[9fans] ok, *init* is running on ts7200

[9fans] ok, *init* is running on ts7200

[Ronald G. Minnich]

although not much else is happening.

virtual address caches are touchy when it comes time to flush them and
change mappings ...

in particulsr, you have to
	flush()
	change mappings
	invalidate TLBs

in careful order. Note that PACs are not nearly so picky.

anyway ... where I'm at is init code is in memory, I'm trying to switch to
it, at some point mmuswitch is called again, then again, and that third
time ... no joy, the system dies.


but it's closer .

ron

Re: [9fans] ok, *init* is running on ts7200

[Ronald G. Minnich]

On Fri, 18 Mar 2005, Ronald G. Minnich wrote:

> in particulsr, you have to
> 	flush()
> 	change mappings
> 	invalidate TLBs

oh yeah, note that bitsy was not nearly as problematic due to the va==pa
mapping...

ron

Re: [9fans] ok, *init* is running on ts7200

[Charles Forsyth]

it's even more fun on the powerpc where the
traps (before book E, i think that's the one that changed it)
throw away the painfully-crafted virtual memory
environment and leave the processor in real (physical addressing) mode.
it's quite unreal: crawl up out of the swamp again...

Re: [9fans] ok, *init* is running on ts7200

[Ronald G. Minnich]

On Fri, 18 Mar 2005, Charles Forsyth wrote:

> it's even more fun on the powerpc where the
> traps (before book E, i think that's the one that changed it)
> throw away the painfully-crafted virtual memory
> environment and leave the processor in real (physical addressing) mode.

Oh my goodness that's stupid. I wonder why they did that?

ron

Re: [9fans] ok, *init* is running on ts7200

[Charles Forsyth]

>>Oh my goodness that's stupid. I wonder why they did that?

it's certainly annoying.  for x86 fans (and those processors
often need big fans) i should explain that
at least real mode in the powerpc, which sets you back
to the instruction before you enabled vm, isn't as bad as `real'
mode on the x86, which sets you back about 25 years.
the powerpc at least is much the same processor but with
VM switched off (but you've got to jump through a few
hoops to put it on again).  one popular way to switch the mmu back on
is to do a return-from-interrupt back into the trap handler,
having saved the state.  one reason is that just setting
the appropriate bits in the MSR might cause trouble (or might not).
either way, to save the state you've got to
address the appropriate kernel save area.
are those memory accesses cached in that state?
well ... it depends.

Re: [9fans] ok, *init* is running on ts7200

[Charles Forsyth]

anyhow, you possibly don't feel quite so bad about those VACs now.