IA32/X86-64 TLS extensions

54 views
Skip to first unread message

hon...@gmail.com

unread,
Aug 21, 2007, 11:37:28 AM8/21/07
to IA32 System V Application Binary Interface, aol...@redhat.com, dis...@x86-64.org
Hi,

I uploaded IA32/X86-64 TLS extensions implemented in GNU binutils to
IA32 psABI group home page:

http://groups.google.com/group/ia32-abi

The original one is at

http://www.lsd.ic.unicamp.br/~oliva/writeups/TLS/RFC-TLSDESC-x86.txt

Evandro Menezes

unread,
Aug 22, 2007, 11:13:11 AM8/22/07
to IA32 System V Application Binary Interface
HJ,

> I uploaded IA32/X86-64 TLS extensions implemented in GNU binutils to
> IA32 psABI group home page:

BTW, Alexandre posted a patch for the x86-64 psABI a while ago at
https://www.x86-64.org/pipermail/discuss/2005-September/006725.html.

hon...@gmail.com

unread,
Aug 22, 2007, 11:49:57 AM8/22/07
to IA32 System V Application Binary Interface, dis...@x86-64.org
Hi Evandro,

Thanks.

Is x86-64 psABI being actively maintained? Given that those new
relocations are supported in
binutils, we need to update x86-64 psABI. Does x86-64 psABI have a
dedicated mailing list?
Should I create a Google group for x86-64 psABI?

H.J.

Michael Matz

unread,
Aug 22, 2007, 12:06:53 PM8/22/07
to IA32 System V Application Binary Interface, dis...@x86-64.org
Hi HJ,

On Wed, 22 Aug 2007, hon...@gmail.com wrote:

> Is x86-64 psABI being actively maintained?

Yes.

> Given that those new relocations are supported in binutils, we need to
> update x86-64 psABI. Does x86-64 psABI have a dedicated mailing list?

Yes, the dis...@x86-64.org list, which is already CCed here. The patch
from Alexandre wasn't applied at that time because we wanted to wait for
implementations (and agreement) IIRC. I think we should include it now,
and I'll go over it during the next days.


Ciao,
Michael.

H.J. Lu

unread,
Aug 22, 2007, 12:16:00 PM8/22/07
to Michael Matz, IA32 System V Application Binary Interface, dis...@x86-64.org
On Wed, Aug 22, 2007 at 06:06:53PM +0200, Michael Matz wrote:
>
> Hi HJ,
>
> On Wed, 22 Aug 2007, hon...@gmail.com wrote:
>
> > Is x86-64 psABI being actively maintained?
>
> Yes.

The x86-64 psABI used to have an online CVS respository. It is
quite useful for contributing changes like that. Can you bring it
back?

>
> > Given that those new relocations are supported in binutils, we need to
> > update x86-64 psABI. Does x86-64 psABI have a dedicated mailing list?
>
> Yes, the dis...@x86-64.org list, which is already CCed here. The patch
> from Alexandre wasn't applied at that time because we wanted to wait for
> implementations (and agreement) IIRC. I think we should include it now,
> and I'll go over it during the next days.
>
>

Thanks.


H.J.

Michael Matz

unread,
Aug 22, 2007, 12:40:40 PM8/22/07
to H.J. Lu, IA32 System V Application Binary Interface, dis...@x86-64.org
Hi,

On Wed, 22 Aug 2007, H.J. Lu wrote:

> > Yes.
>
> The x86-64 psABI used to have an online CVS respository. It is quite
> useful for contributing changes like that. Can you bring it back?

It's still there, but svn now (
http://www.x86-64.org/viewvc/trunk/x86-64-ABI/ ). But I just see that I
can't access the svn repo itself anymore, and can't login there.
Something has broken the server, I'll ping the maintainers.


Ciao,
Michael.

Alexandre Oliva

unread,
Aug 22, 2007, 5:01:49 PM8/22/07
to ia32...@googlegroups.com, H.J. Lu, dis...@x86-64.org

Thanks for the info, Michael, and thanks, H.J., for having uploaded
the file. I've been meaning to get back to this project for a while,
and I do have updated glibc patches that AFAIK are just waiting for
official acceptance into the ABI documentation for integration.

Sorry that it's been taking me so long to wrap it up. This has all
been a low-priority personal project for most of the time.

I think it's safe to say that the current specs are stable, even
though I do have some experiments that show a lot (but not all) of the
benefits of this new approach could be accomplished without new
relocations or access models. It loses in performance, but it could
be less of an overhaul, especially for platforms that already have a
functional TLS ABI and don't want to support multiple ones.

Are there sources for the IA32 ABI specs that I could patch to
introduce the changes proposed in my document that you uploaded to the
group?

Thanks, and best regards,

--
Alexandre Oliva http://www.lsd.ic.unicamp.br/~oliva/
FSF Latin America Board Member http://www.fsfla.org/
Red Hat Compiler Engineer aoliva@{redhat.com, gcc.gnu.org}
Free Software Evangelist oliva@{lsd.ic.unicamp.br, gnu.org}

Alexandre Oliva

unread,
Aug 22, 2007, 5:10:00 PM8/22/07
to Evandro Menezes, dis...@x86-64.org, IA32 System V Application Binary Interface

> HJ,

Actually, there's a better version of that patch here
http://www.x86-64.org/pipermail/discuss/2006-October/009302.html

Downthread, you've requested some changes, I responded with further
questions, and it looks like I'd never got a reply to those questions.
Presumably you sent them to the discuss@x86_64 list only, and I'm not
on the list. I'm subscribing right now.

H.J. Lu

unread,
Aug 23, 2007, 11:03:33 AM8/23/07
to ia32...@googlegroups.com
On Wed, Aug 22, 2007 at 06:01:49PM -0300, Alexandre Oliva wrote:
>
> Are there sources for the IA32 ABI specs that I could patch to
> introduce the changes proposed in my document that you uploaded to the
> group?
>

I don't have the source for the IA32 psABI. I think we can update
LSB IA32:

http://www.linux-foundation.org/en/Specifications


H.J.

Mats Wichmann

unread,
Aug 23, 2007, 1:23:00 PM8/23/07
to ia32...@googlegroups.com
H.J. Lu wrote:
> On Wed, Aug 22, 2007 at 06:01:49PM -0300, Alexandre Oliva wrote:
>
>> Are there sources for the IA32 ABI specs that I could patch to
>> introduce the changes proposed in my document that you uploaded to the
>> group?
>>
>>
>
> I don't have the source for the IA32 psABI. I think we can update
> LSB IA32:
>
we can, but it will happen a bit slowly as it's necessary for a
condition quite
close to "every Linux distribution has it" to be true; tricky for older
distros
that aren't updating their binutils and so probably limited to the next
major
version.

we could arrange to resume publishing the IA32 architecture-specific piece
(which is basically a Linux delta to the psABI) separately if that would
help,
in advance of that going into the full LSB.

Reply all
Reply to author
Forward
0 new messages