On Mon, Mar 30, 2020 at 5:19 PM Liu Yiding <liuyd.fnst@xxxxxxxxxxxxxx> wrote: > > > On 3/30/20 2:09 PM, Andrii Nakryiko wrote: > > On 3/29/20 5:48 PM, Liu Yiding wrote: > >> Add attachment. > >> > > > > Your BTF seems to be invalid. It has struct perf_ibs, which has a > > first field `struct pmu pmu` field with valid-looking size of 296 > > bytes, **but** the type that field points to is not a complete `struct > > pmu` definition, but rather just forward declaration. The way it is it > > shouldn't be even compilable, because forward declaration of a struct > > doesn't specify the size of a struct, so compiler should have rejected > > it. So it must be that either DWARF generated by compiler isn't > > correct, or there is DWARF -> BTF conversion bug somewhere. Are you > > using any special DWARF Kconfig settings? Maybe you can share your > > full .config and I might try to repro it on my machine. > > > > >> Are you using any special DWARF Kconfig settings? > > Sorry, i'm a newbie at this. I don't know which settings are related to > DWARF. > > Just search keywords. > > ``` > > liuyd@localhost:~$ cat config-5.6.0-rc5 | grep DWARF > # CONFIG_DEBUG_INFO_DWARF4 is not set > > ``` > > I built attached config on a clear ubuntu machine. Error could be > reproduced. So you are right, there is a conflict between kconfigs. > > > >> Maybe you can share your full .config and I might try to repro it on > my machine. > > Thanks a lot. I attached the broken config. Thanks a lot! I think it's due to DEBUG_INFO_REDUCED which produces not entirely correct DWARF. I'm asking Slava to disable this config when BTF is requested in [0]. [0] https://lore.kernel.org/bpf/CAEf4BzadnfAwfa1D0jZb=01Ou783GpK_U7PAYeEJca-L9kdnVA@xxxxxxxxxxxxxx > > > > But either way, that warning you get is a valid one, it should be > > illegal to have non-pointer forward-declared struct as a type for a > > struct member. > > > >> > >> On 3/30/20 8:46 AM, Liu Yiding wrote: > >>> Something wrong with my smtp and this email missed. > >>> > >>> Send again. > >>> > >>> > >>> On 3/27/20 11:09 AM, Liu Yiding wrote: > >>>> Hi, Andrii. > >>>> > >>>> Thanks for your prompt reply! > >>>> > >>>> Please check attatchment for my_btf.bin. > >>>> > >>>> > >>>> On 3/27/20 4:28 AM, Andrii Nakryiko wrote: > >>>>> Would you be able to share BTF of vmlinux that is used to generate > >>>>> vmlinux.h? Please run in verbose mode: `make V=1` and search for > >>>>> `bpftool btf dump file` command. It should point either to > >>>>> /sys/kernel/btf/vmlinux or some other location, depending on how > >>>>> things are set up on your side. > >>>>> > >>>>> If it's /sys/kernel/btf/vmlinux, you can just `cat > >>>>> /sys/kernel/btf/vmlinux > my_btf.bin`. If it's some other file, > >>>>> easiest would be to just share that file. If not, it's possible to > >>>>> extract .BTF ELF section, let me know if you need help with that. > >>>> > > > > > > > -- > Best Regards. > Liu Yiding > > >
