Message-ID: <CAEf4BzYKassG0AP372Q=Qsd+qqy7=YGe2XTXR4zG0c5oQ7Nkeg@mail.gmail.com>
Date:   Fri, 19 Mar 2021 11:39:01 -0700
From:   Andrii Nakryiko <andrii.nakryiko@...il.com>
To:     Jiri Olsa <jolsa@...hat.com>
Cc:     Andrii Nakryiko <andrii@...nel.org>, bpf <bpf@...r.kernel.org>,
        Networking <netdev@...r.kernel.org>,
        Alexei Starovoitov <ast@...com>,
        Daniel Borkmann <daniel@...earbox.net>,
        Kernel Team <kernel-team@...com>
Subject: Re: [PATCH v4 bpf-next 07/12] libbpf: add BPF static linker BTF and
 BTF.ext support

On Fri, Mar 19, 2021 at 9:23 AM Jiri Olsa <jolsa@...hat.com> wrote:
>
> On Thu, Mar 18, 2021 at 12:40:31PM -0700, Andrii Nakryiko wrote:
>
> SNIP
>
> > +
> > +     return NULL;
> > +}
> > +
> > +static int linker_fixup_btf(struct src_obj *obj)
> > +{
> > +     const char *sec_name;
> > +     struct src_sec *sec;
> > +     int i, j, n, m;
> > +
> > +     n = btf__get_nr_types(obj->btf);
>
> hi,
> I'm getting bpftool crash when building tests,
>
> looks like above obj->btf can be NULL:

I lost if (!obj->btf) return 0; somewhere along the rebases. I'll send
a fix shortly. But how did you end up with selftests BPF objects built
without BTF?

>
>         (gdb) r gen object /home/jolsa/linux/tools/testing/selftests/bpf/btf_data2.linked1.o /home/jolsa/linux/tools/testing/selftests/bpf/btf_data2.o
>         Starting program: /home/jolsa/linux/tools/testing/selftests/bpf/tools/sbin/bpftool gen object /home/jolsa/linux/tools/testing/selftests/bpf/btf_data2.linked1.o /home/jolsa/linux/tools/testing/selftests/bpf/btf_data2.o
>
>         Program received signal SIGSEGV, Segmentation fault.
>         btf__get_nr_types (btf=0x0) at btf.c:425
>
>         425             return btf->start_id + btf->nr_types - 1;
>         Missing separate debuginfos, use: dnf debuginfo-install elfutils-libelf-0.182-1.fc33.x86_64 libcap-2.48-2.fc33.x86_64 zlib-1.2.11-23.fc33.x86_64
>         (gdb)
>         (gdb) bt
>         #0  btf__get_nr_types (btf=0x0) at btf.c:425
>         #1  0x000000000043c4ba in linker_fixup_btf (obj=obj@...ry=0x7fffffffda50) at linker.c:1316
>         #2  0x000000000043caf7 in linker_load_obj_file (linker=linker@...ry=0x8612a0, filename=filename@...ry=0x7fffffffe0db "/home/jolsa/linux/tools/testing/selftests/bpf/btf_data2.o",
>             obj=obj@...ry=0x7fffffffda50) at linker.c:653
>         #3  0x000000000043df43 in bpf_linker__add_file (linker=linker@...ry=0x8612a0, filename=filename@...ry=0x7fffffffe0db "/home/jolsa/linux/tools/testing/selftests/bpf/btf_data2.o") at linker.c:412
>         #4  0x000000000040d710 in do_object (argc=0, argv=0x7fffffffdce0) at gen.c:639
>         #5  0x000000000040efd7 in cmd_select (cmds=cmds@...ry=0x50a600 <cmds>, argc=3, argv=0x7fffffffdcc8, help=help@...ry=0x40c6f7 <do_help>) at main.c:134
>         #6  0x000000000040d7c4 in do_gen (argc=<optimized out>, argv=<optimized out>) at gen.c:686
>         #7  0x000000000040efd7 in cmd_select (cmds=cmds@...ry=0x50b400 <cmds>, argc=4, argv=0x7fffffffdcc0, help=help@...ry=0x40ee93 <do_help>) at main.c:134
>         #8  0x000000000040f88e in main (argc=<optimized out>, argv=<optimized out>) at main.c:469
>
> I'm on current bpf-next/master
>
>
> jirka
>
>
> > +     for (i = 1; i <= n; i++) {
> > +             struct btf_var_secinfo *vi;
> > +             struct btf_type *t;
> > +
> > +             t = btf_type_by_id(obj->btf, i);
> > +             if (btf_kind(t) != BTF_KIND_DATASEC)
> > +                     continue;
> > +
> > +             sec_name = btf__str_by_offset(obj->btf, t->name_off);
> > +             sec = find_src_sec_by_name(obj, sec_name);
> > +             if (sec) {
> > +                     /* record actual section size, unless ephemeral */
> > +                     if (sec->shdr)
> > +                             t->size = sec->shdr->sh_size;
> > +             } else {
> > +                     /* BTF can have some sections that are not represented
> > +                      * in ELF, e.g., .kconfig and .ksyms, which are used
> > +                      * for special extern variables.  Here we'll
> > +                      * pre-create "section shells" for them to be able to
> > +                      * keep track of extra per-section metadata later
> > +                      * (e.g., BTF variables).
> > +                      */
> > +                     sec = add_src_sec(obj, sec_name);
> > +                     if (!sec)
> > +                             return -ENOMEM;
> > +
> > +                     sec->ephemeral = true;
> > +                     sec->sec_idx = 0; /* will match UNDEF shndx in ELF */
> > +             }
> > +
> > +             /* remember ELF section and its BTF type ID match */
> > +             sec->sec_type_id = i;
> > +
> > +             /* fix up variable offsets */
> > +             vi = btf_var_secinfos(t);
> > +             for (j = 0, m = btf_vlen(t); j < m; j++, vi++) {
> > +                     const struct btf_type *vt = btf__type_by_id(obj->btf, vi->type);
> > +                     const char *var_name = btf__str_by_offset(obj->btf, vt->name_off);
> > +                     int var_linkage = btf_var(vt)->linkage;
> > +                     Elf64_Sym *sym;
> > +
> > +                     /* no need to patch up static or extern vars */
> > +                     if (var_linkage != BTF_VAR_GLOBAL_ALLOCATED)
> > +                             continue;
> > +
> > +                     sym = find_sym_by_name(obj, sec->sec_idx, STT_OBJECT, var_name);
> > +                     if (!sym) {
> > +                             pr_warn("failed to find symbol for variable '%s' in section '%s'\n", var_name, sec_name);
> > +                             return -ENOENT;
> > +                     }
> > +
> > +                     vi->offset = sym->st_value;
> > +             }
> > +     }
> > +
> > +     return 0;
> > +}
>
> SNIP
>

Powered by blists - more mailing lists