On 30/04/2020 17.03, Christophe Leroy wrote: > > > Le 30/04/2020 à 16:44, Rasmus Villemoes a écrit : >> Hi, >> >> I'm wondering why the sysreset_mpc83xx driver (as well as the similar >> code in arch/powerpc/cpu/mpc83xx/cpu.c) removes the MSR_IR and MSR_DR >> bits from MSR. >> [snip] >> >> But since the arch/powerpc/cpu/mpc83xx/cpu.c code in U-Boot has been >> there forever, I assume I'm missing something fundamental. > > On my mpc8321 it works. > > However I agree it looks odd. > > I think it works by chance because we have a 1:1 MMU mapping, but if you > have a difference mapping I can't see how you can disable Instruction > MMU without going through a write into SRR0/SRR1 and an rfi.
OK, thanks. Can you test if it still works if you change the code to not clear the MSR_IR and MSR_DR bits? (I expect it will, since I assume you'd have noticed if reboot from linux didn't work). I'm actually wondering what the core does after the MSR_IR bit gets cleared. If address translation is disabled, what instruction does it then do immediately after the mtmsr? I mean, the pc has some value, but how does that get interpreted without MSR_IR set? And yes, the (unused, it seems) disable_addr_trans routine in start.S does seem to be somewhat more complicated than a simple RMW of MSR, suggesting that if disabling address translation was really needed before reset, it should be done using that rather than by just clearing the bits directly. Rasmus
