Re: Using get_cycles for add_timer_randomness

[Ralf Baechle <ralf@linux-mips.org>]

On Mon, Aug 16, 2004 at 09:22:28AM -0700, William Lee Irwin III wrote:

> At some point in the past, Arnd Bergmann <arnd@arndb.de> wrote:
> >> I noticed that only i386 and x86-64 are currently using a high resolution
> >> timer source when adding randomness.  Since many architectures have a
> >> working get_cycles() implementation, it seems rather straightforward to use
> >> that.
> 
> On Mon, Aug 16, 2004 at 12:17:56PM -0400, Theodore Ts'o wrote:
> > My only concern about using get_cycles is the speed question; on some
> > architectures, could (particularly those without TSC registers or
> > equivalent hardware support) could get_cycles() be slow enough to cause
> > latency problems?
> 
> AFAICT most of those return 0 from get_cycles().

The cycle counter on MIPS processors [1] is only 32-bit.  I've been
pondering to construct a 64-bit value in get_cycles() which would seriously
slow down the function compared to the current single instruction
implementation; fortunately a quick survey of all get_cycles() users a
while ago did look like non of them actually requires the full 64-bit.

  Ralf

[1] Some older processors don't have at all.