Re: [PATCHSET] 2.6.15-rc7-bird1

[Al Viro]

BTW, here's the state of toolchains, by architecture:

alpha		alpha-linux	works
arm		arm-linux	works
arm26				support is gone in gcc4, tree is not well either
cris		cris-linux	broken [*3][*4], but gets through arch/cris, so
				there's some use for it...
frv		frv-linux	4.1 mostly works [*2], but "U" constraint
				support is still missing in mainline
h8300		h8300-elf	broken [*5][*3]
i386		x86_64-linux	works
ia64		ia64-linux	works
m32r		m32r-linux	works [*2]
m68k		m68k-linux	works [*2]
mips		mips64-linux	works
mips64		mips64-linux	works [*1]
parisc		hppa-linux	works [*4]; gcc-4.1 gives an ICE [*7]
parisc64	hppa64-linux	Linus' tree is broken; gcc builds, but doesn't
				get far in kernel build; might die later.
powerpc		powerpc64-linux	works
ppc		powerpc64-linux	works
s390		s390x-linux	works
sh		sh-linux	works [*4]
sh64		sh64-linux	broken [*6]
sparc		sparc64-linux	works
sparc64		sparc64-linux	works
um              toolchain of corresponding sub-arch
v850		v850e-linux	works [*2][*5]
x86_64		x86_64-linux	works
xtensa, m68knommu: have not tried lately

[*1] except that we get bogus flags passed to as(1) if we are building
for 64bit cpu without CONFIG_BUILD_ELF64.
[*2] 64bit bugs in binutils, fixes exist
[*3] ICE during kernel build
[*4] needs a patch to get rid of dependency on target libc headers; fix exists
[*5] 64bit bugs in gcc, fixes exist
[*6] ICE during libgcc build
[*7] 4.0 works, 4.1 gives an ICE when you build on 64bit host with -Os;
trimmed-down reproducer is
void foo(int sig, int *p)
{
	if (!((1ULL << ((sig)-1)) & 0x300000edcULL))
		*p = 0;
}