2004-10-11 17:07:28

by Chuck Ebbert

[permalink] [raw]

Subject: [PATCH] softdog.c (was: Kernel panic after rmmod softdog (2.6.8.1))

---

Michael Schierl wrote:

> today, when testing the software watchdog, I accidentally removed the
> software watchdog kernel module and got a kernel panic.

Does this fix it? (compiled but not tested)

--- linux-2.6.8.1/drivers/char/watchdog/softdog.c.orig Sun Oct 10 23:08:24 2004
+++ linux-2.6.8.1/drivers/char/watchdog/softdog.c Sun Oct 10 23:10:12 2004
@@ -135,8 +135,9 @@
{
if(test_and_set_bit(0, &timer_alive))
return -EBUSY;
- if (nowayout)
- __module_get(THIS_MODULE);
+
+ __module_get(THIS_MODULE);
+
/*
* Activate timer
*/
@@ -152,6 +153,7 @@
*/
if (expect_close == 42) {
softdog_stop();
+ module_put(THIS_MODULE);
} else {
printk(KERN_CRIT PFX "Unexpected close, not stopping watchdog!\n");
softdog_keepalive();


--Chuck Ebbert 11-Oct-04 12:49:45
Current book: Stephen King: Wizard and Glass

2004-10-11 20:03:48

by Arnd Bergmann

[permalink] [raw]

Subject: Re: [PATCH] softdog.c (was: Kernel panic after rmmod softdog (2.6.8.1))

---

On Maandag 11 Oktober 2004 19:01, Chuck Ebbert wrote:

> --- linux-2.6.8.1/drivers/char/watchdog/softdog.c.orig Sun Oct 10 23:08:24 2004
> +++ linux-2.6.8.1/drivers/char/watchdog/softdog.c Sun Oct 10 23:10:12 2004
> @@ -135,8 +135,9 @@
> {
> if(test_and_set_bit(0, &timer_alive))
> return -EBUSY;
> - if (nowayout)
> - __module_get(THIS_MODULE);
> +
> + __module_get(THIS_MODULE);
> +
> /*
> * Activate timer
> */
> @@ -152,6 +153,7 @@
> */
> if (expect_close == 42) {
> softdog_stop();
> + module_put(THIS_MODULE);
> } else {
> printk(KERN_CRIT PFX "Unexpected close, not stopping watchdog!\n");
> softdog_keepalive();
>
Now if you do open(), close(), open(), write("V"), close(), the module
becomes unremovable, even without nowayout=1. Isn't is possible to
simply add a softdog_stop() call to watchdog_exit()?

Arnd <><

---

Attachments:

(No filename) (0.98 kB)
(No filename) (189.00 B)
signature Download all attachments

2004-10-13 23:59:47

by Chuck Ebbert

[permalink] [raw]

Subject: Re: [PATCH] softdog.c (was: Kernel panic after rmmod softdog (2.6.8.1))

---

Arnd Bergmann wrote:

> Now if you do open(), close(), open(), write("V"), close(), the module
> becomes unremovable, even without nowayout=1. Isn't is possible to
> simply add a softdog_stop() call to watchdog_exit()?

Oh well, at least it can't possibly oops since you can't even remove it. :)

How about this instead?

(Assuming a re-open of the softdog device is allowable; I don't see why not.)

--- linux-2.6.8.1/drivers/char/watchdog/softdog.c.orig Sun Oct 10 23:08:24 2004
+++ linux-2.6.8.1/drivers/char/watchdog/softdog.c Wed Oct 13 14:48:20 2004
@@ -82,9 +82,12 @@

static struct timer_list watchdog_ticktock =
TIMER_INITIALIZER(watchdog_fire, 0, 0);
-static unsigned long timer_alive;
static char expect_close;

+static unsigned long driver_status;
+/* Driver status bits */
+#define SOFTDOG_TIMER_RUNNING 0
+#define SOFTDOG_DEVICE_OPEN 1

/*
* If the timer expires..
@@ -133,9 +136,10 @@

static int softdog_open(struct inode *inode, struct file *file)
{
- if(test_and_set_bit(0, &timer_alive))
+ if (test_and_set_bit(SOFTDOG_DEVICE_OPEN, &driver_status))
return -EBUSY;
- if (nowayout)
+
+ if ( !test_and_set_bit(SOFTDOG_TIMER_RUNNING, &driver_status))
__module_get(THIS_MODULE);
/*
* Activate timer
@@ -152,11 +156,13 @@
*/
if (expect_close == 42) {
softdog_stop();
+ clear_bit(SOFTDOG_TIMER_RUNNING, &driver_status);
+ module_put(THIS_MODULE);
} else {
printk(KERN_CRIT PFX "Unexpected close, not stopping watchdog!\n");
softdog_keepalive();
}
- clear_bit(0, &timer_alive);
+ clear_bit(SOFTDOG_DEVICE_OPEN, &driver_status);
expect_close = 0;
return 0;
}

--Chuck Ebbert 13-Oct-04 19:55:30