“The HiFive Premier P550 and Milk-V Jupiter now work, despite their incoherent DMA. Their CPUs are pretty fast (and Jupiter has 8!) but they take more elapsed time than I think they should; this may be the cost of incoherent DMA.”
“My Jupiter has 8 GB of RAM and 8 1.6 GHz Spacemit X60 cores. The caches are nominally all coherent with each other. Like the Beagle V, the Jupiter’s DMA is incoherent with its caches, which is a shame. To allow manually compensating, it includes the Zicbom and Svpbmt extensions.
The Ethernet controller is Spacemit’s own design, the K1-X. Configuring its clock signals took more effort than it should have: Spacemit’s documentation and the Linux driver weren’t entirely consistent.”
Note, there is no iso or disk image for risc-v, you have to install Plan 9 on another machine or a virtual machine, grab his kernel and compile. You can try booting the Jupiter over network, that’s the way new Plan 9 port are tested. Geoff have a nvme driver, that’s worth trying too.
Skimming through http://www.collyer.net/who/geoff/9/9k-riscv.pdf, and what I read is that the author’s trying to circumvent SBI, which totally disregards the architecture of risc-v. Also he appears to be using the x86 way of thinking that there should be a UART at some fixed address. Instead of treating it as very legacy design of x86, he’s actually in favor of it. The claim of OpenSBI having bugs in sbi_get_hart_status is baseless, too, as there’s no concrete explanation of what went wrong. In general, I don’t think he understands risc-v privileged ISA.