Messages in this thread Patch in this message | | | Subject | Re: [PATCH] perf-probe: dso: Add symbols in .text.* subsections to text symbol map in kenrel modules | From | Evgenii Shatokhin <> | Date | Tue, 23 Feb 2021 18:02:58 +0300 |
| |
On 23.02.2021 10:37, Masami Hiramatsu wrote: > The kernel modules have .text.* subsections such as .text.unlikely. > Since dso__process_kernel_symbol() only identify the symbols in the ".text" > section as the text symbols and inserts it in the default dso in the map, > the symbols in such subsections can not be found by map__find_symbol(). > > This adds the symbols in those subsections to the default dso in the map so > that map__find_symbol() can find them. This solves the perf-probe issue on > probing online module. > > Without this fix, probing on a symbol in .text.unlikely fails. > ---- > # perf probe -m nf_conntrack nf_l4proto_log_invalid > Probe point 'nf_l4proto_log_invalid' not found. > Error: Failed to add events. > ---- > > With this fix, it works because map__find_symbol() can find the symbol > correctly. > ---- > # perf probe -m nf_conntrack nf_l4proto_log_invalid > Added new event: > probe:nf_l4proto_log_invalid (on nf_l4proto_log_invalid in nf_conntrack) > > You can now use it in all perf tools, such as: > > perf record -e probe:nf_l4proto_log_invalid -aR sleep 1 > > ---- > > Reported-by: Evgenii Shatokhin <eshatokhin@virtuozzo.com> > Signed-off-by: Masami Hiramatsu <mhiramat@kernel.org>
Thanks for the fix!
It looks like it helps, at least with nf_conntrack in kernel 5.11: --------------------- # ./perf probe -v -m nf_conntrack nf_ct_resolve_clash probe-definition(0): nf_ct_resolve_clash symbol:nf_ct_resolve_clash file:(null) line:0 offset:0 return:0 lazy:(null) 0 arguments Failed to get build-id from nf_conntrack. Cache open error: -1 Open Debuginfo file: /lib/modules/5.11.0-test01/kernel/net/netfilter/nf_conntrack.ko Try to find probe point from debuginfo. Matched function: nf_ct_resolve_clash [33616] Probe point found: nf_ct_resolve_clash+0 Found 1 probe_trace_events. Opening /sys/kernel/tracing//kprobe_events write=1 Opening /sys/kernel/tracing//README write=0 Writing event: p:probe/nf_ct_resolve_clash nf_conntrack:nf_ct_resolve_clash+0 Added new event: probe:nf_ct_resolve_clash (on nf_ct_resolve_clash in nf_conntrack)
You can now use it in all perf tools, such as:
perf record -e probe:nf_ct_resolve_clash -aR sleep 1 ---------------------
I guess, the patch is suitable for stable kernel branches as well.
Without the patch, the workaround you suggested earlier (using the full path to nf_conntrack.ko) works too.
> --- > tools/perf/util/symbol-elf.c | 4 +++- > 1 file changed, 3 insertions(+), 1 deletion(-) > > diff --git a/tools/perf/util/symbol-elf.c b/tools/perf/util/symbol-elf.c > index 6dff843fd883..0c1113236913 100644 > --- a/tools/perf/util/symbol-elf.c > +++ b/tools/perf/util/symbol-elf.c > @@ -985,7 +985,9 @@ static int dso__process_kernel_symbol(struct dso *dso, struct map *map, > if (strcmp(section_name, (curr_dso->short_name + dso->short_name_len)) == 0) > return 0; > > - if (strcmp(section_name, ".text") == 0) { > + /* .text and .text.* are included in the text dso */ > + if (strncmp(section_name, ".text", 5) == 0 && > + (section_name[5] == '\0' || section_name[5] == '.')) { > /* > * The initial kernel mapping is based on > * kallsyms and identity maps. Overwrite it to > > . >
Regards, Evgenii
| |