Re: [PATCH tip/sched/core] introduce task_times() to replace task_[us]time() pair

[Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>]

Hidetoshi Seto wrote:
> And just now I noticed that using msecs_to_cputime() is problematic,
> since the type of its return value is "unsigned long" so not 64bit.
> I'll make and post a patch to fix this asap.

Sorry, I was confused.  Please ignore the above.

The problem is that since the type of msecs_to_cputime() argument is
unsigned int, it would be wrong if calculated msecs is greater than
UINT_MAX.

Another problem is that msecs_to_jiffies() returns MAX_JIFFY_OFFSET
if msecs > INT_MAX. 

   5168 #ifndef nsecs_to_cputime
   5169 # define nsecs_to_cputime(__nsecs) \
   5170         msecs_to_cputime(div_u64((__nsecs), NSEC_PER_MSEC))
   5171 #endif

    452 unsigned long msecs_to_jiffies(const unsigned int m)
    453 {
    454         /*
    455          * Negative value, means infinite timeout:
    456          */
    457         if ((int)m < 0)
    458                 return MAX_JIFFY_OFFSET;
     :

What I need here is a simple function can convert 64bit nsecs to
cputime_t, which is unsigned long in asm-generic/cputime.h.
Humm...


Thanks,
H.Seto