Re: [PATCH] watchdog: add CONFIG_WATCHDOG_NOWAYOUT support to PowerPC Book-E watchdog driver

[Josh Boyer <jwboyer@linux.vnet.ibm.com>]

On Fri, Dec 03, 2010 at 10:51:43AM -0600, Timur Tabi wrote:
>Normally, the watchdog is disabled when dev/watchdog is closed, but if
>CONFIG_WATCHDOG_NOWAYOUT is defined, then it means that the watchdog should
>remain enabled.  So we should disable it only if CONFIG_WATCHDOG_NOWAYOUT is
>not defined.
>
>Also ensure that /dev/watchdog is only opened by one process at a time.  That
>way, a second process can't accidentally disable the watchdog while the first
>process has it open.  There shouldn't be any need for more than one process to
>open /dev/watchdog anyway.
>
>Signed-off-by: Timur Tabi <timur@freescale.com>
>---
>
>Kumar, please pick up this patch for 2.6.37.
>
> drivers/watchdog/booke_wdt.c |   16 ++++++++++++++++
> 1 files changed, 16 insertions(+), 0 deletions(-)
>
>diff --git a/drivers/watchdog/booke_wdt.c b/drivers/watchdog/booke_wdt.c
>index d11ffb0..636e013 100644
>--- a/drivers/watchdog/booke_wdt.c
>+++ b/drivers/watchdog/booke_wdt.c
>@@ -193,8 +193,15 @@ static long booke_wdt_ioctl(struct file *file,
> 	return 0;
> }
>
>+/* wdt_is_active stores wether or not the /dev/watchdog device is opened */
>+static unsigned long wdt_is_active;
>+
> static int booke_wdt_open(struct inode *inode, struct file *file)
> {
>+	/* /dev/watchdog can only be opened once */
>+	if (test_and_set_bit(0, &wdt_is_active))
>+		return -EBUSY;
>+
> 	spin_lock(&booke_wdt_lock);
> 	if (booke_wdt_enabled == 0) {
> 		booke_wdt_enabled = 1;

I'm confused why you can't use booke_wdt_enabled for the purposes of the
device having been opened.  It seems the use of the wdt_is_active
basically duplicates this functionalit (and oddly with the bit
manipulation instead of just atomic_inc/dec).

>@@ -210,8 +217,17 @@ static int booke_wdt_open(struct inode *inode, struct file *file)
>
> static int booke_wdt_release(struct inode *inode, struct file *file)
> {
>+#ifndef CONFIG_WATCHDOG_NOWAYOUT
>+	/* Normally, the watchdog is disabled when /dev/watchdog is closed, but
>+	 * if CONFIG_WATCHDOG_NOWAYOUT is defined, then it means that the
>+	 * watchdog should remain enabled.  So we disable it only if
>+	 * CONFIG_WATCHDOG_NOWAYOUT is not defined.
>+	 */
> 	on_each_cpu(__booke_wdt_disable, NULL, 0);
> 	booke_wdt_enabled = 0;
>+#endif
>+
>+	clear_bit(0, &wdt_is_active);

If you were to keep this variable instead of just using
booke_wdt_enabled, wouldn't it be more correct to have the clear_bit
only done inside the #ifndef?  The timer is very much still active if
NOWAYOUT is set...

josh