* [PATCH] perf: Differentiate exec() and non-exec() comm events
@ 2014-05-28 8:45 Adrian Hunter
2014-05-28 8:55 ` Peter Zijlstra
0 siblings, 1 reply; 6+ messages in thread
From: Adrian Hunter @ 2014-05-28 8:45 UTC (permalink / raw)
To: Peter Zijlstra
Cc: Ingo Molnar, Dave Jones, Arnaldo Carvalho de Melo, linux-kernel,
David Ahern, Jiri Olsa, Paul Mackerras
perf tools like 'perf report' can aggregate samples by comm
strings, which generally works. However, there are other
potential use-cases. For example, to pair up 'calls'
with 'returns' accurately (from branch events like Intel BTS)
it is necessary to identify whether the process has exec'd.
Although a comm event is generated when an 'exec' happens
it is also generated whenever the comm string is changed
on a whim (e.g. by prctl PR_SET_NAME). This patch adds a
flag to the comm event to differentiate one case from the
other.
In order to determine whether the kernel supports the new
flag, a selection bit named 'exec' is added to struct
perf_event_attr. The bit does nothing but will cause
perf_event_open() to fail if the bit is set on kernels
that do not have it defined.
Cc: Dave Jones <davej@redhat.com>
Link: http://lkml.kernel.org/r/537D9EBE.7030806@intel.com
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
---
fs/exec.c | 6 +++---
include/linux/perf_event.h | 4 ++--
include/linux/sched.h | 6 +++++-
include/uapi/linux/perf_event.h | 9 +++++++--
kernel/events/core.c | 4 ++--
5 files changed, 19 insertions(+), 10 deletions(-)
diff --git a/fs/exec.c b/fs/exec.c
index eb83064..a9a9591 100644
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -1046,13 +1046,13 @@ EXPORT_SYMBOL_GPL(get_task_comm);
* so that a new one can be started
*/
-void set_task_comm(struct task_struct *tsk, const char *buf)
+void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
{
task_lock(tsk);
trace_task_rename(tsk, buf);
strlcpy(tsk->comm, buf, sizeof(tsk->comm));
task_unlock(tsk);
- perf_event_comm(tsk);
+ perf_event_comm(tsk, exec);
}
int flush_old_exec(struct linux_binprm * bprm)
@@ -1111,7 +1111,7 @@ void setup_new_exec(struct linux_binprm * bprm)
set_dumpable(current->mm, suid_dumpable);
perf_event_exec();
- set_task_comm(current, kbasename(bprm->filename));
+ __set_task_comm(current, kbasename(bprm->filename), true);
/* Set the new mm task size. We have to do that late because it may
* depend on TIF_32BIT which is only updated in flush_thread() on
diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index ec2e29f..3f736a5 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -707,7 +707,7 @@ extern int perf_register_guest_info_callbacks(struct perf_guest_info_callbacks *
extern int perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks);
extern void perf_event_exec(void);
-extern void perf_event_comm(struct task_struct *tsk);
+extern void perf_event_comm(struct task_struct *tsk, bool exec);
extern void perf_event_fork(struct task_struct *tsk);
/* Callchains */
@@ -814,7 +814,7 @@ static inline int perf_unregister_guest_info_callbacks
(struct perf_guest_info_callbacks *callbacks) { return 0; }
static inline void perf_event_mmap(struct vm_area_struct *vma) { }
-static inline void perf_event_comm(struct task_struct *tsk) { }
+static inline void perf_event_comm(struct task_struct *tsk, bool exec) { }
static inline void perf_event_fork(struct task_struct *tsk) { }
static inline void perf_event_init(void) { }
static inline int perf_swevent_get_recursion_context(void) { return -1; }
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 25f54c7..cadcf38 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -2376,7 +2376,11 @@ extern long do_fork(unsigned long, unsigned long, unsigned long, int __user *, i
struct task_struct *fork_idle(int);
extern pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags);
-extern void set_task_comm(struct task_struct *tsk, const char *from);
+extern void __set_task_comm(struct task_struct *tsk, const char *from, bool exec);
+static inline void set_task_comm(struct task_struct *tsk, const char *from)
+{
+ __set_task_comm(tsk, from, false);
+}
extern char *get_task_comm(char *to, struct task_struct *tsk);
#ifdef CONFIG_SMP
diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
index d9cd853..7e55160 100644
--- a/include/uapi/linux/perf_event.h
+++ b/include/uapi/linux/perf_event.h
@@ -302,8 +302,8 @@ struct perf_event_attr {
exclude_callchain_kernel : 1, /* exclude kernel callchains */
exclude_callchain_user : 1, /* exclude user callchains */
mmap2 : 1, /* include mmap with inode data */
-
- __reserved_1 : 40;
+ exec : 1, /* flag comm events that are due to an exec */
+ __reserved_1 : 39;
union {
__u32 wakeup_events; /* wakeup every n events */
@@ -502,7 +502,12 @@ struct perf_event_mmap_page {
#define PERF_RECORD_MISC_GUEST_KERNEL (4 << 0)
#define PERF_RECORD_MISC_GUEST_USER (5 << 0)
+/*
+ * PERF_RECORD_MISC_MMAP_DATA and PERF_RECORD_MISC_COMM_EXEC are used on
+ * different events so can reuse the same bit position.
+ */
#define PERF_RECORD_MISC_MMAP_DATA (1 << 13)
+#define PERF_RECORD_MISC_COMM_EXEC (1 << 13)
/*
* Indicates that the content of PERF_SAMPLE_IP points to
* the actual instruction that triggered the event. See also
diff --git a/kernel/events/core.c b/kernel/events/core.c
index 6a5e064..9efb1e7 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -5090,7 +5090,7 @@ static void perf_event_comm_event(struct perf_comm_event *comm_event)
NULL);
}
-void perf_event_comm(struct task_struct *task)
+void perf_event_comm(struct task_struct *task, bool exec)
{
struct perf_comm_event comm_event;
@@ -5104,7 +5104,7 @@ void perf_event_comm(struct task_struct *task)
.event_id = {
.header = {
.type = PERF_RECORD_COMM,
- .misc = 0,
+ .misc = exec ? PERF_RECORD_MISC_COMM_EXEC : 0,
/* .size */
},
/* .pid */
--
1.8.3.2
^ permalink raw reply related [flat|nested] 6+ messages in thread
* Re: [PATCH] perf: Differentiate exec() and non-exec() comm events
2014-05-28 8:45 [PATCH] perf: Differentiate exec() and non-exec() comm events Adrian Hunter
@ 2014-05-28 8:55 ` Peter Zijlstra
2014-05-28 9:08 ` Adrian Hunter
0 siblings, 1 reply; 6+ messages in thread
From: Peter Zijlstra @ 2014-05-28 8:55 UTC (permalink / raw)
To: Adrian Hunter
Cc: Ingo Molnar, Dave Jones, Arnaldo Carvalho de Melo, linux-kernel,
David Ahern, Jiri Olsa, Paul Mackerras
[-- Attachment #1: Type: text/plain, Size: 1567 bytes --]
On Wed, May 28, 2014 at 11:45:04AM +0300, Adrian Hunter wrote:
> perf tools like 'perf report' can aggregate samples by comm
> strings, which generally works. However, there are other
> potential use-cases. For example, to pair up 'calls'
> with 'returns' accurately (from branch events like Intel BTS)
> it is necessary to identify whether the process has exec'd.
> Although a comm event is generated when an 'exec' happens
> it is also generated whenever the comm string is changed
> on a whim (e.g. by prctl PR_SET_NAME). This patch adds a
> flag to the comm event to differentiate one case from the
> other.
>
> In order to determine whether the kernel supports the new
> flag, a selection bit named 'exec' is added to struct
> perf_event_attr. The bit does nothing but will cause
> perf_event_open() to fail if the bit is set on kernels
> that do not have it defined.
>
> --- a/include/uapi/linux/perf_event.h
> +++ b/include/uapi/linux/perf_event.h
> @@ -302,8 +302,8 @@ struct perf_event_attr {
> exclude_callchain_kernel : 1, /* exclude kernel callchains */
> exclude_callchain_user : 1, /* exclude user callchains */
> mmap2 : 1, /* include mmap with inode data */
> -
> - __reserved_1 : 40;
> + exec : 1, /* flag comm events that are due to an exec */
> + __reserved_1 : 39;
>
Yah.. that's just sad :-(
the only capabilities mask we have is in the mmap() page, so without
mmap()ing we have no way to test that.
Would it make sense to call it comm_exec?
[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] perf: Differentiate exec() and non-exec() comm events
2014-05-28 8:55 ` Peter Zijlstra
@ 2014-05-28 9:08 ` Adrian Hunter
2014-05-28 9:53 ` Peter Zijlstra
0 siblings, 1 reply; 6+ messages in thread
From: Adrian Hunter @ 2014-05-28 9:08 UTC (permalink / raw)
To: Peter Zijlstra
Cc: Ingo Molnar, Dave Jones, Arnaldo Carvalho de Melo, linux-kernel,
David Ahern, Jiri Olsa, Paul Mackerras
On 05/28/2014 11:55 AM, Peter Zijlstra wrote:
> On Wed, May 28, 2014 at 11:45:04AM +0300, Adrian Hunter wrote:
>> perf tools like 'perf report' can aggregate samples by comm
>> strings, which generally works. However, there are other
>> potential use-cases. For example, to pair up 'calls'
>> with 'returns' accurately (from branch events like Intel BTS)
>> it is necessary to identify whether the process has exec'd.
>> Although a comm event is generated when an 'exec' happens
>> it is also generated whenever the comm string is changed
>> on a whim (e.g. by prctl PR_SET_NAME). This patch adds a
>> flag to the comm event to differentiate one case from the
>> other.
>>
>> In order to determine whether the kernel supports the new
>> flag, a selection bit named 'exec' is added to struct
>> perf_event_attr. The bit does nothing but will cause
>> perf_event_open() to fail if the bit is set on kernels
>> that do not have it defined.
>>
>
>> --- a/include/uapi/linux/perf_event.h
>> +++ b/include/uapi/linux/perf_event.h
>> @@ -302,8 +302,8 @@ struct perf_event_attr {
>> exclude_callchain_kernel : 1, /* exclude kernel callchains */
>> exclude_callchain_user : 1, /* exclude user callchains */
>> mmap2 : 1, /* include mmap with inode data */
>> -
>> - __reserved_1 : 40;
>> + exec : 1, /* flag comm events that are due to an exec */
>> + __reserved_1 : 39;
>>
>
> Yah.. that's just sad :-(
>
> the only capabilities mask we have is in the mmap() page, so without
> mmap()ing we have no way to test that.
>
> Would it make sense to call it comm_exec?
Yes, that is better. Do you want me to resend the patch?
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] perf: Differentiate exec() and non-exec() comm events
2014-05-28 9:08 ` Adrian Hunter
@ 2014-05-28 9:53 ` Peter Zijlstra
2014-06-05 9:56 ` Ingo Molnar
0 siblings, 1 reply; 6+ messages in thread
From: Peter Zijlstra @ 2014-05-28 9:53 UTC (permalink / raw)
To: Adrian Hunter
Cc: Ingo Molnar, Dave Jones, Arnaldo Carvalho de Melo, linux-kernel,
David Ahern, Jiri Olsa, Paul Mackerras
[-- Attachment #1: Type: text/plain, Size: 1893 bytes --]
On Wed, May 28, 2014 at 12:08:57PM +0300, Adrian Hunter wrote:
> On 05/28/2014 11:55 AM, Peter Zijlstra wrote:
> > On Wed, May 28, 2014 at 11:45:04AM +0300, Adrian Hunter wrote:
> >> perf tools like 'perf report' can aggregate samples by comm
> >> strings, which generally works. However, there are other
> >> potential use-cases. For example, to pair up 'calls'
> >> with 'returns' accurately (from branch events like Intel BTS)
> >> it is necessary to identify whether the process has exec'd.
> >> Although a comm event is generated when an 'exec' happens
> >> it is also generated whenever the comm string is changed
> >> on a whim (e.g. by prctl PR_SET_NAME). This patch adds a
> >> flag to the comm event to differentiate one case from the
> >> other.
> >>
> >> In order to determine whether the kernel supports the new
> >> flag, a selection bit named 'exec' is added to struct
> >> perf_event_attr. The bit does nothing but will cause
> >> perf_event_open() to fail if the bit is set on kernels
> >> that do not have it defined.
> >>
> >
> >> --- a/include/uapi/linux/perf_event.h
> >> +++ b/include/uapi/linux/perf_event.h
> >> @@ -302,8 +302,8 @@ struct perf_event_attr {
> >> exclude_callchain_kernel : 1, /* exclude kernel callchains */
> >> exclude_callchain_user : 1, /* exclude user callchains */
> >> mmap2 : 1, /* include mmap with inode data */
> >> -
> >> - __reserved_1 : 40;
> >> + exec : 1, /* flag comm events that are due to an exec */
> >> + __reserved_1 : 39;
> >>
> >
> > Yah.. that's just sad :-(
> >
> > the only capabilities mask we have is in the mmap() page, so without
> > mmap()ing we have no way to test that.
> >
> > Would it make sense to call it comm_exec?
>
> Yes, that is better. Do you want me to resend the patch?
Nah, I'll frob it. Thanks!
[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] perf: Differentiate exec() and non-exec() comm events
2014-05-28 9:53 ` Peter Zijlstra
@ 2014-06-05 9:56 ` Ingo Molnar
2014-06-05 9:58 ` Ingo Molnar
0 siblings, 1 reply; 6+ messages in thread
From: Ingo Molnar @ 2014-06-05 9:56 UTC (permalink / raw)
To: Peter Zijlstra
Cc: Adrian Hunter, Ingo Molnar, Dave Jones, Arnaldo Carvalho de Melo,
linux-kernel, David Ahern, Jiri Olsa, Paul Mackerras
* Peter Zijlstra <peterz@infradead.org> wrote:
> On Wed, May 28, 2014 at 12:08:57PM +0300, Adrian Hunter wrote:
> > On 05/28/2014 11:55 AM, Peter Zijlstra wrote:
> > > On Wed, May 28, 2014 at 11:45:04AM +0300, Adrian Hunter wrote:
> > >> perf tools like 'perf report' can aggregate samples by comm
> > >> strings, which generally works. However, there are other
> > >> potential use-cases. For example, to pair up 'calls'
> > >> with 'returns' accurately (from branch events like Intel BTS)
> > >> it is necessary to identify whether the process has exec'd.
> > >> Although a comm event is generated when an 'exec' happens
> > >> it is also generated whenever the comm string is changed
> > >> on a whim (e.g. by prctl PR_SET_NAME). This patch adds a
> > >> flag to the comm event to differentiate one case from the
> > >> other.
> > >>
> > >> In order to determine whether the kernel supports the new
> > >> flag, a selection bit named 'exec' is added to struct
> > >> perf_event_attr. The bit does nothing but will cause
> > >> perf_event_open() to fail if the bit is set on kernels
> > >> that do not have it defined.
> > >>
> > >
> > >> --- a/include/uapi/linux/perf_event.h
> > >> +++ b/include/uapi/linux/perf_event.h
> > >> @@ -302,8 +302,8 @@ struct perf_event_attr {
> > >> exclude_callchain_kernel : 1, /* exclude kernel callchains */
> > >> exclude_callchain_user : 1, /* exclude user callchains */
> > >> mmap2 : 1, /* include mmap with inode data */
> > >> -
> > >> - __reserved_1 : 40;
> > >> + exec : 1, /* flag comm events that are due to an exec */
> > >> + __reserved_1 : 39;
> > >>
> > >
> > > Yah.. that's just sad :-(
> > >
> > > the only capabilities mask we have is in the mmap() page, so without
> > > mmap()ing we have no way to test that.
> > >
> > > Would it make sense to call it comm_exec?
> >
> > Yes, that is better. Do you want me to resend the patch?
>
> Nah, I'll frob it. Thanks!
FYI, this patch breaks pretty much every non-x86 architecture:
/home/mingo/tip/fs/exec.c: In function 'setup_new_exec':
/home/mingo/tip/fs/exec.c:1113: error: implicit declaration of function 'perf_event_exec'
make[2]: *** [fs/exec.o] Error 1
make[2]: *** Waiting for unfinished jobs....
make[1]: *** [fs] Error 2
make[1]: *** Waiting for unfinished jobs....
Thanks,
Ingo
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] perf: Differentiate exec() and non-exec() comm events
2014-06-05 9:56 ` Ingo Molnar
@ 2014-06-05 9:58 ` Ingo Molnar
0 siblings, 0 replies; 6+ messages in thread
From: Ingo Molnar @ 2014-06-05 9:58 UTC (permalink / raw)
To: Peter Zijlstra
Cc: Adrian Hunter, Ingo Molnar, Dave Jones, Arnaldo Carvalho de Melo,
linux-kernel, David Ahern, Jiri Olsa, Paul Mackerras
* Ingo Molnar <mingo@kernel.org> wrote:
> > Nah, I'll frob it. Thanks!
>
> FYI, this patch breaks pretty much every non-x86 architecture:
>
> /home/mingo/tip/fs/exec.c: In function 'setup_new_exec':
> /home/mingo/tip/fs/exec.c:1113: error: implicit declaration of function 'perf_event_exec'
> make[2]: *** [fs/exec.o] Error 1
> make[2]: *** Waiting for unfinished jobs....
> make[1]: *** [fs] Error 2
> make[1]: *** Waiting for unfinished jobs....
Sorry, it was another patch that broke things:
Author: Peter Zijlstra <peterz@infradead.org>
Date: Wed May 21 17:32:19 2014 +0200
perf: Fix perf_event_comm() vs. exec() assumption
Thanks,
Ingo
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2014-06-05 9:58 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2014-05-28 8:45 [PATCH] perf: Differentiate exec() and non-exec() comm events Adrian Hunter
2014-05-28 8:55 ` Peter Zijlstra
2014-05-28 9:08 ` Adrian Hunter
2014-05-28 9:53 ` Peter Zijlstra
2014-06-05 9:56 ` Ingo Molnar
2014-06-05 9:58 ` Ingo Molnar
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).