2.2.15pre4 VM fix

2.2.15pre4 VM fix

[Rik van Riel]

Hi Alan,

with the attached patch I hope to have fixed the 2.2.15pre4
VM problems. I didn't manage to break it myself, but maybe
one of the dear readers has a machine where they are able
to do so...

Please give this patch (against 2.2.15pre4) a solid beating
and report back to us. Thanks all!

regards,

Rik
--
The Internet is not a network of computers. It is a network
of people. That is its real strength.


--- mm/page_alloc.c.orig	Tue Jan 25 00:01:43 2000
+++ mm/page_alloc.c	Wed Jan 26 01:16:21 2000
@@ -210,6 +210,12 @@
 	 */
 	if (!(current->flags & PF_MEMALLOC)) {
 		int freed;
+		if (current->state != TASK_RUNNING && (gfp_mask & __GFP_WAIT)) {
+			printk("gfp called by non-running (%d) task from %p!\n",
+				current->state, __builtin_return_address(0));
+			/* if we're not running, we can't sleep */
+			gfp_mask &= ~__GFP_WAIT;
+		}
 
 		if (nr_free_pages <= freepages.low) {
 			wake_up_interruptible(&kswapd_wait);
@@ -224,6 +230,9 @@
 		current->flags |= PF_MEMALLOC;
 		freed = try_to_free_pages(gfp_mask);
 		current->flags &= ~PF_MEMALLOC;
+
+		if ((gfp_mask & __GFP_MED) && nr_free_pages > freepages.min / 2)
+			goto ok_to_allocate;
 
 		if (!freed && !(gfp_mask & __GFP_HIGH))
 			goto nopage;

--
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.eu.org/Linux-MM/

Re: 2.2.15pre4 VM fix

[Ivan Kokshaysky]

On Wed, Jan 26, 2000 at 01:20:48AM +0100, Rik van Riel wrote:
> 
> Please give this patch (against 2.2.15pre4) a solid beating
> and report back to us. Thanks all!
> 

n_tty_open() has been caught with your patch.
Thanks!

Ivan.
--
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.eu.org/Linux-MM/

Re: 2.2.15pre4 VM fix

[Alan Cox]

> > Please give this patch (against 2.2.15pre4) a solid beating
> > and report back to us. Thanks all!
> 
> n_tty_open() has been caught with your patch.
> Thanks!

Do you know which drivers (serial,tty) you were using it. n_tty_open itself
seems ok, but the caller may be guilty
--
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.eu.org/Linux-MM/

Re: 2.2.15pre4 VM fix

[Theodore Y. Ts'o]

   > > Please give this patch (against 2.2.15pre4) a solid beating
   > > and report back to us. Thanks all!
   > 
   > n_tty_open() has been caught with your patch.
   > Thanks!

   Do you know which drivers (serial,tty) you were using it. n_tty_open itself
   seems ok, but the caller may be guilty

The drivers don't call the line discpline open routine.  That honor is
reserved to the high-level tty layer --- see tty_set_ldisc in tty_io.c

							- Ted
--
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.eu.org/Linux-MM/

Re: 2.2.15pre4 VM fix

[Ivan Kokshaysky]

On Fri, Jan 28, 2000 at 02:40:30PM +0000, Alan Cox wrote:
> > n_tty_open() has been caught with your patch.
> > Thanks!
> 
> Do you know which drivers (serial,tty) you were using it. n_tty_open itself
> seems ok, but the caller may be guilty

It happened when ppp connection was terminated (remote end hangup).
Serial driver is Comtrol Rocketport. The problem is repeatable
(3 times last 20 hours), so I can investigate further to see who
is the caller.

Ivan.
--
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.eu.org/Linux-MM/

Re: 2.2.15pre4 VM fix

[Theodore Y. Ts'o]

   On Fri, Jan 28, 2000 at 02:40:30PM +0000, Alan Cox wrote:
   > > n_tty_open() has been caught with your patch.
   > > Thanks!
   > 
   > Do you know which drivers (serial,tty) you were using it. n_tty_open itself
   > seems ok, but the caller may be guilty

   It happened when ppp connection was terminated (remote end hangup).
   Serial driver is Comtrol Rocketport. The problem is repeatable
   (3 times last 20 hours), so I can investigate further to see who
   is the caller.

I think it's a flase positive.  It's happeninig because tty_do_hangup()
is calling ldisc.open --- which means n_tty_open() inside an interrupt
context.  n_tty_open() makes a check to see whether it is being called
inside an interrupt, and uses GFP_ATOMIC to avoid blocking inside the
interrupt.  


static int n_tty_open(struct tty_struct *tty)
{
	....
		get_zeroed_page(in_interrupt() ? GFP_ATOMIC : GFP_KERNEL);
	...
}

This should be OK, I think.

						- Ted
--
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.eu.org/Linux-MM/