Re: [RFC][PATCH] new timeofday-based soft-timer subsystem

[Nish Aravamudan <nish.aravamudan@gmail.com>]

On 5/3/05, Chris Friesen <cfriesen@nortel.com> wrote:
> Nishanth Aravamudan wrote:
> 
> > but then there is another issue: the restart_block used by
> > sys_nanosleep() only allows for 4 unsigned long arguments, when, in
> > fact, nanoseconds are a 64-bit quantity in the kernel. As long as the
> > nanosleep() request is no more than around 4 seconds, we should be ok
> > using unsigned longs.
> 
> My man page for nanosleep specifies that the "nanoseconds" portion of
> the timespec must be under 1 billion and is of type "long".  Is that no
> longer valid?

Certainly would be, but the problem is if you pass in a timespec ts, where

    ts.tv_sec = 10;
    ts.tv_nsec = 99999;

This will overflow a 32-bit nanosecond representation internally
(10000099999 > 4294967296). Sorry for the confusion, the unsigned long
I was referring to was the internal representation of the nanoseconds
converted from the timespec parameter.

Thanks,
Nish