Hi,
On 17-May-99 Dave Airlie wrote:
>
> Cool that posting helped me understand why I need wbflush, so the next
> question is, can a wbflush() in the wrong place do any harm (apart from
> a performance slowdown),
No, it can't. Nevertheless, wbflush() should be used in as few places as
possible.
> and why in the dec_esp.c driver does the wbflush
> defined not use <asm/wbflush.h> but defines its own wbflush macro for
> the IOASIC decstations, and it is safe to use asm/wbflush.h in its
> place ..?
dec_esp.c is written for and will only work on IOASIC DECstations and the
mechanism for flushing the writeback buffer is the same on all those
machines. I decided to use a shortcut here and avoid an indirect function
call, but the generic wbflush() shouldn't harm.
BTW, I used the same trick in NCR53C9x.c, which probably was a little bit
shortsighted :-). An "#include <asm/wblush.h" definitely belongs here.
Dave, you might want to consider creating something like pmaz-esp.c and
modify hosts.c et al. accordingly. I am not able to see a simple way to
unify an IOASIC and a PMAZ-A driver and the code in *_esp.c is not very
much indeed. What do you think?
Happy hacking.
---
Regards,
Harald
|