[PATCH] ipconfig schedule fix

[PATCH] ipconfig schedule fix

[Magnus Damm]

[-- Attachment #1: Type: text/plain, Size: 299 bytes --]

Hi,

This simple patch makes the DHCP/BOOTP/RARP code sleep with
schedule_timeout() instead of hogging the cpu with cpu_relax() in a
loop. This comes handy when a guest kernel is booted inside QEMU.
Without the patch, the host operating system will consume cpu time
busy waiting.

Thanks,

/ magnus

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: linux-2.6.10-rc2-mm3-ipconfig_schedule.patch --]
[-- Type: text/x-patch; name="linux-2.6.10-rc2-mm3-ipconfig_schedule.patch", Size: 1293 bytes --]

--- linux-2.6.10-rc2-mm3/net/ipv4/ipconfig.c	2004-10-18 23:53:08.000000000 +0200
+++ linux-2.6.10-rc2-mm3-ipconfig_schedule/net/ipv4/ipconfig.c	2004-11-23 23:21:26.154463880 +0100
@@ -1102,8 +1102,8 @@
 
 		jiff = jiffies + (d->next ? CONF_INTER_TIMEOUT : timeout);
 		while (time_before(jiffies, jiff) && !ic_got_reply) {
-			barrier();
-			cpu_relax();
+			set_current_state(TASK_UNINTERRUPTIBLE);
+			schedule_timeout (1);
 		}
 #ifdef IPCONFIG_DHCP
 		/* DHCP isn't done until we get a DHCPACK. */
@@ -1245,7 +1245,6 @@
 
 static int __init ip_auto_config(void)
 {
-	unsigned long jiff;
 	u32 addr;
 
 #ifdef CONFIG_PROC_FS
@@ -1260,18 +1259,16 @@
  try_try_again:
 #endif
 	/* Give hardware a chance to settle */
-	jiff = jiffies + CONF_PRE_OPEN;
-	while (time_before(jiffies, jiff))
-		cpu_relax();
+	set_current_state(TASK_UNINTERRUPTIBLE);
+	schedule_timeout (CONF_PRE_OPEN);
 
 	/* Setup all network devices */
 	if (ic_open_devs() < 0)
 		return -1;
 
 	/* Give drivers a chance to settle */
-	jiff = jiffies + CONF_POST_OPEN;
-	while (time_before(jiffies, jiff))
-		cpu_relax();
+	set_current_state(TASK_UNINTERRUPTIBLE);
+	schedule_timeout (CONF_POST_OPEN);
 
 	/*
 	 * If the config information is insufficient (e.g., our IP address or

Re: [PATCH] ipconfig schedule fix

[David S. Miller]

On Tue, 23 Nov 2004 23:50:35 +0100
Magnus Damm <magnus.damm@gmail.com> wrote:

> This simple patch makes the DHCP/BOOTP/RARP code sleep with
> schedule_timeout() instead of hogging the cpu with cpu_relax() in a
> loop. This comes handy when a guest kernel is booted inside QEMU.
> Without the patch, the host operating system will consume cpu time
> busy waiting.

There are two similar loops involving CONF_PRE_OPEN and CONF_POST_OPEN,
you may wish to hit those while you're at it.

Re: [PATCH] ipconfig schedule fix

[Magnus Damm]

On Tue, 23 Nov 2004 15:18:55 -0800, David S. Miller <davem@davemloft.net> wrote:
> On Tue, 23 Nov 2004 23:50:35 +0100
> 
> 
> Magnus Damm <magnus.damm@gmail.com> wrote:
> 
> > This simple patch makes the DHCP/BOOTP/RARP code sleep with
> > schedule_timeout() instead of hogging the cpu with cpu_relax() in a
> > loop. This comes handy when a guest kernel is booted inside QEMU.
> > Without the patch, the host operating system will consume cpu time
> > busy waiting.
> 
> There are two similar loops involving CONF_PRE_OPEN and CONF_POST_OPEN,
> you may wish to hit those while you're at it.

Huh? You mean _removing_ the timeouts involving CONF_PRE/POST_OPEN? I
thought my patch converted all busy waits into schedule_timeout()...
What am I missing?

/ magnus

Re: [PATCH] ipconfig schedule fix

[David S. Miller]

On Wed, 24 Nov 2004 00:42:17 +0100
Magnus Damm <magnus.damm@gmail.com> wrote:

> On Tue, 23 Nov 2004 15:18:55 -0800, David S. Miller <davem@davemloft.net> wrote:
> > On Tue, 23 Nov 2004 23:50:35 +0100
> > 
> > 
> > Magnus Damm <magnus.damm@gmail.com> wrote:
> > 
> > > This simple patch makes the DHCP/BOOTP/RARP code sleep with
> > > schedule_timeout() instead of hogging the cpu with cpu_relax() in a
> > > loop. This comes handy when a guest kernel is booted inside QEMU.
> > > Without the patch, the host operating system will consume cpu time
> > > busy waiting.
> > 
> > There are two similar loops involving CONF_PRE_OPEN and CONF_POST_OPEN,
> > you may wish to hit those while you're at it.
> 
> Huh? You mean _removing_ the timeouts involving CONF_PRE/POST_OPEN? I
> thought my patch converted all busy waits into schedule_timeout()...
> What am I missing?

No, not removing them, but transforming them likewise into
schedule_timeout() calls.

Re: [PATCH] ipconfig schedule fix

[Magnus Damm]

On Tue, 23 Nov 2004 15:45:13 -0800, David S. Miller <davem@davemloft.net> wrote:
> On Wed, 24 Nov 2004 00:42:17 +0100
> 
> 
> Magnus Damm <magnus.damm@gmail.com> wrote:
> 
> > On Tue, 23 Nov 2004 15:18:55 -0800, David S. Miller <davem@davemloft.net> wrote:
> > > On Tue, 23 Nov 2004 23:50:35 +0100
> > >
> > >
> > > Magnus Damm <magnus.damm@gmail.com> wrote:
> > >
> > > > This simple patch makes the DHCP/BOOTP/RARP code sleep with
> > > > schedule_timeout() instead of hogging the cpu with cpu_relax() in a
> > > > loop. This comes handy when a guest kernel is booted inside QEMU.
> > > > Without the patch, the host operating system will consume cpu time
> > > > busy waiting.
> > >
> > > There are two similar loops involving CONF_PRE_OPEN and CONF_POST_OPEN,
> > > you may wish to hit those while you're at it.
> >
> > Huh? You mean _removing_ the timeouts involving CONF_PRE/POST_OPEN? I
> > thought my patch converted all busy waits into schedule_timeout()...
> > What am I missing?
> 
> No, not removing them, but transforming them likewise into
> schedule_timeout() calls.

Yeah, but isn't the patch already doing that? I tried to find more
occurances of CONF_PRE/POST_OPEN in the
linux-2.6.10-rc2-mm3/net/ipv4/ipconfig.c but I think the patch covers
all of them. Please point out what I am missing, maybe it is getting
too late for me know...

/ magnus