Re: [PATCH 2.4.32] Fix AVM C4 ISDN card init problems with newer CPUs

[Alan Cox <alan@lxorguk.ukuu.org.uk>]

Ar Iau, 2006-08-03 am 19:31 +0200, ysgrifennodd Karsten Keil:
> Yes, that should be go in.
> 
> AVM C4 ISDN NIC: Add three memory barriers, taken from 2.6.7,
> (they are there in 2.6.17.7 too), to fix module initialization
> problems appearing with at least some newer Celerons and
> Pentium III.
> 
> Acked-by: Karsten Keil <kkeil@suse.de>
> Signed-off-by: Jukka Partanen <jspartanen@gmail.com>

NAK: Alan Cox <alan@redhat.com>

Two reasons

#1	You should use cpu_relax in such loops
#2	The readl (which c4inmeml is a pointless #define of) is defined to be
a volatile reference itself

That means that the real bug would appear to be different and either you
have a gcc bug which is possible or you have something stranger going
on, such as the continued polling busying the microcontroller the other
end, in which case you need a delay not the lucky chance that mb() is
slowish on some x86 systems.

So you either want

		cpu_relax + other fixes
	or	udelay(something)


Alan