Skip to contentSkip to conversation

perf test 40 Basic BPF llvm compile dumps core (x86 and s390)

DISCUSSION Thomas Richter 3 participants 3 messages May 28, 2021 – May 30, 2021 BPF

lore
Share & tools
lore
#0 lore raw

I noticed perf test 40.1 dumps core on 5.13.0rc2 and rc3:

[root@f34 perf]# ./perf test -F 40
40: LLVM search and compile :
40.1: Basic BPF llvm compile :
libbpf: elf: skipping unrecognized data section(8) .eh_frame
libbpf: elf: skipping relo section(9) .rel.eh_frame for section(8) .eh_frame Segmentation fault (core dumped)
[root@f34 perf]#

The root cause is a NULL pointer reference in function btf__get_nr_types() as can be seen with gdb:

Breakpoint 1, 0x000000000065f4b1 in btf__get_nr_types

     (btf=btf@entry=0x0) at btf.c:425

425 return btf->start_id + btf->nr_types - 1;

This is the same function and reason why test case 42.1 Basic BPF filtering fails and dumps core too.

The call chain is:
(gdb) where
#0 0x000000000065f4b1 in btf__get_nr_types (btf=btf@entry=0x0) at btf.c:425 #1 btf__find_by_name_kind (btf=btf@entry=0x0,

		type_name=type_name@entry=0x928ab2 ".ksyms",
		kind=kind@entry=15) at btf.c:696
 #2  0x00000000006527fe in add_dummy_ksym_var (btf=0x0) at libbpf.c:3219
 #3  bpf_object__collect_externs (obj=0xd0ea20) at libbpf.c:3266
 #4  __bpf_object__open (path=<optimized out>, path@entry=0x0,
		obj_buf=obj_buf@entry=0xd12fa0,
		obj_buf_sz=obj_buf_sz@entry=1520,
    		opts=opts@entry=0x7fffffffdb30) at libbpf.c:7372
 #5  0x0000000000655415 in __bpf_object__open (opts=0x7fffffffdb30,
		obj_buf_sz=1520, obj_buf=0xd12fa0, path=0x0) at libbpf.c:7337
 #6  bpf_object__open_mem (opts=0x7fffffffdb30, obj_buf_sz=1520,
		obj_buf=0xd12fa0) at libbpf.c:7454
 #7  bpf_object__open_mem (opts=0x7fffffffdb30, obj_buf_sz=1520,
		obj_buf=0xd12fa0) at libbpf.c:7448
 #8  bpf_object__open_buffer (obj_buf=0xd12fa0, obj_buf_sz=1520,
		name=<optimized out>) at libbpf.c:7471
 #9  0x00000000004c8c6e in test__bpf_parsing (obj_buf=0xd12fa0,
		obj_buf_sz=1520) at tests/llvm.c:16
 #10 0x00000000004c8fe2 in test__llvm (test=0xac7d20 <generic_tests+2496>,
		subtest=0) at tests/llvm.c:142

...

I have no knowledge about BPF and why the core dump happens. Before I start digging into this has anybody some hints on where to look?

Thanks a lot.

Signature · 5 lines
Thomas Richter, Dept 3303, IBM s390 Linux Development, Boeblingen, Germany
--
Vorsitzender des Aufsichtsrats: Gregor Pillen
Geschäftsführung: Dirk Wittkopp
Sitz der Gesellschaft: Böblingen / Registergericht: Amtsgericht Stuttgart, HRB 243294
#1 lore raw

Em Fri, May 28, 2021 at 12:48:56PM +0200, Thomas Richter escreveu:

Quoted 10 lines

I noticed perf test 40.1 dumps core on 5.13.0rc2 and rc3:

[root@f34 perf]# ./perf test -F 40
40: LLVM search and compile :
40.1: Basic BPF llvm compile :
libbpf: elf: skipping unrecognized data section(8) .eh_frame
libbpf: elf: skipping relo section(9) .rel.eh_frame for section(8) .eh_frame Segmentation fault (core dumped)
[root@f34 perf]#

Quoted 2 lines

The root cause is a NULL pointer reference in function btf__get_nr_types() as can be seen with gdb:

This looks like a bug in libbpf:

static int bpf_object__collect_externs(struct bpf_object *obj)

{
        struct btf_type *sec, *kcfg_sec = NULL, *ksym_sec = NULL;
        const struct btf_type *t;
        struct extern_desc *ext;
        int i, n, off, dummy_var_btf_id;
        const char *ext_name, *sec_name;
        Elf_Scn *scn;
        GElf_Shdr sh;

        if (!obj->efile.symbols)
                return 0;

        scn = elf_sec_by_idx(obj, obj->efile.symbols_shndx);
        if (elf_sec_hdr(obj, scn, &sh))
                return -LIBBPF_ERRNO__FORMAT;

        dummy_var_btf_id = add_dummy_ksym_var(obj->btf);
        if (dummy_var_btf_id < 0)
                return dummy_var_btf_id;

obj->btf is NULL, so probably btf__find_by_name_kind() should check that and return an error, Andrii?

- Arnaldo

Quoted 44 lines

Breakpoint 1, 0x000000000065f4b1 in btf__get_nr_types

     (btf=btf@entry=0x0) at btf.c:425

425 return btf->start_id + btf->nr_types - 1;

This is the same function and reason why test case 42.1 Basic BPF filtering fails and dumps core too.

The call chain is:
(gdb) where
#0 0x000000000065f4b1 in btf__get_nr_types (btf=btf@entry=0x0) at btf.c:425 #1 btf__find_by_name_kind (btf=btf@entry=0x0,

		type_name=type_name@entry=0x928ab2 ".ksyms",
		kind=kind@entry=15) at btf.c:696
 #2  0x00000000006527fe in add_dummy_ksym_var (btf=0x0) at libbpf.c:3219
 #3  bpf_object__collect_externs (obj=0xd0ea20) at libbpf.c:3266
 #4  __bpf_object__open (path=<optimized out>, path@entry=0x0,
		obj_buf=obj_buf@entry=0xd12fa0,
		obj_buf_sz=obj_buf_sz@entry=1520,
    		opts=opts@entry=0x7fffffffdb30) at libbpf.c:7372
 #5  0x0000000000655415 in __bpf_object__open (opts=0x7fffffffdb30,
		obj_buf_sz=1520, obj_buf=0xd12fa0, path=0x0) at libbpf.c:7337
 #6  bpf_object__open_mem (opts=0x7fffffffdb30, obj_buf_sz=1520,
		obj_buf=0xd12fa0) at libbpf.c:7454
 #7  bpf_object__open_mem (opts=0x7fffffffdb30, obj_buf_sz=1520,
		obj_buf=0xd12fa0) at libbpf.c:7448
 #8  bpf_object__open_buffer (obj_buf=0xd12fa0, obj_buf_sz=1520,
		name=<optimized out>) at libbpf.c:7471
 #9  0x00000000004c8c6e in test__bpf_parsing (obj_buf=0xd12fa0,
		obj_buf_sz=1520) at tests/llvm.c:16
 #10 0x00000000004c8fe2 in test__llvm (test=0xac7d20 <generic_tests+2496>,
		subtest=0) at tests/llvm.c:142

...

I have no knowledge about BPF and why the core dump happens. Before I start digging into this has anybody some hints on where to look?

Thanks a lot.

Signature · 5 lines
Thomas Richter, Dept 3303, IBM s390 Linux Development, Boeblingen, Germany
--
Vorsitzender des Aufsichtsrats: Gregor Pillen
Geschäftsführung: Dirk Wittkopp
Sitz der Gesellschaft: Böblingen / Registergericht: Amtsgericht Stuttgart, HRB 243294
Signature · 1 line
- Arnaldo
#2 lore raw
Quoted 39 lines · Arnaldo Carvalho de Melo

Em Fri, May 28, 2021 at 12:48:56PM +0200, Thomas Richter escreveu:

Quoted 10 lines

I noticed perf test 40.1 dumps core on 5.13.0rc2 and rc3:

[root@f34 perf]# ./perf test -F 40
40: LLVM search and compile :
40.1: Basic BPF llvm compile :
libbpf: elf: skipping unrecognized data section(8) .eh_frame
libbpf: elf: skipping relo section(9) .rel.eh_frame for section(8) .eh_frame Segmentation fault (core dumped)
[root@f34 perf]#

Quoted 2 lines

The root cause is a NULL pointer reference in function btf__get_nr_types() as can be seen with gdb:

This looks like a bug in libbpf:

static int bpf_object__collect_externs(struct bpf_object *obj)

{
        struct btf_type *sec, *kcfg_sec = NULL, *ksym_sec = NULL;
        const struct btf_type *t;
        struct extern_desc *ext;
        int i, n, off, dummy_var_btf_id;
        const char *ext_name, *sec_name;
        Elf_Scn *scn;
        GElf_Shdr sh;

        if (!obj->efile.symbols)
                return 0;

        scn = elf_sec_by_idx(obj, obj->efile.symbols_shndx);
        if (elf_sec_hdr(obj, scn, &sh))
                return -LIBBPF_ERRNO__FORMAT;

        dummy_var_btf_id = add_dummy_ksym_var(obj->btf);
        if (dummy_var_btf_id < 0)
                return dummy_var_btf_id;
Quoted 3 lines

obj->btf is NULL, so probably btf__find_by_name_kind() should check that and return an error, Andrii?

This is already fixed in libbpf v0.4 ([0]). Can you please update pahole's libbpf submodule sha?

[0] https://github.com/libbpf/libbpf/commit/a58b8ca93e39285a2852ca3d60e698598bf7265b

Quoted 50 lines

- Arnaldo

Quoted 36 lines

Breakpoint 1, 0x000000000065f4b1 in btf__get_nr_types

     (btf=btf@entry=0x0) at btf.c:425

425 return btf->start_id + btf->nr_types - 1;

This is the same function and reason why test case 42.1 Basic BPF filtering fails and dumps core too.

The call chain is:
(gdb) where
#0 0x000000000065f4b1 in btf__get_nr_types (btf=btf@entry=0x0) at btf.c:425 #1 btf__find_by_name_kind (btf=btf@entry=0x0,

              type_name=type_name@entry=0x928ab2 ".ksyms",
              kind=kind@entry=15) at btf.c:696
 #2  0x00000000006527fe in add_dummy_ksym_var (btf=0x0) at libbpf.c:3219
 #3  bpf_object__collect_externs (obj=0xd0ea20) at libbpf.c:3266
 #4  __bpf_object__open (path=<optimized out>, path@entry=0x0,
              obj_buf=obj_buf@entry=0xd12fa0,
              obj_buf_sz=obj_buf_sz@entry=1520,
              opts=opts@entry=0x7fffffffdb30) at libbpf.c:7372
 #5  0x0000000000655415 in __bpf_object__open (opts=0x7fffffffdb30,
              obj_buf_sz=1520, obj_buf=0xd12fa0, path=0x0) at libbpf.c:7337
 #6  bpf_object__open_mem (opts=0x7fffffffdb30, obj_buf_sz=1520,
              obj_buf=0xd12fa0) at libbpf.c:7454
 #7  bpf_object__open_mem (opts=0x7fffffffdb30, obj_buf_sz=1520,
              obj_buf=0xd12fa0) at libbpf.c:7448
 #8  bpf_object__open_buffer (obj_buf=0xd12fa0, obj_buf_sz=1520,
              name=<optimized out>) at libbpf.c:7471
 #9  0x00000000004c8c6e in test__bpf_parsing (obj_buf=0xd12fa0,
              obj_buf_sz=1520) at tests/llvm.c:16
 #10 0x00000000004c8fe2 in test__llvm (test=0xac7d20 <generic_tests+2496>,
              subtest=0) at tests/llvm.c:142

...

I have no knowledge about BPF and why the core dump happens. Before I start digging into this has anybody some hints on where to look?

Quoted 8 lines

Thanks a lot.

--
Thomas Richter, Dept 3303, IBM s390 Linux Development, Boeblingen, Germany --
Vorsitzender des Aufsichtsrats: Gregor Pillen
Geschäftsführung: Dirk Wittkopp
Sitz der Gesellschaft: Böblingen / Registergericht: Amtsgericht Stuttgart, HRB 243294

--

- Arnaldo