This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: TLS, 2015 edition
- From: Szabolcs Nagy <nsz at port70 dot net>
- To: Stan Shebs <stanshebs at google dot com>
- Cc: Szabolcs Nagy <szabolcs dot nagy at arm dot com>, "libc-alpha at sourceware dot org" <libc-alpha at sourceware dot org>
- Date: Wed, 1 Apr 2015 01:36:01 +0200
- Subject: Re: TLS, 2015 edition
- Authentication-results: sourceware.org; auth=none
- References: <CA+5-Q5K7M3oJ-vvRP24GiCEHqnohWgqPkc7_6K9CLzc6a_MTcA at mail dot gmail dot com> <551A763A dot 5070909 at arm dot com> <CA+5-Q5K3fmH0AM=qa6uCT0zPvF+XLLopzpk0MzWCSK5Bzn9CNw at mail dot gmail dot com> <551AC46C dot 8060901 at arm dot com> <CA+5-Q5JNL+eeyJt_jb0iB_T393NJvWTZc6t3k1vLKHu4zA6TSA at mail dot gmail dot com>
* Stan Shebs <stanshebs@google.com> [2015-03-31 11:33:06 -0500]:
> On Tue, Mar 31, 2015 at 10:59 AM, Szabolcs Nagy <szabolcs.nagy@arm.com> wrote:
> >
> > can you give more details?
>
> It was getting quite a bit of libc-alpha traffic in late 2013 and
> early 2014, and there is a wiki page:
>
> https://sourceware.org/glibc/wiki/TLSandSignals
>
this is a c11 conformance issue
the cited 7.14.1.1p5 is overly strict i think (it is in conflict with
5.1.2.3p5 and renders read access to global consts and floatig-point
arithmetics ub in an asyc signal handler), but even the strict
7.14.1.1p5 allows writing _Thread_local volatile sig_atomic_t variables
from a signal handler (which is broken in glibc now)
> The following posting is the leader for one of several versions that
> Andrew Hunter posted:
>
> https://sourceware.org/ml/libc-alpha/2013-12/msg00212.html
>
> The Google-local patch is a descendant of that code. (In theory Google
> ought to have a git branch, but I only see ones for Roland?)
>
i dont think making lazy binding signal safe is the right approach
because of the increased complexity
removing lazy binding for tls seems simpler
> Stan