| Message ID | 87egil26sn.fsf@belgarion.home (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Robert Jarzmik <robert.jarzmik@free.fr> writes: > Hello, > > I have a strange regression with linux-next on module insertion. The stack I get > when inserting a module is in [3]. I'm using the gcc in [1], and added a small > print in [2]. This happens on a pxa310 cm-x300 platform. > > As far as I understand the Oops, this is an unaligned access on the following > relocation (0xbf009a26): > *(u32 *)loc += sym->st_value; > > This happens regardless of the module I try, by that fat.ko, pxa3xx-nand.ko, etc > .. I add that I have not changed my toolchain for quite some time. Hi Russell, After digging a bit in the problem, and even if I don't understand why this happens, I know at least _what_ happens. I'd like to know if this is specific to xscale or a wider problem, hence this mail to you. The summary of my problem : - on linux-next (20150828), module insertion Oopses (see [3]) - the Oops is as far as I understand an unaligned access (message above) It is in arch/arm/kernel/module.h, the line: *(u32 *)loc += sym->st_value; - the unalignement comes from a relocation in section __bug_table (see [1]) The address is 0xbf0021ea. - the module readelf in [2] shows the relocation in __bug_table Do you know if other ARM platform are affected, and if an alignement of "1" is normal for a __bug_table section (as opposed to an alignement of 4) ? And I'm wondering how it is I have never seen that before also ... If I have omitted some data, please tell me so that I can complete this report. Cheers.
On Tue, Sep 01, 2015 at 01:00:39AM +0200, Robert Jarzmik wrote: > [2] readelf -Sr pxa3xx_nand.ko > ============================== > There are 36 section headers, starting at offset 0x6a58: > > Section Headers: > [Nr] Name Type Addr Off Size ES Flg Lk Inf Al > [ 0] NULL 00000000 000000 000000 00 0 0 0 > [ 1] .note.gnu.build-i NOTE 00000000 000034 000024 00 A 0 0 4 > [ 2] .text PROGBITS 00000000 000058 0020c4 00 AX 0 0 4 > [ 3] .rel.text REL 00000000 006378 0004b8 08 I 34 2 4 > [ 4] .init.text PROGBITS 00000000 00211c 000014 00 AX 0 0 4 > [ 5] .rel.init.text REL 00000000 006830 000018 08 I 34 4 4 > [ 6] .exit.text PROGBITS 00000000 002130 00000c 00 AX 0 0 4 > [ 7] .rel.exit.text REL 00000000 006848 000010 08 I 34 6 4 > [ 8] .ARM.extab PROGBITS 00000000 00213c 000000 00 A 0 0 1 > [ 9] .ARM.exidx ARM_EXIDX 00000000 00213c 0000b8 00 AL 2 0 4 > [10] .rel.ARM.exidx REL 00000000 006858 0000c0 08 I 34 9 4 > [11] .rodata.str PROGBITS 00000000 0021f4 00003e 01 AMS 0 0 1 > [12] __bug_table PROGBITS 00000000 002232 000018 00 A 0 0 1 This shows that the bug table has an alignment of 1, which is silly as it contains 32-bit values - this really ought to be indicating an alignment of 4. I notice that a bunch of my modules are the same as well. Try adding ".align 2" into arch/arm/include/asm/bug.h, __BUG() macro, just after ".pushsection __bug_table" which should have the effect of increasing the alignment of the section. It probably hasn't been noticed on ARMv6 and later because they'll always fix up these relocations transparently. For ARMv5 and older, you really ought to have the alignment handler enabled - IP networking pretty much relies on this being present to catch corner cases (eg, with TCP options mis-aligning subsequent data.)
=============== rj@belgarion:~/mio_linux/kernel$ arm-linux-gnueabi-gcc -v Using built-in specs. COLLECT_GCC=/usr/bin/arm-linux-gnueabi-gcc COLLECT_LTO_WRAPPER=/usr/lib/gcc/arm-linux-gnueabi/4.9/lto-wrapper Target: arm-linux-gnueabi Configured with: ../src/configure -v --with-pkgversion=' 4.9.2-10' --with-bugurl=file:///usr/share/doc/gcc-4.9/README.Bugs --enable-languages=c,c++,go,fortran,objc --prefix=/usr --program-suffix=-4.9 --enable-shared --enable-linker-build-id --libexecdir=/usr/lib --without-included-gettext --enable-threads=posix --with-gxx-include-dir=/usr/include/c++/4.9 --libdir=/usr/lib --enable-nls --with-sysroot=/ --enable-clocale=gnu --enable-libstdcxx-debug --enable-libstdcxx-time=yes --enable-gnu-unique-object --disable-libitm --disable-libquadmath --enable-plugin --with-system-zlib --enable-objc-gc --enable-multiarch --disable-sjlj-exceptions --with-arch=armv4t --with-float=soft --enable-checking=release --build=x86_64-linux-gnu --host=x86_64-linux-gnu --target=arm-linux-gnueabi --program-prefix=arm-linux-gnueabi- Thread model: posix gcc version 4.9.2 ( 4.9.2-10) [2] Debug patch in arch/arm/kernel/module.c =========================================== diff --git a/arch/arm/kernel/module.c b/arch/arm/kernel/module.c index efdddcb97dd1..aed935c55d63 100644 --- a/arch/arm/kernel/module.c +++ b/arch/arm/kernel/module.c @@ -88,6 +88,9 @@ apply_relocate(Elf32_Shdr *sechdrs, const char *strtab, unsigned int symindex, return -ENOEXEC; } + pr_info("%s():%d dstsec=%p dstsec->sh_addr=%p rel=%p rel->r_offset=%u\n", + __func__, __LINE__, dstsec, dstsec, + dstsec->sh_addr, rel, rel->r_offset); loc = dstsec->sh_addr + rel->r_offset; switch (ELF32_R_TYPE(rel->r_info)) {