vLdpzGOxAKv@casper.infradead.org \
--to=willy@infradead.org \
--cc=brouer@redhat.com \
--cc=hawk@kernel.org \
--cc=ilias.apalodimas@linaro.org \
--cc=jbrouer@redhat.com \
--cc=linux-mm@kvack.org \
--cc=netdev@vger.kernel.org \
--cc=shakeelb@google.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).