RTEMS on MCP750 and MTX-60x, for Eric Valette

[gregory.menke at gsfc.nasa.gov]

Valette Eric writes:
 > gregory.menke at gsfc.nasa.gov wrote:
 > 
 > > I think I'll give your idea a try- some of the stuff in the bootloader
 > > related to identifying the various segments is very strange and I
 > > don't like some aspects of how its implemented. 
 > 
 > 
 > However, this bootlaoder has also unique feature that makes it really 
 > nice ans superior to others (from memory) :
 > 	1) It is fully relocatable. Thus can be loaded at any place. This is 
 > the rason why the linux boot loader does work on many machines. Then it 
 > dynamically finishe part of the relocation,
 > 	2) It worked on more motorolla board than the original linux one in 99,
 >

I spent some time over the weekend looking over the Linux ppc
bootloader- it doesn't do any of the mmu/cache setup that Paubert's
does.  After commenting that stuff out, the mcp750 and mtx603 boards
both boot and run OK.  I think that means I can send the mtx bsp to
Joel without breaking any of the other ppc bsps- then update the
replacement bootloader when I finish working on it.  

Does anyone reading this thread have a motorola 601 board which they
might be interested in using to test this before I submit it?

 
 >  > Since there are newer
 > > ppc devices, it will probably be useful to keep the bootloader
 > > somewhat current- it might make future bsp's a little easier.
 > 
 > Sure. That is the reason I said you probably should invest your time in 
 > a more recent/alive work. I think ppcboot is the obvious candidate but I 
 > do not know if it will work for MBX high end ppc board. Nevertheless, 
 > now that the 8260 has a 603e core, at least concerning 603 CPU init it 
 > should work...

You did say that- sorry I was being a bit thick.

 > 
 > > They also come in SMP versions- ours is not
 > 
 > You mean lossely coupled SMP? or real symetric SMP boards?

The 2 cpus seem to share the same processor bus, but I've not gone
deeply into how they're arranged because ours is just a single
processor board.

 
 > > unfortunately, but wouldn't it be fun to make RTEMS SMP compliant?
 > 
 > Making rtems symetric SMP compliant, would require some major redesign 
 > and at least a walthrough of any file as most looking mechanism assume 
 > IRQ masked ==> enter criticql region. IMHO.

I briefly spoke with Joel about this, his thought was that a n-way SMP
system would probably simultaneously run the n highest priority tasks.
Making it work is another story of course...


 > > I quite like the ppc.  I'm coming from the Mongoose R3000 bsp, which
 > > is a much more primitive processor.  Theres nothing much bad about it,
 > > but the ppc is MUCH fancier.  I wonder how a R4000 compares to the
 > > ppc.
 > 
 > The R3000 has many shortcoming regarding OS support. But now they are 
 > R10000...
 > 

We didn't find the 3000 too horrible, at least we didn't have to
kludge anything.  We would have liked to exploit some of the memory
access control features of the processor, but we felt the overhead for
managing it was too expensive given the cpu clock (12mhz).  On the
other hand, the uarts onboard our R3000 are dreadful.

Gregm