[PATCH] oom: add tracepoints for oom_score_adj

[PATCH] oom: add tracepoints for oom_score_adj

[KAMEZAWA Hiroyuki]

[PATCH] oom: add tracepoints for oom_score_adj

[KAMEZAWA Hiroyuki]

>From 28189e4622fd97324893a0b234183f64472a54d6 Mon Sep 17 00:00:00 2001
From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Date: Wed, 7 Dec 2011 09:58:16 +0900
Subject: [PATCH] oom: trace point for oom_score_adj

oom_score_adj is set to prevent a task from being killed by OOM-Killer.
Some daemons sets this value and their children inerit it sometimes.
Because inheritance of oom_score_adj is done automatically, users
can be confused at seeing the value and finds it's hard to debug.

This patch adds trace point for oom_score_adj. This adds 3 trace
points. at
	- update oom_score_adj
	- fork()
	- rename task->comm(typically, exec())

Outputs will be following.
   bash-2404  [006]   199.620841: oom_score_adj_update: task 2404[bash] updates oom_score_ad  j=-1000
   bash-2404  [006]   205.861287: oom_score_adj_inherited: new_task=2442 oom_score_adj=-1000
   su-2442  [003]   205.861761: oom_score_task_rename: rename task 2442[bash] to [su] oom_  score_adj=-1000
   su-2442  [003]   205.866737: oom_score_adj_inherited: new_task=2444 oom_score_adj=-1000
   bash-2444  [007]   205.868136: oom_score_task_rename: rename task 2444[su] to [bash] oom_  score_adj=-1000
   bash-2444  [007]   205.870407: oom_score_adj_inherited: new_task=2445 oom_score_adj=-1000
   bash-2445  [001]   205.870975: oom_score_adj_inherited: new_task=2446 oom_score_adj=-1000

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
---
 fs/exec.c                  |    5 +++
 fs/proc/base.c             |    3 ++
 include/trace/events/oom.h |   80 ++++++++++++++++++++++++++++++++++++++++++++
 kernel/fork.c              |    5 +++
 mm/oom_kill.c              |    6 +++
 5 files changed, 99 insertions(+), 0 deletions(-)
 create mode 100644 include/trace/events/oom.h

diff --git a/fs/exec.c b/fs/exec.c
index ca141db..562a106 100644
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -59,6 +59,8 @@
 #include <asm/uaccess.h>
 #include <asm/mmu_context.h>
 #include <asm/tlb.h>
+
+#include <trace/events/oom.h>
 #include "internal.h"
 
 int core_uses_pid;
@@ -1054,6 +1056,9 @@ void set_task_comm(struct task_struct *tsk, char *buf)
 {
 	task_lock(tsk);
 
+	if (tsk->signal->oom_score_adj)
+		trace_oom_score_task_rename(tsk, buf);
+
 	/*
 	 * Threads may access current->comm without holding
 	 * the task lock, so write the string carefully.
diff --git a/fs/proc/base.c b/fs/proc/base.c
index 1050b1c..f201e64 100644
--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -87,6 +87,7 @@
 #ifdef CONFIG_HARDWALL
 #include <asm/hardwall.h>
 #endif
+#include <trace/events/oom.h>
 #include "internal.h"
 
 /* NOTE:
@@ -1166,6 +1167,7 @@ static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 	else
 		task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
 								-OOM_DISABLE;
+	trace_oom_score_adj_update(task);
 err_sighand:
 	unlock_task_sighand(task, &flags);
 err_task_lock:
@@ -1253,6 +1255,7 @@ static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
 	task->signal->oom_score_adj = oom_score_adj;
 	if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
 		task->signal->oom_score_adj_min = oom_score_adj;
+	trace_oom_score_adj_update(task);
 	/*
 	 * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
 	 * always attainable.
diff --git a/include/trace/events/oom.h b/include/trace/events/oom.h
new file mode 100644
index 0000000..f5e6f55
--- /dev/null
+++ b/include/trace/events/oom.h
@@ -0,0 +1,80 @@
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM oom
+
+#if !defined(_TRACE_OOM_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_OOM_H
+#include <linux/tracepoint.h>
+
+TRACE_EVENT(oom_score_adj_inherited,
+
+	TP_PROTO(struct task_struct *task),
+	
+	TP_ARGS(task),
+
+	TP_STRUCT__entry(
+		__field(	pid_t,		newpid)
+		__field(	  int,		oom_score_adj)
+	),
+
+	TP_fast_assign(
+		__entry->newpid = task->pid;
+		__entry->oom_score_adj = task->signal->oom_score_adj;
+	),
+
+	TP_printk("new_task=%ld oom_score_adj=%d",
+		__entry->newpid, __entry->oom_score_adj)
+);
+
+TRACE_EVENT(oom_score_task_rename,
+
+	TP_PROTO(struct task_struct *task, char *comm),
+
+	TP_ARGS(task, comm),
+
+	TP_STRUCT__entry(
+		__field(	pid_t,	 pid)
+		__array(         char,   oldcomm,   TASK_COMM_LEN   )
+		__array(         char,   newcomm,   TASK_COMM_LEN   )
+		__field(	  int,   oom_score_adj)
+	),
+
+	TP_fast_assign(
+		__entry->pid = task->pid;
+		 memcpy(__entry->oldcomm, task->comm, TASK_COMM_LEN);
+		 memcpy(__entry->newcomm, comm, TASK_COMM_LEN);
+		__entry->oom_score_adj = task->signal->oom_score_adj;
+	),
+
+	TP_printk("rename task %ld[%s] to [%s] oom_score_adj=%d",
+		__entry->pid, __entry->oldcomm, __entry->newcomm,
+		__entry->oom_score_adj)
+);
+
+TRACE_EVENT(oom_score_adj_update,
+
+	TP_PROTO(struct task_struct *task),
+
+	TP_ARGS(task),
+
+	TP_STRUCT__entry(
+		__field(	pid_t,	pid)
+		__array(	char,	comm,	TASK_COMM_LEN )
+		__field(	 int,	oom_score_adj)
+	),
+
+	TP_fast_assign(
+		__entry->pid = task->pid;
+		memcpy(__entry->comm, task->comm, TASK_COMM_LEN);
+		__entry->oom_score_adj = task->signal->oom_score_adj;
+	),
+
+	TP_printk("task %ld[%s] updates oom_score_adj=%d",
+		__entry->pid, __entry->comm, __entry->oom_score_adj)
+);
+
+#endif
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>
+
+
diff --git a/kernel/fork.c b/kernel/fork.c
index e20518d..634aa84 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -76,6 +76,7 @@
 #include <asm/tlbflush.h>
 
 #include <trace/events/sched.h>
+#include <trace/events/oom.h>
 
 /*
  * Protected counters by write_lock_irq(&tasklist_lock)
@@ -1390,6 +1391,10 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 	if (clone_flags & CLONE_THREAD)
 		threadgroup_fork_read_unlock(current);
 	perf_event_fork(p);
+
+	if (!(clone_flags & CLONE_THREAD) && p->signal->oom_score_adj)
+		trace_oom_score_adj_inherited(p);
+
 	return p;
 
 bad_fork_free_pid:
diff --git a/mm/oom_kill.c b/mm/oom_kill.c
index e2e1402..46b6d0a 100644
--- a/mm/oom_kill.c
+++ b/mm/oom_kill.c
@@ -33,6 +33,10 @@
 #include <linux/security.h>
 #include <linux/ptrace.h>
 #include <linux/freezer.h>
+#include <linux/ftrace.h>
+
+#define CREATE_TRACE_POINTS
+#include <trace/events/oom.h>
 
 int sysctl_panic_on_oom;
 int sysctl_oom_kill_allocating_task;
@@ -55,6 +59,7 @@ void compare_swap_oom_score_adj(int old_val, int new_val)
 	spin_lock_irq(&sighand->siglock);
 	if (current->signal->oom_score_adj == old_val)
 		current->signal->oom_score_adj = new_val;
+	trace_oom_score_adj_update(current);
 	spin_unlock_irq(&sighand->siglock);
 }
 
@@ -74,6 +79,7 @@ int test_set_oom_score_adj(int new_val)
 	spin_lock_irq(&sighand->siglock);
 	old_val = current->signal->oom_score_adj;
 	current->signal->oom_score_adj = new_val;
+	trace_oom_score_adj_update(current);
 	spin_unlock_irq(&sighand->siglock);
 
 	return old_val;
-- 
1.7.4.1

Re: [PATCH] oom: add tracepoints for oom_score_adj

[David Rientjes]

On Wed, 7 Dec 2011, KAMEZAWA Hiroyuki wrote:

> From 28189e4622fd97324893a0b234183f64472a54d6 Mon Sep 17 00:00:00 2001
> From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> Date: Wed, 7 Dec 2011 09:58:16 +0900
> Subject: [PATCH] oom: trace point for oom_score_adj
> 
> oom_score_adj is set to prevent a task from being killed by OOM-Killer.
> Some daemons sets this value and their children inerit it sometimes.
> Because inheritance of oom_score_adj is done automatically, users
> can be confused at seeing the value and finds it's hard to debug.
> 
> This patch adds trace point for oom_score_adj. This adds 3 trace
> points. at
> 	- update oom_score_adj
> 	- fork()
> 	- rename task->comm(typically, exec())
> 
> Outputs will be following.
>    bash-2404  [006]   199.620841: oom_score_adj_update: task 2404[bash] updates oom_score_ad  j=-1000
>    bash-2404  [006]   205.861287: oom_score_adj_inherited: new_task=2442 oom_score_adj=-1000
>    su-2442  [003]   205.861761: oom_score_task_rename: rename task 2442[bash] to [su] oom_  score_adj=-1000
>    su-2442  [003]   205.866737: oom_score_adj_inherited: new_task=2444 oom_score_adj=-1000
>    bash-2444  [007]   205.868136: oom_score_task_rename: rename task 2444[su] to [bash] oom_  score_adj=-1000
>    bash-2444  [007]   205.870407: oom_score_adj_inherited: new_task=2445 oom_score_adj=-1000
>    bash-2445  [001]   205.870975: oom_score_adj_inherited: new_task=2446 oom_score_adj=-1000
> 

Little bit of whitespace damage there, but looks good in the code itself.

> Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>

Just minor alterations to the format of the tracepoints from the first 
version, so carry over my

Acked-by: David Rientjes <rientjes@google.com>

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] oom: add tracepoints for oom_score_adj

[David Rientjes]

On Wed, 7 Dec 2011, KAMEZAWA Hiroyuki wrote:

> From 28189e4622fd97324893a0b234183f64472a54d6 Mon Sep 17 00:00:00 2001
> From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> Date: Wed, 7 Dec 2011 09:58:16 +0900
> Subject: [PATCH] oom: trace point for oom_score_adj
> 
> oom_score_adj is set to prevent a task from being killed by OOM-Killer.
> Some daemons sets this value and their children inerit it sometimes.
> Because inheritance of oom_score_adj is done automatically, users
> can be confused at seeing the value and finds it's hard to debug.
> 
> This patch adds trace point for oom_score_adj. This adds 3 trace
> points. at
> 	- update oom_score_adj
> 	- fork()
> 	- rename task->comm(typically, exec())
> 
> Outputs will be following.
>    bash-2404  [006]   199.620841: oom_score_adj_update: task 2404[bash] updates oom_score_ad  j=-1000
>    bash-2404  [006]   205.861287: oom_score_adj_inherited: new_task=2442 oom_score_adj=-1000
>    su-2442  [003]   205.861761: oom_score_task_rename: rename task 2442[bash] to [su] oom_  score_adj=-1000
>    su-2442  [003]   205.866737: oom_score_adj_inherited: new_task=2444 oom_score_adj=-1000
>    bash-2444  [007]   205.868136: oom_score_task_rename: rename task 2444[su] to [bash] oom_  score_adj=-1000
>    bash-2444  [007]   205.870407: oom_score_adj_inherited: new_task=2445 oom_score_adj=-1000
>    bash-2445  [001]   205.870975: oom_score_adj_inherited: new_task=2446 oom_score_adj=-1000
> 

Little bit of whitespace damage there, but looks good in the code itself.

> Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>

Just minor alterations to the format of the tracepoints from the first 
version, so carry over my

Acked-by: David Rientjes <rientjes@google.com>

Re: [PATCH] oom: add tracepoints for oom_score_adj

[KAMEZAWA Hiroyuki]

Sorry, I found a mistake...
I'll post updated one, again.

Thanks,
-Kame


On Wed, 7 Dec 2011 09:54:34 +0900
KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote:

> From 28189e4622fd97324893a0b234183f64472a54d6 Mon Sep 17 00:00:00 2001
> From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> Date: Wed, 7 Dec 2011 09:58:16 +0900
> Subject: [PATCH] oom: trace point for oom_score_adj
> 
> oom_score_adj is set to prevent a task from being killed by OOM-Killer.
> Some daemons sets this value and their children inerit it sometimes.
> Because inheritance of oom_score_adj is done automatically, users
> can be confused at seeing the value and finds it's hard to debug.
> 
> This patch adds trace point for oom_score_adj. This adds 3 trace
> points. at
> 	- update oom_score_adj
> 	- fork()
> 	- rename task->comm(typically, exec())
> 
> Outputs will be following.
>    bash-2404  [006]   199.620841: oom_score_adj_update: task 2404[bash] updates oom_score_ad  j=-1000
>    bash-2404  [006]   205.861287: oom_score_adj_inherited: new_task=2442 oom_score_adj=-1000
>    su-2442  [003]   205.861761: oom_score_task_rename: rename task 2442[bash] to [su] oom_  score_adj=-1000
>    su-2442  [003]   205.866737: oom_score_adj_inherited: new_task=2444 oom_score_adj=-1000
>    bash-2444  [007]   205.868136: oom_score_task_rename: rename task 2444[su] to [bash] oom_  score_adj=-1000
>    bash-2444  [007]   205.870407: oom_score_adj_inherited: new_task=2445 oom_score_adj=-1000
>    bash-2445  [001]   205.870975: oom_score_adj_inherited: new_task=2446 oom_score_adj=-1000
> 
> Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> ---
>  fs/exec.c                  |    5 +++
>  fs/proc/base.c             |    3 ++
>  include/trace/events/oom.h |   80 ++++++++++++++++++++++++++++++++++++++++++++
>  kernel/fork.c              |    5 +++
>  mm/oom_kill.c              |    6 +++
>  5 files changed, 99 insertions(+), 0 deletions(-)
>  create mode 100644 include/trace/events/oom.h
> 
> diff --git a/fs/exec.c b/fs/exec.c
> index ca141db..562a106 100644
> --- a/fs/exec.c
> +++ b/fs/exec.c
> @@ -59,6 +59,8 @@
>  #include <asm/uaccess.h>
>  #include <asm/mmu_context.h>
>  #include <asm/tlb.h>
> +
> +#include <trace/events/oom.h>
>  #include "internal.h"
>  
>  int core_uses_pid;
> @@ -1054,6 +1056,9 @@ void set_task_comm(struct task_struct *tsk, char *buf)
>  {
>  	task_lock(tsk);
>  
> +	if (tsk->signal->oom_score_adj)
> +		trace_oom_score_task_rename(tsk, buf);
> +
>  	/*
>  	 * Threads may access current->comm without holding
>  	 * the task lock, so write the string carefully.
> diff --git a/fs/proc/base.c b/fs/proc/base.c
> index 1050b1c..f201e64 100644
> --- a/fs/proc/base.c
> +++ b/fs/proc/base.c
> @@ -87,6 +87,7 @@
>  #ifdef CONFIG_HARDWALL
>  #include <asm/hardwall.h>
>  #endif
> +#include <trace/events/oom.h>
>  #include "internal.h"
>  
>  /* NOTE:
> @@ -1166,6 +1167,7 @@ static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
>  	else
>  		task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
>  								-OOM_DISABLE;
> +	trace_oom_score_adj_update(task);
>  err_sighand:
>  	unlock_task_sighand(task, &flags);
>  err_task_lock:
> @@ -1253,6 +1255,7 @@ static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
>  	task->signal->oom_score_adj = oom_score_adj;
>  	if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
>  		task->signal->oom_score_adj_min = oom_score_adj;
> +	trace_oom_score_adj_update(task);
>  	/*
>  	 * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
>  	 * always attainable.
> diff --git a/include/trace/events/oom.h b/include/trace/events/oom.h
> new file mode 100644
> index 0000000..f5e6f55
> --- /dev/null
> +++ b/include/trace/events/oom.h
> @@ -0,0 +1,80 @@
> +#undef TRACE_SYSTEM
> +#define TRACE_SYSTEM oom
> +
> +#if !defined(_TRACE_OOM_H) || defined(TRACE_HEADER_MULTI_READ)
> +#define _TRACE_OOM_H
> +#include <linux/tracepoint.h>
> +
> +TRACE_EVENT(oom_score_adj_inherited,
> +
> +	TP_PROTO(struct task_struct *task),
> +	
> +	TP_ARGS(task),
> +
> +	TP_STRUCT__entry(
> +		__field(	pid_t,		newpid)
> +		__field(	  int,		oom_score_adj)
> +	),
> +
> +	TP_fast_assign(
> +		__entry->newpid = task->pid;
> +		__entry->oom_score_adj = task->signal->oom_score_adj;
> +	),
> +
> +	TP_printk("new_task=%ld oom_score_adj=%d",
> +		__entry->newpid, __entry->oom_score_adj)
> +);
> +
> +TRACE_EVENT(oom_score_task_rename,
> +
> +	TP_PROTO(struct task_struct *task, char *comm),
> +
> +	TP_ARGS(task, comm),
> +
> +	TP_STRUCT__entry(
> +		__field(	pid_t,	 pid)
> +		__array(         char,   oldcomm,   TASK_COMM_LEN   )
> +		__array(         char,   newcomm,   TASK_COMM_LEN   )
> +		__field(	  int,   oom_score_adj)
> +	),
> +
> +	TP_fast_assign(
> +		__entry->pid = task->pid;
> +		 memcpy(__entry->oldcomm, task->comm, TASK_COMM_LEN);
> +		 memcpy(__entry->newcomm, comm, TASK_COMM_LEN);
> +		__entry->oom_score_adj = task->signal->oom_score_adj;
> +	),
> +
> +	TP_printk("rename task %ld[%s] to [%s] oom_score_adj=%d",
> +		__entry->pid, __entry->oldcomm, __entry->newcomm,
> +		__entry->oom_score_adj)
> +);
> +
> +TRACE_EVENT(oom_score_adj_update,
> +
> +	TP_PROTO(struct task_struct *task),
> +
> +	TP_ARGS(task),
> +
> +	TP_STRUCT__entry(
> +		__field(	pid_t,	pid)
> +		__array(	char,	comm,	TASK_COMM_LEN )
> +		__field(	 int,	oom_score_adj)
> +	),
> +
> +	TP_fast_assign(
> +		__entry->pid = task->pid;
> +		memcpy(__entry->comm, task->comm, TASK_COMM_LEN);
> +		__entry->oom_score_adj = task->signal->oom_score_adj;
> +	),
> +
> +	TP_printk("task %ld[%s] updates oom_score_adj=%d",
> +		__entry->pid, __entry->comm, __entry->oom_score_adj)
> +);
> +
> +#endif
> +
> +/* This part must be outside protection */
> +#include <trace/define_trace.h>
> +
> +
> diff --git a/kernel/fork.c b/kernel/fork.c
> index e20518d..634aa84 100644
> --- a/kernel/fork.c
> +++ b/kernel/fork.c
> @@ -76,6 +76,7 @@
>  #include <asm/tlbflush.h>
>  
>  #include <trace/events/sched.h>
> +#include <trace/events/oom.h>
>  
>  /*
>   * Protected counters by write_lock_irq(&tasklist_lock)
> @@ -1390,6 +1391,10 @@ static struct task_struct *copy_process(unsigned long clone_flags,
>  	if (clone_flags & CLONE_THREAD)
>  		threadgroup_fork_read_unlock(current);
>  	perf_event_fork(p);
> +
> +	if (!(clone_flags & CLONE_THREAD) && p->signal->oom_score_adj)
> +		trace_oom_score_adj_inherited(p);
> +
>  	return p;
>  
>  bad_fork_free_pid:
> diff --git a/mm/oom_kill.c b/mm/oom_kill.c
> index e2e1402..46b6d0a 100644
> --- a/mm/oom_kill.c
> +++ b/mm/oom_kill.c
> @@ -33,6 +33,10 @@
>  #include <linux/security.h>
>  #include <linux/ptrace.h>
>  #include <linux/freezer.h>
> +#include <linux/ftrace.h>
> +
> +#define CREATE_TRACE_POINTS
> +#include <trace/events/oom.h>
>  
>  int sysctl_panic_on_oom;
>  int sysctl_oom_kill_allocating_task;
> @@ -55,6 +59,7 @@ void compare_swap_oom_score_adj(int old_val, int new_val)
>  	spin_lock_irq(&sighand->siglock);
>  	if (current->signal->oom_score_adj == old_val)
>  		current->signal->oom_score_adj = new_val;
> +	trace_oom_score_adj_update(current);
>  	spin_unlock_irq(&sighand->siglock);
>  }
>  
> @@ -74,6 +79,7 @@ int test_set_oom_score_adj(int new_val)
>  	spin_lock_irq(&sighand->siglock);
>  	old_val = current->signal->oom_score_adj;
>  	current->signal->oom_score_adj = new_val;
> +	trace_oom_score_adj_update(current);
>  	spin_unlock_irq(&sighand->siglock);
>  
>  	return old_val;
> -- 
> 1.7.4.1
> 
> 
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to majordomo@kvack.org.  For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
> Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
> 

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] oom: add tracepoints for oom_score_adj

[KAMEZAWA Hiroyuki]

Sorry, I found a mistake...
I'll post updated one, again.

Thanks,
-Kame


On Wed, 7 Dec 2011 09:54:34 +0900
KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote:

> From 28189e4622fd97324893a0b234183f64472a54d6 Mon Sep 17 00:00:00 2001
> From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> Date: Wed, 7 Dec 2011 09:58:16 +0900
> Subject: [PATCH] oom: trace point for oom_score_adj
> 
> oom_score_adj is set to prevent a task from being killed by OOM-Killer.
> Some daemons sets this value and their children inerit it sometimes.
> Because inheritance of oom_score_adj is done automatically, users
> can be confused at seeing the value and finds it's hard to debug.
> 
> This patch adds trace point for oom_score_adj. This adds 3 trace
> points. at
> 	- update oom_score_adj
> 	- fork()
> 	- rename task->comm(typically, exec())
> 
> Outputs will be following.
>    bash-2404  [006]   199.620841: oom_score_adj_update: task 2404[bash] updates oom_score_ad  j=-1000
>    bash-2404  [006]   205.861287: oom_score_adj_inherited: new_task=2442 oom_score_adj=-1000
>    su-2442  [003]   205.861761: oom_score_task_rename: rename task 2442[bash] to [su] oom_  score_adj=-1000
>    su-2442  [003]   205.866737: oom_score_adj_inherited: new_task=2444 oom_score_adj=-1000
>    bash-2444  [007]   205.868136: oom_score_task_rename: rename task 2444[su] to [bash] oom_  score_adj=-1000
>    bash-2444  [007]   205.870407: oom_score_adj_inherited: new_task=2445 oom_score_adj=-1000
>    bash-2445  [001]   205.870975: oom_score_adj_inherited: new_task=2446 oom_score_adj=-1000
> 
> Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> ---
>  fs/exec.c                  |    5 +++
>  fs/proc/base.c             |    3 ++
>  include/trace/events/oom.h |   80 ++++++++++++++++++++++++++++++++++++++++++++
>  kernel/fork.c              |    5 +++
>  mm/oom_kill.c              |    6 +++
>  5 files changed, 99 insertions(+), 0 deletions(-)
>  create mode 100644 include/trace/events/oom.h
> 
> diff --git a/fs/exec.c b/fs/exec.c
> index ca141db..562a106 100644
> --- a/fs/exec.c
> +++ b/fs/exec.c
> @@ -59,6 +59,8 @@
>  #include <asm/uaccess.h>
>  #include <asm/mmu_context.h>
>  #include <asm/tlb.h>
> +
> +#include <trace/events/oom.h>
>  #include "internal.h"
>  
>  int core_uses_pid;
> @@ -1054,6 +1056,9 @@ void set_task_comm(struct task_struct *tsk, char *buf)
>  {
>  	task_lock(tsk);
>  
> +	if (tsk->signal->oom_score_adj)
> +		trace_oom_score_task_rename(tsk, buf);
> +
>  	/*
>  	 * Threads may access current->comm without holding
>  	 * the task lock, so write the string carefully.
> diff --git a/fs/proc/base.c b/fs/proc/base.c
> index 1050b1c..f201e64 100644
> --- a/fs/proc/base.c
> +++ b/fs/proc/base.c
> @@ -87,6 +87,7 @@
>  #ifdef CONFIG_HARDWALL
>  #include <asm/hardwall.h>
>  #endif
> +#include <trace/events/oom.h>
>  #include "internal.h"
>  
>  /* NOTE:
> @@ -1166,6 +1167,7 @@ static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
>  	else
>  		task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
>  								-OOM_DISABLE;
> +	trace_oom_score_adj_update(task);
>  err_sighand:
>  	unlock_task_sighand(task, &flags);
>  err_task_lock:
> @@ -1253,6 +1255,7 @@ static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
>  	task->signal->oom_score_adj = oom_score_adj;
>  	if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
>  		task->signal->oom_score_adj_min = oom_score_adj;
> +	trace_oom_score_adj_update(task);
>  	/*
>  	 * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
>  	 * always attainable.
> diff --git a/include/trace/events/oom.h b/include/trace/events/oom.h
> new file mode 100644
> index 0000000..f5e6f55
> --- /dev/null
> +++ b/include/trace/events/oom.h
> @@ -0,0 +1,80 @@
> +#undef TRACE_SYSTEM
> +#define TRACE_SYSTEM oom
> +
> +#if !defined(_TRACE_OOM_H) || defined(TRACE_HEADER_MULTI_READ)
> +#define _TRACE_OOM_H
> +#include <linux/tracepoint.h>
> +
> +TRACE_EVENT(oom_score_adj_inherited,
> +
> +	TP_PROTO(struct task_struct *task),
> +	
> +	TP_ARGS(task),
> +
> +	TP_STRUCT__entry(
> +		__field(	pid_t,		newpid)
> +		__field(	  int,		oom_score_adj)
> +	),
> +
> +	TP_fast_assign(
> +		__entry->newpid = task->pid;
> +		__entry->oom_score_adj = task->signal->oom_score_adj;
> +	),
> +
> +	TP_printk("new_task=%ld oom_score_adj=%d",
> +		__entry->newpid, __entry->oom_score_adj)
> +);
> +
> +TRACE_EVENT(oom_score_task_rename,
> +
> +	TP_PROTO(struct task_struct *task, char *comm),
> +
> +	TP_ARGS(task, comm),
> +
> +	TP_STRUCT__entry(
> +		__field(	pid_t,	 pid)
> +		__array(         char,   oldcomm,   TASK_COMM_LEN   )
> +		__array(         char,   newcomm,   TASK_COMM_LEN   )
> +		__field(	  int,   oom_score_adj)
> +	),
> +
> +	TP_fast_assign(
> +		__entry->pid = task->pid;
> +		 memcpy(__entry->oldcomm, task->comm, TASK_COMM_LEN);
> +		 memcpy(__entry->newcomm, comm, TASK_COMM_LEN);
> +		__entry->oom_score_adj = task->signal->oom_score_adj;
> +	),
> +
> +	TP_printk("rename task %ld[%s] to [%s] oom_score_adj=%d",
> +		__entry->pid, __entry->oldcomm, __entry->newcomm,
> +		__entry->oom_score_adj)
> +);
> +
> +TRACE_EVENT(oom_score_adj_update,
> +
> +	TP_PROTO(struct task_struct *task),
> +
> +	TP_ARGS(task),
> +
> +	TP_STRUCT__entry(
> +		__field(	pid_t,	pid)
> +		__array(	char,	comm,	TASK_COMM_LEN )
> +		__field(	 int,	oom_score_adj)
> +	),
> +
> +	TP_fast_assign(
> +		__entry->pid = task->pid;
> +		memcpy(__entry->comm, task->comm, TASK_COMM_LEN);
> +		__entry->oom_score_adj = task->signal->oom_score_adj;
> +	),
> +
> +	TP_printk("task %ld[%s] updates oom_score_adj=%d",
> +		__entry->pid, __entry->comm, __entry->oom_score_adj)
> +);
> +
> +#endif
> +
> +/* This part must be outside protection */
> +#include <trace/define_trace.h>
> +
> +
> diff --git a/kernel/fork.c b/kernel/fork.c
> index e20518d..634aa84 100644
> --- a/kernel/fork.c
> +++ b/kernel/fork.c
> @@ -76,6 +76,7 @@
>  #include <asm/tlbflush.h>
>  
>  #include <trace/events/sched.h>
> +#include <trace/events/oom.h>
>  
>  /*
>   * Protected counters by write_lock_irq(&tasklist_lock)
> @@ -1390,6 +1391,10 @@ static struct task_struct *copy_process(unsigned long clone_flags,
>  	if (clone_flags & CLONE_THREAD)
>  		threadgroup_fork_read_unlock(current);
>  	perf_event_fork(p);
> +
> +	if (!(clone_flags & CLONE_THREAD) && p->signal->oom_score_adj)
> +		trace_oom_score_adj_inherited(p);
> +
>  	return p;
>  
>  bad_fork_free_pid:
> diff --git a/mm/oom_kill.c b/mm/oom_kill.c
> index e2e1402..46b6d0a 100644
> --- a/mm/oom_kill.c
> +++ b/mm/oom_kill.c
> @@ -33,6 +33,10 @@
>  #include <linux/security.h>
>  #include <linux/ptrace.h>
>  #include <linux/freezer.h>
> +#include <linux/ftrace.h>
> +
> +#define CREATE_TRACE_POINTS
> +#include <trace/events/oom.h>
>  
>  int sysctl_panic_on_oom;
>  int sysctl_oom_kill_allocating_task;
> @@ -55,6 +59,7 @@ void compare_swap_oom_score_adj(int old_val, int new_val)
>  	spin_lock_irq(&sighand->siglock);
>  	if (current->signal->oom_score_adj == old_val)
>  		current->signal->oom_score_adj = new_val;
> +	trace_oom_score_adj_update(current);
>  	spin_unlock_irq(&sighand->siglock);
>  }
>  
> @@ -74,6 +79,7 @@ int test_set_oom_score_adj(int new_val)
>  	spin_lock_irq(&sighand->siglock);
>  	old_val = current->signal->oom_score_adj;
>  	current->signal->oom_score_adj = new_val;
> +	trace_oom_score_adj_update(current);
>  	spin_unlock_irq(&sighand->siglock);
>  
>  	return old_val;
> -- 
> 1.7.4.1
> 
> 
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to majordomo@kvack.org.  For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
> Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
> 

[PATCH v3] oom: add tracepoints for oom_score_adj

[KAMEZAWA Hiroyuki]

Fixed compile warning and use TRACE_EVENT_CONDITION() rather than open-coded 'if'
==

[PATCH v3] oom: add tracepoints for oom_score_adj

[KAMEZAWA Hiroyuki]

Fixed compile warning and use TRACE_EVENT_CONDITION() rather than open-coded 'if'
==
>From 2609d1c3c3a3bad2aae24bf5dd48e065bb859be4 Mon Sep 17 00:00:00 2001
From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Date: Wed, 7 Dec 2011 09:58:16 +0900
Subject: [PATCH] oom: trace point for oom_score_adj

oom_score_adj is set to prevent a task from being killed by OOM-Killer.
Some daemons sets this value and their children inerit it sometimes.
Because inheritance of oom_score_adj is done automatically, users
can be confused at seeing the value and finds it's hard to debug.

This patch adds trace point for oom_score_adj. This adds 3 trace
points. at
	- update oom_score_adj
	- fork()
	- rename task->comm(typically, exec())

Outputs will be following.
bash-2404  [006]   199.620841: oom_score_adj_update: task 2404[bash] updates oom_score_ad  j=-1000
bash-2404  [006]   205.861287: oom_score_adj_inherited: new_task=2442 oom_score_adj=-1000
su-2442  [003]   205.861761: oom_score_task_rename: rename task 2442[bash] to [su] oom_  score_adj=-1000
su-2442  [003]   205.866737: oom_score_adj_inherited: new_task=2444 oom_score_adj=-1000
bash-2444  [007]   205.868136: oom_score_task_rename: rename task 2444[su] to [bash] oom_  score_adj=-1000
bash-2444  [007]   205.870407: oom_score_adj_inherited: new_task=2445 oom_score_adj=-1000
bash-2445  [001]   205.870975: oom_score_adj_inherited: new_task=2446 oom_score_adj=-1000

Changelog v2->v3:
  - use TRACE_EVENT_CONDITION to check condition.
  - use %d for pid.(for fixing warning..)

Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
---
 fs/exec.c                  |    4 ++
 fs/proc/base.c             |    3 ++
 include/trace/events/oom.h |   84 ++++++++++++++++++++++++++++++++++++++++++++
 kernel/fork.c              |    5 +++
 mm/oom_kill.c              |    6 +++
 5 files changed, 102 insertions(+), 0 deletions(-)
 create mode 100644 include/trace/events/oom.h

diff --git a/fs/exec.c b/fs/exec.c
index ca141db..9e99cf9 100644
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -59,6 +59,8 @@
 #include <asm/uaccess.h>
 #include <asm/mmu_context.h>
 #include <asm/tlb.h>
+
+#include <trace/events/oom.h>
 #include "internal.h"
 
 int core_uses_pid;
@@ -1054,6 +1056,8 @@ void set_task_comm(struct task_struct *tsk, char *buf)
 {
 	task_lock(tsk);
 
+	trace_oom_score_task_rename(tsk, buf);
+
 	/*
 	 * Threads may access current->comm without holding
 	 * the task lock, so write the string carefully.
diff --git a/fs/proc/base.c b/fs/proc/base.c
index 1050b1c..f201e64 100644
--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -87,6 +87,7 @@
 #ifdef CONFIG_HARDWALL
 #include <asm/hardwall.h>
 #endif
+#include <trace/events/oom.h>
 #include "internal.h"
 
 /* NOTE:
@@ -1166,6 +1167,7 @@ static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 	else
 		task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
 								-OOM_DISABLE;
+	trace_oom_score_adj_update(task);
 err_sighand:
 	unlock_task_sighand(task, &flags);
 err_task_lock:
@@ -1253,6 +1255,7 @@ static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
 	task->signal->oom_score_adj = oom_score_adj;
 	if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
 		task->signal->oom_score_adj_min = oom_score_adj;
+	trace_oom_score_adj_update(task);
 	/*
 	 * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
 	 * always attainable.
diff --git a/include/trace/events/oom.h b/include/trace/events/oom.h
new file mode 100644
index 0000000..b52b45d
--- /dev/null
+++ b/include/trace/events/oom.h
@@ -0,0 +1,84 @@
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM oom
+
+#if !defined(_TRACE_OOM_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_OOM_H
+#include <linux/tracepoint.h>
+
+TRACE_EVENT_CONDITION(oom_score_adj_inherited,
+
+	TP_PROTO(struct task_struct *task),
+	
+	TP_ARGS(task),
+
+	TP_CONDITION(task->signal->oom_score_adj != 0),
+
+	TP_STRUCT__entry(
+		__field(	pid_t,		newpid)
+		__field(	  int,		oom_score_adj)
+	),
+
+	TP_fast_assign(
+		__entry->newpid = task->pid;
+		__entry->oom_score_adj = task->signal->oom_score_adj;
+	),
+
+	TP_printk("new_task=%d oom_score_adj=%d",
+		__entry->newpid, __entry->oom_score_adj)
+);
+
+TRACE_EVENT_CONDITION(oom_score_task_rename,
+
+	TP_PROTO(struct task_struct *task, char *comm),
+
+	TP_ARGS(task, comm),
+
+	TP_CONDITION(task->signal->oom_score_adj != 0),
+
+	TP_STRUCT__entry(
+		__field(	pid_t,	 pid)
+		__array(         char,   oldcomm,   TASK_COMM_LEN   )
+		__array(         char,   newcomm,   TASK_COMM_LEN   )
+		__field(	  int,   oom_score_adj)
+	),
+
+	TP_fast_assign(
+		__entry->pid = task->pid;
+		 memcpy(__entry->oldcomm, task->comm, TASK_COMM_LEN);
+		 memcpy(__entry->newcomm, comm, TASK_COMM_LEN);
+		__entry->oom_score_adj = task->signal->oom_score_adj;
+	),
+
+	TP_printk("rename task %d[%s] to [%s] oom_score_adj=%d",
+		__entry->pid, __entry->oldcomm, __entry->newcomm,
+		__entry->oom_score_adj)
+);
+
+TRACE_EVENT(oom_score_adj_update,
+
+	TP_PROTO(struct task_struct *task),
+
+	TP_ARGS(task),
+
+	TP_STRUCT__entry(
+		__field(	pid_t,	pid)
+		__array(	char,	comm,	TASK_COMM_LEN )
+		__field(	 int,	oom_score_adj)
+	),
+
+	TP_fast_assign(
+		__entry->pid = task->pid;
+		memcpy(__entry->comm, task->comm, TASK_COMM_LEN);
+		__entry->oom_score_adj = task->signal->oom_score_adj;
+	),
+
+	TP_printk("task %d[%s] updates oom_score_adj=%d",
+		__entry->pid, __entry->comm, __entry->oom_score_adj)
+);
+
+#endif
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>
+
+
diff --git a/kernel/fork.c b/kernel/fork.c
index e20518d..758e5db 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -76,6 +76,7 @@
 #include <asm/tlbflush.h>
 
 #include <trace/events/sched.h>
+#include <trace/events/oom.h>
 
 /*
  * Protected counters by write_lock_irq(&tasklist_lock)
@@ -1390,6 +1391,10 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 	if (clone_flags & CLONE_THREAD)
 		threadgroup_fork_read_unlock(current);
 	perf_event_fork(p);
+
+	if (!(clone_flags & CLONE_THREAD))
+		trace_oom_score_adj_inherited(p);
+
 	return p;
 
 bad_fork_free_pid:
diff --git a/mm/oom_kill.c b/mm/oom_kill.c
index e2e1402..46b6d0a 100644
--- a/mm/oom_kill.c
+++ b/mm/oom_kill.c
@@ -33,6 +33,10 @@
 #include <linux/security.h>
 #include <linux/ptrace.h>
 #include <linux/freezer.h>
+#include <linux/ftrace.h>
+
+#define CREATE_TRACE_POINTS
+#include <trace/events/oom.h>
 
 int sysctl_panic_on_oom;
 int sysctl_oom_kill_allocating_task;
@@ -55,6 +59,7 @@ void compare_swap_oom_score_adj(int old_val, int new_val)
 	spin_lock_irq(&sighand->siglock);
 	if (current->signal->oom_score_adj == old_val)
 		current->signal->oom_score_adj = new_val;
+	trace_oom_score_adj_update(current);
 	spin_unlock_irq(&sighand->siglock);
 }
 
@@ -74,6 +79,7 @@ int test_set_oom_score_adj(int new_val)
 	spin_lock_irq(&sighand->siglock);
 	old_val = current->signal->oom_score_adj;
 	current->signal->oom_score_adj = new_val;
+	trace_oom_score_adj_update(current);
 	spin_unlock_irq(&sighand->siglock);
 
 	return old_val;
-- 
1.7.4.1

Re: [PATCH] oom: add tracepoints for oom_score_adj

[KOSAKI Motohiro]

On 12/6/2011 7:54 PM, KAMEZAWA Hiroyuki wrote:
>>From 28189e4622fd97324893a0b234183f64472a54d6 Mon Sep 17 00:00:00 2001
> From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> Date: Wed, 7 Dec 2011 09:58:16 +0900
> Subject: [PATCH] oom: trace point for oom_score_adj
> 
> oom_score_adj is set to prevent a task from being killed by OOM-Killer.
> Some daemons sets this value and their children inerit it sometimes.
> Because inheritance of oom_score_adj is done automatically, users
> can be confused at seeing the value and finds it's hard to debug.
> 
> This patch adds trace point for oom_score_adj. This adds 3 trace
> points. at
> 	- update oom_score_adj


> 	- fork()
> 	- rename task->comm(typically, exec())

I don't think they have oom specific thing. Can you please add generic fork and
task rename tracepoint instead?



> 
> Outputs will be following.
>    bash-2404  [006]   199.620841: oom_score_adj_update: task 2404[bash] updates oom_score_ad  j=-1000

"task 2404[bash]" don't look good to me.

In almost case, we use either

 - [pid] comm
 - pid:comm
 - comm:pid
 - comm-pid    (ftrace specific)

Why do we need to introduce alternative printing style?


>    bash-2404  [006]   205.861287: oom_score_adj_inherited: new_task=2442 oom_score_adj=-1000
>    su-2442  [003]   205.861761: oom_score_task_rename: rename task 2442[bash] to [su] oom_  score_adj=-1000
>    su-2442  [003]   205.866737: oom_score_adj_inherited: new_task=2444 oom_score_adj=-1000
>    bash-2444  [007]   205.868136: oom_score_task_rename: rename task 2444[su] to [bash] oom_  score_adj=-1000
>    bash-2444  [007]   205.870407: oom_score_adj_inherited: new_task=2445 oom_score_adj=-1000
>    bash-2445  [001]   205.870975: oom_score_adj_inherited: new_task=2446 oom_score_adj=-1000
> 
> Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> ---
>  fs/exec.c                  |    5 +++
>  fs/proc/base.c             |    3 ++
>  include/trace/events/oom.h |   80 ++++++++++++++++++++++++++++++++++++++++++++
>  kernel/fork.c              |    5 +++
>  mm/oom_kill.c              |    6 +++
>  5 files changed, 99 insertions(+), 0 deletions(-)
>  create mode 100644 include/trace/events/oom.h
> 
> diff --git a/fs/exec.c b/fs/exec.c
> index ca141db..562a106 100644
> --- a/fs/exec.c
> +++ b/fs/exec.c
> @@ -59,6 +59,8 @@
>  #include <asm/uaccess.h>
>  #include <asm/mmu_context.h>
>  #include <asm/tlb.h>
> +
> +#include <trace/events/oom.h>
>  #include "internal.h"
>  
>  int core_uses_pid;
> @@ -1054,6 +1056,9 @@ void set_task_comm(struct task_struct *tsk, char *buf)
>  {
>  	task_lock(tsk);
>  
> +	if (tsk->signal->oom_score_adj)
> +		trace_oom_score_task_rename(tsk, buf);
> +
>  	/*
>  	 * Threads may access current->comm without holding
>  	 * the task lock, so write the string carefully.
> diff --git a/fs/proc/base.c b/fs/proc/base.c
> index 1050b1c..f201e64 100644
> --- a/fs/proc/base.c
> +++ b/fs/proc/base.c
> @@ -87,6 +87,7 @@
>  #ifdef CONFIG_HARDWALL
>  #include <asm/hardwall.h>
>  #endif
> +#include <trace/events/oom.h>
>  #include "internal.h"
>  
>  /* NOTE:
> @@ -1166,6 +1167,7 @@ static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
>  	else
>  		task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
>  								-OOM_DISABLE;
> +	trace_oom_score_adj_update(task);
>  err_sighand:
>  	unlock_task_sighand(task, &flags);
>  err_task_lock:
> @@ -1253,6 +1255,7 @@ static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
>  	task->signal->oom_score_adj = oom_score_adj;
>  	if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
>  		task->signal->oom_score_adj_min = oom_score_adj;
> +	trace_oom_score_adj_update(task);
>  	/*
>  	 * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
>  	 * always attainable.
> diff --git a/include/trace/events/oom.h b/include/trace/events/oom.h
> new file mode 100644
> index 0000000..f5e6f55
> --- /dev/null
> +++ b/include/trace/events/oom.h
> @@ -0,0 +1,80 @@
> +#undef TRACE_SYSTEM
> +#define TRACE_SYSTEM oom
> +
> +#if !defined(_TRACE_OOM_H) || defined(TRACE_HEADER_MULTI_READ)
> +#define _TRACE_OOM_H
> +#include <linux/tracepoint.h>
> +
> +TRACE_EVENT(oom_score_adj_inherited,
> +
> +	TP_PROTO(struct task_struct *task),
> +	
> +	TP_ARGS(task),
> +
> +	TP_STRUCT__entry(
> +		__field(	pid_t,		newpid)
> +		__field(	  int,		oom_score_adj)
> +	),
> +
> +	TP_fast_assign(
> +		__entry->newpid = task->pid;
> +		__entry->oom_score_adj = task->signal->oom_score_adj;
> +	),
> +
> +	TP_printk("new_task=%ld oom_score_adj=%d",
> +		__entry->newpid, __entry->oom_score_adj)
> +);
> +
> +TRACE_EVENT(oom_score_task_rename,
> +
> +	TP_PROTO(struct task_struct *task, char *comm),
> +
> +	TP_ARGS(task, comm),
> +
> +	TP_STRUCT__entry(
> +		__field(	pid_t,	 pid)
> +		__array(         char,   oldcomm,   TASK_COMM_LEN   )
> +		__array(         char,   newcomm,   TASK_COMM_LEN   )
> +		__field(	  int,   oom_score_adj)
> +	),
> +
> +	TP_fast_assign(
> +		__entry->pid = task->pid;
> +		 memcpy(__entry->oldcomm, task->comm, TASK_COMM_LEN);
> +		 memcpy(__entry->newcomm, comm, TASK_COMM_LEN);
> +		__entry->oom_score_adj = task->signal->oom_score_adj;
> +	),
> +
> +	TP_printk("rename task %ld[%s] to [%s] oom_score_adj=%d",
> +		__entry->pid, __entry->oldcomm, __entry->newcomm,
> +		__entry->oom_score_adj)
> +);
> +
> +TRACE_EVENT(oom_score_adj_update,
> +
> +	TP_PROTO(struct task_struct *task),
> +
> +	TP_ARGS(task),
> +
> +	TP_STRUCT__entry(
> +		__field(	pid_t,	pid)
> +		__array(	char,	comm,	TASK_COMM_LEN )
> +		__field(	 int,	oom_score_adj)
> +	),
> +
> +	TP_fast_assign(
> +		__entry->pid = task->pid;
> +		memcpy(__entry->comm, task->comm, TASK_COMM_LEN);
> +		__entry->oom_score_adj = task->signal->oom_score_adj;
> +	),
> +
> +	TP_printk("task %ld[%s] updates oom_score_adj=%d",
> +		__entry->pid, __entry->comm, __entry->oom_score_adj)
> +);
> +
> +#endif
> +
> +/* This part must be outside protection */
> +#include <trace/define_trace.h>
> +
> +
> diff --git a/kernel/fork.c b/kernel/fork.c
> index e20518d..634aa84 100644
> --- a/kernel/fork.c
> +++ b/kernel/fork.c
> @@ -76,6 +76,7 @@
>  #include <asm/tlbflush.h>
>  
>  #include <trace/events/sched.h>
> +#include <trace/events/oom.h>
>  
>  /*
>   * Protected counters by write_lock_irq(&tasklist_lock)
> @@ -1390,6 +1391,10 @@ static struct task_struct *copy_process(unsigned long clone_flags,
>  	if (clone_flags & CLONE_THREAD)
>  		threadgroup_fork_read_unlock(current);
>  	perf_event_fork(p);
> +
> +	if (!(clone_flags & CLONE_THREAD) && p->signal->oom_score_adj)
> +		trace_oom_score_adj_inherited(p);
> +
>  	return p;
>  
>  bad_fork_free_pid:
> diff --git a/mm/oom_kill.c b/mm/oom_kill.c
> index e2e1402..46b6d0a 100644
> --- a/mm/oom_kill.c
> +++ b/mm/oom_kill.c
> @@ -33,6 +33,10 @@
>  #include <linux/security.h>
>  #include <linux/ptrace.h>
>  #include <linux/freezer.h>
> +#include <linux/ftrace.h>
> +
> +#define CREATE_TRACE_POINTS
> +#include <trace/events/oom.h>
>  
>  int sysctl_panic_on_oom;
>  int sysctl_oom_kill_allocating_task;
> @@ -55,6 +59,7 @@ void compare_swap_oom_score_adj(int old_val, int new_val)
>  	spin_lock_irq(&sighand->siglock);
>  	if (current->signal->oom_score_adj == old_val)
>  		current->signal->oom_score_adj = new_val;
> +	trace_oom_score_adj_update(current);
>  	spin_unlock_irq(&sighand->siglock);
>  }
>  
> @@ -74,6 +79,7 @@ int test_set_oom_score_adj(int new_val)
>  	spin_lock_irq(&sighand->siglock);
>  	old_val = current->signal->oom_score_adj;
>  	current->signal->oom_score_adj = new_val;
> +	trace_oom_score_adj_update(current);
>  	spin_unlock_irq(&sighand->siglock);
>  
>  	return old_val;

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] oom: add tracepoints for oom_score_adj

[KOSAKI Motohiro]

On 12/6/2011 7:54 PM, KAMEZAWA Hiroyuki wrote:
>>From 28189e4622fd97324893a0b234183f64472a54d6 Mon Sep 17 00:00:00 2001
> From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> Date: Wed, 7 Dec 2011 09:58:16 +0900
> Subject: [PATCH] oom: trace point for oom_score_adj
> 
> oom_score_adj is set to prevent a task from being killed by OOM-Killer.
> Some daemons sets this value and their children inerit it sometimes.
> Because inheritance of oom_score_adj is done automatically, users
> can be confused at seeing the value and finds it's hard to debug.
> 
> This patch adds trace point for oom_score_adj. This adds 3 trace
> points. at
> 	- update oom_score_adj


> 	- fork()
> 	- rename task->comm(typically, exec())

I don't think they have oom specific thing. Can you please add generic fork and
task rename tracepoint instead?



> 
> Outputs will be following.
>    bash-2404  [006]   199.620841: oom_score_adj_update: task 2404[bash] updates oom_score_ad  j=-1000

"task 2404[bash]" don't look good to me.

In almost case, we use either

 - [pid] comm
 - pid:comm
 - comm:pid
 - comm-pid    (ftrace specific)

Why do we need to introduce alternative printing style?


>    bash-2404  [006]   205.861287: oom_score_adj_inherited: new_task=2442 oom_score_adj=-1000
>    su-2442  [003]   205.861761: oom_score_task_rename: rename task 2442[bash] to [su] oom_  score_adj=-1000
>    su-2442  [003]   205.866737: oom_score_adj_inherited: new_task=2444 oom_score_adj=-1000
>    bash-2444  [007]   205.868136: oom_score_task_rename: rename task 2444[su] to [bash] oom_  score_adj=-1000
>    bash-2444  [007]   205.870407: oom_score_adj_inherited: new_task=2445 oom_score_adj=-1000
>    bash-2445  [001]   205.870975: oom_score_adj_inherited: new_task=2446 oom_score_adj=-1000
> 
> Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> ---
>  fs/exec.c                  |    5 +++
>  fs/proc/base.c             |    3 ++
>  include/trace/events/oom.h |   80 ++++++++++++++++++++++++++++++++++++++++++++
>  kernel/fork.c              |    5 +++
>  mm/oom_kill.c              |    6 +++
>  5 files changed, 99 insertions(+), 0 deletions(-)
>  create mode 100644 include/trace/events/oom.h
> 
> diff --git a/fs/exec.c b/fs/exec.c
> index ca141db..562a106 100644
> --- a/fs/exec.c
> +++ b/fs/exec.c
> @@ -59,6 +59,8 @@
>  #include <asm/uaccess.h>
>  #include <asm/mmu_context.h>
>  #include <asm/tlb.h>
> +
> +#include <trace/events/oom.h>
>  #include "internal.h"
>  
>  int core_uses_pid;
> @@ -1054,6 +1056,9 @@ void set_task_comm(struct task_struct *tsk, char *buf)
>  {
>  	task_lock(tsk);
>  
> +	if (tsk->signal->oom_score_adj)
> +		trace_oom_score_task_rename(tsk, buf);
> +
>  	/*
>  	 * Threads may access current->comm without holding
>  	 * the task lock, so write the string carefully.
> diff --git a/fs/proc/base.c b/fs/proc/base.c
> index 1050b1c..f201e64 100644
> --- a/fs/proc/base.c
> +++ b/fs/proc/base.c
> @@ -87,6 +87,7 @@
>  #ifdef CONFIG_HARDWALL
>  #include <asm/hardwall.h>
>  #endif
> +#include <trace/events/oom.h>
>  #include "internal.h"
>  
>  /* NOTE:
> @@ -1166,6 +1167,7 @@ static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
>  	else
>  		task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
>  								-OOM_DISABLE;
> +	trace_oom_score_adj_update(task);
>  err_sighand:
>  	unlock_task_sighand(task, &flags);
>  err_task_lock:
> @@ -1253,6 +1255,7 @@ static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
>  	task->signal->oom_score_adj = oom_score_adj;
>  	if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
>  		task->signal->oom_score_adj_min = oom_score_adj;
> +	trace_oom_score_adj_update(task);
>  	/*
>  	 * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
>  	 * always attainable.
> diff --git a/include/trace/events/oom.h b/include/trace/events/oom.h
> new file mode 100644
> index 0000000..f5e6f55
> --- /dev/null
> +++ b/include/trace/events/oom.h
> @@ -0,0 +1,80 @@
> +#undef TRACE_SYSTEM
> +#define TRACE_SYSTEM oom
> +
> +#if !defined(_TRACE_OOM_H) || defined(TRACE_HEADER_MULTI_READ)
> +#define _TRACE_OOM_H
> +#include <linux/tracepoint.h>
> +
> +TRACE_EVENT(oom_score_adj_inherited,
> +
> +	TP_PROTO(struct task_struct *task),
> +	
> +	TP_ARGS(task),
> +
> +	TP_STRUCT__entry(
> +		__field(	pid_t,		newpid)
> +		__field(	  int,		oom_score_adj)
> +	),
> +
> +	TP_fast_assign(
> +		__entry->newpid = task->pid;
> +		__entry->oom_score_adj = task->signal->oom_score_adj;
> +	),
> +
> +	TP_printk("new_task=%ld oom_score_adj=%d",
> +		__entry->newpid, __entry->oom_score_adj)
> +);
> +
> +TRACE_EVENT(oom_score_task_rename,
> +
> +	TP_PROTO(struct task_struct *task, char *comm),
> +
> +	TP_ARGS(task, comm),
> +
> +	TP_STRUCT__entry(
> +		__field(	pid_t,	 pid)
> +		__array(         char,   oldcomm,   TASK_COMM_LEN   )
> +		__array(         char,   newcomm,   TASK_COMM_LEN   )
> +		__field(	  int,   oom_score_adj)
> +	),
> +
> +	TP_fast_assign(
> +		__entry->pid = task->pid;
> +		 memcpy(__entry->oldcomm, task->comm, TASK_COMM_LEN);
> +		 memcpy(__entry->newcomm, comm, TASK_COMM_LEN);
> +		__entry->oom_score_adj = task->signal->oom_score_adj;
> +	),
> +
> +	TP_printk("rename task %ld[%s] to [%s] oom_score_adj=%d",
> +		__entry->pid, __entry->oldcomm, __entry->newcomm,
> +		__entry->oom_score_adj)
> +);
> +
> +TRACE_EVENT(oom_score_adj_update,
> +
> +	TP_PROTO(struct task_struct *task),
> +
> +	TP_ARGS(task),
> +
> +	TP_STRUCT__entry(
> +		__field(	pid_t,	pid)
> +		__array(	char,	comm,	TASK_COMM_LEN )
> +		__field(	 int,	oom_score_adj)
> +	),
> +
> +	TP_fast_assign(
> +		__entry->pid = task->pid;
> +		memcpy(__entry->comm, task->comm, TASK_COMM_LEN);
> +		__entry->oom_score_adj = task->signal->oom_score_adj;
> +	),
> +
> +	TP_printk("task %ld[%s] updates oom_score_adj=%d",
> +		__entry->pid, __entry->comm, __entry->oom_score_adj)
> +);
> +
> +#endif
> +
> +/* This part must be outside protection */
> +#include <trace/define_trace.h>
> +
> +
> diff --git a/kernel/fork.c b/kernel/fork.c
> index e20518d..634aa84 100644
> --- a/kernel/fork.c
> +++ b/kernel/fork.c
> @@ -76,6 +76,7 @@
>  #include <asm/tlbflush.h>
>  
>  #include <trace/events/sched.h>
> +#include <trace/events/oom.h>
>  
>  /*
>   * Protected counters by write_lock_irq(&tasklist_lock)
> @@ -1390,6 +1391,10 @@ static struct task_struct *copy_process(unsigned long clone_flags,
>  	if (clone_flags & CLONE_THREAD)
>  		threadgroup_fork_read_unlock(current);
>  	perf_event_fork(p);
> +
> +	if (!(clone_flags & CLONE_THREAD) && p->signal->oom_score_adj)
> +		trace_oom_score_adj_inherited(p);
> +
>  	return p;
>  
>  bad_fork_free_pid:
> diff --git a/mm/oom_kill.c b/mm/oom_kill.c
> index e2e1402..46b6d0a 100644
> --- a/mm/oom_kill.c
> +++ b/mm/oom_kill.c
> @@ -33,6 +33,10 @@
>  #include <linux/security.h>
>  #include <linux/ptrace.h>
>  #include <linux/freezer.h>
> +#include <linux/ftrace.h>
> +
> +#define CREATE_TRACE_POINTS
> +#include <trace/events/oom.h>
>  
>  int sysctl_panic_on_oom;
>  int sysctl_oom_kill_allocating_task;
> @@ -55,6 +59,7 @@ void compare_swap_oom_score_adj(int old_val, int new_val)
>  	spin_lock_irq(&sighand->siglock);
>  	if (current->signal->oom_score_adj == old_val)
>  		current->signal->oom_score_adj = new_val;
> +	trace_oom_score_adj_update(current);
>  	spin_unlock_irq(&sighand->siglock);
>  }
>  
> @@ -74,6 +79,7 @@ int test_set_oom_score_adj(int new_val)
>  	spin_lock_irq(&sighand->siglock);
>  	old_val = current->signal->oom_score_adj;
>  	current->signal->oom_score_adj = new_val;
> +	trace_oom_score_adj_update(current);
>  	spin_unlock_irq(&sighand->siglock);
>  
>  	return old_val;

Re: [PATCH] oom: add tracepoints for oom_score_adj

[KAMEZAWA Hiroyuki]

On Wed, 07 Dec 2011 11:52:02 -0500
KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> wrote:

> On 12/6/2011 7:54 PM, KAMEZAWA Hiroyuki wrote:
> >>From 28189e4622fd97324893a0b234183f64472a54d6 Mon Sep 17 00:00:00 2001
> > From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> > Date: Wed, 7 Dec 2011 09:58:16 +0900
> > Subject: [PATCH] oom: trace point for oom_score_adj
> > 
> > oom_score_adj is set to prevent a task from being killed by OOM-Killer.
> > Some daemons sets this value and their children inerit it sometimes.
> > Because inheritance of oom_score_adj is done automatically, users
> > can be confused at seeing the value and finds it's hard to debug.
> > 
> > This patch adds trace point for oom_score_adj. This adds 3 trace
> > points. at
> > 	- update oom_score_adj
> 
> 
> > 	- fork()
> > 	- rename task->comm(typically, exec())
> 
> I don't think they have oom specific thing. Can you please add generic fork and
> task rename tracepoint instead?
> 
I think it makes oom-targeted debug difficult.
This tracehook using task->signal->oom_score_adj as filter.
This reduces traces much and makes debugging easier.
 
If you need another trace point for other purpose, another trace point
should be better. For generic purpose, oom_socre_adj filtering will not
be necessary.





> > 
> > Outputs will be following.
> >    bash-2404  [006]   199.620841: oom_score_adj_update: task 2404[bash] updates oom_score_ad  j=-1000
> 
> "task 2404[bash]" don't look good to me.
> 
> In almost case, we use either
> 
>  - [pid] comm
>  - pid:comm
>  - comm:pid
>  - comm-pid    (ftrace specific)
> 
> Why do we need to introduce alternative printing style?
> 

No reason. ok, I'll fix.

Thanks,
-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] oom: add tracepoints for oom_score_adj

[KAMEZAWA Hiroyuki]

On Wed, 07 Dec 2011 11:52:02 -0500
KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> wrote:

> On 12/6/2011 7:54 PM, KAMEZAWA Hiroyuki wrote:
> >>From 28189e4622fd97324893a0b234183f64472a54d6 Mon Sep 17 00:00:00 2001
> > From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> > Date: Wed, 7 Dec 2011 09:58:16 +0900
> > Subject: [PATCH] oom: trace point for oom_score_adj
> > 
> > oom_score_adj is set to prevent a task from being killed by OOM-Killer.
> > Some daemons sets this value and their children inerit it sometimes.
> > Because inheritance of oom_score_adj is done automatically, users
> > can be confused at seeing the value and finds it's hard to debug.
> > 
> > This patch adds trace point for oom_score_adj. This adds 3 trace
> > points. at
> > 	- update oom_score_adj
> 
> 
> > 	- fork()
> > 	- rename task->comm(typically, exec())
> 
> I don't think they have oom specific thing. Can you please add generic fork and
> task rename tracepoint instead?
> 
I think it makes oom-targeted debug difficult.
This tracehook using task->signal->oom_score_adj as filter.
This reduces traces much and makes debugging easier.
 
If you need another trace point for other purpose, another trace point
should be better. For generic purpose, oom_socre_adj filtering will not
be necessary.





> > 
> > Outputs will be following.
> >    bash-2404  [006]   199.620841: oom_score_adj_update: task 2404[bash] updates oom_score_ad  j=-1000
> 
> "task 2404[bash]" don't look good to me.
> 
> In almost case, we use either
> 
>  - [pid] comm
>  - pid:comm
>  - comm:pid
>  - comm-pid    (ftrace specific)
> 
> Why do we need to introduce alternative printing style?
> 

No reason. ok, I'll fix.

Thanks,
-Kame

[PATCH v4] oom: add tracepoints for oom_score_adj

[KAMEZAWA Hiroyuki]

On Thu, 8 Dec 2011 10:47:05 +0900
KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote:

> On Wed, 07 Dec 2011 11:52:02 -0500
> KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> wrote:
  - [pid] comm
> >  - pid:comm
> >  - comm:pid
> >  - comm-pid    (ftrace specific)
> > 
> > Why do we need to introduce alternative printing style?
> > 
> 

v4 here
==

[PATCH v4] oom: add tracepoints for oom_score_adj

[KAMEZAWA Hiroyuki]

On Thu, 8 Dec 2011 10:47:05 +0900
KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote:

> On Wed, 07 Dec 2011 11:52:02 -0500
> KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> wrote:
  - [pid] comm
> >  - pid:comm
> >  - comm:pid
> >  - comm-pid    (ftrace specific)
> > 
> > Why do we need to introduce alternative printing style?
> > 
> 

v4 here
==
>From 5dc1f8c879ae424d5853af255df8860494209e39 Mon Sep 17 00:00:00 2001
From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Date: Wed, 7 Dec 2011 09:58:16 +0900
Subject: [PATCH] oom: trace point for oom_score_adj

oom_score_adj is set to prevent a task from being killed by OOM-Killer.
Some daemons sets this value and their children inerit it sometimes.
Because inheritance of oom_score_adj is done automatically, users
can be confused at seeing the value and finds it's hard to debug.

This patch adds trace point for oom_score_adj. This adds 3 trace
points. at
	- update oom_score_adj
	- fork()
	- rename task->comm(typically, exec())

At fork and rename, this trace has filters as task->signal->oom_score_adj != 0.
Then, users can extract information required for fixing oom problem easily.

<...>-2456  [002]    87.347612: oom_score_adj_update: task 2456:bash updates oom_score_adj=-1000
<...>-2456  [002]    90.228660: oom_score_adj_inherited: new_task=2480 oom_score_adj=-1000
ls-2480  [007]    90.229122: oom_score_task_rename: rename task 2480:bash to ls oom_score_adj=-1000
<...>-2456  [006]   111.024606: oom_score_adj_inherited: new_task=2482 oom_score_adj=-1000
bash-2482  [003]   111.025174: oom_score_task_rename: rename task 2482:bash to bash oom_score_adj=-1000

Changelog v3->v4:
  - change format from pid[comm] to pid:comm
  - updated patch description.

Changelog v2->v3:
  - use TRACE_EVENT_CONDITION to check condition.
  - use %d for pid.

Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
---
 fs/exec.c                  |    4 ++
 fs/proc/base.c             |    3 ++
 include/trace/events/oom.h |   84 ++++++++++++++++++++++++++++++++++++++++++++
 kernel/fork.c              |    5 +++
 mm/oom_kill.c              |    6 +++
 5 files changed, 102 insertions(+), 0 deletions(-)
 create mode 100644 include/trace/events/oom.h

diff --git a/fs/exec.c b/fs/exec.c
index ca141db..9e99cf9 100644
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -59,6 +59,8 @@
 #include <asm/uaccess.h>
 #include <asm/mmu_context.h>
 #include <asm/tlb.h>
+
+#include <trace/events/oom.h>
 #include "internal.h"
 
 int core_uses_pid;
@@ -1054,6 +1056,8 @@ void set_task_comm(struct task_struct *tsk, char *buf)
 {
 	task_lock(tsk);
 
+	trace_oom_score_task_rename(tsk, buf);
+
 	/*
 	 * Threads may access current->comm without holding
 	 * the task lock, so write the string carefully.
diff --git a/fs/proc/base.c b/fs/proc/base.c
index 1050b1c..f201e64 100644
--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -87,6 +87,7 @@
 #ifdef CONFIG_HARDWALL
 #include <asm/hardwall.h>
 #endif
+#include <trace/events/oom.h>
 #include "internal.h"
 
 /* NOTE:
@@ -1166,6 +1167,7 @@ static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 	else
 		task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
 								-OOM_DISABLE;
+	trace_oom_score_adj_update(task);
 err_sighand:
 	unlock_task_sighand(task, &flags);
 err_task_lock:
@@ -1253,6 +1255,7 @@ static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
 	task->signal->oom_score_adj = oom_score_adj;
 	if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
 		task->signal->oom_score_adj_min = oom_score_adj;
+	trace_oom_score_adj_update(task);
 	/*
 	 * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
 	 * always attainable.
diff --git a/include/trace/events/oom.h b/include/trace/events/oom.h
new file mode 100644
index 0000000..275155c
--- /dev/null
+++ b/include/trace/events/oom.h
@@ -0,0 +1,84 @@
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM oom
+
+#if !defined(_TRACE_OOM_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_OOM_H
+#include <linux/tracepoint.h>
+
+TRACE_EVENT_CONDITION(oom_score_adj_inherited,
+
+	TP_PROTO(struct task_struct *task),
+	
+	TP_ARGS(task),
+
+	TP_CONDITION(task->signal->oom_score_adj != 0),
+
+	TP_STRUCT__entry(
+		__field(	pid_t,		newpid)
+		__field(	  int,		oom_score_adj)
+	),
+
+	TP_fast_assign(
+		__entry->newpid = task->pid;
+		__entry->oom_score_adj = task->signal->oom_score_adj;
+	),
+
+	TP_printk("new_task=%d oom_score_adj=%d",
+		__entry->newpid, __entry->oom_score_adj)
+);
+
+TRACE_EVENT_CONDITION(oom_score_task_rename,
+
+	TP_PROTO(struct task_struct *task, char *comm),
+
+	TP_ARGS(task, comm),
+
+	TP_CONDITION(task->signal->oom_score_adj != 0),
+
+	TP_STRUCT__entry(
+		__field(	pid_t,	 pid)
+		__array(         char,   oldcomm,   TASK_COMM_LEN   )
+		__array(         char,   newcomm,   TASK_COMM_LEN   )
+		__field(	  int,   oom_score_adj)
+	),
+
+	TP_fast_assign(
+		__entry->pid = task->pid;
+		 memcpy(__entry->oldcomm, task->comm, TASK_COMM_LEN);
+		 memcpy(__entry->newcomm, comm, TASK_COMM_LEN);
+		__entry->oom_score_adj = task->signal->oom_score_adj;
+	),
+
+	TP_printk("rename task %d:%s to %s oom_score_adj=%d",
+		__entry->pid, __entry->oldcomm, __entry->newcomm,
+		__entry->oom_score_adj)
+);
+
+TRACE_EVENT(oom_score_adj_update,
+
+	TP_PROTO(struct task_struct *task),
+
+	TP_ARGS(task),
+
+	TP_STRUCT__entry(
+		__field(	pid_t,	pid)
+		__array(	char,	comm,	TASK_COMM_LEN )
+		__field(	 int,	oom_score_adj)
+	),
+
+	TP_fast_assign(
+		__entry->pid = task->pid;
+		memcpy(__entry->comm, task->comm, TASK_COMM_LEN);
+		__entry->oom_score_adj = task->signal->oom_score_adj;
+	),
+
+	TP_printk("task %d:%s updates oom_score_adj=%d",
+		__entry->pid, __entry->comm, __entry->oom_score_adj)
+);
+
+#endif
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>
+
+
diff --git a/kernel/fork.c b/kernel/fork.c
index e20518d..758e5db 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -76,6 +76,7 @@
 #include <asm/tlbflush.h>
 
 #include <trace/events/sched.h>
+#include <trace/events/oom.h>
 
 /*
  * Protected counters by write_lock_irq(&tasklist_lock)
@@ -1390,6 +1391,10 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 	if (clone_flags & CLONE_THREAD)
 		threadgroup_fork_read_unlock(current);
 	perf_event_fork(p);
+
+	if (!(clone_flags & CLONE_THREAD))
+		trace_oom_score_adj_inherited(p);
+
 	return p;
 
 bad_fork_free_pid:
diff --git a/mm/oom_kill.c b/mm/oom_kill.c
index e2e1402..46b6d0a 100644
--- a/mm/oom_kill.c
+++ b/mm/oom_kill.c
@@ -33,6 +33,10 @@
 #include <linux/security.h>
 #include <linux/ptrace.h>
 #include <linux/freezer.h>
+#include <linux/ftrace.h>
+
+#define CREATE_TRACE_POINTS
+#include <trace/events/oom.h>
 
 int sysctl_panic_on_oom;
 int sysctl_oom_kill_allocating_task;
@@ -55,6 +59,7 @@ void compare_swap_oom_score_adj(int old_val, int new_val)
 	spin_lock_irq(&sighand->siglock);
 	if (current->signal->oom_score_adj == old_val)
 		current->signal->oom_score_adj = new_val;
+	trace_oom_score_adj_update(current);
 	spin_unlock_irq(&sighand->siglock);
 }
 
@@ -74,6 +79,7 @@ int test_set_oom_score_adj(int new_val)
 	spin_lock_irq(&sighand->siglock);
 	old_val = current->signal->oom_score_adj;
 	current->signal->oom_score_adj = new_val;
+	trace_oom_score_adj_update(current);
 	spin_unlock_irq(&sighand->siglock);
 
 	return old_val;
-- 
1.7.4.1

Re: [PATCH v4] oom: add tracepoints for oom_score_adj

[Peter Zijlstra]

On Thu, Dec 08, 2011 at 03:32:30PM +0900, KAMEZAWA Hiroyuki wrote:
> From 5dc1f8c879ae424d5853af255df8860494209e39 Mon Sep 17 00:00:00 2001
> From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> Date: Wed, 7 Dec 2011 09:58:16 +0900
> Subject: [PATCH] oom: trace point for oom_score_adj
> 
> oom_score_adj is set to prevent a task from being killed by OOM-Killer.
> Some daemons sets this value and their children inerit it sometimes.
> Because inheritance of oom_score_adj is done automatically, users
> can be confused at seeing the value and finds it's hard to debug.
> 
> This patch adds trace point for oom_score_adj. This adds 3 trace
> points. at
> 	- update oom_score_adj
> 	- fork()
> 	- rename task->comm(typically, exec())
> 

And nobody was bothered by the fact that we already had fork and exec tracepoints?

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH v4] oom: add tracepoints for oom_score_adj

[Peter Zijlstra]

On Thu, Dec 08, 2011 at 03:32:30PM +0900, KAMEZAWA Hiroyuki wrote:
> From 5dc1f8c879ae424d5853af255df8860494209e39 Mon Sep 17 00:00:00 2001
> From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> Date: Wed, 7 Dec 2011 09:58:16 +0900
> Subject: [PATCH] oom: trace point for oom_score_adj
> 
> oom_score_adj is set to prevent a task from being killed by OOM-Killer.
> Some daemons sets this value and their children inerit it sometimes.
> Because inheritance of oom_score_adj is done automatically, users
> can be confused at seeing the value and finds it's hard to debug.
> 
> This patch adds trace point for oom_score_adj. This adds 3 trace
> points. at
> 	- update oom_score_adj
> 	- fork()
> 	- rename task->comm(typically, exec())
> 

And nobody was bothered by the fact that we already had fork and exec tracepoints?

Re: [PATCH] oom: add tracepoints for oom_score_adj

[KOSAKI Motohiro]

On 12/7/2011 8:47 PM, KAMEZAWA Hiroyuki wrote:
> On Wed, 07 Dec 2011 11:52:02 -0500
> KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> wrote:
> 
>> On 12/6/2011 7:54 PM, KAMEZAWA Hiroyuki wrote:
>>> >From 28189e4622fd97324893a0b234183f64472a54d6 Mon Sep 17 00:00:00 2001
>>> From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
>>> Date: Wed, 7 Dec 2011 09:58:16 +0900
>>> Subject: [PATCH] oom: trace point for oom_score_adj
>>>
>>> oom_score_adj is set to prevent a task from being killed by OOM-Killer.
>>> Some daemons sets this value and their children inerit it sometimes.
>>> Because inheritance of oom_score_adj is done automatically, users
>>> can be confused at seeing the value and finds it's hard to debug.
>>>
>>> This patch adds trace point for oom_score_adj. This adds 3 trace
>>> points. at
>>> 	- update oom_score_adj
>>
>>
>>> 	- fork()
>>> 	- rename task->comm(typically, exec())
>>
>> I don't think they have oom specific thing. Can you please add generic fork and
>> task rename tracepoint instead?
>>
> I think it makes oom-targeted debug difficult.
> This tracehook using task->signal->oom_score_adj as filter.
> This reduces traces much and makes debugging easier.
>  
> If you need another trace point for other purpose, another trace point
> should be better. For generic purpose, oom_socre_adj filtering will not
> be necessary.

see Documentation/trace/event.txt 5. Event filgtering

Now, both ftrace and perf have good filter feature. Isn't this enough?




--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] oom: add tracepoints for oom_score_adj

[KOSAKI Motohiro]

On 12/7/2011 8:47 PM, KAMEZAWA Hiroyuki wrote:
> On Wed, 07 Dec 2011 11:52:02 -0500
> KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> wrote:
> 
>> On 12/6/2011 7:54 PM, KAMEZAWA Hiroyuki wrote:
>>> >From 28189e4622fd97324893a0b234183f64472a54d6 Mon Sep 17 00:00:00 2001
>>> From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
>>> Date: Wed, 7 Dec 2011 09:58:16 +0900
>>> Subject: [PATCH] oom: trace point for oom_score_adj
>>>
>>> oom_score_adj is set to prevent a task from being killed by OOM-Killer.
>>> Some daemons sets this value and their children inerit it sometimes.
>>> Because inheritance of oom_score_adj is done automatically, users
>>> can be confused at seeing the value and finds it's hard to debug.
>>>
>>> This patch adds trace point for oom_score_adj. This adds 3 trace
>>> points. at
>>> 	- update oom_score_adj
>>
>>
>>> 	- fork()
>>> 	- rename task->comm(typically, exec())
>>
>> I don't think they have oom specific thing. Can you please add generic fork and
>> task rename tracepoint instead?
>>
> I think it makes oom-targeted debug difficult.
> This tracehook using task->signal->oom_score_adj as filter.
> This reduces traces much and makes debugging easier.
>  
> If you need another trace point for other purpose, another trace point
> should be better. For generic purpose, oom_socre_adj filtering will not
> be necessary.

see Documentation/trace/event.txt 5. Event filgtering

Now, both ftrace and perf have good filter feature. Isn't this enough?

Re: [PATCH] oom: add tracepoints for oom_score_adj

[KAMEZAWA Hiroyuki]

On Thu, 08 Dec 2011 12:33:35 -0500
KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> wrote:

> On 12/7/2011 8:47 PM, KAMEZAWA Hiroyuki wrote:
> > On Wed, 07 Dec 2011 11:52:02 -0500
> > KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> wrote:
> > 
> >> On 12/6/2011 7:54 PM, KAMEZAWA Hiroyuki wrote:
> >>> >From 28189e4622fd97324893a0b234183f64472a54d6 Mon Sep 17 00:00:00 2001
> >>> From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> >>> Date: Wed, 7 Dec 2011 09:58:16 +0900
> >>> Subject: [PATCH] oom: trace point for oom_score_adj
> >>>
> >>> oom_score_adj is set to prevent a task from being killed by OOM-Killer.
> >>> Some daemons sets this value and their children inerit it sometimes.
> >>> Because inheritance of oom_score_adj is done automatically, users
> >>> can be confused at seeing the value and finds it's hard to debug.
> >>>
> >>> This patch adds trace point for oom_score_adj. This adds 3 trace
> >>> points. at
> >>> 	- update oom_score_adj
> >>
> >>
> >>> 	- fork()
> >>> 	- rename task->comm(typically, exec())
> >>
> >> I don't think they have oom specific thing. Can you please add generic fork and
> >> task rename tracepoint instead?
> >>
> > I think it makes oom-targeted debug difficult.
> > This tracehook using task->signal->oom_score_adj as filter.
> > This reduces traces much and makes debugging easier.
> >  
> > If you need another trace point for other purpose, another trace point
> > should be better. For generic purpose, oom_socre_adj filtering will not
> > be necessary.
> 
> see Documentation/trace/event.txt 5. Event filgtering
> 
> Now, both ftrace and perf have good filter feature. Isn't this enough?
> 

Could you make patch ? Then, I stop this and go other probelm.

Thanks,
-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] oom: add tracepoints for oom_score_adj

[KAMEZAWA Hiroyuki]

On Thu, 08 Dec 2011 12:33:35 -0500
KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> wrote:

> On 12/7/2011 8:47 PM, KAMEZAWA Hiroyuki wrote:
> > On Wed, 07 Dec 2011 11:52:02 -0500
> > KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> wrote:
> > 
> >> On 12/6/2011 7:54 PM, KAMEZAWA Hiroyuki wrote:
> >>> >From 28189e4622fd97324893a0b234183f64472a54d6 Mon Sep 17 00:00:00 2001
> >>> From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> >>> Date: Wed, 7 Dec 2011 09:58:16 +0900
> >>> Subject: [PATCH] oom: trace point for oom_score_adj
> >>>
> >>> oom_score_adj is set to prevent a task from being killed by OOM-Killer.
> >>> Some daemons sets this value and their children inerit it sometimes.
> >>> Because inheritance of oom_score_adj is done automatically, users
> >>> can be confused at seeing the value and finds it's hard to debug.
> >>>
> >>> This patch adds trace point for oom_score_adj. This adds 3 trace
> >>> points. at
> >>> 	- update oom_score_adj
> >>
> >>
> >>> 	- fork()
> >>> 	- rename task->comm(typically, exec())
> >>
> >> I don't think they have oom specific thing. Can you please add generic fork and
> >> task rename tracepoint instead?
> >>
> > I think it makes oom-targeted debug difficult.
> > This tracehook using task->signal->oom_score_adj as filter.
> > This reduces traces much and makes debugging easier.
> >  
> > If you need another trace point for other purpose, another trace point
> > should be better. For generic purpose, oom_socre_adj filtering will not
> > be necessary.
> 
> see Documentation/trace/event.txt 5. Event filgtering
> 
> Now, both ftrace and perf have good filter feature. Isn't this enough?
> 

Could you make patch ? Then, I stop this and go other probelm.

Thanks,
-Kame

Re: [PATCH] oom: add tracepoints for oom_score_adj

[KAMEZAWA Hiroyuki]

On Fri, 9 Dec 2011 08:41:03 +0900
KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote:

> On Thu, 08 Dec 2011 12:33:35 -0500
> KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> wrote:
 see Documentation/trace/event.txt 5. Event filgtering
> > 
> > Now, both ftrace and perf have good filter feature. Isn't this enough?
> > 
> 
> Could you make patch ? Then, I stop this and go other probelm.
> 

Hmm, core of patch should be like this. But need some works on

 - How to debug oom in Documenation especially for trace-cmd users.
 - Other trace points sutable for 'task' tracing. maybe 'exit, stop, freeze' ?
 - at creating new task, what other members should be printed out ?
 - At renaming...don't we need reason for renaming ?

Hm. BTW, do you know how to write filtering in
/etc/sysconfig/trace-cmd.config ?

Re: [PATCH] oom: add tracepoints for oom_score_adj

[KAMEZAWA Hiroyuki]

On Fri, 9 Dec 2011 08:41:03 +0900
KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote:

> On Thu, 08 Dec 2011 12:33:35 -0500
> KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> wrote:
 see Documentation/trace/event.txt 5. Event filgtering
> > 
> > Now, both ftrace and perf have good filter feature. Isn't this enough?
> > 
> 
> Could you make patch ? Then, I stop this and go other probelm.
> 

Hmm, core of patch should be like this. But need some works on

 - How to debug oom in Documenation especially for trace-cmd users.
 - Other trace points sutable for 'task' tracing. maybe 'exit, stop, freeze' ?
 - at creating new task, what other members should be printed out ?
 - At renaming...don't we need reason for renaming ?

Hm. BTW, do you know how to write filtering in
/etc/sysconfig/trace-cmd.config ?




>From 3e19265f25dff2c86829ae8afbcc19d02bcefae0 Mon Sep 17 00:00:00 2001
From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Date: Fri, 9 Dec 2011 09:44:23 +0900
Subject: [PATCH] trace point for oom debugging.

beta version.
---
 fs/exec.c                   |    4 +++
 fs/proc/base.c              |    3 ++
 include/trace/events/oom.h  |   35 ++++++++++++++++++++++++
 include/trace/events/task.h |   63 +++++++++++++++++++++++++++++++++++++++++++
 kernel/fork.c               |    6 ++++
 mm/oom_kill.c               |    6 ++++
 6 files changed, 117 insertions(+), 0 deletions(-)
 create mode 100644 include/trace/events/oom.h
 create mode 100644 include/trace/events/task.h

diff --git a/fs/exec.c b/fs/exec.c
index ca141db..fd0bfbd 100644
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -59,6 +59,8 @@
 #include <asm/uaccess.h>
 #include <asm/mmu_context.h>
 #include <asm/tlb.h>
+
+#include <trace/events/task.h>
 #include "internal.h"
 
 int core_uses_pid;
@@ -1054,6 +1056,8 @@ void set_task_comm(struct task_struct *tsk, char *buf)
 {
 	task_lock(tsk);
 
+	trace_task_rename(tsk, buf);
+
 	/*
 	 * Threads may access current->comm without holding
 	 * the task lock, so write the string carefully.
diff --git a/fs/proc/base.c b/fs/proc/base.c
index 1050b1c..f201e64 100644
--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -87,6 +87,7 @@
 #ifdef CONFIG_HARDWALL
 #include <asm/hardwall.h>
 #endif
+#include <trace/events/oom.h>
 #include "internal.h"
 
 /* NOTE:
@@ -1166,6 +1167,7 @@ static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 	else
 		task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
 								-OOM_DISABLE;
+	trace_oom_score_adj_update(task);
 err_sighand:
 	unlock_task_sighand(task, &flags);
 err_task_lock:
@@ -1253,6 +1255,7 @@ static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
 	task->signal->oom_score_adj = oom_score_adj;
 	if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
 		task->signal->oom_score_adj_min = oom_score_adj;
+	trace_oom_score_adj_update(task);
 	/*
 	 * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
 	 * always attainable.
diff --git a/include/trace/events/oom.h b/include/trace/events/oom.h
new file mode 100644
index 0000000..bb75e5c
--- /dev/null
+++ b/include/trace/events/oom.h
@@ -0,0 +1,35 @@
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM oom
+
+#if !defined(_TRACE_OOM_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_OOM_H
+#include <linux/tracepoint.h>
+
+TRACE_EVENT(oom_score_adj_update,
+
+	TP_PROTO(struct task_struct *task),
+
+	TP_ARGS(task),
+
+	TP_STRUCT__entry(
+		__field(	pid_t,	pid)
+		__array(	char,	comm,	TASK_COMM_LEN )
+		__field(	 int,	oom_score_adj)
+	),
+
+	TP_fast_assign(
+		__entry->pid = task->pid;
+		memcpy(__entry->comm, task->comm, TASK_COMM_LEN);
+		__entry->oom_score_adj = task->signal->oom_score_adj;
+	),
+
+	TP_printk("pid=%d comm=%s oom_score_adj=%d",
+		__entry->pid, __entry->comm, __entry->oom_score_adj)
+);
+
+#endif
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>
+
+
diff --git a/include/trace/events/task.h b/include/trace/events/task.h
new file mode 100644
index 0000000..df8f8f8
--- /dev/null
+++ b/include/trace/events/task.h
@@ -0,0 +1,63 @@
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM task
+
+#if !defined(_TRACE_TASK_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_TASK_H
+#include <linux/tracepoint.h>
+
+TRACE_EVENT(task_newtask,
+
+	TP_PROTO(struct task_struct *task, unsigned long clone_flags),
+
+	TP_ARGS(task, clone_flags),
+
+	TP_STRUCT__entry(
+		__field(	pid_t,	pid)
+		__array(	char,	comm, TASK_COMM_LEN)
+		__field( unsigned long, clone_flags)
+		__field(	int,    oom_score_adj)
+	),
+
+	TP_fast_assign(
+		__entry->pid = task->pid;
+		memcpy(__entry->comm, task->comm, TASK_COMM_LEN);
+		__entry->clone_flags = clone_flags;
+		__entry->oom_score_adj = task->signal->oom_score_adj;
+	),
+
+	TP_printk("pid=%d comm=%s clone_flags=%ld oom_score_adj=%d",
+		__entry->pid, __entry->comm,
+		__entry->clone_flags, __entry->oom_score_adj)
+);
+
+TRACE_EVENT(task_rename,
+
+	TP_PROTO(struct task_struct *task, char *comm),
+
+	TP_ARGS(task, comm),
+
+	TP_STRUCT__entry(
+		__field(	pid_t,	pid)
+		__array(	char, oldcomm,  TASK_COMM_LEN)
+		__array(	char, newcomm,  TASK_COMM_LEN)
+		__field(	int, oom_score_adj)
+	),
+
+	TP_fast_assign(
+		__entry->pid = task->pid;
+		memcpy(entry->oldcomm, task->comm, TASK_COMM_LEN);
+		memcpy(entry->newcomm, comm, TASK_COMM_LEN);
+		__entry->oom_score_adj = task->signal->oom_score_adj;
+	),
+
+	TP_printk("pid=%d oldcomm=%s newcomm=%s oom_score_adj=%d",
+		__entry->pid, __entry->oldcomm,
+		__entry->newcomm, __entry->oom_score_adj)
+);
+
+#endif
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>
+
+
diff --git a/kernel/fork.c b/kernel/fork.c
index e20518d..d93ef13 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -77,6 +77,9 @@
 
 #include <trace/events/sched.h>
 
+#define CREATE_TRACE_POINTS
+#include <trace/events/task.h>
+
 /*
  * Protected counters by write_lock_irq(&tasklist_lock)
  */
@@ -1390,6 +1393,9 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 	if (clone_flags & CLONE_THREAD)
 		threadgroup_fork_read_unlock(current);
 	perf_event_fork(p);
+
+	trace_task_newtask(p, clone_flags);
+
 	return p;
 
 bad_fork_free_pid:
diff --git a/mm/oom_kill.c b/mm/oom_kill.c
index e2e1402..46b6d0a 100644
--- a/mm/oom_kill.c
+++ b/mm/oom_kill.c
@@ -33,6 +33,10 @@
 #include <linux/security.h>
 #include <linux/ptrace.h>
 #include <linux/freezer.h>
+#include <linux/ftrace.h>
+
+#define CREATE_TRACE_POINTS
+#include <trace/events/oom.h>
 
 int sysctl_panic_on_oom;
 int sysctl_oom_kill_allocating_task;
@@ -55,6 +59,7 @@ void compare_swap_oom_score_adj(int old_val, int new_val)
 	spin_lock_irq(&sighand->siglock);
 	if (current->signal->oom_score_adj == old_val)
 		current->signal->oom_score_adj = new_val;
+	trace_oom_score_adj_update(current);
 	spin_unlock_irq(&sighand->siglock);
 }
 
@@ -74,6 +79,7 @@ int test_set_oom_score_adj(int new_val)
 	spin_lock_irq(&sighand->siglock);
 	old_val = current->signal->oom_score_adj;
 	current->signal->oom_score_adj = new_val;
+	trace_oom_score_adj_update(current);
 	spin_unlock_irq(&sighand->siglock);
 
 	return old_val;
-- 
1.7.4.1