On Wed, 26 Jul 2023 15:46:03 +0100 Alan Maguire <alan.maguire@xxxxxxxxxx> wrote: > On 26/07/2023 12:07, Masami Hiramatsu (Google) wrote: > > Hi Jiri, > > > > On Wed, 26 Jul 2023 12:04:08 +0200 > > Jiri Olsa <olsajiri@xxxxxxxxx> wrote: > > > >> On Wed, Jul 26, 2023 at 10:25:34AM +0900, Masami Hiramatsu wrote: > >>> Hello, > >>> (I resend this because kconfig was too big) > >>> > >>> I found that BTF is not generated for gcc-built kernel with that latest > >>> pahole (v1.25). > >> > >> hi, > >> I can't reproduce on my setup with your .config > >> > >> does 'bpftool btf dump file ./vmlinux' show any error? > >> > >> is there any error in the kernel build output? > > > > Yes, here it is. I saw these 2 lines. > > > > die__process: DW_TAG_compile_unit, DW_TAG_type_unit, DW_TAG_partial_unit or DW_TAG_skeleton_unit expected got INVALID (0x0)! > > die__process: DW_TAG_compile_unit, DW_TAG_type_unit, DW_TAG_partial_unit or DW_TAG_skeleton_unit expected got INVALID (0x0)! > > This is strange, looks like some CUs were encoded incorrectly or we are > parsing incorrectly. The error originates in die__process() and happens > if the dwarf_tag() associated with the DIE isn't an expected unit; it's > not even a valid tag value (0) it looks like. I've not built with gcc 13 > yet so it's possible that's the reason you're seeing this, I'll try to > reproduce it.. And this warning message is not good for debugging. It would better dump the message with DIE index so that we can analyze the DWARF with other tools. > >>> When I'm using the distro origin pahole (v1.22) it works. (I also checked > >>> v1.23 and v1.24, both partially generated BTF) > >>> > >>> e.g. > >>> > >>> # echo 'f kfree $arg*' >> /sys/kernel/tracing/dynamic_events > >>> sh: write error: Invalid argument > >>> > >>> # cat /sys/kernel/tracing/error_log > >>> [ 21.595724] trace_fprobe: error: BTF is not available or not supported > >>> Command: f kfree $arg* > >>> ^ > >>> [ 21.596032] trace_fprobe: error: Invalid $-valiable specified > >>> Command: f kfree $arg* > >>> ^ > >>> > >>> / # strings /sys/kernel/btf/vmlinux | grep kfree > >> > >> hm, if you have this file present, you have BTF in > > > > Yes, it seems that the BTF itself is generated, but many entries seems > > dropped compared with pahole v1.22. So, if a given symbol has BTF, (e.g. > > kfree_rcu_batch_init) it works. > > > > Yep, BPF generation is more selective about what it emits in 1.25 to > avoid cases where a kernel function signature is ambiguous (multiple > functions of the same name with different signatures) or where it has > unexpected register use. You can observe this via pahole's --verbose > option (a lot of outut is emitted): > > In a built kernel directory (where unstripped vmlinux is present): > $ PAHOLE_FLAGS=$(./scripts/pahole_flags) > $ PAHOLE=/usr/local/bin/pahole > $ pahole --verbose -J $PAHOLE_FLAGS vmlinux > /tmp/pahole.out So this will generate BTF from vmlinux DWARF info. > If you want to investigate why a function has been left out, look for > "skipping" verbose output like this: > > skipping addition of 'access_error'(access_error) due to multiple > inconsistent function prototypes > skipping addition of > 'acpi_ex_convert_to_object_type_string'(acpi_ex_convert_to_object_type_string.isra.0) > due to unexpected register used for parameter Ah, that's nice. Let me try. > > FWIW I see most of the below in my BTF output on bpf-next, though I am > missing a few, possibly due to differing config options since they don't > appear in kallsyms either. I don't see the DWARF tag label not handled > messages so it's possible that's a symptom of something > > I suspect however some form of corruption in the DWARF representation > may be the reason a lot of these are missing. Would be worth trying > to "objdump -g vmlinux >vmlinux.dwarf" (file will be huge tho) I suspect. OK, let me dump the dwarf and check what information it has for "kfree". Thank you, -- Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>
