Properly support the ru_maxrss field of the rusage structure returned by
getrusage(). This patch includes documentation both of the getrusage()
implementation in general and of the ru_maxrss implementation
specifically. This implementation matches that of FreeBSD, which is the
only other OS of which I'm aware that implements this field.
Like a number of other folks, we recently had a need for a non-/proc way
of getting the RSS of a process and happened on getrusage(). Unlike the
rest, though, we fixed the system call to do what we want. I wrote a
wrong implementation and submitted it while I was sick; this time I took
my time and I think I got it right.
A test program that has been run against a number of systems (of which
only FreeBSD and Linux 2.6 with this patch applied passed) is also
available at
http://www.exit.com/Archives/Linux/
Signed-off-by: Frank Mayhar <[email protected]>
---
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 3de7901..85735af 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -518,6 +518,7 @@ 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;
+ unsigned long cmaxrss;
/*
* Cumulative ns of scheduled CPU time for dead threads in the
@@ -1027,6 +1028,8 @@ #endif
struct timespec real_start_time; /* boot based time */
/* mm fault and swap info: this can arguably be seen as either mm-specific or thread-specific */
unsigned long min_flt, maj_flt;
+/* maxrss gets the hiwater_rss in do_exit() */
+ unsigned long maxrss;
cputime_t it_prof_expires, it_virt_expires;
unsigned long long it_sched_expires;
diff --git a/kernel/exit.c b/kernel/exit.c
index 06b24b3..390d834 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -127,6 +127,8 @@ static void __exit_signal(struct task_st
sig->inblock += task_io_get_inblock(tsk);
sig->oublock += task_io_get_oublock(tsk);
sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
+ if (tsk->maxrss > sig->cmaxrss)
+ sig->cmaxrss = tsk->maxrss;
sig = NULL; /* Marker for below. */
}
@@ -957,6 +959,14 @@ fastcall NORET_TYPE void do_exit(long co
if (tsk->mm) {
update_hiwater_rss(tsk->mm);
update_hiwater_vm(tsk->mm);
+ BUG_ON(tsk->maxrss != 0);
+ /*
+ * Store the hiwater_rss in a task field, since when we need
+ * it in __exit_signal() the mm structure is gone; we can't
+ * stuff it in the signal structure since then we would be
+ * racing with wait_task_zombie().
+ */
+ tsk->maxrss = tsk->mm->hiwater_rss;
}
group_dead = atomic_dec_and_test(&tsk->signal->live);
if (group_dead) {
@@ -1265,6 +1275,17 @@ static int wait_task_zombie(struct task_
psig->coublock +=
task_io_get_oublock(p) +
sig->oublock + sig->coublock;
+ /*
+ * Save the maximum RSS of this task and all its terminated,
+ * waited-for children. Don't accumulate the RSS since, one,
+ * other operating systems (FreeBSD, AIX) do it this way and,
+ * two, the cumulative RSS of a long-lived process with lots
+ * of sequential child process would be meaningless.
+ */
+ if (sig->cmaxrss > psig->cmaxrss)
+ psig->cmaxrss = sig->cmaxrss;
+ if (p->maxrss > psig->cmaxrss)
+ psig->cmaxrss = p->maxrss;
spin_unlock_irq(&p->parent->sighand->siglock);
}
diff --git a/kernel/fork.c b/kernel/fork.c
index 7332e23..fae76dd 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -536,6 +536,7 @@ static int copy_mm(unsigned long clone_f
tsk->min_flt = tsk->maj_flt = 0;
tsk->nvcsw = tsk->nivcsw = 0;
+ tsk->maxrss = 0;
tsk->mm = NULL;
tsk->active_mm = NULL;
@@ -881,6 +882,7 @@ static inline int copy_signal(unsigned l
sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
sig->sum_sched_runtime = 0;
+ sig->cmaxrss = 0;
INIT_LIST_HEAD(&sig->cpu_timers[0]);
INIT_LIST_HEAD(&sig->cpu_timers[1]);
INIT_LIST_HEAD(&sig->cpu_timers[2]);
diff --git a/kernel/sys.c b/kernel/sys.c
index 1b33b05..e450bab 100644
--- a/kernel/sys.c
+++ b/kernel/sys.c
@@ -2099,6 +2099,7 @@ static void k_getrusage(struct task_stru
r->ru_majflt = p->signal->cmaj_flt;
r->ru_inblock = p->signal->cinblock;
r->ru_oublock = p->signal->coublock;
+ r->ru_maxrss = p->signal->cmaxrss;
if (who == RUSAGE_CHILDREN)
break;
@@ -2124,11 +2125,15 @@ static void k_getrusage(struct task_stru
r->ru_oublock += task_io_get_oublock(t);
t = next_thread(t);
} while (t != p);
+ update_hiwater_rss(p->mm);
+ if (r->ru_maxrss < p->mm->hiwater_rss)
+ r->ru_maxrss = p->mm->hiwater_rss;
break;
default:
BUG();
}
+ r->ru_maxrss <<= PAGE_SHIFT - 10; /* Convert from pages to kilobytes. */
unlock_task_sighand(p, &flags);
rcu_read_unlock();
--
Frank Mayhar <[email protected]>
Google, Inc.
Frank Mayhar wrote:
> Properly support the ru_maxrss field of the rusage structure returned by
> getrusage().
> Signed-off-by: Frank Mayhar <[email protected]>
Acked-by: Rik van Riel <[email protected]>
--
All Rights Reversed
On Thu, 20 Sep 2007 10:57:19 -0700 Frank Mayhar <[email protected]> wrote:
> Properly support the ru_maxrss field of the rusage structure returned by
> getrusage(). This patch includes documentation both of the getrusage()
> implementation in general and of the ru_maxrss implementation
> specifically. This implementation matches that of FreeBSD, which is the
> only other OS of which I'm aware that implements this field.
>
> Like a number of other folks, we recently had a need for a non-/proc way
> of getting the RSS of a process and happened on getrusage(). Unlike the
> rest, though, we fixed the system call to do what we want. I wrote a
> wrong implementation and submitted it while I was sick; this time I took
> my time and I think I got it right.
>
> A test program that has been run against a number of systems (of which
> only FreeBSD and Linux 2.6 with this patch applied passed) is also
> available at
> http://www.exit.com/Archives/Linux/
>
Nice patch, but the Vaio always wins ;)
It boots OK, but when I ssh into the machine:
[ 56.310698] sonypi: ioport 0x1080 busy, using sony-laptop? if not use check_ioport=0
[ 56.314186] sonypi: failed to request ioports
[ 56.317362] sonypi: probe of sonypi failed with error -16
[ 57.017983] ipw2200: Radio Frequency Kill Switch is On:
[ 57.017985] Kill switch must be turned off for wireless networking to work.
[ 57.124331] ipw2200: Failed to send WEP_KEY: Aborted due to RF kill switch.
[ 57.242396] ipw2200: Failed to send WEP_KEY: Command timed out.
[ 60.343396] BUG: unable to handle kernel NULL pointer dereference at virtual address 0000009c
[ 60.343573] printing eip: c012ba9c *pde = 00000000
[ 60.343680] Oops: 0000 [#1] PREEMPT
[ 60.343763] last sysfs file: /devices/system/cpu/cpu0/cpufreq/scaling_setspeed
[ 60.343894] Modules linked in: ipw2200 sonypi ipv6 autofs4 hidp l2cap bluetooth sunrpc nf_conntrack_netbios_ns ipt_REJECT nf_conntrack_ipv4 xt_state nf_conntrack nfnetlink xt_tcpudp iptable_filter ip_tables x_tables acpi_cpufreq sbs sbshc nvram ohci1394 ieee1394 ehci_hcd uhci_hcd sg joydev snd_hda_intel snd_seq_dummy snd_seq_oss snd_seq_midi_event snd_seq snd_seq_device snd_pcm_oss snd_mixer_oss ieee80211 sr_mod cdrom snd_pcm ieee80211_crypt snd_timer i2c_i801 i2c_core snd pcspkr soundcore piix snd_page_alloc battery button ac generic ext3 jbd ide_disk ide_core
[ 60.345225]
[ 60.345258] Pid: 2503, comm: zsh Not tainted (2.6.23-rc7-mm1 #9)
[ 60.345367] EIP: 0060:[<c012ba9c>] EFLAGS: 00010046 CPU: 0
[ 60.345472] EIP is at getrusage+0x181/0x21f
[ 60.345551] EAX: 00000000 EBX: 00000000 ECX: c2370ec0 EDX: 00000000
[ 60.345665] ESI: c2370ec0 EDI: 00000002 EBP: c3cc3f1c ESP: c3cc3ec0
[ 60.345779] DS: 007b ES: 007b FS: 0000 GS: 0033 SS: 0068
[ 60.345878] Process zsh (pid: 2503, ti=c3cc2000 task=c3cc0e00 task.ti=c3cc2000)
[ 60.346009] last branch before last exception/interrupt
[ 60.346112] from c012ba9c (getrusage+0x181/0x21f)
[ 60.346210] to c0325f70 (__kprobes_text_start+0x0/0x8)
[ 60.346317] Stack: bfc698f4 00000000 00000000 00000000 00000000 00000000 00000000 00000000
[ 60.346539] 00000000 00000159 00000000 00000000 00000100 00000000 00000000 00000000
[ 60.346759] 00000000 00000001 00000001 00000296 00000000 c341ca40 c2370ec0 c3cc3fa0
[ 60.346981] Call Trace:
[ 60.347036] [<c0104f55>] show_trace_log_lvl+0x1a/0x2f
[ 60.347141] [<c0105007>] show_stack_log_lvl+0x9d/0xa5
[ 60.347245] [<c010513d>] show_registers+0x12e/0x26a
[ 60.347346] [<c010537f>] die+0x106/0x1ec
[ 60.347430] [<c0118b30>] do_page_fault+0x483/0x56b
[ 60.347529] [<c0325fe2>] error_code+0x6a/0x70
[ 60.347620] [<c012347c>] do_wait+0x5ae/0x9fa
[ 60.347712] [<c01238f8>] sys_wait4+0x30/0x32
[ 60.347933] [<c0103f92>] syscall_call+0x7/0xb
[ 60.348024] =======================
[ 60.348091] Code: 81 5c 0c 00 00 8b 91 60 0c 00 00 0f ac d0 09 01 45 d8 c1 ea 09 8b 89 20 01 00 00 81 e9 20 01 00 00 39 f1 75 94 8b 86 b4 00 00 00 <8b> 90 9c 00 00 00 03 90 98 00 00 00 39 90 a0 00 00 00 73 06 89
[ 60.348909] EIP: [<c012ba9c>] getrusage+0x181/0x21f SS:ESP 0068:c3cc3ec0
[ 60.349049] note: zsh[2503] exited with preempt_count 2
[ 60.349218] BUG: scheduling while atomic: zsh/2503/0x10000003
[ 60.349323] INFO: lockdep is turned off.
[ 60.349397] [<c0104f55>] show_trace_log_lvl+0x1a/0x2f
[ 60.349501] [<c01059db>] show_trace+0x12/0x14
[ 60.349592] [<c0105afa>] dump_stack+0x15/0x17
[ 60.354070] [<c011b257>] __schedule_bug+0x65/0x6c
[ 60.358647] [<c0322ede>] __sched_text_start+0x5e/0x30d
[ 60.363137] [<c011b27f>] __cond_resched+0x21/0x3b
[ 60.367704] [<c0323815>] cond_resched+0x26/0x31
[ 60.372158] [<c016237b>] unmap_vmas+0x36c/0x44a
[ 60.376555] [<c0164be8>] exit_mmap+0x68/0xf0
[ 60.380895] [<c011f106>] mmput+0x34/0x93
[ 60.385293] [<c0122763>] exit_mm+0xb7/0xbc
[ 60.389543] [<c0123b95>] do_exit+0x212/0x6ea
[ 60.393717] [<c010545d>] die+0x1e4/0x1ec
[ 60.397868] [<c0118b30>] do_page_fault+0x483/0x56b
[ 60.402058] [<c0325fe2>] error_code+0x6a/0x70
[ 60.406018] [<c012347c>] do_wait+0x5ae/0x9fa
[ 60.409780] [<c01238f8>] sys_wait4+0x30/0x32
[ 60.413375] [<c0103f92>] syscall_call+0x7/0xb
[ 60.416877] =======================
[ 60.524087] BUG: scheduling while atomic: zsh/2503/0x10000003
[ 60.527489] INFO: lockdep is turned off.
[ 60.530845] [<c0104f55>] show_trace_log_lvl+0x1a/0x2f
[ 60.534230] [<c01059db>] show_trace+0x12/0x14
[ 60.537543] [<c0105afa>] dump_stack+0x15/0x17
[ 60.540851] [<c011b257>] __schedule_bug+0x65/0x6c
[ 60.544142] [<c0322ede>] __sched_text_start+0x5e/0x30d
[ 60.547391] [<c011b27f>] __cond_resched+0x21/0x3b
[ 60.550589] [<c0323815>] cond_resched+0x26/0x31
[ 60.553772] [<c0122a6f>] put_files_struct+0x6b/0xa8
[ 60.556932] [<c0123bd6>] do_exit+0x253/0x6ea
[ 60.560039] [<c010545d>] die+0x1e4/0x1ec
[ 60.563147] [<c0118b30>] do_page_fault+0x483/0x56b
[ 60.566273] [<c0325fe2>] error_code+0x6a/0x70
[ 60.569318] [<c012347c>] do_wait+0x5ae/0x9fa
[ 60.572357] [<c01238f8>] sys_wait4+0x30/0x32
[ 60.575411] [<c0103f92>] syscall_call+0x7/0xb
[ 60.578461] =======================
Linux version 2.6.23-rc7-mm1 (akpm@box) (gcc version 4.1.0) #10 PREEMPT Mon Sep 24 00:51:00 PDT 2007
Distro is FC5, config is http://userweb.kernel.org/~akpm/config-sony.txt