updated gcc-3.4 patches for 2.4.27-rc4

updated gcc-3.4 patches for 2.4.27-rc4

[Mikael Pettersson]

[Resend. 1st post seems to have gone into a /dev/null somewhere.]

The gcc-3.4 patches for the 2.4.27-rc4 kernel have been updated:
http://www.csd.uu.se/~mikpe/linux/patches/2.4/patch-gcc34-fastcall-fixes-2.4.27-rc4
http://www.csd.uu.se/~mikpe/linux/patches/2.4/patch-gcc34-lvalue-fixes-2.4.27-rc4
http://www.csd.uu.se/~mikpe/linux/patches/2.4/patch-gcc34-misc-fixes-2.4.27-rc4
Download and apply all three of them.

After a careful review of all patches, the following changes were made:
- i386' smp_send_reschedule() now has proper fastcall declarations.
  The previous patches removed the FASTCALL from <linux/smp.h>,
  but the new patches instead preserve the behaviour from gcc < 3.4.
- Several functions had their fastcall annotations moved around a
  little to match exactly their 2.6.8-rc2 versions.
- A few void* casts used for cast-as-lvalue fixes were reformatted
  to match exactly their 2.6.8-rc2 versions.
- Cleaned up the cast-as-lvalue fix in
  arch/x86_64/ia32/ia32_ioctl.c:blkpg_ioctl_trans().

The cast-as-lvalue fixes for kernel/sysctl.c result in code doing
pointer arithmetic on void* values. There has been some discussion
about replacing those with char* values. I decided against that
because (1) the diff became larger, and (2) the functions in sysctl.c
already do void* pointer arithmetic, and there is no strong reason
for eliminating that (ab)use in the 2.4 kernel.

The cast-as-lvalue fixes are also needed to silence warnings from
gcc-3.3.4.

/Mikael

Re: updated gcc-3.4 patches for 2.4.27-rc4

[Adrian Bunk]

On Tue, Aug 03, 2004 at 09:02:22PM +0200, Mikael Pettersson wrote:
> [Resend. 1st post seems to have gone into a /dev/null somewhere.]
> 
> The gcc-3.4 patches for the 2.4.27-rc4 kernel have been updated:
>...
> http://www.csd.uu.se/~mikpe/linux/patches/2.4/patch-gcc34-misc-fixes-2.4.27-rc4
>...

This still contains the compiler.h patch which shouldn't be there.

> /Mikael

cu
Adrian

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed

Re: updated gcc-3.4 patches for 2.4.27-rc4

[Mikael Pettersson]

On Tue, 3 Aug 2004 21:34:54 +0200, Adrian Bunk wrote:
> On Tue, Aug 03, 2004 at 09:02:22PM +0200, Mikael Pettersson wrote:
> > [Resend. 1st post seems to have gone into a /dev/null somewhere.]
> > 
> > The gcc-3.4 patches for the 2.4.27-rc4 kernel have been updated:
> >...
> > http://www.csd.uu.se/~mikpe/linux/patches/2.4/patch-gcc34-misc-fixes-2.4.27-rc4
> >...
> 
> This still contains the compiler.h patch which shouldn't be there.

It's currently needed since the inlining failures caused by
__attribute__((always_inline)) haven't been fixed yet in 2.4.
This patch kit is used for production, you know :-) Things
must work.

I'll remove the compiler.h patch later when I have time to deal
with the inlining failures.

/Mikael

Re: updated gcc-3.4 patches for 2.4.27-rc4

[O.Sezer]

 > 
http://www.csd.uu.se/~mikpe/linux/patches/2.4/patch-gcc34-misc-fixes-2.4.27-rc4

Does /drivers/usb/gadget/net2280.c line 794 not need the same
change for min() ?
FWIW: The -pac tree does a #define MIN(x,y) ((x) > (y) ? (y) : (x))
at the beginning and then changes the three instances of min() to
MIN()

Özkan Sezer

Re: updated gcc-3.4 patches for 2.4.27-rc4

[Mikael Pettersson]

On Wed, 04 Aug 2004 01:25:26 +0300, O.Sezer wrote:
> > 
>http://www.csd.uu.se/~mikpe/linux/patches/2.4/patch-gcc34-misc-fixes-2.4.27-rc4
>
>Does /drivers/usb/gadget/net2280.c line 794 not need the same
>change for min() ?

No. The min() that was changed had an argument that was a
bitfield in a struct. min() uses typeof, but typeof (like
sizeof) doesn't work on bitfields.

The arguments to the min() at line 794 are plain variables
or struct fields. They don't have any problem with typeof.

/Mikael