TLS redux

[Joseph S. Myers]

On Thu, 30 Jan 2014, Carlos O'Donell wrote:

> I thought we ran into a similar bug for ARM and NEON?

https://sourceware.org/bugzilla/show_bug.cgi?id=15792

(this is distinct from the AArch64 bug someone referenced).

I believe the issue relates to:

(a) call-clobbered registers used for argument passing (including for 
vector types etc.);

(b) other call-clobbered registers, if there are functions whose ABI has a 
restricted clobber set that may be called through the PLT;

and that the places needing to handle saving and restoring are:

(i) _dl_runtime_resolve;

(ii) _dl_runtime_profile;

(iii) the TLS descriptors code in dl-tlsdesc.S.

Perhaps someone could confirm this and write up a description of how 
architecture maintainers can determine if their architecture has any 
issues in this regard, then send a message asking for architecture 
maintainers to check and either confirm there are no issues or file a bug 
describing the issues present?  Then someone could track which 
architectures have reported back.

-- 
Joseph S. Myers
joseph@codesourcery.com