[Linux-ia64] Re: compiler version for kernel builds

[Linux-ia64] Re: compiler version for kernel builds

[David Mosberger]

>>>>> On Thu, 14 Mar 2002 15:46:04 -0800, Jesse Barnes <jbarnes@sgi.com> said:

  Jesse> David, which compiler version do you use for kernel builds
  Jesse> these days?

I usually use 3.0, though I try to remember to test with 2.96 from
time to time.  I haven't had much luck with 3.1 yet: it builds a
working kernel for the simulator, but the kernel for the real machine
dies early on.  Haven't had time yet to investigate.

  Jesse> Also, is there a document somewhere that
  Jesse> describes building that version as a cross compiler?  Last
  Jesse> time I tried building 3.0.2 as an x86 hosted ia64 cross
  Jesse> compiler, I ran into some trouble.

We posted something on this list a while ago (search for "recipe" or
something like that).  We haven't built a cross-compiler recently, so
the recipe may not be perfectly accurate anymore.

	--david

Re: [Linux-ia64] Re: compiler version for kernel builds

[Gary Hade]

On Thu, Mar 14, 2002 at 07:56:07PM -0800, David Mosberger wrote:
> >>>>> On Thu, 14 Mar 2002 15:46:04 -0800, Jesse Barnes <jbarnes@sgi.com> said:
> 
>   Jesse> David, which compiler version do you use for kernel builds
>   Jesse> these days?
> 
> I usually use 3.0, though I try to remember to test with 2.96 from
> time to time.  I haven't had much luck with 3.1 yet: it builds a
> working kernel for the simulator, but the kernel for the real machine
> dies early on.  Haven't had time yet to investigate.

I can confirm that gcc-3_1-branch (as of yesterday) is still 
unable to build a working kernel.

A few days ago the most current 3.1 source wouldn't even compile 
the 2.4.17 kernel source.  It was dying with an internal compiler 
error while attempting to compile linux/fs/jbd/journal.c.  I 
submitted GCC PR 5892 which was fixed on 3/13. 

Using a gcc built yesterday from the gcc-3_1-branch containing 
the above fix I hit a link-time error due to an unresolved reference 
to __xchg_called_with_bad_pointer.  We determined that this problem 
was due to a change in the compiler inlining behavior where 
static inline functions over a certain size are no longer being 
inlined by default.  It appears that we will need to use 
-finline-limit=n to force inlining of larger functions.  After 
adding -finline-limit 00 the kernel compiles and links.  
Unfortunately, it hangs right away when you try to boot it.

I'll let you if we learn more.

Gary

> 
>   Jesse> Also, is there a document somewhere that
>   Jesse> describes building that version as a cross compiler?  Last
>   Jesse> time I tried building 3.0.2 as an x86 hosted ia64 cross
>   Jesse> compiler, I ran into some trouble.
> 
> We posted something on this list a while ago (search for "recipe" or
> something like that).  We haven't built a cross-compiler recently, so
> the recipe may not be perfectly accurate anymore.
> 
> 	--david
> 
> 
> _______________________________________________
> Linux-IA64 mailing list
> Linux-IA64@linuxia64.org
> http://lists.linuxia64.org/lists/listinfo/linux-ia64

-- 
Gary Hade
IBM Linux Technology Center
503-578-4503  IBM T/L: 775-4503
garyhade@us.ibm.com
http://www.ibm.com/linux/ltc

Re: [Linux-ia64] Re: compiler version for kernel builds

[Jes Sorensen]

David Mosberger <davidm@napali.hpl.hp.com> writes:

> >>>>> On Thu, 14 Mar 2002 15:46:04 -0800, Jesse Barnes <jbarnes@sgi.com> said:
> 
>   Jesse> David, which compiler version do you use for kernel builds
>   Jesse> these days?
> 
> I usually use 3.0, though I try to remember to test with 2.96 from
> time to time.  I haven't had much luck with 3.1 yet: it builds a
> working kernel for the simulator, but the kernel for the real machine
> dies early on.  Haven't had time yet to investigate.

I have just tested 3.0.5 (20020316) and I can't get a booting
2.5.7pre1 kernel built with that toolchain.

Seems it's not just 3.1+ thats borked ;-(

Jes
c

Re: [Linux-ia64] Re: compiler version for kernel builds

[Gary Hade]

FYI, earlier this afternoon I was able to successfully build 
and boot an IA-64 2.4.18 kernel built with:
	gcc version 3.1 20020327 (prerelease)
 
The GCC problem that was previously preventing a successful 
boot was isolated by Chris McDermott and Steve Christiansen.
It was fixed earlier today by Richard Henderson.
  GCC PR 6054 (GCC 3.1 for ia64 fails to restore gp after 
                 indirect call in Linux kernel)
 
Gary


On Fri, Mar 15, 2002 at 02:44:36PM -0800, Gary Hade wrote:
> On Thu, Mar 14, 2002 at 07:56:07PM -0800, David Mosberger wrote:
> > >>>>> On Thu, 14 Mar 2002 15:46:04 -0800, Jesse Barnes <jbarnes@sgi.com> said:
> > 
> >   Jesse> David, which compiler version do you use for kernel builds
> >   Jesse> these days?
> > 
> > I usually use 3.0, though I try to remember to test with 2.96 from
> > time to time.  I haven't had much luck with 3.1 yet: it builds a
> > working kernel for the simulator, but the kernel for the real machine
> > dies early on.  Haven't had time yet to investigate.
> 
> I can confirm that gcc-3_1-branch (as of yesterday) is still 
> unable to build a working kernel.
> 
> A few days ago the most current 3.1 source wouldn't even compile 
> the 2.4.17 kernel source.  It was dying with an internal compiler 
> error while attempting to compile linux/fs/jbd/journal.c.  I 
> submitted GCC PR 5892 which was fixed on 3/13. 
> 
> Using a gcc built yesterday from the gcc-3_1-branch containing 
> the above fix I hit a link-time error due to an unresolved reference 
> to __xchg_called_with_bad_pointer.  We determined that this problem 
> was due to a change in the compiler inlining behavior where 
> static inline functions over a certain size are no longer being 
> inlined by default.  It appears that we will need to use 
> -finline-limit=n to force inlining of larger functions.  After 
> adding -finline-limit 00 the kernel compiles and links.  
> Unfortunately, it hangs right away when you try to boot it.
> 
> I'll let you if we learn more.
> 
> Gary
> 
> > 
> >   Jesse> Also, is there a document somewhere that
> >   Jesse> describes building that version as a cross compiler?  Last
> >   Jesse> time I tried building 3.0.2 as an x86 hosted ia64 cross
> >   Jesse> compiler, I ran into some trouble.
> > 
> > We posted something on this list a while ago (search for "recipe" or
> > something like that).  We haven't built a cross-compiler recently, so
> > the recipe may not be perfectly accurate anymore.
> > 
> > 	--david
> > 
> > 
> > _______________________________________________
> > Linux-IA64 mailing list
> > Linux-IA64@linuxia64.org
> > http://lists.linuxia64.org/lists/listinfo/linux-ia64
> 
> -- 
> Gary Hade
> IBM Linux Technology Center
> 503-578-4503  IBM T/L: 775-4503
> garyhade@us.ibm.com
> http://www.ibm.com/linux/ltc
> 
> _______________________________________________
> Linux-IA64 mailing list
> Linux-IA64@linuxia64.org
> http://lists.linuxia64.org/lists/listinfo/linux-ia64

Re: [Linux-ia64] Re: compiler version for kernel builds

[David Mosberger]

>>>>> On Wed, 27 Mar 2002 15:45:39 -0800, Gary Hade <garyhade@us.ibm.com> said:

  Gary> FYI, earlier this afternoon I was able to successfully build
  Gary> and boot an IA-64 2.4.18 kernel built with: gcc version 3.1
  Gary> 20020327 (prerelease)

  Gary> The GCC problem that was previously preventing a successful
  Gary> boot was isolated by Chris McDermott and Steve Christiansen.
  Gary> It was fixed earlier today by Richard Henderson.  GCC PR 6054
  Gary> (GCC 3.1 for ia64 fails to restore gp after indirect call in
  Gary> Linux kernel)

Very good.  One thing less to worry about.

Thanks,

	--david