Re: [rfc] aio-core for 2.5.29 (Re: async-io API registration for

[Linus Torvalds]

On Thu, 1 Aug 2002, Chris Friesen wrote:
>
> Now if we had a constant monotonic source of time--say 64-bit nanoseconds since
> boot--this wouldn't be a problem.

Well, we do have such a monotonic time sequence already, and that's the
one that the kernel always uses internally.

It's called "jiffies64".

However, "jiffies" are not really real time, they are only a "reasonable
abstraction thereof", and while they imply ordering ("time_after()" works
fine inside the kernel), they do _not_ imply real time.

In other words, there is no way to move from time -> jiffies and back.

But we could certainly export jiffies64 as a "nanosecond-like" thing. All
it takes is one 32x64-bit multiply. It won't be "true nanoseconds", but it
will be a "reasonable approximation" (ie the rate may be off by several
percentage points, since nothing is correcting for it. But the "no
correction" is part of the _advantage_ too).

		Linus