This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
[PATCH 0/2] New pthread rwlock that is more scalable.
- From: Torvald Riegel <triegel at redhat dot com>
- To: GLIBC Devel <libc-alpha at sourceware dot org>
- Date: Wed, 27 Jul 2016 23:38:53 +0200
- Subject: [PATCH 0/2] New pthread rwlock that is more scalable.
- Authentication-results: sourceware.org; auth=none
This reimplements the POSIX Threads reader--writer lock in a way that
will make it more scalable. In particular, the state changes of an
rwlock are not serialized through a critical section anymore but use
atomic operations (in the fast paths, just one atomic read-modify-write
is needed per lock acquisition or lock release).
I have not actually done performance testing; my focus so far was one
ensuring correctness of a new algorithm for which I'm confident that it
will scale better and will have comparable or less overheads in the
uncontended case.
Patch 1/2 adds an additional atomic operation we need.
Patch 2/2 is the rwlock rewrite.
Tested on x86_64-linux using the glibc tests.
We intend to do further testing in Rawhide. Reviews before that testing
concludes are nonetheless appreciated :)