[parisc-linux] New binutils available from cvs

[parisc-linux] New binutils available from cvs

[Alan Modra]

Hello everyone,
   I have just uploaded a new binutils to puffin.external.hp.com cvs.  The
new binutils, which you can get with "cvs co binutils-2.10", is a merge of
current public sourceware.cygnus.com cvs with the puffin tree and a few
fixes of my own.

What's changed:

o Quite a few fixes have been added to the sourceware archive since the
  previous puffin import.  I'll not try to list these here.  One warning
  though: some little fixes and general code clean-up I checked in to the
  sourceware archive broke the hppa assembler.  Jeff Law has fixes from 
  me for the problem, but so far hasn't reviewed and OK'd them, so the
  sourceware hppa support is currently broken.

o Some parts of bfd/elf32-hppa.c have been rewritten, since the previous
  code just couldn't do the right thing.  eg. hppa_elf_relocate_insn used
  to inspect the insn it was given to see whether it was a branch, but
  there's no guarantee that insn was actually code.  It could have come
  from a .word directive, and in fact one place in pa-linux was broken
  due to this bug.

o The binutils hppa testsuite has been modified to (mostly) work for
  hppa1.1-linux.

o ld/emultempl/hppalinux.em has the new orphan section handling.

o I've put in a few fixes for 64 bit parisc, mostly just things I noticed
  while trying to make sense of the hppa backend.

o There are various incomplete pieces of code for things we'll need to
  eventually support.

***BIG WARNING**
  This binutils goes back to using the standard hppa comment char and line
separator char.  Existing puffin parisc code needs to be modified to suit,
otherwise you'll get lots of assembly errors (easy to fix), or silent
removal of code (where multiple asm instructions were placed on one line).

  I'll be shortly checking in a patch to puffin linux-2.3 and palo cvs
that should allow either the new or the old binutils to be used.  I guess
there is plenty of other source that needs fixing similarly.

Enjoy,
  Alan Modra.

PS. If you check in patches to the new binutils, please _please_ put in a
ChangeLog entry too.
-- 
Linuxcare.  Support for the Revolution.

Re: [parisc-linux] New binutils available from cvs

[Philipp Rumpf]

On Thu, May 18, 2000 at 12:41:13PM +1000, Alan Modra wrote:
> Hello everyone,
>    I have just uploaded a new binutils to puffin.external.hp.com cvs.  The
> new binutils, which you can get with "cvs co binutils-2.10", is a merge of

Any reason for renaming cvs modules all the time instead of staying with
one name and perhaps not pissing off people with slow, expensive
connections ?

> ***BIG WARNING**
>   This binutils goes back to using the standard hppa comment char and line
> separator char.  Existing puffin parisc code needs to be modified to suit,

This
 a) makes certain glbic code harder to work (which already does work, at
    that)
 b) is inconsistent with every single other architecture that does run
    Linux
 c) is inconsistent with my tree, and basically forces me to distribute
    binutils and gcc patches in addition to the gcc one-liner I already
    have.

All in all this strikes me as not too smart.

BTW, did you fix gcc to work with -mdisable-fpregs and actually make the
kernel use that in the CVS tree ?  It caused a lot of oopses here before
I worked around it (patch is at
ftp://linux-parisc.sourceforge.net/pub/linux-parisc/) (why we got oopses
and what is wrong should be quite obvious if you know RTL - I don't and
it seems reasonably obvious to me, I just can't fix it).

	Philipp Rumpf

Re: [parisc-linux] New binutils available from cvs

[Alan Modra]

On Thu, 18 May 2000, Philipp Rumpf wrote:

> On Thu, May 18, 2000 at 12:41:13PM +1000, Alan Modra wrote:
> > Hello everyone,
> >    I have just uploaded a new binutils to puffin.external.hp.com cvs.  The
> > new binutils, which you can get with "cvs co binutils-2.10", is a merge of
> 
> Any reason for renaming cvs modules all the time instead of staying with
> one name and perhaps not pissing off people with slow, expensive
> connections ?

Mainly so people can still easily use the old binutils.  .diff.gz came to
1.8M anyway.  I can put the diff somewhere you can get it, if you like.

> > ***BIG WARNING**
> >   This binutils goes back to using the standard hppa comment char and line
> > separator char.  Existing puffin parisc code needs to be modified to suit,
> 
> This
>  a) makes certain glbic code harder to work (which already does work, at
>     that)

Can you explain this?

>  b) is inconsistent with every single other architecture that does run
>     Linux
>  c) is inconsistent with my tree, and basically forces me to distribute
>     binutils and gcc patches in addition to the gcc one-liner I already
>     have.
> 
> All in all this strikes me as not too smart.

Using `!' for comments
a) breaks fp condition codes.
b) is inconsistent with existing hp assembler.

All in all .... :-)

Item a) is the killer.  It might be possible to work around this problem
and make `!' start a comment only when the preceding non-whitespace
character is not a comma, but this sort of horrible hack hasn't much hope
of ever being accepted back into FSF binutils.

> BTW, did you fix gcc to work with -mdisable-fpregs and actually make the
> kernel use that in the CVS tree ?  It caused a lot of oopses here before
> I worked around it (patch is at
> ftp://linux-parisc.sourceforge.net/pub/linux-parisc/) (why we got oopses
> and what is wrong should be quite obvious if you know RTL - I don't and
> it seems reasonably obvious to me, I just can't fix it).

No, I haven't done anything with gcc

-- 
Linuxcare.  Support for the Revolution.

Re: [parisc-linux] New binutils available from cvs

[Philipp Rumpf]

On Thu, May 18, 2000 at 08:37:20PM +1000, Alan Modra wrote:
> On Thu, 18 May 2000, Philipp Rumpf wrote:
> 
> > On Thu, May 18, 2000 at 12:41:13PM +1000, Alan Modra wrote:
> > > Hello everyone,
> > >    I have just uploaded a new binutils to puffin.external.hp.com cvs.  The
> > > new binutils, which you can get with "cvs co binutils-2.10", is a merge of
> > 
> > Any reason for renaming cvs modules all the time instead of staying with
> > one name and perhaps not pissing off people with slow, expensive
> > connections ?
> 
> Mainly so people can still easily use the old binutils.  .diff.gz came to
> 1.8M anyway.  I can put the diff somewhere you can get it, if you like.
> 
> > > ***BIG WARNING**
> > >   This binutils goes back to using the standard hppa comment char and line
> > > separator char.  Existing puffin parisc code needs to be modified to suit,
> > 
> > This
> >  a) makes certain glbic code harder to work (which already does work, at
> >     that)
> 
> Can you explain this?
> 
> >  b) is inconsistent with every single other architecture that does run
> >     Linux
> >  c) is inconsistent with my tree, and basically forces me to distribute
> >     binutils and gcc patches in addition to the gcc one-liner I already
> >     have.
> > 
> > All in all this strikes me as not too smart.
> 
> Using `!' for comments
> a) breaks fp condition codes.

No, it doesn't.  The condition code for fp "not equal" is "<>", not "!=".

> b) is inconsistent with existing hp assembler.

Your point ?

We broke existing assembly language / used the smarter of two possibilities
on almost all architectures.

> All in all .... :-)

All in all even _if_ you have a reason to change it back it could have used
some discussion.  So far you haven't.

> Item a) is the killer.  It might be possible to work around this problem
> and make `!' start a comment only when the preceding non-whitespace
> character is not a comma, but this sort of horrible hack hasn't much hope
> of ever being accepted back into FSF binutils.

Any idea which illegal drug the person who decided "!=" was the fp not
equal condition code was on ;) ?

> > BTW, did you fix gcc to work with -mdisable-fpregs and actually make the
> > kernel use that in the CVS tree ?  It caused a lot of oopses here before
> > I worked around it (patch is at
> > ftp://linux-parisc.sourceforge.net/pub/linux-parisc/) (why we got oopses
> > and what is wrong should be quite obvious if you know RTL - I don't and
> > it seems reasonably obvious to me, I just can't fix it).
> 
> No, I haven't done anything with gcc

Anyone else ?  I think you (plural) should at least have a look causing
random data corruption.

> Linuxcare.  Support for the Revolution.

And it's just in historical spirit that those actually supporting the
revolution lose their jobs ... (yes, I think the signature is quite 
inconsiderate.)


	Philipp Rumpf

Re: [parisc-linux] New binutils available from cvs

[Alan Modra]

On Thu, 18 May 2000, Philipp Rumpf wrote:

> > Using `!' for comments
> > a) breaks fp condition codes.
> 
> No, it doesn't.  The condition code for fp "not equal" is "<>", not "!=".

I'm talking about floating point compare instruction conditions, for
example: "fcmp,dbl,!?>= %fr4,%fr5"

-- 
Linuxcare.  Support for the Revolution.

Re: [parisc-linux] New binutils available from cvs

[Philipp Rumpf]

> I'm talking about floating point compare instruction conditions, for
> example: "fcmp,dbl,!?>= %fr4,%fr5"

Oh, I see.  (I stupidly only checked the condition codes tables in the
architecture manual, the floating point conditions were in the FPU
section).

So, '!' isn't usable as either line comment or '\n'-equivalent
character without hacking.  Let's just not use it ?

	Philipp Rumpf

Re: [parisc-linux] New binutils available from cvs

[Alan Modra]

On Thu, 18 May 2000, Philipp Rumpf wrote:

> > I'm talking about floating point compare instruction conditions, for
> > example: "fcmp,dbl,!?>= %fr4,%fr5"
> 
> Oh, I see.  (I stupidly only checked the condition codes tables in the
> architecture manual, the floating point conditions were in the FPU
> section).
> 
> So, '!' isn't usable as either line comment or '\n'-equivalent
> character without hacking.  Let's just not use it ?

It works quite OK as a line separator.  The assembler line separator is
not a '\n' equivalent.

-- 
Linuxcare.  Support for the Revolution.

Re: [parisc-linux] New binutils available from cvs

[Jeffrey A Law]

  In message <Pine.LNX.4.21.0005182055320.6449-100000@front.linuxcare.com.au>yo
u write:
  > On Thu, 18 May 2000, Philipp Rumpf wrote:
  > 
  > > > Using `!' for comments
  > > > a) breaks fp condition codes.
  > > 
  > > No, it doesn't.  The condition code for fp "not equal" is "<>", not "!=".
  > 
  > I'm talking about floating point compare instruction conditions, for
  > example: "fcmp,dbl,!?>= %fr4,%fr5"
Alan is correct.  A large number of the FP comparisons use '!'.  

jeff

Re: [parisc-linux] New binutils available from cvs

[Grant Grundler]

Philipp Rumpf wrote:
> > b) is inconsistent with existing hp assembler.
> 
> Your point ?

I think Alan meant it breaks with existing GNU toolchain for parisc.

And even if he didn't, asking people to change assembler files
for "internal" reasons (glibc/whatever) seems like an extra hurdle
that just doesn't have to be. I'd like to make it easier for people
to migrate from HP-UX (OpenBSD/FreeBSD/mklinux/whatever) to parisc-linux.

...
> All in all even _if_ you have a reason to change it back it could have used
> some discussion.  So far you haven't.

Alan has.  But no one replied publicly. See
http://puffin.external.hp.com/mailing-lists/parisc-linux/2000/05-May/0038.html

Paul Bame and I both replied privately that he should go ahead
and make the change. So you can blame us if you like. :^)

...
> Any idea which illegal drug the person who decided "!=" was the fp not
> equal condition code was on ;) ?

No...but I suspect the FP was developed *seperately* from the PA-RISC CPU.
Don't need drugs to get divergence - it's almost guaranteed to happen
one way or another.

> > Linuxcare.  Support for the Revolution.
> 
> And it's just in historical spirit that those actually supporting the
> revolution lose their jobs ... (yes, I think the signature is quite 
> inconsiderate.)

Without companies *like* Linuxcare, linux wouldn't have a rat's ass
chance in the business world. I think the signature is quite accurate.
And anyone with even 1% clue about linux is certainly not having
employment problems right now.  (I'm proof of that! :^)

grant

Grant Grundler
Unix Development Lab
+1.408.447.7253

Re: [parisc-linux] New binutils available from cvs

[Tor Arntsen]

On May 18, 11:45, Alan Modra wrote:
>On Thu, 18 May 2000, Philipp Rumpf wrote:
>> Any reason for renaming cvs modules all the time instead of staying with
>> one name and perhaps not pissing off people with slow, expensive
>> connections ?
>
>Mainly so people can still easily use the old binutils.  .diff.gz came to
[...]

That's what tags and branches are for!  *Please* use CVS the way it was
meant to be used, don't use it like some old SCCS system!

It's now close to impossible to track these things.  cvs co -c doesn't show
the modules either, without keeping *very* close tracking of the mailing
list we won't even know what we should/can use.

>1.8M anyway.  I can put the diff somewhere you can get it, if you like.

Sigh.  What *for*?  See above, that's why CVS has branches and tags which
makes it easy for users to:
a) See what's in there (cvs status -v)
b) Switch between versions with ease and with minimal bandwidth
c) Keep several versions locally (copy a tree and switch to other branch or 
   rev.)

-Tor

Re: [parisc-linux] New binutils available from cvs

[willy]

On Thu, May 18, 2000 at 07:26:51PM +0200, Tor Arntsen wrote:
> That's what tags and branches are for!  *Please* use CVS the way it was
> meant to be used, don't use it like some old SCCS system!

the trouble is that none of us are sufficiently skilled with CVS to cope
with the tags and branches.  there've been sufficient problems before when
we've tried to use them to convince us that it really isn't worthwhile
for this particular developmnt community.  i should note that prumpf was
one of the people who complained the loudest about the use of CVS tags.