perf: Default use of build IDs and improvements
Ian Rogers
reviewerv3 cover perf: Default use of build IDs and improvements
Build ID mmap2 events have been available since Linux v5.12 and avoid certain races. Enable these by default as discussed in:
https://lore.kernel.org/linux-perf-users/CAP-5=fXP7jN_QrGUcd55_QH5J-Y-FCaJ6=NaHVtyx0oyNh8_-Q@mail.gmail.com/
The dso_id is used to indentify a DSO that may change by being overwritten. The inode generation isn't present in /proc/pid/maps and so was already only optionally filled in. With build ID mmap events the other major, minor and inode varialbes aren't filled in. Change the dso_id implementation to make optional values explicit, rather than injecting a dso_id we want to improve it during find operations, add the buildid to the dso_id for sorting and so that matching fails when build IDs vary between DSOs.
Mark the callchain for buildids and not just the sample IP, fixing missing DSOs.
Fix sample__for_each_callchain_node to populate the map even when symbols aren't computed.
Other minor bits of build_id clean up.
v3: Ensure the struct build_id is initialized empty prior to use as
read paths may fail (Namhyung).
v2: Make marking DSOs still the default even with the defaulted build
ID mmap. The command line option still disables this to avoid regressions. Add callchain patches and jitdump fix.
Ian Rogers (9):
perf callchain: Always populate the addr_location map when adding IP perf build-id: Reduce size of "size" variable
perf build-id: Truncate to avoid overflowing the build_id data perf build-id: Change sprintf functions to snprintf
perf build-id: Mark DSO in sample callchains
perf build-id: Ensure struct build_id is empty before use perf dso: Move build_id to dso_id
perf jitdump: Directly mark the jitdump DSO
perf record: Make --buildid-mmap the default
tools/perf/bench/inject-buildid.c | 2 +- tools/perf/builtin-buildid-cache.c | 20 ++-- tools/perf/builtin-buildid-list.c | 6 +- tools/perf/builtin-inject.c | 36 +++--- tools/perf/builtin-record.c | 33 ++++-- tools/perf/builtin-report.c | 11 +- tools/perf/include/perf/perf_dlfilter.h | 2 +- tools/perf/tests/pe-file-parsing.c | 2 +- tools/perf/tests/sdt.c | 2 +- tools/perf/tests/symbols.c | 4 +- tools/perf/util/build-id.c | 63 ++++++---- tools/perf/util/build-id.h | 8 +- tools/perf/util/debuginfo.c | 2 +- tools/perf/util/disasm.c | 2 +- tools/perf/util/dso.c | 111 ++++++++++-------- tools/perf/util/dso.h | 75 ++++++------ tools/perf/util/dsos.c | 20 ++-- tools/perf/util/event.c | 2 +- tools/perf/util/header.c | 2 +- tools/perf/util/jitdump.c | 21 +++- tools/perf/util/machine.c | 34 +++--- tools/perf/util/map.c | 15 ++- tools/perf/util/map.h | 5 +- tools/perf/util/probe-event.c | 3 +- tools/perf/util/probe-file.c | 4 +- tools/perf/util/probe-finder.c | 3 +- .../scripting-engines/trace-event-python.c | 7 +- tools/perf/util/sort.c | 27 +++-- tools/perf/util/symbol-minimal.c | 2 +- tools/perf/util/symbol.c | 7 +- tools/perf/util/symbol_conf.h | 2 +- tools/perf/util/synthetic-events.c | 44 ++++--- tools/perf/util/thread.c | 8 +- tools/perf/util/thread.h | 2 +- 34 files changed, 337 insertions(+), 250 deletions(-)
Signature · 1 line
2.49.0.901.g37484f566f-goog
Ian Rogers
reviewerPATCH v3 1/9 perf callchain: Always populate the addr_location map when adding IP
Dropping symbols also meant the callchain maps wasn't populated, but the callchain map is needed to find the DSO. Plumb the symbols option better, falling back to thread__find_map rather than thread__find_symbol when symbols are disabled.
--- tools/perf/util/machine.c | 6 ++++-- tools/perf/util/thread.c | 8 ++++++-- tools/perf/util/thread.h | 2 +- 3 files changed, 11 insertions(+), 5 deletions(-)
tools/perf/util/machine.c +4 -2
@@ -1976,7 +1976,7 @@ static void ip__resolve_ams(struct thread *thread,* Thus, we have to try consecutively until we find a match * or else, the symbol is unknown */ -thread__find_cpumode_addr_location (thread, ip, &al); +thread__find_cpumode_addr_location (thread, ip,/*symbols=*/ true, &al); ams->addr = ip; ams->al_addr = al.addr;
@@ -2078,7 +2078,7 @@ static int add_callchain_ip(struct thread *thread, al.sym =NULL ; al.srcline =NULL ;if (!cpumode) { -thread__find_cpumode_addr_location (thread, ip, &al); +thread__find_cpumode_addr_location (thread, ip, symbols, &al); }else {if (ip >=PERF_CONTEXT_MAX ) {switch (ip) {
@@ -2106,6 +2106,8 @@ static int add_callchain_ip(struct thread *thread, }if (symbols)thread__find_symbol (thread, *cpumode, ip, &al); +else +thread__find_map (thread, *cpumode, ip, &al); }if (al.sym !=NULL ) {
tools/perf/util/thread.c +6 -2
@@ -410,7 +410,7 @@ int thread__fork(struct thread *thread, struct thread *parent, u64 timestamp, bo }void thread__find_cpumode_addr_location (struct thread *thread,u64 addr, -struct addr_location *al) +bool symbols,struct addr_location *al) {size_t i;const u8 cpumodes[] = {
@@ -421,7 +421,11 @@ void thread__find_cpumode_addr_location(struct thread *thread, u64 addr, };for (i =0 ; i <ARRAY_SIZE (cpumodes); i++) { -thread__find_symbol (thread, cpumodes[i], addr, al); +if (symbols) +thread__find_symbol (thread, cpumodes[i], addr, al); +else +thread__find_map (thread, cpumodes[i], addr, al); +if (al->map)break ; }
tools/perf/util/thread.h +1 -1
@@ -126,7 +126,7 @@ struct symbol *thread__find_symbol_fb(struct thread *thread, u8 cpumode, u64 addr, struct addr_location *al);void thread__find_cpumode_addr_location (struct thread *thread,u64 addr, -struct addr_location *al); +bool symbols,struct addr_location *al);int thread__memcpy (struct thread *thread,struct machine *machine,void *buf,u64 ip,int len,bool *is64bit);
Signature · 1 line
2.49.0.901.g37484f566f-goog
Ian Rogers
reviewerPATCH v3 2/9 perf build-id: Reduce size of "size" variable
In reply to Ian RogersLater clean up of the dso_id to include a build_id will suffer from alignment and size issues. The size can only hold up to a value of BUILD_ID_SIZE (20) and the mmap2 event uses a byte for the value.
--- tools/perf/util/build-id.h | 2 +- tools/perf/util/synthetic-events.c | 8 ++++++-- 2 files changed, 7 insertions(+), 3 deletions(-)
tools/perf/util/build-id.h +1 -1
@@ -13,7 +13,7 @@struct build_id {u8 data[BUILD_ID_SIZE ]; -size_t size; +u8 size; };struct dso ;
tools/perf/util/synthetic-events.c +6 -2
@@ -2248,7 +2248,9 @@ int perf_event__synthesize_build_id(const struct perf_tool *tool,memset (&ev,0 , len); - ev.build_id.size =min (bid->size,sizeof (ev.build_id.build_id)); + ev.build_id.size = bid->size; +if (ev.build_id.size >sizeof (ev.build_id.build_id)) + ev.build_id.size =sizeof (ev.build_id.build_id);memcpy (ev.build_id.build_id, bid->data, ev.build_id.size); ev.build_id.header.type =PERF_RECORD_HEADER_BUILD_ID ; ev.build_id.header.misc = misc |PERF_RECORD_MISC_BUILD_ID_SIZE ;
@@ -2308,7 +2310,9 @@ int perf_event__synthesize_mmap2_build_id(const struct perf_tool *tool, ev.mmap2.len = len; ev.mmap2.pgoff = pgoff; - ev.mmap2.build_id_size =min (bid->size,sizeof (ev.mmap2.build_id)); + ev.mmap2.build_id_size = bid->size; +if (ev.mmap2.build_id_size >sizeof (ev.mmap2.build_id)) + ev.build_id.size =sizeof (ev.mmap2.build_id);memcpy (ev.mmap2.build_id, bid->data, ev.mmap2.build_id_size); ev.mmap2.prot = prot;
Signature · 1 line
2.49.0.901.g37484f566f-goog
Ian Rogers
reviewerPATCH v3 3/9 perf build-id: Truncate to avoid overflowing the build_id data
In reply to Ian RogersWarning when the build_id data would be overflowed would lead to memory corruption, switch to truncation.
--- tools/perf/util/build-id.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)
tools/perf/util/build-id.c +4 -1
@@ -951,7 +951,10 @@ bool perf_session__read_build_ids(struct perf_session *session, bool with_hits)void build_id__init (struct build_id *bid,const u8 *data,size_t size) { -WARN_ON (size >BUILD_ID_SIZE ); +if (size >BUILD_ID_SIZE ) { +pr_debug ("Truncating build_id size from %zd\n" , size); + size =BUILD_ID_SIZE ; + }memcpy (bid->data, data, size); bid->size = size; }
Signature · 1 line
2.49.0.901.g37484f566f-goog
Ian Rogers
reviewerPATCH v3 4/9 perf build-id: Change sprintf functions to snprintf
In reply to Ian RogersPass in a size argument rather than implying all build id strings must be SBUILD_ID_SIZE.
--- tools/perf/builtin-buildid-cache.c | 12 +++---- tools/perf/builtin-buildid-list.c | 6 ++-- tools/perf/util/build-id.c | 33 ++++++++----------- tools/perf/util/build-id.h | 6 ++-- tools/perf/util/disasm.c | 2 +- tools/perf/util/dso.c | 4 +-- tools/perf/util/dsos.c | 2 +- tools/perf/util/event.c | 2 +- tools/perf/util/header.c | 2 +- tools/perf/util/map.c | 2 +- tools/perf/util/probe-file.c | 4 +-- .../scripting-engines/trace-event-python.c | 7 ++-- tools/perf/util/symbol.c | 2 +- 13 files changed, 38 insertions(+), 46 deletions(-)
tools/perf/builtin-buildid-cache.c +6 -6
@@ -31,7 +31,7 @@#include <linux/string.h> #include <linux/err.h> -static int build_id_cache__kcore_buildid (const char *proc_dir,char *sbuildid) +static int build_id_cache__kcore_buildid (const char *proc_dir,char *sbuildid,size_t sbuildid_size) {char root_dir[PATH_MAX ];char *p;
@@ -42,7 +42,7 @@ static int build_id_cache__kcore_buildid(const char *proc_dir, char *sbuildid)if (!p)return -1 ; *p ='\0' ; -return sysfs__sprintf_build_id (root_dir, sbuildid); +return sysfs__snprintf_build_id (root_dir, sbuildid, sbuildid_size); }static int build_id_cache__kcore_dir (char *dir,size_t sz)
@@ -128,7 +128,7 @@ static int build_id_cache__add_kcore(const char *filename, bool force)return -1 ; *p ='\0' ; -if (build_id_cache__kcore_buildid (from_dir, sbuildid) <0 ) +if (build_id_cache__kcore_buildid (from_dir, sbuildid,sizeof (sbuildid)) <0 )return -1 ; scnprintf(to_dir, sizeof(to_dir), "%s/%s/%s",
@@ -187,7 +187,7 @@ static int build_id_cache__add_file(const char *filename, struct nsinfo *nsi)return -1 ; } -build_id__sprintf (&bid, sbuild_id); +build_id__snprintf (&bid, sbuild_id,sizeof (sbuild_id)); err =build_id_cache__add_s (sbuild_id, filename, nsi, false, false); pr_debug("Adding %s %s: %s\n", sbuild_id, filename,
@@ -211,7 +211,7 @@ static int build_id_cache__remove_file(const char *filename, struct nsinfo *nsi)return -1 ; } -build_id__sprintf (&bid, sbuild_id); +build_id__snprintf (&bid, sbuild_id,sizeof (sbuild_id)); err =build_id_cache__remove_s (sbuild_id);pr_debug ("Removing %s %s: %s\n" , sbuild_id, filename, err ?"FAIL" :"Ok" );
@@ -317,7 +317,7 @@ static int build_id_cache__update_file(const char *filename, struct nsinfo *nsi) } err =0 ; -build_id__sprintf (&bid, sbuild_id); +build_id__snprintf (&bid, sbuild_id,sizeof (sbuild_id));if (build_id_cache__cached (sbuild_id)) err =build_id_cache__remove_s (sbuild_id);
tools/perf/builtin-buildid-list.c +3 -3
@@ -31,7 +31,7 @@ static int buildid__map_cb(struct map *map, void *arg __maybe_unused)memset (bid_buf,0 ,sizeof (bid_buf));if (dso__has_build_id (dso)) -build_id__sprintf (dso__bid_const (dso), bid_buf); +build_id__snprintf (dso__bid_const (dso), bid_buf,sizeof (bid_buf));printf ("%s %16" PRIx64" %16" PRIx64, bid_buf,map__start (map),map__end (map));if (dso_long_name !=NULL )printf (" %s" , dso_long_name);
@@ -57,7 +57,7 @@ static int sysfs__fprintf_build_id(FILE *fp)char sbuild_id[SBUILD_ID_SIZE ];int ret; - ret =sysfs__sprintf_build_id ("/" , sbuild_id); + ret =sysfs__snprintf_build_id ("/" , sbuild_id,sizeof (sbuild_id));if (ret !=sizeof (sbuild_id))return ret <0 ? ret : -EINVAL ;
@@ -69,7 +69,7 @@ static int filename__fprintf_build_id(const char *name, FILE *fp)char sbuild_id[SBUILD_ID_SIZE ];int ret; - ret =filename__sprintf_build_id (name, sbuild_id); + ret =filename__snprintf_build_id (name, sbuild_id,sizeof (sbuild_id));if (ret !=sizeof (sbuild_id))return ret <0 ? ret : -EINVAL ;
tools/perf/util/build-id.c +13 -20
@@ -67,24 +67,17 @@ int build_id__mark_dso_hit(const struct perf_tool *tool __maybe_unused,return 0 ; } -int build_id__sprintf (const struct build_id *build_id,char *bf) +int build_id__snprintf (const struct build_id *build_id,char *bf,size_t bf_size) { -char *bid = bf; -const u8 *raw = build_id->data; -size_t i; - - bf[0 ] =0x0 ; +size_t offs =0 ; -for (i =0 ; i < build_id->size; ++i) { -sprintf (bid,"%02x" , *raw); - ++raw; - bid +=2 ; - } +for (size_t i =0 ; i < build_id->size && offs < bf_size; ++i) + offs +=snprintf (bf + offs, bf_size - offs,"%02x" , build_id->data[i]); -return (bid - bf) +1 ; +return offs; } -int sysfs__sprintf_build_id (const char *root_dir,char *sbuild_id) +int sysfs__snprintf_build_id (const char *root_dir,char *sbuild_id,size_t sbuild_id_size) {char notes[PATH_MAX ];struct build_id bid;
@@ -99,10 +92,10 @@ int sysfs__sprintf_build_id(const char *root_dir, char *sbuild_id)if (ret <0 )return ret; -return build_id__sprintf (&bid, sbuild_id); +return build_id__snprintf (&bid, sbuild_id, sbuild_id_size); } -int filename__sprintf_build_id (const char *pathname,char *sbuild_id) +int filename__snprintf_build_id (const char *pathname,char *sbuild_id,size_t sbuild_id_size) {struct build_id bid;int ret;
@@ -111,7 +104,7 @@ int filename__sprintf_build_id(const char *pathname, char *sbuild_id)if (ret <0 )return ret; -return build_id__sprintf (&bid, sbuild_id); +return build_id__snprintf (&bid, sbuild_id, sbuild_id_size); }/* asnprintf consolidates asprintf and snprintf */
@@ -212,9 +205,9 @@ static bool build_id_cache__valid_id(char *sbuild_id)return false;if (!strcmp (pathname,DSO__NAME_KALLSYMS )) - ret =sysfs__sprintf_build_id ("/" , real_sbuild_id); + ret =sysfs__snprintf_build_id ("/" , real_sbuild_id,sizeof (real_sbuild_id));else if (pathname[0 ] =='/' ) - ret =filename__sprintf_build_id (pathname, real_sbuild_id); + ret =filename__snprintf_build_id (pathname, real_sbuild_id,sizeof (real_sbuild_id));else ret = -EINVAL ;/* Should we support other special DSO cache? */ if (ret >=0 )
@@ -243,7 +236,7 @@ char *__dso__build_id_filename(const struct dso *dso, char *bf, size_t size,if (!dso__has_build_id (dso))return NULL ; -build_id__sprintf (dso__bid_const (dso), sbuild_id); +build_id__snprintf (dso__bid_const (dso), sbuild_id,sizeof (sbuild_id)); linkname =build_id_cache__linkname (sbuild_id,NULL ,0 );if (!linkname)return NULL ;
@@ -769,7 +762,7 @@ static int build_id_cache__add_b(const struct build_id *bid, {char sbuild_id[SBUILD_ID_SIZE ]; -build_id__sprintf (bid, sbuild_id); +build_id__snprintf (bid, sbuild_id,sizeof (sbuild_id));return __build_id_cache__add_s (sbuild_id, name, nsi, is_kallsyms, is_vdso, proper_name, root_dir);
tools/perf/util/build-id.h +3 -3
@@ -21,10 +21,10 @@ struct feat_fd;struct nsinfo ;void build_id__init (struct build_id *bid,const u8 *data,size_t size); -int build_id__sprintf (const struct build_id *build_id,char *bf); +int build_id__snprintf (const struct build_id *build_id,char *bf,size_t bf_size);bool build_id__is_defined (const struct build_id *bid); -int sysfs__sprintf_build_id (const char *root_dir,char *sbuild_id); -int filename__sprintf_build_id (const char *pathname,char *sbuild_id); +int sysfs__snprintf_build_id (const char *root_dir,char *sbuild_id,size_t sbuild_id_size); +int filename__snprintf_build_id (const char *pathname,char *sbuild_id,size_t sbuild_id_size);char *build_id_cache__kallsyms_path (const char *sbuild_id,char *bf,size_t size);
tools/perf/util/disasm.c +1 -1
@@ -1218,7 +1218,7 @@ int symbol__strerror_disassemble(struct map_symbol *ms, int errnum, char *buf, schar *build_id_msg =NULL ;if (dso__has_build_id (dso)) { -build_id__sprintf (dso__bid (dso), bf +15 ); +build_id__snprintf (dso__bid (dso), bf +15 ,sizeof (bf) -15 ); build_id_msg = bf; } scnprintf(buf, buflen,
tools/perf/util/dso.c +2 -2
@@ -217,7 +217,7 @@ int dso__read_binary_type_filename(const struct dso *dso,break ; } -build_id__sprintf (dso__bid_const (dso), build_id_hex); +build_id__snprintf (dso__bid_const (dso), build_id_hex,sizeof (build_id_hex)); len =__symbol__join_symfs (filename, size,"/usr/lib/debug/.build-id/" );snprintf (filename + len, size - len,"%.2s/%s.debug" , build_id_hex, build_id_hex +2 );
@@ -1697,7 +1697,7 @@ static size_t dso__fprintf_buildid(struct dso *dso, FILE *fp) {char sbuild_id[SBUILD_ID_SIZE ]; -build_id__sprintf (dso__bid (dso), sbuild_id); +build_id__snprintf (dso__bid (dso), sbuild_id,sizeof (sbuild_id));return fprintf (fp,"%s" , sbuild_id); }
tools/perf/util/dsos.c +1 -1
@@ -370,7 +370,7 @@ static int dsos__fprintf_buildid_cb(struct dso *dso, void *data)if (args->skip && args->skip (dso, args->parm))return 0 ; -build_id__sprintf (dso__bid (dso), sbuild_id); +build_id__snprintf (dso__bid (dso), sbuild_id,sizeof (sbuild_id)); args->ret +=fprintf (args->fp,"%-40s %s\n" , sbuild_id,dso__long_name (dso));return 0 ; }
tools/perf/util/event.c +1 -1
@@ -329,7 +329,7 @@ size_t perf_event__fprintf_mmap2(union perf_event *event, FILE *fp)build_id__init (&bid, event->mmap2.build_id, event->mmap2.build_id_size); -build_id__sprintf (&bid, sbuild_id); +build_id__snprintf (&bid, sbuild_id,sizeof (sbuild_id)); return fprintf(fp, " %d/%d: [%#" PRI_lx64 "(%#" PRI_lx64 ") @ %#" PRI_lx64 " <%s>]: %c%c%c%c %s\n",
tools/perf/util/header.c +1 -1
@@ -2291,7 +2291,7 @@ static int __event_process_build_id(struct perf_record_header_build_id *bev,free (m.name); } -build_id__sprintf (dso__bid (dso), sbuild_id); +build_id__snprintf (dso__bid (dso), sbuild_id,sizeof (sbuild_id));pr_debug ("build id event received for %s: %s [%zu]\n" ,dso__long_name (dso), sbuild_id, size);dso__put (dso);
tools/perf/util/map.c +1 -1
@@ -354,7 +354,7 @@ int map__load(struct map *map)if (dso__has_build_id (dso)) {char sbuild_id[SBUILD_ID_SIZE ]; -build_id__sprintf (dso__bid (dso), sbuild_id); +build_id__snprintf (dso__bid (dso), sbuild_id,sizeof (sbuild_id));pr_debug ("%s with build id %s not found" , name, sbuild_id); }else pr_debug ("Failed to open %s" , name);
tools/perf/util/probe-file.c +2 -2
@@ -448,10 +448,10 @@ static int probe_cache__open(struct probe_cache *pcache, const char *target,if (!target || !strcmp (target,DSO__NAME_KALLSYMS )) { target =DSO__NAME_KALLSYMS ; is_kallsyms = true; - ret =sysfs__sprintf_build_id ("/" , sbuildid); + ret =sysfs__snprintf_build_id ("/" , sbuildid,sizeof (sbuildid)); }else {nsinfo__mountns_enter (nsi, &nsc); - ret =filename__sprintf_build_id (target, sbuildid); + ret =filename__snprintf_build_id (target, sbuildid,sizeof (sbuildid));nsinfo__mountns_exit (&nsc); }
tools/perf/util/scripting-engines/trace-event-python.c +3 -4
@@ -780,14 +780,13 @@ static void set_sym_in_dict(PyObject *dict, struct addr_location *al, const char *sym_field, const char *symoff_field, const char *map_pgoff) { - char sbuild_id[SBUILD_ID_SIZE]; - if (al->map) { + char sbuild_id[SBUILD_ID_SIZE];struct dso *dso =map__dso (al->map);pydict_set_item_string_decref (dict, dso_field,_PyUnicode_FromString (dso__name (dso))); -build_id__sprintf (dso__bid (dso), sbuild_id); +build_id__snprintf (dso__bid (dso), sbuild_id,sizeof (sbuild_id));pydict_set_item_string_decref (dict, dso_bid_field,_PyUnicode_FromString (sbuild_id)); pydict_set_item_string_decref(dict, dso_map_start,
@@ -1238,7 +1237,7 @@ static int python_export_dso(struct db_export *dbe, struct dso *dso,char sbuild_id[SBUILD_ID_SIZE ];PyObject *t; -build_id__sprintf (dso__bid (dso), sbuild_id); +build_id__snprintf (dso__bid (dso), sbuild_id,sizeof (sbuild_id)); t =tuple_new (5 );
tools/perf/util/symbol.c +1 -1
@@ -2143,7 +2143,7 @@ static char *dso__find_kallsyms(struct dso *dso, struct map *map) goto proc_kallsyms; } -build_id__sprintf (dso__bid (dso), sbuild_id); +build_id__snprintf (dso__bid (dso), sbuild_id,sizeof (sbuild_id));/* Find kallsyms in build-id cache with kcore */ scnprintf(path, sizeof(path), "%s/%s/%s",
Signature · 1 line
2.49.0.901.g37484f566f-goog
Ian Rogers
reviewerPATCH v3 5/9 perf build-id: Mark DSO in sample callchains
In reply to Ian RogersPreviously only the sample IP's map DSO would be marked hit for the purposes of populating the build ID cache. Walk the call chain to mark all IPs and DSOs.
--- tools/perf/util/build-id.c | 17 ++++++++++++++++- 1 file changed, 16 insertions(+), 1 deletion(-)
tools/perf/util/build-id.c +16 -1
@@ -42,10 +42,20 @@static bool no_buildid_cache; +static int mark_dso_hit_callback (struct callchain_cursor_node *node,void *data__maybe_unused ) +{ +struct map *map = node->ms.map; + +if (map) +dso__set_hit (map__dso (map)); + +return 0 ; +} +int build_id__mark_dso_hit (const struct perf_tool *tool__maybe_unused ,union perf_event *event,struct perf_sample *sample, -struct evsel *evsel__maybe_unused , +struct evsel *evsel,struct machine *machine) {struct addr_location al;
@@ -63,6 +73,11 @@ int build_id__mark_dso_hit(const struct perf_tool *tool __maybe_unused,dso__set_hit (map__dso (al.map));addr_location__exit (&al); + +sample__for_each_callchain_node (thread, evsel, sample,PERF_MAX_STACK_DEPTH , +/*symbols=*/ false, mark_dso_hit_callback,/*data=*/ NULL ); + +thread__put (thread);return 0 ; }
Signature · 1 line
2.49.0.901.g37484f566f-goog
Namhyung Kim
maintainerRe: [PATCH v3 5/9] perf build-id: Mark DSO in sample callchains
Quoted 3 lines · Ian Rogers
Previously only the sample IP's map DSO would be marked hit for the purposes of populating the build ID cache. Walk the call chain to mark all IPs and DSOs.
I think this is correct, but I'm afraid it'd also increase the processing time. Do you happen to have any numbers?
Thanks,
Namhyung
Quoted 47 lines
Signed-off-by Ian Rogers <[email protected]>--- tools/perf/util/build-id.c | 17 ++++++++++++++++- 1 file changed, 16 insertions(+), 1 deletion(-)tools/perf/util/build-id.c +16
-1@@ -42,10 +42,20 @@static bool no_buildid_cache; +static int mark_dso_hit_callback (struct callchain_cursor_node *node,void *data__maybe_unused ) +{ +struct map *map = node->ms.map; + +if (map) +dso__set_hit (map__dso (map)); + +return 0 ; +} +int build_id__mark_dso_hit (const struct perf_tool *tool__maybe_unused ,union perf_event *event,struct perf_sample *sample, -struct evsel *evsel__maybe_unused , +struct evsel *evsel,struct machine *machine) {struct addr_location al;@@ -63,6 +73,11 @@ int build_id__mark_dso_hit(const struct perf_tool *tool __maybe_unused,dso__set_hit (map__dso (al.map));addr_location__exit (&al); + +sample__for_each_callchain_node (thread, evsel, sample,PERF_MAX_STACK_DEPTH , +/*symbols=*/ false, mark_dso_hit_callback,/*data=*/ NULL ); + +thread__put (thread);return 0 ; }Signature · 1 line
2.49.0.901.g37484f566f-goog
Ian Rogers
reviewerRe: [PATCH v3 5/9] perf build-id: Mark DSO in sample callchains
Quoted 8 lines · Namhyung Kim
Quoted 3 lines · Ian Rogers
Previously only the sample IP's map DSO would be marked hit for the purposes of populating the build ID cache. Walk the call chain to mark all IPs and DSOs.
I think this is correct, but I'm afraid it'd also increase the processing time. Do you happen to have any numbers?
It increases time spent processing the data file but to get a large data file I had to run for multiple seconds and I struggled to get the performance cost of this to be in the milliseconds (ie a tiny fraction of the record time). Ultimately I found the change imperceptible and couldn't think of a good command line to make it perceptible.
If the time is spent populating ~/.debug because more DSOs are marked then this is fixing a bug and isn't a problem with the patch.
My personal opinion is that it is somewhat surprising perf record is post-processing the perf.data file at all, and -B and -N would be my expected defaults - just as --buildid-mmap implies --no-buildid (-B). I didn't want to modify existing behaviors in these changes, however, in this case I was just trying to make the existing behavior correct, similar to fixing the same bug in perf inject.
Thanks,
Ian
Quoted 50 lines
Thanks,
NamhyungQuoted 47 lines
Signed-off-by Ian Rogers <[email protected]>--- tools/perf/util/build-id.c | 17 ++++++++++++++++- 1 file changed, 16 insertions(+), 1 deletion(-)tools/perf/util/build-id.c +16
-2@@ -42,10 +42,20 @@static bool no_buildid_cache; +static int mark_dso_hit_callback (struct callchain_cursor_node *node,void *data__maybe_unused ) +{ +struct map *map = node->ms.map; + +if (map) +dso__set_hit (map__dso (map)); + +return 0 ; +} +int build_id__mark_dso_hit (const struct perf_tool *tool__maybe_unused ,union perf_event *event,struct perf_sample *sample, -struct evsel *evsel__maybe_unused , +struct evsel *evsel,struct machine *machine) {struct addr_location al;@@ -63,6 +73,11 @@ int build_id__mark_dso_hit(const struct perf_tool *tool __maybe_unused,dso__set_hit (map__dso (al.map));addr_location__exit (&al); + +sample__for_each_callchain_node (thread, evsel, sample,PERF_MAX_STACK_DEPTH , +/*symbols=*/ false, mark_dso_hit_callback,/*data=*/ NULL ); + +thread__put (thread);return 0 ; } --2.49.0.901.g37484f566f-goog
Namhyung Kim
maintainerRe: [PATCH v3 5/9] perf build-id: Mark DSO in sample callchains
Quoted 7 lines · Ian Rogers
Quoted 6 lines · Namhyung Kim
Quoted 3 lines · Ian Rogers
Previously only the sample IP's map DSO would be marked hit for the purposes of populating the build ID cache. Walk the call chain to mark all IPs and DSOs.
Quoted 8 lines
Quoted 2 lines
I think this is correct, but I'm afraid it'd also increase the processing time. Do you happen to have any numbers?
It increases time spent processing the data file but to get a large data file I had to run for multiple seconds and I struggled to get the performance cost of this to be in the milliseconds (ie a tiny fraction of the record time). Ultimately I found the change imperceptible and couldn't think of a good command line to make it perceptible.
The worst case would be dwarf unwinding. Maybe we can skip the processing if it takes too long..
Quoted 3 lines
If the time is spent populating ~/.debug because more DSOs are marked then this is fixing a bug and isn't a problem with the patch.
Right, it's a good thing.
Quoted 4 lines
My personal opinion is that it is somewhat surprising
perf recordis post-processing the perf.data file at all, and -B and -N would be my expected defaults - just as --buildid-mmap implies --no-buildid (-B).
Otherwise nobody will run perf buildid-cache to add the info. :)
Quoted 3 lines
I didn't want to modify existing behaviors in these changes, however, in this case I was just trying to make the existing behavior correct, similar to fixing the same bug in
perf inject.
Thanks for your work,
Namhyung
Ian Rogers
reviewerRe: [PATCH v3 5/9] perf build-id: Mark DSO in sample callchains
Quoted 12 lines · Namhyung Kim
Quoted 10 lines · Ian Rogers
Quoted 8 lines · Namhyung Kim
Quoted 3 lines · Ian Rogers
Previously only the sample IP's map DSO would be marked hit for the purposes of populating the build ID cache. Walk the call chain to mark all IPs and DSOs.
I think this is correct, but I'm afraid it'd also increase the processing time. Do you happen to have any numbers?
Quoted 8 lines
Quoted 5 lines
It increases time spent processing the data file but to get a large data file I had to run for multiple seconds and I struggled to get the performance cost of this to be in the milliseconds (ie a tiny fraction of the record time). Ultimately I found the change imperceptible and couldn't think of a good command line to make it perceptible.
The worst case would be dwarf unwinding. Maybe we can skip the processing if it takes too long..
This doesn't sound unreasonable but is somewhat beyond the scope of what I wanted to do here, which relates to migrating from inodes to buildids as identifiers for DSOs. It would be useful to get a bug report on this being too slow.
Quoted 4 lines
Quoted 3 lines
If the time is spent populating ~/.debug because more DSOs are marked then this is fixing a bug and isn't a problem with the patch.
Quoted 8 lines
Right, it's a good thing.
Quoted 4 lines
My personal opinion is that it is somewhat surprising
perf recordis post-processing the perf.data file at all, and -B and -N would be my expected defaults - just as --buildid-mmap implies --no-buildid (-B).Otherwise nobody will run perf buildid-cache to add the info. :)
Right, but we know it is high overhead as we run a number of tests with -B:
https://web.git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/tests/shell/record_sideband.sh?h=perf-tools-next#n25
https://web.git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/tests/shell/test_arm_spe.sh?h=perf-tools-next#n94
I agree that populating the buildid cache immediately after perf record minimizes a gap where DSOs may change prior to perf report. If the DSO changes midway through perf record it doesn't help as the buildid information is only computed at the end of perf record.
Isn't there an argument that because we have build IDs we don't care about the record to report race any more as debuginfod can find the prior version by means of the build ID? What I mean:
Current perf default way:
1) Run perf record with mmap2 inode rather the buildid data
1.1) To avoid any DSOs in the perf.data from not being available populate the buildid header in the perf.data and the buildid cache at the end of perf record
2) Replace some DSO that's got a sample in the perf.data with a different DSO
3) Run perf report
3.1) The DSO's buildid is known via the header and the buildid cache already contains the DSO
Current way with -N:
1) Run perf record with mmap2 inode rather the buildid data
1.1) To ensure DSOs have buildid data populate the buildid header in the perf.data at the end of perf record
2) Replace some DSO that's got a sample in the perf.data with a different DSO
3) Run perf report
3.1) The DSO's buildid is known via the header, debuginfod can populate the buildid cache as needed
With --buildid-mmap:
1) Run perf record with mmap2 buildid data
1.1) No need to post process file to gather buildids
2) Replace some DSO that's got a sample in the perf.data with a different DSO
3) Run perf report
3.1) The DSO's buildid is known via the mmap2 events, debuginfod can populate the buildid cache as needed
With the current way and the current way with -N there is a race with the DSO changing midway through perf record. The buildid mmap closes the race.
With -N and --buildid-mmap the buildid cache is populated based on use by a tool trying to read the DSO, rather than ahead of time at the end of perf record.
Is the risk of the race that much of an issue? I'm not sure, that's why I'd say default to using -B (if we didn't switch to buildid mmaps by default, but this series does that). You could opt into to covering the race by adding a flag so the data is processed at the end of perf record. You could use perf inject to add the build IDs.
As you say there's the cost at the end of perf record and I'm not sure it is worth it, which is why I'd expect the default to be to opt into having the cost. With --buildid-mmap I'm not seeing a race to cover but this series populates the buildid cache as I know you've argued that perf record should do this by default.
Thanks,
Ian
Quoted 7 lines
Quoted 3 lines
I didn't want to modify existing behaviors in these changes, however, in this case I was just trying to make the existing behavior correct, similar to fixing the same bug in
perf inject.Thanks for your work,
Namhyung
Namhyung Kim
maintainerRe: [PATCH v3 5/9] perf build-id: Mark DSO in sample callchains
Quoted 93 lines · Ian Rogers
Quoted 12 lines · Namhyung Kim
Quoted 10 lines · Ian Rogers
Quoted 8 lines · Namhyung Kim
Quoted 3 lines · Ian Rogers
Previously only the sample IP's map DSO would be marked hit for the purposes of populating the build ID cache. Walk the call chain to mark all IPs and DSOs.
I think this is correct, but I'm afraid it'd also increase the processing time. Do you happen to have any numbers?
Quoted 8 lines
Quoted 5 lines
It increases time spent processing the data file but to get a large data file I had to run for multiple seconds and I struggled to get the performance cost of this to be in the milliseconds (ie a tiny fraction of the record time). Ultimately I found the change imperceptible and couldn't think of a good command line to make it perceptible.
The worst case would be dwarf unwinding. Maybe we can skip the processing if it takes too long..
This doesn't sound unreasonable but is somewhat beyond the scope of what I wanted to do here, which relates to migrating from inodes to buildids as identifiers for DSOs. It would be useful to get a bug report on this being too slow.
Quoted 4 lines
Quoted 3 lines
If the time is spent populating ~/.debug because more DSOs are marked then this is fixing a bug and isn't a problem with the patch.
Quoted 8 lines
Right, it's a good thing.
Quoted 4 lines
My personal opinion is that it is somewhat surprising
perf recordis post-processing the perf.data file at all, and -B and -N would be my expected defaults - just as --buildid-mmap implies --no-buildid (-B).Otherwise nobody will run perf buildid-cache to add the info. :)
Right, but we know it is high overhead as we run a number of tests with -B:
https://web.git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/tests/shell/record_sideband.sh?h=perf-tools-next#n25
https://web.git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/tests/shell/test_arm_spe.sh?h=perf-tools-next#n94I agree that populating the buildid cache immediately after perf record minimizes a gap where DSOs may change prior to perf report. If the DSO changes midway through perf record it doesn't help as the buildid information is only computed at the end of perf record.
Isn't there an argument that because we have build IDs we don't care about the record to report race any more as debuginfod can find the prior version by means of the build ID? What I mean:
Current perf default way:
1) Run perf record with mmap2 inode rather the buildid data
1.1) To avoid any DSOs in the perf.data from not being available populate the buildid header in the perf.data and the buildid cache at the end of perf record
2) Replace some DSO that's got a sample in the perf.data with a different DSO
3) Run perf report
3.1) The DSO's buildid is known via the header and the buildid cache already contains the DSOCurrent way with -N:
1) Run perf record with mmap2 inode rather the buildid data
1.1) To ensure DSOs have buildid data populate the buildid header in the perf.data at the end of perf record
2) Replace some DSO that's got a sample in the perf.data with a different DSO
3) Run perf report
3.1) The DSO's buildid is known via the header, debuginfod can populate the buildid cache as neededWith --buildid-mmap:
1) Run perf record with mmap2 buildid data
1.1) No need to post process file to gather buildids
2) Replace some DSO that's got a sample in the perf.data with a different DSO
3) Run perf report
3.1) The DSO's buildid is known via the mmap2 events, debuginfod can populate the buildid cache as neededWith the current way and the current way with -N there is a race with the DSO changing midway through perf record. The buildid mmap closes the race.
With -N and --buildid-mmap the buildid cache is populated based on use by a tool trying to read the DSO, rather than ahead of time at the end of perf record.Is the risk of the race that much of an issue? I'm not sure, that's why I'd say default to using -B (if we didn't switch to buildid mmaps by default, but this series does that). You could opt into to covering the race by adding a flag so the data is processed at the end of perf record. You could use perf inject to add the build IDs.
Quoted 5 lines
As you say there's the cost at the end of perf record and I'm not sure it is worth it, which is why I'd expect the default to be to opt into having the cost. With --buildid-mmap I'm not seeing a race to cover but this series populates the buildid cache as I know you've argued that perf record should do this by default.
Yeah, I just wanted to keep the original behavior to save the used binaries in the build-ID cache. But we can use debuginfod if it's available as you said.
I think it's a separate topic whether we prefer the build-ID cache or debuginfod. We should be able to change the default behavior with a config option later. So let's keep the caching for now.
Thanks,
Namhyung
Ian Rogers
reviewerPATCH v3 6/9 perf build-id: Ensure struct build_id is empty before use
In reply to Ian RogersIf a build ID is read then not all code paths may ensure it is empty before use. Initialize the build_id to be zero-ed unless there is clear initialization such as a call to build_id__init.
--- tools/perf/bench/inject-buildid.c | 2 +- tools/perf/builtin-buildid-cache.c | 8 ++++---- tools/perf/tests/pe-file-parsing.c | 2 +- tools/perf/tests/sdt.c | 2 +- tools/perf/util/build-id.c | 6 +++--- tools/perf/util/debuginfo.c | 2 +- tools/perf/util/probe-event.c | 3 ++- tools/perf/util/probe-finder.c | 3 ++- tools/perf/util/symbol-minimal.c | 2 +- tools/perf/util/symbol.c | 5 +++-- tools/perf/util/synthetic-events.c | 2 +- 11 files changed, 20 insertions(+), 17 deletions(-)
tools/perf/bench/inject-buildid.c +1 -1
@@ -80,7 +80,7 @@ static int add_dso(const char *fpath, const struct stat *sb __maybe_unused, int typeflag, struct FTW *ftwbuf __maybe_unused) { struct bench_dso *dso = &dsos[nr_dsos]; -struct build_id bid; +struct build_id bid = { .size =0 , };if (typeflag ==FTW_D || typeflag ==FTW_SL )return 0 ;
tools/perf/builtin-buildid-cache.c +4 -4
@@ -175,7 +175,7 @@ static int build_id_cache__add_kcore(const char *filename, bool force)static int build_id_cache__add_file (const char *filename,struct nsinfo *nsi) {char sbuild_id[SBUILD_ID_SIZE ]; -struct build_id bid; +struct build_id bid = { .size =0 , };int err;struct nscookie nsc;
@@ -198,7 +198,7 @@ static int build_id_cache__add_file(const char *filename, struct nsinfo *nsi)static int build_id_cache__remove_file (const char *filename,struct nsinfo *nsi) {char sbuild_id[SBUILD_ID_SIZE ]; -struct build_id bid; +struct build_id bid = { .size =0 , };struct nscookie nsc;int err;
@@ -275,7 +275,7 @@ static int build_id_cache__purge_all(void)static bool dso__missing_buildid_cache (struct dso *dso,int parm__maybe_unused ) {char filename[PATH_MAX ]; -struct build_id bid; +struct build_id bid = { .size =0 , };if (!dso__build_id_filename (dso, filename,sizeof (filename), false))return true;
@@ -303,7 +303,7 @@ static int build_id_cache__fprintf_missing(struct perf_session *session, FILE *fstatic int build_id_cache__update_file (const char *filename,struct nsinfo *nsi) {char sbuild_id[SBUILD_ID_SIZE ]; -struct build_id bid; +struct build_id bid = { .size =0 , };struct nscookie nsc;int err;
tools/perf/tests/pe-file-parsing.c +1 -1
@@ -24,7 +24,7 @@ static int run_dir(const char *d) {char filename[PATH_MAX ];char debugfile[PATH_MAX ]; -struct build_id bid; +struct build_id bid = { .size =0 , };char debuglink[PATH_MAX ];char expect_build_id[] = {0x5a ,0x0f ,0xd8 ,0x82 ,0xb5 ,0x30 ,0x84 ,0x22 ,
tools/perf/tests/sdt.c +1 -1
@@ -28,7 +28,7 @@ static int target_function(void)static int build_id_cache__add_file (const char *filename) {char sbuild_id[SBUILD_ID_SIZE ]; -struct build_id bid; +struct build_id bid = { .size =0 , };int err; err =filename__read_build_id (filename, &bid);
tools/perf/util/build-id.c +3 -3
@@ -95,7 +95,7 @@ int build_id__snprintf(const struct build_id *build_id, char *bf, size_t bf_sizeint sysfs__snprintf_build_id (const char *root_dir,char *sbuild_id,size_t sbuild_id_size) {char notes[PATH_MAX ]; -struct build_id bid; +struct build_id bid = { .size =0 , };int ret;if (!root_dir)
@@ -112,7 +112,7 @@ int sysfs__snprintf_build_id(const char *root_dir, char *sbuild_id, size_t sbuilint filename__snprintf_build_id (const char *pathname,char *sbuild_id,size_t sbuild_id_size) { -struct build_id bid; +struct build_id bid = { .size =0 , };int ret; ret =filename__read_build_id (pathname, &bid);
@@ -849,7 +849,7 @@ static int filename__read_build_id_ns(const char *filename,static bool dso__build_id_mismatch (struct dso *dso,const char *name) { -struct build_id bid; +struct build_id bid = { .size =0 , };bool ret = false;mutex_lock (dso__lock (dso));
tools/perf/util/debuginfo.c +1 -1
@@ -103,7 +103,7 @@ struct debuginfo *debuginfo__new(const char *path)char buf[PATH_MAX ], nil ='\0' ;struct dso *dso;struct debuginfo *dinfo =NULL ; -struct build_id bid; +struct build_id bid = { .size =0 };/* Try to open distro debuginfo files */ dso =dso__new (path);
tools/perf/util/probe-event.c +2 -1
@@ -1063,7 +1063,6 @@ static int sprint_line_description(char *sbuf, size_t size, struct line_range *lstatic int __show_line_range (struct line_range *lr,const char *module,bool user) { -struct build_id bid;int l =1 ;struct int_node *ln;struct debuginfo *dinfo;
@@ -1088,6 +1087,8 @@ static int __show_line_range(struct line_range *lr, const char *module, ret = -ENOENT ; }if (dinfo->build_id) { +struct build_id bid; +build_id__init (&bid, dinfo->build_id,BUILD_ID_SIZE );build_id__sprintf (&bid, sbuild_id); }
tools/perf/util/probe-finder.c +2 -1
@@ -848,7 +848,6 @@ static int probe_point_lazy_walker(const char *fname, int lineno,/* Find probe points from lazy pattern */ static int find_probe_point_lazy (Dwarf_Die *sp_die,struct probe_finder *pf) { -struct build_id bid;char sbuild_id[SBUILD_ID_SIZE ] ="" ;int ret =0 ;char *fpath;
@@ -858,6 +857,8 @@ static int find_probe_point_lazy(Dwarf_Die *sp_die, struct probe_finder *pf) comp_dir =cu_get_comp_dir (&pf->cu_die);if (pf->dbg->build_id) { +struct build_id bid; +build_id__init (&bid, pf->dbg->build_id,BUILD_ID_SIZE );build_id__sprintf (&bid, sbuild_id); }
tools/perf/util/symbol-minimal.c +1 -1
@@ -343,7 +343,7 @@ int dso__load_sym(struct dso *dso, struct map *map __maybe_unused, struct symsrc *runtime_ss __maybe_unused, int kmodule __maybe_unused) { -struct build_id bid; +struct build_id bid = { .size =0 , };int ret; ret =fd__is_64_bit (ss->fd);
tools/perf/util/symbol.c +3 -2
@@ -1804,7 +1804,6 @@ int dso__load(struct dso *dso, struct map *map)struct symsrc *syms_ss =NULL , *runtime_ss =NULL ;bool kmod;bool perfmap; -struct build_id bid;struct nscookie nsc;char newmapname[PATH_MAX ];const char *map_path =dso__long_name (dso);
@@ -1865,6 +1864,8 @@ int dso__load(struct dso *dso, struct map *map)*/ if (!dso__has_build_id (dso) &&is_regular_file (dso__long_name (dso))) { +struct build_id bid = { .size =0 , }; +__symbol__join_symfs (name,PATH_MAX ,dso__long_name (dso));if (filename__read_build_id (name, &bid) >0 )dso__set_build_id (dso, &bid);
@@ -2113,7 +2114,7 @@ static bool filename__readable(const char *file)static char *dso__find_kallsyms (struct dso *dso,struct map *map) { -struct build_id bid; +struct build_id bid = { .size =0 , };char sbuild_id[SBUILD_ID_SIZE ];bool is_host = false;char path[PATH_MAX ];
tools/perf/util/synthetic-events.c +1 -1
@@ -368,7 +368,7 @@ static void perf_record_mmap2__read_build_id(struct perf_record_mmap2 *event, struct machine *machine, bool is_kernel) { -struct build_id bid; +struct build_id bid = { .size =0 , };struct nsinfo *nsi;struct nscookie nc;struct dso *dso =NULL ;
Signature · 1 line
2.49.0.901.g37484f566f-goog
Ian Rogers
reviewerPATCH v3 7/9 perf dso: Move build_id to dso_id
In reply to Ian RogersThe dso_id previously contained the major, minor, inode and inode generation information from a mmap2 event - the inode generation would be zero when reading from /proc/pid/maps. The build_id was in the dso. With build ID mmap2 events these fields wouldn't be initialized which would largely mean the special empty case where any dso would match for equality. This isn't desirable as if a dso is replaced we want the comparison to yield a difference.
To support detecting the difference between DSOs based on build_id, move the build_id out of the DSO and into the dso_id. The dso_id is also stored in the DSO so nothing is lost. Capture in the dso_id what parts have been initialized and rename dso_id__inject to dso_id__improve_id so that it is clear the dso_id is being improved upon with additional information. With the build_id in the dso_id, use memcmp to compare for equality.
--- tools/perf/builtin-buildid-list.c | 2 +- tools/perf/builtin-inject.c | 36 ++++---- tools/perf/builtin-report.c | 11 ++- tools/perf/include/perf/perf_dlfilter.h | 2 +- tools/perf/tests/symbols.c | 4 +- tools/perf/util/build-id.c | 4 +- tools/perf/util/dso.c | 109 +++++++++++++----------- tools/perf/util/dso.h | 75 ++++++++-------- tools/perf/util/dsos.c | 18 ++-- tools/perf/util/machine.c | 28 +++--- tools/perf/util/map.c | 13 ++- tools/perf/util/map.h | 5 +- tools/perf/util/sort.c | 27 +++--- tools/perf/util/synthetic-events.c | 18 ++-- 14 files changed, 197 insertions(+), 155 deletions(-)
tools/perf/builtin-buildid-list.c +1 -1
@@ -31,7 +31,7 @@ static int buildid__map_cb(struct map *map, void *arg __maybe_unused)memset (bid_buf,0 ,sizeof (bid_buf));if (dso__has_build_id (dso)) -build_id__snprintf (dso__bid_const (dso), bid_buf,sizeof (bid_buf)); +build_id__snprintf (dso__bid (dso), bid_buf,sizeof (bid_buf));printf ("%s %16" PRIx64" %16" PRIx64, bid_buf,map__start (map),map__end (map));if (dso_long_name !=NULL )printf (" %s" , dso_long_name);
tools/perf/builtin-inject.c +21 -15
@@ -587,15 +587,17 @@ static int perf_event__repipe_mmap2(const struct perf_tool *tool,struct perf_sample *sample, struct machine *machine) { - struct dso_id id; -struct dso_id *dso_id =NULL ; + struct dso_id id = dso_id_empty; -if (!(event->header.misc &PERF_RECORD_MISC_MMAP_BUILD_ID )) { +if (event->header.misc &PERF_RECORD_MISC_MMAP_BUILD_ID ) { +build_id__init (&id.build_id, event->mmap2.build_id, event->mmap2.build_id_size); + }else { id.maj = event->mmap2.maj; id.min = event->mmap2.min; id.ino = event->mmap2.ino; id.ino_generation = event->mmap2.ino_generation; - dso_id = &id; + id.mmap2_valid = true; + id.mmap2_ino_generation_valid = true; } return perf_event__repipe_common_mmap(
@@ -603,7 +605,7 @@ static int perf_event__repipe_mmap2(const struct perf_tool *tool, event->mmap2.pid, event->mmap2.tid, event->mmap2.start, event->mmap2.len, event->mmap2.pgoff, event->mmap2.flags, event->mmap2.prot, - event->mmap2.filename, dso_id, + event->mmap2.filename, &id, perf_event__process_mmap2); }
@@ -671,19 +673,20 @@ static int perf_event__repipe_tracing_data(struct perf_session *session,static int dso__read_build_id (struct dso *dso) {struct nscookie nsc; +struct build_id bid = { .size =0 , };if (dso__has_build_id (dso))return 0 ;mutex_lock (dso__lock (dso));nsinfo__mountns_enter (dso__nsinfo (dso), &nsc); -if (filename__read_build_id (dso__long_name (dso),dso__bid (dso)) >0 ) -dso__set_has_build_id (dso); +if (filename__read_build_id (dso__long_name (dso), &bid) >0 ) +dso__set_build_id (dso, &bid);else if (dso__nsinfo (dso)) {char *new_name =dso__filename_with_chroot (dso,dso__long_name (dso)); -if (new_name &&filename__read_build_id (new_name,dso__bid (dso)) >0 ) -dso__set_has_build_id (dso); +if (new_name &&filename__read_build_id (new_name, &bid) >0 ) +dso__set_build_id (dso, &bid);free (new_name); }nsinfo__mountns_exit (&nsc);
@@ -732,23 +735,26 @@ static bool perf_inject__lookup_known_build_id(struct perf_inject *inject, struct dso *dso) {struct str_node *pos; -int bid_len;strlist__for_each_entry (pos, inject->known_build_ids) { +struct build_id bid;const char *build_id, *dso_name; +size_t bid_len; build_id =skip_spaces (pos->s); dso_name =strchr (build_id,' ' ); bid_len = dso_name - pos->s; +if (bid_len >sizeof (bid.data)) + bid_len =sizeof (bid.data); dso_name =skip_spaces (dso_name);if (strcmp (dso__long_name (dso), dso_name))continue ; -for (int ix =0 ;2 * ix +1 < bid_len; ++ix) { -dso__bid (dso)->data[ix] = (hex (build_id[2 * ix]) <<4 | -hex (build_id[2 * ix +1 ])); +for (size_t ix =0 ;2 * ix +1 < bid_len; ++ix) { + bid.data[ix] = (hex (build_id[2 * ix]) <<4 | +hex (build_id[2 * ix +1 ])); } -dso__bid (dso)->size = bid_len /2 ; -dso__set_has_build_id (dso); + bid.size = bid_len /2 ; +dso__set_build_id (dso, &bid);return true; }return false;
tools/perf/builtin-report.c +9 -2
@@ -858,17 +858,24 @@ static int maps__fprintf_task_cb(struct map *map, void *data)struct maps__fprintf_task_args *args = data;const struct dso *dso =map__dso (map);u32 prot =map__prot (map); +const struct dso_id *dso_id =dso__id_const (dso);int ret; +char buf[SBUILD_ID_SIZE ]; + +if (dso_id->mmap2_valid) +snprintf (buf,sizeof (buf),"%" PRIu64, dso_id->ino); +else +build_id__snprintf (&dso_id->build_id, buf,sizeof (buf)); ret =fprintf (args->fp, -"%*s %" PRIx64"-%" PRIx64" %c%c%c%c %08" PRIx64" %" PRIu64" %s\n" , +"%*s %" PRIx64"-%" PRIx64" %c%c%c%c %08" PRIx64" %s %s\n" , args->indent,"" ,map__start (map),map__end (map), prot &PROT_READ ?'r' :'-' , prot &PROT_WRITE ?'w' :'-' , prot &PROT_EXEC ?'x' :'-' ,map__flags (map) ?'s' :'p' ,map__pgoff (map), -dso__id_const (dso)->ino,dso__name (dso)); + buf,dso__name (dso));if (ret <0 )return ret;
tools/perf/include/perf/perf_dlfilter.h +1 -1
@@ -87,7 +87,7 @@ struct perf_dlfilter_al {__u8 is_64_bit;/* Only valid if dso is not NULL */ __u8 is_kernel_ip;/* True if in kernel space */ __u32 buildid_size; -__u8 *buildid; +const __u8 *buildid;/* Below members are only populated by resolve_ip() */ __u8 filtered;/* True if this sample event will be filtered out */ const char *comm;
tools/perf/tests/symbols.c +2 -2
@@ -96,8 +96,8 @@ static int create_map(struct test_info *ti, char *filename, struct map **map_p)dso__put (dso);/* Create a dummy map at 0x100000 */ - *map_p =map__new (ti->machine,0x100000 ,0xffffffff ,0 ,NULL , -PROT_EXEC ,0 ,NULL , filename, ti->thread); + *map_p =map__new (ti->machine,0x100000 ,0xffffffff ,0 , &dso_id_empty, +PROT_EXEC ,/*flags=*/ 0 , filename, ti->thread);if (!*map_p) {pr_debug ("Failed to create map!" );return TEST_FAIL ;
tools/perf/util/build-id.c +2 -2
@@ -251,7 +251,7 @@ char *__dso__build_id_filename(const struct dso *dso, char *bf, size_t size,if (!dso__has_build_id (dso))return NULL ; -build_id__snprintf (dso__bid_const (dso), sbuild_id,sizeof (sbuild_id)); +build_id__snprintf (dso__bid (dso), sbuild_id,sizeof (sbuild_id)); linkname =build_id_cache__linkname (sbuild_id,NULL ,0 );if (!linkname)return NULL ;
@@ -334,7 +334,7 @@ static int machine__write_buildid_table_cb(struct dso *dso, void *data) } in_kernel =dso__kernel (dso) ||is_kernel_module (name,PERF_RECORD_MISC_CPUMODE_UNKNOWN ); -return write_buildid (name, name_len,dso__bid (dso), args->machine->pid, +return write_buildid (name, name_len, &dso__id (dso)->build_id, args->machine->pid, in_kernel ? args->kmisc : args->umisc, args->fd); }
tools/perf/util/dso.c +60 -49
180 lines not on this page · raw message
tools/perf/util/dso.h +41 -34
136 lines not on this page · raw message
tools/perf/util/dsos.c +10 -8
@@ -72,6 +72,7 @@ static int dsos__read_build_ids_cb(struct dso *dso, void *data) {struct dsos__read_build_ids_cb_args *args = data;struct nscookie nsc; +struct build_id bid = { .size =0 , };if (args->with_hits && !dso__hit (dso) && !dso__is_vdso (dso))return 0 ;
@@ -80,15 +81,15 @@ static int dsos__read_build_ids_cb(struct dso *dso, void *data)return 0 ; }nsinfo__mountns_enter (dso__nsinfo (dso), &nsc); -if (filename__read_build_id (dso__long_name (dso),dso__bid (dso)) >0 ) { +if (filename__read_build_id (dso__long_name (dso), &bid) >0 ) { +dso__set_build_id (dso, &bid); args->have_build_id = true; -dso__set_has_build_id (dso); }else if (errno ==ENOENT &&dso__nsinfo (dso)) {char *new_name =dso__filename_with_chroot (dso,dso__long_name (dso)); -if (new_name &&filename__read_build_id (new_name,dso__bid (dso)) >0 ) { +if (new_name &&filename__read_build_id (new_name, &bid) >0 ) { +dso__set_build_id (dso, &bid); args->have_build_id = true; -dso__set_has_build_id (dso); }free (new_name); }
@@ -284,7 +285,7 @@ struct dso *dsos__find(struct dsos *dsos, const char *name, bool cmp_short)struct dso *res;down_read (&dsos->lock); - res =__dsos__find_id (dsos, name,NULL , cmp_short,/*write_locked=*/ false); + res =__dsos__find_id (dsos, name, &dso_id_empty, cmp_short,/*write_locked=*/ false);up_read (&dsos->lock);return res; }
@@ -341,8 +342,8 @@ static struct dso *__dsos__findnew_id(struct dsos *dsos, const char *name, const {struct dso *dso =__dsos__find_id (dsos, name, id, false,/*write_locked=*/ true); -if (dso &&dso_id__empty (dso__id (dso)) && !dso_id__empty (id)) -__dso__inject_id (dso, id); +if (dso) +__dso__improve_id (dso, id);return dso ? dso :__dsos__addnew_id (dsos, name, id); }
@@ -433,7 +434,8 @@ struct dso *dsos__findnew_module_dso(struct dsos *dsos,down_write (&dsos->lock); - dso =__dsos__find_id (dsos, m->name,NULL ,/*cmp_short=*/ true,/*write_locked=*/ true); + dso =__dsos__find_id (dsos, m->name, &dso_id_empty,/*cmp_short=*/ true, +/*write_locked=*/ true);if (dso) {up_write (&dsos->lock);return dso;
tools/perf/util/machine.c +14 -14
@@ -1696,21 +1696,21 @@ int machine__process_mmap2_event(struct machine *machine, {struct thread *thread;struct map *map; -struct dso_id dso_id = { - .maj = event->mmap2.maj, - .min = event->mmap2.min, - .ino = event->mmap2.ino, - .ino_generation = event->mmap2.ino_generation, - }; -struct build_id __bid, *bid =NULL ; +struct dso_id dso_id = dso_id_empty;int ret =0 ;if (dump_trace)perf_event__fprintf_mmap2 (event, stdout);if (event->header.misc &PERF_RECORD_MISC_MMAP_BUILD_ID ) { - bid = &__bid; -build_id__init (bid, event->mmap2.build_id, event->mmap2.build_id_size); +build_id__init (&dso_id.build_id, event->mmap2.build_id, event->mmap2.build_id_size); + }else { + dso_id.maj = event->mmap2.maj; + dso_id.min = event->mmap2.min; + dso_id.ino = event->mmap2.ino; + dso_id.ino_generation = event->mmap2.ino_generation; + dso_id.mmap2_valid = true; + dso_id.mmap2_ino_generation_valid = true; } if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
@@ -1722,7 +1722,7 @@ int machine__process_mmap2_event(struct machine *machine, };strlcpy (xm.name, event->mmap2.filename,KMAP_NAME_LEN ); - ret =machine__process_kernel_mmap_event (machine, &xm, bid); + ret =machine__process_kernel_mmap_event (machine, &xm, &dso_id.build_id);if (ret <0 ) goto out_problem;return 0 ;
@@ -1736,7 +1736,7 @@ int machine__process_mmap2_event(struct machine *machine, map =map__new (machine, event->mmap2.start, event->mmap2.len, event->mmap2.pgoff, &dso_id, event->mmap2.prot, - event->mmap2.flags, bid, + event->mmap2.flags, event->mmap2.filename, thread);if (map ==NULL )
@@ -1794,8 +1794,8 @@ int machine__process_mmap_event(struct machine *machine, union perf_event *event prot =PROT_EXEC ; map =map__new (machine, event->mmap.start, - event->mmap.len, event->mmap.pgoff, -NULL , prot,0 ,NULL , event->mmap.filename, thread); + event->mmap.len, event->mmap.pgoff, + &dso_id_empty, prot,/*flags=*/ 0 , event->mmap.filename, thread);if (map ==NULL ) goto out_problem_map;
@@ -3157,7 +3157,7 @@ struct dso *machine__findnew_dso_id(struct machine *machine, const char *filenamstruct dso *machine__findnew_dso (struct machine *machine,const char *filename) { -return machine__findnew_dso_id (machine, filename,NULL ); +return machine__findnew_dso_id (machine, filename, &dso_id_empty); }char *machine__resolve_kernel_addr (void *vmachine,unsigned long long *addrp,char **modp)
tools/perf/util/map.c +6 -7
@@ -120,8 +120,8 @@ static void map__init(struct map *map, u64 start, u64 end, u64 pgoff, }struct map *map__new (struct machine *machine,u64 start,u64 len, -u64 pgoff,struct dso_id *id, -u32 prot,u32 flags,struct build_id *bid, +u64 pgoff,const struct dso_id *id, +u32 prot,u32 flags,char *filename,struct thread *thread) {struct map *result;
@@ -132,7 +132,7 @@ struct map *map__new(struct machine *machine, u64 start, u64 len, map =zalloc (sizeof (*map));if (ADD_RC_CHK (result, map)) {char newfilename[PATH_MAX ]; -struct dso *dso, *header_bid_dso; +struct dso *dso;int anon, no_dso, vdso, android; android =is_android_lib (filename);
@@ -189,16 +189,15 @@ struct map *map__new(struct machine *machine, u64 start, u64 len,dso__set_nsinfo (dso, nsi);mutex_unlock (dso__lock (dso)); -if (build_id__is_defined (bid)) { -dso__set_build_id (dso, bid); - }else { +if (!build_id__is_defined (&id->build_id)) {/* * If the mmap event had no build ID, search for an existing dso from the * build ID header by name. Otherwise only the dso loaded at the time of * reading the header will have the build ID set and all future mmaps will * have it missing. */ - header_bid_dso =dsos__find (&machine->dsos, filename, false); +struct dso *header_bid_dso =dsos__find (&machine->dsos, filename, false); +if (header_bid_dso &&dso__header_build_id (header_bid_dso)) {dso__set_build_id (dso,dso__bid (header_bid_dso));dso__set_header_build_id (dso,1 );
tools/perf/util/map.h +2 -3
@@ -173,11 +173,10 @@ struct thread;__map__for_each_symbol_by_name (map, sym_name, (pos), idx) struct dso_id; -struct build_id ;struct map *map__new (struct machine *machine,u64 start,u64 len, -u64 pgoff,struct dso_id *id,u32 prot,u32 flags, -struct build_id *bid,char *filename,struct thread *thread); +u64 pgoff,const struct dso_id *id,u32 prot,u32 flags, +char *filename,struct thread *thread);struct map *map__new2 (u64 start,struct dso *dso);void map__delete (struct map *map);struct map *map__clone (struct map *map);
tools/perf/util/sort.c +16 -11
@@ -1709,22 +1709,27 @@ sort__dcacheline_cmp(struct hist_entry *left, struct hist_entry *right)if (rc)return rc;/* -* Addresses with no major/minor numbers are assumed to be +* Addresses with no major/minor numbers or build ID are assumed to be * anonymous in userspace. Sort those on pid then address. * * The kernel and non-zero major/minor mapped areas are * assumed to be unity mapped. Sort those on address. */ +if (left->cpumode !=PERF_RECORD_MISC_KERNEL && (map__flags (l_map) &MAP_SHARED ) ==0 ) { +const struct dso_id *dso_id =dso__id_const (l_dso); -if ((left->cpumode !=PERF_RECORD_MISC_KERNEL ) && - (!(map__flags (l_map) &MAP_SHARED )) && !dso__id (l_dso)->maj && !dso__id (l_dso)->min && - !dso__id (l_dso)->ino && !dso__id (l_dso)->ino_generation) { -/* userspace anonymous */ +if (!dso_id->mmap2_valid) + dso_id =dso__id_const (r_dso); -if (thread__pid (left->thread) >thread__pid (right->thread)) -return -1 ; -if (thread__pid (left->thread) <thread__pid (right->thread)) -return 1 ; +if (!build_id__is_defined (&dso_id->build_id) && + (!dso_id->mmap2_valid || (dso_id->maj ==0 && dso_id->min ==0 ))) { +/* userspace anonymous */ + +if (thread__pid (left->thread) >thread__pid (right->thread)) +return -1 ; +if (thread__pid (left->thread) <thread__pid (right->thread)) +return 1 ; + } } addr:
@@ -1749,6 +1754,7 @@ static int hist_entry__dcacheline_snprintf(struct hist_entry *he, char *bf,if (he->mem_info) {struct map *map =mem_info__daddr (he->mem_info)->ms.map;struct dso *dso = map ?map__dso (map) :NULL ; +const struct dso_id *dso_id = dso ?dso__id_const (dso) : &dso_id_empty; addr =cl_address (mem_info__daddr (he->mem_info)->al_addr, chk_double_cl); ms = &mem_info__daddr (he->mem_info)->ms;
@@ -1757,8 +1763,7 @@ static int hist_entry__dcacheline_snprintf(struct hist_entry *he, char *bf,if ((he->cpumode !=PERF_RECORD_MISC_KERNEL ) && map && !(map__prot (map) &PROT_EXEC ) && (map__flags (map) &MAP_SHARED ) && - (dso__id (dso)->maj ||dso__id (dso)->min ||dso__id (dso)->ino || -dso__id (dso)->ino_generation)) + (!dso_id->mmap2_valid || (dso_id->maj ==0 && dso_id->min ==0 ))) level ='s' ;else if (!map) level ='X' ;
tools/perf/util/synthetic-events.c +12 -6
@@ -372,7 +372,7 @@ static void perf_record_mmap2__read_build_id(struct perf_record_mmap2 *event,struct nsinfo *nsi;struct nscookie nc;struct dso *dso =NULL ; -struct dso_id id; +struct dso_id dso_id = dso_id_empty;int rc;if (is_kernel) {
@@ -380,12 +380,18 @@ static void perf_record_mmap2__read_build_id(struct perf_record_mmap2 *event, goto out; } - id.maj = event->maj; - id.min = event->min; - id.ino = event->ino; - id.ino_generation = event->ino_generation; +if (event->header.misc &PERF_RECORD_MISC_MMAP_BUILD_ID ) { +build_id__init (&dso_id.build_id, event->build_id, event->build_id_size); + }else { + dso_id.maj = event->maj; + dso_id.min = event->min; + dso_id.ino = event->ino; + dso_id.ino_generation = event->ino_generation; + dso_id.mmap2_valid = true; + dso_id.mmap2_ino_generation_valid = true; + }; - dso =dsos__findnew_id (&machine->dsos, event->filename, &id); + dso =dsos__findnew_id (&machine->dsos, event->filename, &dso_id);if (dso &&dso__has_build_id (dso)) { bid = *dso__bid (dso); rc =0 ;
Signature · 1 line
2.49.0.901.g37484f566f-goog
Ian Rogers
reviewerPATCH v3 8/9 perf jitdump: Directly mark the jitdump DSO
In reply to Ian RogersThe DSO being generated was being accessed through a thread's maps, this is unnecessary as the dso can just be directly found. This avoids problems with passing a NULL evsel which may be inspected to determine properties of a callchain when using the buildid DSO marking code.
--- tools/perf/util/jitdump.c | 21 +++++++++++++++++---- 1 file changed, 17 insertions(+), 4 deletions(-)
tools/perf/util/jitdump.c +17 -4
@@ -14,9 +14,9 @@#include <sys/mman.h> #include <linux/stringify.h> -#include "build-id.h" #include "event.h" #include "debug.h" +#include "dso.h" #include "evlist.h" #include "namespaces.h" #include "symbol.h"
@@ -531,9 +531,22 @@ static int jit_repipe_code_load(struct jit_buf_desc *jd, union jr_entry *jr)/* * mark dso as use to generate buildid in the header */ -if (!ret) -build_id__mark_dso_hit (tool, event, &sample,NULL , jd->machine); - +if (!ret) { +struct dso_id dso_id = { + { + .maj = event->mmap2.maj, + .min = event->mmap2.min, + .ino = event->mmap2.ino, + .ino_generation = event->mmap2.ino_generation, + }, + .mmap2_valid = true, + .mmap2_ino_generation_valid = true, + }; +struct dso *dso =machine__findnew_dso_id (jd->machine, filename, &dso_id); + +if (dso) +dso__set_hit (dso); + } out:perf_sample__exit (&sample);free (event);
Signature · 1 line
2.49.0.901.g37484f566f-goog
Ian Rogers
reviewerPATCH v3 9/9 perf record: Make --buildid-mmap the default
In reply to Ian RogersSupport for build IDs in mmap2 perf events has been present since Linux v5.12:
https://lore.kernel.org/lkml/[email protected]/
Build ID mmap events don't avoid the need to inject build IDs for DSO touched by samples as the build ID cache is populated by perf record. They can avoid some cases of symbol mis-resolution caused by the file system changing from when a sample occurred and when the DSO is sought. To disable build ID scanning
Unlike the --buildid-mmap option, this doesn't disable the build ID cache but it does disable the processing of samples looking for DSOs to inject build IDs for. To disable the build ID cache the -B (--no-buildid) option should be used.
Making this option the default was raised on the list in:
https://lore.kernel.org/linux-perf-users/CAP-5=fXP7jN_QrGUcd55_QH5J-Y-FCaJ6=NaHVtyx0oyNh8_-Q@mail.gmail.com/
--- tools/perf/builtin-record.c | 33 +++++++++++++++++++----------- tools/perf/util/symbol_conf.h | 2 +- tools/perf/util/synthetic-events.c | 16 +++++++-------- 3 files changed, 30 insertions(+), 21 deletions(-)
tools/perf/builtin-record.c +21 -12
@@ -169,6 +169,7 @@ struct record {bool no_buildid_cache_set;bool buildid_all;bool buildid_mmap; +bool buildid_mmap_set;bool timestamp_filename;bool timestamp_boundary;bool off_cpu;
@@ -1795,6 +1796,7 @@ record__finish_output(struct record *rec) data->dir.files[i].size =lseek (data->dir.files[i].fd,0 ,SEEK_CUR ); } +/* Buildid scanning disabled or build ID in kernel and synthesized map events. */ if (!rec->no_buildid) {process_buildids (rec);
@@ -2966,6 +2968,8 @@ static int perf_record_config(const char *var, const char *value, void *cb) rec->no_buildid = true; else if (!strcmp(value, "mmap")) rec->buildid_mmap = true; + else if (!strcmp(value, "no-mmap")) + rec->buildid_mmap = false; else return -1;return 0 ;
@@ -3349,6 +3353,7 @@ static struct record record = { .ctl_fd_ack = -1, .synth = PERF_SYNTH_ALL, }, + .buildid_mmap = true, };const char record_callchain_help[] =CALLCHAIN_RECORD_HELP
@@ -3514,8 +3519,8 @@ static struct option __record_options[] = { "file", "vmlinux pathname"),OPT_BOOLEAN (0 ,"buildid-all" , &record.buildid_all,"Record build-id of all DSOs regardless of hits" ), -OPT_BOOLEAN (0 ,"buildid-mmap" , &record.buildid_mmap, -"Record build-id in map events" ), +OPT_BOOLEAN_SET (0 ,"buildid-mmap" , &record.buildid_mmap, &record.buildid_mmap_set, +"Legacy record build-id in map events option which is now the default. Behaves indentically to --no-buildid. Disable with --no-buildid-mmap" ),OPT_BOOLEAN (0 ,"timestamp-filename" , &record.timestamp_filename,"append timestamp to output filename" ), OPT_BOOLEAN(0, "timestamp-boundary", &record.timestamp_boundary,
@@ -4042,19 +4047,23 @@ int cmd_record(int argc, const char **argv) record.opts.record_switch_events = true; } +if (!rec->buildid_mmap) { +pr_debug ("Disabling build id in synthesized mmap2 events.\n" ); + symbol_conf.no_buildid_mmap2 = true; + }else if (rec->buildid_mmap_set) { +/* +* Explicitly passing --buildid-mmap disables buildid processing +* and cache generation. +*/ + rec->no_buildid = true; + } +if (rec->buildid_mmap && !perf_can_record_build_id ()) { +pr_warning ("Missing support for build id in kernel mmap events. Disable this warning with --no-buildid-mmap\n" ); + rec->buildid_mmap = false; + }if (rec->buildid_mmap) { -if (!perf_can_record_build_id ()) { -pr_err ("Failed: no support to record build id in mmap events, update your kernel.\n" ); - err = -EINVAL ; - goto out_opts; - } -pr_debug ("Enabling build id in mmap2 events.\n" ); -/* Enable mmap build id synthesizing. */ - symbol_conf.buildid_mmap2 = true;/* Enable perf_event_attr::build_id bit. */ rec->opts.build_id = true; -/* Disable build id cache. */ - rec->no_buildid = true; }if (rec->opts.record_cgroup && !perf_can_record_cgroup ()) {
tools/perf/util/symbol_conf.h +1 -1
@@ -43,7 +43,7 @@ struct symbol_conf { report_individual_block, inline_name, disable_add2line_warn, - buildid_mmap2, + no_buildid_mmap2, guest_code, lazy_load_kernel_maps, keep_exited_threads,
tools/perf/util/synthetic-events.c +8 -8
@@ -532,7 +532,7 @@ int perf_event__synthesize_mmap_events(const struct perf_tool *tool, event->mmap2.pid = tgid; event->mmap2.tid = pid; -if (symbol_conf.buildid_mmap2) +if (!symbol_conf.no_buildid_mmap2)perf_record_mmap2__read_build_id (&event->mmap2, machine, false);if (perf_tool__process_synth_event (tool, event, machine, process) !=0 ) {
@@ -690,7 +690,7 @@ static int perf_event__synthesize_modules_maps_cb(struct map *map, void *data)return 0 ; dso =map__dso (map); -if (symbol_conf.buildid_mmap2) { +if (!symbol_conf.no_buildid_mmap2) { size =PERF_ALIGN (dso__long_name_len (dso) +1 ,sizeof (u64)); event->mmap2.header.type =PERF_RECORD_MMAP2 ; event->mmap2.header.size = (sizeof(event->mmap2) -
@@ -734,9 +734,9 @@ int perf_event__synthesize_modules(const struct perf_tool *tool, perf_event__han .process = process, .machine = machine, }; -size_t size = symbol_conf.buildid_mmap2 - ?sizeof (args.event->mmap2) - :sizeof (args.event->mmap); +size_t size = symbol_conf.no_buildid_mmap2 + ?sizeof (args.event->mmap) + :sizeof (args.event->mmap2); args.event =zalloc (size + machine->id_hdr_size);if (args.event ==NULL ) {
@@ -1124,8 +1124,8 @@ static int __perf_event__synthesize_kernel_mmap(const struct perf_tool *tool, struct machine *machine) {union perf_event *event; -size_t size = symbol_conf.buildid_mmap2 ? -sizeof (event->mmap2) :sizeof (event->mmap); +size_t size = symbol_conf.no_buildid_mmap2 ? +sizeof (event->mmap) :sizeof (event->mmap2);struct map *map =machine__kernel_map (machine);struct kmap *kmap;int err;
@@ -1159,7 +1159,7 @@ static int __perf_event__synthesize_kernel_mmap(const struct perf_tool *tool, event->header.misc =PERF_RECORD_MISC_GUEST_KERNEL ; } -if (symbol_conf.buildid_mmap2) { +if (!symbol_conf.no_buildid_mmap2) { size =snprintf (event->mmap2.filename,sizeof (event->mmap2.filename),"%s%s" , machine->mmap_name, kmap->ref_reloc_sym->name) +1 ; size =PERF_ALIGN (size,sizeof (u64));
Signature · 1 line
2.49.0.901.g37484f566f-goog
Namhyung Kim
maintainerRe: [PATCH v3 9/9] perf record: Make --buildid-mmap the default
Quoted 124 lines · Ian Rogers
Support for build IDs in mmap2 perf events has been present since Linux v5.12:
https://lore.kernel.org/lkml/[email protected]/
Build ID mmap events don't avoid the need to inject build IDs for DSO touched by samples as the build ID cache is populated by perf record. They can avoid some cases of symbol mis-resolution caused by the file system changing from when a sample occurred and when the DSO is sought. To disable build ID scanningUnlike the --buildid-mmap option, this doesn't disable the build ID cache but it does disable the processing of samples looking for DSOs to inject build IDs for. To disable the build ID cache the -B (--no-buildid) option should be used.
Making this option the default was raised on the list in:
https://lore.kernel.org/linux-perf-users/CAP-5=fXP7jN_QrGUcd55_QH5J-Y-FCaJ6=NaHVtyx0oyNh8_-Q@mail.gmail.com/Signed-off-by Ian Rogers <[email protected]>--- tools/perf/builtin-record.c | 33 +++++++++++++++++++----------- tools/perf/util/symbol_conf.h | 2 +- tools/perf/util/synthetic-events.c | 16 +++++++-------- 3 files changed, 30 insertions(+), 21 deletions(-)tools/perf/builtin-record.c +21
-12@@ -169,6 +169,7 @@ struct record {bool no_buildid_cache_set;bool buildid_all;bool buildid_mmap; +bool buildid_mmap_set;bool timestamp_filename;bool timestamp_boundary;bool off_cpu;@@ -1795,6 +1796,7 @@ record__finish_output(struct record *rec) data->dir.files[i].size =lseek (data->dir.files[i].fd,0 ,SEEK_CUR ); } +/* Buildid scanning disabled or build ID in kernel and synthesized map events. */ if (!rec->no_buildid) {process_buildids (rec);@@ -2966,6 +2968,8 @@ static int perf_record_config(const char *var, const char *value, void *cb) rec->no_buildid = true; else if (!strcmp(value, "mmap")) rec->buildid_mmap = true; + else if (!strcmp(value, "no-mmap")) + rec->buildid_mmap = false; else return -1;return 0 ;@@ -3349,6 +3353,7 @@ static struct record record = { .ctl_fd_ack = -1, .synth = PERF_SYNTH_ALL, }, + .buildid_mmap = true, };const char record_callchain_help[] =CALLCHAIN_RECORD_HELP @@ -3514,8 +3519,8 @@ static struct option __record_options[] = { "file", "vmlinux pathname"),OPT_BOOLEAN (0 ,"buildid-all" , &record.buildid_all,"Record build-id of all DSOs regardless of hits" ), -OPT_BOOLEAN (0 ,"buildid-mmap" , &record.buildid_mmap, -"Record build-id in map events" ), +OPT_BOOLEAN_SET (0 ,"buildid-mmap" , &record.buildid_mmap, &record.buildid_mmap_set, +"Legacy record build-id in map events option which is now the default. Behaves indentically to --no-buildid. Disable with --no-buildid-mmap" ),OPT_BOOLEAN (0 ,"timestamp-filename" , &record.timestamp_filename,"append timestamp to output filename" ), OPT_BOOLEAN(0, "timestamp-boundary", &record.timestamp_boundary,@@ -4042,19 +4047,23 @@ int cmd_record(int argc, const char **argv) record.opts.record_switch_events = true; } +if (!rec->buildid_mmap) { +pr_debug ("Disabling build id in synthesized mmap2 events.\n" ); + symbol_conf.no_buildid_mmap2 = true; + }else if (rec->buildid_mmap_set) { +/* +* Explicitly passing --buildid-mmap disables buildid processing +* and cache generation. +*/ + rec->no_buildid = true; + } +if (rec->buildid_mmap && !perf_can_record_build_id ()) { +pr_warning ("Missing support for build id in kernel mmap events. Disable this warning with --no-buildid-mmap\n" ); + rec->buildid_mmap = false; + }if (rec->buildid_mmap) { -if (!perf_can_record_build_id ()) { -pr_err ("Failed: no support to record build id in mmap events, update your kernel.\n" ); - err = -EINVAL ; - goto out_opts; - } -pr_debug ("Enabling build id in mmap2 events.\n" ); -/* Enable mmap build id synthesizing. */ - symbol_conf.buildid_mmap2 = true;/* Enable perf_event_attr::build_id bit. */ rec->opts.build_id = true; -/* Disable build id cache. */ - rec->no_buildid = true; }if (rec->opts.record_cgroup && !perf_can_record_cgroup ()) {tools/perf/util/symbol_conf.h +1
-1@@ -43,7 +43,7 @@ struct symbol_conf { report_individual_block, inline_name, disable_add2line_warn, - buildid_mmap2, + no_buildid_mmap2, guest_code, lazy_load_kernel_maps, keep_exited_threads,tools/perf/util/synthetic-events.c +0
-0
tools/perf/util/synthetic-events.c int perf_event__synthesize_mmap_events(const struct perf_tool *tool, · line 535
Quoted 5 lines
tools/perf/util/synthetic-events.c +1
-1@@ -532,7 +532,7 @@ int perf_event__synthesize_mmap_events(const struct perf_tool *tool, event->mmap2.pid = tgid; event->mmap2.tid = pid; -if (symbol_conf.buildid_mmap2) +if (!symbol_conf.no_buildid_mmap2)
I find the double negation confusing. Can we keep it positive?
Thanks,
Namhyung
Quoted 48 lines
tools/perf/util/synthetic-events.c +7
-7perf_record_mmap2__read_build_id (&event->mmap2, machine, false);if (perf_tool__process_synth_event (tool, event, machine, process) !=0 ) {@@ -690,7 +690,7 @@ static int perf_event__synthesize_modules_maps_cb(struct map *map, void *data)return 0 ; dso =map__dso (map); -if (symbol_conf.buildid_mmap2) { +if (!symbol_conf.no_buildid_mmap2) { size =PERF_ALIGN (dso__long_name_len (dso) +1 ,sizeof (u64)); event->mmap2.header.type =PERF_RECORD_MMAP2 ; event->mmap2.header.size = (sizeof(event->mmap2) -@@ -734,9 +734,9 @@ int perf_event__synthesize_modules(const struct perf_tool *tool, perf_event__han .process = process, .machine = machine, }; -size_t size = symbol_conf.buildid_mmap2 - ?sizeof (args.event->mmap2) - :sizeof (args.event->mmap); +size_t size = symbol_conf.no_buildid_mmap2 + ?sizeof (args.event->mmap) + :sizeof (args.event->mmap2); args.event =zalloc (size + machine->id_hdr_size);if (args.event ==NULL ) {@@ -1124,8 +1124,8 @@ static int __perf_event__synthesize_kernel_mmap(const struct perf_tool *tool, struct machine *machine) {union perf_event *event; -size_t size = symbol_conf.buildid_mmap2 ? -sizeof (event->mmap2) :sizeof (event->mmap); +size_t size = symbol_conf.no_buildid_mmap2 ? +sizeof (event->mmap) :sizeof (event->mmap2);struct map *map =machine__kernel_map (machine);struct kmap *kmap;int err;@@ -1159,7 +1159,7 @@ static int __perf_event__synthesize_kernel_mmap(const struct perf_tool *tool, event->header.misc =PERF_RECORD_MISC_GUEST_KERNEL ; } -if (symbol_conf.buildid_mmap2) { +if (!symbol_conf.no_buildid_mmap2) { size =snprintf (event->mmap2.filename,sizeof (event->mmap2.filename),"%s%s" , machine->mmap_name, kmap->ref_reloc_sym->name) +1 ; size =PERF_ALIGN (size,sizeof (u64));Signature · 1 line
2.49.0.901.g37484f566f-goog
Ian Rogers
reviewerRe: [PATCH v3 9/9] perf record: Make --buildid-mmap the default
Quoted 129 lines · Namhyung Kim
Quoted 127 lines · Ian Rogers
Support for build IDs in mmap2 perf events has been present since Linux v5.12:
https://lore.kernel.org/lkml/[email protected]/
Build ID mmap events don't avoid the need to inject build IDs for DSO touched by samples as the build ID cache is populated by perf record. They can avoid some cases of symbol mis-resolution caused by the file system changing from when a sample occurred and when the DSO is sought. To disable build ID scanningUnlike the --buildid-mmap option, this doesn't disable the build ID cache but it does disable the processing of samples looking for DSOs to inject build IDs for. To disable the build ID cache the -B (--no-buildid) option should be used.
Making this option the default was raised on the list in:
https://lore.kernel.org/linux-perf-users/CAP-5=fXP7jN_QrGUcd55_QH5J-Y-FCaJ6=NaHVtyx0oyNh8_-Q@mail.gmail.com/Signed-off-by Ian Rogers <[email protected]>--- tools/perf/builtin-record.c | 33 +++++++++++++++++++----------- tools/perf/util/symbol_conf.h | 2 +- tools/perf/util/synthetic-events.c | 16 +++++++-------- 3 files changed, 30 insertions(+), 21 deletions(-)tools/perf/builtin-record.c +21
-12@@ -169,6 +169,7 @@ struct record {bool no_buildid_cache_set;bool buildid_all;bool buildid_mmap; +bool buildid_mmap_set;bool timestamp_filename;bool timestamp_boundary;bool off_cpu;@@ -1795,6 +1796,7 @@ record__finish_output(struct record *rec) data->dir.files[i].size =lseek (data->dir.files[i].fd,0 ,SEEK_CUR ); } +/* Buildid scanning disabled or build ID in kernel and synthesized map events. */ if (!rec->no_buildid) {process_buildids (rec);@@ -2966,6 +2968,8 @@ static int perf_record_config(const char *var, const char *value, void *cb) rec->no_buildid = true; else if (!strcmp(value, "mmap")) rec->buildid_mmap = true; + else if (!strcmp(value, "no-mmap")) + rec->buildid_mmap = false; else return -1;return 0 ;@@ -3349,6 +3353,7 @@ static struct record record = { .ctl_fd_ack = -1, .synth = PERF_SYNTH_ALL, }, + .buildid_mmap = true, };const char record_callchain_help[] =CALLCHAIN_RECORD_HELP @@ -3514,8 +3519,8 @@ static struct option __record_options[] = { "file", "vmlinux pathname"),OPT_BOOLEAN (0 ,"buildid-all" , &record.buildid_all,"Record build-id of all DSOs regardless of hits" ), -OPT_BOOLEAN (0 ,"buildid-mmap" , &record.buildid_mmap, -"Record build-id in map events" ), +OPT_BOOLEAN_SET (0 ,"buildid-mmap" , &record.buildid_mmap, &record.buildid_mmap_set, +"Legacy record build-id in map events option which is now the default. Behaves indentically to --no-buildid. Disable with --no-buildid-mmap" ),OPT_BOOLEAN (0 ,"timestamp-filename" , &record.timestamp_filename,"append timestamp to output filename" ), OPT_BOOLEAN(0, "timestamp-boundary", &record.timestamp_boundary,@@ -4042,19 +4047,23 @@ int cmd_record(int argc, const char **argv) record.opts.record_switch_events = true; } +if (!rec->buildid_mmap) { +pr_debug ("Disabling build id in synthesized mmap2 events.\n" ); + symbol_conf.no_buildid_mmap2 = true; + }else if (rec->buildid_mmap_set) { +/* +* Explicitly passing --buildid-mmap disables buildid processing +* and cache generation. +*/ + rec->no_buildid = true; + } +if (rec->buildid_mmap && !perf_can_record_build_id ()) { +pr_warning ("Missing support for build id in kernel mmap events. Disable this warning with --no-buildid-mmap\n" ); + rec->buildid_mmap = false; + }if (rec->buildid_mmap) { -if (!perf_can_record_build_id ()) { -pr_err ("Failed: no support to record build id in mmap events, update your kernel.\n" ); - err = -EINVAL ; - goto out_opts; - } -pr_debug ("Enabling build id in mmap2 events.\n" ); -/* Enable mmap build id synthesizing. */ - symbol_conf.buildid_mmap2 = true;/* Enable perf_event_attr::build_id bit. */ rec->opts.build_id = true; -/* Disable build id cache. */ - rec->no_buildid = true; }if (rec->opts.record_cgroup && !perf_can_record_cgroup ()) {tools/perf/util/symbol_conf.h +1
-1@@ -43,7 +43,7 @@ struct symbol_conf { report_individual_block, inline_name, disable_add2line_warn, - buildid_mmap2, + no_buildid_mmap2, guest_code, lazy_load_kernel_maps, keep_exited_threads,tools/perf/util/synthetic-events.c +0
-0@@ -532,7 +532,7 @@ int perf_event__synthesize_mmap_events(const struct perf_tool *tool, event->mmap2.pid = tgid; event->mmap2.tid = pid;
Quoted 4 lines
tools/perf/util/synthetic-events.c int perf_event__synthesize_mmap_events(const struct perf_tool *tool, · line 535
Quoted 2 lines
tools/perf/util/synthetic-events.c +1
-1-if (symbol_conf.buildid_mmap2) +if (!symbol_conf.no_buildid_mmap2)I find the double negation confusing. Can we keep it positive?
Agreed double negation is broadly bad. Here I changed buildid_mmap2 to no_buildid_mmap2 to convey that using build ID with mmap2 events was the default and needed opting out of. Getting rid of the 'no_' is of course possible, it will mean all symbol_confs will need to initialize the value to true (which has minor binary size implications), I also think it makes the intention of the code harder to understand.
Thanks,
Ian
Quoted 52 lines
Thanks,
NamhyungQuoted 48 lines
tools/perf/util/synthetic-events.c +7
-8perf_record_mmap2__read_build_id (&event->mmap2, machine, false);if (perf_tool__process_synth_event (tool, event, machine, process) !=0 ) {@@ -690,7 +690,7 @@ static int perf_event__synthesize_modules_maps_cb(struct map *map, void *data)return 0 ; dso =map__dso (map); -if (symbol_conf.buildid_mmap2) { +if (!symbol_conf.no_buildid_mmap2) { size =PERF_ALIGN (dso__long_name_len (dso) +1 ,sizeof (u64)); event->mmap2.header.type =PERF_RECORD_MMAP2 ; event->mmap2.header.size = (sizeof(event->mmap2) -@@ -734,9 +734,9 @@ int perf_event__synthesize_modules(const struct perf_tool *tool, perf_event__han .process = process, .machine = machine, }; -size_t size = symbol_conf.buildid_mmap2 - ?sizeof (args.event->mmap2) - :sizeof (args.event->mmap); +size_t size = symbol_conf.no_buildid_mmap2 + ?sizeof (args.event->mmap) + :sizeof (args.event->mmap2); args.event =zalloc (size + machine->id_hdr_size);if (args.event ==NULL ) {@@ -1124,8 +1124,8 @@ static int __perf_event__synthesize_kernel_mmap(const struct perf_tool *tool, struct machine *machine) {union perf_event *event; -size_t size = symbol_conf.buildid_mmap2 ? -sizeof (event->mmap2) :sizeof (event->mmap); +size_t size = symbol_conf.no_buildid_mmap2 ? +sizeof (event->mmap) :sizeof (event->mmap2);struct map *map =machine__kernel_map (machine);struct kmap *kmap;int err;@@ -1159,7 +1159,7 @@ static int __perf_event__synthesize_kernel_mmap(const struct perf_tool *tool, event->header.misc =PERF_RECORD_MISC_GUEST_KERNEL ; } -if (symbol_conf.buildid_mmap2) { +if (!symbol_conf.no_buildid_mmap2) { size =snprintf (event->mmap2.filename,sizeof (event->mmap2.filename),"%s%s" , machine->mmap_name, kmap->ref_reloc_sym->name) +1 ; size =PERF_ALIGN (size,sizeof (u64)); --2.49.0.901.g37484f566f-goog
Namhyung Kim
maintainerRe: [PATCH v3 9/9] perf record: Make --buildid-mmap the default
Quoted 135 lines · Ian Rogers
Quoted 129 lines · Namhyung Kim
Quoted 127 lines · Ian Rogers
Support for build IDs in mmap2 perf events has been present since Linux v5.12:
https://lore.kernel.org/lkml/[email protected]/
Build ID mmap events don't avoid the need to inject build IDs for DSO touched by samples as the build ID cache is populated by perf record. They can avoid some cases of symbol mis-resolution caused by the file system changing from when a sample occurred and when the DSO is sought. To disable build ID scanningUnlike the --buildid-mmap option, this doesn't disable the build ID cache but it does disable the processing of samples looking for DSOs to inject build IDs for. To disable the build ID cache the -B (--no-buildid) option should be used.
Making this option the default was raised on the list in:
https://lore.kernel.org/linux-perf-users/CAP-5=fXP7jN_QrGUcd55_QH5J-Y-FCaJ6=NaHVtyx0oyNh8_-Q@mail.gmail.com/Signed-off-by Ian Rogers <[email protected]>--- tools/perf/builtin-record.c | 33 +++++++++++++++++++----------- tools/perf/util/symbol_conf.h | 2 +- tools/perf/util/synthetic-events.c | 16 +++++++-------- 3 files changed, 30 insertions(+), 21 deletions(-)tools/perf/builtin-record.c +21
-12@@ -169,6 +169,7 @@ struct record {bool no_buildid_cache_set;bool buildid_all;bool buildid_mmap; +bool buildid_mmap_set;bool timestamp_filename;bool timestamp_boundary;bool off_cpu;@@ -1795,6 +1796,7 @@ record__finish_output(struct record *rec) data->dir.files[i].size =lseek (data->dir.files[i].fd,0 ,SEEK_CUR ); } +/* Buildid scanning disabled or build ID in kernel and synthesized map events. */ if (!rec->no_buildid) {process_buildids (rec);@@ -2966,6 +2968,8 @@ static int perf_record_config(const char *var, const char *value, void *cb) rec->no_buildid = true; else if (!strcmp(value, "mmap")) rec->buildid_mmap = true; + else if (!strcmp(value, "no-mmap")) + rec->buildid_mmap = false; else return -1;return 0 ;@@ -3349,6 +3353,7 @@ static struct record record = { .ctl_fd_ack = -1, .synth = PERF_SYNTH_ALL, }, + .buildid_mmap = true, };const char record_callchain_help[] =CALLCHAIN_RECORD_HELP @@ -3514,8 +3519,8 @@ static struct option __record_options[] = { "file", "vmlinux pathname"),OPT_BOOLEAN (0 ,"buildid-all" , &record.buildid_all,"Record build-id of all DSOs regardless of hits" ), -OPT_BOOLEAN (0 ,"buildid-mmap" , &record.buildid_mmap, -"Record build-id in map events" ), +OPT_BOOLEAN_SET (0 ,"buildid-mmap" , &record.buildid_mmap, &record.buildid_mmap_set, +"Legacy record build-id in map events option which is now the default. Behaves indentically to --no-buildid. Disable with --no-buildid-mmap" ),OPT_BOOLEAN (0 ,"timestamp-filename" , &record.timestamp_filename,"append timestamp to output filename" ), OPT_BOOLEAN(0, "timestamp-boundary", &record.timestamp_boundary,@@ -4042,19 +4047,23 @@ int cmd_record(int argc, const char **argv) record.opts.record_switch_events = true; } +if (!rec->buildid_mmap) { +pr_debug ("Disabling build id in synthesized mmap2 events.\n" ); + symbol_conf.no_buildid_mmap2 = true; + }else if (rec->buildid_mmap_set) { +/* +* Explicitly passing --buildid-mmap disables buildid processing +* and cache generation. +*/ + rec->no_buildid = true; + } +if (rec->buildid_mmap && !perf_can_record_build_id ()) { +pr_warning ("Missing support for build id in kernel mmap events. Disable this warning with --no-buildid-mmap\n" ); + rec->buildid_mmap = false; + }if (rec->buildid_mmap) { -if (!perf_can_record_build_id ()) { -pr_err ("Failed: no support to record build id in mmap events, update your kernel.\n" ); - err = -EINVAL ; - goto out_opts; - } -pr_debug ("Enabling build id in mmap2 events.\n" ); -/* Enable mmap build id synthesizing. */ - symbol_conf.buildid_mmap2 = true;/* Enable perf_event_attr::build_id bit. */ rec->opts.build_id = true; -/* Disable build id cache. */ - rec->no_buildid = true; }if (rec->opts.record_cgroup && !perf_can_record_cgroup ()) {tools/perf/util/symbol_conf.h +1
-1@@ -43,7 +43,7 @@ struct symbol_conf { report_individual_block, inline_name, disable_add2line_warn, - buildid_mmap2, + no_buildid_mmap2, guest_code, lazy_load_kernel_maps, keep_exited_threads,tools/perf/util/synthetic-events.c +0
-0@@ -532,7 +532,7 @@ int perf_event__synthesize_mmap_events(const struct perf_tool *tool, event->mmap2.pid = tgid; event->mmap2.tid = pid;Quoted 4 lines
tools/perf/util/synthetic-events.c int perf_event__synthesize_mmap_events(const struct perf_tool *tool, · line 535
Quoted 2 lines
tools/perf/util/synthetic-events.c +1
-1-if (symbol_conf.buildid_mmap2) +if (!symbol_conf.no_buildid_mmap2)I find the double negation confusing. Can we keep it positive?
Quoted 6 lines
Agreed double negation is broadly bad. Here I changed
buildid_mmap2tono_buildid_mmap2to convey that using build ID with mmap2 events was the default and needed opting out of. Getting rid of the 'no_' is of course possible, it will mean allsymbol_confswill need to initialize the value to true (which has minor binary size implications), I also think it makes the intention of the code harder to understand.
Well I think perf record is the only place to set it. But I won't insist on it strongly. Let's go with negative if you like.
Thanks,
Namhyung
Ian Rogers
reviewerQuoted 20 lines · Ian Rogers
Build ID mmap2 events have been available since Linux v5.12 and avoid certain races. Enable these by default as discussed in:
https://lore.kernel.org/linux-perf-users/CAP-5=fXP7jN_QrGUcd55_QH5J-Y-FCaJ6=NaHVtyx0oyNh8_-Q@mail.gmail.com/The
dso_idis used to indentify a DSO that may change by being overwritten. The inode generation isn't present in/proc/pid/mapsand so was already only optionally filled in. With build ID mmap events the other major, minor and inode varialbes aren't filled in. Change thedso_idimplementation to make optional values explicit, rather than injecting adso_idwe want to improve it during find operations, add the buildid to thedso_idfor sorting and so that matching fails when build IDs vary between DSOs.Mark the callchain for buildids and not just the sample IP, fixing missing DSOs.
Fix
sample__for_each_callchain_nodeto populate the map even when symbols aren't computed.
Quoted 8 lines
Other minor bits of
build_idclean up.v3: Ensure the struct
build_idis initialized empty prior to use as
read paths may fail (Namhyung).v2: Make marking DSOs still the default even with the defaulted build
ID mmap. The command line option still disables this to avoid regressions. Add callchain patches and jitdump fix.
Ping. Thanks,
Ian
Quoted 50 lines
Ian Rogers (9):
perf callchain: Always populate theaddr_locationmap when adding IP perf build-id: Reduce size of "size" variable
perf build-id: Truncate to avoid overflowing thebuild_iddata perf build-id: Change sprintf functions to snprintf
perf build-id: Mark DSO in sample callchains
perf build-id: Ensure structbuild_idis empty before use perf dso: Movebuild_idtodso_id
perf jitdump: Directly mark the jitdump DSO
perf record: Make --buildid-mmap the defaulttools/perf/bench/inject-buildid.c | 2 +- tools/perf/builtin-buildid-cache.c | 20 ++-- tools/perf/builtin-buildid-list.c | 6 +- tools/perf/builtin-inject.c | 36 +++--- tools/perf/builtin-record.c | 33 ++++-- tools/perf/builtin-report.c | 11 +- tools/perf/include/perf/perf_dlfilter.h | 2 +- tools/perf/tests/pe-file-parsing.c | 2 +- tools/perf/tests/sdt.c | 2 +- tools/perf/tests/symbols.c | 4 +- tools/perf/util/build-id.c | 63 ++++++---- tools/perf/util/build-id.h | 8 +- tools/perf/util/debuginfo.c | 2 +- tools/perf/util/disasm.c | 2 +- tools/perf/util/dso.c | 111 ++++++++++-------- tools/perf/util/dso.h | 75 ++++++------ tools/perf/util/dsos.c | 20 ++-- tools/perf/util/event.c | 2 +- tools/perf/util/header.c | 2 +- tools/perf/util/jitdump.c | 21 +++- tools/perf/util/machine.c | 34 +++--- tools/perf/util/map.c | 15 ++- tools/perf/util/map.h | 5 +- tools/perf/util/probe-event.c | 3 +- tools/perf/util/probe-file.c | 4 +- tools/perf/util/probe-finder.c | 3 +- .../scripting-engines/trace-event-python.c | 7 +- tools/perf/util/sort.c | 27 +++-- tools/perf/util/symbol-minimal.c | 2 +- tools/perf/util/symbol.c | 7 +- tools/perf/util/symbol_conf.h | 2 +- tools/perf/util/synthetic-events.c | 44 ++++--- tools/perf/util/thread.c | 8 +- tools/perf/util/thread.h | 2 +- 34 files changed, 337 insertions(+), 250 deletions(-)--
2.49.0.901.g37484f566f-goog
Arnaldo Carvalho de Melo
maintainerQuoted 26 lines · Ian Rogers
Quoted 25 lines · Ian Rogers
Build ID mmap2 events have been available since Linux v5.12 and avoid certain races. Enable these by default as discussed in:
https://lore.kernel.org/linux-perf-users/CAP-5=fXP7jN_QrGUcd55_QH5J-Y-FCaJ6=NaHVtyx0oyNh8_-Q@mail.gmail.com/The
dso_idis used to indentify a DSO that may change by being overwritten. The inode generation isn't present in/proc/pid/mapsand so was already only optionally filled in. With build ID mmap events the other major, minor and inode varialbes aren't filled in. Change thedso_idimplementation to make optional values explicit, rather than injecting adso_idwe want to improve it during find operations, add the buildid to thedso_idfor sorting and so that matching fails when build IDs vary between DSOs.Mark the callchain for buildids and not just the sample IP, fixing missing DSOs.
Fix
sample__for_each_callchain_nodeto populate the map even when symbols aren't computed.Other minor bits of
build_idclean up.v3: Ensure the struct
build_idis initialized empty prior to use as
read paths may fail (Namhyung).
Quoted 6 lines
Quoted 3 lines
v2: Make marking DSOs still the default even with the defaulted build
ID mmap. The command line option still disables this to avoid regressions. Add callchain patches and jitdump fix.Ping. Thanks,
Ian
⬢ [acme@toolbx perf-tools-next]$ git am ./v3_20250428_irogers_perf_default_use_of_build_ids_and_improvements.mbx
Applying: perf callchain: Always populate the addr_location map when adding IP
Applying: perf build-id: Reduce size of "size" variable
Applying: perf build-id: Truncate to avoid overflowing the build_id data
Applying: perf build-id: Change sprintf functions to snprintf
Applying: perf build-id: Mark DSO in sample callchains
Applying: perf build-id: Ensure struct build_id is empty before use
Applying: perf dso: Move build_id to dso_id
Applying: perf jitdump: Directly mark the jitdump DSO
Applying: perf record: Make --buildid-mmap the default
error: patch failed: tools/perf/builtin-record.c:3349
error: tools/perf/builtin-record.c: patch does not apply
Patch failed at 0009 perf record: Make --buildid-mmap the default
hint: Use 'git am --show-current-patch=diff' to see the failed patch
hint: When you have resolved this problem, run "git am --continue".
hint: If you prefer to skip this patch, run "git am --skip" instead.
hint: To restore the original branch and stop patching, run "git am --abort".
hint: Disable this message with "git config set advice.mergeConflict false" ⬢ [acme@toolbx perf-tools-next]$
- Arnaldo
Ian Rogers
reviewerQuoted 46 lines · Arnaldo Carvalho de Melo
Quoted 26 lines · Ian Rogers
Quoted 25 lines · Ian Rogers
Build ID mmap2 events have been available since Linux v5.12 and avoid certain races. Enable these by default as discussed in:
https://lore.kernel.org/linux-perf-users/CAP-5=fXP7jN_QrGUcd55_QH5J-Y-FCaJ6=NaHVtyx0oyNh8_-Q@mail.gmail.com/The
dso_idis used to indentify a DSO that may change by being overwritten. The inode generation isn't present in/proc/pid/mapsand so was already only optionally filled in. With build ID mmap events the other major, minor and inode varialbes aren't filled in. Change thedso_idimplementation to make optional values explicit, rather than injecting adso_idwe want to improve it during find operations, add the buildid to thedso_idfor sorting and so that matching fails when build IDs vary between DSOs.Mark the callchain for buildids and not just the sample IP, fixing missing DSOs.
Fix
sample__for_each_callchain_nodeto populate the map even when symbols aren't computed.Other minor bits of
build_idclean up.v3: Ensure the struct
build_idis initialized empty prior to use as
read paths may fail (Namhyung).Quoted 6 lines
Quoted 3 lines
v2: Make marking DSOs still the default even with the defaulted build
ID mmap. The command line option still disables this to avoid regressions. Add callchain patches and jitdump fix.Ping. Thanks,
Ian⬢ [acme@toolbx perf-tools-next]$ git am ./v3_20250428_irogers_perf_default_use_of_build_ids_and_improvements.mbx
Applying: perf callchain: Always populate theaddr_locationmap when adding IP
Applying: perf build-id: Reduce size of "size" variable
Applying: perf build-id: Truncate to avoid overflowing thebuild_iddata
Applying: perf build-id: Change sprintf functions to snprintf
Applying: perf build-id: Mark DSO in sample callchains
Applying: perf build-id: Ensure structbuild_idis empty before use
Applying: perf dso: Movebuild_idtodso_id
Applying: perf jitdump: Directly mark the jitdump DSO
Applying: perf record: Make --buildid-mmap the default
error: patch failed:tools/perf/builtin-record.c:3349
Quoted 8 lines
error:
tools/perf/builtin-record.c: patch does not apply
Patch failed at 0009 perf record: Make --buildid-mmap the default
hint: Use 'git am --show-current-patch=diff' to see the failed patch
hint: When you have resolved this problem, run "git am --continue".
hint: If you prefer to skip this patch, run "git am --skip" instead.
hint: To restore the original branch and stop patching, run "git am --abort".
hint: Disable this message with "git config set advice.mergeConflict false" ⬢ [acme@toolbx perf-tools-next]$
Thanks, I'll send the rebase in v4. I saw you had a branch on perf-tools-next.git.
Thanks,
Ian
Quoted 1 line
- Arnaldo