Hi Keith, On 01.02.22 08:18, Keith Hui wrote: > Read my story then tell me what I need to know and do, and if I've > taken on too much. > > With my P8Z77-M board committed, I've restarted another decade-old > platform coreboot port. > > This time it's the M4A785TD-M EVO, another Asus board, also one where > another variant *was* already supported, but not that exact one. > > I said "was" because two years ago we dropped AMD family10 CPUs and a > number of northbridge/southbridge serving them because they have not > gained some important features.
unfortunately, by now there may be more to do than implementing those features. For instance the old (deprecated) allocator had some support for special things surrounding hyper-transport, IIRC. But I don't know AMD platforms as well to tell if this affects your system. Also, what could be considered current standards most probably changed. > > However this board is prime candidate for a port because it's 64-bit > but predates UEFI or PSP; its 1MB SPI flash was picked up by flashrom > with no extra work and the southbridge can take up to 16MB SPI meaning > I COULD fit a miniature Linux system in there. As I bought the board > along with a Phenom II 975 black CPU and DDR3 memory, it has as much > power as P8Z77-M and its i7-3770k and my daily machine the X230 > tablet, meaning it is still useful as a daily. > > I popped the heatsinks and have visually confirmed the chips I'm > dealing with: RS880 and SB710. Phenom II x4 975 black is a family 10h. > > I backed up their codes as present before the purge and would work off > of them. I wonder what held us up back then, was it a software limit > or just lack of attention? My goal is to bring the three pieces up to > current coreboot standards and do everything natively. I would guess "software limit". Much unlike the code around P2B these platforms were hard to maintain. I can't tell if the code just decayed or never was easy to maintain from the beginning. > > For the job I got these manuals: > RS880: Databook, Register programming requirements, RRG, BKDG > SB7x0: RRG, BKDG, Databook > Fam10: BKDG, revision guide > Plus the JEDEC DDR3 specs where info on mode registers are. > > So far I've done the bootblock inits of the board and SB710. I made > sure the SB710 code would enable the entire flash regardless of size. > Next I'll want to do CAR setup. The BKDG has specific instructions to > not use MMX or SSE instructions during CAR. Can we compile only some > units without SSE? That should certainly be possible. For instance for MMX we have an opt-out in the top-level Makefile: ifneq ($(CONFIG_MMX),y) CFLAGS_x86_32 += -mno-mmx endif You could have something like CFLAGS_bootblock += -mno-mmx > > What would be my next big hurdle, with a goal of coming up with a > native raminit like ivybridge? > > Is it time for a reality check before I dive too deep and start to > neglect Z77, or worse, P2B? Very hard to tell. It would keep you very busy to say the least. It might be worth a look how much code can be shared with other platforms. There's some effort for fam15, AIUI, and even the code of the newest Zen SoCs shows signs of rather old hardware blocks that could still be com- patible. Nico _______________________________________________ coreboot mailing list -- coreboot@coreboot.org To unsubscribe send an email to coreboot-le...@coreboot.org
