We're working on an upstream kernel feature [1] which adds a new relocation section to kernel modules and when loading the .ko's symbol information into crash utility (which includes gdb, which includes bfd) it's hitting an assertion in elf.c that isn't giving much information to debug. Here is the assertion when running crash (with a bit of extra debugging): % insmod /lib/modules/5.1.0-rc4+/kernel/samples/livepatch/livepatch-annotated-sample.ko % crash ... crash> mod -s livepatch_annotated_sample /lib/modules/5.1.0-rc4+/kernel/samples/livepatch/livepatch-annotated-sample.ko shindex=34, name=.klp.rela.vmlinux..text *p_hdr(0x57f6150) == NULL = 0, this_idx=3 BFD: BFD (GNU Binutils) 2.23.52.20130312 assertion fail elf.c:1881 MODULE NAME SIZE OBJECT FILE ffffffffc05cb380 livepatch_annotated_sample 16384 /lib/modules/5.1.0-rc4+/kernel/samples/livepatch/livepatch-annotated-sample.ko and the corresponding assertion, with extra debugging, in elf.c: 1545 bfd_section_from_shdr (bfd *abfd, unsigned int shindex) .... 1563 switch (hdr->sh_type) 1564 { .... 1784 case SHT_REL: 1785 case SHT_RELA: .... 1871 esdt = elf_section_data (target_sect); 1872 if (hdr->sh_type == SHT_RELA) 1873 p_hdr = &esdt->rela.hdr; 1874 else 1875 p_hdr = &esdt->rel.hdr; 1876 1877 if (*p_hdr != NULL) { 1878 printf("shindex=%d\nname=%s\n", shindex, name); 1879 printf("*p_hdr(%p) == NULL = %d, this_idx=%d\n", *p_hdr, (*p_hdr == NULL), esdt->this_idx); 1880 } 1881 BFD_ASSERT (*p_hdr == NULL); I noticed that it is processing our new section, index 34, .klp.rela.vmlinux..text and finds an existing p_hdr to section index 3, .text: % eu-readelf --relocs /lib/modules/5.1.0-rc4+/kernel/samples/livepatch/livepatch-annotated-sample.ko Relocation section [ 4] '.rela.text' for section [ 3] '.text' at offset 0xc8 contains 7 entries: Offset Type Value Addend Name 0x0000000000000001 X86_64_PC32 000000000000000000 -4 __fentry__ 0x000000000000000f X86_64_32S 000000000000000000 +0 .rodata.str1.1 0x0000000000000014 X86_64_PC32 000000000000000000 -4 seq_printf 0x0000000000000021 X86_64_PC32 000000000000000000 -4 __fentry__ 0x0000000000000028 X86_64_32S 000000000000000000 +0 .data 0x000000000000002d X86_64_PC32 000000000000000000 -4 klp_enable_patch 0x0000000000000041 X86_64_PC32 000000000000000000 -4 __fentry__ Relocation section [34] '.klp.rela.vmlinux..text' for section [ 3] '.text' at offset 0x4a080 contains 1 entry: Offset Type Value Addend Name 0x0000000000000008 X86_64_PC32 000000000000000000 -4 .klp.sym.vmlinux.saved_command_line,0 Unfortunately it is less than trivial to move crash utility's version of gdb / binutils forward, so I couldn't directly test with a newer version of binutils. However, elf.c :: bfd_section_from_shdr() still seems to contain this assertion, though in a slightly more direct format: /* PR 17512: file: 0b4f81b7. */ if (*p_hdr != NULL) goto fail; There other binutils utilities that are also reporting problems, but with an error message too vague to determine why they are failing. They are perhaps related, and possibly easier to debug/verify for our purposes here. Here's what a fresh clone + build of binutils-gdb tree versions of objdump and gdb think of our new object file [2]: % git describe HEAD users/ARM/embedded-gdb-master-2018q4-978-g48574d91bf12 % ./binutils/objdump -D /tmp/bug/livepatch-annotated-sample.ko ./binutils/objdump: /tmp/bug/livepatch-annotated-sample.ko: bad value % ./gdb/gdb -q /tmp/bug/livepatch-annotated-sample.ko "/tmp/bug/livepatch-annotated-sample.ko": not in executable format: bad value Both of those tools work as expected with an ordinary kernel module [3] (ie, one with any extra relocation sections to '.text'). [1] https://lore.kernel.org/lkml/20190410155058.9437-1-joe.lawrence@redhat.com/ [2] http://people.redhat.com/~jolawren/bug/livepatch-annotated-sample.ko [3] http://people.redhat.com/~jolawren/bug/livepatch-sample.ko
The master branch has been updated by Nick Clifton <nickc@sourceware.org>: https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=a7ba389645d178c43100ec47e513389ae8bf8f93 commit a7ba389645d178c43100ec47e513389ae8bf8f93 Author: Nick Clifton <nickc@redhat.com> Date: Fri Aug 23 13:22:02 2019 +0100 Stop the BFD library from failing when encountering a second set of relocs for the same section. PR 24456 * elf.c (bfd_section_from_shdr): Issue an informative warning message and continue processing other sections after encountering a reloc section for a section which already has other relocs associated with it.
Hi Joe, Sorry for dropping this PR. I have now checked in a patch which makes the BFD library generate a more helpful warning message, and then continue processing the rest of the input file. The downside though, is that this patch does not enable processing of the extra reloc section - it is just ignored - but I feel that that is better than just failing outright. % objdump -D livepatch-annotated-sample.ko > /dev/null objdump: livepatch-annotated-sample.ko: warning: multiple relocation sections for section .text found - ignoring all but the first % echo $? 0 I hope that this will suffice for your purposes. Cheers Nick
Hi Nick, Thanks for revisiting this one! I can confirm that with the latest master branch, objdump can now process the .ko's that we've generated with multiple section relocations. Even if those follow-up relocations aren't processed, it is very helpful to at least dump assembly, sections, etc. when debugging. Thanks, -- Joe
OK, I am going to close this PR. If it turns out that there is a need to process the other reloc sections then this PR can be reopened, or a new one filed. I suspect however that enabling support for them will be difficult, as it is likely to invole a lot of rewriting of the internals of the BFD library.
Created attachment 11987 [details] Test binary