Skip to contentSkip to conversation

x86: don't compile with gcc-3.3.3

RFC PATCH In review [email protected] 4 participants 11 messages Sep 13, 2010 x86

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 x86: don't compile with gcc-3.3.3

hpa commented on bug 16506[1] :
"Please note that gcc-3.3.3 is known broken on x86; gcc-3.4 is the oldest version which is known to *not* be broken."

References:

	[0]: https://bugzilla.kernel.org/show_bug.cgi?id=16633
	[1]: https://bugzilla.kernel.org/show_bug.cgi?id=16506#c28

If that is indeed so, we should abort the build? No?

---
 arch/x86/Makefile |    8 +++++++-
 1 files changed, 7 insertions(+), 1 deletions(-)
arch/x86/Makefile +7 -1
index 8aa1b59..1eb2d99 100644
@@ -153,7 +153,7 @@ all: bzImage
 # KBUILD_IMAGE specify target image being built
 KBUILD_IMAGE := $(boot)/bzImage
 
-bzImage: vmlinux
+bzImage: checkgccversion vmlinux
 ifeq ($(CONFIG_X86_DECODER_SELFTEST),y)
 	$(Q)$(MAKE) $(build)=arch/x86/tools posttest
 endif
@@ -172,6 +172,12 @@ PHONY += vdso_install
 vdso_install:
 	$(Q)$(MAKE) $(build)=arch/x86/vdso $@
 
+checkgccversion:
+	$(Q)if test "$(call cc-fullversion)" = "040403" ; then \
+		echo "*** GCC-3.3.3 is known broken. Please upgrade." ; \
+		echo "*** GCC-3.4 is the earliest version known *not* broken." ; \
+		false ; \
+	fi
 archclean:
 	$(Q)rm -rf $(objtree)/arch/i386
 	$(Q)rm -rf $(objtree)/arch/x86_64
Signature · 1 line
1.7.2
#1 lore raw
Quoted 9 lines · On Mon, 2010-09-13 at 10:31 +0200, [email protected] wrote:

hpa commented on bug 16506[1] :
"Please note that gcc-3.3.3 is known broken on x86; gcc-3.4 is the oldest version which is known to *not* be broken."

References:

	[0]: https://bugzilla.kernel.org/show_bug.cgi?id=16633
	[1]: https://bugzilla.kernel.org/show_bug.cgi?id=16506#c28

If that is indeed so, we should abort the build? No?

Does it in fact still build with 3.4? I seem to recall some talk about pushing the minimum version to 4.x for x86, although I can't remember where..

#2 lore raw
Quoted 9 lines · Peter Zijlstra
Quoted 8 lines · On Mon, 2010-09-13 at 10:31 +0200, [email protected] wrote:

hpa commented on bug 16506[1] :
"Please note that gcc-3.3.3 is known broken on x86; gcc-3.4 is the oldest version which is known to *not* be broken."

References:

	[0]: https://bugzilla.kernel.org/show_bug.cgi?id=16633
	[1]: https://bugzilla.kernel.org/show_bug.cgi?id=16506#c28
Quoted 5 lines
Quoted 1 line

If that is indeed so, we should abort the build? No?

Does it in fact still build with 3.4? I seem to recall some talk about pushing the minimum version to 4.x for x86, although I can't remember where..

It does indeed still build with 3.4, although it is giving us a bunch of headaches to *make it so*, and I would personally be really glad if the consensus is we can just axe it.

The only reason to not abort the build for gcc 3.3.3 is that soem "enterprise" distros have been shipping gcc 3.3.3 with backported fixes from 3.4, which of course still identifies themselves as gcc 3.3.3, and so technically it is iompossioble to tell if any particular "gcc 3.3.3" is actually broken or not.

However, as far as I can tell, most of the people who build current x86 kernels with gcc 3.x are people who are testing building current kernels with gcc 3.x. Some of the embedded systems are different, because for some strange reasons most of the embedded world seem stuck on gcc 3.4 or so.

-hpa

Signature · 2 lines
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.
#3 lore raw
Quoted 26 lines · On Mon, 13 Sep 2010 08:56:39 -0700 "H. Peter Anvin" <[email protected]> wrote:
Quoted 9 lines · Peter Zijlstra
Quoted 8 lines · On Mon, 2010-09-13 at 10:31 +0200, [email protected] wrote:

hpa commented on bug 16506[1] :
"Please note that gcc-3.3.3 is known broken on x86; gcc-3.4 is the oldest version which is known to *not* be broken."

References:

	[0]: https://bugzilla.kernel.org/show_bug.cgi?id=16633
	[1]: https://bugzilla.kernel.org/show_bug.cgi?id=16506#c28
Quoted 5 lines
Quoted 1 line

If that is indeed so, we should abort the build? No?

Does it in fact still build with 3.4? I seem to recall some talk about pushing the minimum version to 4.x for x86, although I can't remember where..

It does indeed still build with 3.4, although it is giving us a bunch of headaches to *make it so*, and I would personally be really glad if the consensus is we can just axe it.

The only reason to not abort the build for gcc 3.3.3 is that soem "enterprise" distros have been shipping gcc 3.3.3 with backported fixes from 3.4, which of course still identifies themselves as gcc 3.3.3, and so technically it is iompossioble to tell if any particular "gcc 3.3.3" is actually broken or not.

Quoted 4 lines

However, as far as I can tell, most of the people who build current x86 kernels with gcc 3.x are people who are testing building current kernels with gcc 3.x. Some of the embedded systems are different, because for some strange reasons most of the embedded world seem stuck on gcc 3.4 or so.

Well, all _I_ can add, that there are people (2 I know
from the above bug reports to be precise) who spent hours (? well I guess, at least some minutes) debugging old gcc bugs and just to have some (please take no offense, I'm just trying to tell how it must look to them) random guy tell them in the bugzilla "na na, we know about this already and we don't fix it, na na" ...

Also the patch above checks actually for 4.4.3 which is obvioulsy wrong ;)

The enterprise distro issue is a little bit of an annoyance.. but they should perhaps give us an opportunity to distinguish between their (working) gcc and the stock broken one. I'm totally not opposed to any ugly hack to make that distinguishing over here. But not fixing the issues above seems to hurt the mainline linux kernel.

Cheers,
Flo

Quoted 3 lines

-hpa

#4 lore raw
Quoted 10 lines · On Mon, 13 Sep 2010 18:21:05 +0200 Florian Mickler <[email protected]> wrote:
Quoted 7 lines · On Mon, 13 Sep 2010 08:56:39 -0700 "H. Peter Anvin" <[email protected]> wrote:

On 09/13/2010 01:39 AM, Peter Zijlstra wrote:

The only reason to not abort the build for gcc 3.3.3 is that soem "enterprise" distros have been shipping gcc 3.3.3 with backported fixes from 3.4, which of course still identifies themselves as gcc 3.3.3, and so technically it is iompossioble to tell if any particular "gcc 3.3.3" is actually broken or not.

What about issuing a warning like

"gcc 3.3.3 is broken because of bug [bugdescription/url]. if you know your build is not affected by this bug do:
touch .gcc333bugfixNNNN but please be aware that you are not supported by the mainline kernel development comunity if you do"

and implementing the accompanying logic?

Cheers,
Flo

#5 lore raw
Quoted 11 lines · Florian Mickler

What about issuing a warning like

"gcc 3.3.3 is broken because of bug [bugdescription/url]. if you know your build is not affected by this bug do:
touch .gcc333bugfixNNNN but please be aware that you are not supported by the mainline kernel development comunity if you do"

and implementing the accompanying logic?

URK.

The workload of maintaining that crap would be astonishing, especially for something at the tail end of supportable.

I don't think that is even remotely realistic.

-hpa

#6 lore raw
Quoted 13 lines · On Mon, 13 Sep 2010 10:25:43 -0700 "H. Peter Anvin" <[email protected]> wrote:
Quoted 11 lines · Florian Mickler

What about issuing a warning like

"gcc 3.3.3 is broken because of bug [bugdescription/url]. if you know your build is not affected by this bug do:
touch .gcc333bugfixNNNN but please be aware that you are not supported by the mainline kernel development comunity if you do"

and implementing the accompanying logic?

Quoted 8 lines

URK.

The workload of maintaining that crap would be astonishing, especially for something at the tail end of supportable.

I don't think that is even remotely realistic.

-hpa

I can imagine that even without having expirienced anything alike :)

Should I then resend this RFC-Patch updated to abort the x86-build for gcc -lt 3.4?

Regards,
Flo

#7 lore raw
Quoted 6 lines · Florian Mickler

I can imagine that even without having expirienced anything alike :)

Should I then resend this RFC-Patch updated to abort the x86-build for gcc -lt 3.4?

I'm doing a patch series which is a little more complete; I'll post it shortly.

-hpa

#8 lore raw
Quoted 4 lines · On Mon, 13 Sep 2010 13:00:57 -0700 "H. Peter Anvin" <[email protected]> wrote:
Quoted 3 lines · Florian Mickler

I can imagine that even without having expirienced anything alike :)

Quoted 8 lines
Quoted 3 lines

Should I then resend this RFC-Patch updated to abort the x86-build for gcc -lt 3.4?

I'm doing a patch series which is a little more complete; I'll post it shortly.

-hpa

ok, cool.

Regards,
Flo

#9 lore raw
In reply to Peter Zijlstra
Quoted 9 lines · Peter Zijlstra
Quoted 8 lines · On Mon, 2010-09-13 at 10:31 +0200, [email protected] wrote:

hpa commented on bug 16506[1] :
"Please note that gcc-3.3.3 is known broken on x86; gcc-3.4 is the oldest version which is known to *not* be broken."

References:

	[0]: https://bugzilla.kernel.org/show_bug.cgi?id=16633
	[1]: https://bugzilla.kernel.org/show_bug.cgi?id=16506#c28
Quoted 5 lines
Quoted 1 line

If that is indeed so, we should abort the build? No?

Does it in fact still build with 3.4? I seem to recall some talk about pushing the minimum version to 4.x for x86, although I can't remember where..

I'm still building kernels with gcc 3.4.3 (for the ARMs which don't require a later compiler) or gcc 4.3.2+patches for those which do.

ARM gcc 3.4.3 is certainly noticably faster than gcc 4.3.2 even on x86 - and as long as ARM gcc 3.4.3 works, and there's not great pain in allowing it to build the arch-independent stuff, I see no reason to deny it across the entire kernel build.

I think we should have a minimum compiler version for the generic kernel, and individual minimum compiler versions for the architectures, so that arches can specify a higher minimum compiler version if they have specific problems there.

#10 lore raw
Quoted 13 lines · Russell King - ARM Linux
Quoted 4 lines

Does it in fact still build with 3.4? I seem to recall some talk about pushing the minimum version to 4.x for x86, although I can't remember where..

I'm still building kernels with gcc 3.4.3 (for the ARMs which don't require a later compiler) or gcc 4.3.2+patches for those which do.

ARM gcc 3.4.3 is certainly noticably faster than gcc 4.3.2 even on x86 - and as long as ARM gcc 3.4.3 works, and there's not great pain in allowing it to build the arch-independent stuff, I see no reason to deny it across the entire kernel build.

Quoted 4 lines

I think we should have a minimum compiler version for the generic kernel, and individual minimum compiler versions for the architectures, so that arches can specify a higher minimum compiler version if they have specific problems there.

Yes, I don't think anyone has proposed bumping past 3.4 for the generic kernel at this time.

-hpa