On Thu, Jul 11, 2019 at 11:04:35PM +0200, Jann Horn wrote: > On Thu, Jul 11, 2019 at 11:00 PM Arnd Bergmann <a...@arndb.de> wrote: > > > > On Thu, Jul 11, 2019 at 7:26 PM Josh Poimboeuf <jpoim...@redhat.com> wrote: > > > > > > On Thu, Jul 11, 2019 at 02:40:06PM +0200, Arnd Bergmann wrote: > > > > During randconfig testing with clang-9, I came across an object file > > > > that makes objtool segfault, see attachment. Let me know if you need > > > > more information to > > > > debug this. > > > > > > > > I also get a ton of objtool warnings building random configurations, > > > > but Nick > > > > mentioned that there is still a bug related to asm-goto in the build > > > > I'm using > > > > that may be the root cause. Once I have a fixed clang-9 build, I can > > > > have a look > > > > at those as well. > > > > > > Seg fault fix: > > > > Thanks for the fix! testing it over night now, will let you know tomorrow > > if problems remain. > > > > I wonder if this is also related to several warnings I get about switch > > tables like: > > > > drivers/usb/misc/sisusbvga/sisusb.o: warning: objtool: > > sisusb_write_mem_bulk()+0x588: can't find switch jump table > > > > drivers/gpu/drm/amd/amdgpu/../display/dc/dce110/dce110_mem_input_v.o: > > warning: objtool: dce_mem_input_v_program_pte_vm()+0x46e: can't find > > switch jump table > > drivers/gpu/drm/amd/amdgpu/../display/dc/dce110/dce110_opp_csc_v.o: > > warning: objtool: dce110_opp_v_set_csc_default()+0x714: can't find > > switch jump table > > drivers/gpu/drm/nouveau/nvkm/subdev/clk/nv50.o: warning: objtool: > > nv50_clk_read()+0x15c: can't find switch jump table > > drivers/media/dvb-frontends/cxd2880/cxd2880_tnrdmd_dvbt2.o: warning: > > objtool: x_tune_dvbt2_demod_setting()+0x992: can't find switch jump > > table > > drivers/media/tuners/mt2063.o: warning: objtool: > > MT2063_SetReceiverMode()+0x24d: can't find switch jump table > > drivers/mmc/host/tifm_sd.o: warning: objtool: tifm_sd_exec()+0x7e: > > can't find switch jump table > > drivers/mtd/nand/raw/fsl_ifc_nand.o: warning: objtool: > > fsl_ifc_nand_probe()+0x4c7: can't find switch jump table > > drivers/net/can/at91_can.o: warning: objtool: at91_irq()+0x347: can't > > find switch jump table > > drivers/net/phy/phylink.o: warning: objtool: > > phylink_mac_config()+0x2b5: can't find switch jump table > > drivers/regulator/max8973-regulator.o: warning: objtool: > > max8973_probe()+0x736: can't find switch jump table > > drivers/regulator/tps80031-regulator.o: warning: objtool: > > tps80031_regulator_probe()+0x143: can't find switch jump table > > drivers/tty/cyclades.o: warning: objtool: cy_set_line_char()+0x86c: > > can't find switch jump table > > drivers/tty/serial/jsm/jsm_cls.o: warning: objtool: cls_param()+0x10b: > > can't find switch jump table > > drivers/tty/serial/jsm/jsm_neo.o: warning: objtool: neo_param()+0x151: > > can't find switch jump table > > drivers/usb/core/hub.o: warning: objtool: hub_probe()+0x920: can't > > find switch jump table > > drivers/usb/misc/sisusbvga/sisusb.o: warning: objtool: > > sisusb_write_mem_bulk()+0x4db: can't find switch jump table > > kernel/rcu/tree.o: warning: objtool: rcu_note_context_switch()+0x6b8: > > can't find switch jump table > > lib/zstd/decompress.o: warning: objtool: > > ZSTD_decodeLiteralsBlock()+0x5e: can't find switch jump table > > > > If you want to have a look, I can provide object files and/or reduced test > > cases for this. My guess is that it is unrelated to the warnings that Nick > > saw for asm-goto. > > I was playing around with building the kernel with LLVM a few months > ago and used this local patch, but didn't get around to submitting > upstream because I couldn't reproduce the problem for some reason. I > think the warnings you're getting sound like what I saw back then: > https://gist.github.com/thejh/0434662728afb95d72455bf30ece5817 > > Quoting the commit message from that patch: > > ==== > With clang from git master, code can be generated where a function contains > two indirect jump instructions that use the same switch table. To deal with > this case and similar ones properly, convert the switch table parsing to > use two passes: > ==== > > Does that sound like what you're seeing?
Thanks Jann, I like this approach. Hopefully it also works with GCC. The switch tables (and jump tables in general) have been a hot topic in objtool lately. I have several other patches pending which touch this code. I'll integrate your patch with the others and try to do some more testing in GCC. -- Josh