Re: [PATCH] eliminate nswap and cnswap

Re: [PATCH] eliminate nswap and cnswap

[Albert Cahalan]

> The nswap and cnswap variables counters have never
> been incremented as Linux doesn't do task swapping.

I'm pretty sure they were used for paging activity.
We don't eliminate support for "swap space", do we?

Somebody must have broken nswap and cnswap while
hacking on some vm code. I hate to see the variables
get completely ripped out of the kernel instead of
getting fixed.

Re: [PATCH] eliminate nswap and cnswap

[Andrew Morton]

Albert Cahalan <albert@users.sourceforge.net> wrote:
>
> > The nswap and cnswap variables counters have never
> > been incremented as Linux doesn't do task swapping.
> 
> I'm pretty sure they were used for paging activity.
> We don't eliminate support for "swap space", do we?
> 
> Somebody must have broken nswap and cnswap while
> hacking on some vm code. I hate to see the variables
> get completely ripped out of the kernel instead of
> getting fixed.

There's nothing in 2.4 which increments these, nor was there in 2.6.  Which
tends to imply that they weren't very important.

We could sort-of do this - move them into mm_struct (doing it in
task_struct was always wrong) and increment them in the VM.  But we'd need
some reason why these statistics are interesting, and we'd need an
explanation of what nswap and cnswap are actually supposed to represent.  

Re: [PATCH] eliminate nswap and cnswap

[Matt Mackall]

On Mon, Apr 12, 2004 at 08:42:23PM -0700, Andrew Morton wrote:
> Albert Cahalan <albert@users.sourceforge.net> wrote:
> >
> > > The nswap and cnswap variables counters have never
> > > been incremented as Linux doesn't do task swapping.
> > 
> > I'm pretty sure they were used for paging activity.
> > We don't eliminate support for "swap space", do we?
> > 
> > Somebody must have broken nswap and cnswap while
> > hacking on some vm code. I hate to see the variables
> > get completely ripped out of the kernel instead of
> > getting fixed.
> 
> There's nothing in 2.4 which increments these, nor was there in 2.6.  Which
> tends to imply that they weren't very important.
> 
> We could sort-of do this - move them into mm_struct (doing it in
> task_struct was always wrong) and increment them in the VM.  But we'd need
> some reason why these statistics are interesting, and we'd need an
> explanation of what nswap and cnswap are actually supposed to represent.  

It's used in 2.0 and 2.2 to satisfy the getrusage(2) syscall. The
getrusage page only documents it as 'swaps', but says:

       The above struct was taken from BSD 4.3 Reno. Not all fields
       are meaningful under Linux. Right now (Linux 2.4) only the
       fields ru_utime, ru_stime, ru_minflt, ru_majflt, and ru_nswap
       are maintained.

the BSD manpage says:

ru_nswap
    the number of times a process was swapped out of main memory.

Which means it's a count of full process swaps like ancestral UNIX,
which Linux has never had.

So the 2.0/2.2 attempt to increment said variables was actually bogus.

-- 
Matt Mackall : http://www.selenic.com : Linux development and consulting

Re: [PATCH] eliminate nswap and cnswap

[Albert Cahalan]

On Tue, 2004-04-13 at 00:27, Matt Mackall wrote:
> On Mon, Apr 12, 2004 at 08:42:23PM -0700, Andrew Morton wrote:
> > Albert Cahalan <albert@users.sourceforge.net> wrote:
> > >
> > > > The nswap and cnswap variables counters have never
> > > > been incremented as Linux doesn't do task swapping.
> > > 
> > > I'm pretty sure they were used for paging activity.
> > > We don't eliminate support for "swap space", do we?
> > > 
> > > Somebody must have broken nswap and cnswap while
> > > hacking on some vm code. I hate to see the variables
> > > get completely ripped out of the kernel instead of
> > > getting fixed.
> > 
> > There's nothing in 2.4 which increments these, nor was there in 2.6.  Which
> > tends to imply that they weren't very important.
> > 
> > We could sort-of do this - move them into mm_struct (doing it in
> > task_struct was always wrong) and increment them in the VM.  But we'd need
> > some reason why these statistics are interesting, and we'd need an
> > explanation of what nswap and cnswap are actually supposed to represent.  
> 
> It's used in 2.0 and 2.2 to satisfy the getrusage(2) syscall. The
> getrusage page only documents it as 'swaps', but says:
> 
>        The above struct was taken from BSD 4.3 Reno. Not all fields
>        are meaningful under Linux. Right now (Linux 2.4) only the
>        fields ru_utime, ru_stime, ru_minflt, ru_majflt, and ru_nswap
>        are maintained.
> 
> the BSD manpage says:
> 
> ru_nswap
>     the number of times a process was swapped out of main memory.
> 
> Which means it's a count of full process swaps like ancestral UNIX,
> which Linux has never had.

Well, that's the BSD man page.

I've always considered "swapped" to mean "paged"
when dealing with Linux. We still have our swap
partitions and swapon command, not paging partitions
and pageon command.

I suppose "paged out the area with the stack pointer"
would be a good approximation to the BSD behavior.

> So the 2.0/2.2 attempt to increment said variables was actually bogus.

If the variables counted paging, I consider them
correct for Linux.

Re: [PATCH] eliminate nswap and cnswap

[Bill Davidsen]

Albert Cahalan wrote:
>>The nswap and cnswap variables counters have never
>>been incremented as Linux doesn't do task swapping.
> 
> 
> I'm pretty sure they were used for paging activity.
> We don't eliminate support for "swap space", do we?
> 
> Somebody must have broken nswap and cnswap while
> hacking on some vm code. I hate to see the variables
> get completely ripped out of the kernel instead of
> getting fixed.

Since Linux doesn't swap, "fixed" would mean returning zero for these 
values. I don't thing even BSD swaps anymore, does it? In any case, 
Linux never did.

-- 
    -bill davidsen (davidsen@tmr.com)
"The secret to procrastination is to put things off until the
  last possible moment - but no longer"  -me

[PATCH] eliminate nswap and cnswap

[Matt Mackall]

The nswap and cnswap variables counters have never been incremented as
Linux doesn't do task swapping.

 tiny-mpm/arch/alpha/kernel/osf_sys.c |    3 ---
 tiny-mpm/fs/proc/array.c             |    4 ++--
 tiny-mpm/include/linux/sched.h       |    2 +-
 tiny-mpm/kernel/acct.c               |    2 +-
 tiny-mpm/kernel/exit.c               |    1 -
 tiny-mpm/kernel/fork.c               |    1 -
 tiny-mpm/kernel/sys.c                |    3 ---
 7 files changed, 4 insertions(+), 12 deletions(-)

diff -puN arch/alpha/kernel/osf_sys.c~kill-nswap arch/alpha/kernel/osf_sys.c
--- tiny/arch/alpha/kernel/osf_sys.c~kill-nswap	2004-03-20 12:14:34.000000000 -0600
+++ tiny-mpm/arch/alpha/kernel/osf_sys.c	2004-03-20 12:14:34.000000000 -0600
@@ -1095,14 +1095,12 @@ osf_getrusage(int who, struct rusage32 *
 		jiffies_to_timeval32(current->stime, &r.ru_stime);
 		r.ru_minflt = current->min_flt;
 		r.ru_majflt = current->maj_flt;
-		r.ru_nswap = current->nswap;
 		break;
 	case RUSAGE_CHILDREN:
 		jiffies_to_timeval32(current->cutime, &r.ru_utime);
 		jiffies_to_timeval32(current->cstime, &r.ru_stime);
 		r.ru_minflt = current->cmin_flt;
 		r.ru_majflt = current->cmaj_flt;
-		r.ru_nswap = current->cnswap;
 		break;
 	default:
 		jiffies_to_timeval32(current->utime + current->cutime,
@@ -1111,7 +1109,6 @@ osf_getrusage(int who, struct rusage32 *
 				   &r.ru_stime);
 		r.ru_minflt = current->min_flt + current->cmin_flt;
 		r.ru_majflt = current->maj_flt + current->cmaj_flt;
-		r.ru_nswap = current->nswap + current->cnswap;
 		break;
 	}
 
diff -puN fs/proc/array.c~kill-nswap fs/proc/array.c
--- tiny/fs/proc/array.c~kill-nswap	2004-03-20 12:14:34.000000000 -0600
+++ tiny-mpm/fs/proc/array.c	2004-03-20 12:14:34.000000000 -0600
@@ -388,8 +388,8 @@ int proc_pid_stat(struct task_struct *ta
 		sigign      .sig[0] & 0x7fffffffUL,
 		sigcatch    .sig[0] & 0x7fffffffUL,
 		wchan,
-		task->nswap,
-		task->cnswap,
+		0UL,
+		0UL,
 		task->exit_signal,
 		task_cpu(task),
 		task->rt_priority,
diff -puN include/linux/sched.h~kill-nswap include/linux/sched.h
--- tiny/include/linux/sched.h~kill-nswap	2004-03-20 12:14:34.000000000 -0600
+++ tiny-mpm/include/linux/sched.h	2004-03-20 12:14:34.000000000 -0600
@@ -428,7 +428,7 @@ struct task_struct {
 	unsigned long nvcsw, nivcsw, cnvcsw, cnivcsw; /* context switch counts */
 	u64 start_time;
 /* mm fault and swap info: this can arguably be seen as either mm-specific or thread-specific */
-	unsigned long min_flt, maj_flt, nswap, cmin_flt, cmaj_flt, cnswap;
+	unsigned long min_flt, maj_flt, cmin_flt, cmaj_flt;
 /* process credentials */
 	uid_t uid,euid,suid,fsuid;
 	gid_t gid,egid,sgid,fsgid;
diff -puN kernel/acct.c~kill-nswap kernel/acct.c
--- tiny/kernel/acct.c~kill-nswap	2004-03-20 12:14:34.000000000 -0600
+++ tiny-mpm/kernel/acct.c	2004-03-20 12:14:34.000000000 -0600
@@ -376,7 +376,7 @@ static void do_acct_process(long exitcod
 	ac.ac_rw = encode_comp_t(ac.ac_io / 1024);
 	ac.ac_minflt = encode_comp_t(current->min_flt);
 	ac.ac_majflt = encode_comp_t(current->maj_flt);
-	ac.ac_swaps = encode_comp_t(current->nswap);
+	ac.ac_swaps = encode_comp_t(0);
 	ac.ac_exitcode = exitcode;
 
 	/*
diff -puN kernel/exit.c~kill-nswap kernel/exit.c
--- tiny/kernel/exit.c~kill-nswap	2004-03-20 12:14:34.000000000 -0600
+++ tiny-mpm/kernel/exit.c	2004-03-20 12:14:34.000000000 -0600
@@ -92,7 +92,6 @@ repeat: 
 	p->parent->cstime += p->stime + p->cstime;
 	p->parent->cmin_flt += p->min_flt + p->cmin_flt;
 	p->parent->cmaj_flt += p->maj_flt + p->cmaj_flt;
-	p->parent->cnswap += p->nswap + p->cnswap;
 	p->parent->cnvcsw += p->nvcsw + p->cnvcsw;
 	p->parent->cnivcsw += p->nivcsw + p->cnivcsw;
 	sched_exit(p);
diff -puN kernel/fork.c~kill-nswap kernel/fork.c
--- tiny/kernel/fork.c~kill-nswap	2004-03-20 12:14:34.000000000 -0600
+++ tiny-mpm/kernel/fork.c	2004-03-20 12:14:34.000000000 -0600
@@ -512,7 +512,6 @@ static int copy_mm(unsigned long clone_f
 
 	tsk->min_flt = tsk->maj_flt = 0;
 	tsk->cmin_flt = tsk->cmaj_flt = 0;
-	tsk->nswap = tsk->cnswap = 0;
 	tsk->nvcsw = tsk->nivcsw = tsk->cnvcsw = tsk->cnivcsw = 0;
 
 	tsk->mm = NULL;
diff -puN kernel/sys.c~kill-nswap kernel/sys.c
--- tiny/kernel/sys.c~kill-nswap	2004-03-20 12:14:34.000000000 -0600
+++ tiny-mpm/kernel/sys.c	2004-03-20 12:14:34.000000000 -0600
@@ -1521,7 +1521,6 @@ int getrusage(struct task_struct *p, int
 			r.ru_nivcsw = p->nivcsw;
 			r.ru_minflt = p->min_flt;
 			r.ru_majflt = p->maj_flt;
-			r.ru_nswap = p->nswap;
 			break;
 		case RUSAGE_CHILDREN:
 			jiffies_to_timeval(p->cutime, &r.ru_utime);
@@ -1530,7 +1529,6 @@ int getrusage(struct task_struct *p, int
 			r.ru_nivcsw = p->cnivcsw;
 			r.ru_minflt = p->cmin_flt;
 			r.ru_majflt = p->cmaj_flt;
-			r.ru_nswap = p->cnswap;
 			break;
 		default:
 			jiffies_to_timeval(p->utime + p->cutime, &r.ru_utime);
@@ -1539,7 +1537,6 @@ int getrusage(struct task_struct *p, int
 			r.ru_nivcsw = p->nivcsw + p->cnivcsw;
 			r.ru_minflt = p->min_flt + p->cmin_flt;
 			r.ru_majflt = p->maj_flt + p->cmaj_flt;
-			r.ru_nswap = p->nswap + p->cnswap;
 			break;
 	}
 	return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;

_

-- 
Matt Mackall : http://www.selenic.com : Linux development and consulting