Re: [PATCH] Re: time() glitch on 2.4.18: solved

[Alan Cox]

On Tue, 2002-11-05 at 18:57, Richard B. Johnson wrote:
> No! You will break many machines. You cannot use out_p() when
> writing the latch it __must__ be out(). the "_p" puts a write
> to another port between the two writes. That will screw up
> the internal state-machine of most PITs including AMD-SC520.

The delay is required for the PIT. Your AMD-SC520 is simply a bit wacko.
The right way to fix this is actually to switch inb_p to use udelay(8)
and to load a conservative bogomip number at boot time (we inb_p before
we compute udelay values in a few spots)

BTW if your SC520 is broken that way I just broke 2.5.4x support for it
fixing some more standards compliant platforms 8)