Re: [hatari-devel] Add full support for STF/STE MMU

[ Thread Index | Date Index | More Archives ]

Just tell me if you need a falcon test.

Le 04/01/2017 à 23:42, Nicolas Pomarède a écrit :

I completed the new part that emulates STF/STE MMU and the different address translations we can get depending on the MMU settings for each of the 2 banks (128 KB, 512 KB or 2048 KB).

Thanks a lot to Christian for his documentation on the address mapping to the CAS/RAS signals.

On STF, the MMU will default to the more versatile version where bank0 and bank1 can have different values (ie the non-IMP version)

On STE, the MMU will always use bank0 size for bank1.

Although very rarely used, this emulation of $ff8001 will allow TOS to boot and detect RAM without having to patch the memvalid variables ourselves.

When STF/STE is used with 4 MB or less, then we now use another versions of the lget/wget/bget/lput/wput/bput functions for RAM which do an additionnal address translation based on the MMU setting. These translations were compared to a real STF for all 9 possible settings, Hatari results should be similar to real HW.

When MMU is configured to match the physical RAM size, then we default to the usual memory access functions (without translation), so there will be no speed penalty in 99% of the time, only a slight overhead during boot when TOS/Emutos change MMU config to detect RAM size.

This emulation can be useful to Emutos devs if they want to test their RAM detection code without using real HW.

It's also interesting as it now allows STF/STE to boot completely without any patch in TOS (assuming gemdos HD is not used, only floppy), thus making Hatari even more "stealth" and closer to the original HW.

Code is not 100% completete yet, I will add later code for the TT translations too (according to TOS 3, there're only 2 possible MMU settings) ; as for Falcon, it seems TOS 4 doesn't use FF8001 at all, so this register might be ignored (this needs more tests)


Mail converted by MHonArc 2.6.19+