New binutils for kernel

New binutils for kernel

[Ralf Baechle]

On Tue, Aug 20, 2002 at 04:19:35PM +0200, Maciej W. Rozycki wrote:

> On Mon, 19 Aug 2002, Ralf Baechle wrote:
> 
> > >  Are you sure?  I believe the patch effectively forces everyone to use
> > > binutils 2.13 for mips64.  Is it really acceptable now? 
> > 
> > In the past week I ended up more and more kludging around binutils bugs.
> > We need something newer and distributions seem to be all at ~ 2.12 at least.
> 
>  While 2.12 may be OK from the file format point of view, there are
> serious bugs leading to bad code.  So bad the kernel doesn't work.  It's
> really 2.13 that is needed.  I have another less important fix that will
> hopefully go in to 2.13.1 and all gcc versions are broken without yet
> another fix (it bites in mm/mmap.c; not sure if fatally). 
> 
> > So I guess it's time to bite the bullet?
> 
>  Since I'm using 2.13 anyway, it's alike to me.  But it should be
> discussed at the list, IMO.

Yep.  It won't hurt most of us kernel hackers very much but in particular
the distribution people may want to comment.

So any comments?

  Ralf

Re: New binutils for kernel

[Daniel Jacobowitz]

On Tue, Aug 20, 2002 at 04:29:59PM +0200, Ralf Baechle wrote:
> On Tue, Aug 20, 2002 at 04:19:35PM +0200, Maciej W. Rozycki wrote:
> 
> > On Mon, 19 Aug 2002, Ralf Baechle wrote:
> > 
> > > >  Are you sure?  I believe the patch effectively forces everyone to use
> > > > binutils 2.13 for mips64.  Is it really acceptable now? 
> > > 
> > > In the past week I ended up more and more kludging around binutils bugs.
> > > We need something newer and distributions seem to be all at ~ 2.12 at least.
> > 
> >  While 2.12 may be OK from the file format point of view, there are
> > serious bugs leading to bad code.  So bad the kernel doesn't work.  It's
> > really 2.13 that is needed.  I have another less important fix that will
> > hopefully go in to 2.13.1 and all gcc versions are broken without yet
> > another fix (it bites in mm/mmap.c; not sure if fatally). 
> > 
> > > So I guess it's time to bite the bullet?
> > 
> >  Since I'm using 2.13 anyway, it's alike to me.  But it should be
> > discussed at the list, IMO.
> 
> Yep.  It won't hurt most of us kernel hackers very much but in particular
> the distribution people may want to comment.
> 
> So any comments?

Well, I think 2.13's a good idea, but it's very new.  I'd say that was
acceptable as long as you're looking at MIPS64 only, not at MIPS32. 

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: New binutils for kernel

[Ralf Baechle]

On Tue, Aug 20, 2002 at 10:50:51AM -0400, Daniel Jacobowitz wrote:

> > So any comments?
> 
> Well, I think 2.13's a good idea, but it's very new.  I'd say that was
> acceptable as long as you're looking at MIPS64 only, not at MIPS32. 

Such considerations have kept us back at antique levels of binutils.  And
juggling with several different versions for userland, and two kernel
flavours is evil ...

  Ralf

Re: New binutils for kernel

[Daniel Jacobowitz]

On Tue, Aug 20, 2002 at 04:58:35PM +0200, Ralf Baechle wrote:
> On Tue, Aug 20, 2002 at 10:50:51AM -0400, Daniel Jacobowitz wrote:
> 
> > > So any comments?
> > 
> > Well, I think 2.13's a good idea, but it's very new.  I'd say that was
> > acceptable as long as you're looking at MIPS64 only, not at MIPS32. 
> 
> Such considerations have kept us back at antique levels of binutils.  And
> juggling with several different versions for userland, and two kernel
> flavours is evil ...

Please keep in mind that 2.13's been out for all of half a month now. 
Almost no one's using it yet for significant development... if you
decide to require it, we can cope, though.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: New binutils for kernel

[Maciej W. Rozycki]

On Tue, 20 Aug 2002, Ralf Baechle wrote:

> > Well, I think 2.13's a good idea, but it's very new.  I'd say that was
> > acceptable as long as you're looking at MIPS64 only, not at MIPS32. 
> 
> Such considerations have kept us back at antique levels of binutils.  And
> juggling with several different versions for userland, and two kernel
> flavours is evil ...

 Any version since 2.11, possibly older, should work just fine for 32-bit
MIPS.  I don't think there are any significant interface changes between
2.11 and 2.13, so if 2.13 works then 2.11 will not bail out either.  Thus
there is no need to force 2.13 for 32-bit MIPS, but I think it is
acceptable to stop caring about versions older than 2.11 in the nearby
future.

-- 
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: macro@ds2.pg.gda.pl, PGP key available        +

Re: New binutils for kernel

[Daniel Jacobowitz]

On Tue, Aug 20, 2002 at 05:08:20PM +0200, Maciej W. Rozycki wrote:
> On Tue, 20 Aug 2002, Ralf Baechle wrote:
> 
> > > Well, I think 2.13's a good idea, but it's very new.  I'd say that was
> > > acceptable as long as you're looking at MIPS64 only, not at MIPS32. 
> > 
> > Such considerations have kept us back at antique levels of binutils.  And
> > juggling with several different versions for userland, and two kernel
> > flavours is evil ...
> 
>  Any version since 2.11, possibly older, should work just fine for 32-bit
> MIPS.  I don't think there are any significant interface changes between
> 2.11 and 2.13, so if 2.13 works then 2.11 will not bail out either.  Thus
> there is no need to force 2.13 for 32-bit MIPS, but I think it is
> acceptable to stop caring about versions older than 2.11 in the nearby
> future.

Sure.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: New binutils for kernel

[Ralf Baechle]

On Tue, Aug 20, 2002 at 11:11:35AM -0400, Daniel Jacobowitz wrote:

> > > > Well, I think 2.13's a good idea, but it's very new.  I'd say that was
> > > > acceptable as long as you're looking at MIPS64 only, not at MIPS32. 
> > > 
> > > Such considerations have kept us back at antique levels of binutils.  And
> > > juggling with several different versions for userland, and two kernel
> > > flavours is evil ...
> > 
> >  Any version since 2.11, possibly older, should work just fine for 32-bit
> > MIPS.  I don't think there are any significant interface changes between
> > 2.11 and 2.13, so if 2.13 works then 2.11 will not bail out either.  Thus
> > there is no need to force 2.13 for 32-bit MIPS, but I think it is
> > acceptable to stop caring about versions older than 2.11 in the nearby
> > future.
> 
> Sure.

Sounds like we'll then recommend 2.13 for the fearless and 64-bit developers
and 2.11 for everybody else.

  Ralf

Re: New binutils for kernel

[H. J. Lu]

On Tue, Aug 20, 2002 at 04:58:35PM +0200, Ralf Baechle wrote:
> On Tue, Aug 20, 2002 at 10:50:51AM -0400, Daniel Jacobowitz wrote:
> 
> > > So any comments?
> > 
> > Well, I think 2.13's a good idea, but it's very new.  I'd say that was
> > acceptable as long as you're looking at MIPS64 only, not at MIPS32. 
> 
> Such considerations have kept us back at antique levels of binutils.  And
> juggling with several different versions for userland, and two kernel
> flavours is evil ...

I have no problems with binutils 2.13.90.0.4 on Linux/mips. I rebuild
everything, from kernel to rpms, for Linux/mipsel with it. They seem
to run fine.


H.J.

Re: New binutils for kernel

[Karsten Merker]

On Tue, Aug 20, 2002 at 04:29:59PM +0200, Ralf Baechle wrote:

[requiring binutils 2.13]]

> Yep.  It won't hurt most of us kernel hackers very much but in particular
> the distribution people may want to comment.

Well, I guess that covers me :-).

Debian Woody, the current stable Debian release (AFAIK the only full-blown
distribution for Linux/mips that targets at "normal users" and not only at
developers) ships with binutils 2.12.90.0.1. As the Debian policy requires
that no new program versions are to be introduced for the stable release,
Debian will not be able to switch to binutils 2.13 there. Bugfixes for
packages in the release are accepted, but not new upstream versions; so to
get any fixes in, they would need to be backported to binutils 2.12.

For the unstable distribution (currently binutils 2.12.90.0.15), switching
to binutils 2.13 should IMHO be possible in a reasonable timeframe, if it
does not break other things. Caveat: as Debian requires all 11 released
architectures to be in sync, any new version would have to work properly on
_all_ supported platforms. The decision about introducing a new version of a
package into Debian is taken by the package maintainer, so I am going to ask
him about his plans.

Regards,
Karsten
-- 
#include <standard_disclaimer>
Nach Paragraph 28 Abs. 3 Bundesdatenschutzgesetz widerspreche ich der Nutzung
oder Uebermittlung meiner Daten fuer Werbezwecke oder fuer die Markt- oder
Meinungsforschung.