distinct tgid/tid I/O statistics (was: taskstats and /proc/.../io asymmetry?)

PATCH In review Andrea Righi 2 participants 6 messages May 16, 2008 – May 19, 2008

Subsystems (5)

Linux-next tree MM core Scheduler exec and binfmt +1 more subsystems

lore

Share & tools

lore

Andrea Righi

author

2008-05-16 20:38 UTC #0 lore raw

PATCH 1/1 distinct tgid/tid I/O statistics (was: taskstats and /proc/.../io asymmetry?)

Quoted 16 lines · Balbir Singh

Quoted 15 lines · Mark Seger
If you look at /proc/pid/stat, you can get the total CPU consumed by a process. If you look at /proc/pid/task/tid/stat you can get the cpu consumed by a thread and if the tid is that of the parent you only gets its consumption as opposed to all its children.
I was surprised to see that the way process I/O is reported doesn't follow this model. There are no /prod/pid/task/tid/io entries but rather you need to look in /proc/tid/io. While I view this as a minor inconvenience, I can certainly live with it. However, /proc/pid/io does not show the aggregate I/O numbers for the whole process and that both surprises and disappoints. This means if I have a process that starts a bunch of worker threads that do the real work and I want to find the top I/O consumers I can't simply walk the /proc/pid tree but rather have to look at all the threads of each process and add them up.

Quoted 8 lines

Quoted 2 lines
Or am I missing something?
I looked through the code and your argument seems to be correct. The behaviour is inconsistent w.r.t. other statistics like utime and stime. We currently accumulate tgid information in signal_struct, we need to do something similar for io as well. If nobody gets to it by the time I finish my backlog, I'll try and get to it.

Balbir, Mark,

what do you think about the following approach?
Patch against 2.6.25.4, tested in KVM.

-Andrea

--
Report per-thread I/O statistics in /proc/pid/task/tid/io and aggregate parent I/O statistics in /proc/pid/io. This approach follows the same model used to account per-process and per-thread CPU times.

As a practial application, this allows for example to quickly find the top I/O consumer when a process spawns many child threads that perform the actual I/O work, because the aggregated I/O statistics can always be found in /proc/pid/io.

Bug reported by Mark Seger <[email protected]>.

Signed-off-by Andrea Righi <[email protected]>

---

fs/proc/base.c        |   83 +++++++++++++++++++++++++++++++++++++++++++------
include/linux/sched.h |    4 ++
kernel/exit.c         |   27 ++++++++++++++++
kernel/fork.c         |    6 +++
4 files changed, 110 insertions(+), 10 deletions(-)

fs/proc/base.c +0 -0

index 81d7d14..23763c5 100644
@@ -2255,8 +2255,58 @@ static int proc_base_fill_cache(struct file *filp, void *dirent,

}

#ifdef CONFIG_TASK_IO_ACCOUNTING
-static int proc_pid_io_accounting(struct task_struct *task, char *buffer) +static int do_io_accounting(struct task_struct *task, char *buffer, int whole) {
+ unsigned long flags;
+ u64 rchar, wchar, syscr, syscw;
+ struct task_io_accounting ioac;
+
+ if (!whole) {
+#ifdef CONFIG_TASK_XACCT
+ rchar = task->rchar;
+ wchar = task->wchar;
+ syscr = task->syscr;
+ syscw = task->syscw;
+#endif
+ memcpy(&ioac, &task->ioac, sizeof(ioac));
+ } else {
+ rchar = wchar = syscr = syscw = 0;
+ memset(&ioac, 0, sizeof(ioac));
+ rcu_read_lock();
+ if (lock_task_sighand(task, &flags)) {
+ struct signal_struct *sig = task->signal;
+ struct task_struct *t = task;
+
+ do {
+#ifdef CONFIG_TASK_XACCT
+ rchar += t->rchar;
+ wchar += t->wchar;
+ syscr += t->syscr;
+ syscw += t->syscw;
+#endif
+ ioac.read_bytes += t->ioac.read_bytes;
+ ioac.write_bytes += t->ioac.write_bytes;
+ ioac.cancelled_write_bytes +=
+ t->ioac.cancelled_write_bytes;
+ t = next_thread(t);
+ } while (t != task);
+
+#ifdef CONFIG_TASK_XACCT
+ rchar += sig->rchar;
+ wchar += sig->wchar;
+ syscr += sig->syscr;
+ syscw += sig->syscw;
+#endif
+ ioac.read_bytes += sig->ioac.read_bytes;
+ ioac.write_bytes += sig->ioac.write_bytes;
+ ioac.cancelled_write_bytes +=
+ sig->ioac.cancelled_write_bytes;
+
+ unlock_task_sighand(task, &flags);
+ }
+ rcu_read_unlock();
+ }
+
return sprintf(buffer,

#ifdef CONFIG_TASK_XACCT
			"rchar: %llu\n"
@@ -2268,16 +2318,26 @@ static int proc_pid_io_accounting(struct task_struct *task, char *buffer)
			"write_bytes: %llu\n"
			"cancelled_write_bytes: %llu\n",

#ifdef CONFIG_TASK_XACCT
- (unsigned long long)task->rchar,
- (unsigned long long)task->wchar,
- (unsigned long long)task->syscr,
- (unsigned long long)task->syscw,
+ (unsigned long long)rchar,
+ (unsigned long long)wchar,
+ (unsigned long long)syscr,
+ (unsigned long long)syscw,
#endif
- (unsigned long long)task->ioac.read_bytes,
- (unsigned long long)task->ioac.write_bytes,
- (unsigned long long)task->ioac.cancelled_write_bytes);
+ (unsigned long long)ioac.read_bytes,
+ (unsigned long long)ioac.write_bytes,
+ (unsigned long long)ioac.cancelled_write_bytes);
+}
+
+static int proc_tid_io_accounting(struct task_struct *task, char *buffer) +{
+ return do_io_accounting(task, buffer, 0);
}
-#endif
+
+static int proc_tgid_io_accounting(struct task_struct *task, char *buffer) +{
+ return do_io_accounting(task, buffer, 1);
+}
+#endif /* CONFIG_TASK_IO_ACCOUNTING */

/*
* Thread groups
@@ -2348,7 +2408,7 @@ static const struct pid_entry tgid_base_stuff[] = {
REG("coredump_filter", S_IRUGO|S_IWUSR, coredump_filter),
#endif
#ifdef CONFIG_TASK_IO_ACCOUNTING
- INF("io", S_IRUGO, pid_io_accounting),
+ INF("io", S_IRUGO, tgid_io_accounting),
#endif
};

@@ -2675,6 +2735,9 @@ static const struct pid_entry tid_base_stuff[] = {

#ifdef CONFIG_FAULT_INJECTION
	REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),

#endif
+#ifdef CONFIG_TASK_IO_ACCOUNTING
+ INF("io", S_IRUGO, tid_io_accounting),
+#endif
};

static int proc_tid_base_readdir(struct file * filp,

include/linux/sched.h +0 -0

index 6a1e7af..ebf8b45 100644
@@ -504,6 +504,10 @@ struct signal_struct {

	unsigned long nvcsw, nivcsw, cnvcsw, cnivcsw;
	unsigned long min_flt, maj_flt, cmin_flt, cmaj_flt;
	unsigned long inblock, oublock, cinblock, coublock;

+#ifdef CONFIG_TASK_XACCT
+ u64 rchar, wchar, syscr, syscw;
+#endif
+ struct task_io_accounting ioac;

	/*
	 * Cumulative ns of scheduled CPU time for dead threads in the

kernel/exit.c +0 -0

index 073005b..1a35583 100644
@@ -114,6 +114,18 @@ static void __exit_signal(struct task_struct *tsk)

		sig->nivcsw += tsk->nivcsw;
		sig->inblock += task_io_get_inblock(tsk);
		sig->oublock += task_io_get_oublock(tsk);

+#ifdef CONFIG_TASK_XACCT
+ sig->rchar += tsk->rchar;
+ sig->wchar += tsk->wchar;
+ sig->syscr += tsk->syscr;
+ sig->syscw += tsk->syscw;
+#endif /* CONFIG_TASK_XACCT */
+#ifdef CONFIG_TASK_IO_ACCOUNTING
+ sig->ioac.read_bytes += tsk->ioac.read_bytes;
+ sig->ioac.write_bytes += tsk->ioac.write_bytes;
+ sig->ioac.cancelled_write_bytes +=
+ tsk->ioac.cancelled_write_bytes;
+#endif /* CONFIG_TASK_IO_ACCOUNTING */

		sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
		sig = NULL; /* Marker for below. */
	}
@@ -1250,6 +1262,21 @@ static int wait_task_zombie(struct task_struct *p, int noreap,
		psig->coublock +=
			task_io_get_oublock(p) +
			sig->oublock + sig->coublock;

+#ifdef CONFIG_TASK_XACCT
+ psig->rchar += p->rchar + sig->rchar;
+ psig->wchar += p->wchar + sig->wchar;
+ psig->syscr += p->syscr + sig->syscr;
+ psig->syscw += p->syscw + sig->syscw;
+#endif /* CONFIG_TASK_XACCT */
+#ifdef CONFIG_TASK_IO_ACCOUNTING
+ psig->ioac.read_bytes +=
+ p->ioac.read_bytes + sig->ioac.read_bytes;
+ psig->ioac.write_bytes +=
+ p->ioac.write_bytes + sig->ioac.write_bytes;
+ psig->ioac.cancelled_write_bytes +=
+ p->ioac.cancelled_write_bytes +
+ sig->ioac.cancelled_write_bytes;
+#endif /* CONFIG_TASK_IO_ACCOUNTING */

		spin_unlock_irq(&p->parent->sighand->siglock);
	}

kernel/fork.c +0 -0

index 9c042f9..b8f408e 100644
@@ -924,6 +924,12 @@ static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)

	sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
	sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
	sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;

+#ifdef CONFIG_TASK_XACCT
+ sig->rchar = sig->wchar = sig->syscr = sig->syscw = 0;
+#endif
+#ifdef CONFIG_TASK_IO_ACCOUNTING
+ memset(&sig->ioac, 0, sizeof(sig->ioac));
+#endif

	sig->sum_sched_runtime = 0;
	INIT_LIST_HEAD(&sig->cpu_timers[0]);
	INIT_LIST_HEAD(&sig->cpu_timers[1]);

Signature · 1 line

1.5.4.3

Andrea Righi

author

2008-05-19 16:36 UTC #1 lore raw

Quoted 1 line · Andrea Righi

Patch against 2.6.25.4, tested in KVM.

Tested in latest linux-next (next-20080519) as well.

-Andrea

Balbir Singh

Acked-by on v1 1/1

2008-05-19 17:50 UTC #2 lore raw

In reply to Andrea Righi

Quoted 30 lines · Andrea Righi

Quoted 19 lines · Balbir Singh
Quoted 15 lines · Mark Seger
If you look at /proc/pid/stat, you can get the total CPU consumed by a process. If you look at /proc/pid/task/tid/stat you can get the cpu consumed by a thread and if the tid is that of the parent you only gets its consumption as opposed to all its children.
I was surprised to see that the way process I/O is reported doesn't follow this model. There are no /prod/pid/task/tid/io entries but rather you need to look in /proc/tid/io. While I view this as a minor inconvenience, I can certainly live with it. However, /proc/pid/io does not show the aggregate I/O numbers for the whole process and that both surprises and disappoints. This means if I have a process that starts a bunch of worker threads that do the real work and I want to find the top I/O consumers I can't simply walk the /proc/pid tree but rather have to look at all the threads of each process and add them up.
Quoted 2 lines
Or am I missing something?
Quoted 9 lines
I looked through the code and your argument seems to be correct. The behaviour
is inconsistent w.r.t. other statistics like utime and stime. We currently
accumulate tgid information in signal_struct, we need to do something similar
for io as well. If nobody gets to it by the time I finish my backlog, I'll try
and get to it.

Quoted 19 lines

Balbir, Mark,

what do you think about the following approach?
Patch against 2.6.25.4, tested in KVM.

-Andrea

Signature · 10 lines

Report per-thread I/O statistics in /proc/pid/task/tid/io and aggregate
parent I/O statistics in /proc/pid/io. This approach follows the same
model used to account per-process and per-thread CPU times.

As a practial application, this allows for example to quickly find the
top I/O consumer when a process spawns many child threads that perform
the actual I/O work, because the aggregated I/O statistics can always be
found in /proc/pid/io.

Bug reported by Mark Seger <[email protected]>.

Comments below. Overall looks pretty well covered. Have you tested the patches? Mark could you please test these and see?

Acked-by Balbir Singh <[email protected]>

Quoted 18 lines

Signed-off-by Andrea Righi <[email protected]>

---
fs/proc/base.c | 83
+++++++++++++++++++++++++++++++++++++++++++------

include/linux/sched.h |    4 ++
kernel/exit.c         |   27 ++++++++++++++++
kernel/fork.c         |    6 +++
4 files changed, 110 insertions(+), 10 deletions(-)

fs/proc/base.c +0 -0

index 81d7d14..23763c5 100644
@@ -2255,8 +2255,58 @@ static int proc_base_fill_cache(struct file
 *filp, void *dirent,
 }
 
 #ifdef CONFIG_TASK_IO_ACCOUNTING

fs/proc/base.c static int proc_base_fill_cache(struct file · line 2267

Quoted 10 lines

fs/proc/base.c +7 -1

-static int proc_pid_io_accounting(struct task_struct *task, char *buffer)
+static int do_io_accounting(struct task_struct *task, char *buffer, int
 whole)
 {
+    unsigned long flags;
+    u64 rchar, wchar, syscr, syscw;
+    struct task_io_accounting ioac;
+
+    if (!whole) {
+#ifdef CONFIG_TASK_XACCT

Can't we abstract this into a function for ease?

fs/proc/base.c static int proc_base_fill_cache(struct file · line 2278

Quoted 11 lines

fs/proc/base.c +11 -0

+        rchar = task->rchar;
+        wchar = task->wchar;
+        syscr = task->syscr;
+        syscw = task->syscw;
+#endif
+        memcpy(&ioac, &task->ioac, sizeof(ioac));
+    } else {
+        rchar = wchar = syscr = syscw = 0;
+        memset(&ioac, 0, sizeof(ioac));
+        rcu_read_lock();
+        if (lock_task_sighand(task, &flags)) {

If lock(), what happens otherwise?

Quoted 170 lines

fs/proc/base.c +33 -0

+        struct signal_struct *sig = task->signal;
+        struct task_struct *t = task;
+
+            do {
+#ifdef CONFIG_TASK_XACCT
+                rchar += t->rchar;
+                wchar += t->wchar;
+                syscr += t->syscr;
+                syscw += t->syscw;
+#endif
+                ioac.read_bytes += t->ioac.read_bytes;
+                ioac.write_bytes += t->ioac.write_bytes;
+                ioac.cancelled_write_bytes +=
+                        t->ioac.cancelled_write_bytes;
+                t = next_thread(t);
+            } while (t != task);
+
+#ifdef CONFIG_TASK_XACCT
+            rchar += sig->rchar;
+            wchar += sig->wchar;
+            syscr += sig->syscr;
+            syscw += sig->syscw;
+#endif
+            ioac.read_bytes += sig->ioac.read_bytes;
+            ioac.write_bytes += sig->ioac.write_bytes;
+            ioac.cancelled_write_bytes +=
+                    sig->ioac.cancelled_write_bytes;
+
+            unlock_task_sighand(task, &flags);
+        }
+        rcu_read_unlock();
+    }
+
    return sprintf(buffer,

#ifdef CONFIG_TASK_XACCT
            "rchar: %llu\n"

fs/proc/base.c +18 -8

@@ -2268,16 +2318,26 @@ static int proc_pid_io_accounting(struct
 task_struct *task, char *buffer)
            "write_bytes: %llu\n"
            "cancelled_write_bytes: %llu\n",
 #ifdef CONFIG_TASK_XACCT
-            (unsigned long long)task->rchar,
-            (unsigned long long)task->wchar,
-            (unsigned long long)task->syscr,
-            (unsigned long long)task->syscw,
+            (unsigned long long)rchar,
+            (unsigned long long)wchar,
+            (unsigned long long)syscr,
+            (unsigned long long)syscw,
 #endif
-            (unsigned long long)task->ioac.read_bytes,
-            (unsigned long long)task->ioac.write_bytes,
-            (unsigned long long)task->ioac.cancelled_write_bytes);
+            (unsigned long long)ioac.read_bytes,
+            (unsigned long long)ioac.write_bytes,
+            (unsigned long long)ioac.cancelled_write_bytes);
+}
+
+static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
+{
+    return do_io_accounting(task, buffer, 0);
 }
-#endif
+
+static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
+{
+    return do_io_accounting(task, buffer, 1);
+}
+#endif /* CONFIG_TASK_IO_ACCOUNTING */
 
 /*

* Thread groups

fs/proc/base.c +4 -1

@@ -2348,7 +2408,7 @@ static const struct pid_entry tgid_base_stuff[] = {
    REG("coredump_filter", S_IRUGO|S_IWUSR, coredump_filter),
 #endif
 #ifdef CONFIG_TASK_IO_ACCOUNTING
-    INF("io",    S_IRUGO, pid_io_accounting),
+    INF("io",    S_IRUGO, tgid_io_accounting),
 #endif
 };

@@ -2675,6 +2735,9 @@ static const struct pid_entry tid_base_stuff[] = {
 #ifdef CONFIG_FAULT_INJECTION
    REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),
 #endif
+#ifdef CONFIG_TASK_IO_ACCOUNTING
+    INF("io",    S_IRUGO, tid_io_accounting),
+#endif
 };
 
 static int proc_tid_base_readdir(struct file * filp,

include/linux/sched.h +4 -0

index 6a1e7af..ebf8b45 100644
@@ -504,6 +504,10 @@ struct signal_struct {
    unsigned long nvcsw, nivcsw, cnvcsw, cnivcsw;
    unsigned long min_flt, maj_flt, cmin_flt, cmaj_flt;
    unsigned long inblock, oublock, cinblock, coublock;
+#ifdef CONFIG_TASK_XACCT
+    u64 rchar, wchar, syscr, syscw;
+#endif
+    struct task_io_accounting ioac;
 
    /*
     * Cumulative ns of scheduled CPU time for dead threads in the

kernel/exit.c +27 -0

index 073005b..1a35583 100644
@@ -114,6 +114,18 @@ static void __exit_signal(struct task_struct *tsk)
        sig->nivcsw += tsk->nivcsw;
        sig->inblock += task_io_get_inblock(tsk);
        sig->oublock += task_io_get_oublock(tsk);
+#ifdef CONFIG_TASK_XACCT
+        sig->rchar += tsk->rchar;
+        sig->wchar += tsk->wchar;
+        sig->syscr += tsk->syscr;
+        sig->syscw += tsk->syscw;
+#endif /* CONFIG_TASK_XACCT */
+#ifdef CONFIG_TASK_IO_ACCOUNTING
+        sig->ioac.read_bytes += tsk->ioac.read_bytes;
+        sig->ioac.write_bytes += tsk->ioac.write_bytes;
+        sig->ioac.cancelled_write_bytes +=
+                    tsk->ioac.cancelled_write_bytes;
+#endif /* CONFIG_TASK_IO_ACCOUNTING */
        sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
        sig = NULL; /* Marker for below. */
    }

@@ -1250,6 +1262,21 @@ static int wait_task_zombie(struct task_struct
 *p, int noreap,
        psig->coublock +=
            task_io_get_oublock(p) +
            sig->oublock + sig->coublock;
+#ifdef CONFIG_TASK_XACCT
+        psig->rchar += p->rchar + sig->rchar;
+        psig->wchar += p->wchar + sig->wchar;
+        psig->syscr += p->syscr + sig->syscr;
+        psig->syscw += p->syscw + sig->syscw;
+#endif /* CONFIG_TASK_XACCT */
+#ifdef CONFIG_TASK_IO_ACCOUNTING
+        psig->ioac.read_bytes +=
+            p->ioac.read_bytes + sig->ioac.read_bytes;
+        psig->ioac.write_bytes +=
+            p->ioac.write_bytes + sig->ioac.write_bytes;
+        psig->ioac.cancelled_write_bytes +=
+                p->ioac.cancelled_write_bytes +
+                sig->ioac.cancelled_write_bytes;
+#endif /* CONFIG_TASK_IO_ACCOUNTING */
        spin_unlock_irq(&p->parent->sighand->siglock);
    }

kernel/fork.c +6 -0

index 9c042f9..b8f408e 100644
@@ -924,6 +924,12 @@ static int copy_signal(unsigned long clone_flags,
 struct task_struct *tsk)
    sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
    sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
    sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
+#ifdef CONFIG_TASK_XACCT
+    sig->rchar = sig->wchar = sig->syscr = sig->syscw = 0;
+#endif
+#ifdef CONFIG_TASK_IO_ACCOUNTING
+    memset(&sig->ioac, 0, sizeof(sig->ioac));
+#endif
    sig->sum_sched_runtime = 0;
    INIT_LIST_HEAD(&sig->cpu_timers[0]);

    INIT_LIST_HEAD(&sig->cpu_timers[1]);

Signature · 4 lines

	Warm Regards,
	Balbir Singh
	Linux Technology Center
	IBM, ISTL

Andrea Righi

author

2008-05-19 18:33 UTC #3 lore raw

Re: [PATCH] distinct tgid/tid I/O statistics

Quoted 5 lines · Balbir Singh

Review
Quoted 3 lines
(quoted diff) +3 -0
+
+    if (!whole) {
+#ifdef CONFIG_TASK_XACCT
Can't we abstract this into a function for ease?

Maybe:

#ifdef CONFIG_TASK_XACCT
#define _task_xacct 1
#else
#define _task_xacct 0
#endif

and:

rchar = __task_xacct ? task->rchar : 0;
wchar = __task_xacct ? task->wchar : 0;
syscr = __task_xacct ? task->syscr : 0;
syscw = __task_xacct ? task->syscw : 0;

But it's very ugly anyway. It seems a similar problem has been raised here:

http://lkml.org/lkml/2008/5/17/154

Apparently with no solution.

Quoted 13 lines

Review

Quoted 12 lines

(quoted diff) +11 -0

+        rchar = task->rchar;
+        wchar = task->wchar;
+        syscr = task->syscr;
+        syscw = task->syscw;
+#endif
+        memcpy(&ioac, &task->ioac, sizeof(ioac));
+    } else {
+        rchar = wchar = syscr = syscw = 0;
+        memset(&ioac, 0, sizeof(ioac));
+        rcu_read_lock();
+        if (lock_task_sighand(task, &flags)) {

Quoted 2 lines

If lock(), what happens otherwise?

mmmh... we must surely move the threads stats accounting code out of the if statement. OK, I'll post a new patch soon.

Thanks,
-Andrea

Quoted 32 lines

(quoted diff) +32 -0

+        struct signal_struct *sig = task->signal;
+        struct task_struct *t = task;
+
+            do {
+#ifdef CONFIG_TASK_XACCT
+                rchar += t->rchar;
+                wchar += t->wchar;
+                syscr += t->syscr;
+                syscw += t->syscw;
+#endif
+                ioac.read_bytes += t->ioac.read_bytes;
+                ioac.write_bytes += t->ioac.write_bytes;
+                ioac.cancelled_write_bytes +=
+                        t->ioac.cancelled_write_bytes;
+                t = next_thread(t);
+            } while (t != task);
+
+#ifdef CONFIG_TASK_XACCT
+            rchar += sig->rchar;
+            wchar += sig->wchar;
+            syscr += sig->syscr;
+            syscw += sig->syscw;
+#endif
+            ioac.read_bytes += sig->ioac.read_bytes;
+            ioac.write_bytes += sig->ioac.write_bytes;
+            ioac.cancelled_write_bytes +=
+                    sig->ioac.cancelled_write_bytes;
+
+            unlock_task_sighand(task, &flags);
+        }
+        rcu_read_unlock();
+    }

Andrea Righi

author

2008-05-19 20:09 UTC #4 lore raw

Re: [PATCH] distinct tgid/tid I/O statistics

Quoted 6 lines · Andrea Righi

Quoted 4 lines · Balbir Singh
Review
Quoted 3 lines
(quoted diff) +3 -0
+
+    if (!whole) {
+#ifdef CONFIG_TASK_XACCT
Can't we abstract this into a function for ease?

BTW, CONFIG_TASK_IO_ACCOUNTING depends on CONFIG_TASK_XACCT, so I think we can just remove those #ifdefs inside CONFIG_TASK_IO_ACCOUNTING.

-Andrea

Andrea Righi

author

2008-05-19 21:58 UTC #5 lore raw

PATCH v2 1/1 distinct tgid/tid I/O statistics

Changes based on Balbir Singh feedback:
- fix: correcly account children threads i/o activity
- removed CONFIG_TASK_XACCT #ifdefs inside CONFIG_TASK_IO_ACCOUNTING

Tested in latest Linus git.
---
Subject: [PATCH] distinct tgid/tid I/O statistics
From: Andrea Righi <[email protected]>

Report per-thread I/O statistics in /proc/pid/task/tid/io and aggregate parent I/O statistics in /proc/pid/io. This approach follows the same model used to account per-process and per-thread CPU times.

Bug reported by Mark Seger <[email protected]>.

Signed-off-by Andrea Righi <[email protected]>

---

fs/proc/base.c        |   84 ++++++++++++++++++++++++++++++++++++++++--------
include/linux/sched.h |    4 ++
kernel/exit.c         |   27 ++++++++++++++++
kernel/fork.c         |    6 +++
4 files changed, 107 insertions(+), 14 deletions(-)

fs/proc/base.c +0 -0

index 808cbdc..b905c6d 100644
@@ -2356,29 +2356,82 @@ static int proc_base_fill_cache(struct file *filp, void *dirent,

}

#ifdef CONFIG_TASK_IO_ACCOUNTING
-static int proc_pid_io_accounting(struct task_struct *task, char *buffer) +static int do_io_accounting(struct task_struct *task, char *buffer, int whole) {
+ unsigned long flags;
+ u64 rchar, wchar, syscr, syscw;
+ struct task_io_accounting ioac;
+
+ if (!whole) {
+ rchar = task->rchar;
+ wchar = task->wchar;
+ syscr = task->syscr;
+ syscw = task->syscw;
+ memcpy(&ioac, &task->ioac, sizeof(ioac));
+ } else {
+ struct task_struct *t = task;
+ rchar = wchar = syscr = syscw = 0;
+ memset(&ioac, 0, sizeof(ioac));
+
+ rcu_read_lock();
+ do {
+ rchar += t->rchar;
+ wchar += t->wchar;
+ syscr += t->syscr;
+ syscw += t->syscw;
+
+ ioac.read_bytes += t->ioac.read_bytes;
+ ioac.write_bytes += t->ioac.write_bytes;
+ ioac.cancelled_write_bytes +=
+ t->ioac.cancelled_write_bytes;
+ t = next_thread(t);
+ } while (t != task);
+
+ if (lock_task_sighand(task, &flags)) {
+ struct signal_struct *sig = task->signal;
+
+ rchar += sig->rchar;
+ wchar += sig->wchar;
+ syscr += sig->syscr;
+ syscw += sig->syscw;
+
+ ioac.read_bytes += sig->ioac.read_bytes;
+ ioac.write_bytes += sig->ioac.write_bytes;
+ ioac.cancelled_write_bytes +=
+ sig->ioac.cancelled_write_bytes;
+
+ unlock_task_sighand(task, &flags);
+ }
+ rcu_read_unlock();
+ }
+
return sprintf(buffer,
-#ifdef CONFIG_TASK_XACCT

			"rchar: %llu\n"
			"wchar: %llu\n"
			"syscr: %llu\n"
			"syscw: %llu\n"
-#endif
			"read_bytes: %llu\n"
			"write_bytes: %llu\n"
			"cancelled_write_bytes: %llu\n",

-#ifdef CONFIG_TASK_XACCT
- (unsigned long long)task->rchar,
- (unsigned long long)task->wchar,
- (unsigned long long)task->syscr,
- (unsigned long long)task->syscw,
-#endif
- (unsigned long long)task->ioac.read_bytes,
- (unsigned long long)task->ioac.write_bytes,
- (unsigned long long)task->ioac.cancelled_write_bytes);
+ (unsigned long long)rchar,
+ (unsigned long long)wchar,
+ (unsigned long long)syscr,
+ (unsigned long long)syscw,
+ (unsigned long long)ioac.read_bytes,
+ (unsigned long long)ioac.write_bytes,
+ (unsigned long long)ioac.cancelled_write_bytes);
}
-#endif
+
+static int proc_tid_io_accounting(struct task_struct *task, char *buffer) +{
+ return do_io_accounting(task, buffer, 0);
+}
+
+static int proc_tgid_io_accounting(struct task_struct *task, char *buffer) +{
+ return do_io_accounting(task, buffer, 1);
+}
+#endif /* CONFIG_TASK_IO_ACCOUNTING */

/*
* Thread groups
@@ -2450,7 +2503,7 @@ static const struct pid_entry tgid_base_stuff[] = {
REG("coredump_filter", S_IRUGO|S_IWUSR, coredump_filter),
#endif
#ifdef CONFIG_TASK_IO_ACCOUNTING
- INF("io", S_IRUGO, pid_io_accounting),
+ INF("io", S_IRUGO, tgid_io_accounting),
#endif
};

@@ -2778,6 +2831,9 @@ static const struct pid_entry tid_base_stuff[] = {

#ifdef CONFIG_FAULT_INJECTION
	REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject),

#endif
+#ifdef CONFIG_TASK_IO_ACCOUNTING
+ INF("io", S_IRUGO, tid_io_accounting),
+#endif
};

static int proc_tid_base_readdir(struct file * filp,

include/linux/sched.h +0 -0

index 5395a61..d4d9adf 100644
@@ -504,6 +504,10 @@ struct signal_struct {

	unsigned long nvcsw, nivcsw, cnvcsw, cnivcsw;
	unsigned long min_flt, maj_flt, cmin_flt, cmaj_flt;
	unsigned long inblock, oublock, cinblock, coublock;

+#ifdef CONFIG_TASK_XACCT
+ u64 rchar, wchar, syscr, syscw;
+#endif
+ struct task_io_accounting ioac;

	/*
	 * Cumulative ns of scheduled CPU time for dead threads in the

kernel/exit.c +0 -0

index 1510f78..1f3c0ec 100644
@@ -120,6 +120,18 @@ static void __exit_signal(struct task_struct *tsk)

		sig->nivcsw += tsk->nivcsw;
		sig->inblock += task_io_get_inblock(tsk);
		sig->oublock += task_io_get_oublock(tsk);

		sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
		sig = NULL; /* Marker for below. */
	}
@@ -1321,6 +1333,21 @@ static int wait_task_zombie(struct task_struct *p, int noreap,
		psig->coublock +=
			task_io_get_oublock(p) +
			sig->oublock + sig->coublock;

		spin_unlock_irq(&p->parent->sighand->siglock);
	}

kernel/fork.c +0 -0

index 933e60e..bad9981 100644
@@ -915,6 +915,12 @@ static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)

	sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
	sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
	sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;

+#ifdef CONFIG_TASK_XACCT
+ sig->rchar = sig->wchar = sig->syscr = sig->syscw = 0;
+#endif
+#ifdef CONFIG_TASK_IO_ACCOUNTING
+ memset(&sig->ioac, 0, sizeof(sig->ioac));
+#endif

	sig->sum_sched_runtime = 0;
	INIT_LIST_HEAD(&sig->cpu_timers[0]);
	INIT_LIST_HEAD(&sig->cpu_timers[1]);