This is the mail archive of the
mailing list for the glibc project.
Re: powerpc __tls_get_addr call optimization
- From: Rich Felker <dalias at libc dot org>
- To: "H.J. Lu" <hjl dot tools at gmail dot com>
- Cc: Carlos O'Donell <carlos at redhat dot com>, Alan Modra <amodra at gmail dot com>, GNU C Library <libc-alpha at sourceware dot org>
- Date: Fri, 20 Mar 2015 14:01:14 -0400
- Subject: Re: powerpc __tls_get_addr call optimization
- Authentication-results: sourceware.org; auth=none
- References: <20150319025631 dot GC28603 at bubble dot grove dot modra dot org> <550B94FC dot 3070903 at redhat dot com> <20150320075502 dot GC26234 at bubble dot grove dot modra dot org> <20150320152712 dot GK23507 at brightrain dot aerifal dot cx> <550C4152 dot 5020206 at redhat dot com> <CAMe9rOqrTWXs=rped5Tm-__Ezu9w0_D=JQpkGRRwKJgKOMAyPQ at mail dot gmail dot com> <20150320161448 dot GL23507 at brightrain dot aerifal dot cx> <CAMe9rOqXk=kF5cUO2bQG0ebto5=2QeSZ6Kj=daRXxUL-n2UzyQ at mail dot gmail dot com> <550C4909 dot 7070908 at redhat dot com> <CAMe9rOqeFhbXa_yHG6Oh29cQJPs6Q7H5Bg6wvZJ36=as_R8ktA at mail dot gmail dot com>
On Fri, Mar 20, 2015 at 09:24:21AM -0700, H.J. Lu wrote:
> >>> Not that I know of. i386 TLSDESC was broken in binutils for several
> >>> years and only recently fixed... Until a couple months ago nobody
> >>> noticed. :-(
> >>> This situation really should be set right (with proper tests and
> >>> timeline for changing the default to TLSDESC) so we can put an end to
> >>> the invalid use of IE-model in shared libraries.
> >> Another thing, x86 and x86-64 TLS DESC spec should be
> >> in x86 and x86-64 psABIs, not a URL.
> > Agreed. As should the TLS specification instead of a URL reference to
> > tls.pdf which is going to get out of date.
> TLS spec is too big to be included in x86 psABIs unless
> Ulrich contributed patches for tex source to x86 psABIs.
Are you sure? His TLS docs contain a lot of informative content that
should not be taken as spec. There's no reason for a psABI to document
optimizations a linker can make. Simply documenting the semantics of
the relocation types and the actual ABI constraints they impose
(mainly, location of static TLS relative to the thread-pointer) should
be possible in a fairly compact text suitable for inclusion in the
psABI. Of course actually writing that is a bit of work...