perf top: Decay all events in the evlist
Namhyung Kim
maintainerPATCH 1/2 perf top: Decay all events in the evlist
Currently perf top only decays entries in a selected evsel. I don't know whether it's intended (maybe due to performance reason?) but anyway it might show incorrect output when event group is used since users will see leader event is decayed but others are not.
This patch moves the decay code into evlist__resort_hists() so that stdio and tui code shared the logic.
--- tools/perf/builtin-top.c | 38 +++++++++++++------------------------- 1 file changed, 13 insertions(+), 25 deletions(-)
tools/perf/builtin-top.c +13 -25
@@ -264,13 +264,23 @@ static void perf_top__show_details(struct perf_top *top)pthread_mutex_unlock (¬es->lock); } -static void evlist__resort_hists (struct evlist *evlist) +static void evlist__resort_hists (struct perf_top *t) { +struct evlist *evlist = t->evlist;struct evsel *pos;evlist__for_each_entry (evlist, pos) {struct hists *hists =evsel__hists (pos); +if (evlist->enabled) { +if (t->zero) { +hists__delete_entries (hists); + }else { +hists__decay_entries (hists, t->hide_user_symbols, + t->hide_kernel_symbols); + } + } +hists__collapse_resort (hists,NULL );/* Non-group events are considered as leader */
@@ -319,16 +329,7 @@ static void perf_top__print_sym_table(struct perf_top *top)return ; } -if (top->evlist->enabled) { -if (top->zero) { -hists__delete_entries (hists); - }else { -hists__decay_entries (hists, top->hide_user_symbols, - top->hide_kernel_symbols); - } - } - -evlist__resort_hists (top->evlist); +evlist__resort_hists (top);hists__output_recalc_col_len (hists, top->print_entries - printed);putchar ('\n' );
@@ -576,24 +577,11 @@ static bool perf_top__handle_keypress(struct perf_top *top, int c)static void perf_top__sort_new_samples (void *arg) {struct perf_top *t = arg; -struct evsel *evsel = t->sym_evsel; -struct hists *hists;if (t->evlist->selected !=NULL ) t->sym_evsel = t->evlist->selected; - hists =evsel__hists (evsel); - -if (t->evlist->enabled) { -if (t->zero) { -hists__delete_entries (hists); - }else { -hists__decay_entries (hists, t->hide_user_symbols, - t->hide_kernel_symbols); - } - } - -evlist__resort_hists (t->evlist); +evlist__resort_hists (t);if (t->lost || t->drop)pr_warning ("Too slow to read ring buffer (change period (-c/-F) or limit CPUs (-C)\n" );
Signature · 1 line
2.23.0.187.g17f5b7556c-goog
Namhyung Kim
maintainerPATCH 2/2 perf top: Fix event group with more than two events
The event group feature links relevant hist entries among events so that they can be displayed together. During the link process, each hist entry in non-leader events is connected to a hist entry in the leader event. This is done in order of events specified in the command line so it assumes that events are linked in the order.
But perf top can break the assumption since it does the link process multiple times. For example, a hist entry can be in the third event only at first so it's linked after the leader. Some time later, second event has a hist entry for it and it'll be linked after the entry of the third event.
This makes the code compilicated to deal with such unordered entries. This patch simply unlink all the entries after it's printed so that they can assume the correct order after the repeated link process. Also it'd be easy to deal with decaying old entries IMHO.
--- tools/perf/builtin-top.c | 6 ++++++ tools/perf/util/hist.c | 39 +++++++++++++++++++++------------------ tools/perf/util/hist.h | 1 + 3 files changed, 28 insertions(+), 18 deletions(-)
tools/perf/builtin-top.c +6 -0
@@ -272,6 +272,12 @@ static void evlist__resort_hists(struct perf_top *t)evlist__for_each_entry (evlist, pos) {struct hists *hists =evsel__hists (pos); +/* +* unlink existing entries so that they can be linked +* in a correct order in hists__match() below. +*/ +hists__unlink (hists); +if (evlist->enabled) {if (t->zero) {hists__delete_entries (hists);
tools/perf/util/hist.c +21 -18
@@ -2436,7 +2436,7 @@ void hists__match(struct hists *leader, struct hists *other) {struct rb_root_cached *root;struct rb_node *nd; -struct hist_entry *pos, *pair, *pos_pair, *tmp_pair; +struct hist_entry *pos, *pair;if (symbol_conf.report_hierarchy) {/* hierarchy report always collapses entries */
@@ -2453,24 +2453,8 @@ void hists__match(struct hists *leader, struct hists *other) pos = rb_entry(nd, struct hist_entry, rb_node_in); pair =hists__find_entry (other, pos); -if (pair &&list_empty (&pair->pairs.node)) { -list_for_each_entry_safe (pos_pair, tmp_pair, &pos->pairs.head, pairs.node) { -if (pos_pair->hists == other) { -/* -* XXX maybe decayed entries can appear -* here? but then we would have use -* after free, as decayed entries are -* freed see hists__delete_entry -*/ -BUG_ON (!pos_pair->dummy); -list_del_init (&pos_pair->pairs.node); -hist_entry__delete (pos_pair); -break ; - } - } - +if (pair)hist_entry__add_pair (pair, pos); - } } }
@@ -2555,6 +2539,25 @@ int hists__link(struct hists *leader, struct hists *other)return 0 ; } +int hists__unlink (struct hists *hists) +{ +struct rb_root_cached *root; +struct rb_node *nd; +struct hist_entry *pos; + +if (hists__has (hists, need_collapse)) + root = &hists->entries_collapsed; +else + root = hists->entries_in; + +for (nd =rb_first_cached (root); nd; nd =rb_next (nd)) { + pos = rb_entry(nd, struct hist_entry, rb_node_in); +list_del_init (&pos->pairs.node); + } + +return 0 ; +} +void hist__account_cycles (struct branch_stack *bs,struct addr_location *al,struct perf_sample *sample,bool nonany_branch_mode) {
tools/perf/util/hist.h +1 -0
@@ -217,6 +217,7 @@ void hists__calc_col_len(struct hists *hists, struct hist_entry *he);void hists__match (struct hists *leader,struct hists *other);int hists__link (struct hists *leader,struct hists *other); +int hists__unlink (struct hists *hists);struct hists_evsel {struct evsel evsel;
Signature · 1 line
2.23.0.187.g17f5b7556c-goog
Re: [PATCH 2/2] perf top: Fix event group with more than two events
Em Wed, Aug 28, 2019 at 08:15:55AM +0900, Namhyung Kim escreveu:
Quoted 12 lines
The event group feature links relevant hist entries among events so that they can be displayed together. During the link process, each hist entry in non-leader events is connected to a hist entry in the leader event. This is done in order of events specified in the command line so it assumes that events are linked in the order.
But perf top can break the assumption since it does the link process multiple times. For example, a hist entry can be in the third event only at first so it's linked after the leader. Some time later, second event has a hist entry for it and it'll be linked after the entry of the third event.
Quoted 4 lines
This makes the code compilicated to deal with such unordered entries. This patch simply unlink all the entries after it's printed so that they can assume the correct order after the repeated link process. Also it'd be easy to deal with decaying old entries IMHO.
Excellent, I just tested it with:
# perf top --show-total-period -e '{cycles,instructions,cache-references,cache-misses}'And the total period goes down for all evsels when that symbol is quiescent between screen refreshes, and no crash, all seems to be working as expected, thanks for fixing this!
Also when pressing 'A' the annotation also seems to be working.
All applied.
- Arnaldo
Quoted 103 lines
Signed-off-by Namhyung Kim <[email protected]>--- tools/perf/builtin-top.c | 6 ++++++ tools/perf/util/hist.c | 39 +++++++++++++++++++++------------------ tools/perf/util/hist.h | 1 + 3 files changed, 28 insertions(+), 18 deletions(-)tools/perf/builtin-top.c +6
-0@@ -272,6 +272,12 @@ static void evlist__resort_hists(struct perf_top *t)evlist__for_each_entry (evlist, pos) {struct hists *hists =evsel__hists (pos); +/* +* unlink existing entries so that they can be linked +* in a correct order in hists__match() below. +*/ +hists__unlink (hists); +if (evlist->enabled) {if (t->zero) {hists__delete_entries (hists);tools/perf/util/hist.c +21
-18@@ -2436,7 +2436,7 @@ void hists__match(struct hists *leader, struct hists *other) {struct rb_root_cached *root;struct rb_node *nd; -struct hist_entry *pos, *pair, *pos_pair, *tmp_pair; +struct hist_entry *pos, *pair;if (symbol_conf.report_hierarchy) {/* hierarchy report always collapses entries */ @@ -2453,24 +2453,8 @@ void hists__match(struct hists *leader, struct hists *other) pos = rb_entry(nd, struct hist_entry, rb_node_in); pair =hists__find_entry (other, pos); -if (pair &&list_empty (&pair->pairs.node)) { -list_for_each_entry_safe (pos_pair, tmp_pair, &pos->pairs.head, pairs.node) { -if (pos_pair->hists == other) { -/* -* XXX maybe decayed entries can appear -* here? but then we would have use -* after free, as decayed entries are -* freed see hists__delete_entry -*/ -BUG_ON (!pos_pair->dummy); -list_del_init (&pos_pair->pairs.node); -hist_entry__delete (pos_pair); -break ; - } - } - +if (pair)hist_entry__add_pair (pair, pos); - } } }@@ -2555,6 +2539,25 @@ int hists__link(struct hists *leader, struct hists *other)return 0 ; } +int hists__unlink (struct hists *hists) +{ +struct rb_root_cached *root; +struct rb_node *nd; +struct hist_entry *pos; + +if (hists__has (hists, need_collapse)) + root = &hists->entries_collapsed; +else + root = hists->entries_in; + +for (nd =rb_first_cached (root); nd; nd =rb_next (nd)) { + pos = rb_entry(nd, struct hist_entry, rb_node_in); +list_del_init (&pos->pairs.node); + } + +return 0 ; +} +void hist__account_cycles (struct branch_stack *bs,struct addr_location *al,struct perf_sample *sample,bool nonany_branch_mode) {tools/perf/util/hist.h +1
-0@@ -217,6 +217,7 @@ void hists__calc_col_len(struct hists *hists, struct hist_entry *he);void hists__match (struct hists *leader,struct hists *other);int hists__link (struct hists *leader,struct hists *other); +int hists__unlink (struct hists *hists);struct hists_evsel {struct evsel evsel;Signature · 1 line
2.23.0.187.g17f5b7556c-goog
Signature · 1 line
- Arnaldo
[tip: perf/core] perf top: Fix event group with more than two events
In reply to Namhyung KimThe following commit has been merged into the perf/core branch of tip:
The following commit has been merged into the perf/core branch of tip:
Commit-ID: be5863b7d9281bbb932542d16b7d758357fde267 Gitweb: https://git.kernel.org/tip/be5863b7d9281bbb932542d16b7d758357fde267 Author: Namhyung Kim <[email protected]> AuthorDate: Wed, 28 Aug 2019 08:15:55 +09:00 Committer: Arnaldo Carvalho de Melo <[email protected]> CommitterDate: Wed, 28 Aug 2019 18:15:03 -03:00
perf top: Fix event group with more than two events
The event group feature links relevant hist entries among events so that they can be displayed together. During the link process, each hist entry in non-leader events is connected to a hist entry in the leader event. This is done in order of events specified in the command line so it assumes that events are linked in the order.
But 'perf top' can break the assumption since it does the link process multiple times. For example, a hist entry can be in the third event only at first so it's linked after the leader. Some time later, second event has a hist entry for it and it'll be linked after the entry of the third event.
This makes the code compilicated to deal with such unordered entries. This patch simply unlink all the entries after it's printed so that they can assume the correct order after the repeated link process. Also it'd be easy to deal with decaying old entries IMHO.
--- tools/perf/builtin-top.c | 6 ++++++- tools/perf/util/hist.c | 39 +++++++++++++++++++++------------------ tools/perf/util/hist.h | 1 +- 3 files changed, 28 insertions(+), 18 deletions(-)
tools/perf/builtin-top.c +6 -0
@@ -273,6 +273,12 @@ static void perf_top__resort_hists(struct perf_top *t)evlist__for_each_entry (evlist, pos) {struct hists *hists =evsel__hists (pos); +/* +* unlink existing entries so that they can be linked +* in a correct order in hists__match() below. +*/ +hists__unlink (hists); +if (evlist->enabled) {if (t->zero) {hists__delete_entries (hists);
tools/perf/util/hist.c +21 -18
@@ -2439,7 +2439,7 @@ void hists__match(struct hists *leader, struct hists *other) {struct rb_root_cached *root;struct rb_node *nd; -struct hist_entry *pos, *pair, *pos_pair, *tmp_pair; +struct hist_entry *pos, *pair;if (symbol_conf.report_hierarchy) {/* hierarchy report always collapses entries */
@@ -2456,24 +2456,8 @@ void hists__match(struct hists *leader, struct hists *other) pos = rb_entry(nd, struct hist_entry, rb_node_in); pair =hists__find_entry (other, pos); -if (pair &&list_empty (&pair->pairs.node)) { -list_for_each_entry_safe (pos_pair, tmp_pair, &pos->pairs.head, pairs.node) { -if (pos_pair->hists == other) { -/* -* XXX maybe decayed entries can appear -* here? but then we would have use -* after free, as decayed entries are -* freed see hists__delete_entry -*/ -BUG_ON (!pos_pair->dummy); -list_del_init (&pos_pair->pairs.node); -hist_entry__delete (pos_pair); -break ; - } - } - +if (pair)hist_entry__add_pair (pair, pos); - } } }
@@ -2558,6 +2542,25 @@ int hists__link(struct hists *leader, struct hists *other)return 0 ; } +int hists__unlink (struct hists *hists) +{ +struct rb_root_cached *root; +struct rb_node *nd; +struct hist_entry *pos; + +if (hists__has (hists, need_collapse)) + root = &hists->entries_collapsed; +else + root = hists->entries_in; + +for (nd =rb_first_cached (root); nd; nd =rb_next (nd)) { + pos = rb_entry(nd, struct hist_entry, rb_node_in); +list_del_init (&pos->pairs.node); + } + +return 0 ; +} +void hist__account_cycles (struct branch_stack *bs,struct addr_location *al,struct perf_sample *sample,bool nonany_branch_mode) {
tools/perf/util/hist.h +1 -0
@@ -217,6 +217,7 @@ void hists__calc_col_len(struct hists *hists, struct hist_entry *he);void hists__match (struct hists *leader,struct hists *other);int hists__link (struct hists *leader,struct hists *other); +int hists__unlink (struct hists *hists);struct hists_evsel {struct evsel evsel;
Em Wed, Aug 28, 2019 at 08:15:54AM +0900, Namhyung Kim escreveu:
Quoted 18 lines
Currently perf top only decays entries in a selected evsel. I don't know whether it's intended (maybe due to performance reason?) but anyway it might show incorrect output when event group is used since users will see leader event is decayed but others are not.
This patch moves the decay code into
evlist__resort_hists()so that stdio and tui code shared the logic.Signed-off-by Namhyung Kim <[email protected]>--- tools/perf/builtin-top.c | 38 +++++++++++++------------------------- 1 file changed, 13 insertions(+), 25 deletions(-)tools/perf/builtin-top.c +0
-0
tools/perf/builtin-top.c static void perf_top__show_details(struct perf_top *top) · line 267
Quoted 5 lines
tools/perf/builtin-top.c +1
-1@@ -264,13 +264,23 @@ static void perf_top__show_details(struct perf_top *top)pthread_mutex_unlock (¬es->lock); } -static void evlist__resort_hists (struct evlist *evlist) +static void evlist__resort_hists (struct perf_top *t)
Since this now operates on the perf_top struct, I'll rename it to perf_top__resort_hists(), ok? No need to send an updated patch.
- Arnaldo
Quoted 65 lines
tools/perf/builtin-top.c +12
-24{ +struct evlist *evlist = t->evlist;struct evsel *pos;evlist__for_each_entry (evlist, pos) {struct hists *hists =evsel__hists (pos); +if (evlist->enabled) { +if (t->zero) { +hists__delete_entries (hists); + }else { +hists__decay_entries (hists, t->hide_user_symbols, + t->hide_kernel_symbols); + } + } +hists__collapse_resort (hists,NULL );/* Non-group events are considered as leader */ @@ -319,16 +329,7 @@ static void perf_top__print_sym_table(struct perf_top *top)return ; } -if (top->evlist->enabled) { -if (top->zero) { -hists__delete_entries (hists); - }else { -hists__decay_entries (hists, top->hide_user_symbols, - top->hide_kernel_symbols); - } - } - -evlist__resort_hists (top->evlist); +evlist__resort_hists (top);hists__output_recalc_col_len (hists, top->print_entries - printed);putchar ('\n' );@@ -576,24 +577,11 @@ static bool perf_top__handle_keypress(struct perf_top *top, int c)static void perf_top__sort_new_samples (void *arg) {struct perf_top *t = arg; -struct evsel *evsel = t->sym_evsel; -struct hists *hists;if (t->evlist->selected !=NULL ) t->sym_evsel = t->evlist->selected; - hists =evsel__hists (evsel); - -if (t->evlist->enabled) { -if (t->zero) { -hists__delete_entries (hists); - }else { -hists__decay_entries (hists, t->hide_user_symbols, - t->hide_kernel_symbols); - } - } - -evlist__resort_hists (t->evlist); +evlist__resort_hists (t);if (t->lost || t->drop)pr_warning ("Too slow to read ring buffer (change period (-c/-F) or limit CPUs (-C)\n" );Signature · 1 line
2.23.0.187.g17f5b7556c-goog
Signature · 1 line
- Arnaldo
Namhyung Kim
maintainerHi Arnaldo,
Quoted 23 lines · Arnaldo Carvalho de Melo
Em Wed, Aug 28, 2019 at 08:15:54AM +0900, Namhyung Kim escreveu:
Quoted 21 lines
Currently perf top only decays entries in a selected evsel. I don't know whether it's intended (maybe due to performance reason?) but anyway it might show incorrect output when event group is used since users will see leader event is decayed but others are not.
This patch moves the decay code into
evlist__resort_hists()so that stdio and tui code shared the logic.Signed-off-by Namhyung Kim <[email protected]>--- tools/perf/builtin-top.c | 38 +++++++++++++------------------------- 1 file changed, 13 insertions(+), 25 deletions(-)tools/perf/builtin-top.c +0
-0@@ -264,13 +264,23 @@ static void perf_top__show_details(struct perf_top *top)pthread_mutex_unlock (¬es->lock); }
Quoted 5 lines
tools/perf/builtin-top.c static void perf_top__show_details(struct perf_top *top) · line 267
Quoted 2 lines
tools/perf/builtin-top.c +1
-1-static void evlist__resort_hists (struct evlist *evlist) +static void evlist__resort_hists (struct perf_top *t)Since this now operates on the
perf_topstruct, I'll rename it toperf_top__resort_hists(), ok? No need to send an updated patch.
Right. Thanks for doing this!
Namhyung
The following commit has been merged into the perf/core branch of tip:
The following commit has been merged into the perf/core branch of tip:
Commit-ID: ea4385f804eadce3f4fd8698d4ffd9e85fb6d5e0 Gitweb: https://git.kernel.org/tip/ea4385f804eadce3f4fd8698d4ffd9e85fb6d5e0 Author: Namhyung Kim <[email protected]> AuthorDate: Wed, 28 Aug 2019 08:15:54 +09:00 Committer: Arnaldo Carvalho de Melo <[email protected]> CommitterDate: Wed, 28 Aug 2019 18:15:03 -03:00
perf top: Decay all events in the evlist
Currently perf top only decays entries in a selected evsel. I don't know whether it's intended (maybe due to performance reason?) but anyway it might show incorrect output when event group is used since users will see leader event is decayed but others are not.
This patch moves the decay code into perf_top__resort_hists() so that stdio and TUI code shared the logic.
--- tools/perf/builtin-top.c | 38 +++++++++++++------------------------- 1 file changed, 13 insertions(+), 25 deletions(-)
tools/perf/builtin-top.c +13 -25
@@ -265,13 +265,23 @@ static void perf_top__show_details(struct perf_top *top)pthread_mutex_unlock (¬es->lock); } -static void evlist__resort_hists (struct evlist *evlist) +static void perf_top__resort_hists (struct perf_top *t) { +struct evlist *evlist = t->evlist;struct evsel *pos;evlist__for_each_entry (evlist, pos) {struct hists *hists =evsel__hists (pos); +if (evlist->enabled) { +if (t->zero) { +hists__delete_entries (hists); + }else { +hists__decay_entries (hists, t->hide_user_symbols, + t->hide_kernel_symbols); + } + } +hists__collapse_resort (hists,NULL );/* Non-group events are considered as leader */
@@ -320,16 +330,7 @@ static void perf_top__print_sym_table(struct perf_top *top)return ; } -if (top->evlist->enabled) { -if (top->zero) { -hists__delete_entries (hists); - }else { -hists__decay_entries (hists, top->hide_user_symbols, - top->hide_kernel_symbols); - } - } - -evlist__resort_hists (top->evlist); +perf_top__resort_hists (top);hists__output_recalc_col_len (hists, top->print_entries - printed);putchar ('\n' );
@@ -577,24 +578,11 @@ static bool perf_top__handle_keypress(struct perf_top *top, int c)static void perf_top__sort_new_samples (void *arg) {struct perf_top *t = arg; -struct evsel *evsel = t->sym_evsel; -struct hists *hists;if (t->evlist->selected !=NULL ) t->sym_evsel = t->evlist->selected; - hists =evsel__hists (evsel); - -if (t->evlist->enabled) { -if (t->zero) { -hists__delete_entries (hists); - }else { -hists__decay_entries (hists, t->hide_user_symbols, - t->hide_kernel_symbols); - } - } - -evlist__resort_hists (t->evlist); +perf_top__resort_hists (t);if (t->lost || t->drop)pr_warning ("Too slow to read ring buffer (change period (-c/-F) or limit CPUs (-C)\n" );