http://bugzilla.opensuse.org/show_bug.cgi?id=970239 Ludwig Nussel changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |lnussel@suse.com Assignee|kernel-maintainers@forge.pr |rguenther@suse.com |ovo.novell.com | -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=970239 http://bugzilla.opensuse.org/show_bug.cgi?id=970239#c3 --- Comment #3 from Dominique Leuenberger --- One part of Fedora's fixing was apparently the importing of H.J.Lu's kernel LTO patch (there is no source reference - one apparently knows him) http://pkgs.fedoraproject.org/cgit/rpms/binutils.git/tree/binutils-2.26-lto.... Another correlation done on the fedora bug is the enabling/disabling of debug in the kernel (idsabled seems to work, first reported by AdamW on 22.2, kernel disabled it in fedora's package on 21.2) -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=970239 http://bugzilla.opensuse.org/show_bug.cgi?id=970239#c7 Felix Miata changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |mrmazda@earthlink.net --- Comment #7 from Felix Miata --- I have several i586 TW installations I can test this with. -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=970239 http://bugzilla.opensuse.org/show_bug.cgi?id=970239#c9 --- Comment #9 from Dominique Leuenberger --- (In reply to Dominique Leuenberger from comment #8)

The error printed on screen comes from arch/x86/boot/compressed/misc.c (if I found the right spot)

293 phdrs = malloc(sizeof(*phdrs) * ehdr.e_phnum); 294 if (!phdrs) 295 error("Failed to allocate space for phdrs");

Something must have gone awfully wrong by the time we are here... objdump -a vmlinz: Size of program headers: 32 (bytes) Number of program headers: 3 -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=970239 http://bugzilla.opensuse.org/show_bug.cgi?id=970239#c11 Fabian Vogt changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |fvogt@suse.com --- Comment #11 from Fabian Vogt --- (In reply to Dominique Leuenberger from comment #9)

(In reply to Dominique Leuenberger from comment #8)

...

The error printed on screen comes from arch/x86/boot/compressed/misc.c (if I found the right spot)

293 phdrs = malloc(sizeof(*phdrs) * ehdr.e_phnum); 294 if (!phdrs) 295 error("Failed to allocate space for phdrs");

Something must have gone awfully wrong by the time we are here...

objdump -a vmlinz: Size of program headers: 32 (bytes) Number of program headers: 3

I had a quick GDB session: malloc returns NULL because free_mem_ptr gets overwritten at some point, probably during decompression. -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=970239 http://bugzilla.opensuse.org/show_bug.cgi?id=970239#c15 --- Comment #15 from Dominique Leuenberger --- (In reply to Fabian Vogt from comment #13)

You might be right, as it is actually broken in a way (and it affects x86 only): https://sourceware.org/bugzilla/show_bug.cgi?id=19520 The fix is in 2.26.1. (I wonder whether using GOLD would work as well here) I'm currently trying to compile a kernel to reproduce it (and not succeeding in doing so with tinyconfig), so I can try that option after the first "successful failure".

I can provide a VM disk image which has a couple kernels installed (most from RPM and one self-built inside the VM. The locally built one failed to boot in my tests, so the kernel tree / config seems 'fine' to reproduce -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=970239 http://bugzilla.opensuse.org/show_bug.cgi?id=970239#c17 --- Comment #17 from Fabian Vogt --- After some research, experiments and many kernel and binutils builds I finally found the root cause. Like richard said, it's the introduction of the new relocation type(s). gas generates a R_386_GOT32X relocation for several global symbols, like "free_mem_ptr". R_386_GOT32X is basically a fancy R_386_GOTOFF with some more information that the linker can use for optimization, like converting GOT-relative "mov 0x10(%ebx), %ecx" to absolute "lea $0xcafebeef, %ecx" for example. The kernel's decompressor code is built with "-fPIC" so that a GOT is used where applicable, but for some cases ld optimizes that away by converting the the GOT-relative free_mem_ptr reference to an absolute one, breaking the runtime relocation in the process. The bug is in binutils-2.26/bfd/elf32-i386.c:2992. Modifying the if condition to be false fixes the decompressor relocation. (Disclaimer: The following may be entirely based on false assumptions) AFAICS, the issue is in the design of the build process. ld does not know that a GOT was explicitly requested by "-fPIC" so it is free to optimize it. By specifying "-pie" or "-shared" ld behaves correctly by not removing the GOT reference, but not in this case, as the output format is not position independant. -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=970239 http://bugzilla.opensuse.org/show_bug.cgi?id=970239#c19 Markus Trippelsdorf changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |octoploid@yandex.com --- Comment #19 from Markus Trippelsdorf --- (In reply to Fabian Vogt from comment #16)

(In reply to Ismail Donmez from comment #14)

...

(In reply to Fabian Vogt from comment #13)

...

The fix is in 2.26.1. (I wonder whether using GOLD would work as well here)

...

Unless something changed gold won't be able to link the linux kernel, afaik it doesn't support its ldscript usage.

What a pity, I'm using gold everywhere and now I'm used to the speed boost...

gold is able to link the kernel for several years already... -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=970239 http://bugzilla.opensuse.org/show_bug.cgi?id=970239#c22 Dominique Leuenberger changed: What |Removed |Added ---------------------------------------------------------------------------- Assignee|rguenther@suse.com |kernel-maintainers@forge.pr | |ovo.novell.com --- Comment #22 from Dominique Leuenberger --- (In reply to Hendrik Woltersdorf from comment #21) I don't think you meant to reset the assignee to Richard -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=970239 VERSINI Arnaud changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |arnaud@versini.eu -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=970239 http://bugzilla.opensuse.org/show_bug.cgi?id=970239#c29 --- Comment #29 from Fabian Vogt --- I just retested the patch on the binutils bug report and it only works together with ld from git master, using it with 2.26 yields an unbootable kernel. -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=970239 http://bugzilla.opensuse.org/show_bug.cgi?id=970239#c30 --- Comment #30 from Dominique Leuenberger --- @kernel team: Can you please consider to provide a kernel with the so far found work arounds by fvogt? Even though this is likely not the final solution yet (binutils changes are possible, some more kernel patches) we should try to get something together in a working form asap, as we're blocking all tumbleweed snapshots for a week already. -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=970239 http://bugzilla.opensuse.org/show_bug.cgi?id=970239#c32 Fabian Vogt changed: What |Removed |Added ---------------------------------------------------------------------------- Flags|needinfo?(fvogt@suse.com) | --- Comment #32 from Fabian Vogt --- Created attachment 668976 --> http://bugzilla.opensuse.org/attachment.cgi?id=668976&action=edit Patch for use with binutils 2.26.0 This is a patch that disables relax relocations in the 32-bit loader. However, it only works with binutils <= 2.26 as newer versions (I tried git master) optimize non-relax relocations as well and thus need a different patch: https://sourceware.org/bugzilla/attachment.cgi?id=9089 That patch does not work with anything older than git master though... -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=970239 http://bugzilla.opensuse.org/show_bug.cgi?id=970239#c38 Dieter Nützel changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |Dieter@nuetzel-hh.de --- Comment #38 from Dieter Nützel --- (In reply to Takashi Iwai from comment #37)

OK, I took Fabian's patch as is, and pushed to my git branch. Now waiting for pull and the package submission.

Hip, hip, hooray! Hello Takashi, et al., you all saved my faith in S.u.S.E, SuSE, openSUSE...;-) And I'm with you since ~1993, first on floppies, and then on CDs. My home server VIA C7-D stopped working (boot) under openSUSE 13.2 plus stable kernel 4.4.5-1.2 on Friday night, 11.03.2016. I had running, plugging, carrying monitors, keyboards and systems, all that shit a hole night... i686-pae 4.4.5-1.1 GOOD 4.4.5-1.2 BAD Then I had have to go back to kernel-pae-3.16.6-2.1.i686 with USB recovery. And lost network configuration. After that was fixed I was on 3.16.7-35-pae, until now. I hopped for the new 4.5.0 series, but... 4.5.0-1.g95a9976-pae BAD and now you... Linux XXXX 2.6.65-2.gb2c9ae5-pae #1 SMP PREEMPT Wed Mar 16 17:30:21 UTC 2016 (b2c9ae5) i686 i686 i386 GNU/Linux Running. So you got from me: Tested-by: Dieter Nützel Thank you so much! Normally I'm on my Xeon machine for DRM/Mesa/Radeon development;-) Linux XXXX 4.5.0-2.gb2c9ae5-default #1 SMP PREEMPT Wed Mar 16 17:30:21 UTC 2016 (b2c9ae5) x86_64 x86_64 x86_64 GNU/Linux -Dieter PS If I (we) only could find a solution for my openSUSE 13.2 wicked/DSL/modem problem... (all newer then wicked-0.6.19-18.1 versions can't bring up dsl0) -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=970239 http://bugzilla.opensuse.org/show_bug.cgi?id=970239#c43 Bruno Friedmann changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |bruno@ioda-net.ch --- Comment #43 from Bruno Friedmann --- @Dominique could you also post a status update here (and certainly perhaps on ml) to a rough estimate for the light at the end of tunnel The bug is marked resolved, normally it should have been mentionned in some changes and sr on obs to have a trace here of its status. -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=970239 http://bugzilla.opensuse.org/show_bug.cgi?id=970239#c45 --- Comment #45 from Hendrik Woltersdorf --- (In reply to Dieter Nützel from comment #39)

Linux XXXX 4.5.0-2.gb2c9ae5-pae #1 SMP PREEMPT Wed Mar 16 17:30:21 UTC 2016 (b2c9ae5) i686 i686 i386 GNU/Linux

I've downloaded this kernel package too and installed it on my hardware: - Intel(R) Pentium(R) M processor 730 - gfx card: ATI Mobility FireGL V5000 564B (M26) (PCIE) with the free radeon driver - encrypted disks (everything but /boot): encrypted single partitions with etx4 and xfs and a "LVM on LUKS" container And it works! -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.opensuse.org/show_bug.cgi?id=970239 cees vorstenbosch changed: What |Removed |Added ---------------------------------------------------------------------------- See Also|https://bugzilla.redhat.com | |/show_bug.cgi?id=1302071 | -- You are receiving this mail because: You are on the CC list for the bug.