This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [patch] Fix BZ#16374 -- don't use mmap for FILE buffers
- From: Paul Pluzhnikov <ppluzhnikov at google dot com>
- To: Florian Weimer <fweimer at redhat dot com>
- Cc: GLIBC Devel <libc-alpha at sourceware dot org>
- Date: Mon, 16 Feb 2015 12:53:32 -0800
- Subject: Re: [patch] Fix BZ#16374 -- don't use mmap for FILE buffers
- Authentication-results: sourceware.org; auth=none
- References: <CALoOobNomWyxd9Oz3=kHq0vyBpmfxSyj_cFBxyahCJSs1cZBzQ at mail dot gmail dot com> <54E236D9 dot 3010807 at redhat dot com> <CALoOobPVr0PAkzDtQbXXGP7Vyu-Ls+V-1JGXpi8yBbyryqYf5g at mail dot gmail dot com> <54E24E99 dot 2090207 at redhat dot com>
On Mon, Feb 16, 2015 at 12:10 PM, Florian Weimer <fweimer@redhat.com> wrote:
> I think it will still be a net win, but fopen/fclose pairs might now run
> into the issue discussed in the âReduce worst-case behaviour with
> madvise and refault overheadâ thread. :-P
Thanks for pointing out that thread.
I don't believe we'll run into that here: AFAICT the mmap threshold is
DEFAULT_MMAP_THRESHOLD_MIN currently at 128KB, while the buffers we
are malloc()ing here are _IO_BUFSIZ of 8KB.
--
Paul Pluzhnikov