perf dso: Fix dso comparison
PATCH 1/1 perf dso: Fix dso comparison
Perf gets dso details from two different sources. 1st, from builid headers in perf.data and 2nd from MMAP2 samples. Dso from buildid header does not have dso_id detail. And dso from MMAP2 samples does not have buildid information. If detail of the same dso is present at both the places, filename is common.
Previously, __dsos__findnew_link_by_longname_id() used to compare only long or short names, but Commit 0e3149f86b99 ("perf dso: Move dso_id from 'struct map' to 'struct dso'") also added a dso_id comparison. Because of that, now perf is creating two different dso objects of the same file, one from buildid header (with dso_id but without buildid) and second from MMAP2 sample (with buildid but without dso_id).
This is causing issues with archive, buildid-list etc subcommands. Fix this by comparing dso_id only when it's present. And incase dso is present in 'dsos' list without dso_id, inject dso_id detail as well.
Before:
$ sudo ./perf buildid-list -H 0000000000000000000000000000000000000000 /usr/bin/ls 0000000000000000000000000000000000000000 /usr/lib64/ld-2.30.so 0000000000000000000000000000000000000000 /usr/lib64/libc-2.30.so
$ ./perf archive perf archive: no build-ids found
After:
$ ./perf buildid-list -H b6b1291d0cead046ed0fa5734037fa87a579adee /usr/bin/ls 641f0c90cfa15779352f12c0ec3c7a2b2b6f41e8 /usr/lib64/ld-2.30.so 675ace3ca07a0b863df01f461a7b0984c65c8b37 /usr/lib64/libc-2.30.so
$ ./perf archive Now please run:
$ tar xvf perf.data.tar.bz2 -C ~/.debug
wherever you need to run 'perf report' on.
--- tools/perf/util/dsos.c | 22 +++++++++++++++++++++- 1 file changed, 21 insertions(+), 1 deletion(-)
tools/perf/util/dsos.c +21 -1
@@ -26,13 +26,29 @@ static int __dso_id__cmp(struct dso_id *a, struct dso_id *b)return 0 ; } +static bool is_empty_dso_id (struct dso_id *id) +{ +if (!id) +return true; + +return !id->maj && !id->min && !id->ino && !id->ino_generation; +} + +static void inject_dso_id (struct dso *dso,struct dso_id *id) +{ + dso->id.maj = id->maj; + dso->id.min = id->min; + dso->id.ino = id->ino; + dso->id.ino_generation = id->ino_generation; +} +static int dso_id__cmp (struct dso_id *a,struct dso_id *b) {/* * The second is always dso->id, so zeroes if not set, assume passing * NULL for a means a zeroed id */ -if (a ==NULL ) +if (is_empty_dso_id (a) ||is_empty_dso_id (b))return 0 ;return __dso_id__cmp (a, b);
@@ -249,6 +265,10 @@ struct dso *__dsos__addnew(struct dsos *dsos, const char *name)static struct dso *__dsos__findnew_id (struct dsos *dsos,const char *name,struct dso_id *id) {struct dso *dso =__dsos__find_id (dsos, name, id, false); + +if (dso &&is_empty_dso_id (&dso->id) && !is_empty_dso_id (id)) +inject_dso_id (dso, id); +return dso ? dso :__dsos__addnew_id (dsos, name, id); }
Signature · 1 line
2.24.1
Quoted 34 lines · Ravi Bangoria
Perf gets dso details from two different sources. 1st, from builid headers in perf.data and 2nd from MMAP2 samples. Dso from buildid header does not have
dso_iddetail. And dso from MMAP2 samples does not have buildid information. If detail of the same dso is present at both the places, filename is common.Previously,
__dsos__findnew_link_by_longname_id()used to compare only long or short names, but Commit0e3149f86b99("perf dso: Movedso_idfrom 'struct map' to 'struct dso'") also added adso_idcomparison. Because of that, now perf is creating two different dso objects of the same file, one from buildid header (withdso_idbut without buildid) and second from MMAP2 sample (with buildid but withoutdso_id).This is causing issues with archive, buildid-list etc subcommands. Fix this by comparing
dso_idonly when it's present. And incase dso is present in 'dsos' list withoutdso_id, injectdso_iddetail as well.Before:
$ sudo ./perf buildid-list -H 0000000000000000000000000000000000000000 /usr/bin/ls 0000000000000000000000000000000000000000 /usr/lib64/ld-2.30.so 0000000000000000000000000000000000000000 /usr/lib64/libc-2.30.so$ ./perf archive perf archive: no build-ids foundAfter:
$ ./perf buildid-list -H b6b1291d0cead046ed0fa5734037fa87a579adee /usr/bin/ls 641f0c90cfa15779352f12c0ec3c7a2b2b6f41e8 /usr/lib64/ld-2.30.so 675ace3ca07a0b863df01f461a7b0984c65c8b37 /usr/lib64/libc-2.30.so
Quoted 8 lines
$ ./perf archive Now please run:$ tar xvf perf.data.tar.bz2 -C ~/.debugwherever you need to run 'perf report' on.
Reported-by Naveen N. Rao <[email protected]>
looks good, do we need to add the dso_id check to sort__dso_cmp?
thanks,
jirka
Quoted 38 lines · Jiri Olsa
Quoted 37 lines · Ravi Bangoria
Perf gets dso details from two different sources. 1st, from builid headers in perf.data and 2nd from MMAP2 samples. Dso from buildid header does not have
dso_iddetail. And dso from MMAP2 samples does not have buildid information. If detail of the same dso is present at both the places, filename is common.Previously,
__dsos__findnew_link_by_longname_id()used to compare only long or short names, but Commit0e3149f86b99("perf dso: Movedso_idfrom 'struct map' to 'struct dso'") also added adso_idcomparison. Because of that, now perf is creating two different dso objects of the same file, one from buildid header (withdso_idbut without buildid) and second from MMAP2 sample (with buildid but withoutdso_id).This is causing issues with archive, buildid-list etc subcommands. Fix this by comparing
dso_idonly when it's present. And incase dso is present in 'dsos' list withoutdso_id, injectdso_iddetail as well.Before:
$ sudo ./perf buildid-list -H 0000000000000000000000000000000000000000 /usr/bin/ls 0000000000000000000000000000000000000000 /usr/lib64/ld-2.30.so 0000000000000000000000000000000000000000 /usr/lib64/libc-2.30.so$ ./perf archive perf archive: no build-ids foundAfter:
$ ./perf buildid-list -H b6b1291d0cead046ed0fa5734037fa87a579adee /usr/bin/ls 641f0c90cfa15779352f12c0ec3c7a2b2b6f41e8 /usr/lib64/ld-2.30.so 675ace3ca07a0b863df01f461a7b0984c65c8b37 /usr/lib64/libc-2.30.so$ ./perf archive Now please run:
Quoted 7 lines
Quoted 5 lines
$ tar xvf perf.data.tar.bz2 -C ~/.debugwherever you need to run 'perf report' on.
Reported-by Naveen N. Rao <[email protected]>looks good, do we need to add the
dso_idcheck tosort__dso_cmp?
I guess with different filename there is no need to compare dso_id. But for same filename, adding dso_id cmp will separate out the samples:
Ex, Without dso_id compare:
$ ./perf report -s dso,dso_size -v
66.63% /home/ravi/a.out 4096
33.36% /home/ravi/Workspace/linux/tools/perf/a.out 4096 $ ./perf report -s dso,dso_size
99.99% a.out 4096With below diff:
- return strcmp(dso_name_l, dso_name_r); + ret = strcmp(dso_name_l, dso_name_r); + if (ret)
+ return ret;
+ else
+ return dso__cmp_id(dso_l, dso_r);
$ ./perf report -s dso,dso_size
99.99% a.out 4096
33.36% a.out 4096though, the o/p also depends which other sort keys are used along with dso key. Do you think this change makes sense?
Ravi
On Tue, Mar 24, 2020 at 06:07:23PM +0530, Ravi Bangoria wrote:
SNIP
Quoted 26 lines
Quoted 1 line
looks good, do we need to add the
dso_idcheck tosort__dso_cmp?I guess with different filename there is no need to compare
dso_id. But for same filename, addingdso_idcmp will separate out the samples:Ex, Without
dso_idcompare:$ ./perf report -s dso,dso_size -v 66.63% /home/ravi/a.out 4096 33.36% /home/ravi/Workspace/linux/tools/perf/a.out 4096$ ./perf report -s dso,dso_size 99.99% a.out 4096With below diff:
- return strcmp(dso_name_l, dso_name_r); + ret = strcmp(dso_name_l, dso_name_r); + if (ret)+ return ret;
+ else
+ returndso__cmp_id(dso_l, dso_r);
Quoted 6 lines
$ ./perf report -s dso,dso_size 99.99% a.out 4096 33.36% a.out 4096though, the o/p also depends which other sort keys are used along with dso key. Do you think this change makes sense?
the above behaviour is something I'd expect from 'dso' sort key to do - separate out different dsos, even with the same name
jirka
Em Tue, Mar 24, 2020 at 02:22:58PM +0100, Jiri Olsa escreveu:
Quoted 34 lines · Ravi Bangoria
SNIP
Quoted 30 lines
Quoted 1 line
looks good, do we need to add the
dso_idcheck tosort__dso_cmp?I guess with different filename there is no need to compare
dso_id. But for same filename, addingdso_idcmp will separate out the samples:Ex, Without
dso_idcompare:$ ./perf report -s dso,dso_size -v 66.63% /home/ravi/a.out 4096 33.36% /home/ravi/Workspace/linux/tools/perf/a.out 4096$ ./perf report -s dso,dso_size 99.99% a.out 4096With below diff:
- return strcmp(dso_name_l, dso_name_r); + ret = strcmp(dso_name_l, dso_name_r); + if (ret)+ return ret;
+ else
+ returndso__cmp_id(dso_l, dso_r);$ ./perf report -s dso,dso_size 99.99% a.out 4096 33.36% a.out 4096
Quoted 6 lines
Quoted 2 lines
though, the o/p also depends which other sort keys are used along with dso key. Do you think this change makes sense?
the above behaviour is something I'd expect from 'dso' sort key to do - separate out different dsos, even with the same name
This specific one can be resolved using -v when long_name will be used, the biggest problem is when long_name is the same (and thus short_name), which can happen when developing some software, i.e. compile+rebuild and get a different content, same short/long name, in that case we should use some diferentiator, the build-id comes to mind, but one that could be more useful would be file timestamp, meaning, hey, the older version is actually better, which one, lemme look at the build-id, and even the source code if developed with -g, by using the copy we stored in the build-id cache (~/.debug), which would be really useful workflow.
- Arnaldo
Quoted 2 lines
jirka
Signature · 1 line
- Arnaldo
Quoted 34 lines · Jiri Olsa
On Tue, Mar 24, 2020 at 06:07:23PM +0530, Ravi Bangoria wrote:
SNIP
Quoted 30 lines
Quoted 1 line
looks good, do we need to add the
dso_idcheck tosort__dso_cmp?I guess with different filename there is no need to compare
dso_id. But for same filename, addingdso_idcmp will separate out the samples:Ex, Without
dso_idcompare:$ ./perf report -s dso,dso_size -v 66.63% /home/ravi/a.out 4096 33.36% /home/ravi/Workspace/linux/tools/perf/a.out 4096$ ./perf report -s dso,dso_size 99.99% a.out 4096With below diff:
- return strcmp(dso_name_l, dso_name_r); + ret = strcmp(dso_name_l, dso_name_r); + if (ret)+ return ret;
+ else
+ returndso__cmp_id(dso_l, dso_r);$ ./perf report -s dso,dso_size 99.99% a.out 4096 33.36% a.out 4096
Quoted 6 lines
Quoted 2 lines
though, the o/p also depends which other sort keys are used along with dso key. Do you think this change makes sense?
the above behaviour is something I'd expect from 'dso' sort key to do - separate out different dsos, even with the same name
Yes it does that as well...
$ ./perf report -s dso
66.63% a.out
33.36% a.outRavi
Em Tue, Mar 24, 2020 at 11:48:43AM +0100, Jiri Olsa escreveu:
Quoted 38 lines · Ravi Bangoria
Quoted 37 lines
Perf gets dso details from two different sources. 1st, from builid headers in perf.data and 2nd from MMAP2 samples. Dso from buildid header does not have
dso_iddetail. And dso from MMAP2 samples does not have buildid information. If detail of the same dso is present at both the places, filename is common.Previously,
__dsos__findnew_link_by_longname_id()used to compare only long or short names, but Commit0e3149f86b99("perf dso: Movedso_idfrom 'struct map' to 'struct dso'") also added adso_idcomparison. Because of that, now perf is creating two different dso objects of the same file, one from buildid header (withdso_idbut without buildid) and second from MMAP2 sample (with buildid but withoutdso_id).This is causing issues with archive, buildid-list etc subcommands. Fix this by comparing
dso_idonly when it's present. And incase dso is present in 'dsos' list withoutdso_id, injectdso_iddetail as well.Before:
$ sudo ./perf buildid-list -H 0000000000000000000000000000000000000000 /usr/bin/ls 0000000000000000000000000000000000000000 /usr/lib64/ld-2.30.so 0000000000000000000000000000000000000000 /usr/lib64/libc-2.30.so$ ./perf archive perf archive: no build-ids foundAfter:
$ ./perf buildid-list -H b6b1291d0cead046ed0fa5734037fa87a579adee /usr/bin/ls 641f0c90cfa15779352f12c0ec3c7a2b2b6f41e8 /usr/lib64/ld-2.30.so 675ace3ca07a0b863df01f461a7b0984c65c8b37 /usr/lib64/libc-2.30.so$ ./perf archive Now please run:
Quoted 7 lines
Quoted 5 lines
$ tar xvf perf.data.tar.bz2 -C ~/.debugwherever you need to run 'perf report' on.
Reported-by Naveen N. Rao <[email protected]>looks good, do we need to add the
dso_idcheck tosort__dso_cmp?
Jiri:
Humm, you mean sort__dso_cmp() -> _sort__dso_cmp() should consider the dso_id and not just its name? Humm, when "dso" sort key is used that means just the short_name (or long_name, if verbose), if we use the ID for "dso" then we need to somehow show the id in the output, otherwise we'd have multiple lines with the same DSO name, when multiple versions exist... Perhaps we should do a first pass, figure out if there are DSOs with the same name/different IDs and mark them for showing the ID to differentiate them on the output? But this is something that should be dealt with in a separece cset, I think.
With that in mind, can I add your Acked-by for this patch, with my changes described below?
Ravi:
I'm applying it with the changes below, to keep namespacing consistency, ok?
- Arnaldo
tools/perf/util/dsos.c +5 -5
@@ -26,7 +26,7 @@ static int __dso_id__cmp(struct dso_id *a, struct dso_id *b)return 0 ; } -static bool is_empty_dso_id (struct dso_id *id) +static bool dso_id__empty (struct dso_id *id) {if (!id)return true;
@@ -34,7 +34,7 @@ static bool is_empty_dso_id(struct dso_id *id)return !id->maj && !id->min && !id->ino && !id->ino_generation; } -static void inject_dso_id (struct dso *dso,struct dso_id *id) +static void dso__inject_id (struct dso *dso,struct dso_id *id) { dso->id.maj = id->maj; dso->id.min = id->min;
@@ -48,7 +48,7 @@ static int dso_id__cmp(struct dso_id *a, struct dso_id *b)* The second is always dso->id, so zeroes if not set, assume passing * NULL for a means a zeroed id */ -if (is_empty_dso_id (a) ||is_empty_dso_id (b)) +if (dso_id__empty (a) ||dso_id__empty (b))return 0 ;return __dso_id__cmp (a, b);
@@ -266,8 +266,8 @@ static struct dso *__dsos__findnew_id(struct dsos *dsos, const char *name, struc {struct dso *dso =__dsos__find_id (dsos, name, id, false); -if (dso &&is_empty_dso_id (&dso->id) && !is_empty_dso_id (id)) -inject_dso_id (dso, id); +if (dso &&dso_id__empty (&dso->id) && !dso_id__empty (id)) +dso__inject_id (dso, id);return dso ? dso :__dsos__addnew_id (dsos, name, id); }
Jiri Olsa
Acked-byQuoted 49 lines · Arnaldo Carvalho de Melo
Em Tue, Mar 24, 2020 at 11:48:43AM +0100, Jiri Olsa escreveu:
Quoted 38 lines · Ravi Bangoria
Quoted 37 lines
Perf gets dso details from two different sources. 1st, from builid headers in perf.data and 2nd from MMAP2 samples. Dso from buildid header does not have
dso_iddetail. And dso from MMAP2 samples does not have buildid information. If detail of the same dso is present at both the places, filename is common.Previously,
__dsos__findnew_link_by_longname_id()used to compare only long or short names, but Commit0e3149f86b99("perf dso: Movedso_idfrom 'struct map' to 'struct dso'") also added adso_idcomparison. Because of that, now perf is creating two different dso objects of the same file, one from buildid header (withdso_idbut without buildid) and second from MMAP2 sample (with buildid but withoutdso_id).This is causing issues with archive, buildid-list etc subcommands. Fix this by comparing
dso_idonly when it's present. And incase dso is present in 'dsos' list withoutdso_id, injectdso_iddetail as well.Before:
$ sudo ./perf buildid-list -H 0000000000000000000000000000000000000000 /usr/bin/ls 0000000000000000000000000000000000000000 /usr/lib64/ld-2.30.so 0000000000000000000000000000000000000000 /usr/lib64/libc-2.30.so$ ./perf archive perf archive: no build-ids foundAfter:
$ ./perf buildid-list -H b6b1291d0cead046ed0fa5734037fa87a579adee /usr/bin/ls 641f0c90cfa15779352f12c0ec3c7a2b2b6f41e8 /usr/lib64/ld-2.30.so 675ace3ca07a0b863df01f461a7b0984c65c8b37 /usr/lib64/libc-2.30.so$ ./perf archive Now please run:Quoted 7 lines
Quoted 5 lines
$ tar xvf perf.data.tar.bz2 -C ~/.debugwherever you need to run 'perf report' on.
Reported-by Naveen N. Rao <[email protected]>looks good, do we need to add the
dso_idcheck tosort__dso_cmp?Jiri:
Quoted 9 lines
Humm, you mean
sort__dso_cmp()->_sort__dso_cmp()should consider thedso_idand not just its name? Humm, when "dso" sort key is used that means just theshort_name(orlong_name, if verbose), if we use the ID for "dso" then we need to somehow show the id in the output, otherwise we'd have multiple lines with the same DSO name, when multiple versions exist... Perhaps we should do a first pass, figure out if there are DSOs with the same name/different IDs and mark them for showing the ID to differentiate them on the output? But this is something that should be dealt with in a separece cset, I think.
true, also Ravi pointed out in the other answer,
it depends on what we want.. compare the name or
different binaries with same name
Quoted 3 lines
With that in mind, can I add your Acked-by for this patch, with my changes described below?
yep,
jirka
Quoted 50 lines
Ravi:
I'm applying it with the changes below, to keep namespacing consistency, ok?
- Arnaldo
tools/perf/util/dsos.c +5
-5@@ -26,7 +26,7 @@ static int __dso_id__cmp(struct dso_id *a, struct dso_id *b)return 0 ; } -static bool is_empty_dso_id (struct dso_id *id) +static bool dso_id__empty (struct dso_id *id) {if (!id)return true;@@ -34,7 +34,7 @@ static bool is_empty_dso_id(struct dso_id *id)return !id->maj && !id->min && !id->ino && !id->ino_generation; } -static void inject_dso_id (struct dso *dso,struct dso_id *id) +static void dso__inject_id (struct dso *dso,struct dso_id *id) { dso->id.maj = id->maj; dso->id.min = id->min;@@ -48,7 +48,7 @@ static int dso_id__cmp(struct dso_id *a, struct dso_id *b)* The second is always dso->id, so zeroes if not set, assume passing * NULL for a means a zeroed id */ -if (is_empty_dso_id (a) ||is_empty_dso_id (b)) +if (dso_id__empty (a) ||dso_id__empty (b))return 0 ;return __dso_id__cmp (a, b);@@ -266,8 +266,8 @@ static struct dso *__dsos__findnew_id(struct dsos *dsos, const char *name, struc {struct dso *dso =__dsos__find_id (dsos, name, id, false); -if (dso &&is_empty_dso_id (&dso->id) && !is_empty_dso_id (id)) -inject_dso_id (dso, id); +if (dso &&dso_id__empty (&dso->id) && !dso_id__empty (id)) +dso__inject_id (dso, id);return dso ? dso :__dsos__addnew_id (dsos, name, id); }
Em Tue, Mar 24, 2020 at 10:00:52AM -0300, Arnaldo Carvalho de Melo escreveu:
Quoted 42 lines
Em Tue, Mar 24, 2020 at 11:48:43AM +0100, Jiri Olsa escreveu:
Quoted 41 lines · Ravi Bangoria
Quoted 40 lines
Perf gets dso details from two different sources. 1st, from builid headers in perf.data and 2nd from MMAP2 samples. Dso from buildid header does not have
dso_iddetail. And dso from MMAP2 samples does not have buildid information. If detail of the same dso is present at both the places, filename is common.Previously,
__dsos__findnew_link_by_longname_id()used to compare only long or short names, but Commit0e3149f86b99("perf dso: Movedso_idfrom 'struct map' to 'struct dso'") also added adso_idcomparison. Because of that, now perf is creating two different dso objects of the same file, one from buildid header (withdso_idbut without buildid) and second from MMAP2 sample (with buildid but withoutdso_id).This is causing issues with archive, buildid-list etc subcommands. Fix this by comparing
dso_idonly when it's present. And incase dso is present in 'dsos' list withoutdso_id, injectdso_iddetail as well.Before:
$ sudo ./perf buildid-list -H 0000000000000000000000000000000000000000 /usr/bin/ls 0000000000000000000000000000000000000000 /usr/lib64/ld-2.30.so 0000000000000000000000000000000000000000 /usr/lib64/libc-2.30.so$ ./perf archive perf archive: no build-ids foundAfter:
$ ./perf buildid-list -H b6b1291d0cead046ed0fa5734037fa87a579adee /usr/bin/ls 641f0c90cfa15779352f12c0ec3c7a2b2b6f41e8 /usr/lib64/ld-2.30.so 675ace3ca07a0b863df01f461a7b0984c65c8b37 /usr/lib64/libc-2.30.so$ ./perf archive Now please run:$ tar xvf perf.data.tar.bz2 -C ~/.debugwherever you need to run 'perf report' on.
This improves the situation, but something else is still amiss:
[root@five ~]# perf buildid-list -H | grep ^000000000000 | while read -a line ; do path=${line[1]} ; nlines=$(perf buildid-list -H -i perf.data | grep $path | tee /tmp/lines | wc -l) ;[ $nlines -eq 2 ] && cat /tmp/lines ; done 641f0c90cfa15779352f12c0ec3c7a2b2b6f41e8 /usr/lib64/ld-2.30.so
0000000000000000000000000000000000000000 /usr/lib64/ld-2.30.sod40c8da7371d8adea464ae2b099590b2c4465574 /usr/lib64/libpthread-2.30.so
0000000000000000000000000000000000000000 /usr/lib64/libpthread-2.30.soaec38b95c2b305c9f1943b2dd988aeb58c17a5d9 /usr/lib64/libm-2.30.so
0000000000000000000000000000000000000000 /usr/lib64/libm-2.30.so675ace3ca07a0b863df01f461a7b0984c65c8b37 /usr/lib64/libc-2.30.so
0000000000000000000000000000000000000000 /usr/lib64/libc-2.30.so8286f22591b0be26730eea306a22a0f30475590b /usr/bin/bash
0000000000000000000000000000000000000000 /usr/bin/bash
[root@five ~]#
I.e. DSOs that have not changed:
[root@five ~]# grep libpthread /proc/*/maps | cut -d' ' -f6- | sort | uniq -c
690 /usr/lib64/libpthread-2.30.so
[root@five ~]#
[root@five ~]# file /usr/lib64/libpthread-2.30.so/usr/lib64/libpthread-2.30.so: ELF 64-bit LSB shared object, x86-64, version 1 (SYSV), dynamically linked, interpreter /lib64/ld-linux-x86-64.so.2, BuildID[sha1]=d40c8da7371d8adea464ae2b099590b2c4465574, for GNU/Linux 3.2.0, not stripped [root@five ~]#
[root@five ~]# perf buildid-list -i /usr/lib64/libpthread-2.30.sod40c8da7371d8adea464ae2b099590b2c4465574
[root@five ~]#
Appear with/without build-id in 'perf buildid-list'.
I'm checking a bit more to see if I figure this out, I'll keep your patch in, maybe this is a diferent issue, will combine if needed.
- Arnaldo
Quoted 39 lines · Ravi Bangoria
Perf gets dso details from two different sources. 1st, from builid headers in perf.data and 2nd from MMAP2 samples. Dso from buildid header does not have
dso_iddetail. And dso from MMAP2 samples does not have buildid information. If detail of the same dso is present at both the places, filename is common.Previously,
__dsos__findnew_link_by_longname_id()used to compare only long or short names, but Commit0e3149f86b99("perf dso: Movedso_idfrom 'struct map' to 'struct dso'") also added adso_idcomparison. Because of that, now perf is creating two different dso objects of the same file, one from buildid header (withdso_idbut without buildid) and second from MMAP2 sample (with buildid but withoutdso_id).This is causing issues with archive, buildid-list etc subcommands. Fix this by comparing
dso_idonly when it's present. And incase dso is present in 'dsos' list withoutdso_id, injectdso_iddetail as well.Before:
$ sudo ./perf buildid-list -H 0000000000000000000000000000000000000000 /usr/bin/ls 0000000000000000000000000000000000000000 /usr/lib64/ld-2.30.so 0000000000000000000000000000000000000000 /usr/lib64/libc-2.30.so$ ./perf archive perf archive: no build-ids foundAfter:
$ ./perf buildid-list -H b6b1291d0cead046ed0fa5734037fa87a579adee /usr/bin/ls 641f0c90cfa15779352f12c0ec3c7a2b2b6f41e8 /usr/lib64/ld-2.30.so 675ace3ca07a0b863df01f461a7b0984c65c8b37 /usr/lib64/libc-2.30.so$ ./perf archive Now please run:$ tar xvf perf.data.tar.bz2 -C ~/.debug
Quoted 8 lines
wherever you need to run 'perf report' on.
Reported-by Naveen N. Rao <[email protected]>Fixes 0e3149f86b99 ("perf dso: Move dso_id from 'struct map' to 'struct dso'")Signed-off-by Ravi Bangoria <[email protected]>--- tools/perf/util/dsos.c | 22 +++++++++++++++++++++- 1 file changed, 21 insertions(+), 1 deletion(-)
Thanks. This fixes the issue I was facing with 'perf archive' not picking up the right binaries. So:
- Naveen
The following commit has been merged into the perf/urgent branch of tip:
The following commit has been merged into the perf/urgent branch of tip:
Commit-ID: 0d33b34352531ff7029c58eda2321340c0ea3f5f Gitweb: https://git.kernel.org/tip/0d33b34352531ff7029c58eda2321340c0ea3f5f Author: Ravi Bangoria <[email protected]> AuthorDate: Tue, 24 Mar 2020 09:54:24 +05:30 Committer: Arnaldo Carvalho de Melo <[email protected]> CommitterDate: Tue, 24 Mar 2020 10:57:38 -03:00
perf dso: Fix dso comparison
Perf gets dso details from two different sources. 1st, from builid headers in perf.data and 2nd from MMAP2 samples. Dso from buildid header does not have dso_id detail. And dso from MMAP2 samples does not have buildid information. If detail of the same dso is present at both the places, filename is common.
Previously, __dsos__findnew_link_by_longname_id() used to compare only long or short names, but Commit 0e3149f86b99 ("perf dso: Move dso_id from 'struct map' to 'struct dso'") also added a dso_id comparison. Because of that, now perf is creating two different dso objects of the same file, one from buildid header (with dso_id but without buildid) and second from MMAP2 sample (with buildid but without dso_id).
This is causing issues with archive, buildid-list etc subcommands. Fix this by comparing dso_id only when it's present. And incase dso is present in 'dsos' list without dso_id, inject dso_id detail as well.
Before:
$ sudo ./perf buildid-list -H 0000000000000000000000000000000000000000 /usr/bin/ls 0000000000000000000000000000000000000000 /usr/lib64/ld-2.30.so 0000000000000000000000000000000000000000 /usr/lib64/libc-2.30.so
$ ./perf archive perf archive: no build-ids found
After:
$ ./perf buildid-list -H b6b1291d0cead046ed0fa5734037fa87a579adee /usr/bin/ls 641f0c90cfa15779352f12c0ec3c7a2b2b6f41e8 /usr/lib64/ld-2.30.so 675ace3ca07a0b863df01f461a7b0984c65c8b37 /usr/lib64/libc-2.30.so
$ ./perf archive Now please run:
$ tar xvf perf.data.tar.bz2 -C ~/.debug
wherever you need to run 'perf report' on.
Committer notes:
Renamed is_empty_dso_id() to dso_id__empty() and inject_dso_id() to dso__inject_id() to keep namespacing consistent.
--- tools/perf/util/dsos.c | 22 +++++++++++++++++++++- 1 file changed, 21 insertions(+), 1 deletion(-)
tools/perf/util/dsos.c +21 -1
@@ -26,13 +26,29 @@ static int __dso_id__cmp(struct dso_id *a, struct dso_id *b)return 0 ; } +static bool dso_id__empty (struct dso_id *id) +{ +if (!id) +return true; + +return !id->maj && !id->min && !id->ino && !id->ino_generation; +} + +static void dso__inject_id (struct dso *dso,struct dso_id *id) +{ + dso->id.maj = id->maj; + dso->id.min = id->min; + dso->id.ino = id->ino; + dso->id.ino_generation = id->ino_generation; +} +static int dso_id__cmp (struct dso_id *a,struct dso_id *b) {/* * The second is always dso->id, so zeroes if not set, assume passing * NULL for a means a zeroed id */ -if (a ==NULL ) +if (dso_id__empty (a) ||dso_id__empty (b))return 0 ;return __dso_id__cmp (a, b);
@@ -249,6 +265,10 @@ struct dso *__dsos__addnew(struct dsos *dsos, const char *name)static struct dso *__dsos__findnew_id (struct dsos *dsos,const char *name,struct dso_id *id) {struct dso *dso =__dsos__find_id (dsos, name, id, false); + +if (dso &&dso_id__empty (&dso->id) && !dso_id__empty (id)) +dso__inject_id (dso, id); +return dso ? dso :__dsos__addnew_id (dsos, name, id); }