From: Al Viro <viro@ZenIV.linux.org.uk>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Eric Dumazet <eric.dumazet@gmail.com>,
David Miller <davem@davemloft.net>,
Stephen Hemminger <stephen@networkplumber.org>,
Network Development <netdev@vger.kernel.org>,
David Howells <dhowells@redhat.com>,
linux-fsdevel <linux-fsdevel@vger.kernel.org>
Subject: Re: [Bug 106241] New: shutdown(3)/close(3) behaviour is incorrect for sockets in accept(3)
Date: Fri, 30 Oct 2015 21:02:15 +0000 [thread overview]
Message-ID: <20151030210215.GI22011@ZenIV.linux.org.uk> (raw)
In-Reply-To: <CA+55aFx_BQG4FzSuyL2KAz=c+R0cPv0ZpjboT_=yKjZNGmTUEg@mail.gmail.com>
On Fri, Oct 30, 2015 at 10:18:12AM -0700, Linus Torvalds wrote:
> I do wonder if we couldn't just speed up the bitmap allocator by an
> order of magnitude. It would be nicer to be able to make existing
> loads faster without any new "don't bother with POSIX semantics" flag.
>
> We could, for example, extend the "open_fds" bitmap with a
> "second-level" bitmap that has information on when the first level is
> full. We traverse the open_fd's bitmap one word at a time anyway, we
> could have a second-level bitmap that has one bit per word to say
> whether that word is already full.
Your variant has 1:64 ratio; obviously better than now, but we can actually
do 1:bits-per-cacheline quite easily.
I've been playing with a variant that has more than two bitmaps, and
AFAICS it
a) does not increase the amount of cacheline pulled and
b) keeps it well-bound even in the absolutely worst case
(128M-odd descriptors filled, followed by close(0);dup2(1,0); -
in that case it ends up accessing the 7 cachelines worth of
bitmaps; your variant will read through 4000-odd cachelines of
the summary bitmap alone; the mainline is even worse).
> The advantage of the above is that it should just work for existing
> binaries. It may not be quite as optimal as just introducing a new
> "don't care about POSIX" feature, but quite frankly, if it cuts down
> the bad case of "find_next_zero_bit()" by a factror of 64 (and then
> adds a *small* expense factor on top of that), I suspect it should be
> "good enough" even for your nasty case.
>
> What do you think? Willing to try the above approach (with any
> inevitable bug-fixes) and see how it compares?
>
> Obviously in addition to any fixes to my pseudo-code above you'd need
> to add the allocations for the new "full_fds_bits" etc, but I think it
> should be easy to make the full_fds_bit allocation be *part* of the
> "open_fds" allocation, so you wouldn't need a new allocation in
> alloc_fdtable(). We already do that whole "use a single allocation" to
> combine open_fds with close_on_exec into one single allocation.
I'll finish testing what I've got and post it; it costs 3 extra pointers
in the files_struct and a bit fatter bitmap allocation (less than 0.2%
extra). All the arguments regarding the unmodified binaries apply, of
course, and so far it looks fairly compact...
next prev parent reply other threads:[~2015-10-30 21:02 UTC|newest]
Thread overview: 138+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-10-19 16:59 Fw: [Bug 106241] New: shutdown(3)/close(3) behaviour is incorrect for sockets in accept(3) Stephen Hemminger
2015-10-19 23:33 ` Eric Dumazet
2015-10-20 1:12 ` Alan Burlison
2015-10-20 1:45 ` Eric Dumazet
2015-10-20 9:59 ` Alan Burlison
2015-10-20 11:24 ` David Miller
2015-10-20 11:39 ` Alan Burlison
2015-10-20 13:19 ` Fw: " Eric Dumazet
2015-10-20 13:45 ` Alan Burlison
2015-10-20 15:30 ` Eric Dumazet
2015-10-20 18:31 ` Alan Burlison
2015-10-20 18:42 ` Eric Dumazet
2015-10-21 10:25 ` David Laight
2015-10-21 10:49 ` Alan Burlison
2015-10-21 11:28 ` Eric Dumazet
2015-10-21 13:03 ` Alan Burlison
2015-10-21 13:29 ` Eric Dumazet
2015-10-21 3:49 ` Al Viro
2015-10-21 14:38 ` Alan Burlison
2015-10-21 15:30 ` David Miller
2015-10-21 16:04 ` Casper.Dik
2015-10-21 21:18 ` Eric Dumazet
2015-10-21 21:28 ` Al Viro
2015-10-21 16:32 ` Fw: " Eric Dumazet
2015-10-21 18:51 ` Al Viro
2015-10-21 20:33 ` Casper.Dik
2015-10-22 4:21 ` Al Viro
2015-10-22 10:55 ` Alan Burlison
2015-10-22 18:16 ` Al Viro
2015-10-22 20:15 ` Alan Burlison
2015-11-02 10:03 ` David Laight
2015-11-02 10:29 ` Al Viro
2015-10-21 22:28 ` Alan Burlison
2015-10-22 1:29 ` David Miller
2015-10-22 4:17 ` Alan Burlison
2015-10-22 4:44 ` Al Viro
2015-10-22 6:03 ` Al Viro
2015-10-22 6:34 ` Casper.Dik
2015-10-22 17:21 ` Al Viro
2015-10-22 18:24 ` Casper.Dik
2015-10-22 19:07 ` Al Viro
2015-10-22 19:51 ` Casper.Dik
2015-10-22 21:57 ` Al Viro
2015-10-23 9:52 ` Casper.Dik
2015-10-23 13:02 ` Eric Dumazet
2015-10-23 13:20 ` Casper.Dik
2015-10-23 13:48 ` Eric Dumazet
2015-10-23 14:13 ` Eric Dumazet
2015-10-23 13:35 ` Alan Burlison
2015-10-23 14:21 ` Eric Dumazet
2015-10-23 15:46 ` Alan Burlison
2015-10-23 16:00 ` Eric Dumazet
2015-10-23 16:07 ` Alan Burlison
2015-10-23 16:19 ` Eric Dumazet
2015-10-23 16:40 ` Alan Burlison
2015-10-23 17:47 ` Eric Dumazet
2015-10-23 17:59 ` [PATCH net-next] af_unix: do not report POLLOUT on listeners Eric Dumazet
2015-10-25 13:45 ` David Miller
2015-10-24 2:30 ` [Bug 106241] New: shutdown(3)/close(3) behaviour is incorrect for sockets in accept(3) Al Viro
2015-10-27 9:08 ` Casper.Dik
2015-10-27 10:52 ` Alan Burlison
2015-10-27 12:01 ` Eric Dumazet
2015-10-27 12:27 ` Alan Burlison
2015-10-27 12:44 ` Eric Dumazet
2015-10-27 13:42 ` David Miller
2015-10-27 13:37 ` Alan Burlison
2015-10-27 13:59 ` David Miller
2015-10-27 14:13 ` Alan Burlison
2015-10-27 14:39 ` David Miller
2015-10-27 14:39 ` Alan Burlison
2015-10-27 15:04 ` David Miller
2015-10-27 15:53 ` Alan Burlison
2015-10-27 23:17 ` Al Viro
2015-10-28 0:13 ` Eric Dumazet
2015-10-28 12:35 ` Al Viro
2015-10-28 13:24 ` Eric Dumazet
2015-10-28 14:47 ` Eric Dumazet
2015-10-28 21:13 ` Al Viro
2015-10-28 21:44 ` Eric Dumazet
2015-10-28 22:33 ` Al Viro
2015-10-28 23:08 ` Eric Dumazet
2015-10-29 0:15 ` Al Viro
2015-10-29 3:29 ` Eric Dumazet
2015-10-29 4:16 ` Al Viro
2015-10-29 12:35 ` Eric Dumazet
2015-10-29 13:48 ` Eric Dumazet
2015-10-30 17:18 ` Linus Torvalds
2015-10-30 21:02 ` Al Viro [this message]
2015-10-30 21:23 ` Linus Torvalds
2015-10-30 21:50 ` Linus Torvalds
2015-10-30 22:33 ` Al Viro
2015-10-30 23:52 ` Linus Torvalds
2015-10-31 0:09 ` Al Viro
2015-10-31 15:59 ` Eric Dumazet
2015-10-31 19:34 ` Al Viro
2015-10-31 19:54 ` Linus Torvalds
2015-10-31 20:29 ` Al Viro
2015-11-02 0:24 ` Al Viro
2015-11-02 0:59 ` Linus Torvalds
2015-11-02 2:14 ` Eric Dumazet
2015-11-02 6:22 ` Al Viro
2015-10-31 20:45 ` Eric Dumazet
2015-10-31 21:23 ` Linus Torvalds
2015-10-31 21:51 ` Al Viro
2015-10-31 22:34 ` Eric Dumazet
2015-10-31 1:07 ` Eric Dumazet
2015-10-28 16:04 ` Alan Burlison
2015-10-29 14:58 ` David Holland
2015-10-29 15:18 ` Alan Burlison
2015-10-29 16:01 ` David Holland
2015-10-29 16:15 ` Alan Burlison
2015-10-29 17:07 ` Al Viro
2015-10-29 17:12 ` Alan Burlison
2015-10-30 1:54 ` David Miller
2015-10-30 1:55 ` David Miller
2015-10-30 5:44 ` David Holland
2015-10-30 17:43 ` David Laight
2015-10-30 21:09 ` Al Viro
2015-11-04 15:54 ` David Laight
2015-11-04 16:27 ` Al Viro
2015-11-06 15:07 ` David Laight
2015-11-06 19:31 ` Al Viro
2015-10-22 6:51 ` Casper.Dik
2015-10-22 11:18 ` Alan Burlison
2015-10-22 11:15 ` Alan Burlison
2015-10-22 6:15 ` Casper.Dik
2015-10-22 11:30 ` Eric Dumazet
2015-10-22 11:58 ` Alan Burlison
2015-10-22 12:10 ` Eric Dumazet
2015-10-22 13:12 ` David Miller
2015-10-22 13:14 ` Alan Burlison
2015-10-22 17:05 ` Al Viro
2015-10-22 17:39 ` Alan Burlison
2015-10-22 18:56 ` Al Viro
2015-10-22 19:50 ` Casper.Dik
2015-10-23 17:09 ` Al Viro
2015-10-23 18:30 ` Fw: " David Holland
2015-10-23 19:51 ` Al Viro
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=20151030210215.GI22011@ZenIV.linux.org.uk \
--to=viro@zeniv.linux.org.uk \
--cc=davem@davemloft.net \
--cc=dhowells@redhat.com \
--cc=eric.dumazet@gmail.com \
--cc=linux-fsdevel@vger.kernel.org \
--cc=netdev@vger.kernel.org \
--cc=stephen@networkplumber.org \
--cc=torvalds@linux-foundation.org \
/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).