Re: [update] [patch] linux: Cache coherency fixes

[Ralf Baechle <ralf@oss.sgi.com>]

On Thu, Aug 01, 2002 at 07:14:05PM +0200, Maciej W. Rozycki wrote:

> > R10000.
> 
>  OK.  Any specs anywhere?

techpubs.sgi.com should have a somewhat older manual (must have!) and
www.necel.com as well.  The geniouses at NEC stripped the description of
the cache instruction from their manual so it you really want both manuals.

> > Back in time I prefered CONFIG_NONCOHERENT_IO over CONFIG_COHERENT_IO
> > because the noncoherent case needs additional code and in general I'm
> > trying to reduce the number of the #if !defined conditionals for easier
> > readability.
> 
>  Hmm, what's wrong with "#ifndef"?  Not much less readable than "#ifdef",
> IMO. 

Just a small detail.  Nest conditions several times and the spaghetti
starts :-)

> Basically:
> 
> 1. Does the CPU support coherency?
> 
> 2. If so, does the system?
> 
> I'm going to express it this way in the config script.

Have fun expressing if a R4000 variant supports coherency :-)  You can't
if you don't want to introduce even more R4000 types or subtypes.

>  Well, inferring from docs and my experience it's not needed.  A system
> may simply require areas used for DMA to be marked as non-coherent by
> CPUs.  Often uncached accesses are used to prevent spoiling the caches
> anyway.

None such MIPS system known where this is a sensible mode of operation -
and I've hacked quite a number of platforms.  Anyway, if there were such
systems they'd either have to be considered as coherent or as non-coherent.
Our current model doesn't permit any finer grained configuration and unless
such a system actually exists I don't think we should introduce one.

Btw, I've seen a fairly new system in which the non-coherent bits in some
agent are not working at all - it's easier to implement that way ...

> > Using a non-coherent mode on IP27 may result in nice, hard to trackdown bus
> > errors.
> 
>  Weird, but I accept it as a fact.  Still a bus error expresses more than
> a hang. ;-) 

Not so weired.  The system is still operating in coherent mode; that also
means the directory caches still think they know where a particular memory
address is cached.  It's possible to operate an IP27 in non-coherent mode
for I/O only by also reconfiguring the whole chipset but that turned out to
be always slower and harder to get the software correct, so it's not used
except for debugging with a logic analyzer; details are mindbogglingly
complex.

  Ralf