make AT_SYSINFO platform-independent

make AT_SYSINFO platform-independent

[David Mosberger]

How about moving the AT_SYSINFO macro from asm-i386/elf.h to
linux/elf.h?  Several architectures can benefit from it (certainly
pa-risc and ia64) and since glibc also defines it in a
non-platformspecific fashion, there really is no point not doing the
same in the kernel.  I suppose it would be nice if we could renumber
it from 32 to 18, but that would require updating glibc, which is
probably too painful.

	--david

===== include/asm-i386/elf.h 1.5 vs edited =====
--- 1.5/include/asm-i386/elf.h	Thu Jan  2 07:22:48 2003
+++ edited/include/asm-i386/elf.h	Fri Jan 10 22:40:15 2003
@@ -96,12 +96,6 @@
 
 #define ELF_PLATFORM  (system_utsname.machine)
 
-/*
- * Architecture-neutral AT_ values in 0-17, leave some room
- * for more of them, start the x86-specific ones at 32.
- */
-#define AT_SYSINFO	32
-
 #ifdef __KERNEL__
 #define SET_PERSONALITY(ex, ibcs2) set_personality((ibcs2)?PER_SVR4:PER_LINUX)
 
===== include/linux/elf.h 1.16 vs edited =====
--- 1.16/include/linux/elf.h	Thu Dec 26 14:07:53 2002
+++ edited/include/linux/elf.h	Fri Jan 10 22:40:45 2003
@@ -182,6 +182,7 @@
 #define AT_PLATFORM 15  /* string identifying CPU for optimizations */
 #define AT_HWCAP  16    /* arch dependent hints at CPU capabilities */
 #define AT_CLKTCK 17	/* frequency at which times() increments */
+#define AT_SYSINFO	32	/* pointer to kernel's sysinfo */
 
 typedef struct dynamic{
   Elf32_Sword d_tag;

Re: make AT_SYSINFO platform-independent

[Christoph Hellwig]

On Fri, Jan 10, 2003 at 10:45:26PM -0800, David Mosberger wrote:
> How about moving the AT_SYSINFO macro from asm-i386/elf.h to
> linux/elf.h?  Several architectures can benefit from it (certainly
> pa-risc and ia64) and since glibc also defines it in a
> non-platformspecific fashion, there really is no point not doing the
> same in the kernel.  I suppose it would be nice if we could renumber
> it from 32 to 18, but that would require updating glibc, which is
> probably too painful.

I think it should be updated.  There is no released glibc or stable kernel
with that number yet.

Re: make AT_SYSINFO platform-independent

[Linus Torvalds]

On Sat, 11 Jan 2003, Christoph Hellwig wrote:
>
> On Fri, Jan 10, 2003 at 10:45:26PM -0800, David Mosberger wrote:
> > How about moving the AT_SYSINFO macro from asm-i386/elf.h to
> > linux/elf.h?  Several architectures can benefit from it (certainly
> > pa-risc and ia64) and since glibc also defines it in a
> > non-platformspecific fashion, there really is no point not doing the
> > same in the kernel.  I suppose it would be nice if we could renumber
> > it from 32 to 18, but that would require updating glibc, which is
> > probably too painful.
> 
> I think it should be updated.  There is no released glibc or stable kernel
> with that number yet.

Sounds like an uncommonly bad idea to me, since the range 19-22 is already 
used at least by PPC.

Yeah, 18 itself may be free (although I wonder _why_? Maybe it's some old 
value that is no longer used by the kernel but old binaries may know 
about?) but the fact is that there just isn't room in the low numbers, 
which was why I put it up higher.

		Linus

Re: make AT_SYSINFO platform-independent

[Ulrich Drepper]

Christoph Hellwig wrote:

> I think it should be updated.  There is no released glibc or stable kernel
> with that number yet.

Actually, we've included the support already in some published code.  If
you want to complain to somebody about this, do it to the person who is
responsible for distributing this code.  His email address is

   torvalds@transmeta.com

Make sure you react with the same nastiness as if I would have made the
decision, OK?

-- 
--------------.                        ,-.            444 Castro Street
Ulrich Drepper \    ,-----------------'   \ Mountain View, CA 94041 USA
Red Hat         `--' drepper at redhat.com `---------------------------

Re: make AT_SYSINFO platform-independent

[Christoph Hellwig]

On Sat, Jan 11, 2003 at 10:52:46AM -0800, Ulrich Drepper wrote:
> Christoph Hellwig wrote:
> 
> > I think it should be updated.  There is no released glibc or stable kernel
> > with that number yet.
> 
> Actually, we've included the support already in some published code.  If
> you want to complain to somebody about this, do it to the person who is
> responsible for distributing this code.  His email address is
> 
>    torvalds@transmeta.com
> 
> Make sure you react with the same nastiness as if I would have made the
> decision, OK?

-EWHATSUP

Who is we?  And why should I care who selected that number.  There is no
glibc release yet and no stable kernel release yet with that number.
Interfaces change during 2.5 (see the sys_security and largepage syscall
removal) and that's okay.  There was no nastiness involved in my
suggestion.

Re: make AT_SYSINFO platform-independent

[Ulrich Drepper]

Christoph Hellwig wrote:

> Who is we?  And why should I care who selected that number.

It's not about who selected the number, it's about who's responsible for
distributing binaries which use the currently used number.


> There is no
> glibc release yet and no stable kernel release yet with that number.

Of course there is.  You better don't talk about thinks you don't know.


> Interfaces change during 2.5 (see the sys_security and largepage syscall
> removal) and that's okay.  There was no nastiness involved in my
> suggestion.

You don't get it.  I would never have distributed binaries with the
current code if it wouldn't have been requested along with the guarantee
that the current interface is stable and final.

-- 
--------------.                        ,-.            444 Castro Street
Ulrich Drepper \    ,-----------------'   \ Mountain View, CA 94041 USA
Red Hat         `--' drepper at redhat.com `---------------------------

Re: make AT_SYSINFO platform-independent

[Christoph Hellwig]

On Sat, Jan 11, 2003 at 11:06:41AM -0800, Ulrich Drepper wrote:
> > glibc release yet and no stable kernel release yet with that number.
> 
> Of course there is.  You better don't talk about thinks you don't know.

The latest glibc release, 2.3.1 does not contain AT_SYSINFO support.
I might be not as briliant as you but I'm certainly not stupid.

> > Interfaces change during 2.5 (see the sys_security and largepage syscall
> > removal) and that's okay.  There was no nastiness involved in my
> > suggestion.
> 
> You don't get it.  I would never have distributed binaries with the
> current code if it wouldn't have been requested along with the guarantee
> that the current interface is stable and final.

Sorry, but having not yet fixed interfaces is the whole point of
development kernels.  If vendors can't resist using the latest and gratest
features they get screwed.  RH seems to have the manpower to maintain their
mistakes for a long enough time, so what's the issue?

Re: make AT_SYSINFO platform-independent

[Ulrich Drepper]

Christoph Hellwig wrote:
> If vendors can't resist using the latest and gratest
> features they get screwed.

You still don't get it.  RH hasn't added the support to binaries we ship
because we wanted or needed it but instead because Linus requested it.
I see no reason to not believe Linus when he says an interface is
stable.  And that's exactly what he said (or wrote in this case).

-- 
--------------.                        ,-.            444 Castro Street
Ulrich Drepper \    ,-----------------'   \ Mountain View, CA 94041 USA
Red Hat         `--' drepper at redhat.com `---------------------------

Re: make AT_SYSINFO platform-independent

[Christoph Hellwig]

On Sat, Jan 11, 2003 at 11:24:42AM -0800, Ulrich Drepper wrote:
> I see no reason to not believe Linus when he says an interface is
> stable.  And that's exactly what he said (or wrote in this case).

Linus had no problem with changing his opinion after strong words in the
past :)

But what exactly is the problem with a changing defintion for you.  There
is neither a stable kernel nor a formal glibc release with this defined
out yet.  That means the ABI isn't formalized and the number can be
allocated where it belongs.

Who cares what Linus said in the past as long as Linux development goes
to the right direction?

Re: make AT_SYSINFO platform-independent

[Linus Torvalds]

On Sat, 11 Jan 2003, Christoph Hellwig wrote:
> 
> -EWHATSUP

Ehh, Uli is a bit sensitive, and thinks everybody blames him for 
everything. I told him the AT_SYSINFO interface is stable, so he wants to 
make it clear that _I_ am the one to blame, not him.

Anyway, as I said in another mail, I think 18 is a bad choice, and I'll 
leave it at 32.

		Linus

Re: make AT_SYSINFO platform-independent

[Christoph Hellwig]

On Sat, Jan 11, 2003 at 11:29:14AM -0800, Linus Torvalds wrote:
> 
> On Sat, 11 Jan 2003, Christoph Hellwig wrote:
> > 
> > -EWHATSUP
> 
> Ehh, Uli is a bit sensitive, and thinks everybody blames him for 
> everything. I told him the AT_SYSINFO interface is stable, so he wants to 
> make it clear that _I_ am the one to blame, not him.

Oh, I have absolutely no problem to blame you 8)

> Anyway, as I said in another mail, I think 18 is a bad choice, and I'll 
> leave it at 32.

But please at least fixup the comment that talks about x86-specific entries
starting at 32 then.

Re: make AT_SYSINFO platform-independent

[David Mosberger]

>>>>> On Sat, 11 Jan 2003 11:29:14 -0800 (PST), Linus Torvalds <torvalds@transmeta.com> said:

  Linus> Anyway, as I said in another mail, I think 18 is a bad
  Linus> choice, and I'll leave it at 32.

But you will move it into <linux/elf.h>?

	--david

Re: make AT_SYSINFO platform-independent

[Alan Cox]

On Sat, 2003-01-11 at 18:57, Christoph Hellwig wrote:
> > Actually, we've included the support already in some published code.  If
> > you want to complain to somebody about this, do it to the person who is
> > responsible for distributing this code.  His email address is
> > 
> >    torvalds@transmeta.com
> > 
> > Make sure you react with the same nastiness as if I would have made the
> > decision, OK?
> 
> -EWHATSUP
> 
> Who is we?  And why should I care who selected that number.  There is no

Linus promised it was stable. 

Re: make AT_SYSINFO platform-independent

[Jeff Garzik]

On Sun, Jan 12, 2003 at 12:21:39AM +0000, Alan Cox wrote:
> Linus promised it was stable. 

Sort of like 2.4.0?

Oh sorry.  I really couldn't resist.  ;-)

	Jeff