Re: [PATCH V2 8/9] perf tools: Adjust kallsyms for relocated kernel

[Arnaldo Carvalho de Melo]

Summary: I'm applying this patch as it is reported to fix problems, but
answering anyway to outline my reasoning, that I eventually should put
in place to simplify the reloc handling:

Em Thu, Jan 30, 2014 at 10:10:50AM +0200, Adrian Hunter escreveu:
> On 29/01/14 21:08, Arnaldo Carvalho de Melo wrote:
> > Em Wed, Jan 29, 2014 at 04:14:43PM +0200, Adrian Hunter escreveu:
> >> If the kernel is relocated at boot time, kallsyms
> >> will not match data recorded previously.  That
> >> does not matter for modules because they are
> >> corrected anyway.  It also does not matter if
> >> vmlinux is being used for symbols. But if perf
> >> tools has only kallsyms then the symbols will not
> >> match.  Fix by applying the delta gained by
> >> comparing the old and current addresses of the
> >> relocation reference symbol.
> > 
> > Don't we have map->reloc? Can't we use it here, i.e. be consistent and
> > have the unrelocated address + the relocation, and then, when using
> > map->{map,unmap} take that into account, as you did to the objdump
> > variants, no?
> 
> map->reloc is calculated from vmlinux.  No vmlinux, no map->reloc.

I see that this is how it is being used now, yes.

> e.g. we don't need to calculate objdump addresses if we have no
> object code anyway.

But map->reloc doesn't need to be used just for objdump purposes.

> Also without vmlinux we do not know what the actual relocation was, we
> just know the difference between where the kernel was located when the
> data was recorded and where it is now.  That is the "delta".

A relocation is a delta as well, no? I'm just trying to figure out why
can't we avoid adding this 'delta' thing and instead use map->reloc for
that.
 
> This is the case where all we have is kallsyms.  In that case
> perf tools uses identity maps (identity__map_ip).  An alternative
> would be to change to use map__[un]map_ip() and set map->pgoff to
> match the current kallsyms. i.e. map->pgoff = map->start + delta

Why not use map->reloc for that? When we don't find any delta, then it
is zero and all is as today, when we find the delta, be it from vmlinux
or from a difference in the ref_reloc_symbol, we set it up, no?
 
> So we need either to change the symbols to match the mapping or
> change the mapping to match the symbols.  I went with the former
> because we are already changing the module symbols so it seemed
> consistent.

Well, for modules what we do is apply a relocation in place, like you do
with this delta, no? I.e. what I'm saying is that it should work both
ways, but we should try to be consistent accross the board, i.e. fully
use map->reloc and stop changing the address we get from the original
symtab.

Anyway, that can be done in a followup patch as your patch has been
reported to fix real bugs, thanks.

- Arnaldo
 
> > 
> > I applied the others, testing now.
> > 
> > - Arnaldo
> >  
> >> Signed-off-by: Adrian Hunter <adrian.hunter@xxxxxxxxx>
> >> ---
> >>  tools/perf/util/symbol.c | 40 ++++++++++++++++++++++++++++++++++++++--
> >>  1 file changed, 38 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c
> >> index 4ac1f87..a9d758a 100644
> >> --- a/tools/perf/util/symbol.c
> >> +++ b/tools/perf/util/symbol.c
> >> @@ -627,7 +627,7 @@ static int dso__split_kallsyms_for_kcore(struct dso *dso, struct map *map,
> >>   * kernel range is broken in several maps, named [kernel].N, as we don't have
> >>   * the original ELF section names vmlinux have.
> >>   */
> >> -static int dso__split_kallsyms(struct dso *dso, struct map *map,
> >> +static int dso__split_kallsyms(struct dso *dso, struct map *map, u64 delta,
> >>  			       symbol_filter_t filter)
> >>  {
> >>  	struct map_groups *kmaps = map__kmap(map)->kmaps;
> >> @@ -692,6 +692,12 @@ static int dso__split_kallsyms(struct dso *dso, struct map *map,
> >>  			char dso_name[PATH_MAX];
> >>  			struct dso *ndso;
> >>  
> >> +			if (delta) {
> >> +				/* Kernel was relocated at boot time */
> >> +				pos->start -= delta;
> >> +				pos->end -= delta;
> >> +			}
> >> +
> >>  			if (count == 0) {
> >>  				curr_map = map;
> >>  				goto filter_symbol;
> >> @@ -721,6 +727,10 @@ static int dso__split_kallsyms(struct dso *dso, struct map *map,
> >>  			curr_map->map_ip = curr_map->unmap_ip = identity__map_ip;
> >>  			map_groups__insert(kmaps, curr_map);
> >>  			++kernel_range;
> >> +		} else if (delta) {
> >> +			/* Kernel was relocated at boot time */
> >> +			pos->start -= delta;
> >> +			pos->end -= delta;
> >>  		}
> >>  filter_symbol:
> >>  		if (filter && filter(curr_map, pos)) {
> >> @@ -1130,15 +1140,41 @@ out_err:
> >>  	return -EINVAL;
> >>  }
> >>  
> >> +/*
> >> + * If the kernel is relocated at boot time, kallsyms won't match.  Compute the
> >> + * delta based on the relocation reference symbol.
> >> + */
> >> +static int kallsyms__delta(struct map *map, const char *filename, u64 *delta)
> >> +{
> >> +	struct kmap *kmap = map__kmap(map);
> >> +	u64 addr;
> >> +
> >> +	if (!kmap->ref_reloc_sym || !kmap->ref_reloc_sym->name)
> >> +		return 0;
> >> +
> >> +	addr = kallsyms__get_function_start(filename,
> >> +					    kmap->ref_reloc_sym->name);
> >> +	if (!addr)
> >> +		return -1;
> >> +
> >> +	*delta = addr - kmap->ref_reloc_sym->addr;
> >> +	return 0;
> >> +}
> >> +
> >>  int dso__load_kallsyms(struct dso *dso, const char *filename,
> >>  		       struct map *map, symbol_filter_t filter)
> >>  {
> >> +	u64 delta = 0;
> >> +
> >>  	if (symbol__restricted_filename(filename, "/proc/kallsyms"))
> >>  		return -1;
> >>  
> >>  	if (dso__load_all_kallsyms(dso, filename, map) < 0)
> >>  		return -1;
> >>  
> >> +	if (kallsyms__delta(map, filename, &delta))
> >> +		return -1;
> >> +
> >>  	symbols__fixup_duplicate(&dso->symbols[map->type]);
> >>  	symbols__fixup_end(&dso->symbols[map->type]);
> >>  
> >> @@ -1150,7 +1186,7 @@ int dso__load_kallsyms(struct dso *dso, const char *filename,
> >>  	if (!dso__load_kcore(dso, map, filename))
> >>  		return dso__split_kallsyms_for_kcore(dso, map, filter);
> >>  	else
> >> -		return dso__split_kallsyms(dso, map, filter);
> >> +		return dso__split_kallsyms(dso, map, delta, filter);
> >>  }
> >>  
> >>  static int dso__load_perf_map(struct dso *dso, struct map *map,
> >> -- 
> >> 1.7.11.7
> > 
> > 
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/