This is the mail archive of the
binutils@sourceware.org
mailing list for the binutils project.
Re: PowerPC TLS ABI
- From: Alan Modra <amodra at bigpond dot net dot au>
- To: binutils at sourceware dot org, Peter Bergner <bergner at us dot ibm dot com>, David Edelsohn <dje dot gcc at gmail dot com>, Mark Mendell <mendell at ca dot ibm dot com>, Yaakov Yaari <YAARI at il dot ibm dot com>
- Date: Tue, 24 Feb 2009 08:30:15 +1030
- Subject: Re: PowerPC TLS ABI
- References: <20090223141128.GD6783@bubble.grove.modra.org> <20090223150526.GA15177@caradoc.them.org>
On Mon, Feb 23, 2009 at 10:05:26AM -0500, Daniel Jacobowitz wrote:
> On Tue, Feb 24, 2009 at 12:41:28AM +1030, Alan Modra wrote:
> > sequences. In fact if we use two different marker relocs then we can
> > partition the instructions into those used for global dynamic and
> > those used for local dynamic tls. This is sufficient because every
> > occurrence of global dynamic code for a given symbol will be subject
> > to the same optimization, and likewise for local dynamic. The linker
> > doesn't need to know exactly which sequence a given __tls_get_addr
> > call belong to, just that it belongs with a given symbol.
>
> For correctness, I believe this means the linker has to transform
> every TLS reference to a given symbol or none of them - is that right?
Yes.
> Are there any ways the linker can fail, e.g. to find the expected
> instruction sequence?
Not unless the ABI is violated, in which case all bets are off.
--
Alan Modra
Australia Development Lab, IBM