Re: powerpc ABI change
On Wed, 2014-05-14 at 18:30 +0200, Bastian Blank wrote: > On Wed, May 14, 2014 at 02:05:34PM +0100, Ben Hutchings wrote: > > Since 'asm goto' has been around for a while, I suspect that the real > > failure was in linking a 64-bit executable on 32-bit powerpc. > > Can't be. This script is called without any parameter, only the > compiler. Also it does not link binaries, only compile it. OK, then I have no idea why this check failed. I have worked around it, anyhow. Ben. -- Ben Hutchings We get into the habit of living before acquiring the habit of thinking. - Albert Camus signature.asc Description: This is a digitally signed message part
Re: powerpc ABI change
On Wed, May 14, 2014 at 02:05:34PM +0100, Ben Hutchings wrote: > Since 'asm goto' has been around for a while, I suspect that the real > failure was in linking a 64-bit executable on 32-bit powerpc. Can't be. This script is called without any parameter, only the compiler. Also it does not link binaries, only compile it. Bastian -- I have never understood the female capacity to avoid a direct answer to any question. -- Spock, "This Side of Paradise", stardate 3417.3 -- To UNSUBSCRIBE, email to debian-kernel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: https://lists.debian.org/20140514163014.ga24...@mail.waldi.eu.org
Re: powerpc ABI change
On Tue, 2014-05-13 at 22:11 +0200, Bastian Blank wrote: > On Tue, May 13, 2014 at 09:04:45PM +0200, Bastian Blank wrote: > > Are you sure that powerpc is the only affected? > > This looks pretty much like an unstable compiler support check. This > script hides any output, so we have no way to see what happened: > | ifeq ($(shell $(CONFIG_SHELL) $(srctree)/scripts/gcc-goto.sh $(CC)), y) > | KBUILD_CFLAGS += -DCC_HAVE_ASM_GOTO > | endif > > The list of added symbols is consistent with support for jump labels and > this support adds two fields to "struct module", so almost every symbol > is affected. I've worked around this for 3.14.4-1 by forcing CONFIG_JUMP_LABEL off for powerpc/powerpc64 (where the test failed) and adding a version to the compiler build-dependency for ppc64 (where the test passed). Since 'asm goto' has been around for a while, I suspect that the real failure was in linking a 64-bit executable on 32-bit powerpc. Ben. -- Ben Hutchings We get into the habit of living before acquiring the habit of thinking. - Albert Camus signature.asc Description: This is a digitally signed message part
Re: powerpc ABI change
On Tue, May 13, 2014 at 09:04:45PM +0200, Bastian Blank wrote: > Are you sure that powerpc is the only affected? This looks pretty much like an unstable compiler support check. This script hides any output, so we have no way to see what happened: | ifeq ($(shell $(CONFIG_SHELL) $(srctree)/scripts/gcc-goto.sh $(CC)), y) | KBUILD_CFLAGS += -DCC_HAVE_ASM_GOTO | endif The list of added symbols is consistent with support for jump labels and this support adds two fields to "struct module", so almost every symbol is affected. Bastian -- Vulcans worship peace above all. -- McCoy, "Return to Tomorrow", stardate 4768.3 signature.asc Description: Digital signature
Re: powerpc ABI change
On Mon, May 12, 2014 at 02:00:29AM +0100, Ben Hutchings wrote: > I tried building 3.14.3-1 on powerpc, and that failed with a large ABI > change (about 7000 out of 13000 symbols changed). Okay, I can reproduce this with 3.14.2-1 in the powerpc64 image (not the powerpc image). > So I think that this is due to a compiler change that results in > different type definitions after preprocessing: either a header is > testing for a newly defined macro, or Kbuild is automatically using a > new compiler option which in turn affects conditions in a header. Are you sure that powerpc is the only affected? Bastian -- You can't evaluate a man by logic alone. -- McCoy, "I, Mudd", stardate 4513.3 -- To UNSUBSCRIBE, email to debian-kernel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: https://lists.debian.org/20140513190445.ga19...@mail.waldi.eu.org
powerpc ABI change
I tried building 3.14.3-1 on powerpc, and that failed with a large ABI change (about 7000 out of 13000 symbols changed). After searching for the source of the ABI change and finding no relevant source or config changes, I tried rebuilding 3.14.2-1 with the ABI reference files in place, and that also failed with the same changes (minus a few that I understand and can be ignored). Surprisingly, 3.14.2-1 was built with gcc-4.8 4.8.2-14 which is several months out of date. I'm now building with 4.8.2-21, which has many backported changes related to the powerpc port (and others). So I think that this is due to a compiler change that results in different type definitions after preprocessing: either a header is testing for a newly defined macro, or Kbuild is automatically using a new compiler option which in turn affects conditions in a header. If you can't work it out either, it should be possible to bump the ABI version for a single architecture. But we haven't tried that for a while. Ben. -- Ben Hutchings Sturgeon's Law: Ninety percent of everything is crap. signature.asc Description: This is a digitally signed message part
