On Thu, May 11, 2006 at 12:21:00PM -0700, David S. Miller wrote: > From: Martin Habets <errandir_news@xxxxxxxxxxxxxxxxx> > Date: Thu, 11 May 2006 19:07:32 +0100 > > > + case R_SPARC_UA32: /* Like R_SPARC_32 but unaligned */ > > + uv = (u8 *) v; > > + location[0] = uv[0]; > > + location[1] = uv[1]; > > + location[2] = uv[2]; > > + location[3] = uv[3]; > > + break; > > + > > This is not correct. > > The relocation puts "value v" at "location" not "data at address v" at > "location", the latter of which is what your patch is trying to do. Hmm, yeah. It did not feel right after a night's sleep. > Simply adding "case R_SPARC_UA32:" right under "case R_SPARC_32:" > would work just fine since the R_SPARC_32 code already handles > arbitrary alignment of "location" which is the crucial difference > between R_SPARC_32 and R_SPARC_UA32. > > I still wonder where you're getting that relocation from. It must be > emitted by some dwarf2 frame unwind stuff. It's a shame that such > bloat is aparently being emitted into objects by default. I've checked the .ko files, R_SPARC_UA32 is only used in the '.rela.eh_frame' section. The build command is: gcc -Wp,-MD,lib/.crc32.o.d -nostdinc -isystem /usr/lib/gcc-lib/sparc-linux/3.3.5/include -D__KERNEL__ -Iinclude -include include/linux/autoconf.h -Wall -Wundef -Wstrict-prototypes -Wno-trigraphs -fno-strict-aliasing -fno-common -O2 -fomit-frame-pointer -fasynchronous-unwind-tables -m32 -pipe -mno-fpu -fcall-used-g5 -fcall-used-g7 -DMODULE -D"KBUILD_STR(s)=#s" -D"KBUILD_BASENAME=KBUILD_STR(crc32)" -D"KBUILD_MODNAME=KBUILD_STR(crc32)" -c -o lib/crc32.o lib/crc32.c The -fasynchronous-unwind-tables caught my eye. Leaving it out caused the '.rela.eh_frame' section not to be generated at all (module loads ok as well). I've tried -funwind-tables, but that generated the same '.rela.eh_frame' section as with -fasynchronous-unwind-tables. > Can you send me a sample object that has such a relocation again? I > don't want to add support for this relocation if it really is not > needed. I'm hoping we can add a compiler option that will make this > cruft not get output by the compiler since it is superfluous and of > no use in the kernel build. I've uploaded the following to http://www.mph.eclipse.co.uk/pub/linux/ crc32.o crc32.ko crc32.2 these are all generated using the standard build options (without any modifications by me). -- Martin - To unsubscribe from this list: send the line "unsubscribe sparclinux" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
