Re: [PATCH] next_timer_interrupt: simpler overflow handling

[Zachary Amsden <zach@vmware.com>]

Keir Fraser wrote:
> Having seen the patch applied to 2.6.17 to fix the overflowing 
> comparison in next_timer_interrupt() it occurred to me that a much 
> simpler fix is to not set hr_expires to MAX_JIFFY_OFFSET. It's way 
> further out from jiffies than necessary, which is why it's caused 
> problems. I instead propose that we initialise it to LONG_MAX>>1, just 
> as we already do for the non-hr expires variable. This will allow safe 
> comparison with any timer value in the range jiffies+/-(LONG_MAX>>1) 
> which is plenty of range around jiffies (+/- 12 days if HZ=1000 and 
> long is 32 bits).
>
> The advantages are simpler code, and uniform initialisation of expires 
> and hr_expires variables.

Even simpler would be to just make MAX_JIFFY_OFFSET be (LONG_MAX >> 1) 
and use this for both.  In fact, it appears it used to be, judging by 
the comment in jiffies.h:

 * The maximum jiffie value is (MAX_INT >> 1).  Here we translate that

But seeing as this could have unanticipated side effects, I like this 
fix better.

Acked-By: Zachary Amsden <zach@vmware.com>