[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20170623135547.GB4622@kernel.org>
Date: Fri, 23 Jun 2017 10:55:47 -0300
From: Arnaldo Carvalho de Melo <acme@...nel.org>
To: Namhyung Kim <namhyung@...nel.org>
Cc: Ingo Molnar <mingo@...nel.org>,
Peter Zijlstra <a.p.zijlstra@...llo.nl>,
Jiri Olsa <jolsa@...nel.org>,
LKML <linux-kernel@...r.kernel.org>, kernel-team@....com,
Masami Hiramatsu <mhiramat@...nel.org>,
Andi Kleen <andi@...stfloor.org>,
Adrian Hunter <adrian.hunter@...el.com>,
Wang Nan <wangnan0@...wei.com>
Subject: Re: [PATCH/RFC 6/9] perf symbols: Use already loaded module dso when
loading kcore
Em Fri, Jun 23, 2017 at 02:48:24PM +0900, Namhyung Kim escreveu:
> Even every module has loaded onto same addresses, some modules can be
> changed and reloaded.
Can you rephrase the above statement?
You mean even if a module is reloaded at the same address it can have a
different symtab and thus misresolve symbols if we get a sample for one
module version and try to resolve symbols using a symtab for another
module version?
> In that case it needs to access to the old module
> in the build-id cache.
Right, we need to make sure that during a 'perf record' session a módule
isn't reloaded, because at 'perf record' exit we will record that module
build-id in a perf.data header and grab a copy (or a hardlink, if
possible) into the ~/.debug build-id cache, then, at 'perf report' time
we will read that build-id perf.data table and lookup in the build-id
cache to get the matching symtab (and full elf .ko file to do things
like annotation).
- Arnaldo
> Cc: Adrian Hunter <adrian.hunter@...el.com>
> Cc: Wang Nan <wangnan0@...wei.com>
> Signed-off-by: Namhyung Kim <namhyung@...nel.org>
> ---
> tools/perf/util/symbol.c | 45 +++++++++++++++++++++++----------------------
> 1 file changed, 23 insertions(+), 22 deletions(-)
>
> diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c
> index ce79a51f25bf..fe46eb782297 100644
> --- a/tools/perf/util/symbol.c
> +++ b/tools/perf/util/symbol.c
> @@ -1155,6 +1155,7 @@ static int dso__load_kcore(struct dso *dso, struct map *map,
> int err, fd;
> char kcore_filename[PATH_MAX];
> struct symbol *sym;
> + struct map_groups old_mg;
>
> if (!kmaps)
> return -EINVAL;
> @@ -1196,15 +1197,8 @@ static int dso__load_kcore(struct dso *dso, struct map *map,
> goto out_err;
> }
>
> - /* Remove old maps */
> - old_map = map_groups__first(kmaps, map->type);
> - while (old_map) {
> - struct map *next = map_groups__next(old_map);
> -
> - if (old_map != map)
> - map_groups__remove(kmaps, old_map);
> - old_map = next;
> - }
> + old_mg = *kmaps;
> + kmaps->maps[map->type].entries = RB_ROOT;
>
> /* Find the kernel map using the first symbol */
> sym = dso__first_symbol(dso, map->type);
> @@ -1223,24 +1217,31 @@ static int dso__load_kcore(struct dso *dso, struct map *map,
> while (!list_empty(&md.maps)) {
> new_map = list_entry(md.maps.next, struct map, node);
> list_del_init(&new_map->node);
> - if (new_map == replacement_map) {
> - map->start = new_map->start;
> - map->end = new_map->end;
> - map->pgoff = new_map->pgoff;
> - map->map_ip = new_map->map_ip;
> - map->unmap_ip = new_map->unmap_ip;
> - /* Ensure maps are correctly ordered */
> - map__get(map);
> - map_groups__remove(kmaps, map);
> - map_groups__insert(kmaps, map);
> - map__put(map);
> - } else {
> - map_groups__insert(kmaps, new_map);
> +
> + map_groups__insert(kmaps, new_map);
> +
> + if (new_map != replacement_map) {
> + old_map = map_groups__find(&old_mg, new_map->type, new_map->start);
> + if (old_map && dso__loaded(old_map->dso, old_map->type)) {
> + new_map->pgoff = old_map->pgoff;
> +
> + dso__put(new_map->dso);
> + new_map->dso = dso__get(old_map->dso);
> + }
> }
>
> map__put(new_map);
> }
>
> + /* Remove old maps */
> + old_map = map_groups__first(&old_mg, map->type);
> + while (old_map) {
> + struct map *next = map_groups__next(old_map);
> +
> + map_groups__remove(&old_mg, old_map);
> + old_map = next;
> + }
> +
> /*
> * Set the data type and long name so that kcore can be read via
> * dso__data_read_addr().
> --
> 2.13.1
Powered by blists - more mailing lists