perf: Differentiate exec() and non-exec() comm events
Subsystems (4)
Adrian Hunter
reviewerPATCH 1/1 perf: Differentiate exec() and non-exec() comm events
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.
--- 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(-)
fs/exec.c +3 -3
@@ -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
include/linux/perf_event.h +2 -2
@@ -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 ; }
include/linux/sched.h +5 -1
@@ -2376,7 +2376,11 @@ extern long do_fork(unsigned long, unsigned long, unsigned long, int __user *, istruct 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
include/uapi/linux/perf_event.h +7 -2
@@ -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
kernel/events/core.c +2 -2
@@ -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 */
Signature · 1 line
1.8.3.2
Peter Zijlstra
maintainerQuoted 21 lines · Adrian Hunter
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 causeperf_event_open()to fail if the bit is set on kernels that do not have it defined.
include/uapi/linux/perf_event.h struct perf_event_attr { · line 307
Quoted 8 lines
include/uapi/linux/perf_event.h +2
-2@@ -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?
Adrian Hunter
reviewerQuoted 33 lines · Peter Zijlstra
Quoted 21 lines · Adrian Hunter
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 causeperf_event_open()to fail if the bit is set on kernels that do not have it defined.include/uapi/linux/perf_event.h struct perf_event_attr { · line 307
Quoted 8 lines
include/uapi/linux/perf_event.h +2
-2@@ -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 :-(
Quoted 4 lines
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?
Peter Zijlstra
maintainerQuoted 34 lines · Adrian Hunter
Quoted 33 lines · Peter Zijlstra
Quoted 21 lines · Adrian Hunter
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 causeperf_event_open()to fail if the bit is set on kernels that do not have it defined.include/uapi/linux/perf_event.h struct perf_event_attr { · line 307
Quoted 8 lines
include/uapi/linux/perf_event.h +2
-2@@ -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 :-(
Quoted 6 lines
Quoted 4 lines
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!
Quoted 35 lines · Peter Zijlstra
Quoted 34 lines · Adrian Hunter
Quoted 33 lines · Peter Zijlstra
Quoted 21 lines · Adrian Hunter
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 causeperf_event_open()to fail if the bit is set on kernels that do not have it defined.include/uapi/linux/perf_event.h struct perf_event_attr { · line 307
Quoted 8 lines
include/uapi/linux/perf_event.h +2
-2@@ -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 :-(
Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
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
Quoted 10 lines · Ingo Molnar
Quoted 1 line
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 <[email protected]> Date: Wed May 21 17:32:19 2014 +0200
perf: Fix perf_event_comm() vs. exec() assumption
Thanks,
Ingo