[PATCH] ip2: avoid add_timer() with pending timer

[PATCH] ip2: avoid add_timer() with pending timer

[Akinobu Mita]

add_timer() is not suppose to be called when the timer is pending.
ip2 driver attempts to avoid that condition by setting and resetting
a flag (TimerOn) in timer function. But there is some gap between
add_timer() and setting TimerOn.

This patch fix this problem by using mod_timer() and remove TimerOn
which has been unnecessary by this change.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Michael H. Warfield <mhw@wittsend.com>
---
 drivers/char/ip2/ip2main.c |   18 ++++--------------
 1 file changed, 4 insertions(+), 14 deletions(-)

Index: 2.6-mmotm/drivers/char/ip2/ip2main.c
===================================================================
--- 2.6-mmotm.orig/drivers/char/ip2/ip2main.c
+++ 2.6-mmotm/drivers/char/ip2/ip2main.c
@@ -252,7 +252,6 @@ static unsigned long bh_counter = 0;
  */
 #define  POLL_TIMEOUT   (jiffies + 1)
 static DEFINE_TIMER(PollTimer, ip2_poll, 0, 0);
-static char  TimerOn;
 
 #ifdef IP2DEBUG_TRACE
 /* Trace (debug) buffer data */
@@ -372,10 +371,7 @@ ip2_cleanup_module(void)
 	printk (KERN_DEBUG "Unloading %s: version %s\n", pcName, pcVersion );
 #endif
 	/* Stop poll timer if we had one. */
-	if ( TimerOn ) {
-		del_timer ( &PollTimer );
-		TimerOn = 0;
-	}
+	del_timer_sync(&PollTimer);
 
 	/* Reset the boards we have. */
 	for( i = 0; i < IP2_MAX_BOARDS; ++i ) {
@@ -746,10 +742,8 @@ ip2_loadmain(int *iop, int *irqp)
 			}
 			if ( ip2config.irq[i] == CIR_POLL ) {
 retry:
-				if (!TimerOn) {
-					PollTimer.expires = POLL_TIMEOUT;
-					add_timer ( &PollTimer );
-					TimerOn = 1;
+				if (!timer_pending(&PollTimer)) {
+					mod_timer(&PollTimer, POLL_TIMEOUT);
 					printk( KERN_INFO "IP2: polling\n");
 				}
 			} else {
@@ -1250,16 +1244,12 @@ ip2_poll(unsigned long arg)
 {
 	ip2trace (ITRC_NO_PORT, ITRC_INTR, 100, 0 );
 
-	TimerOn = 0; // it's the truth but not checked in service
-
 	// Just polled boards, IRQ = 0 will hit all non-interrupt boards.
 	// It will NOT poll boards handled by hard interrupts.
 	// The issue of queued BH interrupts is handled in ip2_interrupt().
 	ip2_polled_interrupt();
 
-	PollTimer.expires = POLL_TIMEOUT;
-	add_timer( &PollTimer );
-	TimerOn = 1;
+	mod_timer(&PollTimer, POLL_TIMEOUT);
 
 	ip2trace (ITRC_NO_PORT, ITRC_INTR, ITRC_RETURN, 0 );
 }

Re: [PATCH] ip2: avoid add_timer() with pending timer

[Alan Cox]

On Mon, 14 Jul 2008 11:49:55 +0900
Akinobu Mita <akinobu.mita@gmail.com> wrote:

> add_timer() is not suppose to be called when the timer is pending.
> ip2 driver attempts to avoid that condition by setting and resetting
> a flag (TimerOn) in timer function. But there is some gap between
> add_timer() and setting TimerOn.
> 
> This patch fix this problem by using mod_timer() and remove TimerOn
> which has been unnecessary by this change.
> 
> Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
> Cc: Michael H. Warfield <mhw@wittsend.com>

Acked-by: Alan Cox <alan@redhat.com>

Queued for ttydev tree

Alan

Re: [PATCH] ip2: avoid add_timer() with pending timer

[Andrew Morton]

On Mon, 14 Jul 2008 11:49:55 +0900 Akinobu Mita <akinobu.mita@gmail.com> wrote:

> @@ -746,10 +742,8 @@ ip2_loadmain(int *iop, int *irqp)
>  			}
>  			if ( ip2config.irq[i] == CIR_POLL ) {
>  retry:
> -				if (!TimerOn) {
> -					PollTimer.expires = POLL_TIMEOUT;
> -					add_timer ( &PollTimer );
> -					TimerOn = 1;
> +				if (!timer_pending(&PollTimer)) {
> +					mod_timer(&PollTimer, POLL_TIMEOUT);
>  					printk( KERN_INFO "IP2: polling\n");

This change to ip2_loadmain() might be racy if it actually did
anything.  But afaict timer_pending() can never be true here (we're
only called from module_init()) so it can and should be removed.

Re: [PATCH] ip2: avoid add_timer() with pending timer

[Akinobu Mita]

On Mon, Jul 14, 2008 at 10:04:02AM -0700, Andrew Morton wrote:
> On Mon, 14 Jul 2008 11:49:55 +0900 Akinobu Mita <akinobu.mita@gmail.com> wrote:
> 
> > @@ -746,10 +742,8 @@ ip2_loadmain(int *iop, int *irqp)
> >  			}
> >  			if ( ip2config.irq[i] == CIR_POLL ) {
> >  retry:
> > -				if (!TimerOn) {
> > -					PollTimer.expires = POLL_TIMEOUT;
> > -					add_timer ( &PollTimer );
> > -					TimerOn = 1;
> > +				if (!timer_pending(&PollTimer)) {
> > +					mod_timer(&PollTimer, POLL_TIMEOUT);
> >  					printk( KERN_INFO "IP2: polling\n");
> 
> This change to ip2_loadmain() might be racy if it actually did
> anything.  But afaict timer_pending() can never be true here (we're
> only called from module_init()) so it can and should be removed.

Yes, ip2_loadmain() is only called from module_init(). But this add_timer()
is in the loop for each device. So it may be called more than once.

And yes, I can remove timer_pending() here because I switched to use
mod_timer() instead of add_timer(). But this PollTimer works as periodic
timer and it's global timer. So it will not cause a problem.