[PATCH] fix vsyscall settimeofday

[PATCH] fix vsyscall settimeofday

[Daniel Walker]

I've only seen this on x86_64 .

The vsyscall state only gets updated when a timer interrupts comes in . So
if the time is set long before the next timer, there will be a period when
a gettimeofday() won't reflect the correct time.

I added an explicit update_vsyscall() during the settimeofday(), that way
the vsyscall state doesn't get stale.

Any thought John?

Signed-Off-By: Daniel Walker <dwalker@mvista.com>

---
 kernel/timer.c |    2 ++
 1 file changed, 2 insertions(+)

Index: linux-2.6.20/kernel/timer.c
===================================================================
--- linux-2.6.20.orig/kernel/timer.c
+++ linux-2.6.20/kernel/timer.c
@@ -861,6 +861,8 @@ int do_settimeofday(struct timespec *tv)
 	clock->error = 0;
 	ntp_clear();
 
+	update_vsyscall(&xtime, clock);
+
 	write_sequnlock_irqrestore(&xtime_lock, flags);
 
 	/* signal hrtimers about time change */
-- 

Re: [PATCH] fix vsyscall settimeofday

[john stultz]

On Mon, 2007-03-05 at 08:56 -0800, Daniel Walker wrote:
> I've only seen this on x86_64 .
> 
> The vsyscall state only gets updated when a timer interrupts comes in . So
> if the time is set long before the next timer, there will be a period when
> a gettimeofday() won't reflect the correct time.
> 
> I added an explicit update_vsyscall() during the settimeofday(), that way
> the vsyscall state doesn't get stale.
> 
> Any thought John?

Oh! Yes, very good catch, Daniel!

Thanks so much!
-john


> Signed-Off-By: Daniel Walker <dwalker@mvista.com>

Acked-by: John Stultz <johnstul@us.ibm.com>


> ---
>  kernel/timer.c |    2 ++
>  1 file changed, 2 insertions(+)
> 
> Index: linux-2.6.20/kernel/timer.c
> ===================================================================
> --- linux-2.6.20.orig/kernel/timer.c
> +++ linux-2.6.20/kernel/timer.c
> @@ -861,6 +861,8 @@ int do_settimeofday(struct timespec *tv)
>  	clock->error = 0;
>  	ntp_clear();
> 
> +	update_vsyscall(&xtime, clock);
> +
>  	write_sequnlock_irqrestore(&xtime_lock, flags);
> 
>  	/* signal hrtimers about time change */