Effectively, guarded means "don't prefetch" for code fetches.  It also means that
no speculative data fetches will be performed to a guarded space.  Speculative
data fetches occur when the processor is speculatively executing a chunk of code
as a result of a predicted branch whose condition is not yet known to be as
predicted.  Setting the guarded attribute on an area prevents such speculation.
This issue probably is nearly moot on the nearly non-superscalar (dumb as a rock)
8xx processors, but is MUCH more important (i.e., VITAL) on the 82xx processors
with the 603ish core and on all "real" PowerPC processors.

a

"Mark S. Mathews" wrote:

Hey Dan,

On Tue, 30 Nov 1999, Dan Malek wrote:

quoted

Mark S. Mathews wrote:

quoted

quoted

We've been working with the embedded 2.2.13 kernel on an RPX-Lite CW with
a XPC823ZT66A processor running at the 50MHz/8MHz setting.

It's not only the RPX Lite.....I have a variety of 8xx boards
and when I have trouble like this with a particular card, it
occurs in all of the boards.

Hmmmm.  The way ours "works for awhile", I'm wondering if there's a
problem w/ the way the 8xx is handling the WAIT# when the MMU is enabled
(?)

quoted

quoted

............ but our
accesses to the common memory regions of the card are twitchy.

Same thing I have seen.  The I/O and attribute regions seem
to work OK, but memory regions don't.....on more than one
type of card.

This is good to know.  Our card supports I/O or memory access to the
shared memory.  We'll shift over to the I/O and try that.

quoted

quoted

.....but eventually we wind up with a machine-check.

Which points to some kind of bus timing or protocol problem.

We've been running _lots_ of experiments with the timing settings.  So far
3,10,6 seems to work best....but it still fails.

quoted

quoted

runs well on the x86, our PowerBook, and on a different 860 based platform
(non-Linux, no MMU) so we're fairly confident it isn't the code.

Well now, that's interesting (the no MMU, not the non-Linux
part :-).  With the MMU disabled, the accesses behave as guarded.

I've seen the 'guarded' thing around in the sources, but I'm not sure what
it's all about.  Guess I should look. ;-)

quoted

This is something I have not properly implemented on the 8xx,
and with my somewhat sloppy use of eieio() and synchronization,
I am always waiting for this to come back and haunt me.  Notice
how I buried this important fact in this paragraph.  I will now
properly implement this (yet tonight).  Tell me the kernel version
you are using and I will send some updates for your testing.

embedded kernel 2.2.13

quoted

How does someone (like me :-) determine what a PCMCIA interface
in something like a PowerBook uses for bus timing?

I'm not really sure.  Have to ask David, it's probably buried in pcmcia-cs
somewhere.

quoted

quoted

One specific question...when setting up the PCMCIA bus timings, the 823
book lists the settings in units of "clock cycles".  Which clock?

It is the CLKOUT (system/bus) clock.  On the 66MHz processor,
this better be 33 MHz (processor clock / 2).  For 50 MHz or
less, the CLKOUT is the processor clock.

BIG help!  Thanks a million.

We'll keep trying and let you know.

-Mark

Mark S. Mathews

AbsoluteValue Software     Web:   http://www.absoval.com
P.O. Box 941149            e-mail: mark@absoval.com
Maitland, FL 32794-1149    Phone:  407.644.8582
USA                        Fax:    407.539.1294

--
Alan Mimms     Packet Engines, Inc.     Spokane, Washington [99214-0497]
  USA, Earth, Sol, Milky Way, The Local Group, Virgo Supercluster, U0
Despite the cost of living, have you noticed how popular it remains?
  -- Steven Wright?


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/