Skip to contentSkip to conversation

perf tools: Add traceevent plugins support

PATCH Superseded Jiri Olsa 3 participants 77 messages Nov 21, 2013 – Nov 27, 2013 1 message not in the archive perf events

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Jiri Olsa

author
#0 lore raw

cover perf tools: Add traceevent plugins support

hi,
backporting traceevent plugin support from trace-cmd.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

Initial RFC is here:
http://marc.info/?l=linux-kernel&m=137803257921043&w=2

Having plugins allow traceevent to properly parse 'print fmt' line of tracepoint format and display proper/additional data in perf script command (check examples below).

Plugins backported from trace-cmd:
plugin_function.c
plugin_hrtimer.c
plugin_jbd2.c
plugin_kmem.c
plugin_kvm.c
plugin_mac80211.c
plugin_sched_switch.c

Plugins newly added:
plugin_cfg80211.c
plugin_scsi.c
plugin_xen.c

Changes from RFC:
- added comment about repo/HEAD of the backport source
- ommited options plugin support
- change plugins installation path to:

    $(perfexecdir)/traceevent/plugins

The code is available in here:
git://git.kernel.org/pub/scm/linux/kernel/git/jolsa/perf.git
perf/core_plugins

thanks,
jirka

Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>

Examples (diffs of 'perf script' output):

* data was generated by 'perf record -e 'cfg80211:*' -a' --- script.cfg80211.old
+++ script.cfg80211.new
- ifconfig 2705 [003] 662.896560: cfg80211:cfg80211_tx_mlme_mgmt: [FAILED TO PARSE] name=wlan0 ifindex=3 frame=ARRAY[c0, 00,
+ ifconfig 2705 [003] 662.896560: cfg80211:cfg80211_tx_mlme_mgmt: netdev:wlan0(3), ftype:0xc0
- kworker/u16:0 1697 [002] 664.808210: cfg80211:cfg80211_rx_mlme_mgmt: [FAILED TO PARSE] name=wlan0 ifindex=3 frame=ARRAY[b0, 00,
+ kworker/u16:0 1697 [002] 664.808210: cfg80211:cfg80211_rx_mlme_mgmt: netdev:wlan0(3), ftype:0xb0

* data was generated by 'perf record -e sched:sched_switch -a' --- script.sched_switch.old
+++ script.sched_switch.new
- perf 577 [002] 30965.311852: sched:sched_switch: prev_comm=perf prev_pid=577 prev_prio=120 prev_state=S ==> next_comm=sw
+ perf 577 [002] 30965.311852: sched:sched_switch: perf:577 [120] S ==> swapper/2:0 [120]

* data was generated by 'perf record -e ftrace:function ls' --- script.function.old
+++ script.function.new
- ls 10781 [001] 32667.291379: ftrace:function: ffffffff811adb80 <-- ffffffff811afc48
- ls 10781 [001] 32667.291379: ftrace:function: ffffffff811b35d0 <-- ffffffff811adb9b
- ls 10781 [001] 32667.291380: ftrace:function: ffffffff811b3520 <-- ffffffff811b35e8
- ls 10781 [001] 32667.291380: ftrace:function: ffffffff811b2720 <-- ffffffff811b3549
- ls 10781 [001] 32667.291381: ftrace:function: ffffffff81297e10 <-- ffffffff811b356c
- ls 10781 [001] 32667.291381: ftrace:function: ffffffff81298f40 <-- ffffffff81297e2c
- ls 10781 [001] 32667.291382: ftrace:function: ffffffff81076160 <-- ffffffff811afbf0
- ls 10781 [001] 32667.291383: ftrace:function: ffffffff811c3eb0 <-- ffffffff811afbfc
- ls 10781 [001] 32667.291383: ftrace:function: ffffffff8164e100 <-- ffffffff811c3ed8
- ls 10781 [001] 32667.291384: ftrace:function: ffffffff811a5d10 <-- ffffffff811c3f53
- ls 10781 [001] 32667.291384: ftrace:function: ffffffff811e8e70 <-- ffffffff811a5d58
- ls 10781 [001] 32667.291385: ftrace:function: ffffffff811f38e0 <-- ffffffff811a5d63
- ls 10781 [001] 32667.291385: ftrace:function: ffffffff811a9ff0 <-- ffffffff811a5d6b
- ls 10781 [001] 32667.291386: ftrace:function: ffffffff811a9fa0 <-- ffffffff811aa015
- ls 10781 [001] 32667.291387: ftrace:function: ffffffff810851c0 <-- ffffffff811aa053
- ls 10781 [001] 32667.291387: ftrace:function: ffffffff81090e00 <-- ffffffff81085211
+ ls 10781 [001] 32667.291379: ftrace:function: would_dump <-- setup_new_exec
+ ls 10781 [001] 32667.291379: ftrace:function: inode_permission <-- would_dump
+ ls 10781 [001] 32667.291380: ftrace:function: __inode_permission <-- inode_permission
+ ls 10781 [001] 32667.291380: ftrace:function: generic_permission <-- __inode_permission
+ ls 10781 [001] 32667.291381: ftrace:function: security_inode_permission <-- __inode_permission
+ ls 10781 [001] 32667.291381: ftrace:function: cap_inode_permission <-- security_inode_permission
+ ls 10781 [001] 32667.291382: ftrace:function: flush_signal_handlers <-- setup_new_exec
+ ls 10781 [001] 32667.291383: ftrace:function: do_close_on_exec <-- setup_new_exec
+ ls 10781 [001] 32667.291383: ftrace:function: _raw_spin_lock <-- do_close_on_exec
+ ls 10781 [001] 32667.291384: ftrace:function: filp_close <-- do_close_on_exec
+ ls 10781 [001] 32667.291384: ftrace:function: dnotify_flush <-- filp_close
+ ls 10781 [001] 32667.291385: ftrace:function: locks_remove_posix <-- filp_close
+ ls 10781 [001] 32667.291385: ftrace:function: fput <-- filp_close
+ ls 10781 [001] 32667.291386: ftrace:function: file_sb_list_del <-- fput
+ ls 10781 [001] 32667.291387: ftrace:function: task_work_add <-- fput
+ ls 10781 [001] 32667.291387: ftrace:function: kick_process <-- task_work_add

* data was generated by 'perf record -e 'kvm:*,kvmmmu:*' -a'
--- script.kvm.old
+++ script.kvm.new
qemu-system-x86 17414 [000] 6868.995053: kvm:kvm_exit: reason EPT_VIOLATION rip 0xfff0 info 184 0 qemu-system-x86 17414 [000] 6868.995109: kvm:kvm_emulate_insn: f0000:c46b:e4 71 (real)
- qemu-system-x86 3006 [002] 10562.079422: kvmmmu:kvm_mmu_get_page: [FAILED TO PARSE] mmu_valid_gen=0x2 gfn=0 role=122884 root_count=0
+ qemu-system-x86 3006 [002] 10562.079422: kvmmmu:kvm_mmu_get_page: new sp gfn 0 0/4 q0 direct --- !pge !nxe root 0 sync
- qemu-system-x86 3006 [002] 10562.080502: kvmmmu:kvm_mmu_prepare_zap_page: [FAILED TO PARSE] mmu_valid_gen=0x2 gfn=0 role=122884 root
+ qemu-system-x86 3006 [002] 10562.080502: kvmmmu:kvm_mmu_prepare_zap_page: 0/4 q0 direct --- !pge !nxe root 1 sync qemu-system-x86 3290 [002] 10708.755312: kvmmmu:fast_page_fault: [FAILED TO PARSE] vcpu_id=0 gva=4094486080 error_code=3 sptep=0xfff
- insmod 2576 [001] 781.731666: kvmmmu:kvm_mmu_sync_page: [FAILED TO PARSE] mmu_valid_gen=0x1 gfn=2 role=24624 root_count=1
+ insmod 2576 [001] 781.731666: kvmmmu:kvm_mmu_sync_page: 3/0 q0 --- !pge !nxe root 10 unsync
- insmod 2576 [001] 781.731668: kvmmmu:kvm_mmu_unsync_page: [FAILED TO PARSE] mmu_valid_gen=0x1 gfn=2 role=24624 root_count
+ insmod 2576 [001] 781.731668: kvmmmu:kvm_mmu_unsync_page: 3/0 q0 --- !pge !nxe root 10 unsync

---
Jiri Olsa (22):
tools lib traceevent: Add plugin support
tools lib traceevent: Add plugin build support
tools lib traceevent: Add traceevent_host_bigendian function tools lib traceevent: Add jbd2 plugin
tools lib traceevent: Add hrtimer plugin
tools lib traceevent: Add kmem plugin
tools lib traceevent: Add kvm plugin
tools lib traceevent: Add mac80211 plugin
tools lib traceevent: Add sched_switch plugin
tools lib traceevent: Add function plugin
tools lib traceevent: Add xen plugin
tools lib traceevent: Add scsi plugin
tools lib traceevent: Add cfg80211 plugin
tools lib traceevent: Update kvm plugin with is_writable_pte helper tools lib traceevent: Change pevent_parse_event to return event format tools lib traceevent: Harmonize the install messages in lib-traceevent perf tools: Add build and install plugins targets
perf tools: Add udis86 disassembler feature check
perf tools: Add filename__read_str util function
perf tools: Add trace-event object
perf tools: Add trace-event global object for tracepoint interface perf tools: Overload pr_stat traceevent print function

 tools/lib/traceevent/Makefile                  |  61 ++++++++++--
 tools/lib/traceevent/event-parse.c             |  32 +++----
 tools/lib/traceevent/event-parse.h             |  21 +++-
 tools/lib/traceevent/event-plugin.c            | 202 +++++++++++++++++++++++++++++++++++++++
 tools/lib/traceevent/plugin_cfg80211.c         |  23 +++++
 tools/lib/traceevent/plugin_function.c         | 151 +++++++++++++++++++++++++++++
 tools/lib/traceevent/plugin_hrtimer.c          |  93 ++++++++++++++++++
 tools/lib/traceevent/plugin_jbd2.c             |  66 +++++++++++++
 tools/lib/traceevent/plugin_kmem.c             |  72 ++++++++++++++
 tools/lib/traceevent/plugin_kvm.c              | 436 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 tools/lib/traceevent/plugin_mac80211.c         |  95 +++++++++++++++++++
 tools/lib/traceevent/plugin_sched_switch.c     | 148 +++++++++++++++++++++++++++++
 tools/lib/traceevent/plugin_scsi.c             | 423 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 tools/lib/traceevent/plugin_xen.c              | 130 +++++++++++++++++++++++++
 tools/perf/Makefile.perf                       |  16 +++-
 tools/perf/builtin-script.c                    |   2 +-
 tools/perf/builtin-trace.c                     |   5 +-
 tools/perf/config/Makefile                     |  10 ++
 tools/perf/config/feature-checks/Makefile      |   6 +-
 tools/perf/config/feature-checks/test-all.c    |   5 +
 tools/perf/config/feature-checks/test-udis86.c |  11 +++
 tools/perf/util/debug.c                        |  16 ++++
 tools/perf/util/debug.h                        |   2 +
 tools/perf/util/evsel.c                        |  44 +--------
 tools/perf/util/header.c                       |   8 +-
 tools/perf/util/python-ext-sources             |   1 +
 tools/perf/util/session.h                      |   3 +-
 tools/perf/util/trace-event-parse.c            |  17 +---
 tools/perf/util/trace-event-read.c             |  20 ++--
 tools/perf/util/trace-event.c                  |  82 ++++++++++++++++
 tools/perf/util/trace-event.h                  |  15 ++-
 tools/perf/util/util.c                         |  41 ++++++++
 tools/perf/util/util.h                         |   1 +
 33 files changed, 2144 insertions(+), 114 deletions(-)
 create mode 100644 tools/lib/traceevent/event-plugin.c
 create mode 100644 tools/lib/traceevent/plugin_cfg80211.c
 create mode 100644 tools/lib/traceevent/plugin_function.c
 create mode 100644 tools/lib/traceevent/plugin_hrtimer.c
 create mode 100644 tools/lib/traceevent/plugin_jbd2.c
 create mode 100644 tools/lib/traceevent/plugin_kmem.c
 create mode 100644 tools/lib/traceevent/plugin_kvm.c
 create mode 100644 tools/lib/traceevent/plugin_mac80211.c
 create mode 100644 tools/lib/traceevent/plugin_sched_switch.c
 create mode 100644 tools/lib/traceevent/plugin_scsi.c
 create mode 100644 tools/lib/traceevent/plugin_xen.c
 create mode 100644 tools/perf/config/feature-checks/test-udis86.c
 create mode 100644 tools/perf/util/trace-event.c
author
#1 lore raw

PATCH 1/22 tools lib traceevent: Add plugin support

Backporting plugin support for traceevent lib.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

It's now possible to use following interface to load plugins (shared objects) to enahnce pevent object functionality.

The plugin interface/hooks are as follows:
(taken from event-parse.h comments)

- 'pevent_plugin_loader' (required)
The function name to initialized the plugin.

int pevent_plugin_loader(struct pevent *pevent)

- 'pevent_plugin_unloader' (optional)
The function called just before unloading

int pevent_plugin_unloader(void)

- 'pevent_plugin_options' (optional)
Plugin options that can be set before loading

struct plugin_option pevent_plugin_options[] = {
{
.name = "option-name",
.plugin_alias = "overide-file-name", (optional) .description = "description of option to show users",
},
{
.name = NULL,
},
};

Array must end with .name = NULL;

The plugin_alias (below) can be used to give a shorter name to access the vairable. Useful if a plugin handles more than one event.

NOTE options support is not backported yet.

- 'pevent_plugin_alias' (optional)
The name to use for finding options (uses filename if not defined)

New traceevent functions are added to search and load available plugins:

struct plugin_list* traceevent_load_plugins(struct pevent *pevent)
- loads plusing for 'struct pevent' object and returns loaded plugins list

void traceevent_unload_plugins(struct plugin_list *plugin_list);
- unload plugin list

Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/lib/traceevent/Makefile       |   6 +-
 tools/lib/traceevent/event-parse.h  |   5 +
 tools/lib/traceevent/event-plugin.c | 202 ++++++++++++++++++++++++++++++++++++
 3 files changed, 212 insertions(+), 1 deletion(-)
 create mode 100644 tools/lib/traceevent/event-plugin.c
tools/lib/traceevent/Makefile +5 -1
index fc15020..2ccb5bc 100644
@@ -180,7 +180,11 @@ $(obj)/%.o: $(src)/%.c
 %.o: $(src)/%.c
 	$(Q)$(call do_compile)
 
-PEVENT_LIB_OBJS = event-parse.o trace-seq.o parse-filter.o parse-utils.o
+PEVENT_LIB_OBJS  = event-parse.o
+PEVENT_LIB_OBJS += event-plugin.o
+PEVENT_LIB_OBJS += trace-seq.o
+PEVENT_LIB_OBJS += parse-filter.o
+PEVENT_LIB_OBJS += parse-utils.o
 PEVENT_LIB_OBJS += kbuffer-parse.o
 
 ALL_OBJS = $(PEVENT_LIB_OBJS)
tools/lib/traceevent/event-parse.h +5 -0
index 8d73d25..a288860 100644
@@ -377,6 +377,11 @@ enum pevent_errno {
 };
 #undef _PE
 
+struct plugin_list;
+
+struct plugin_list *traceevent_load_plugins(struct pevent *pevent);
+void traceevent_unload_plugins(struct plugin_list *plugin_list);
+
 struct cmdline;
 struct cmdline_list;
 struct func_map;
tools/lib/traceevent/event-plugin.c new file +202 -0

202 lines not on this page · raw message

Signature · 1 line
1.8.3.1
#2 lore raw

Re: [PATCH 01/22] tools lib traceevent: Add plugin support

Quoted 57 lines · On Thu, 21 Nov 2013 12:00:59 +0100 Jiri Olsa <[email protected]> wrote:

Backporting plugin support for traceevent lib.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

It's now possible to use following interface to load plugins (shared objects) to enahnce pevent object functionality.

The plugin interface/hooks are as follows:
(taken from event-parse.h comments)

- 'pevent_plugin_loader' (required)
The function name to initialized the plugin.

int pevent_plugin_loader(struct pevent *pevent)

- 'pevent_plugin_unloader' (optional)
The function called just before unloading

int pevent_plugin_unloader(void)

- 'pevent_plugin_options' (optional)
Plugin options that can be set before loading

struct plugin_option pevent_plugin_options[] = {
{
.name = "option-name",
.plugin_alias = "overide-file-name", (optional) .description = "description of option to show users",
},
{
.name = NULL,
},
};

Array must end with .name = NULL;

The plugin_alias (below) can be used to give a shorter name to access the vairable. Useful if a plugin handles more than one event.

NOTE options support is not backported yet.

- 'pevent_plugin_alias' (optional)
The name to use for finding options (uses filename if not defined)

New traceevent functions are added to search and load available plugins:

struct plugin_list* traceevent_load_plugins(struct pevent *pevent)
- loads plusing for 'struct pevent' object and returns loaded plugins list

void traceevent_unload_plugins(struct plugin_list *plugin_list);
- unload plugin list

Quoted 9 lines
Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>

As I authored the original code, you can add my:

-- Steve

Quoted 8 lines
Cc David Ahern <[email protected]>
---
 tools/lib/traceevent/Makefile       |   6 +-
 tools/lib/traceevent/event-parse.h  |   5 +
 tools/lib/traceevent/event-plugin.c | 202 ++++++++++++++++++++++++++++++++++++
 3 files changed, 212 insertions(+), 1 deletion(-)
 create mode 100644 tools/lib/traceevent/event-plugin.c
maintainer
#3 lore raw

Re: [PATCH 01/22] tools lib traceevent: Add plugin support

In reply to Jiri Olsa

Hi Jiri,

2013-11-21 (목), 12:00 +0100, Jiri Olsa:

Quoted 49 lines

Backporting plugin support for traceevent lib.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

It's now possible to use following interface to load plugins (shared objects) to enahnce pevent object functionality.

The plugin interface/hooks are as follows:
(taken from event-parse.h comments)

- 'pevent_plugin_loader' (required)
The function name to initialized the plugin.

int pevent_plugin_loader(struct pevent *pevent)

- 'pevent_plugin_unloader' (optional)
The function called just before unloading

int pevent_plugin_unloader(void)

- 'pevent_plugin_options' (optional)
Plugin options that can be set before loading

struct plugin_option pevent_plugin_options[] = {
{
.name = "option-name",
.plugin_alias = "overide-file-name", (optional) .description = "description of option to show users",
},
{
.name = NULL,
},
};

Array must end with .name = NULL;

The plugin_alias (below) can be used to give a shorter name to access the vairable. Useful if a plugin handles more than one event.

NOTE options support is not backported yet.

- 'pevent_plugin_alias' (optional)
The name to use for finding options (uses filename if not defined)

New traceevent functions are added to search and load available plugins:

Quoted 7 lines

struct plugin_list* traceevent_load_plugins(struct pevent *pevent)
- loads plusing for 'struct pevent' object and returns loaded plugins list

void traceevent_unload_plugins(struct plugin_list *plugin_list);
- unload plugin list

First of all, thank you for your nice work! :)

Quoted 1 line

[SNIP[

tools/lib/traceevent/event-plugin.c line 50

Quoted 12 lines
tools/lib/traceevent/event-plugin.c new file +12 -0
+static void
+load_plugin(struct pevent *pevent, const char *path,
+	    const char *file, void *data)
+{
+	struct plugin_list **plugin_list = data;
+	pevent_plugin_load_func func;
+	struct plugin_list *list;
+	const char *alias;
+	char *plugin;
+	void *handle;
+
+	plugin = malloc_or_die(strlen(path) + strlen(file) + 2);

I'd like not to see this malloc_or_die() anymore in a new code. Just returning after showing a warning looks enough here.

tools/lib/traceevent/event-plugin.c line 56

Quoted 6 lines
tools/lib/traceevent/event-plugin.c new file +6 -0
+
+	strcpy(plugin, path);
+	strcat(plugin, "/");
+	strcat(plugin, file);
+
+	handle = dlopen(plugin, RTLD_NOW | RTLD_GLOBAL);

Why RTLD_NOW and RTLD_GLOBAL? Hmm.. maybe using _NOW is needed to prevent a runtime error, but not sure why _GLOBAL is needed.

IIUC _GLOBAL is for exporting symbols to *other libraries*. Is it intended for this plugin support?

tools/lib/traceevent/event-plugin.c line 65

Quoted 9 lines
tools/lib/traceevent/event-plugin.c new file +9 -0
+	if (!handle) {
+		warning("could not load plugin '%s'\n%s\n",
+			plugin, dlerror());
+		goto out_free;
+	}
+
+	alias = dlsym(handle, PEVENT_PLUGIN_ALIAS_NAME);
+	if (!alias)
+		alias = file;

So this 'alias' is not used anywhere in the current code, right? Do you plan to add the option processing soon?

tools/lib/traceevent/event-plugin.c line 74

Quoted 9 lines
tools/lib/traceevent/event-plugin.c new file +9 -0
+
+	func = dlsym(handle, PEVENT_PLUGIN_LOADER_NAME);
+	if (!func) {
+		warning("could not find func '%s' in plugin '%s'\n%s\n",
+			PEVENT_PLUGIN_LOADER_NAME, plugin, dlerror());
+		goto out_free;
+	}
+
+	list = malloc_or_die(sizeof(*list));

Ditto. Please check return value and handle error properly.

Quoted 10 lines
tools/lib/traceevent/event-plugin.c new file +10 -0
+	list->next = *plugin_list;
+	list->handle = handle;
+	list->name = plugin;
+	*plugin_list = list;
+
+	pr_stat("registering plugin: %s", plugin);
+	func(pevent);
+	return;
+
+ out_free:

tools/lib/traceevent/event-plugin.c line 94

Quoted 10 lines
tools/lib/traceevent/event-plugin.c new file +10 -0
+	free(plugin);
+}
+
+static void
+load_plugins_dir(struct pevent *pevent, const char *suffix,
+		 const char *path,
+		 void (*load_plugin)(struct pevent *pevent,
+				     const char *path,
+				     const char *name,
+				     void *data),

Doesn't gcc complain about the name shadows other function? Anyway I think we should avoid using same name - probably using typedef for the function might be helpful.

Quoted 1 line
(quoted diff) +1 -0
+		 void *data)

[SNIP]

tools/lib/traceevent/event-plugin.c line 166

Quoted 9 lines
tools/lib/traceevent/event-plugin.c new file +9 -0
+	/*
+	 * Now let the home directory override the environment
+	 * or system defaults.
+	 */
+	home = getenv("HOME");
+	if (!home)
+		return;
+
+	path = malloc_or_die(strlen(home) + strlen(LOCAL_PLUGIN_DIR) + 2);

Ditto.

Thanks,
Namhyung

Quoted 36 lines
tools/lib/traceevent/event-plugin.c new file +36 -0
+
+	strcpy(path, home);
+	strcat(path, "/");
+	strcat(path, LOCAL_PLUGIN_DIR);
+
+	load_plugins_dir(pevent, suffix, path, load_plugin, data);
+
+	free(path);
+}
+
+struct plugin_list*
+traceevent_load_plugins(struct pevent *pevent)
+{
+	struct plugin_list *list = NULL;
+
+	load_plugins(pevent, ".so", load_plugin, &list);
+	return list;
+}
+
+void
+traceevent_unload_plugins(struct plugin_list *plugin_list)
+{
+	pevent_plugin_unload_func func;
+	struct plugin_list *list;
+
+	while (plugin_list) {
+		list = plugin_list;
+		plugin_list = list->next;
+		func = dlsym(list->handle, PEVENT_PLUGIN_UNLOADER_NAME);
+		if (func)
+			func();
+		dlclose(list->handle);
+		free(list->name);
+		free(list);
+	}
+}
#4 lore raw

Re: [PATCH 01/22] tools lib traceevent: Add plugin support

Quoted 17 lines · On Fri, 22 Nov 2013 23:17:06 +0900 Namhyung Kim <[email protected]> wrote:
Quoted 1 line

[SNIP[

tools/lib/traceevent/event-plugin.c line 50

Quoted 12 lines
tools/lib/traceevent/event-plugin.c new file +12 -0
+static void
+load_plugin(struct pevent *pevent, const char *path,
+	    const char *file, void *data)
+{
+	struct plugin_list **plugin_list = data;
+	pevent_plugin_load_func func;
+	struct plugin_list *list;
+	const char *alias;
+	char *plugin;
+	void *handle;
+
+	plugin = malloc_or_die(strlen(path) + strlen(file) + 2);

I'd like not to see this malloc_or_die() anymore in a new code. Just returning after showing a warning looks enough here.

Yeah I agree. This is a relic from my code. I think it's OK to add here, as it is pretty much direct port of my code, and then we can just add a patch against it to remove it.

Quoted 10 lines

tools/lib/traceevent/event-plugin.c line 56

Quoted 7 lines
tools/lib/traceevent/event-plugin.c new file +6 -0
+
+	strcpy(plugin, path);
+	strcat(plugin, "/");
+	strcat(plugin, file);
+
+	handle = dlopen(plugin, RTLD_NOW | RTLD_GLOBAL);

Why RTLD_NOW and RTLD_GLOBAL? Hmm.. maybe using _NOW is needed to prevent a runtime error, but not sure why _GLOBAL is needed.

Yes, we want to make sure all symbols defined are available at time of load, otherwise bail out.

Quoted 3 lines

IIUC _GLOBAL is for exporting symbols to *other libraries*. Is it intended for this plugin support?

That was the plan. To have one plugin supply a set of functions that other plugins may use. That is what GLOBAL is for, right? I don't recall if I every did this, but it was something I wanted for future work.

Now if we don't need it, we could remove it, but is it bad to have?

Quoted 14 lines

tools/lib/traceevent/event-plugin.c line 65

Quoted 11 lines
tools/lib/traceevent/event-plugin.c new file +9 -0
+	if (!handle) {
+		warning("could not load plugin '%s'\n%s\n",
+			plugin, dlerror());
+		goto out_free;
+	}
+
+	alias = dlsym(handle, PEVENT_PLUGIN_ALIAS_NAME);
+	if (!alias)
+		alias = file;

So this 'alias' is not used anywhere in the current code, right? Do you plan to add the option processing soon?

I believe he is, and hopefully he will, because I want to forward port this code to have trace-cmd use it. The end result should be a separate library that we all can use.

It may be best to add the above alias when we add option support. But it doesn't hurt to include it now, as long as we plan to add that in the near future.

Quoted 12 lines

tools/lib/traceevent/event-plugin.c line 74

Quoted 10 lines
tools/lib/traceevent/event-plugin.c new file +9 -0
+
+	func = dlsym(handle, PEVENT_PLUGIN_LOADER_NAME);
+	if (!func) {
+		warning("could not find func '%s' in plugin '%s'\n%s\n",
+			PEVENT_PLUGIN_LOADER_NAME, plugin, dlerror());
+		goto out_free;
+	}
+
+	list = malloc_or_die(sizeof(*list));

Ditto. Please check return value and handle error properly.

I agree, but lets make that a separate patch, such that we see the changes between the backport and updates.

Quoted 18 lines
Quoted 18 lines
tools/lib/traceevent/event-plugin.c new file +16 -0
+	list->next = *plugin_list;
+	list->handle = handle;
+	list->name = plugin;
+	*plugin_list = list;
+
+	pr_stat("registering plugin: %s", plugin);
+	func(pevent);
+	return;
+
+ out_free:
+	free(plugin);
+}
+
+static void
+load_plugins_dir(struct pevent *pevent, const char *suffix,
+		 const char *path,
Quoted 8 lines

tools/lib/traceevent/event-plugin.c line 94

Quoted 4 lines
tools/lib/traceevent/event-plugin.c new file +4 -0
+		 void (*load_plugin)(struct pevent *pevent,
+				     const char *path,
+				     const char *name,
+				     void *data),

Doesn't gcc complain about the name shadows other function? Anyway I think we should avoid using same name - probably using typedef for the function might be helpful.

I agree here too, but again, lets keep the backport and fixes as separate patches.

-- Steve

Quoted 18 lines
Quoted 3 lines
(quoted diff) +1 -0
+		 void *data)

[SNIP]

tools/lib/traceevent/event-plugin.c line 166

Quoted 9 lines
tools/lib/traceevent/event-plugin.c new file +9 -0
+	/*
+	 * Now let the home directory override the environment
+	 * or system defaults.
+	 */
+	home = getenv("HOME");
+	if (!home)
+		return;
+
+	path = malloc_or_die(strlen(home) + strlen(LOCAL_PLUGIN_DIR) + 2);

Ditto.

Thanks,
Namhyung

maintainer
#5 lore raw

Re: [PATCH 01/22] tools lib traceevent: Add plugin support

Quoted 21 lines · Steven Rostedt
Quoted 17 lines · On Fri, 22 Nov 2013 23:17:06 +0900 Namhyung Kim <[email protected]> wrote:
Quoted 1 line

[SNIP[

tools/lib/traceevent/event-plugin.c line 50

Quoted 12 lines
tools/lib/traceevent/event-plugin.c new file +12 -0
+static void
+load_plugin(struct pevent *pevent, const char *path,
+	    const char *file, void *data)
+{
+	struct plugin_list **plugin_list = data;
+	pevent_plugin_load_func func;
+	struct plugin_list *list;
+	const char *alias;
+	char *plugin;
+	void *handle;
+
+	plugin = malloc_or_die(strlen(path) + strlen(file) + 2);

I'd like not to see this malloc_or_die() anymore in a new code. Just returning after showing a warning looks enough here.

Quoted 3 lines

Yeah I agree. This is a relic from my code. I think it's OK to add here, as it is pretty much direct port of my code, and then we can just add a patch against it to remove it.

Okay. I agree that it'd be better to make them separate patches.

Quoted 19 lines
Quoted 11 lines

tools/lib/traceevent/event-plugin.c line 56

Quoted 8 lines
tools/lib/traceevent/event-plugin.c new file +6 -0
+
+	strcpy(plugin, path);
+	strcat(plugin, "/");
+	strcat(plugin, file);
+
+	handle = dlopen(plugin, RTLD_NOW | RTLD_GLOBAL);

Why RTLD_NOW and RTLD_GLOBAL? Hmm.. maybe using _NOW is needed to prevent a runtime error, but not sure why _GLOBAL is needed.

Yes, we want to make sure all symbols defined are available at time of load, otherwise bail out.

Quoted 3 lines

IIUC _GLOBAL is for exporting symbols to *other libraries*. Is it intended for this plugin support?

Quoted 6 lines

That was the plan. To have one plugin supply a set of functions that other plugins may use. That is what GLOBAL is for, right? I don't recall if I every did this, but it was something I wanted for future work.

Now if we don't need it, we could remove it, but is it bad to have?

I might be slow down symbol resolution of new plugins tiny bit. But I don't think it's a real problem as its effect will be negligible.

I don't object the code but just want to know your intention. :)

Thanks,
Namhyung

author
#6 lore raw

Re: [PATCH 01/22] tools lib traceevent: Add plugin support

In reply to Steven Rostedt

On Sat, Nov 23, 2013 at 03:12:19AM -0500, Steven Rostedt wrote:

SNIP

Quoted 13 lines
Quoted 4 lines
Quoted 4 lines
tools/lib/traceevent/event-plugin.c new file +4 -0
+	if (!handle) {
+		warning("could not load plugin '%s'\n%s\n",
+			plugin, dlerror());
+		goto out_free;
Quoted 8 lines

tools/lib/traceevent/event-plugin.c line 65

Quoted 5 lines
tools/lib/traceevent/event-plugin.c new file +5 -0
+	}
+
+	alias = dlsym(handle, PEVENT_PLUGIN_ALIAS_NAME);
+	if (!alias)
+		alias = file;

So this 'alias' is not used anywhere in the current code, right? Do you plan to add the option processing soon?

Quoted 3 lines

I believe he is, and hopefully he will, because I want to forward port this code to have trace-cmd use it. The end result should be a separate library that we all can use.

yep, thats the plan ;-)

so I get the the deal is in v2 I keep the current code and add patches addressing comments on top of it..

jirka

author
#7 lore raw

PATCH 2/22 tools lib traceevent: Add plugin build support

In reply to Jiri Olsa

Backporting missing pieces of plugin building infrastructure:
- adding Makefile 'plugins' target to build all defined plugins
- adding Makefile 'install_plugins' target as 'install_lib' target dependency
- link plugin objects with shared object building

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

Plugins are by default installed into following locations:
'$(HOME)/.traceevent/plugins'
- if we are installing under $(HOME)
'$(prefix)/lib/traceevent/plugins'
- otherwise

This path is propagated to the plugin object as a plugins search path.

Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/lib/traceevent/Makefile | 42 +++++++++++++++++++++++++++++++++++++-----
 1 file changed, 37 insertions(+), 5 deletions(-)
tools/lib/traceevent/Makefile +37 -5
index 2ccb5bc..ea88c19 100644
@@ -43,6 +43,20 @@ man_dir_SQ = '$(subst ','\'',$(man_dir))'
 export man_dir man_dir_SQ INSTALL
 export DESTDIR DESTDIR_SQ
 
+# Set plugin_dir to preffered global plugin location
+# If we install under $HOME directory we go under
+# $(HOME)/.traceevent/plugins
+ifeq ($(plugin_dir),)
+ifeq ($(prefix),$(HOME))
+override plugin_dir = $(HOME)/.traceevent/plugins
+else
+override plugin_dir = $(prefix)/lib/traceevent/plugins
+endif
+endif
+
+PLUGIN_DIR = -DPLUGIN_DIR="$(DESTDIR)/$(plugin_dir)"
+PLUGIN_DIR_SQ = '$(subst ','\'',$(PLUGIN_DIR))'
+
 # copy a bit from Linux kbuild
 
 ifeq ("$(origin V)", "command line")
@@ -96,6 +110,7 @@ export prefix bindir src obj
 # Shell quotes
 bindir_SQ = $(subst ','\'',$(bindir))
 bindir_relative_SQ = $(subst ','\'',$(bindir_relative))
+plugin_dir_SQ = $(subst ','\'',$(plugin_dir))
 
 LIB_FILE = libtraceevent.a libtraceevent.so
 
@@ -138,8 +153,8 @@ else
   print_app_build =		echo '  BUILD    '$(OBJ);
   print_fpic_compile =		echo '  CC FPIC  '$(OBJ);
   print_shared_lib_compile =	echo '  BUILD    SHARED LIB '$(OBJ);
-  print_plugin_obj_compile =	echo '  BUILD    PLUGIN OBJ '$(OBJ);
-  print_plugin_build =		echo '  BUILD    PLUGIN     '$(OBJ);
+  print_plugin_obj_compile =	echo '  CC FPIC  '$(OBJ);
+  print_plugin_build =		echo '  BUILD    PLUGIN '$(OBJ);
   print_static_lib_build =	echo '  BUILD    STATIC LIB '$(OBJ);
   print_install =		echo '  INSTALL  '$1'	to	$(DESTDIR_SQ)$2';
 endif
@@ -187,9 +202,11 @@ PEVENT_LIB_OBJS += parse-filter.o
 PEVENT_LIB_OBJS += parse-utils.o
 PEVENT_LIB_OBJS += kbuffer-parse.o
 
-ALL_OBJS = $(PEVENT_LIB_OBJS)
+PLUGINS := $(PLUGIN_OBJS:.o=.so)
+
+ALL_OBJS = $(PEVENT_LIB_OBJS) $(PLUGIN_OBJS)
 
-CMD_TARGETS = $(LIB_FILE)
+CMD_TARGETS = $(LIB_FILE) $(PLUGINS)
 
 TARGETS = $(CMD_TARGETS)
 
@@ -204,9 +221,17 @@ libtraceevent.so: $(PEVENT_LIB_OBJS)
 libtraceevent.a: $(PEVENT_LIB_OBJS)
 	$(Q)$(do_build_static_lib)
 
+plugins: $(PLUGINS)
+
 $(PEVENT_LIB_OBJS): %.o: $(src)/%.c TRACEEVENT-CFLAGS
 	$(Q)$(do_fpic_compile)
 
+$(PLUGIN_OBJS): %.o : $(src)/%.c
+	$(Q)$(do_compile_plugin_obj)
+
+$(PLUGINS): %.so: %.o
+	$(Q)$(do_plugin_build)
+
 define make_version.h
 	(echo '/* This file is automatically generated. Do not modify. */';		\
 	echo \#define VERSION_CODE $(shell						\
@@ -294,9 +319,16 @@ define do_install
 	$(INSTALL) $1 '$(DESTDIR_SQ)$2'
 endef
 
-install_lib: all_cmd
+install_lib: all_cmd install_plugins
 	$(Q)$(call do_install,$(LIB_FILE),$(bindir_SQ))
 
+PLUGINS_INSTALL = $(subst .so,.install,$(PLUGINS))
+
+$(PLUGINS_INSTALL): %.install : %.so force
+	$(Q)$(call do_install,$<,$(plugin_dir_SQ))
+
+install_plugins: $(PLUGINS_INSTALL)
+
 install: install_lib
 
 clean:
Signature · 1 line
1.8.3.1
author
#8 lore raw

PATCH 3/22 tools lib traceevent: Add traceevent_host_bigendian function

In reply to Jiri Olsa

Adding traceevent_host_bigendian function to get host endianity. It's used in following patches.

Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/lib/traceevent/event-parse.h | 9 +++++++++
 1 file changed, 9 insertions(+)
tools/lib/traceevent/event-parse.h +9 -0
index a288860..0ad7f6b 100644
@@ -527,6 +527,15 @@ __data2host8(struct pevent *pevent, unsigned long long data)
 	__data2host8(pevent, __val);				\
 })
 
+static inline int traceevent_host_bigendian(void)
+{
+	unsigned char str[] = { 0x1, 0x2, 0x3, 0x4 };
+	unsigned int *ptr;
+
+	ptr = (unsigned int *)str;
+	return *ptr == 0x01020304;
+}
+
 /* taken from kernel/trace/trace.h */
 enum trace_flag_type {
 	TRACE_FLAG_IRQS_OFF		= 0x01,
Signature · 1 line
1.8.3.1
maintainer
#9 lore raw

Re: [PATCH 03/22] tools lib traceevent: Add traceevent_host_bigendian function

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 2 lines

Adding traceevent_host_bigendian function to get host endianity. It's used in following patches.

[SNIP]

tools/lib/traceevent/event-parse.h __data2host8(struct pevent *pevent, unsigned long long data) · line 536

Quoted 7 lines
tools/lib/traceevent/event-parse.h +7 -0
+static inline int traceevent_host_bigendian(void)
+{
+	unsigned char str[] = { 0x1, 0x2, 0x3, 0x4 };
+	unsigned int *ptr;
+
+	ptr = (unsigned int *)str;
+	return *ptr == 0x01020304;

Is it safe for every architecture supported - especially ones that require stricter alignment? I know many architectures/compilers align stack but not sure doing this is safe for all architecture.

Thanks,
Namhyung

#10 lore raw

Re: [PATCH 03/22] tools lib traceevent: Add traceevent_host_bigendian function

Quoted 5 lines · On Fri, 22 Nov 2013 23:22:52 +0900 Namhyung Kim <[email protected]> wrote:

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 2 lines

Adding traceevent_host_bigendian function to get host endianity. It's used in following patches.

[SNIP]

Quoted 11 lines

tools/lib/traceevent/event-parse.h __data2host8(struct pevent *pevent, unsigned long long data) · line 536

Quoted 7 lines
tools/lib/traceevent/event-parse.h +7 -0
+static inline int traceevent_host_bigendian(void)
+{
+	unsigned char str[] = { 0x1, 0x2, 0x3, 0x4 };
+	unsigned int *ptr;
+
+	ptr = (unsigned int *)str;
+	return *ptr == 0x01020304;

Is it safe for every architecture supported - especially ones that require stricter alignment? I know many architectures/compilers align stack but not sure doing this is safe for all architecture.

Would you prefer this (I tested it on both a big and little endian)

{
	unsigned char str[] = { 0x1, 0x2, 0x3, 0x4 };
	unsigned int val;

	memcpy(&val, str, 4);
	return val == 0x01020304;
}

-- Steve

maintainer
#11 lore raw

Re: [PATCH 03/22] tools lib traceevent: Add traceevent_host_bigendian function

Hi Steve,

Quoted 22 lines · Steven Rostedt
Quoted 5 lines · On Fri, 22 Nov 2013 23:22:52 +0900 Namhyung Kim <[email protected]> wrote:

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 2 lines

Adding traceevent_host_bigendian function to get host endianity. It's used in following patches.

[SNIP]

Quoted 11 lines

tools/lib/traceevent/event-parse.h __data2host8(struct pevent *pevent, unsigned long long data) · line 536

Quoted 7 lines
tools/lib/traceevent/event-parse.h +7 -0
+static inline int traceevent_host_bigendian(void)
+{
+	unsigned char str[] = { 0x1, 0x2, 0x3, 0x4 };
+	unsigned int *ptr;
+
+	ptr = (unsigned int *)str;
+	return *ptr == 0x01020304;

Is it safe for every architecture supported - especially ones that require stricter alignment? I know many architectures/compilers align stack but not sure doing this is safe for all architecture.

Would you prefer this (I tested it on both a big and little endian)

Quoted 7 lines
{
	unsigned char str[] = { 0x1, 0x2, 0x3, 0x4 };
	unsigned int val;

	memcpy(&val, str, 4);
	return val == 0x01020304;
}

Yeah, looks good to me.

Thanks,
Namhyung

author
#12 lore raw

Re: [PATCH 03/22] tools lib traceevent: Add traceevent_host_bigendian function

Quoted 29 lines · Namhyung Kim

Hi Steve,

Quoted 26 lines · Steven Rostedt
Quoted 5 lines · On Fri, 22 Nov 2013 23:22:52 +0900 Namhyung Kim <[email protected]> wrote:

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 2 lines

Adding traceevent_host_bigendian function to get host endianity. It's used in following patches.

[SNIP]

Quoted 11 lines

tools/lib/traceevent/event-parse.h __data2host8(struct pevent *pevent, unsigned long long data) · line 536

Quoted 7 lines
tools/lib/traceevent/event-parse.h +7 -0
+static inline int traceevent_host_bigendian(void)
+{
+	unsigned char str[] = { 0x1, 0x2, 0x3, 0x4 };
+	unsigned int *ptr;
+
+	ptr = (unsigned int *)str;
+	return *ptr == 0x01020304;

Is it safe for every architecture supported - especially ones that require stricter alignment? I know many architectures/compilers align stack but not sure doing this is safe for all architecture.

Would you prefer this (I tested it on both a big and little endian)

{
	unsigned char str[] = { 0x1, 0x2, 0x3, 0x4 };
	unsigned int val;
Quoted 5 lines
Quoted 3 lines
	memcpy(&val, str, 4);
	return val == 0x01020304;
}

Yeah, looks good to me.

ok, will change to this one

thanks,
jirka

author
#13 lore raw

PATCH 4/22 tools lib traceevent: Add jbd2 plugin

In reply to Jiri Olsa

Backporting jbd2 plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds field resolving functions for following tracepoint events:
jbd2:jbd2_checkpoint_stats jbd2:jbd2_run_stats

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'jbd2:jbd2_run_stats,jbd2:jbd2_checkpoint_stats' -a')

--- script.jbd2.old
+++ script.jbd2.new

-     jbd2/dm-3-8   576 [000]  2983.748423: jbd2:jbd2_checkpoint_stats: [FAILED TO PARSE] dev=265289731 tid=0x3f2bbb chp_time=0x0 forced_to_close=0 written=0 dropped=2
+     jbd2/dm-3-8   576 [000]  2983.748423: jbd2:jbd2_checkpoint_stats: dev 253,3 tid 4139963 chp_time 0 forced_to_close 0 written 0 dropped 2
-     jbd2/dm-3-8   576 [000]  2983.852789: jbd2:jbd2_run_stats: [FAILED TO PARSE] dev=265289731 tid=0x3f2bc0 wait=0x0 request_delay=0x0 running=0x138d locked=0x0 flushing=0x0 logging=0x68 handle_count=2995 blocks=17 blocks_logged=18
+     jbd2/dm-3-8   576 [000]  2983.852789: jbd2:jbd2_run_stats: dev 253,3 tid 4139968 wait 0 request_delay 0 running 5005 locked 0 flushing 0 logging 104 handle_count 2995 blocks 17 blocks_logged 18

This plugin also adds jbd2_dev_to_name function, which was removed from jdb2 tracepoints recently. Keeping it in for backward compatibility reasons.

Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/lib/traceevent/Makefile      |  2 ++
 tools/lib/traceevent/plugin_jbd2.c | 66 ++++++++++++++++++++++++++++++++++++++
 2 files changed, 68 insertions(+)
 create mode 100644 tools/lib/traceevent/plugin_jbd2.c
tools/lib/traceevent/Makefile +2 -0
index ea88c19..7aabc24 100644
@@ -202,6 +202,8 @@ PEVENT_LIB_OBJS += parse-filter.o
 PEVENT_LIB_OBJS += parse-utils.o
 PEVENT_LIB_OBJS += kbuffer-parse.o
 
+PLUGIN_OBJS = plugin_jbd2.o
+
 PLUGINS := $(PLUGIN_OBJS:.o=.so)
 
 ALL_OBJS = $(PEVENT_LIB_OBJS) $(PLUGIN_OBJS)
tools/lib/traceevent/plugin_jbd2.c new file +66 -0

66 lines not on this page · raw message

Signature · 1 line
1.8.3.1
#14 lore raw

Re: [PATCH 04/22] tools lib traceevent: Add jbd2 plugin

Quoted 24 lines · On Thu, 21 Nov 2013 12:01:02 +0100 Jiri Olsa <[email protected]> wrote:

Backporting jbd2 plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds field resolving functions for following tracepoint events:
jbd2:jbd2_checkpoint_stats jbd2:jbd2_run_stats

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'jbd2:jbd2_run_stats,jbd2:jbd2_checkpoint_stats' -a')

script.jbd2.new renamed from script.jbd2.old +2 -2
-     jbd2/dm-3-8   576 [000]  2983.748423: jbd2:jbd2_checkpoint_stats: [FAILED TO PARSE] dev=265289731 tid=0x3f2bbb chp_time=0x0 forced_to_close=0 written=0 dropped=2
+     jbd2/dm-3-8   576 [000]  2983.748423: jbd2:jbd2_checkpoint_stats: dev 253,3 tid 4139963 chp_time 0 forced_to_close 0 written 0 dropped 2
-     jbd2/dm-3-8   576 [000]  2983.852789: jbd2:jbd2_run_stats: [FAILED TO PARSE] dev=265289731 tid=0x3f2bc0 wait=0x0 request_delay=0x0 running=0x138d locked=0x0 flushing=0x0 logging=0x68 handle_count=2995 blocks=17 blocks_logged=18
+     jbd2/dm-3-8   576 [000]  2983.852789: jbd2:jbd2_run_stats: dev 253,3 tid 4139968 wait 0 request_delay 0 running 5005 locked 0 flushing 0 logging 104 handle_count 2995 blocks 17 blocks_logged 18

This plugin also adds jbd2_dev_to_name function, which was removed from jdb2 tracepoints recently. Keeping it in for backward compatibility reasons.

Quoted 9 lines
Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>

-- Steve

Quoted 2 lines
Cc David Ahern <[email protected]>

---

maintainer
#15 lore raw

Re: [PATCH 04/22] tools lib traceevent: Add jbd2 plugin

In reply to Jiri Olsa

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 23 lines

Backporting jbd2 plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds field resolving functions for following tracepoint events:
jbd2:jbd2_checkpoint_stats jbd2:jbd2_run_stats

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'jbd2:jbd2_run_stats,jbd2:jbd2_checkpoint_stats' -a')

script.jbd2.new renamed from script.jbd2.old +2 -2
-     jbd2/dm-3-8   576 [000]  2983.748423: jbd2:jbd2_checkpoint_stats: [FAILED TO PARSE] dev=265289731 tid=0x3f2bbb chp_time=0x0 forced_to_close=0 written=0 dropped=2
+     jbd2/dm-3-8   576 [000]  2983.748423: jbd2:jbd2_checkpoint_stats: dev 253,3 tid 4139963 chp_time 0 forced_to_close 0 written 0 dropped 2
-     jbd2/dm-3-8   576 [000]  2983.852789: jbd2:jbd2_run_stats: [FAILED TO PARSE] dev=265289731 tid=0x3f2bc0 wait=0x0 request_delay=0x0 running=0x138d locked=0x0 flushing=0x0 logging=0x68 handle_count=2995 blocks=17 blocks_logged=18
+     jbd2/dm-3-8   576 [000]  2983.852789: jbd2:jbd2_run_stats: dev 253,3 tid 4139968 wait 0 request_delay 0 running 5005 locked 0 flushing 0 logging 104 handle_count 2995 blocks 17 blocks_logged 18

This plugin also adds jbd2_dev_to_name function, which was removed from jdb2 tracepoints recently. Keeping it in for backward compatibility reasons.

[SNIP]

script.jbd2.new

Quoted 8 lines
script.jbd2.new renamed from script.jbd2.old +8 -0
+#define MINORBITS	20
+#define MINORMASK	((1U << MINORBITS) - 1)
+
+#define MAJOR(dev)	((unsigned int) ((dev) >> MINORBITS))
+#define MINOR(dev)	((unsigned int) ((dev) & MINORMASK))
+
+unsigned long long process_jbd2_dev_to_name(struct trace_seq *s,
+					    unsigned long long *args)

Isn't it better to make these functions static?

Quoted 11 lines
script.jbd2.new renamed from script.jbd2.old +11 -0
+{
+	unsigned int dev = args[0];
+
+	trace_seq_printf(s, "%d:%d", MAJOR(dev), MINOR(dev));
+	return 0;
+}
+
+unsigned long long process_jiffies_to_msecs(struct trace_seq *s,
+					    unsigned long long *args)
+{
+	unsigned long long jiffies = args[0];

script.jbd2.new

Quoted 10 lines
script.jbd2.new renamed from script.jbd2.old +10 -0
+
+	trace_seq_printf(s, "%lld", jiffies);
+	return jiffies;
+}
+
+int PEVENT_PLUGIN_LOADER(struct pevent *pevent)
+{
+	pevent_register_print_function(pevent,
+				       process_jbd2_dev_to_name,
+				       PEVENT_FUNC_ARG_STRING,

Actually the function returns long long not string. But it seems the current code doesn't care about the return type.

script.jbd2.new

Quoted 12 lines
script.jbd2.new renamed from script.jbd2.old +12 -0
+				       "jbd2_dev_to_name",
+				       PEVENT_FUNC_ARG_INT,
+				       PEVENT_FUNC_ARG_VOID);
+
+	pevent_register_print_function(pevent,
+				       process_jiffies_to_msecs,
+				       PEVENT_FUNC_ARG_LONG,
+				       "jiffies_to_msecs",
+				       PEVENT_FUNC_ARG_LONG,
+				       PEVENT_FUNC_ARG_VOID);
+	return 0;
+}

Shouldn't it unregister these functions when unloaded?

Thanks,
Namhyung

#16 lore raw

Re: [PATCH 04/22] tools lib traceevent: Add jbd2 plugin

Quoted 11 lines · On Fri, 22 Nov 2013 23:27:57 +0900 Namhyung Kim <[email protected]> wrote:

[SNIP]

tools/lib/traceevent/plugin_jbd2.c line 33

Quoted 8 lines
tools/lib/traceevent/plugin_jbd2.c new file +8 -0
+#define MINORBITS	20
+#define MINORMASK	((1U << MINORBITS) - 1)
+
+#define MAJOR(dev)	((unsigned int) ((dev) >> MINORBITS))
+#define MINOR(dev)	((unsigned int) ((dev) & MINORMASK))
+
+unsigned long long process_jbd2_dev_to_name(struct trace_seq *s,
+					    unsigned long long *args)

Isn't it better to make these functions static?

Yeah, but that was my mistake. Lets keep the fix separate from the backport though. Thanks!

Quoted 18 lines
Quoted 18 lines
tools/lib/traceevent/plugin_jbd2.c new file +16 -0
+{
+	unsigned int dev = args[0];
+
+	trace_seq_printf(s, "%d:%d", MAJOR(dev), MINOR(dev));
+	return 0;
+}
+
+unsigned long long process_jiffies_to_msecs(struct trace_seq *s,
+					    unsigned long long *args)
+{
+	unsigned long long jiffies = args[0];
+
+	trace_seq_printf(s, "%lld", jiffies);
+	return jiffies;
+}
+
Quoted 8 lines

tools/lib/traceevent/plugin_jbd2.c line 54

Quoted 5 lines
tools/lib/traceevent/plugin_jbd2.c new file +5 -0
+int PEVENT_PLUGIN_LOADER(struct pevent *pevent)
+{
+	pevent_register_print_function(pevent,
+				       process_jbd2_dev_to_name,
+				       PEVENT_FUNC_ARG_STRING,

Actually the function returns long long not string. But it seems the current code doesn't care about the return type.

Actually it's not representing what process_jbd2_dev_to_name() returns (which will always return unsigned long long), but what "jbd2_dev_to_name()" returns that is (was) defined in the kernel. That was:

const char *jbd2_dev_to_name(dev_t device)

When registering a function to handle, you need to express the prototype of that function (not the handler). The third argument is the ret_type of that function.

But this is interesting, the ret_type doesn't seem to be used in event_parse.c. The return value of the callback is only done in eval_num_arg() where we could put a warning if the ret_type is not a number.

Funny thing is, this patch is irrelevant with todays kernel due to this commit:

commit 4862fd6047ed02e2726667c54d35f538eecc56aa

Author: Theodore Ts'o <[email protected]>
Date:   Sun Jul 10 22:05:08 2011 -0400

jbd2: remove jbd2_dev_to_name() from jbd2 tracepoints

Using function calls in TP_printk causes perf heartburn, so print the MAJOR/MINOR device numbers instead.

Quoted 16 lines
Quoted 4 lines
tools/lib/traceevent/plugin_jbd2.c new file +2 -0
+				       "jbd2_dev_to_name",
+				       PEVENT_FUNC_ARG_INT,

tools/lib/traceevent/plugin_jbd2.c line 66

Quoted 10 lines
tools/lib/traceevent/plugin_jbd2.c new file +10 -0
+				       PEVENT_FUNC_ARG_VOID);
+
+	pevent_register_print_function(pevent,
+				       process_jiffies_to_msecs,
+				       PEVENT_FUNC_ARG_LONG,
+				       "jiffies_to_msecs",
+				       PEVENT_FUNC_ARG_LONG,
+				       PEVENT_FUNC_ARG_VOID);
+	return 0;
+}

Shouldn't it unregister these functions when unloaded?

Probably ;-)

-- Steve

Quoted 4 lines

Thanks,
Namhyung

maintainer
#17 lore raw

Re: [PATCH 04/22] tools lib traceevent: Add jbd2 plugin

Quoted 42 lines · Steven Rostedt
Quoted 25 lines · On Fri, 22 Nov 2013 23:27:57 +0900 Namhyung Kim <[email protected]> wrote:

[SNIP]

Quoted 24 lines
tools/lib/traceevent/plugin_jbd2.c new file +24 -0
+#define MINORBITS	20
+#define MINORMASK	((1U << MINORBITS) - 1)
+
+#define MAJOR(dev)	((unsigned int) ((dev) >> MINORBITS))
+#define MINOR(dev)	((unsigned int) ((dev) & MINORMASK))
+
+unsigned long long process_jbd2_dev_to_name(struct trace_seq *s,
+					    unsigned long long *args)
+{
+	unsigned int dev = args[0];
+
+	trace_seq_printf(s, "%d:%d", MAJOR(dev), MINOR(dev));
+	return 0;
+}
+
+unsigned long long process_jiffies_to_msecs(struct trace_seq *s,
+					    unsigned long long *args)
+{
+	unsigned long long jiffies = args[0];
+
+	trace_seq_printf(s, "%lld", jiffies);
+	return jiffies;
+}
+
Quoted 8 lines

tools/lib/traceevent/plugin_jbd2.c line 54

Quoted 5 lines
tools/lib/traceevent/plugin_jbd2.c new file +5 -0
+int PEVENT_PLUGIN_LOADER(struct pevent *pevent)
+{
+	pevent_register_print_function(pevent,
+				       process_jbd2_dev_to_name,
+				       PEVENT_FUNC_ARG_STRING,

Actually the function returns long long not string. But it seems the current code doesn't care about the return type.

Actually it's not representing what process_jbd2_dev_to_name() returns (which will always return unsigned long long), but what "jbd2_dev_to_name()" returns that is (was) defined in the kernel. That was:

Quoted 5 lines

const char *jbd2_dev_to_name(dev_t device)

When registering a function to handle, you need to express the prototype of that function (not the handler). The third argument is the ret_type of that function.

Aha, got it. Thank you for the explanation.

Quoted 4 lines

But this is interesting, the ret_type doesn't seem to be used in event_parse.c. The return value of the callback is only done in eval_num_arg() where we could put a warning if the ret_type is not a number.

Yes. :)

Thanks,
Namhyung

author
#18 lore raw

Re: [PATCH 04/22] tools lib traceevent: Add jbd2 plugin

In reply to Steven Rostedt
Quoted 15 lines · Steven Rostedt
Quoted 11 lines · On Fri, 22 Nov 2013 23:27:57 +0900 Namhyung Kim <[email protected]> wrote:

[SNIP]

tools/lib/traceevent/plugin_jbd2.c line 33

Quoted 8 lines
tools/lib/traceevent/plugin_jbd2.c new file +8 -0
+#define MINORBITS	20
+#define MINORMASK	((1U << MINORBITS) - 1)
+
+#define MAJOR(dev)	((unsigned int) ((dev) >> MINORBITS))
+#define MINOR(dev)	((unsigned int) ((dev) & MINORMASK))
+
+unsigned long long process_jbd2_dev_to_name(struct trace_seq *s,
+					    unsigned long long *args)

Isn't it better to make these functions static?

Quoted 2 lines

Yeah, but that was my mistake. Lets keep the fix separate from the backport though. Thanks!

ok

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
tools/lib/traceevent/plugin_jbd2.c new file +4 -0
+{
+	unsigned int dev = args[0];
+
+	trace_seq_printf(s, "%d:%d", MAJOR(dev), MINOR(dev));

SNIP

Quoted 7 lines

jbd2: remove jbd2_dev_to_name() from jbd2 tracepoints

Using function calls in TP_printk causes perf heartburn, so print the MAJOR/MINOR device numbers instead.

yep, it's rather old change (v3.1-rc1) but I thought we want to keep that for backward compatibility

and it's small piece of code anyway.. ;-)

Quoted 20 lines
Quoted 5 lines
Quoted 14 lines

tools/lib/traceevent/plugin_jbd2.c line 66

Quoted 12 lines
tools/lib/traceevent/plugin_jbd2.c new file +12 -0
+				       "jbd2_dev_to_name",
+				       PEVENT_FUNC_ARG_INT,
+				       PEVENT_FUNC_ARG_VOID);
+
+	pevent_register_print_function(pevent,
+				       process_jiffies_to_msecs,
+				       PEVENT_FUNC_ARG_LONG,
+				       "jiffies_to_msecs",
+				       PEVENT_FUNC_ARG_LONG,
+				       PEVENT_FUNC_ARG_VOID);
+	return 0;
+}

Shouldn't it unregister these functions when unloaded?

Quoted 1 line

Probably ;-)

will add,

thanks,
jirka

author
#19 lore raw

Re: [PATCH 04/22] tools lib traceevent: Add jbd2 plugin

Quoted 5 lines · Jiri Olsa
Quoted 3 lines
Quoted 1 line

Shouldn't it unregister these functions when unloaded?

Probably ;-)

will add,

hum... haven't found exported interface to unregister/remove for print function, seems like this is getting freed in the end at:

void pevent_free(struct pevent *pevent)

jirka

author
#20 lore raw

PATCH 5/22 tools lib traceevent: Add hrtimer plugin

In reply to Jiri Olsa

Backporting hrtimer plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds function field resolving for following tracepoint events:
timer:hrtimer_expire_entry timer:hrtimer_start

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'timer:hrtimer*' -a')

--- script.hrtimer.old
+++ script.hrtimer.new

-         swapper     0 [000] 27405.519092: timer:hrtimer_start: [FAILED TO PARSE] hrtimer=0xffff88021e20e800 function=0xffffffff810c0e10 expires=27398383000000 softexpires=27398383000000
+         swapper     0 [000] 27405.519103: timer:hrtimer_start: hrtimer=0xffff88021e20e800 function=tick_sched_timer expires=27398383000000 softexpires=27398383000000
-         swapper     0 [001] 27405.519544: timer:hrtimer_expire_entry: [FAILED TO PARSE] hrtimer=0xffff880211334058 now=27398294182491 function=0xffffffff81086f20
+         swapper     0 [001] 27405.519544: timer:hrtimer_expire_entry: hrtimer=0xffff880211334058 now=27398294182491 function=posix_timer_fn/0x0

Check the 'function' field is translated into the function name.

Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/lib/traceevent/Makefile         |  3 +-
 tools/lib/traceevent/plugin_hrtimer.c | 93 +++++++++++++++++++++++++++++++++++
 2 files changed, 95 insertions(+), 1 deletion(-)
 create mode 100644 tools/lib/traceevent/plugin_hrtimer.c
tools/lib/traceevent/Makefile +2 -1
index 7aabc24..df087f3 100644
@@ -202,7 +202,8 @@ PEVENT_LIB_OBJS += parse-filter.o
 PEVENT_LIB_OBJS += parse-utils.o
 PEVENT_LIB_OBJS += kbuffer-parse.o
 
-PLUGIN_OBJS = plugin_jbd2.o
+PLUGIN_OBJS  = plugin_jbd2.o
+PLUGIN_OBJS += plugin_hrtimer.o
 
 PLUGINS := $(PLUGIN_OBJS:.o=.so)
 
tools/lib/traceevent/plugin_hrtimer.c new file +93 -0

93 lines not on this page · raw message

Signature · 1 line
1.8.3.1
#21 lore raw

Re: [PATCH 05/22] tools lib traceevent: Add hrtimer plugin

Quoted 22 lines · On Thu, 21 Nov 2013 12:01:03 +0100 Jiri Olsa <[email protected]> wrote:

Backporting hrtimer plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds function field resolving for following tracepoint events:
timer:hrtimer_expire_entry timer:hrtimer_start

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'timer:hrtimer*' -a')

script.hrtimer.new renamed from script.hrtimer.old +2 -2
-         swapper     0 [000] 27405.519092: timer:hrtimer_start: [FAILED TO PARSE] hrtimer=0xffff88021e20e800 function=0xffffffff810c0e10 expires=27398383000000 softexpires=27398383000000
+         swapper     0 [000] 27405.519103: timer:hrtimer_start: hrtimer=0xffff88021e20e800 function=tick_sched_timer expires=27398383000000 softexpires=27398383000000
-         swapper     0 [001] 27405.519544: timer:hrtimer_expire_entry: [FAILED TO PARSE] hrtimer=0xffff880211334058 now=27398294182491 function=0xffffffff81086f20
+         swapper     0 [001] 27405.519544: timer:hrtimer_expire_entry: hrtimer=0xffff880211334058 now=27398294182491 function=posix_timer_fn/0x0

Check the 'function' field is translated into the function name.

Quoted 9 lines
Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>

-- Steve

Quoted 1 line
Cc David Ahern <[email protected]>
maintainer
#22 lore raw

Re: [PATCH 05/22] tools lib traceevent: Add hrtimer plugin

In reply to Jiri Olsa

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 21 lines

Backporting hrtimer plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds function field resolving for following tracepoint events:
timer:hrtimer_expire_entry timer:hrtimer_start

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'timer:hrtimer*' -a')

script.hrtimer.new renamed from script.hrtimer.old +2 -2
-         swapper     0 [000] 27405.519092: timer:hrtimer_start: [FAILED TO PARSE] hrtimer=0xffff88021e20e800 function=0xffffffff810c0e10 expires=27398383000000 softexpires=27398383000000
+         swapper     0 [000] 27405.519103: timer:hrtimer_start: hrtimer=0xffff88021e20e800 function=tick_sched_timer expires=27398383000000 softexpires=27398383000000
-         swapper     0 [001] 27405.519544: timer:hrtimer_expire_entry: [FAILED TO PARSE] hrtimer=0xffff880211334058 now=27398294182491 function=0xffffffff81086f20
+         swapper     0 [001] 27405.519544: timer:hrtimer_expire_entry: hrtimer=0xffff880211334058 now=27398294182491 function=posix_timer_fn/0x0

Check the 'function' field is translated into the function name.

[SNIP]

script.hrtimer.new

Quoted 12 lines
script.hrtimer.new renamed from script.hrtimer.old +12 -0
+	if (!fn) {
+		trace_seq_printf(s, " function=MISSING");
+	} else {
+		unsigned long long function;
+		const char *func;
+
+		if (pevent_read_number_field(fn, data, &function))
+			trace_seq_printf(s, " function=INVALID");
+
+		func = pevent_find_function(pevent, function);
+
+		trace_seq_printf(s, " function=%s", func);

Shouldn't it be like below?

		if (pevent_read_number_field(fn, data, &function))
			trace_seq_printf(s, " function=INVALID");
		else {
			func = pevent_find_function(pevent, function);
			trace_seq_printf(s, " function=%s", func);
		}

Otherwise it'll print "function=XXX" twice if pevent_read_number_field() failed.

Thanks,
Namhyung

Quoted 20 lines
script.hrtimer.new renamed from script.hrtimer.old +20 -0
+	}
+
+	trace_seq_printf(s, " expires=");
+	pevent_print_num_field(s, "%llu", event, "expires", record, 1);
+
+	trace_seq_printf(s, " softexpires=");
+	pevent_print_num_field(s, "%llu", event, "softexpires", record, 1);
+	return 0;
+}
+
+int PEVENT_PLUGIN_LOADER(struct pevent *pevent)
+{
+	pevent_register_event_handler(pevent, -1,
+				      "timer", "hrtimer_expire_entry",
+				      timer_expire_handler, NULL);
+
+	pevent_register_event_handler(pevent, -1, "timer", "hrtimer_start",
+				      timer_start_handler, NULL);
+	return 0;
+}
#23 lore raw

Re: [PATCH 05/22] tools lib traceevent: Add hrtimer plugin

Quoted 46 lines · On Fri, 22 Nov 2013 23:33:29 +0900 Namhyung Kim <[email protected]> wrote:

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 21 lines

Backporting hrtimer plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds function field resolving for following tracepoint events:
timer:hrtimer_expire_entry timer:hrtimer_start

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'timer:hrtimer*' -a')

script.hrtimer.new renamed from script.hrtimer.old +2 -2
-         swapper     0 [000] 27405.519092: timer:hrtimer_start: [FAILED TO PARSE] hrtimer=0xffff88021e20e800 function=0xffffffff810c0e10 expires=27398383000000 softexpires=27398383000000
+         swapper     0 [000] 27405.519103: timer:hrtimer_start: hrtimer=0xffff88021e20e800 function=tick_sched_timer expires=27398383000000 softexpires=27398383000000
-         swapper     0 [001] 27405.519544: timer:hrtimer_expire_entry: [FAILED TO PARSE] hrtimer=0xffff880211334058 now=27398294182491 function=0xffffffff81086f20
+         swapper     0 [001] 27405.519544: timer:hrtimer_expire_entry: hrtimer=0xffff880211334058 now=27398294182491 function=posix_timer_fn/0x0

Check the 'function' field is translated into the function name.

[SNIP]

script.hrtimer.new

Quoted 12 lines
script.hrtimer.new renamed from script.hrtimer.old +12 -0
+	if (!fn) {
+		trace_seq_printf(s, " function=MISSING");
+	} else {
+		unsigned long long function;
+		const char *func;
+
+		if (pevent_read_number_field(fn, data, &function))
+			trace_seq_printf(s, " function=INVALID");
+
+		func = pevent_find_function(pevent, function);
+
+		trace_seq_printf(s, " function=%s", func);

Shouldn't it be like below?

		if (pevent_read_number_field(fn, data, &function))
			trace_seq_printf(s, " function=INVALID");
		else {
			func = pevent_find_function(pevent, function);
			trace_seq_printf(s, " function=%s", func);
		}
Quoted 3 lines

Otherwise it'll print "function=XXX" twice if pevent_read_number_field() failed.

Nice catch, but lets keep the backport and fixes separate.

Actually, this was one of the first plugins I wrote, and since then, I added a function tat does this for you. The proper change is to rip this out and use:

pevent_print_func_field()

instead. That function does the checks properly.

-- Steve

author
#24 lore raw

PATCH 6/22 tools lib traceevent: Add kmem plugin

In reply to Jiri Olsa

Backporting kmem plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds call_site field resolving for following tracepoint events:
kmem:kfree
kmem:kmalloc
kmem:kmalloc_node
kmem:kmem_cache_alloc kmem:kmem_cache_alloc_node kmem:kmem_cache_free

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'kmem:*' -a')

--- script.kmem.old
+++ script.kmem.new

-            perf 27846 [001] 29643.403319: kmem:kfree: call_site=ffffffff810e64f6 ptr=(nil)
+            perf 27846 [001] 29643.403238: kmem:kfree: (__audit_syscall_exit+0x1f6) call_site=ffffffff810e64f6 ptr=(nil)
-            perf 27846 [001] 29643.403337: kmem:kmem_cache_alloc: call_site=ffffffff812ff0c5 ptr=0xffff88020e155630 bytes_req=560 bytes_alloc=568 gfp_flags=GFP_KERNEL
+            perf 27846 [001] 29643.403337: kmem:kmem_cache_alloc: (radix_tree_preload+0x35) call_site=ffffffff812ff0c5 ptr=0xffff88020e155630 bytes_req=560 bytes_alloc=568 gfp_flags=GFP_KERNEL
-            perf 27846 [001] 29643.403342: kmem:kmem_cache_free: call_site=ffffffff8126ec61 ptr=0xffff88020dffe750
+            perf 27846 [001] 29643.403342: kmem:kmem_cache_free: (jbd2_journal_stop+0x221) call_site=ffffffff8126ec61 ptr=0xffff88020dffe750
-         firefox   954 [000] 29643.445477: kmem:kmem_cache_alloc_node: call_site=ffffffff8153c64e ptr=0xffff8801cecb4000 bytes_req=256 bytes_alloc=256 gfp_flags=GFP_KERNEL|GFP_REPEAT node=-1
+         firefox   954 [000] 29643.445477: kmem:kmem_cache_alloc_node: (__alloc_skb+0x4e) call_site=ffffffff8153c64e ptr=0xffff8801cecb4000 bytes_req=256 bytes_alloc=256 gfp_flags=GFP_KERNEL|GFP_REPEAT node=-1
-            perf 27846 [001] 29643.445510: kmem:kmalloc: call_site=ffffffff81250642 ptr=0xffff88020fd6c300 bytes_req=96 bytes_alloc=96 gfp_flags=GFP_NOFS|GFP_ZERO
+            perf 27846 [001] 29643.445510: kmem:kmalloc: (ext4_ext_find_extent+0x362) call_site=ffffffff81250642 ptr=0xffff88020fd6c300 bytes_req=96 bytes_alloc=96 gfp_flags=GFP_NOFS|GFP_ZERO
-               X   888 [002] 29643.445546: kmem:kmalloc_node: call_site=ffffffff8153c67e ptr=0xffff880103828e00 bytes_req=384 bytes_alloc=512 gfp_flags=GFP_KERNEL|GFP_NOWARN|GFP_REPEAT|GFP_NOMEMALLOC node=-1
+               X   888 [002] 29643.445546: kmem:kmalloc_node: (__alloc_skb+0x7e) call_site=ffffffff8153c67e ptr=0xffff880103828e00 bytes_req=384 bytes_alloc=512 gfp_flags=GFP_KERNEL|GFP_NOWARN|GFP_REPEAT|GFP_NOMEMALLOC node=-1
Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/lib/traceevent/Makefile      |  1 +
 tools/lib/traceevent/plugin_kmem.c | 72 ++++++++++++++++++++++++++++++++++++++
 2 files changed, 73 insertions(+)
 create mode 100644 tools/lib/traceevent/plugin_kmem.c
tools/lib/traceevent/Makefile +1 -0
index df087f3..1ff8bb2 100644
@@ -204,6 +204,7 @@ PEVENT_LIB_OBJS += kbuffer-parse.o
 
 PLUGIN_OBJS  = plugin_jbd2.o
 PLUGIN_OBJS += plugin_hrtimer.o
+PLUGIN_OBJS += plugin_kmem.o
 
 PLUGINS := $(PLUGIN_OBJS:.o=.so)
 
tools/lib/traceevent/plugin_kmem.c new file +72 -0

72 lines not on this page · raw message

Signature · 1 line
1.8.3.1
#25 lore raw

Re: [PATCH 06/22] tools lib traceevent: Add kmem plugin

Quoted 95 lines · On Thu, 21 Nov 2013 12:01:04 +0100 Jiri Olsa <[email protected]> wrote:

Backporting kmem plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds call_site field resolving for following tracepoint events:
kmem:kfree
kmem:kmalloc
kmem:kmalloc_node
kmem:kmem_cache_alloc kmem:kmem_cache_alloc_node kmem:kmem_cache_free

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'kmem:*' -a')

script.kmem.new renamed from script.kmem.old +6 -6
-            perf 27846 [001] 29643.403319: kmem:kfree: call_site=ffffffff810e64f6 ptr=(nil)
+            perf 27846 [001] 29643.403238: kmem:kfree: (__audit_syscall_exit+0x1f6) call_site=ffffffff810e64f6 ptr=(nil)
-            perf 27846 [001] 29643.403337: kmem:kmem_cache_alloc: call_site=ffffffff812ff0c5 ptr=0xffff88020e155630 bytes_req=560 bytes_alloc=568 gfp_flags=GFP_KERNEL
+            perf 27846 [001] 29643.403337: kmem:kmem_cache_alloc: (radix_tree_preload+0x35) call_site=ffffffff812ff0c5 ptr=0xffff88020e155630 bytes_req=560 bytes_alloc=568 gfp_flags=GFP_KERNEL
-            perf 27846 [001] 29643.403342: kmem:kmem_cache_free: call_site=ffffffff8126ec61 ptr=0xffff88020dffe750
+            perf 27846 [001] 29643.403342: kmem:kmem_cache_free: (jbd2_journal_stop+0x221) call_site=ffffffff8126ec61 ptr=0xffff88020dffe750
-         firefox   954 [000] 29643.445477: kmem:kmem_cache_alloc_node: call_site=ffffffff8153c64e ptr=0xffff8801cecb4000 bytes_req=256 bytes_alloc=256 gfp_flags=GFP_KERNEL|GFP_REPEAT node=-1
+         firefox   954 [000] 29643.445477: kmem:kmem_cache_alloc_node: (__alloc_skb+0x4e) call_site=ffffffff8153c64e ptr=0xffff8801cecb4000 bytes_req=256 bytes_alloc=256 gfp_flags=GFP_KERNEL|GFP_REPEAT node=-1
-            perf 27846 [001] 29643.445510: kmem:kmalloc: call_site=ffffffff81250642 ptr=0xffff88020fd6c300 bytes_req=96 bytes_alloc=96 gfp_flags=GFP_NOFS|GFP_ZERO
+            perf 27846 [001] 29643.445510: kmem:kmalloc: (ext4_ext_find_extent+0x362) call_site=ffffffff81250642 ptr=0xffff88020fd6c300 bytes_req=96 bytes_alloc=96 gfp_flags=GFP_NOFS|GFP_ZERO
-               X   888 [002] 29643.445546: kmem:kmalloc_node: call_site=ffffffff8153c67e ptr=0xffff880103828e00 bytes_req=384 bytes_alloc=512 gfp_flags=GFP_KERNEL|GFP_NOWARN|GFP_REPEAT|GFP_NOMEMALLOC node=-1
+               X   888 [002] 29643.445546: kmem:kmalloc_node: (__alloc_skb+0x7e) call_site=ffffffff8153c67e ptr=0xffff880103828e00 bytes_req=384 bytes_alloc=512 gfp_flags=GFP_KERNEL|GFP_NOWARN|GFP_REPEAT|GFP_NOMEMALLOC node=-1
Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/lib/traceevent/Makefile      |  1 +
 tools/lib/traceevent/plugin_kmem.c | 72 ++++++++++++++++++++++++++++++++++++++
 2 files changed, 73 insertions(+)
 create mode 100644 tools/lib/traceevent/plugin_kmem.c
tools/lib/traceevent/Makefile +1 -0
index df087f3..1ff8bb2 100644
@@ -204,6 +204,7 @@ PEVENT_LIB_OBJS += kbuffer-parse.o
 
 PLUGIN_OBJS  = plugin_jbd2.o
 PLUGIN_OBJS += plugin_hrtimer.o
+PLUGIN_OBJS += plugin_kmem.o
 
 PLUGINS := $(PLUGIN_OBJS:.o=.so)
 
tools/lib/traceevent/plugin_kmem.c new file +29 -0
new file mode 100644
index 0000000..7115c80
@@ -0,0 +1,72 @@
+/*
+ * Copyright (C) 2009 Red Hat Inc, Steven Rostedt <[email protected]>
+ *
+ * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License (not later!)
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this program; if not,  see <http://www.gnu.org/licenses>
+ *
+ * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ */
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "event-parse.h"
+
+static int call_site_handler(struct trace_seq *s, struct pevent_record *record,
+			     struct event_format *event, void *context)
+{
+	struct format_field *field;

tools/lib/traceevent/plugin_kmem.c line 39

Quoted 10 lines
tools/lib/traceevent/plugin_kmem.c new file +10 -0
+	unsigned long long val, addr;
+	void *data = record->data;
+	const char *func;
+
+	field = pevent_find_field(event, "call_site");
+	if (!field)
+		return 1;
+
+	if (pevent_read_number_field(field, data, &val))
+		return 1;

This was one of the older plugins. We can simplify this with:

	if (pevent_get_field_val(s, event, "call_site",
			 record, &val, 0))
		return 1;

Would would replace the above two checks.

But we can fix this later. As I'm sure there's other places that can use these updates.

-- Steve

Quoted 33 lines
tools/lib/traceevent/plugin_kmem.c new file +33 -0
+
+	func = pevent_find_function(event->pevent, val);
+	if (!func)
+		return 1;
+
+	addr = pevent_find_function_address(event->pevent, val);
+
+	trace_seq_printf(s, "(%s+0x%x) ", func, (int)(val - addr));
+	return 1;
+}
+
+int PEVENT_PLUGIN_LOADER(struct pevent *pevent)
+{
+	pevent_register_event_handler(pevent, -1, "kmem", "kfree",
+				      call_site_handler, NULL);
+
+	pevent_register_event_handler(pevent, -1, "kmem", "kmalloc",
+				      call_site_handler, NULL);
+
+	pevent_register_event_handler(pevent, -1, "kmem", "kmalloc_node",
+				      call_site_handler, NULL);
+
+	pevent_register_event_handler(pevent, -1, "kmem", "kmem_cache_alloc",
+				      call_site_handler, NULL);
+
+	pevent_register_event_handler(pevent, -1, "kmem",
+				      "kmem_cache_alloc_node",
+				      call_site_handler, NULL);
+
+	pevent_register_event_handler(pevent, -1, "kmem", "kmem_cache_free",
+				      call_site_handler, NULL);
+	return 0;
+}
maintainer
#26 lore raw

Re: [PATCH 06/22] tools lib traceevent: Add kmem plugin

In reply to Jiri Olsa

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 21 lines

Backporting kmem plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds call_site field resolving for following tracepoint events:
kmem:kfree
kmem:kmalloc
kmem:kmalloc_node
kmem:kmem_cache_alloc kmem:kmem_cache_alloc_node kmem:kmem_cache_free

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'kmem:*' -a')

script.kmem.new renamed from script.kmem.old +1 -1
-            perf 27846 [001] 29643.403319: kmem:kfree: call_site=ffffffff810e64f6 ptr=(nil)
+            perf 27846 [001] 29643.403238: kmem:kfree: (__audit_syscall_exit+0x1f6) call_site=ffffffff810e64f6 ptr=(nil)

script.kmem.new

Quoted 10 lines
script.kmem.new renamed from script.kmem.old +5 -5
-            perf 27846 [001] 29643.403337: kmem:kmem_cache_alloc: call_site=ffffffff812ff0c5 ptr=0xffff88020e155630 bytes_req=560 bytes_alloc=568 gfp_flags=GFP_KERNEL
+            perf 27846 [001] 29643.403337: kmem:kmem_cache_alloc: (radix_tree_preload+0x35) call_site=ffffffff812ff0c5 ptr=0xffff88020e155630 bytes_req=560 bytes_alloc=568 gfp_flags=GFP_KERNEL
-            perf 27846 [001] 29643.403342: kmem:kmem_cache_free: call_site=ffffffff8126ec61 ptr=0xffff88020dffe750
+            perf 27846 [001] 29643.403342: kmem:kmem_cache_free: (jbd2_journal_stop+0x221) call_site=ffffffff8126ec61 ptr=0xffff88020dffe750
-         firefox   954 [000] 29643.445477: kmem:kmem_cache_alloc_node: call_site=ffffffff8153c64e ptr=0xffff8801cecb4000 bytes_req=256 bytes_alloc=256 gfp_flags=GFP_KERNEL|GFP_REPEAT node=-1
+         firefox   954 [000] 29643.445477: kmem:kmem_cache_alloc_node: (__alloc_skb+0x4e) call_site=ffffffff8153c64e ptr=0xffff8801cecb4000 bytes_req=256 bytes_alloc=256 gfp_flags=GFP_KERNEL|GFP_REPEAT node=-1
-            perf 27846 [001] 29643.445510: kmem:kmalloc: call_site=ffffffff81250642 ptr=0xffff88020fd6c300 bytes_req=96 bytes_alloc=96 gfp_flags=GFP_NOFS|GFP_ZERO
+            perf 27846 [001] 29643.445510: kmem:kmalloc: (ext4_ext_find_extent+0x362) call_site=ffffffff81250642 ptr=0xffff88020fd6c300 bytes_req=96 bytes_alloc=96 gfp_flags=GFP_NOFS|GFP_ZERO
-               X   888 [002] 29643.445546: kmem:kmalloc_node: call_site=ffffffff8153c67e ptr=0xffff880103828e00 bytes_req=384 bytes_alloc=512 gfp_flags=GFP_KERNEL|GFP_NOWARN|GFP_REPEAT|GFP_NOMEMALLOC node=-1
+               X   888 [002] 29643.445546: kmem:kmalloc_node: (__alloc_skb+0x7e) call_site=ffffffff8153c67e ptr=0xffff880103828e00 bytes_req=384 bytes_alloc=512 gfp_flags=GFP_KERNEL|GFP_NOWARN|GFP_REPEAT|GFP_NOMEMALLOC node=-1

It'd be great if the "call_site" in the output changes to display function names instead of hex addresses directly.

Thanks,
Namhyung

#27 lore raw

Re: [PATCH 06/22] tools lib traceevent: Add kmem plugin

Quoted 27 lines · On Fri, 22 Nov 2013 23:38:17 +0900 Namhyung Kim <[email protected]> wrote:

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 26 lines

Backporting kmem plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds call_site field resolving for following tracepoint events:
kmem:kfree
kmem:kmalloc
kmem:kmalloc_node
kmem:kmem_cache_alloc kmem:kmem_cache_alloc_node kmem:kmem_cache_free

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'kmem:*' -a')

script.kmem.new renamed from script.kmem.old +3 -4
-            perf 27846 [001] 29643.403319: kmem:kfree: call_site=ffffffff810e64f6 ptr=(nil)
+            perf 27846 [001] 29643.403238: kmem:kfree: (__audit_syscall_exit+0x1f6) call_site=ffffffff810e64f6 ptr=(nil)
-            perf 27846 [001] 29643.403337: kmem:kmem_cache_alloc: call_site=ffffffff812ff0c5 ptr=0xffff88020e155630 bytes_req=560 bytes_alloc=568 gfp_flags=GFP_KERNEL
+            perf 27846 [001] 29643.403337: kmem:kmem_cache_alloc: (radix_tree_preload+0x35) call_site=ffffffff812ff0c5 ptr=0xffff88020e155630 bytes_req=560 bytes_alloc=568 gfp_flags=GFP_KERNEL
-            perf 27846 [001] 29643.403342: kmem:kmem_cache_free: call_site=ffffffff8126ec61 ptr=0xffff88020dffe750
+            perf 27846 [001] 29643.403342: kmem:kmem_cache_free: (jbd2_journal_stop+0x221) call_site=ffffffff8126ec61 ptr=0xffff88020dffe750
-         firefox   954 [000] 29643.445477: kmem:kmem_cache_alloc_node: call_site=ffffffff8153c64e ptr=0xffff8801cecb4000 bytes_req=256 bytes_alloc=256 gfp_flags=GFP_KERNEL|GFP_REPEAT node=-1
Quoted 9 lines

script.kmem.new

Quoted 5 lines
script.kmem.new renamed from script.kmem.old +3 -2
+         firefox   954 [000] 29643.445477: kmem:kmem_cache_alloc_node: (__alloc_skb+0x4e) call_site=ffffffff8153c64e ptr=0xffff8801cecb4000 bytes_req=256 bytes_alloc=256 gfp_flags=GFP_KERNEL|GFP_REPEAT node=-1
-            perf 27846 [001] 29643.445510: kmem:kmalloc: call_site=ffffffff81250642 ptr=0xffff88020fd6c300 bytes_req=96 bytes_alloc=96 gfp_flags=GFP_NOFS|GFP_ZERO
+            perf 27846 [001] 29643.445510: kmem:kmalloc: (ext4_ext_find_extent+0x362) call_site=ffffffff81250642 ptr=0xffff88020fd6c300 bytes_req=96 bytes_alloc=96 gfp_flags=GFP_NOFS|GFP_ZERO
-               X   888 [002] 29643.445546: kmem:kmalloc_node: call_site=ffffffff8153c67e ptr=0xffff880103828e00 bytes_req=384 bytes_alloc=512 gfp_flags=GFP_KERNEL|GFP_NOWARN|GFP_REPEAT|GFP_NOMEMALLOC node=-1
+               X   888 [002] 29643.445546: kmem:kmalloc_node: (__alloc_skb+0x7e) call_site=ffffffff8153c67e ptr=0xffff880103828e00 bytes_req=384 bytes_alloc=512 gfp_flags=GFP_KERNEL|GFP_NOWARN|GFP_REPEAT|GFP_NOMEMALLOC node=-1

It'd be great if the "call_site" in the output changes to display function names instead of hex addresses directly.

Actually, that's what's in the ().

kmem:kmalloc_node: (__alloc_skb+0x7e) call_site=ffffffff8153c67e

This uses a short cut, where we don't overwrite the entire handler, in case the TP_printk() gets new fields.

If the registered handler for an event, like "call_site_handler" (see how we use it for all of tracepoints) returns >0, that tells the library that we only added extra information, and to print the tracepoint as it is normally.

The better solution here is to use "%pS" or something in the actual tracepoint instead.

-- Steve

maintainer
#28 lore raw

Re: [PATCH 06/22] tools lib traceevent: Add kmem plugin

Quoted 10 lines · Steven Rostedt
Quoted 3 lines · On Fri, 22 Nov 2013 23:38:17 +0900 Namhyung Kim <[email protected]> wrote:

It'd be great if the "call_site" in the output changes to display function names instead of hex addresses directly.

Actually, that's what's in the ().

   kmem:kmalloc_node: (__alloc_skb+0x7e) call_site=ffffffff8153c67e
Quoted 7 lines

This uses a short cut, where we don't overwrite the entire handler, in case the TP_printk() gets new fields.

If the registered handler for an event, like "call_site_handler" (see how we use it for all of tracepoints) returns >0, that tells the library that we only added extra information, and to print the tracepoint as it is normally.

Yeah, I know. But just want to say that it'd be better if it's displayed like below.

kmem:kmalloc_node: call_site=__alloc_skb+0x7e ...

But it requires to write new handlers for each event..

Quoted 3 lines

The better solution here is to use "%pS" or something in the actual tracepoint instead.

Agreed.

Thanks,
Namhyung

author
#29 lore raw

PATCH 7/22 tools lib traceevent: Add kvm plugin

In reply to Jiri Olsa

Backporting kvm plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds field resolving functions for following tracepoint events:
kvm:kvm_exit
kvm:kvm_emulate_insn
kvm:kvm_nested_vmexit kvm:kvm_nested_vmexit_inject kvmmmu:kvm_mmu_get_page kvmmmu:kvm_mmu_sync_page kvmmmu:kvm_mmu_unsync_page kvmmmu:kvm_mmu_zap_page kvmmmu:kvm_mmu_prepare_zap_page

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'kvm:*,kvmmmu:*' -a')

--- script.kvm.old
+++ script.kvm.new
qemu-system-x86 17414 [000] 6868.995053: kvm:kvm_exit: reason EPT_VIOLATION rip 0xfff0 info 184 0 qemu-system-x86 17414 [000] 6868.995109: kvm:kvm_emulate_insn: f0000:c46b:e4 71 (real)

- qemu-system-x86  3006 [002] 10562.079422: kvmmmu:kvm_mmu_get_page: [FAILED TO PARSE] mmu_valid_gen=0x2 gfn=0 role=122884 root_count=0 unsync=0 created=1
+ qemu-system-x86  3006 [002] 10562.079422: kvmmmu:kvm_mmu_get_page: new sp gfn 0 0/4 q0 direct --- !pge !nxe root 0 sync
- qemu-system-x86  3006 [002] 10562.080502: kvmmmu:kvm_mmu_prepare_zap_page: [FAILED TO PARSE] mmu_valid_gen=0x2 gfn=0 role=122884 root_count=1 unsync=0
+ qemu-system-x86  3006 [002] 10562.080502: kvmmmu:kvm_mmu_prepare_zap_page: 0/4 q0 direct --- !pge !nxe root 1 sync
  qemu-system-x86  3290 [002] 10708.755312: kvmmmu:fast_page_fault: [FAILED TO PARSE] vcpu_id=0 gva=4094486080 error_code=3 sptep=0xffff88019f1e3670 old_spte=336391285 new_spte=336391287 retry=1
-          insmod  2576 [001]   781.731666: kvmmmu:kvm_mmu_sync_page: [FAILED TO PARSE] mmu_valid_gen=0x1 gfn=2 role=24624 root_count=10 unsync=1
+          insmod  2576 [001]   781.731666: kvmmmu:kvm_mmu_sync_page: 3/0 q0 --- !pge !nxe root 10 unsync
-          insmod  2576 [001]   781.731668: kvmmmu:kvm_mmu_unsync_page: [FAILED TO PARSE] mmu_valid_gen=0x1 gfn=2 role=24624 root_count=10 unsync=1
+          insmod  2576 [001]   781.731668: kvmmmu:kvm_mmu_unsync_page: 3/0 q0 --- !pge !nxe root 10 unsync

Note:
- kvm_mmu_zap_page is replaced by kvm_mmu_prepare_zap_page in current kernel, keeping it for backward compatibility
- some of the tracepoints keep the same output even with the plugin handling: kvm:kvm_exit, kvm:kvm_emulate_insn
- the 'kvmmmu:fast_page_fault' is still broken because of missing is_writable_pte function and is fixed in another patch
- ommited following tracepoints from backport because the output was buggy
kvm:kvm_nested_vmexit
kvm:kvm_nested_vmexit_inject

Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/lib/traceevent/Makefile     |   1 +
 tools/lib/traceevent/plugin_kvm.c | 419 ++++++++++++++++++++++++++++++++++++++
 2 files changed, 420 insertions(+)
 create mode 100644 tools/lib/traceevent/plugin_kvm.c
tools/lib/traceevent/Makefile +1 -0
index 1ff8bb2..2ed515f 100644
@@ -205,6 +205,7 @@ PEVENT_LIB_OBJS += kbuffer-parse.o
 PLUGIN_OBJS  = plugin_jbd2.o
 PLUGIN_OBJS += plugin_hrtimer.o
 PLUGIN_OBJS += plugin_kmem.o
+PLUGIN_OBJS += plugin_kvm.o
 
 PLUGINS := $(PLUGIN_OBJS:.o=.so)
 
tools/lib/traceevent/plugin_kvm.c new file +419 -0

419 lines not on this page · raw message

Signature · 1 line
1.8.3.1
#30 lore raw

Re: [PATCH 07/22] tools lib traceevent: Add kvm plugin

Quoted 46 lines · On Thu, 21 Nov 2013 12:01:05 +0100 Jiri Olsa <[email protected]> wrote:

Backporting kvm plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds field resolving functions for following tracepoint events:
kvm:kvm_exit
kvm:kvm_emulate_insn
kvm:kvm_nested_vmexit kvm:kvm_nested_vmexit_inject kvmmmu:kvm_mmu_get_page kvmmmu:kvm_mmu_sync_page kvmmmu:kvm_mmu_unsync_page kvmmmu:kvm_mmu_zap_page kvmmmu:kvm_mmu_prepare_zap_page

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'kvm:*,kvmmmu:*' -a')

script.kvm.new renamed from script.kvm.old +4 -4
  qemu-system-x86 17414 [000]  6868.995053: kvm:kvm_exit: reason EPT_VIOLATION rip 0xfff0 info 184 0
  qemu-system-x86 17414 [000]  6868.995109: kvm:kvm_emulate_insn: f0000:c46b:e4 71 (real)
- qemu-system-x86  3006 [002] 10562.079422: kvmmmu:kvm_mmu_get_page: [FAILED TO PARSE] mmu_valid_gen=0x2 gfn=0 role=122884 root_count=0 unsync=0 created=1
+ qemu-system-x86  3006 [002] 10562.079422: kvmmmu:kvm_mmu_get_page: new sp gfn 0 0/4 q0 direct --- !pge !nxe root 0 sync
- qemu-system-x86  3006 [002] 10562.080502: kvmmmu:kvm_mmu_prepare_zap_page: [FAILED TO PARSE] mmu_valid_gen=0x2 gfn=0 role=122884 root_count=1 unsync=0
+ qemu-system-x86  3006 [002] 10562.080502: kvmmmu:kvm_mmu_prepare_zap_page: 0/4 q0 direct --- !pge !nxe root 1 sync
  qemu-system-x86  3290 [002] 10708.755312: kvmmmu:fast_page_fault: [FAILED TO PARSE] vcpu_id=0 gva=4094486080 error_code=3 sptep=0xffff88019f1e3670 old_spte=336391285 new_spte=336391287 retry=1
-          insmod  2576 [001]   781.731666: kvmmmu:kvm_mmu_sync_page: [FAILED TO PARSE] mmu_valid_gen=0x1 gfn=2 role=24624 root_count=10 unsync=1
+          insmod  2576 [001]   781.731666: kvmmmu:kvm_mmu_sync_page: 3/0 q0 --- !pge !nxe root 10 unsync
-          insmod  2576 [001]   781.731668: kvmmmu:kvm_mmu_unsync_page: [FAILED TO PARSE] mmu_valid_gen=0x1 gfn=2 role=24624 root_count=10 unsync=1
+          insmod  2576 [001]   781.731668: kvmmmu:kvm_mmu_unsync_page: 3/0 q0 --- !pge !nxe root 10 unsync

Note:
- kvm_mmu_zap_page is replaced by kvm_mmu_prepare_zap_page in current kernel, keeping it for backward compatibility
- some of the tracepoints keep the same output even with the plugin handling: kvm:kvm_exit, kvm:kvm_emulate_insn
- the 'kvmmmu:fast_page_fault' is still broken because of missing is_writable_pte function and is fixed in another patch
- ommited following tracepoints from backport because the output was buggy
kvm:kvm_nested_vmexit
kvm:kvm_nested_vmexit_inject

Quoted 9 lines
Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>

-- Steve

Quoted 2 lines
Cc David Ahern <[email protected]>

---

maintainer
#31 lore raw

Re: [PATCH 07/22] tools lib traceevent: Add kvm plugin

In reply to Jiri Olsa

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 45 lines

Backporting kvm plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds field resolving functions for following tracepoint events:
kvm:kvm_exit
kvm:kvm_emulate_insn
kvm:kvm_nested_vmexit kvm:kvm_nested_vmexit_inject kvmmmu:kvm_mmu_get_page kvmmmu:kvm_mmu_sync_page kvmmmu:kvm_mmu_unsync_page kvmmmu:kvm_mmu_zap_page kvmmmu:kvm_mmu_prepare_zap_page

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'kvm:*,kvmmmu:*' -a')

script.kvm.new renamed from script.kvm.old +4 -4
  qemu-system-x86 17414 [000]  6868.995053: kvm:kvm_exit: reason EPT_VIOLATION rip 0xfff0 info 184 0
  qemu-system-x86 17414 [000]  6868.995109: kvm:kvm_emulate_insn: f0000:c46b:e4 71 (real)
- qemu-system-x86  3006 [002] 10562.079422: kvmmmu:kvm_mmu_get_page: [FAILED TO PARSE] mmu_valid_gen=0x2 gfn=0 role=122884 root_count=0 unsync=0 created=1
+ qemu-system-x86  3006 [002] 10562.079422: kvmmmu:kvm_mmu_get_page: new sp gfn 0 0/4 q0 direct --- !pge !nxe root 0 sync
- qemu-system-x86  3006 [002] 10562.080502: kvmmmu:kvm_mmu_prepare_zap_page: [FAILED TO PARSE] mmu_valid_gen=0x2 gfn=0 role=122884 root_count=1 unsync=0
+ qemu-system-x86  3006 [002] 10562.080502: kvmmmu:kvm_mmu_prepare_zap_page: 0/4 q0 direct --- !pge !nxe root 1 sync
  qemu-system-x86  3290 [002] 10708.755312: kvmmmu:fast_page_fault: [FAILED TO PARSE] vcpu_id=0 gva=4094486080 error_code=3 sptep=0xffff88019f1e3670 old_spte=336391285 new_spte=336391287 retry=1
-          insmod  2576 [001]   781.731666: kvmmmu:kvm_mmu_sync_page: [FAILED TO PARSE] mmu_valid_gen=0x1 gfn=2 role=24624 root_count=10 unsync=1
+          insmod  2576 [001]   781.731666: kvmmmu:kvm_mmu_sync_page: 3/0 q0 --- !pge !nxe root 10 unsync
-          insmod  2576 [001]   781.731668: kvmmmu:kvm_mmu_unsync_page: [FAILED TO PARSE] mmu_valid_gen=0x1 gfn=2 role=24624 root_count=10 unsync=1
+          insmod  2576 [001]   781.731668: kvmmmu:kvm_mmu_unsync_page: 3/0 q0 --- !pge !nxe root 10 unsync

Note:
- kvm_mmu_zap_page is replaced by kvm_mmu_prepare_zap_page in current kernel, keeping it for backward compatibility
- some of the tracepoints keep the same output even with the plugin handling: kvm:kvm_exit, kvm:kvm_emulate_insn
- the 'kvmmmu:fast_page_fault' is still broken because of missing is_writable_pte function and is fixed in another patch
- ommited following tracepoints from backport because the output was buggy
kvm:kvm_nested_vmexit
kvm:kvm_nested_vmexit_inject

[SNIP]

Quoted 28 lines
script.kvm.new renamed from script.kvm.old +28 -0
+static int kvm_mmu_print_role(struct trace_seq *s, struct pevent_record *record,
+			      struct event_format *event, void *context)
+{
+	unsigned long long val;
+	static const char *access_str[] = {
+		"---", "--x", "w--", "w-x", "-u-", "-ux", "wu-", "wux"
+	};
+	union kvm_mmu_page_role role;
+
+	if (pevent_get_field_val(s, event, "role", record, &val, 1) < 0)
+		return -1;
+
+	role.word = (int)val;
+
+	/*
+	 * We can only use the structure if file is of the same
+	 * endianess.
+	 */
+	if (pevent_is_file_bigendian(event->pevent) ==
+	    pevent_is_host_bigendian(event->pevent)) {
+
+		trace_seq_printf(s, "%u/%u q%u%s %s%s %spge %snxe",
+				 role.level,
+				 role.glevels,
+				 role.quadrant,
+				 role.direct ? " direct" : "",
+				 access_str[role.access],
+				 role.invalid ? " invalid" : "",

script.kvm.new

Quoted 10 lines
script.kvm.new renamed from script.kvm.old +10 -0
+				 role.cr4_pge ? "" : "!",
+				 role.nxe ? "" : "!");
+	} else
+		trace_seq_printf(s, "WORD: %08x", role.word);
+
+	pevent_print_num_field(s, " root %u ",  event,
+			       "root_count", record, 1);
+
+	if (pevent_get_field_val(s, event, "unsync", record, &val, 1) < 0)
+		return -1;

Hmm.. it seems this returning from the middle of function can make output hard to parse. How about printing 'unknown' if the field not found?

script.kvm.new

Quoted 2 lines
script.kvm.new renamed from script.kvm.old +2 -0
+
+	trace_seq_printf(s, "%s%c",  val ? "unsync" : "sync", 0);

Why did you print %c (0) at the end?

Thanks,
Namhyung

Quoted 2 lines
script.kvm.new renamed from script.kvm.old +2 -0
+	return 0;
+}
#32 lore raw

Re: [PATCH 07/22] tools lib traceevent: Add kvm plugin

Quoted 9 lines · On Fri, 22 Nov 2013 23:45:24 +0900 Namhyung Kim <[email protected]> wrote:

tools/lib/traceevent/plugin_kvm.c line 368

Quoted 5 lines
tools/lib/traceevent/plugin_kvm.c new file +5 -0
+	pevent_print_num_field(s, " root %u ",  event,
+			       "root_count", record, 1);
+
+	if (pevent_get_field_val(s, event, "unsync", record, &val, 1) < 0)
+		return -1;

Hmm.. it seems this returning from the middle of function can make output hard to parse. How about printing 'unknown' if the field not found?

I wasn't fully correct in my last email. If the handler returns anything but zero, the TP_print() will go in affect. Thus, if this fails, then the default print will happen anyway.

It shouldn't fail. But yeah, we could think of adding "unknow" and such too. That can go as a separate patch.

Quoted 5 lines

tools/lib/traceevent/plugin_kvm.c line 370

Quoted 3 lines
tools/lib/traceevent/plugin_kvm.c new file +2 -0
+
+	trace_seq_printf(s, "%s%c",  val ? "unsync" : "sync", 0);

Why did you print %c (0) at the end?

I simulated exactly what was (and still is) in the kvm tracepoint:

arch/x86/kvm/mmutrace.h:

        trace_seq_printf(p, "sp gfn %llx %u%s q%u%s %s%s"               \
                         " %snxe root %u %s%c",                         \
                         __entry->gfn, role.level,                      \
                         role.cr4_pae ? " pae" : "",                    \
                         role.quadrant,                                 \
                         role.direct ? " direct" : "",                  \
                         access_str[role.access],                       \
                         role.invalid ? " invalid" : "",                \
                         role.nxe ? "" : "!",                           \
                         __entry->root_count,                           \
                         __entry->unsync ? "unsync" : "sync", 0);	\

-- Steve

Quoted 8 lines

Thanks,
Namhyung

Quoted 2 lines
(quoted diff) +2 -0
+	return 0;
+}
author
#33 lore raw

PATCH 8/22 tools lib traceevent: Add mac80211 plugin

In reply to Jiri Olsa

Backporting mac80211 plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds changed field resolving for mac80211:drv_bss_info_changed tracepoint event.

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'mac80211:drv_bss_info_changed' -a')

--- script.mac80211.old
+++ script.mac80211.new

-        ifconfig  3711 [000]  1290.446492: mac80211:drv_bss_info_changed: phy0 vif:wlan0(2) changed:0x309f
+        ifconfig  3711 [000]  1290.446492: mac80211:drv_bss_info_changed: phy0 vif:wlan0(2)
+                                                                 assoc:0 aid:2 cts:0 shortpre:0 shortslot:0 dtimper:1
+                                                                 bcnint:102 assoc_cap:0x431 basic_rates:0xf enable_beacon:0
+                                                                 ht_operation_mode:0

Omitting the mac80211:drv_config tracepoint handling because the kernel tracepoint changed its prototype and the plugin handler is no longer working.

Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/lib/traceevent/Makefile          |  1 +
 tools/lib/traceevent/plugin_mac80211.c | 95 ++++++++++++++++++++++++++++++++++
 2 files changed, 96 insertions(+)
 create mode 100644 tools/lib/traceevent/plugin_mac80211.c
tools/lib/traceevent/Makefile +1 -0
index 2ed515f..a26c89e 100644
@@ -206,6 +206,7 @@ PLUGIN_OBJS  = plugin_jbd2.o
 PLUGIN_OBJS += plugin_hrtimer.o
 PLUGIN_OBJS += plugin_kmem.o
 PLUGIN_OBJS += plugin_kvm.o
+PLUGIN_OBJS += plugin_mac80211.o
 
 PLUGINS := $(PLUGIN_OBJS:.o=.so)
 
tools/lib/traceevent/plugin_mac80211.c new file +95 -0

95 lines not on this page · raw message

Signature · 1 line
1.8.3.1
author
#34 lore raw

PATCH 9/22 tools lib traceevent: Add sched_switch plugin

In reply to Jiri Olsa

Backporting sched_switch plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds fields resolving for sched:sched_switch tracepoint event.

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e sched:sched_switch -a')

--- script.sched_switch.old
+++ script.sched_switch.new

-            perf   577 [002] 30965.311852: sched:sched_switch: prev_comm=perf prev_pid=577 prev_prio=120 prev_state=S ==> next_comm=swapper/2 next_pid=0 next_prio=120
+            perf   577 [002] 30965.311852: sched:sched_switch: perf:577 [120] S ==> swapper/2:0 [120]
Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/lib/traceevent/Makefile              |   1 +
 tools/lib/traceevent/plugin_sched_switch.c | 148 +++++++++++++++++++++++++++++
 2 files changed, 149 insertions(+)
 create mode 100644 tools/lib/traceevent/plugin_sched_switch.c
tools/lib/traceevent/Makefile +1 -0
index a26c89e..1dc6934 100644
@@ -207,6 +207,7 @@ PLUGIN_OBJS += plugin_hrtimer.o
 PLUGIN_OBJS += plugin_kmem.o
 PLUGIN_OBJS += plugin_kvm.o
 PLUGIN_OBJS += plugin_mac80211.o
+PLUGIN_OBJS += plugin_sched_switch.o
 
 PLUGINS := $(PLUGIN_OBJS:.o=.so)
 
tools/lib/traceevent/plugin_sched_switch.c new file +148 -0

148 lines not on this page · raw message

Signature · 1 line
1.8.3.1
#35 lore raw

Re: [PATCH 09/22] tools lib traceevent: Add sched_switch plugin

Quoted 16 lines · On Thu, 21 Nov 2013 12:01:07 +0100 Jiri Olsa <[email protected]> wrote:

Backporting sched_switch plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds fields resolving for sched:sched_switch tracepoint event.

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e sched:sched_switch -a')

script.sched_switch.new renamed from script.sched_switch.old +1 -1
-            perf   577 [002] 30965.311852: sched:sched_switch: prev_comm=perf prev_pid=577 prev_prio=120 prev_state=S ==> next_comm=swapper/2 next_pid=0 next_prio=120
+            perf   577 [002] 30965.311852: sched:sched_switch: perf:577 [120] S ==> swapper/2:0 [120]
Quoted 9 lines
Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>

-- Steve

Quoted 7 lines
Cc David Ahern <[email protected]>
---
 tools/lib/traceevent/Makefile              |   1 +
 tools/lib/traceevent/plugin_sched_switch.c | 148 +++++++++++++++++++++++++++++
 2 files changed, 149 insertions(+)
 create mode 100644 tools/lib/traceevent/plugin_sched_switch.c

A message that is not in the archive (sent off-list or not yet received): 1385132034.1747.58.camel@leonhard

#37 lore raw

Re: [PATCH 09/22] tools lib traceevent: Add sched_switch plugin

Quoted 46 lines · On Fri, 22 Nov 2013 23:53:54 +0900 Namhyung Kim <[email protected]> wrote:

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 15 lines

Backporting sched_switch plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds fields resolving for
sched:sched_switch tracepoint event.

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e sched:sched_switch -a')

script.sched_switch.new renamed from script.sched_switch.old +1 -1
-            perf   577 [002] 30965.311852: sched:sched_switch: prev_comm=perf prev_pid=577 prev_prio=120 prev_state=S ==> next_comm=swapper/2 next_pid=0 next_prio=120
+            perf   577 [002] 30965.311852: sched:sched_switch: perf:577 [120] S ==> swapper/2:0 [120]

[SNIP]

Quoted 25 lines
script.sched_switch.new renamed from script.sched_switch.old +25 -0
+
+static int sched_switch_handler(struct trace_seq *s,
+				struct pevent_record *record,
+				struct event_format *event, void *context)
+{
+	struct format_field *field;
+	unsigned long long val;
+
+	if (pevent_get_field_val(s, event, "prev_pid", record, &val, 1))
+		return trace_seq_putc(s, '!');
+
+	field = pevent_find_any_field(event, "prev_comm");
+	if (field) {
+		write_and_save_comm(field, record, s, val);
+		trace_seq_putc(s, ':');
+	}
+	trace_seq_printf(s, "%lld ", val);
+
+	if (pevent_get_field_val(s, event, "prev_prio", record, &val, 0) == 0)
+		trace_seq_printf(s, "[%lld] ", val);
+
+	if (pevent_get_field_val(s,  event, "prev_state", record, &val, 0) == 0)
+		write_state(s, val);
+
+	trace_seq_puts(s, " ==> ");
Quoted 9 lines

script.sched_switch.new

Quoted 3 lines
script.sched_switch.new renamed from script.sched_switch.old +3 -0
+
+	if (pevent_get_field_val(s, event, "next_pid", record, &val, 1))
+		return trace_seq_putc(s, '!');

Like I said in the previous reply, it'd better to avoid returning from the middle of function IMHO. If it's really critical info, check it at the beginning of the function and bail out before print anything.

Well, it shouldn't lose any data, as I said in a previous reply, it will print (reprint) the data as if the plugin didn't exist.

-- Steve

author
#38 lore raw

PATCH 10/22 tools lib traceevent: Add function plugin

In reply to Jiri Olsa

Backporting function plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds function and parent function fields resolving for ftrace:function tracepoint event.

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e ftrace:function ls')

--- script.function.old
+++ script.function.new

-              ls 10781 [001] 32667.291379: ftrace:function:  ffffffff811adb80 <-- ffffffff811afc48
-              ls 10781 [001] 32667.291379: ftrace:function:  ffffffff811b35d0 <-- ffffffff811adb9b
-              ls 10781 [001] 32667.291380: ftrace:function:  ffffffff811b3520 <-- ffffffff811b35e8
-              ls 10781 [001] 32667.291380: ftrace:function:  ffffffff811b2720 <-- ffffffff811b3549
-              ls 10781 [001] 32667.291381: ftrace:function:  ffffffff81297e10 <-- ffffffff811b356c
-              ls 10781 [001] 32667.291381: ftrace:function:  ffffffff81298f40 <-- ffffffff81297e2c
-              ls 10781 [001] 32667.291382: ftrace:function:  ffffffff81076160 <-- ffffffff811afbf0
-              ls 10781 [001] 32667.291383: ftrace:function:  ffffffff811c3eb0 <-- ffffffff811afbfc
-              ls 10781 [001] 32667.291383: ftrace:function:  ffffffff8164e100 <-- ffffffff811c3ed8
-              ls 10781 [001] 32667.291384: ftrace:function:  ffffffff811a5d10 <-- ffffffff811c3f53
-              ls 10781 [001] 32667.291384: ftrace:function:  ffffffff811e8e70 <-- ffffffff811a5d58
-              ls 10781 [001] 32667.291385: ftrace:function:  ffffffff811f38e0 <-- ffffffff811a5d63
-              ls 10781 [001] 32667.291385: ftrace:function:  ffffffff811a9ff0 <-- ffffffff811a5d6b
-              ls 10781 [001] 32667.291386: ftrace:function:  ffffffff811a9fa0 <-- ffffffff811aa015
-              ls 10781 [001] 32667.291387: ftrace:function:  ffffffff810851c0 <-- ffffffff811aa053
-              ls 10781 [001] 32667.291387: ftrace:function:  ffffffff81090e00 <-- ffffffff81085211
+              ls 10781 [001] 32667.291379: ftrace:function: would_dump <-- setup_new_exec
+              ls 10781 [001] 32667.291379: ftrace:function:    inode_permission <-- would_dump
+              ls 10781 [001] 32667.291380: ftrace:function:       __inode_permission <-- inode_permission
+              ls 10781 [001] 32667.291380: ftrace:function:          generic_permission <-- __inode_permission
+              ls 10781 [001] 32667.291381: ftrace:function:          security_inode_permission <-- __inode_permission
+              ls 10781 [001] 32667.291381: ftrace:function:             cap_inode_permission <-- security_inode_permission
+              ls 10781 [001] 32667.291382: ftrace:function: flush_signal_handlers <-- setup_new_exec
+              ls 10781 [001] 32667.291383: ftrace:function: do_close_on_exec <-- setup_new_exec
+              ls 10781 [001] 32667.291383: ftrace:function:    _raw_spin_lock <-- do_close_on_exec
+              ls 10781 [001] 32667.291384: ftrace:function:    filp_close <-- do_close_on_exec
+              ls 10781 [001] 32667.291384: ftrace:function:       dnotify_flush <-- filp_close
+              ls 10781 [001] 32667.291385: ftrace:function:       locks_remove_posix <-- filp_close
+              ls 10781 [001] 32667.291385: ftrace:function:       fput <-- filp_close
+              ls 10781 [001] 32667.291386: ftrace:function:          file_sb_list_del <-- fput
+              ls 10781 [001] 32667.291387: ftrace:function:          task_work_add <-- fput
+              ls 10781 [001] 32667.291387: ftrace:function:             kick_process <-- task_work_add

Removing options support as it's not backported yet. Currently this plugin supports 2 options:
'parent' to display parent function 'indent' to show function call indents

Enabling both of them by default.

Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/lib/traceevent/Makefile          |   1 +
 tools/lib/traceevent/plugin_function.c | 151 +++++++++++++++++++++++++++++++++
 2 files changed, 152 insertions(+)
 create mode 100644 tools/lib/traceevent/plugin_function.c
tools/lib/traceevent/Makefile +1 -0
index 1dc6934..5bc4a85 100644
@@ -208,6 +208,7 @@ PLUGIN_OBJS += plugin_kmem.o
 PLUGIN_OBJS += plugin_kvm.o
 PLUGIN_OBJS += plugin_mac80211.o
 PLUGIN_OBJS += plugin_sched_switch.o
+PLUGIN_OBJS += plugin_function.o
 
 PLUGINS := $(PLUGIN_OBJS:.o=.so)
 
tools/lib/traceevent/plugin_function.c new file +151 -0

151 lines not on this page · raw message

Signature · 1 line
1.8.3.1
#39 lore raw

Re: [PATCH 10/22] tools lib traceevent: Add function plugin

Quoted 46 lines · On Thu, 21 Nov 2013 12:01:08 +0100 Jiri Olsa <[email protected]> wrote:

Backporting function plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds function and parent function fields resolving for ftrace:function tracepoint event.

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e ftrace:function ls')

script.function.new renamed from script.function.old +16 -16
-              ls 10781 [001] 32667.291379: ftrace:function:  ffffffff811adb80 <-- ffffffff811afc48
-              ls 10781 [001] 32667.291379: ftrace:function:  ffffffff811b35d0 <-- ffffffff811adb9b
-              ls 10781 [001] 32667.291380: ftrace:function:  ffffffff811b3520 <-- ffffffff811b35e8
-              ls 10781 [001] 32667.291380: ftrace:function:  ffffffff811b2720 <-- ffffffff811b3549
-              ls 10781 [001] 32667.291381: ftrace:function:  ffffffff81297e10 <-- ffffffff811b356c
-              ls 10781 [001] 32667.291381: ftrace:function:  ffffffff81298f40 <-- ffffffff81297e2c
-              ls 10781 [001] 32667.291382: ftrace:function:  ffffffff81076160 <-- ffffffff811afbf0
-              ls 10781 [001] 32667.291383: ftrace:function:  ffffffff811c3eb0 <-- ffffffff811afbfc
-              ls 10781 [001] 32667.291383: ftrace:function:  ffffffff8164e100 <-- ffffffff811c3ed8
-              ls 10781 [001] 32667.291384: ftrace:function:  ffffffff811a5d10 <-- ffffffff811c3f53
-              ls 10781 [001] 32667.291384: ftrace:function:  ffffffff811e8e70 <-- ffffffff811a5d58
-              ls 10781 [001] 32667.291385: ftrace:function:  ffffffff811f38e0 <-- ffffffff811a5d63
-              ls 10781 [001] 32667.291385: ftrace:function:  ffffffff811a9ff0 <-- ffffffff811a5d6b
-              ls 10781 [001] 32667.291386: ftrace:function:  ffffffff811a9fa0 <-- ffffffff811aa015
-              ls 10781 [001] 32667.291387: ftrace:function:  ffffffff810851c0 <-- ffffffff811aa053
-              ls 10781 [001] 32667.291387: ftrace:function:  ffffffff81090e00 <-- ffffffff81085211
+              ls 10781 [001] 32667.291379: ftrace:function: would_dump <-- setup_new_exec
+              ls 10781 [001] 32667.291379: ftrace:function:    inode_permission <-- would_dump
+              ls 10781 [001] 32667.291380: ftrace:function:       __inode_permission <-- inode_permission
+              ls 10781 [001] 32667.291380: ftrace:function:          generic_permission <-- __inode_permission
+              ls 10781 [001] 32667.291381: ftrace:function:          security_inode_permission <-- __inode_permission
+              ls 10781 [001] 32667.291381: ftrace:function:             cap_inode_permission <-- security_inode_permission
+              ls 10781 [001] 32667.291382: ftrace:function: flush_signal_handlers <-- setup_new_exec
+              ls 10781 [001] 32667.291383: ftrace:function: do_close_on_exec <-- setup_new_exec
+              ls 10781 [001] 32667.291383: ftrace:function:    _raw_spin_lock <-- do_close_on_exec
+              ls 10781 [001] 32667.291384: ftrace:function:    filp_close <-- do_close_on_exec
+              ls 10781 [001] 32667.291384: ftrace:function:       dnotify_flush <-- filp_close
+              ls 10781 [001] 32667.291385: ftrace:function:       locks_remove_posix <-- filp_close
+              ls 10781 [001] 32667.291385: ftrace:function:       fput <-- filp_close
+              ls 10781 [001] 32667.291386: ftrace:function:          file_sb_list_del <-- fput
+              ls 10781 [001] 32667.291387: ftrace:function:          task_work_add <-- fput
+              ls 10781 [001] 32667.291387: ftrace:function:             kick_process <-- task_work_add
Quoted 3 lines

Removing options support as it's not backported yet. Currently this plugin supports 2 options:
'parent' to display parent function

The number of times I enabled the parent option, I should make it default too.

Thanks!

Quoted 4 lines

'indent' to show function call indents

Enabling both of them by default.

Quoted 9 lines
Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>

-- Steve

Quoted 7 lines
Cc David Ahern <[email protected]>
---
 tools/lib/traceevent/Makefile          |   1 +
 tools/lib/traceevent/plugin_function.c | 151 +++++++++++++++++++++++++++++++++
 2 files changed, 152 insertions(+)
 create mode 100644 tools/lib/traceevent/plugin_function.c
maintainer
#40 lore raw

Re: [PATCH 10/22] tools lib traceevent: Add function plugin

In reply to Jiri Olsa

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 52 lines

Backporting function plugin.

Backported from Steven Rostedt's trace-cmd repo (HEAD 0f2c2fb):
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/trace-cmd.git

This plugin adds function and parent function fields resolving for ftrace:function tracepoint event.

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e ftrace:function ls')

script.function.new renamed from script.function.old +16 -16
-              ls 10781 [001] 32667.291379: ftrace:function:  ffffffff811adb80 <-- ffffffff811afc48
-              ls 10781 [001] 32667.291379: ftrace:function:  ffffffff811b35d0 <-- ffffffff811adb9b
-              ls 10781 [001] 32667.291380: ftrace:function:  ffffffff811b3520 <-- ffffffff811b35e8
-              ls 10781 [001] 32667.291380: ftrace:function:  ffffffff811b2720 <-- ffffffff811b3549
-              ls 10781 [001] 32667.291381: ftrace:function:  ffffffff81297e10 <-- ffffffff811b356c
-              ls 10781 [001] 32667.291381: ftrace:function:  ffffffff81298f40 <-- ffffffff81297e2c
-              ls 10781 [001] 32667.291382: ftrace:function:  ffffffff81076160 <-- ffffffff811afbf0
-              ls 10781 [001] 32667.291383: ftrace:function:  ffffffff811c3eb0 <-- ffffffff811afbfc
-              ls 10781 [001] 32667.291383: ftrace:function:  ffffffff8164e100 <-- ffffffff811c3ed8
-              ls 10781 [001] 32667.291384: ftrace:function:  ffffffff811a5d10 <-- ffffffff811c3f53
-              ls 10781 [001] 32667.291384: ftrace:function:  ffffffff811e8e70 <-- ffffffff811a5d58
-              ls 10781 [001] 32667.291385: ftrace:function:  ffffffff811f38e0 <-- ffffffff811a5d63
-              ls 10781 [001] 32667.291385: ftrace:function:  ffffffff811a9ff0 <-- ffffffff811a5d6b
-              ls 10781 [001] 32667.291386: ftrace:function:  ffffffff811a9fa0 <-- ffffffff811aa015
-              ls 10781 [001] 32667.291387: ftrace:function:  ffffffff810851c0 <-- ffffffff811aa053
-              ls 10781 [001] 32667.291387: ftrace:function:  ffffffff81090e00 <-- ffffffff81085211
+              ls 10781 [001] 32667.291379: ftrace:function: would_dump <-- setup_new_exec
+              ls 10781 [001] 32667.291379: ftrace:function:    inode_permission <-- would_dump
+              ls 10781 [001] 32667.291380: ftrace:function:       __inode_permission <-- inode_permission
+              ls 10781 [001] 32667.291380: ftrace:function:          generic_permission <-- __inode_permission
+              ls 10781 [001] 32667.291381: ftrace:function:          security_inode_permission <-- __inode_permission
+              ls 10781 [001] 32667.291381: ftrace:function:             cap_inode_permission <-- security_inode_permission
+              ls 10781 [001] 32667.291382: ftrace:function: flush_signal_handlers <-- setup_new_exec
+              ls 10781 [001] 32667.291383: ftrace:function: do_close_on_exec <-- setup_new_exec
+              ls 10781 [001] 32667.291383: ftrace:function:    _raw_spin_lock <-- do_close_on_exec
+              ls 10781 [001] 32667.291384: ftrace:function:    filp_close <-- do_close_on_exec
+              ls 10781 [001] 32667.291384: ftrace:function:       dnotify_flush <-- filp_close
+              ls 10781 [001] 32667.291385: ftrace:function:       locks_remove_posix <-- filp_close
+              ls 10781 [001] 32667.291385: ftrace:function:       fput <-- filp_close
+              ls 10781 [001] 32667.291386: ftrace:function:          file_sb_list_del <-- fput
+              ls 10781 [001] 32667.291387: ftrace:function:          task_work_add <-- fput
+              ls 10781 [001] 32667.291387: ftrace:function:             kick_process <-- task_work_add

Removing options support as it's not backported yet. Currently this plugin supports 2 options:
'parent' to display parent function 'indent' to show function call indents

Enabling both of them by default.

[SNIP]

script.function.new

Quoted 2 lines
script.function.new renamed from script.function.old +2 -0
+static struct func_stack {
+	int index;

It seems not used anywhere.

Quoted 10 lines
script.function.new renamed from script.function.old +10 -0
+	int size;
+	char **stack;
+} *fstack;
+
+static int cpus = -1;
+
+#define STK_BLK 10
+
+static void add_child(struct func_stack *stack, const char *child, int pos)
+{

script.function.new

Quoted 10 lines
script.function.new renamed from script.function.old +10 -0
+	int i;
+
+	if (!child)
+		return;
+
+	if (pos < stack->size)
+		free(stack->stack[pos]);
+	else {
+		if (!stack->stack)
+			stack->stack = malloc_or_die(sizeof(char *) * STK_BLK);

Please do not use this function.

script.function.new

Quoted 3 lines
script.function.new renamed from script.function.old +3 -0
+		else
+			stack->stack = realloc(stack->stack, sizeof(char *) *
+					       (stack->size + STK_BLK));

It seems you need to use a local variable to check return value.

script.function.new

Quoted 9 lines
script.function.new renamed from script.function.old +9 -0
+		for (i = stack->size; i < stack->size + STK_BLK; i++)
+			stack->stack[i] = NULL;
+		stack->size += STK_BLK;
+	}
+
+	stack->stack[pos] = strdup(child);
+}
+
+static int get_index(const char *parent, const char *child, int cpu)

It seems this function does not only to get index but also to add function stack. So how about renaming it to something like add_and_get_index()?

script.function.new

Quoted 11 lines
script.function.new renamed from script.function.old +11 -0
+{
+	int i;
+
+	if (cpu < 0)
+		return 0;
+
+	if (cpu > cpus) {
+		if (fstack)
+			fstack = realloc(fstack, sizeof(*fstack) * (cpu + 1));
+		else
+			fstack = malloc_or_die(sizeof(*fstack) * (cpu + 1));

It seems it has same problem as above function.

Quoted 34 lines
script.function.new renamed from script.function.old +34 -0
+
+		/* Account for holes in the cpu count */
+		for (i = cpus + 1; i <= cpu; i++)
+			memset(&fstack[i], 0, sizeof(fstack[i]));
+		cpus = cpu;
+	}
+
+	for (i = 0; i < fstack[cpu].size && fstack[cpu].stack[i]; i++) {
+		if (strcmp(parent, fstack[cpu].stack[i]) == 0) {
+			add_child(&fstack[cpu], child, i+1);
+			return i;
+		}
+	}
+
+	/* Not found */
+	add_child(&fstack[cpu], parent, 0);
+	add_child(&fstack[cpu], child, 1);
+	return 0;
+}
+
+static int function_handler(struct trace_seq *s, struct pevent_record *record,
+			    struct event_format *event, void *context)
+{
+	struct pevent *pevent = event->pevent;
+	unsigned long long function;
+	unsigned long long pfunction;
+	const char *func;
+	const char *parent;
+	int i, index;
+
+	if (pevent_get_field_val(s, event, "ip", record, &function, 1))
+		return trace_seq_putc(s, '!');
+
+	func = pevent_find_function(pevent, function);

script.function.new

Quoted 10 lines
script.function.new renamed from script.function.old +10 -0
+
+	if (pevent_get_field_val(s, event, "parent_ip", record, &pfunction, 1))
+		return trace_seq_putc(s, '!');
+
+	parent = pevent_find_function(pevent, pfunction);
+
+	index = get_index(parent, func, record->cpu);
+
+	for (i = 0; i < index; i++)
+		trace_seq_printf(s, "   ");

What about this?

	trace_seq_printf(s, "%*s", index*3, "");

Thanks,
Namhyung

Quoted 36 lines
script.function.new renamed from script.function.old +36 -0
+
+	if (func)
+		trace_seq_printf(s, "%s", func);
+	else
+		trace_seq_printf(s, "0x%llx", function);
+
+	trace_seq_printf(s, " <-- ");
+	if (parent)
+		trace_seq_printf(s, "%s", parent);
+	else
+		trace_seq_printf(s, "0x%llx", pfunction);
+
+	return 0;
+}
+
+int PEVENT_PLUGIN_LOADER(struct pevent *pevent)
+{
+	pevent_register_event_handler(pevent, -1, "ftrace", "function",
+				      function_handler, NULL);
+	return 0;
+}
+
+void PEVENT_PLUGIN_UNLOADER(void)
+{
+	int i, x;
+
+	for (i = 0; i <= cpus; i++) {
+		for (x = 0; x < fstack[i].size && fstack[i].stack[x]; x++)
+			free(fstack[i].stack[x]);
+		free(fstack[i].stack);
+	}
+
+	free(fstack);
+	fstack = NULL;
+	cpus = -1;
+}
#41 lore raw

Re: [PATCH 10/22] tools lib traceevent: Add function plugin

Quoted 5 lines · On Sat, 23 Nov 2013 00:03:51 +0900 Namhyung Kim <[email protected]> wrote:
Quoted 5 lines

Removing options support as it's not backported yet. Currently this plugin supports 2 options:
'parent' to display parent function 'indent' to show function call indents

Quoted 7 lines
Quoted 1 line

Enabling both of them by default.

[SNIP]

tools/lib/traceevent/plugin_function.c line 28

Quoted 2 lines
tools/lib/traceevent/plugin_function.c new file +2 -0
+static struct func_stack {
+	int index;

It seems not used anywhere.

You're right, it's not! Not sure why I added that. Oh well, again, lets add that as a clean up and keep the backport separate.

Quoted 16 lines
Quoted 16 lines
tools/lib/traceevent/plugin_function.c new file +14 -0
+	int size;
+	char **stack;
+} *fstack;
+
+static int cpus = -1;
+
+#define STK_BLK 10
+
+static void add_child(struct func_stack *stack, const char *child, int pos)
+{
+	int i;
+
+	if (!child)
+		return;
Quoted 8 lines

tools/lib/traceevent/plugin_function.c line 48

Quoted 6 lines
tools/lib/traceevent/plugin_function.c new file +6 -0
+
+	if (pos < stack->size)
+		free(stack->stack[pos]);
+	else {
+		if (!stack->stack)
+			stack->stack = malloc_or_die(sizeof(char *) * STK_BLK);

Please do not use this function.

Again, added as a separate patch (do I hear a theme ;-)

Quoted 84 lines

tools/lib/traceevent/plugin_function.c line 51

Quoted 5 lines
tools/lib/traceevent/plugin_function.c new file +3 -0
+		else
+			stack->stack = realloc(stack->stack, sizeof(char *) *
+					       (stack->size + STK_BLK));

It seems you need to use a local variable to check return value.

tools/lib/traceevent/plugin_function.c line 60

Quoted 9 lines
tools/lib/traceevent/plugin_function.c new file +9 -0
+		for (i = stack->size; i < stack->size + STK_BLK; i++)
+			stack->stack[i] = NULL;
+		stack->size += STK_BLK;
+	}
+
+	stack->stack[pos] = strdup(child);
+}
+
+static int get_index(const char *parent, const char *child, int cpu)

It seems this function does not only to get index but also to add function stack. So how about renaming it to something like add_and_get_index()?

tools/lib/traceevent/plugin_function.c line 71

Quoted 11 lines
tools/lib/traceevent/plugin_function.c new file +11 -0
+{
+	int i;
+
+	if (cpu < 0)
+		return 0;
+
+	if (cpu > cpus) {
+		if (fstack)
+			fstack = realloc(fstack, sizeof(*fstack) * (cpu + 1));
+		else
+			fstack = malloc_or_die(sizeof(*fstack) * (cpu + 1));

It seems it has same problem as above function.

Quoted 34 lines
tools/lib/traceevent/plugin_function.c new file +34 -0
+
+		/* Account for holes in the cpu count */
+		for (i = cpus + 1; i <= cpu; i++)
+			memset(&fstack[i], 0, sizeof(fstack[i]));
+		cpus = cpu;
+	}
+
+	for (i = 0; i < fstack[cpu].size && fstack[cpu].stack[i]; i++) {
+		if (strcmp(parent, fstack[cpu].stack[i]) == 0) {
+			add_child(&fstack[cpu], child, i+1);
+			return i;
+		}
+	}
+
+	/* Not found */
+	add_child(&fstack[cpu], parent, 0);
+	add_child(&fstack[cpu], child, 1);
+	return 0;
+}
+
+static int function_handler(struct trace_seq *s, struct pevent_record *record,
+			    struct event_format *event, void *context)
+{
+	struct pevent *pevent = event->pevent;
+	unsigned long long function;
+	unsigned long long pfunction;
+	const char *func;
+	const char *parent;
+	int i, index;
+
+	if (pevent_get_field_val(s, event, "ip", record, &function, 1))
+		return trace_seq_putc(s, '!');
+
+	func = pevent_find_function(pevent, function);

tools/lib/traceevent/plugin_function.c line 115

Quoted 10 lines
tools/lib/traceevent/plugin_function.c new file +10 -0
+
+	if (pevent_get_field_val(s, event, "parent_ip", record, &pfunction, 1))
+		return trace_seq_putc(s, '!');
+
+	parent = pevent_find_function(pevent, pfunction);
+
+	index = get_index(parent, func, record->cpu);
+
+	for (i = 0; i < index; i++)
+		trace_seq_printf(s, "   ");

What about this?

Quoted 1 line
	trace_seq_printf(s, "%*s", index*3, "");

Nice trick ;-)

I'm fine with all the updates, but lets make that a patch on top.

Thanks,

-- Steve

Quoted 42 lines

Thanks,
Namhyung

Quoted 36 lines
tools/lib/traceevent/plugin_function.c new file +36 -0
+
+	if (func)
+		trace_seq_printf(s, "%s", func);
+	else
+		trace_seq_printf(s, "0x%llx", function);
+
+	trace_seq_printf(s, " <-- ");
+	if (parent)
+		trace_seq_printf(s, "%s", parent);
+	else
+		trace_seq_printf(s, "0x%llx", pfunction);
+
+	return 0;
+}
+
+int PEVENT_PLUGIN_LOADER(struct pevent *pevent)
+{
+	pevent_register_event_handler(pevent, -1, "ftrace", "function",
+				      function_handler, NULL);
+	return 0;
+}
+
+void PEVENT_PLUGIN_UNLOADER(void)
+{
+	int i, x;
+
+	for (i = 0; i <= cpus; i++) {
+		for (x = 0; x < fstack[i].size && fstack[i].stack[x]; x++)
+			free(fstack[i].stack[x]);
+		free(fstack[i].stack);
+	}
+
+	free(fstack);
+	fstack = NULL;
+	cpus = -1;
+}
author
#42 lore raw

PATCH 11/22 tools lib traceevent: Add xen plugin

In reply to Jiri Olsa

Adding xen plugin.

This plugin adds fields resolving for following tracepoint events:
xen:xen_mc_entry xen:xen_mc_extend_args

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'xen:*' ls')

--- script.xen.old
+++ script.xen.new

-         swapper     0 [002]   136.267492: xen:xen_mc_entry: [FAILED TO PARSE] op=3 nargs=2 args=ARRAY[18, 00, 00, 00, 00, 00, 00, 00, 00, e0, d4, 4b, 04, 88, ff, ff, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00]
+         swapper     0 [002]   136.267492: xen:xen_mc_entry: op 3(stack_switch) args [18, 0, 0, 0, 0, 0]
-            perf  1970 [008]   136.273319: xen:xen_mc_extend_args: [FAILED TO PARSE] op=1 args=16 res=1
+            perf  1970 [008]   136.273319: xen:xen_mc_extend_args: extending op 1(mmu_update) by 16 bytes res ???

NOTE We still do not handle the 'sizeof' and fail to parse following xen tracepoints:
xen:xen_mmu_set_pte xen:xen_mmu_set_pte_atomic xen:xen_mmu_set_domain_pte xen:xen_mmu_set_pte_at xen:xen_mmu_set_pmd
xen:xen_mmu_set_pud
xen:xen_mmu_set_pgd xen:xen_mmu_ptep_modify_prot_start xen:xen_mmu_ptep_modify_prot_commit

Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
Cc Jeremy Fitzhardinge <[email protected]>
---
 tools/lib/traceevent/Makefile     |   1 +
 tools/lib/traceevent/plugin_xen.c | 130 ++++++++++++++++++++++++++++++++++++++
 2 files changed, 131 insertions(+)
 create mode 100644 tools/lib/traceevent/plugin_xen.c
tools/lib/traceevent/Makefile +1 -0
index 5bc4a85..f3166a8 100644
@@ -209,6 +209,7 @@ PLUGIN_OBJS += plugin_kvm.o
 PLUGIN_OBJS += plugin_mac80211.o
 PLUGIN_OBJS += plugin_sched_switch.o
 PLUGIN_OBJS += plugin_function.o
+PLUGIN_OBJS += plugin_xen.o
 
 PLUGINS := $(PLUGIN_OBJS:.o=.so)
 
tools/lib/traceevent/plugin_xen.c new file +130 -0

130 lines not on this page · raw message

Signature · 1 line
1.8.3.1
maintainer
#43 lore raw

Re: [PATCH 11/22] tools lib traceevent: Add xen plugin

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 29 lines

Adding xen plugin.

This plugin adds fields resolving for
following tracepoint events:
xen:xen_mc_entry
xen:xen_mc_extend_args

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'xen:*' ls')

script.xen.new renamed from script.xen.old +2 -2
-         swapper     0 [002]   136.267492: xen:xen_mc_entry: [FAILED TO PARSE] op=3 nargs=2 args=ARRAY[18, 00, 00, 00, 00, 00, 00, 00, 00, e0, d4, 4b, 04, 88, ff, ff, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00]
+         swapper     0 [002]   136.267492: xen:xen_mc_entry: op 3(stack_switch) args [18, 0, 0, 0, 0, 0]
-            perf  1970 [008]   136.273319: xen:xen_mc_extend_args: [FAILED TO PARSE] op=1 args=16 res=1
+            perf  1970 [008]   136.273319: xen:xen_mc_extend_args: extending op 1(mmu_update) by 16 bytes res ???

NOTE We still do not handle the 'sizeof' and fail
to parse following xen tracepoints:
xen:xen_mmu_set_pte
xen:xen_mmu_set_pte_atomic
xen:xen_mmu_set_domain_pte
xen:xen_mmu_set_pte_at
xen:xen_mmu_set_pmd
xen:xen_mmu_set_pud
xen:xen_mmu_set_pgd
xen:xen_mmu_ptep_modify_prot_start
xen:xen_mmu_ptep_modify_prot_commit

[SNIP]

script.xen.new

Quoted 7 lines
script.xen.new renamed from script.xen.old +7 -0
+static const char *xen_hypercall_name(unsigned op)
+{
+	if (op < ARRAY_SIZE(xen_hypercall_names) &&
+	    xen_hypercall_names[op] != NULL)
+		return xen_hypercall_names[op];
+
+	return "";

How about defaults to "unknown" if not found?

Thanks
Namhyung

Quoted 21 lines
script.xen.new renamed from script.xen.old +21 -0
+}
+
+unsigned long long process_xen_hypercall_name(struct trace_seq *s,
+					      unsigned long long *args)
+{
+	unsigned int op = args[0];
+
+	trace_seq_printf(s, "%s", xen_hypercall_name(op));
+	return 0;
+}
+
+int PEVENT_PLUGIN_LOADER(struct pevent *pevent)
+{
+	pevent_register_print_function(pevent,
+				       process_xen_hypercall_name,
+				       PEVENT_FUNC_ARG_STRING,
+				       "xen_hypercall_name",
+				       PEVENT_FUNC_ARG_INT,
+				       PEVENT_FUNC_ARG_VOID);
+	return 0;
+}
author
#44 lore raw

Re: [PATCH 11/22] tools lib traceevent: Add xen plugin

Quoted 32 lines · Namhyung Kim

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 29 lines

Adding xen plugin.

This plugin adds fields resolving for
following tracepoint events:
xen:xen_mc_entry
xen:xen_mc_extend_args

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'xen:*' ls')

script.xen.new renamed from script.xen.old +2 -2
-         swapper     0 [002]   136.267492: xen:xen_mc_entry: [FAILED TO PARSE] op=3 nargs=2 args=ARRAY[18, 00, 00, 00, 00, 00, 00, 00, 00, e0, d4, 4b, 04, 88, ff, ff, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00, 00]
+         swapper     0 [002]   136.267492: xen:xen_mc_entry: op 3(stack_switch) args [18, 0, 0, 0, 0, 0]
-            perf  1970 [008]   136.273319: xen:xen_mc_extend_args: [FAILED TO PARSE] op=1 args=16 res=1
+            perf  1970 [008]   136.273319: xen:xen_mc_extend_args: extending op 1(mmu_update) by 16 bytes res ???

NOTE We still do not handle the 'sizeof' and fail
to parse following xen tracepoints:
xen:xen_mmu_set_pte
xen:xen_mmu_set_pte_atomic
xen:xen_mmu_set_domain_pte
xen:xen_mmu_set_pte_at
xen:xen_mmu_set_pmd
xen:xen_mmu_set_pud
xen:xen_mmu_set_pgd
xen:xen_mmu_ptep_modify_prot_start
xen:xen_mmu_ptep_modify_prot_commit

[SNIP]

Quoted 10 lines

script.xen.new

Quoted 7 lines
script.xen.new renamed from script.xen.old +7 -0
+static const char *xen_hypercall_name(unsigned op)
+{
+	if (op < ARRAY_SIZE(xen_hypercall_names) &&
+	    xen_hypercall_names[op] != NULL)
+		return xen_hypercall_names[op];
+
+	return "";

How about defaults to "unknown" if not found?

That's how the kernel 'xen_hypercall_name' handles it.

I dont mind, but I guess we want to change both sides. I've already included Jeremy (kernel side author), so I guess it's up to him ;-)

jirka

author
#45 lore raw

PATCH 12/22 tools lib traceevent: Add scsi plugin

In reply to Jiri Olsa

Adding scsi plugin.

This plugin adds fields resolving fucntions for following tracepoint events:
scsi:scsi_dispatch_cmd_start scsi:scsi_dispatch_cmd_error scsi:scsi_dispatch_cmd_done scsi:scsi_dispatch_cmd_timeout

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'scsi:scsi_dispatch_cmd*' -a)

-         swapper     0 [000]  6620.491019: scsi:scsi_dispatch_cmd_done: [FAILED TO PARSE] host_no=0 channel=0 id=0 lun=0 result=0 opcode=53 cmd_len=10 data_sglen=0 prot_sglen=0 prot_op=0 cmnd=5
+         swapper     0 [000]  6620.491019: scsi:scsi_dispatch_cmd_done: host_no=0 channel=0 id=0 lun=0 data_sgl=0 prot_sgl=0 prot_op=SCSI_PROT_NORMAL cmnd=(SYNCHRONIZE_CACHE - raw=35 00 00 00 00 00 00 00 00 00) result=(driver=DRIVER_OK host=DID_OK message=COMMAND_COMPLETE status=SAM_STAT_GOOD)
-     kworker/0:0 21554 [000]  6620.491126: scsi:scsi_dispatch_cmd_start: [FAILED TO PARSE] host_no=0 channel=0 id=0 lun=0 opcode=42 cmd_len=10 data_sglen=1 prot_sglen=0 prot_op=0 cmnd=*
+     kworker/0:0 21554 [000]  6620.491126: scsi:scsi_dispatch_cmd_start: host_no=0 channel=0 id=0 lun=0 data_sgl=1 prot_sgl=0 prot_op=SCSI_PROT_NORMAL cmnd=(WRITE_10 lba=570899168 txlen=8 protect=0 raw=2a 00 22 07 3a e0 00 00 08 00)
-     jbd2/dm-3-8   593 [002]  6621.607992: scsi:scsi_dispatch_cmd_error: [FAILED TO PARSE] host_no=0 channel=0 id=0 lun=0 rtn=4182 opcode=53 cmd_len=10 data_sglen=0 prot_sglen=0 prot_op=0 cmnd=5
+     jbd2/dm-3-8   593 [002]  6621.607992: scsi:scsi_dispatch_cmd_error: host_no=0 channel=0 id=0 lun=0 data_sgl=0 prot_sgl=0 prot_op=SCSI_PROT_NORMAL cmnd=(SYNCHRONIZE_CACHE - raw=35 00 00 00 00 00 00 00 00 00) rtn=4182

NOTE I couldn't generated scsi_dispatch_cmd_timeout tracepoint,
but it's similar to the rest, so I believe it's ok.

Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
Cc Martin K. Petersen <[email protected]>
---
 tools/lib/traceevent/Makefile      |   1 +
 tools/lib/traceevent/plugin_scsi.c | 423 +++++++++++++++++++++++++++++++++++++
 2 files changed, 424 insertions(+)
 create mode 100644 tools/lib/traceevent/plugin_scsi.c
tools/lib/traceevent/Makefile +1 -0
index f3166a8..88da8af 100644
@@ -210,6 +210,7 @@ PLUGIN_OBJS += plugin_mac80211.o
 PLUGIN_OBJS += plugin_sched_switch.o
 PLUGIN_OBJS += plugin_function.o
 PLUGIN_OBJS += plugin_xen.o
+PLUGIN_OBJS += plugin_scsi.o
 
 PLUGINS := $(PLUGIN_OBJS:.o=.so)
 
tools/lib/traceevent/plugin_scsi.c new file +423 -0

423 lines not on this page · raw message

Signature · 1 line
1.8.3.1
maintainer
#46 lore raw

Re: [PATCH 12/22] tools lib traceevent: Add scsi plugin

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 21 lines

Adding scsi plugin.

This plugin adds fields resolving fucntions for following tracepoint events:
scsi:scsi_dispatch_cmd_start scsi:scsi_dispatch_cmd_error scsi:scsi_dispatch_cmd_done scsi:scsi_dispatch_cmd_timeout

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'scsi:scsi_dispatch_cmd*' -a)

(quoted diff) +3 -3
-         swapper     0 [000]  6620.491019: scsi:scsi_dispatch_cmd_done: [FAILED TO PARSE] host_no=0 channel=0 id=0 lun=0 result=0 opcode=53 cmd_len=10 data_sglen=0 prot_sglen=0 prot_op=0 cmnd=5
+         swapper     0 [000]  6620.491019: scsi:scsi_dispatch_cmd_done: host_no=0 channel=0 id=0 lun=0 data_sgl=0 prot_sgl=0 prot_op=SCSI_PROT_NORMAL cmnd=(SYNCHRONIZE_CACHE - raw=35 00 00 00 00 00 00 00 00 00) result=(driver=DRIVER_OK host=DID_OK message=COMMAND_COMPLETE status=SAM_STAT_GOOD)
-     kworker/0:0 21554 [000]  6620.491126: scsi:scsi_dispatch_cmd_start: [FAILED TO PARSE] host_no=0 channel=0 id=0 lun=0 opcode=42 cmd_len=10 data_sglen=1 prot_sglen=0 prot_op=0 cmnd=*
+     kworker/0:0 21554 [000]  6620.491126: scsi:scsi_dispatch_cmd_start: host_no=0 channel=0 id=0 lun=0 data_sgl=1 prot_sgl=0 prot_op=SCSI_PROT_NORMAL cmnd=(WRITE_10 lba=570899168 txlen=8 protect=0 raw=2a 00 22 07 3a e0 00 00 08 00)
-     jbd2/dm-3-8   593 [002]  6621.607992: scsi:scsi_dispatch_cmd_error: [FAILED TO PARSE] host_no=0 channel=0 id=0 lun=0 rtn=4182 opcode=53 cmd_len=10 data_sglen=0 prot_sglen=0 prot_op=0 cmnd=5
+     jbd2/dm-3-8   593 [002]  6621.607992: scsi:scsi_dispatch_cmd_error: host_no=0 channel=0 id=0 lun=0 data_sgl=0 prot_sgl=0 prot_op=SCSI_PROT_NORMAL cmnd=(SYNCHRONIZE_CACHE - raw=35 00 00 00 00 00 00 00 00 00) rtn=4182

NOTE I couldn't generated scsi_dispatch_cmd_timeout tracepoint,
but it's similar to the rest, so I believe it's ok.

[SNIP]

Quoted 4 lines
tools/lib/traceevent/plugin_scsi.c new file +4 -0
+static const char *
+scsi_trace_rw6(struct trace_seq *p, unsigned char *cdb, int len)
+{
+	const char *ret = p->buffer + p->len;

tools/lib/traceevent/plugin_scsi.c line 164

Quoted 10 lines
tools/lib/traceevent/plugin_scsi.c new file +10 -0
+	sector_t lba = 0, txlen = 0;
+
+	lba |= ((cdb[1] & 0x1F) << 16);
+	lba |=  (cdb[2] << 8);
+	lba |=   cdb[3];
+	txlen = cdb[4];
+
+	trace_seq_printf(p, "lba=%llu txlen=%llu",
+			 (unsigned long long)lba, (unsigned long long)txlen);
+	trace_seq_putc(p, 0);

Why is this _putc(0) needed? It seems you added it to all other functions too.

Quoted 3 lines
tools/lib/traceevent/plugin_scsi.c new file +3 -0
+	return ret;
+}
+
#47 lore raw

Re: [PATCH 12/22] tools lib traceevent: Add scsi plugin

Quoted 10 lines · On Sat, 23 Nov 2013 00:24:50 +0900 Namhyung Kim <[email protected]> wrote:

[SNIP]

Quoted 9 lines
tools/lib/traceevent/plugin_scsi.c new file +9 -0
+static const char *
+scsi_trace_rw6(struct trace_seq *p, unsigned char *cdb, int len)
+{
+	const char *ret = p->buffer + p->len;
+	sector_t lba = 0, txlen = 0;
+
+	lba |= ((cdb[1] & 0x1F) << 16);
+	lba |=  (cdb[2] << 8);
+	lba |=   cdb[3];
Quoted 8 lines

tools/lib/traceevent/plugin_scsi.c line 164

Quoted 5 lines
tools/lib/traceevent/plugin_scsi.c new file +5 -0
+	txlen = cdb[4];
+
+	trace_seq_printf(p, "lba=%llu txlen=%llu",
+			 (unsigned long long)lba, (unsigned long long)txlen);
+	trace_seq_putc(p, 0);

Why is this _putc(0) needed? It seems you added it to all other functions too.

This is because these are copied directly from the kernel, and the kernel versions use the putc(0) too. But there, it is needed, as its up to the helper function to terminate the trace_seq.

I need to add trace_seq_terminate() in the kernel (it's in libtraceevent), so that we can replace the putc(0) with terminate, and not truncate the seq early.

-- Steve

Quoted 7 lines
Quoted 5 lines
tools/lib/traceevent/plugin_scsi.c new file +3 -0
+	return ret;
+}
+
author
#48 lore raw

PATCH 13/22 tools lib traceevent: Add cfg80211 plugin

In reply to Jiri Olsa

Adding cfg80211 plugin.

This plugin adds handler for __le16_to_cpup function t properly parse following tracepoint events:
cfg80211:cfg80211_tx_mlme_mgmt cfg80211:cfg80211_rx_mlme_mgmt cfg80211:cfg80211_rx_unprot_mlme_mgmt

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'cfg80211:*' -a')

--- script.cfg80211.old
+++ script.cfg80211.new

-        ifconfig  2705 [003]   662.896560: cfg80211:cfg80211_tx_mlme_mgmt: [FAILED TO PARSE] name=wlan0 ifindex=3 frame=ARRAY[c0, 00, 00, 00, 00, 3a, 98, a0, 30, 51, 10, 0b, a9, c6, f4, 74, 00, 3a, 98, a0, 30, 51, 00, 00, 03, 00]
+        ifconfig  2705 [003]   662.896560: cfg80211:cfg80211_tx_mlme_mgmt: netdev:wlan0(3), ftype:0xc0
-   kworker/u16:0  1697 [002]   664.808210: cfg80211:cfg80211_rx_mlme_mgmt: [FAILED TO PARSE] name=wlan0 ifindex=3 frame=ARRAY[b0, 00, da, 00, 10, 0b, a9, c6, f4, 74, d8, 24, bd, a1, 26, 31, d8, 24, bd, a1, 26, 31, 10, b7, 00, 00, 02, 00, 00, 00]
+   kworker/u16:0  1697 [002]   664.808210: cfg80211:cfg80211_rx_mlme_mgmt: netdev:wlan0(3), ftype:0xb0
Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/lib/traceevent/Makefile          |  1 +
 tools/lib/traceevent/plugin_cfg80211.c | 23 +++++++++++++++++++++++
 2 files changed, 24 insertions(+)
 create mode 100644 tools/lib/traceevent/plugin_cfg80211.c
tools/lib/traceevent/Makefile +1 -0
index 88da8af..bbb2061 100644
@@ -211,6 +211,7 @@ PLUGIN_OBJS += plugin_sched_switch.o
 PLUGIN_OBJS += plugin_function.o
 PLUGIN_OBJS += plugin_xen.o
 PLUGIN_OBJS += plugin_scsi.o
+PLUGIN_OBJS += plugin_cfg80211.o
 
 PLUGINS := $(PLUGIN_OBJS:.o=.so)
 
tools/lib/traceevent/plugin_cfg80211.c new file +23 -0
new file mode 100644
index 0000000..0bbffb0
@@ -0,0 +1,23 @@
+#include <stdio.h>
+#include <string.h>
+#include <inttypes.h>
+#include <endian.h>
+#include "event-parse.h"
+
+unsigned long long process___le16_to_cpup(struct trace_seq *s,
+					  unsigned long long *args)
+{
+	uint16_t *val = (uint16_t *) args[0];
+	return val ? (long long) le16toh(*val) : 0;
+}
+
+int PEVENT_PLUGIN_LOADER(struct pevent *pevent)
+{
+	pevent_register_print_function(pevent,
+				       process___le16_to_cpup,
+				       PEVENT_FUNC_ARG_INT,
+				       "__le16_to_cpup",
+				       PEVENT_FUNC_ARG_PTR,
+				       PEVENT_FUNC_ARG_VOID);
+	return 0;
+}
Signature · 1 line
1.8.3.1
maintainer
#49 lore raw

Re: [PATCH 13/22] tools lib traceevent: Add cfg80211 plugin

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 17 lines

Adding cfg80211 plugin.

This plugin adds handler for __le16_to_cpup function t properly parse following tracepoint events:
cfg80211:cfg80211_tx_mlme_mgmt cfg80211:cfg80211_rx_mlme_mgmt cfg80211:cfg80211_rx_unprot_mlme_mgmt

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'cfg80211:*' -a')

script.cfg80211.new renamed from script.cfg80211.old +2 -2
-        ifconfig  2705 [003]   662.896560: cfg80211:cfg80211_tx_mlme_mgmt: [FAILED TO PARSE] name=wlan0 ifindex=3 frame=ARRAY[c0, 00, 00, 00, 00, 3a, 98, a0, 30, 51, 10, 0b, a9, c6, f4, 74, 00, 3a, 98, a0, 30, 51, 00, 00, 03, 00]
+        ifconfig  2705 [003]   662.896560: cfg80211:cfg80211_tx_mlme_mgmt: netdev:wlan0(3), ftype:0xc0
-   kworker/u16:0  1697 [002]   664.808210: cfg80211:cfg80211_rx_mlme_mgmt: [FAILED TO PARSE] name=wlan0 ifindex=3 frame=ARRAY[b0, 00, da, 00, 10, 0b, a9, c6, f4, 74, d8, 24, bd, a1, 26, 31, d8, 24, bd, a1, 26, 31, 10, b7, 00, 00, 02, 00, 00, 00]
+   kworker/u16:0  1697 [002]   664.808210: cfg80211:cfg80211_rx_mlme_mgmt: netdev:wlan0(3), ftype:0xb0

[SNIP]

script.cfg80211.new

Quoted 12 lines
script.cfg80211.new renamed from script.cfg80211.old +12 -0
+unsigned long long process___le16_to_cpup(struct trace_seq *s,
+					  unsigned long long *args)
+{
+	uint16_t *val = (uint16_t *) args[0];
+	return val ? (long long) le16toh(*val) : 0;
+}
+
+int PEVENT_PLUGIN_LOADER(struct pevent *pevent)
+{
+	pevent_register_print_function(pevent,
+				       process___le16_to_cpup,
+				       PEVENT_FUNC_ARG_INT,

Just a nitpick. The function should return int or the return type should be _LONG for consistency.

Thanks,
Namhyung

Quoted 5 lines
script.cfg80211.new renamed from script.cfg80211.old +5 -0
+				       "__le16_to_cpup",
+				       PEVENT_FUNC_ARG_PTR,
+				       PEVENT_FUNC_ARG_VOID);
+	return 0;
+}
#50 lore raw

Re: [PATCH 13/22] tools lib traceevent: Add cfg80211 plugin

Quoted 16 lines · On Sat, 23 Nov 2013 00:16:14 +0900 Namhyung Kim <[email protected]> wrote:

[SNIP]

tools/lib/traceevent/plugin_cfg80211.c line 18

Quoted 12 lines
tools/lib/traceevent/plugin_cfg80211.c new file +12 -0
+unsigned long long process___le16_to_cpup(struct trace_seq *s,
+					  unsigned long long *args)
+{
+	uint16_t *val = (uint16_t *) args[0];
+	return val ? (long long) le16toh(*val) : 0;
+}
+
+int PEVENT_PLUGIN_LOADER(struct pevent *pevent)
+{
+	pevent_register_print_function(pevent,
+				       process___le16_to_cpup,
+				       PEVENT_FUNC_ARG_INT,

Just a nitpick. The function should return int or the return type should be _LONG for consistency.

As stated in another reply, the return type is not for the callback, but instead for the function it is representing.

-- Steve

Quoted 11 lines

Thanks,
Namhyung

Quoted 5 lines
tools/lib/traceevent/plugin_cfg80211.c new file +5 -0
+				       "__le16_to_cpup",
+				       PEVENT_FUNC_ARG_PTR,
+				       PEVENT_FUNC_ARG_VOID);
+	return 0;
+}
author
#51 lore raw

PATCH 14/22 tools lib traceevent: Update kvm plugin with is_writable_pte helper

In reply to Jiri Olsa

Adding is_writable_pte print helper function, so the kvmmmu:fast_page_fault print format gets resolved properly.

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'kvm:*,kvmmmu:*' -a')

--- script.kvm.old
+++ script.kvm.new
- qemu-system-x86 3290 [002] 10708.755312: kvmmmu:fast_page_fault: [FAILED TO PARSE] vcpu_id=0 gva=4094486080 error_code=3 sptep=0xffff88019f1e3670 old_spte=336391285 new_spte=336391287 retry=1
+ qemu-system-x86 3290 [002] 10708.755312: kvmmmu:fast_page_fault: vcpu 0 gva f40ce640 error_code P|W sptep 0xffff88019f1e3670 old 0x140cec75 new 140cec77 spurious 0 fixed 1

Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/lib/traceevent/plugin_kvm.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)
tools/lib/traceevent/plugin_kvm.c +17 -0
index be9d9c6..1e37678 100644
@@ -389,6 +389,16 @@ static int kvm_mmu_get_page_handler(struct trace_seq *s,
 	return kvm_mmu_print_role(s, record, event, context);
 }
 
+#define PT_WRITABLE_SHIFT 1
+#define PT_WRITABLE_MASK (1ULL << PT_WRITABLE_SHIFT)
+
+unsigned long long process_is_writable_pte(struct trace_seq *s,
+					   unsigned long long *args)
+{
+	unsigned long pte = args[0];
+	return pte & PT_WRITABLE_MASK;
+}
+
 int PEVENT_PLUGIN_LOADER(struct pevent *pevent)
 {
 	init_disassembler();
@@ -415,5 +425,12 @@ int PEVENT_PLUGIN_LOADER(struct pevent *pevent)
 	pevent_register_event_handler(pevent, -1, "kvmmmu",
 			"kvm_mmu_prepare_zap_page", kvm_mmu_print_role,
 			NULL);
+
+	pevent_register_print_function(pevent,
+				       process_is_writable_pte,
+				       PEVENT_FUNC_ARG_INT,
+				       "is_writable_pte",
+				       PEVENT_FUNC_ARG_LONG,
+				       PEVENT_FUNC_ARG_VOID);
 	return 0;
 }
Signature · 1 line
1.8.3.1
maintainer
#52 lore raw

Re: [PATCH 14/22] tools lib traceevent: Update kvm plugin with is_writable_pte helper

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 10 lines

Adding is_writable_pte print helper function, so the kvmmmu:fast_page_fault print format gets resolved properly.

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'kvm:*,kvmmmu:*' -a')

script.kvm.new renamed from script.kvm.old +1 -1
- qemu-system-x86  3290 [002] 10708.755312: kvmmmu:fast_page_fault: [FAILED TO PARSE] vcpu_id=0 gva=4094486080 error_code=3 sptep=0xffff88019f1e3670 old_spte=336391285 new_spte=336391287 retry=1
+ qemu-system-x86  3290 [002] 10708.755312: kvmmmu:fast_page_fault: vcpu 0 gva f40ce640 error_code P|W sptep 0xffff88019f1e3670 old 0x140cec75 new 140cec77 spurious 0 fixed 1

[SNIP]

Quoted 11 lines
script.kvm.new renamed from script.kvm.old +7 -0
+unsigned long long process_is_writable_pte(struct trace_seq *s,
+					   unsigned long long *args)
+{
+	unsigned long pte = args[0];
+	return pte & PT_WRITABLE_MASK;
+}
+
 int PEVENT_PLUGIN_LOADER(struct pevent *pevent)
 {
 	init_disassembler();

script.kvm.new int PEVENT_PLUGIN_LOADER(struct pevent *pevent) · line 431

Quoted 7 lines
script.kvm.new renamed from script.kvm.old +4 -0
@@ -415,5 +425,12 @@ int PEVENT_PLUGIN_LOADER(struct pevent *pevent)
 	pevent_register_event_handler(pevent, -1, "kvmmmu",
 			"kvm_mmu_prepare_zap_page", kvm_mmu_print_role,
 			NULL);
+
+	pevent_register_print_function(pevent,
+				       process_is_writable_pte,
+				       PEVENT_FUNC_ARG_INT,

Same as the previous reply - return type of the function is inconsistent.

And please also consider making the function static.

Thanks,
Namhyung

Quoted 5 lines
script.kvm.new renamed from script.kvm.old +3 -0
+				       "is_writable_pte",
+				       PEVENT_FUNC_ARG_LONG,
+				       PEVENT_FUNC_ARG_VOID);
 	return 0;
 }
#53 lore raw

Re: [PATCH 14/22] tools lib traceevent: Update kvm plugin with is_writable_pte helper

Quoted 26 lines · On Sat, 23 Nov 2013 00:19:48 +0900 Namhyung Kim <[email protected]> wrote:

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 10 lines

Adding is_writable_pte print helper function, so the kvmmmu:fast_page_fault print format gets resolved properly.

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e 'kvm:*,kvmmmu:*' -a')

script.kvm.new renamed from script.kvm.old +1 -1
- qemu-system-x86  3290 [002] 10708.755312: kvmmmu:fast_page_fault: [FAILED TO PARSE] vcpu_id=0 gva=4094486080 error_code=3 sptep=0xffff88019f1e3670 old_spte=336391285 new_spte=336391287 retry=1
+ qemu-system-x86  3290 [002] 10708.755312: kvmmmu:fast_page_fault: vcpu 0 gva f40ce640 error_code P|W sptep 0xffff88019f1e3670 old 0x140cec75 new 140cec77 spurious 0 fixed 1

[SNIP]

Quoted 13 lines
script.kvm.new renamed from script.kvm.old +7 -0
+unsigned long long process_is_writable_pte(struct trace_seq *s,
+					   unsigned long long *args)
+{
+	unsigned long pte = args[0];
+	return pte & PT_WRITABLE_MASK;
+}
+
 int PEVENT_PLUGIN_LOADER(struct pevent *pevent)
 {
 	init_disassembler();
@@ -415,5 +425,12 @@ int PEVENT_PLUGIN_LOADER(struct pevent *pevent)
 	pevent_register_event_handler(pevent, -1, "kvmmmu",
 			"kvm_mmu_prepare_zap_page", kvm_mmu_print_role,
Quoted 8 lines

script.kvm.new int PEVENT_PLUGIN_LOADER(struct pevent *pevent) · line 431

Quoted 5 lines
script.kvm.new renamed from script.kvm.old +4 -0
 			NULL);
+
+	pevent_register_print_function(pevent,
+				       process_is_writable_pte,
+				       PEVENT_FUNC_ARG_INT,

Same as the previous reply - return type of the function is inconsistent.

Actually it's not.

Quoted 2 lines

And please also consider making the function static.

Agreed.

-- Steve

Quoted 12 lines

Thanks,
Namhyung

Quoted 5 lines
script.kvm.new renamed from script.kvm.old +3 -0
+				       "is_writable_pte",
+				       PEVENT_FUNC_ARG_LONG,
+				       PEVENT_FUNC_ARG_VOID);
 	return 0;
 }
author
#54 lore raw

PATCH 15/22 tools lib traceevent: Change pevent_parse_event to return event format

In reply to Jiri Olsa

Changing pevent_parse_event to return 'struct event_format', for the parsed event if the return pointer is specified.

This way we can remove pevent_parse_format which does the same stuff just without 'struct pevent' object.

The ultimate goal is to always use pevent object when dealing with traceevent library. The reason is that we might need additional processing (like plugins), which is not possible otherwise.

Patches follow to make this happen completely.

Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/lib/traceevent/event-parse.c  | 32 +++++++++++---------------------
 tools/lib/traceevent/event-parse.h  |  7 ++++---
 tools/perf/util/evsel.c             |  2 +-
 tools/perf/util/trace-event-parse.c |  4 ++--
 4 files changed, 18 insertions(+), 27 deletions(-)
tools/lib/traceevent/event-parse.c +11 -21
index 74007ab..4baa0a9 100644
@@ -5130,27 +5130,9 @@ enum pevent_errno __pevent_parse_format(struct event_format **eventp,
 }
 
 /**
- * pevent_parse_format - parse the event format
- * @buf: the buffer storing the event format string
- * @size: the size of @buf
- * @sys: the system the event belongs to
- *
- * This parses the event format and creates an event structure
- * to quickly parse raw data for a given event.
- *
- * These files currently come from:
- *
- * /sys/kernel/debug/tracing/events/.../.../format
- */
-enum pevent_errno pevent_parse_format(struct event_format **eventp, const char *buf,
-				      unsigned long size, const char *sys)
-{
-	return __pevent_parse_format(eventp, NULL, buf, size, sys);
-}
-
-/**
  * pevent_parse_event - parse the event format
  * @pevent: the handle to the pevent
+ * @format: event format object for passed format
  * @buf: the buffer storing the event format string
  * @size: the size of @buf
  * @sys: the system the event belongs to
@@ -5158,11 +5140,16 @@ enum pevent_errno pevent_parse_format(struct event_format **eventp, const char *
  * This parses the event format and creates an event structure
  * to quickly parse raw data for a given event.
  *
+ * Returns  event format object in @format parameter if
+ * specified.
+ *
  * These files currently come from:
  *
  * /sys/kernel/debug/tracing/events/.../.../format
  */
-enum pevent_errno pevent_parse_event(struct pevent *pevent, const char *buf,
+enum pevent_errno pevent_parse_event(struct pevent *pevent,
+				     struct event_format **format,
+				     const char *buf,
 				     unsigned long size, const char *sys)
 {
 	struct event_format *event = NULL;
@@ -5171,7 +5158,7 @@ enum pevent_errno pevent_parse_event(struct pevent *pevent, const char *buf,
 	if (event == NULL)
 		return ret;
 
-	if (add_event(pevent, event)) {
+	if (pevent && add_event(pevent, event)) {
 		ret = PEVENT_ERRNO__MEM_ALLOC_FAILED;
 		goto event_add_failed;
 	}
@@ -5180,6 +5167,9 @@ enum pevent_errno pevent_parse_event(struct pevent *pevent, const char *buf,
 	if (PRINT_ARGS && event->print_fmt.args)
 		print_args(event->print_fmt.args);
 
+	if (format)
+		*format = event;
+
 	return 0;
 
 event_add_failed:
tools/lib/traceevent/event-parse.h +4 -3
index 0ad7f6b..daafb70 100644
@@ -559,10 +559,11 @@ void pevent_print_event(struct pevent *pevent, struct trace_seq *s,
 int pevent_parse_header_page(struct pevent *pevent, char *buf, unsigned long size,
 			     int long_size);
 
-enum pevent_errno pevent_parse_event(struct pevent *pevent, const char *buf,
+enum pevent_errno pevent_parse_event(struct pevent *pevent,
+				     struct event_format **format,
+				     const char *buf,
 				     unsigned long size, const char *sys);
-enum pevent_errno pevent_parse_format(struct event_format **eventp, const char *buf,
-				      unsigned long size, const char *sys);
+
 void pevent_free_format(struct event_format *event);
 
 void *pevent_get_field_raw(struct trace_seq *s, struct event_format *event,
tools/perf/util/evsel.c +1 -1
index 77e38ff..599d1780 100644
@@ -208,7 +208,7 @@ struct event_format *event_format__new(const char *sys, const char *name)
 		size += n;
 	} while (n > 0);
 
-	pevent_parse_format(&format, bf, size, sys);
+	pevent_parse_event(NULL, &format, bf, size, sys);
 
 out_free_bf:
 	free(bf);
tools/perf/util/trace-event-parse.c +2 -2
index 6681f71..c8dae72 100644
@@ -198,13 +198,13 @@ void parse_ftrace_printk(struct pevent *pevent,
 
 int parse_ftrace_file(struct pevent *pevent, char *buf, unsigned long size)
 {
-	return pevent_parse_event(pevent, buf, size, "ftrace");
+	return pevent_parse_event(pevent, NULL, buf, size, "ftrace");
 }
 
 int parse_event_file(struct pevent *pevent,
 		     char *buf, unsigned long size, char *sys)
 {
-	return pevent_parse_event(pevent, buf, size, sys);
+	return pevent_parse_event(pevent, NULL, buf, size, sys);
 }
 
 struct event_format *trace_find_next_event(struct pevent *pevent,
Signature · 1 line
1.8.3.1
#55 lore raw

Re: [PATCH 15/22] tools lib traceevent: Change pevent_parse_event to return event format

Quoted 2 lines · On Thu, 21 Nov 2013 12:01:13 +0100 Jiri Olsa <[email protected]> wrote:

Changing pevent_parse_event to return 'struct event_format', for the parsed event if the return pointer is specified.

Can we not change the API of pevent_parse_event(), I use it quite a bit.

And yes, I'm working hard to keep what's in trace-cmd and the kernel identical.

Quoted 3 lines

This way we can remove pevent_parse_format which does the same stuff just without 'struct pevent' object.

Actually, I don't use pevent_parse_format() at all. What about changing that one to include the pevent?

Quoted 5 lines

The ultimate goal is to always use pevent object when dealing with traceevent library. The reason is that we might need additional processing (like plugins), which is not possible otherwise.

Right, so add a pevent to the pevent_parse_format() and leave the pevent_parse_event() alone. That should make everyone happy :-)

Thanks!

-- Steve

author
#56 lore raw

Re: [PATCH 15/22] tools lib traceevent: Change pevent_parse_event to return event format

Quoted 20 lines · Steven Rostedt
Quoted 2 lines · On Thu, 21 Nov 2013 12:01:13 +0100 Jiri Olsa <[email protected]> wrote:

Changing pevent_parse_event to return 'struct event_format', for the parsed event if the return pointer is specified.

Can we not change the API of pevent_parse_event(), I use it quite a bit.

And yes, I'm working hard to keep what's in trace-cmd and the kernel identical.

Quoted 3 lines

This way we can remove pevent_parse_format which does the same stuff just without 'struct pevent' object.

Actually, I don't use pevent_parse_format() at all. What about changing that one to include the pevent?

Quoted 1 line
Quoted 7 lines
Quoted 4 lines

The ultimate goal is to always use pevent object when dealing with traceevent library. The reason is that we might need additional processing (like plugins), which is not possible otherwise.

Right, so add a pevent to the pevent_parse_format() and leave the pevent_parse_event() alone. That should make everyone happy :-)

ook, np ;-)

thanks,
jirka

author
#57 lore raw

PATCH 16/22 tools lib traceevent: Harmonize the install messages in lib-traceevent

In reply to Jiri Olsa

Removing the 'to ...' part out of the install message, because it does not fit to the rest of the build messages we use.

Before:

  INSTALL  plugin_hrtimer.so    to      /home/jolsa/libexec/perf-core/traceevent/plugins
  INSTALL  plugin_jbd2.so       to      /home/jolsa/libexec/perf-core/traceevent/plugins
  INSTALL  plugin_kmem.so       to      /home/jolsa/libexec/perf-core/traceevent/plugins
  INSTALL  plugin_kvm.so        to      /home/jolsa/libexec/perf-core/traceevent/plugins
  INSTALL  plugin_mac80211.so   to      /home/jolsa/libexec/perf-core/traceevent/plugins
  INSTALL  plugin_sched_switch.so       to      /home/jolsa/libexec/perf-core/traceevent/plugins
  INSTALL  plugin_function.so   to      /home/jolsa/libexec/perf-core/traceevent/plugins
  INSTALL  plugin_xen.so        to      /home/jolsa/libexec/perf-core/traceevent/plugins
  INSTALL  plugin_scsi.so       to      /home/jolsa/libexec/perf-core/traceevent/plugins

Now:

  INSTALL  plugin_jbd2.so
  INSTALL  plugin_hrtimer.so
  INSTALL  plugin_kmem.so
  INSTALL  plugin_kvm.so
  INSTALL  plugin_mac80211.so
  INSTALL  plugin_sched_switch.so
  INSTALL  plugin_function.so
  INSTALL  plugin_xen.so
  INSTALL  plugin_scsi.so
Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/lib/traceevent/Makefile | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/lib/traceevent/Makefile +1 -1
index bbb2061..37da4a9 100644
@@ -156,7 +156,7 @@ else
   print_plugin_obj_compile =	echo '  CC FPIC  '$(OBJ);
   print_plugin_build =		echo '  BUILD    PLUGIN '$(OBJ);
   print_static_lib_build =	echo '  BUILD    STATIC LIB '$(OBJ);
-  print_install =		echo '  INSTALL  '$1'	to	$(DESTDIR_SQ)$2';
+  print_install =		echo '  INSTALL  '$1;
 endif
 
 do_fpic_compile =					\
Signature · 1 line
1.8.3.1
author
#58 lore raw

PATCH 17/22 perf tools: Add build and install plugins targets

In reply to Jiri Olsa

Adding 'plugins' target along with the libtraceevent.a, so plugins are built together with traceevent library.

Adding 'install-traceevent-plugins' Makefile install target, instructing perf to install plugins into:
$(HOME)/libexec/perf-core/traceevent/plugins
- if installed localy under $HOME
$(DESTDIR)/$(prefix)/$(perfexecdir)/traceevent/plugins
- if installed globally

Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/perf/Makefile.perf   | 15 ++++++++++-----
 tools/perf/config/Makefile |  2 ++
 2 files changed, 12 insertions(+), 5 deletions(-)
tools/perf/Makefile.perf +10 -5
index e416ccc..b4f8910 100644
@@ -105,7 +105,7 @@ ifeq ($(config),1)
 include config/Makefile
 endif
 
-export prefix bindir sharedir sysconfdir
+export prefix bindir sharedir sysconfdir DESTDIR
 
 # sparse is architecture-neutral, which means that we need to tell it
 # explicitly what architecture to check for. Fix this up for yours..
@@ -710,13 +710,18 @@ $(LIB_FILE): $(LIB_OBJS)
 # libtraceevent.a
 TE_SOURCES = $(wildcard $(TRACE_EVENT_DIR)*.[ch])
 
-$(LIBTRACEEVENT): $(TE_SOURCES)
-	$(QUIET_SUBDIR0)$(TRACE_EVENT_DIR) $(QUIET_SUBDIR1) O=$(OUTPUT) CFLAGS="-g -Wall $(EXTRA_CFLAGS)" libtraceevent.a
+LIBTRACEEVENT_FLAGS=$(QUIET_SUBDIR1) O=$(OUTPUT) CFLAGS="-g -Wall $(EXTRA_CFLAGS)" plugin_dir=$(plugindir_SQ)
+
+$(LIBTRACEEVENT): $(TE_SOURCES) $(OUTPUT)PERF-CFLAGS
+	$(QUIET_SUBDIR0)$(TRACE_EVENT_DIR) $(LIBTRACEEVENT_FLAGS) libtraceevent.a plugins
 
 $(LIBTRACEEVENT)-clean:
 	$(call QUIET_CLEAN, libtraceevent)
 	@$(MAKE) -C $(TRACE_EVENT_DIR) O=$(OUTPUT) clean >/dev/null
 
+install-traceevent-plugins:
+	$(QUIET_SUBDIR0)$(TRACE_EVENT_DIR) $(LIBTRACEEVENT_FLAGS) install_plugins
+
 LIBLK_SOURCES = $(wildcard $(LK_PATH)*.[ch])
 
 # if subdir is set, we've been called from above so target has been built
@@ -785,7 +790,7 @@ cscope:
 
 ### Detect prefix changes
 TRACK_CFLAGS = $(subst ','\'',$(CFLAGS)):\
-             $(bindir_SQ):$(perfexecdir_SQ):$(template_dir_SQ):$(prefix_SQ)
+             $(bindir_SQ):$(perfexecdir_SQ):$(template_dir_SQ):$(prefix_SQ):$(plugindir_SQ)
 
 $(OUTPUT)PERF-CFLAGS: .FORCE-PERF-CFLAGS
 	@FLAGS='$(TRACK_CFLAGS)'; \
@@ -849,7 +854,7 @@ endif
 		$(INSTALL) -d -m 755 '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/attr'; \
 		$(INSTALL) tests/attr/* '$(DESTDIR_SQ)$(perfexec_instdir_SQ)/tests/attr'
 
-install: install-bin try-install-man
+install: install-bin try-install-man install-traceevent-plugins
 
 install-python_ext:
 	$(PYTHON_WORD) util/setup.py --quiet install --root='/$(DESTDIR_SQ)'
tools/perf/config/Makefile +2 -0
index f7d11a8..8c8a205 100644
@@ -578,6 +578,7 @@ else
 lib = lib
 endif
 libdir = $(prefix)/$(lib)
+plugindir=$(perfexecdir)/traceevent/plugins
 
 # Shell quote (do not use $(call) to accommodate ancient setups);
 ETC_PERFCONFIG_SQ = $(subst ','\'',$(ETC_PERFCONFIG))
@@ -598,3 +599,4 @@ else
 perfexec_instdir = $(prefix)/$(perfexecdir)
 endif
 perfexec_instdir_SQ = $(subst ','\'',$(perfexec_instdir))
+plugindir_SQ= $(subst ','\'',$(prefix)/$(plugindir))
Signature · 1 line
1.8.3.1
maintainer
#59 lore raw

Re: [PATCH 17/22] perf tools: Add build and install plugins targets

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 9 lines

Adding 'plugins' target along with the libtraceevent.a, so plugins are built together with traceevent library.

Adding 'install-traceevent-plugins' Makefile install target, instructing perf to install plugins into:
$(HOME)/libexec/perf-core/traceevent/plugins
- if installed localy under $HOME
$(DESTDIR)/$(prefix)/$(perfexecdir)/traceevent/plugins
- if installed globally

[SNIP]

tools/perf/config/Makefile else · line 581

Quoted 7 lines
tools/perf/config/Makefile +1 -0
@@ -578,6 +578,7 @@ else
 lib = lib
 endif
 libdir = $(prefix)/$(lib)
+plugindir=$(perfexecdir)/traceevent/plugins

Hmm.. it puts traceevent plugins under the perf directory. But I don't think it's a good idea since libtraceevent can be used independently.

What about something like this?

  plugindir = $(perfexecdir)/../traceevent/plugins

Thanks,
Namhyung

author
#60 lore raw

Re: [PATCH 17/22] perf tools: Add build and install plugins targets

Quoted 23 lines · Namhyung Kim

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 9 lines

Adding 'plugins' target along with the libtraceevent.a, so plugins are built together with traceevent library.

Adding 'install-traceevent-plugins' Makefile install target, instructing perf to install plugins into:
$(HOME)/libexec/perf-core/traceevent/plugins
- if installed localy under $HOME
$(DESTDIR)/$(prefix)/$(perfexecdir)/traceevent/plugins
- if installed globally

[SNIP]

tools/perf/config/Makefile else · line 581

Quoted 7 lines
tools/perf/config/Makefile +1 -0
@@ -578,6 +578,7 @@ else
 lib = lib
 endif
 libdir = $(prefix)/$(lib)
+plugindir=$(perfexecdir)/traceevent/plugins

Hmm.. it puts traceevent plugins under the perf directory. But I don't think it's a good idea since libtraceevent can be used independently.

Quoted 3 lines

What about something like this?

  plugindir = $(perfexecdir)/../traceevent/plugins

this is where trace-cmd stores them:

  $(HOME)/.trace-cmd/plugins
  $(prefix)/$(libdir)/trace-cmd/plugins

but I guess we want global place to install plugins, the traceevent by default uses:

  $(HOME)/.traceevent/plugins
  $(prefix)/lib/traceevent/plugins

How about that?

thanks,
jirka

maintainer
#61 lore raw

Re: [PATCH 17/22] perf tools: Add build and install plugins targets

Hi Jiri,

Quoted 33 lines · Jiri Olsa
Quoted 23 lines · Namhyung Kim

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 9 lines

Adding 'plugins' target along with the libtraceevent.a, so plugins are built together with traceevent library.

Adding 'install-traceevent-plugins' Makefile install target, instructing perf to install plugins into:
$(HOME)/libexec/perf-core/traceevent/plugins
- if installed localy under $HOME
$(DESTDIR)/$(prefix)/$(perfexecdir)/traceevent/plugins
- if installed globally

[SNIP]

tools/perf/config/Makefile else · line 581

Quoted 7 lines
tools/perf/config/Makefile +1 -0
@@ -578,6 +578,7 @@ else
 lib = lib
 endif
 libdir = $(prefix)/$(lib)
+plugindir=$(perfexecdir)/traceevent/plugins

Hmm.. it puts traceevent plugins under the perf directory. But I don't think it's a good idea since libtraceevent can be used independently.

Quoted 3 lines

What about something like this?

  plugindir = $(perfexecdir)/../traceevent/plugins

this is where trace-cmd stores them:

  $(HOME)/.trace-cmd/plugins
  $(prefix)/$(libdir)/trace-cmd/plugins
Quoted 7 lines

but I guess we want global place to install plugins, the traceevent by default uses:

  $(HOME)/.traceevent/plugins
  $(prefix)/lib/traceevent/plugins

How about that?

Looks good to me.

Thanks,
Namhyung

author
#62 lore raw

PATCH 18/22 perf tools: Add udis86 disassembler feature check

In reply to Jiri Olsa

Adding udis86 disassembler feature check which support is needed for kvm:kvm_emulate_insn tracepoint.

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e kvm:kvm_emulate_insn -a')

--- script.kvm.old
+++ script.kvm.new
- qemu-system-x86 15519 [003] 5332.470049: kvm:kvm_emulate_insn: 0:ffffffff8103c596:89 b7 00 80 5f ff (prot64)
+ qemu-system-x86 15519 [003] 5332.470049: kvm:kvm_emulate_insn: 0:ffffffff8103c596: mov %esi, -0xa08000(%rdi)

Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/perf/Makefile.perf                       |  2 +-
 tools/perf/config/Makefile                     |  8 ++++++++
 tools/perf/config/feature-checks/Makefile      |  6 +++++-
 tools/perf/config/feature-checks/test-all.c    |  5 +++++
 tools/perf/config/feature-checks/test-udis86.c | 11 +++++++++++
 5 files changed, 30 insertions(+), 2 deletions(-)
 create mode 100644 tools/perf/config/feature-checks/test-udis86.c
tools/perf/Makefile.perf +1 -1
index b4f8910..d465540 100644
@@ -710,7 +710,7 @@ $(LIB_FILE): $(LIB_OBJS)
 # libtraceevent.a
 TE_SOURCES = $(wildcard $(TRACE_EVENT_DIR)*.[ch])
 
-LIBTRACEEVENT_FLAGS=$(QUIET_SUBDIR1) O=$(OUTPUT) CFLAGS="-g -Wall $(EXTRA_CFLAGS)" plugin_dir=$(plugindir_SQ)
+LIBTRACEEVENT_FLAGS=$(QUIET_SUBDIR1) O=$(OUTPUT) CFLAGS="-g -Wall $(EXTRA_CFLAGS) $(LIBTRACEEVENT_CFLAGS)" plugin_dir=$(plugindir_SQ)
 
 $(LIBTRACEEVENT): $(TE_SOURCES) $(OUTPUT)PERF-CFLAGS
 	$(QUIET_SUBDIR0)$(TRACE_EVENT_DIR) $(LIBTRACEEVENT_FLAGS) libtraceevent.a plugins
tools/perf/config/Makefile +8 -0
index 8c8a205..3f9c6b4 100644
@@ -418,6 +418,14 @@ else
   msg := $(warning No timerfd support. Disables 'perf kvm stat live');
 endif
 
+$(call feature_check,udis86)
+ifeq ($(feature-udis86), 1)
+  LIBTRACEEVENT_CFLAGS += -DHAVE_UDIS86
+  EXTLIBS += -ludis86
+else
+  msg := $(warning No udis86 support.);
+endif
+
 disable-python = $(eval $(disable-python_code))
 define disable-python_code
   CFLAGS += -DNO_LIBPYTHON
tools/perf/config/feature-checks/Makefile +5 -1
index 87e7900..b0df0f2 100644
@@ -27,7 +27,8 @@ FILES=					\
 	test-on-exit			\
 	test-stackprotector-all		\
 	test-stackprotector		\
-	test-timerfd
+	test-timerfd			\
+	test-udis86
 
 CC := $(CC) -MD
 
@@ -144,6 +145,9 @@ test-backtrace:
 test-timerfd:
 	$(BUILD)
 
+test-udis86:
+	$(BUILD) -DHAVE_UDIS86 -ludis86
+
 -include *.d
 
 ###############################
tools/perf/config/feature-checks/test-all.c +5 -0
index 59e7a70..0896ae0 100644
@@ -85,6 +85,10 @@
 # include "test-timerfd.c"
 #undef main
 
+#define main main_test_udis86
+# include "test-udis86.c"
+#undef main
+
 int main(int argc, char *argv[])
 {
 	main_test_libpython();
@@ -106,6 +110,7 @@ int main(int argc, char *argv[])
 	main_test_backtrace();
 	main_test_libnuma();
 	main_test_timerfd();
+	main_test_udis86();
 
 	return 0;
 }
tools/perf/config/feature-checks/test-udis86.c new file +11 -0
new file mode 100644
index 0000000..dbc2c26
@@ -0,0 +1,11 @@
+
+/*
+ * test for udis86 functions used by plugin_kvm.c
+ */
+
+#include <udis86.h>
+
+int main(void)
+{
+	return puts("hi");
+}
Signature · 1 line
1.8.3.1
maintainer
#63 lore raw

Re: [PATCH 18/22] perf tools: Add udis86 disassembler feature check

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 10 lines

Adding udis86 disassembler feature check which support is needed for kvm:kvm_emulate_insn tracepoint.

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e kvm:kvm_emulate_insn -a')

script.kvm.new renamed from script.kvm.old +1 -1
- qemu-system-x86 15519 [003]  5332.470049: kvm:kvm_emulate_insn: 0:ffffffff8103c596:89 b7 00 80 5f ff (prot64)
+ qemu-system-x86 15519 [003]  5332.470049: kvm:kvm_emulate_insn: 0:ffffffff8103c596: mov %esi, -0xa08000(%rdi)

[SNIP]

tools/perf/Makefile.perf $(LIB_FILE): $(LIB_OBJS) · line 713

Quoted 8 lines
tools/perf/Makefile.perf +1 -1
@@ -710,7 +710,7 @@ $(LIB_FILE): $(LIB_OBJS)
 # libtraceevent.a
 TE_SOURCES = $(wildcard $(TRACE_EVENT_DIR)*.[ch])
 
-LIBTRACEEVENT_FLAGS=$(QUIET_SUBDIR1) O=$(OUTPUT) CFLAGS="-g -Wall $(EXTRA_CFLAGS)" plugin_dir=$(plugindir_SQ)
+LIBTRACEEVENT_FLAGS=$(QUIET_SUBDIR1) O=$(OUTPUT) CFLAGS="-g -Wall $(EXTRA_CFLAGS) $(LIBTRACEEVENT_CFLAGS)" plugin_dir=$(plugindir_SQ)

What about breaking the line with "+="?

Quoted 8 lines
tools/perf/Makefile.perf +0 -0
 $(LIBTRACEEVENT): $(TE_SOURCES) $(OUTPUT)PERF-CFLAGS
 	$(QUIET_SUBDIR0)$(TRACE_EVENT_DIR) $(LIBTRACEEVENT_FLAGS) libtraceevent.a plugins
tools/perf/config/Makefile +0 -0
index 8c8a205..3f9c6b4 100644

tools/perf/config/Makefile else · line 427

Quoted 10 lines
tools/perf/config/Makefile +7 -0
@@ -418,6 +418,14 @@ else
   msg := $(warning No timerfd support. Disables 'perf kvm stat live');
 endif
 
+$(call feature_check,udis86)
+ifeq ($(feature-udis86), 1)
+  LIBTRACEEVENT_CFLAGS += -DHAVE_UDIS86
+  EXTLIBS += -ludis86
+else
+  msg := $(warning No udis86 support.);
+endif

I think we can just use libbfd for this eventually. But it can be a later work. :)

Thanks,
Namhyung

author
#64 lore raw

Re: [PATCH 18/22] perf tools: Add udis86 disassembler feature check

Quoted 19 lines · Namhyung Kim

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 10 lines

Adding udis86 disassembler feature check which support is needed for kvm:kvm_emulate_insn tracepoint.

The diff of 'perf script' output generated by old and new code:
(data was generated by 'perf record -e kvm:kvm_emulate_insn -a')

script.kvm.new renamed from script.kvm.old +1 -1
- qemu-system-x86 15519 [003]  5332.470049: kvm:kvm_emulate_insn: 0:ffffffff8103c596:89 b7 00 80 5f ff (prot64)
+ qemu-system-x86 15519 [003]  5332.470049: kvm:kvm_emulate_insn: 0:ffffffff8103c596: mov %esi, -0xa08000(%rdi)

[SNIP]

Quoted 6 lines
tools/perf/Makefile.perf +0 -0
@@ -710,7 +710,7 @@ $(LIB_FILE): $(LIB_OBJS)
 # libtraceevent.a
 TE_SOURCES = $(wildcard $(TRACE_EVENT_DIR)*.[ch])
 
Quoted 4 lines

tools/perf/Makefile.perf $(LIB_FILE): $(LIB_OBJS) · line 713

Quoted 2 lines
tools/perf/Makefile.perf +1 -1
-LIBTRACEEVENT_FLAGS=$(QUIET_SUBDIR1) O=$(OUTPUT) CFLAGS="-g -Wall $(EXTRA_CFLAGS)" plugin_dir=$(plugindir_SQ)
+LIBTRACEEVENT_FLAGS=$(QUIET_SUBDIR1) O=$(OUTPUT) CFLAGS="-g -Wall $(EXTRA_CFLAGS) $(LIBTRACEEVENT_CFLAGS)" plugin_dir=$(plugindir_SQ)

What about breaking the line with "+="?

ok

thanks,
jirka

author
#65 lore raw

PATCH 19/22 perf tools: Add filename__read_str util function

In reply to Jiri Olsa

Adding filename__read_str util function to read text file and return it in the char array.

The interface is:
int filename__read_str(const char *filename, char **buf, size_t *sizep)

Returns 0/-1 if the read suceeded/fail respectively.

buf - place to store the data pointer size - place to store data size

Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/perf/util/util.c | 41 +++++++++++++++++++++++++++++++++++++++++
 tools/perf/util/util.h |  1 +
 2 files changed, 42 insertions(+)
tools/perf/util/util.c +41 -0
index 28a0a89..e96cc90 100644
@@ -413,3 +413,44 @@ int filename__read_int(const char *filename, int *value)
 	close(fd);
 	return err;
 }
+
+int filename__read_str(const char *filename, char **buf, size_t *sizep)
+{
+	size_t size = 0, alloc_size = 0;
+	void *bf = NULL, *nbf;
+	int fd, n, err = 0;
+
+	fd = open(filename, O_RDONLY);
+	if (fd < 0)
+		return -errno;
+
+	do {
+		if (size == alloc_size) {
+			alloc_size += BUFSIZ;
+			nbf = realloc(bf, alloc_size);
+			if (!nbf) {
+				err = -ENOMEM;
+				break;
+			}
+
+			bf = nbf;
+		}
+
+		n = read(fd, bf + size, BUFSIZ);
+		if (n < 0) {
+			err = 0;
+			break;
+		}
+
+		size += n;
+	} while (n > 0);
+
+	if (!err) {
+		*sizep = size;
+		*buf   = bf;
+	} else
+		free(bf);
+
+	close(fd);
+	return err;
+}
tools/perf/util/util.h +1 -0
index c8f362d..e54bb10 100644
@@ -307,4 +307,5 @@ char *get_srcline(struct dso *dso, unsigned long addr);
 void free_srcline(char *srcline);
 
 int filename__read_int(const char *filename, int *value);
+int filename__read_str(const char *filename, char **buf, size_t *sizep);
 #endif /* GIT_COMPAT_UTIL_H */
Signature · 1 line
1.8.3.1
maintainer
#66 lore raw

Re: [PATCH 19/22] perf tools: Add filename__read_str util function

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 10 lines

Adding filename__read_str util function to read
text file and return it in the char array.

The interface is:
int filename__read_str(const char *filename, char **buf, size_t *sizep)

Returns 0/-1 if the read suceeded/fail respectively.

buf - place to store the data pointer
size - place to store data size

[SNIP]

Quoted 13 lines
tools/perf/util/util.c +13 -0
+int filename__read_str(const char *filename, char **buf, size_t *sizep)
+{
+	size_t size = 0, alloc_size = 0;
+	void *bf = NULL, *nbf;
+	int fd, n, err = 0;
+
+	fd = open(filename, O_RDONLY);
+	if (fd < 0)
+		return -errno;
+
+	do {
+		if (size == alloc_size) {
+			alloc_size += BUFSIZ;

tools/perf/util/util.c int filename__read_int(const char *filename, int *value) · line 439

Quoted 10 lines
tools/perf/util/util.c +10 -0
+			nbf = realloc(bf, alloc_size);
+			if (!nbf) {
+				err = -ENOMEM;
+				break;
+			}
+
+			bf = nbf;
+		}
+
+		n = read(fd, bf + size, BUFSIZ);

Shouldn't it be "read(fd, bf + size, alloc_size - size)"? Otherwise there might be a problem if read() returned early for some reason with small size and then retry with a full BUFSIZ..

tools/perf/util/util.c int filename__read_int(const char *filename, int *value) · line 441

Quoted 2 lines
tools/perf/util/util.c +2 -0
+		if (n < 0) {
+			err = 0;

I think it needs to check the size also since read() might fail at the first invocation. What about this?

		if (n < 0) {
			if (size)
				err = 0;
			else
				err = -errno;

Thanks,
Namhyung

Quoted 15 lines
tools/perf/util/util.c +15 -0
+			break;
+		}
+
+		size += n;
+	} while (n > 0);
+
+	if (!err) {
+		*sizep = size;
+		*buf   = bf;
+	} else
+		free(bf);
+
+	close(fd);
+	return err;
+}
author
#67 lore raw

Re: [PATCH 19/22] perf tools: Add filename__read_str util function

Quoted 48 lines · Namhyung Kim

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 10 lines

Adding filename__read_str util function to read
text file and return it in the char array.

The interface is:
int filename__read_str(const char *filename, char **buf, size_t *sizep)

Returns 0/-1 if the read suceeded/fail respectively.

buf - place to store the data pointer
size - place to store data size

[SNIP]

Quoted 13 lines
tools/perf/util/util.c +13 -0
+int filename__read_str(const char *filename, char **buf, size_t *sizep)
+{
+	size_t size = 0, alloc_size = 0;
+	void *bf = NULL, *nbf;
+	int fd, n, err = 0;
+
+	fd = open(filename, O_RDONLY);
+	if (fd < 0)
+		return -errno;
+
+	do {
+		if (size == alloc_size) {
+			alloc_size += BUFSIZ;

tools/perf/util/util.c int filename__read_int(const char *filename, int *value) · line 439

Quoted 10 lines
tools/perf/util/util.c +10 -0
+			nbf = realloc(bf, alloc_size);
+			if (!nbf) {
+				err = -ENOMEM;
+				break;
+			}
+
+			bf = nbf;
+		}
+
+		n = read(fd, bf + size, BUFSIZ);

Shouldn't it be "read(fd, bf + size, alloc_size - size)"? Otherwise there might be a problem if read() returned early for some reason with small size and then retry with a full BUFSIZ..

tools/perf/util/util.c int filename__read_int(const char *filename, int *value) · line 441

Quoted 2 lines
tools/perf/util/util.c +2 -0
+		if (n < 0) {
+			err = 0;

I think it needs to check the size also since read() might fail at the first invocation. What about this?

Quoted 6 lines
		if (n < 0) {
			if (size)
				err = 0;
			else
				err = -errno;

ok, added.. plus warning in the 'if (size)' leg,
because we're returning data even if we failed.

thanks,
jirka

author
#68 lore raw

PATCH 20/22 perf tools: Add trace-event object

In reply to Jiri Olsa

Add trace-event object to keep together 'struct pevent' object with its loaded plugins with following interface:

int trace_event__init(struct trace_event *t);
- initalizes 'struct pevent' object and loads plugins for it

void trace_event__cleanup(struct trace_event *t);
- cleanups both 'struct pevent' and plugins

Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/perf/Makefile.perf            |  1 +
 tools/perf/builtin-script.c         |  2 +-
 tools/perf/util/header.c            |  8 ++++----
 tools/perf/util/python-ext-sources  |  1 +
 tools/perf/util/session.h           |  3 ++-
 tools/perf/util/trace-event-parse.c | 13 -------------
 tools/perf/util/trace-event-read.c  | 20 +++++++++++---------
 tools/perf/util/trace-event.c       | 21 +++++++++++++++++++++
 tools/perf/util/trace-event.h       | 13 ++++++++++---
 9 files changed, 51 insertions(+), 31 deletions(-)
 create mode 100644 tools/perf/util/trace-event.c
tools/perf/Makefile.perf +1 -0
index d465540..90790c3 100644
@@ -353,6 +353,7 @@ LIB_OBJS += $(OUTPUT)util/pmu-bison.o
 LIB_OBJS += $(OUTPUT)util/trace-event-read.o
 LIB_OBJS += $(OUTPUT)util/trace-event-info.o
 LIB_OBJS += $(OUTPUT)util/trace-event-scripting.o
+LIB_OBJS += $(OUTPUT)util/trace-event.o
 LIB_OBJS += $(OUTPUT)util/svghelper.o
 LIB_OBJS += $(OUTPUT)util/sort.o
 LIB_OBJS += $(OUTPUT)util/hist.o
tools/perf/builtin-script.c +1 -1
index b392770..0615e2a 100644
@@ -1587,7 +1587,7 @@ int cmd_script(int argc, const char **argv, const char *prefix __maybe_unused)
 			return -1;
 		}
 
-		err = scripting_ops->generate_script(session->pevent,
+		err = scripting_ops->generate_script(session->tevent.pevent,
 						     "perf-script");
 		goto out;
 	}
tools/perf/util/header.c +4 -4
index 1cd0357..0af4470 100644
@@ -2834,11 +2834,11 @@ int perf_session__read_header(struct perf_session *session)
 
 	symbol_conf.nr_events = nr_attrs;
 
-	perf_header__process_sections(header, fd, &session->pevent,
+	perf_header__process_sections(header, fd, &session->tevent,
 				      perf_file_section__process);
 
 	if (perf_evlist__prepare_tracepoint_events(session->evlist,
-						   session->pevent))
+						   session->tevent.pevent))
 		goto out_delete_evlist;
 
 	return 0;
@@ -3003,7 +3003,7 @@ int perf_event__process_tracing_data(struct perf_tool *tool __maybe_unused,
 	lseek(fd, offset + sizeof(struct tracing_data_event),
 	      SEEK_SET);
 
-	size_read = trace_report(fd, &session->pevent,
+	size_read = trace_report(fd, &session->tevent,
 				 session->repipe);
 	padding = PERF_ALIGN(size_read, sizeof(u64)) - size_read;
 
@@ -3025,7 +3025,7 @@ int perf_event__process_tracing_data(struct perf_tool *tool __maybe_unused,
 	}
 
 	perf_evlist__prepare_tracepoint_events(session->evlist,
-					       session->pevent);
+					       session->tevent.pevent);
 
 	return size_read + padding;
 }
tools/perf/util/python-ext-sources +1 -0
index 239036f..595bfc7 100644
@@ -18,4 +18,5 @@ util/cgroup.c
 util/rblist.c
 util/strlist.c
 util/fs.c
+util/trace-event.c
 ../../lib/rbtree.c
tools/perf/util/session.h +2 -1
index 50f6409..83cba9b 100644
@@ -1,6 +1,7 @@
 #ifndef __PERF_SESSION_H
 #define __PERF_SESSION_H
 
+#include "trace-event.h"
 #include "hist.h"
 #include "event.h"
 #include "header.h"
@@ -32,7 +33,7 @@ struct perf_session {
 	struct perf_header	header;
 	struct machines		machines;
 	struct perf_evlist	*evlist;
-	struct pevent		*pevent;
+	struct trace_event	tevent;
 	struct events_stats	stats;
 	bool			repipe;
 	struct ordered_samples	ordered_samples;
tools/perf/util/trace-event-parse.c +0 -13
index c8dae72..044f5cf 100644
@@ -28,19 +28,6 @@
 #include "util.h"
 #include "trace-event.h"
 
-struct pevent *read_trace_init(int file_bigendian, int host_bigendian)
-{
-	struct pevent *pevent = pevent_alloc();
-
-	if (pevent != NULL) {
-		pevent_set_flag(pevent, PEVENT_NSEC_OUTPUT);
-		pevent_set_file_bigendian(pevent, file_bigendian);
-		pevent_set_host_bigendian(pevent, host_bigendian);
-	}
-
-	return pevent;
-}
-
 static int get_common_field(struct scripting_context *context,
 			    int *offset, int *size, const char *type)
 {
tools/perf/util/trace-event-read.c +11 -9
index f211227..e113e18 100644
@@ -343,7 +343,7 @@ static int read_event_files(struct pevent *pevent)
 	return 0;
 }
 
-ssize_t trace_report(int fd, struct pevent **ppevent, bool __repipe)
+ssize_t trace_report(int fd, struct trace_event *tevent, bool __repipe)
 {
 	char buf[BUFSIZ];
 	char test[] = { 23, 8, 68 };
@@ -356,11 +356,9 @@ ssize_t trace_report(int fd, struct pevent **ppevent, bool __repipe)
 	int host_bigendian;
 	int file_long_size;
 	int file_page_size;
-	struct pevent *pevent;
+	struct pevent *pevent = NULL;
 	int err;
 
-	*ppevent = NULL;
-
 	repipe = __repipe;
 	input_fd = fd;
 
@@ -390,12 +388,17 @@ ssize_t trace_report(int fd, struct pevent **ppevent, bool __repipe)
 	file_bigendian = buf[0];
 	host_bigendian = bigendian();
 
-	pevent = read_trace_init(file_bigendian, host_bigendian);
-	if (pevent == NULL) {
-		pr_debug("read_trace_init failed");
+	if (trace_event__init(tevent)) {
+		pr_debug("trace_event__init failed");
 		goto out;
 	}
 
+	pevent = tevent->pevent;
+
+	pevent_set_flag(pevent, PEVENT_NSEC_OUTPUT);
+	pevent_set_file_bigendian(pevent, file_bigendian);
+	pevent_set_host_bigendian(pevent, host_bigendian);
+
 	if (do_read(buf, 1) < 0)
 		goto out;
 	file_long_size = buf[0];
@@ -432,11 +435,10 @@ ssize_t trace_report(int fd, struct pevent **ppevent, bool __repipe)
 		pevent_print_printk(pevent);
 	}
 
-	*ppevent = pevent;
 	pevent = NULL;
 
 out:
 	if (pevent)
-		pevent_free(pevent);
+		trace_event__cleanup(tevent);
 	return size;
 }
tools/perf/util/trace-event.c new file +21 -0
new file mode 100644
index 0000000..a155a77
@@ -0,0 +1,21 @@
+
+#include <traceevent/event-parse.h>
+#include "trace-event.h"
+
+int trace_event__init(struct trace_event *t)
+{
+	struct pevent *pevent = pevent_alloc();
+
+	if (pevent) {
+		t->plugin_list = traceevent_load_plugins(pevent);
+		t->pevent  = pevent;
+	}
+
+	return pevent ? 0 : -1;
+}
+
+void trace_event__cleanup(struct trace_event *t)
+{
+	pevent_free(t->pevent);
+	traceevent_unload_plugins(t->plugin_list);
+}
tools/perf/util/trace-event.h +10 -3
index 04df631..75dded3 100644
@@ -3,17 +3,24 @@
 
 #include <traceevent/event-parse.h>
 #include "parse-events.h"
-#include "session.h"
 
 struct machine;
 struct perf_sample;
 union perf_event;
 struct perf_tool;
 struct thread;
+struct plugin_list;
+
+struct trace_event {
+	struct pevent		*pevent;
+	struct plugin_list	*plugin_list;
+};
+
+int trace_event__init(struct trace_event *t);
+void trace_event__cleanup(struct trace_event *t);
 
 int bigendian(void);
 
-struct pevent *read_trace_init(int file_bigendian, int host_bigendian);
 void event_format__print(struct event_format *event,
 			 int cpu, void *data, int size);
 
@@ -27,7 +34,7 @@ raw_field_value(struct event_format *event, const char *name, void *data);
 void parse_proc_kallsyms(struct pevent *pevent, char *file, unsigned int size);
 void parse_ftrace_printk(struct pevent *pevent, char *file, unsigned int size);
 
-ssize_t trace_report(int fd, struct pevent **pevent, bool repipe);
+ssize_t trace_report(int fd, struct trace_event *tevent, bool repipe);
 
 struct event_format *trace_find_next_event(struct pevent *pevent,
 					   struct event_format *event);
Signature · 1 line
1.8.3.1
author
#69 lore raw

PATCH 21/22 perf tools: Add trace-event global object for tracepoint interface

In reply to Jiri Olsa

In order to get the proper plugins processing we need to use full trace-event interface when creating tracepoint events. So far we were using shortcut to get the parsed format.

Moving current 'event_format__new' function into trace-event object as 'trace_event__tp_format'.

This function uses properly initialized global trace-event object, ensuring proper plugins processing.

Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/perf/builtin-trace.c    |  5 ++--
 tools/perf/util/evsel.c       | 44 ++-----------------------------
 tools/perf/util/trace-event.c | 61 +++++++++++++++++++++++++++++++++++++++++++
 tools/perf/util/trace-event.h |  2 ++
 4 files changed, 68 insertions(+), 44 deletions(-)
tools/perf/builtin-trace.c +3 -2
index e9f345e2..47bf590 100644
@@ -11,6 +11,7 @@
 #include "util/intlist.h"
 #include "util/thread_map.h"
 #include "util/stat.h"
+#include "trace-event.h"
 
 #include <libaudit.h>
 #include <stdlib.h>
@@ -1430,11 +1431,11 @@ static int trace__read_syscall_info(struct trace *trace, int id)
 	sc->fmt  = syscall_fmt__find(sc->name);
 
 	snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->name);
-	sc->tp_format = event_format__new("syscalls", tp_name);
+	sc->tp_format = trace_event__tp_format("syscalls", tp_name);
 
 	if (sc->tp_format == NULL && sc->fmt && sc->fmt->alias) {
 		snprintf(tp_name, sizeof(tp_name), "sys_enter_%s", sc->fmt->alias);
-		sc->tp_format = event_format__new("syscalls", tp_name);
+		sc->tp_format = trace_event__tp_format("syscalls", tp_name);
 	}
 
 	if (sc->tp_format == NULL)
tools/perf/util/evsel.c +2 -42

62 lines not on this page · raw message

tools/perf/util/trace-event.c +61 -0

70 lines not on this page · raw message

tools/perf/util/trace-event.h +2 -0
index 75dded3..3a01618 100644
@@ -18,6 +18,8 @@ struct trace_event {
 
 int trace_event__init(struct trace_event *t);
 void trace_event__cleanup(struct trace_event *t);
+struct event_format*
+trace_event__tp_format(const char *sys, const char *name);
 
 int bigendian(void);
 
Signature · 1 line
1.8.3.1
maintainer
#70 lore raw

Re: [PATCH 21/22] perf tools: Add trace-event global object for tracepoint interface

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 10 lines

In order to get the proper plugins processing we need to use full trace-event interface when creating tracepoint events. So far we were using shortcut to get the parsed format.

Moving current 'event_format__new' function into trace-event object as 'trace_event__tp_format'.

This function uses properly initialized global trace-event object, ensuring proper plugins processing.

I see this when I compile it on my old ubuntu 10.04 box (with gcc 4.4.3)

CC util/trace-event.o
cc1: warnings being treated as errors util/trace-event.c: In function 'trace_event__tp_format':
util/trace-event.c:63: error: not protecting function: no buffer at least 8 bytes long
make[1]: *** [util/trace-event.o] Error 1

Thanks,
Namhyung

author
#71 lore raw

Re: [PATCH 21/22] perf tools: Add trace-event global object for tracepoint interface

Quoted 12 lines · Namhyung Kim

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 10 lines

In order to get the proper plugins processing we need to use full trace-event interface when creating tracepoint events. So far we were using shortcut to get the parsed format.

Moving current 'event_format__new' function into trace-event object as 'trace_event__tp_format'.

This function uses properly initialized global trace-event object, ensuring proper plugins processing.

Quoted 8 lines

I see this when I compile it on my old ubuntu 10.04 box (with gcc 4.4.3)

CC util/trace-event.o
cc1: warnings being treated as errors util/trace-event.c: In function 'trace_event__tp_format':
util/trace-event.c:63: error: not protecting function: no buffer at least 8 bytes long
make[1]: *** [util/trace-event.o] Error 1

hum, what was the stackprotector status?

... stackprotector: [ on ]
... stackprotector-all: [ on ]

thanks,
jirka

maintainer
#72 lore raw

Re: [PATCH 21/22] perf tools: Add trace-event global object for tracepoint interface

Hi Jiri,

2013-11-25 (월), 17:03 +0100, Jiri Olsa:

Quoted 22 lines · Namhyung Kim
Quoted 12 lines

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 10 lines

In order to get the proper plugins processing we need to use full trace-event interface when creating tracepoint events. So far we were using shortcut to get the parsed format.

Moving current 'event_format__new' function into trace-event object as 'trace_event__tp_format'.

This function uses properly initialized global trace-event object, ensuring proper plugins processing.

Quoted 8 lines

I see this when I compile it on my old ubuntu 10.04 box (with gcc 4.4.3)

CC util/trace-event.o
cc1: warnings being treated as errors util/trace-event.c: In function 'trace_event__tp_format':
util/trace-event.c:63: error: not protecting function: no buffer at least 8 bytes long
make[1]: *** [util/trace-event.o] Error 1

Quoted 4 lines

hum, what was the stackprotector status?

... stackprotector: [ on ]
... stackprotector-all: [ on ]

Right. Both of them were turned on.

Auto-detecting system features:

...                     backtrace: [ on  ]
...                         dwarf: [ on  ]
...                fortify-source: [ on  ]
...                         glibc: [ on  ]
...                          gtk2: [ on  ]
...                  gtk2-infobar: [ on  ]
...                      libaudit: [ on  ]
...                        libbfd: [ on  ]
...                        libelf: [ on  ]
...             libelf-getphdrnum: [ OFF ]
...                   libelf-mmap: [ on  ]
...                       libnuma: [ OFF ]
...                       libperl: [ on  ]
...                     libpython: [ on  ]
...             libpython-version: [ on  ]
...                      libslang: [ on  ]
...                     libunwind: [ OFF ]
...                       on-exit: [ on  ]
...                stackprotector: [ on  ]
...            stackprotector-all: [ on  ]
...                       timerfd: [ on  ]

Thanks,
Namhyung

author
#73 lore raw

Re: [PATCH 21/22] perf tools: Add trace-event global object for tracepoint interface

Quoted 46 lines · Namhyung Kim

Hi Jiri,

2013-11-25 (월), 17:03 +0100, Jiri Olsa:

Quoted 22 lines · Namhyung Kim
Quoted 12 lines

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 10 lines

In order to get the proper plugins processing we need to use full trace-event interface when creating tracepoint events. So far we were using shortcut to get the parsed format.

Moving current 'event_format__new' function into trace-event object as 'trace_event__tp_format'.

This function uses properly initialized global trace-event object, ensuring proper plugins processing.

Quoted 8 lines

I see this when I compile it on my old ubuntu 10.04 box (with gcc 4.4.3)

CC util/trace-event.o
cc1: warnings being treated as errors util/trace-event.c: In function 'trace_event__tp_format':
util/trace-event.c:63: error: not protecting function: no buffer at least 8 bytes long
make[1]: *** [util/trace-event.o] Error 1

Quoted 4 lines

hum, what was the stackprotector status?

... stackprotector: [ on ]
... stackprotector-all: [ on ]

Right. Both of them were turned on.

Auto-detecting system features:

...                     backtrace: [ on  ]
...                         dwarf: [ on  ]
...                fortify-source: [ on  ]
...                         glibc: [ on  ]
...                          gtk2: [ on  ]
...                  gtk2-infobar: [ on  ]
...                      libaudit: [ on  ]
...                        libbfd: [ on  ]
...                        libelf: [ on  ]
...             libelf-getphdrnum: [ OFF ]
...                   libelf-mmap: [ on  ]
...                       libnuma: [ OFF ]
...                       libperl: [ on  ]
Quoted 8 lines
...                     libpython: [ on  ]
...             libpython-version: [ on  ]
...                      libslang: [ on  ]
...                     libunwind: [ OFF ]
...                       on-exit: [ on  ]
...                stackprotector: [ on  ]
...            stackprotector-all: [ on  ]
...                       timerfd: [ on  ]

I found that -fstack-protector might not always work:
https://bugs.launchpad.net/ubuntu/+source/gcc-4.5/+bug/691722

they say it was fixed in gcc 4.5.0 which I was not able to prove in gcc release notes..

we could disable the warning like in the below
for gcc < 4.5.0, could you please test?

any other idea? ;-)

thanks,
jirka

---

tools/perf/config/feature-checks/test-stackprotector.c +8 -0
index c9f398d..c0ace8b 100644
@@ -1,5 +1,13 @@
 #include <stdio.h>
 
+#define GCC_VERSION (__GNUC__ * 10000 \
+		+ __GNUC_MINOR__ * 100 \
+		+ __GNUC_PATCHLEVEL__)
+
+#if GCC_VERSION < 40500
+#error "failed"
+#endif
+
 int main(void)
 {
 	return puts("hi");
author
#74 lore raw

Re: [PATCH 21/22] perf tools: Add trace-event global object for tracepoint interface

Quoted 59 lines · Jiri Olsa
Quoted 46 lines · Namhyung Kim

Hi Jiri,

2013-11-25 (월), 17:03 +0100, Jiri Olsa:

Quoted 22 lines · Namhyung Kim
Quoted 12 lines

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 10 lines

In order to get the proper plugins processing we need to use full trace-event interface when creating tracepoint events. So far we were using shortcut to get the parsed format.

Moving current 'event_format__new' function into trace-event object as 'trace_event__tp_format'.

This function uses properly initialized global trace-event object, ensuring proper plugins processing.

Quoted 8 lines

I see this when I compile it on my old ubuntu 10.04 box (with gcc 4.4.3)

CC util/trace-event.o
cc1: warnings being treated as errors util/trace-event.c: In function 'trace_event__tp_format':
util/trace-event.c:63: error: not protecting function: no buffer at least 8 bytes long
make[1]: *** [util/trace-event.o] Error 1

Quoted 4 lines

hum, what was the stackprotector status?

... stackprotector: [ on ]
... stackprotector-all: [ on ]

Right. Both of them were turned on.

Auto-detecting system features:

...                     backtrace: [ on  ]
...                         dwarf: [ on  ]
...                fortify-source: [ on  ]
...                         glibc: [ on  ]
...                          gtk2: [ on  ]
...                  gtk2-infobar: [ on  ]
...                      libaudit: [ on  ]
...                        libbfd: [ on  ]
...                        libelf: [ on  ]
...             libelf-getphdrnum: [ OFF ]
...                   libelf-mmap: [ on  ]
...                       libnuma: [ OFF ]
...                       libperl: [ on  ]
Quoted 8 lines
...                     libpython: [ on  ]
...             libpython-version: [ on  ]
...                      libslang: [ on  ]
...                     libunwind: [ OFF ]
...                       on-exit: [ on  ]
...                stackprotector: [ on  ]
...            stackprotector-all: [ on  ]
...                       timerfd: [ on  ]

I found that -fstack-protector might not always work:
https://bugs.launchpad.net/ubuntu/+source/gcc-4.5/+bug/691722

Quoted 7 lines

they say it was fixed in gcc 4.5.0 which I was not able to prove in gcc release notes..

we could disable the warning like in the below
for gcc < 4.5.0, could you please test?

any other idea? ;-)

or remove -Wstack-protector option completely, as it's just informative seems there's nothing you could do about the warning anyway

jirka

maintainer
#75 lore raw

Re: [PATCH 21/22] perf tools: Add trace-event global object for tracepoint interface

2013-11-27 (수), 16:27 +0100, Jiri Olsa:

Quoted 57 lines · Jiri Olsa
Quoted 56 lines · Namhyung Kim
Quoted 54 lines

Hi Jiri,

2013-11-25 (월), 17:03 +0100, Jiri Olsa:

Quoted 22 lines · Namhyung Kim
Quoted 12 lines

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 10 lines

In order to get the proper plugins processing we need to use full trace-event interface when creating tracepoint events. So far we were using shortcut to get the parsed format.

Moving current 'event_format__new' function into trace-event object as 'trace_event__tp_format'.

This function uses properly initialized global trace-event object, ensuring proper plugins processing.

Quoted 8 lines

I see this when I compile it on my old ubuntu 10.04 box (with gcc 4.4.3)

CC util/trace-event.o
cc1: warnings being treated as errors util/trace-event.c: In function 'trace_event__tp_format':
util/trace-event.c:63: error: not protecting function: no buffer at least 8 bytes long
make[1]: *** [util/trace-event.o] Error 1

Quoted 4 lines

hum, what was the stackprotector status?

... stackprotector: [ on ]
... stackprotector-all: [ on ]

Right. Both of them were turned on.

Auto-detecting system features:

...                     backtrace: [ on  ]
...                         dwarf: [ on  ]
...                fortify-source: [ on  ]
...                         glibc: [ on  ]
...                          gtk2: [ on  ]
...                  gtk2-infobar: [ on  ]
...                      libaudit: [ on  ]
...                        libbfd: [ on  ]
...                        libelf: [ on  ]
...             libelf-getphdrnum: [ OFF ]
...                   libelf-mmap: [ on  ]
...                       libnuma: [ OFF ]
...                       libperl: [ on  ]
...                     libpython: [ on  ]
...             libpython-version: [ on  ]
...                      libslang: [ on  ]
...                     libunwind: [ OFF ]
...                       on-exit: [ on  ]
...                stackprotector: [ on  ]
...            stackprotector-all: [ on  ]
...                       timerfd: [ on  ]
Quoted 8 lines
Quoted 8 lines

I found that -fstack-protector might not always work:
https://bugs.launchpad.net/ubuntu/+source/gcc-4.5/+bug/691722

they say it was fixed in gcc 4.5.0 which I was not able to prove in gcc release notes..

we could disable the warning like in the below
for gcc < 4.5.0, could you please test?

Works well for me, thanks.

Quoted 5 lines
Quoted 2 lines

any other idea? ;-)

or remove -Wstack-protector option completely, as it's just informative seems there's nothing you could do about the warning anyway

Agreed. It seems just adding -fstack-protector is enough.

Thanks,
Namhyung

author
#76 lore raw

PATCH 22/22 perf tools: Overload pr_stat traceevent print function

In reply to Jiri Olsa

The traceevent lib uses pr_stat to display all standard info. It's defined as __weak. Overloading it with perf version plugged into perf output system logic.

Displaying the pr_stat stuff under '-v' option.

Cc Corey Ashford <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc David Ahern <[email protected]>
---
 tools/perf/util/debug.c | 16 ++++++++++++++++
 tools/perf/util/debug.h |  2 ++
 2 files changed, 18 insertions(+)
tools/perf/util/debug.c +16 -0
index 399e74c..dc93bc3 100644
@@ -33,6 +33,22 @@ int eprintf(int level, const char *fmt, ...)
 	return ret;
 }
 
+/*
+ * Overloading libtraceevent standard info print
+ * function, display with -v in perf.
+ */
+void pr_stat(const char *fmt, ...)
+{
+	va_list args;
+
+	if (verbose >= 1) {
+		va_start(args, fmt);
+		vfprintf(stderr, fmt, args);
+		fprintf(stderr, "\n");
+		va_end(args);
+	}
+}
+
 int dump_printf(const char *fmt, ...)
 {
 	va_list args;
tools/perf/util/debug.h +2 -0
index efbd988..443694c 100644
@@ -17,4 +17,6 @@ void trace_event(union perf_event *event);
 int ui__error(const char *format, ...) __attribute__((format(printf, 1, 2)));
 int ui__warning(const char *format, ...) __attribute__((format(printf, 1, 2)));
 
+void pr_stat(const char *fmt, ...);
+
 #endif	/* __PERF_DEBUG_H */
Signature · 1 line
1.8.3.1
maintainer
#77 lore raw

Re: [PATCH 22/22] perf tools: Overload pr_stat traceevent print function

2013-11-21 (목), 12:01 +0100, Jiri Olsa:

Quoted 5 lines

The traceevent lib uses pr_stat to display all standard info. It's defined as __weak. Overloading it with perf version plugged into perf output system logic.

Displaying the pr_stat stuff under '-v' option.

[SNIP]

Quoted 6 lines
tools/perf/util/debug.c +5 -0
+/*
+ * Overloading libtraceevent standard info print
+ * function, display with -v in perf.
+ */
+void pr_stat(const char *fmt, ...)

tools/perf/util/debug.c int eprintf(int level, const char *fmt, ...) · line 50

Quoted 10 lines
tools/perf/util/debug.c +10 -0
+{
+	va_list args;
+
+	if (verbose >= 1) {
+		va_start(args, fmt);
+		vfprintf(stderr, fmt, args);
+		fprintf(stderr, "\n");
+		va_end(args);
+	}
+}

It'd be better if it use or does same as eprintf() like other pr_* functions does.

Thanks,
Namhyung