On Wed, May 2, 2012 at 11:58 AM, Paweł Sikora <pluto@pld-linux.org> wrote: > Hi, > > with the latest binutils release we're observing weird kernel oopses on *i686* architecture. > i've compared the vmlinux and all kernel modules produced by binutils 2.22.52.0.{1,2} > and the only vmlinux differs. afaics there's a diff in size of few elf sections > (.rodata, .rel.rodata, __modver). is it expected or a potential bug? > > here're vmlinux binaries: http://pluto.agmk.net/b.tar.bz2 > Reverting patches from this report: http://sourceware.org/bugzilla/show_bug.cgi?id=13621 (modulo changes to linker.c since then) seem to fix the problem. We're using gcc 4.6.3 and the ooops here happens when something hits the swap. The similiar problem was spotted in Arch linux https://bugs.archlinux.org/task/29694 and their solution was to remove the 13621 patch from arch binutils.
The differences are --- good/System.map 2012-05-02 12:43:53.775355097 -0700 +++ bad/System.map 2012-05-02 12:34:59.780776694 -0700 @@ -50740,7 +50740,7 @@ c0be2640 D dsa_packet_type c0be2680 D edsa_packet_type c0be26c0 D trailer_packet_type c0be2700 D _edata -c0be3000 D __init_begin +c0be3000 A __init_begin c0be3000 T _sinittext c0be3000 T i386_start_kernel c0be307d T reserve_ebda_region @@ -55299,7 +55299,7 @@ c0c7d6c0 d cfd_data c0c7d700 d csd_data c0c7d740 D softnet_data c0c7d800 D __per_cpu_end -c0c7e000 D __init_end +c0c7e000 A __init_end c0c7e000 R __smp_locks c0c84000 B __bss_start c0c84000 R __smp_locks_end The new linker puts __init_begin/__init_end in ABS section.
Does your bootloader relocate the kernel? Absolute symbols break the relocated kernel. Linker may need to keep the output section if there is a symbol defined for it.
My bootloader is good old grub (0.97). To tell the truth - I have no knowledge of its inner workings and no idea if it relocates the kernel :(
Created attachment 6388 [details] A patch Please try this patch.
See http://sourceware.org/ml/binutils/2012-05/msg00019.html I suspect your boot loader is broken. However, this sort of problem is precisely why I put up some resistance to the pr13621 patch.
HJ, your patch is cunning enough but really no better than reverting the pr13621 change, since your patch will likely result in elflint complaints.
FYI, I have reports of the pr13621 patch causing kernel issues using both grub and syslinux. Of course, both bootloaders could be broken...
Add lilo to the list of broken bootloaders... HJ, your patch does the job, my kernel is deep into swap and running fine.
(In reply to comment #6) > HJ, your patch is cunning enough but really no better than reverting the > pr13621 change, since your patch will likely result in elflint complaints My patch avoids ABS symbols only when they are defined in linker script and passes the testcase for PR 13621.
HJ, you no doubt saw my attempt at trying to interpret the ELF spec to rationalize GNU ld generation of SHN_ABS symbols. Cary demolished my argument. So for one, my claim that there might be a bug in the kernel loaders in their treatment of SHN_ABS symbols is rubbish. Also, it means that the pr13621 patch should be completely reverted, because that patch made GNU ld create SHN_ABS symbols from non-SHN_ABS input symbols. Your patch prevents some specific cases of wrong SHN_ABS symbols, but we should be getting rid of all of them.
(In reply to comment #10) > HJ, you no doubt saw my attempt at trying to interpret the ELF spec to > rationalize GNU ld generation of SHN_ABS symbols. Cary demolished my argument. > So for one, my claim that there might be a bug in the kernel loaders in their > treatment of SHN_ABS symbols is rubbish. Also, it means that the pr13621 patch > should be completely reverted, because that patch made GNU ld create SHN_ABS > symbols from non-SHN_ABS input symbols. Your patch prevents some specific > cases of wrong SHN_ABS symbols, but we should be getting rid of all of them. Let's revert the whole PR 13621 patch. Thanks.
Created attachment 6389 [details] A new patch This patch reverts the PR 13621 change and keeps the zero size section if there is a symbol for it.
(In reply to comment #12) > Created attachment 6389 [details] > A new patch > > This patch reverts the PR 13621 change and keeps the zero size > section if there is a symbol for it. Please verify if this patch works.
(In reply to comment #13) > (In reply to comment #12) > > Created attachment 6389 [details] > > A new patch > > > > This patch reverts the PR 13621 change and keeps the zero size > > section if there is a symbol for it. > > Please verify if this patch works. Works for me. kernel 3.3.4 built with this patch runs fine. FYI: c15a3000 T __init_begin c162d000 T __init_end
CVSROOT: /cvs/src Module name: src Changes by: amodra@sourceware.org 2012-05-05 04:51:16 Modified files: bfd : ChangeLog linker.c ld/testsuite : ChangeLog ld/testsuite/ld-elf: warn2.d Removed files: ld/testsuite/ld-elf: zerosize1.d zerosize1.s Log message: PR ld/14052 PR ld/13621 bfd/ * linker.c (_bfd_nearby_section): Revert 2012-02-13 change. ld/testsuite/ * ld-elf/warn2.d: Revert 2012-02-13 change. * ld-elf/zerosize1.d, ld-elf/zerosize1.s: Delete. Patches: http://sourceware.org/cgi-bin/cvsweb.cgi/src/bfd/ChangeLog.diff?cvsroot=src&r1=1.5669&r2=1.5670 http://sourceware.org/cgi-bin/cvsweb.cgi/src/bfd/linker.c.diff?cvsroot=src&r1=1.94&r2=1.95 http://sourceware.org/cgi-bin/cvsweb.cgi/src/ld/testsuite/ChangeLog.diff?cvsroot=src&r1=1.1518&r2=1.1519 http://sourceware.org/cgi-bin/cvsweb.cgi/src/ld/testsuite/ld-elf/warn2.d.diff?cvsroot=src&r1=1.7&r2=1.8 http://sourceware.org/cgi-bin/cvsweb.cgi/src/ld/testsuite/ld-elf/zerosize1.d.diff?cvsroot=src&r1=1.2&r2=NONE http://sourceware.org/cgi-bin/cvsweb.cgi/src/ld/testsuite/ld-elf/zerosize1.s.diff?cvsroot=src&r1=1.1&r2=NONE
Fixed.
CVSROOT: /cvs/src Module name: src Changes by: hjl@sourceware.org 2012-05-05 14:15:21 Modified files: ld/testsuite : ChangeLog Added files: ld/testsuite/ld-elf: pr14052.d pr14052.t Log message: Add a testcase for PR ld/14052 PR ld/14052 * ld-elf/pr14052.d: New file. * ld-elf/pr14052.t: Likewise. Patches: http://sourceware.org/cgi-bin/cvsweb.cgi/src/ld/testsuite/ChangeLog.diff?cvsroot=src&r1=1.1519&r2=1.1520 http://sourceware.org/cgi-bin/cvsweb.cgi/src/ld/testsuite/ld-elf/pr14052.d.diff?cvsroot=src&r1=NONE&r2=1.1 http://sourceware.org/cgi-bin/cvsweb.cgi/src/ld/testsuite/ld-elf/pr14052.t.diff?cvsroot=src&r1=NONE&r2=1.1