Conversation
Palmer Dabbelt
palmer
We've got our first user-visible errata that actually breaks something: https://lore.kernel.org/all/CA+V-a8vT3AjnU1-s0k7ff0Y7WLofpHYnJPF+mKVnUspsrPvQtw@mail.gmail.com/
0
4
4
@tommythorn Yep, that's how RISC-V works. This is the same as the T-Head stuff, the ISA folks were very clear that vendors self-certify their implementations. None of those words in the PDF matter.
0
0
0
@tommythorn I haven't seen the v2, but every other SiFive-based chip has errata and given that the v1 was the most broken of the bunch I'm not super excited on that front.
IMO there's really no way around this: all hardware is a mess, it's just a matter of figuring out how to make it work.
IMO there's really no way around this: all hardware is a mess, it's just a matter of figuring out how to make it work.
0
0
0
@tommythorn that's what I'd hope too, but the RISC-V folks have been pretty adamant that it's up to vendors to decide on this one. The general idea was vendors would never ship broken hardware because that would be embarrassing for them, but it's not like that's ever worked before.
IMO we're way better off taking support for the craziness than telling them to fork the software stack. Everything we have is for embedded land right now, so if we push back upstream then vendors will likely just ignore us and ship whatever workarounds they need in their SDKs. Then we'll just have a bunch of incompatible RISC-V-like ports floating around, which would be a huge mess.
IMO we're way better off taking support for the craziness than telling them to fork the software stack. Everything we have is for embedded land right now, so if we push back upstream then vendors will likely just ignore us and ship whatever workarounds they need in their SDKs. Then we'll just have a bunch of incompatible RISC-V-like ports floating around, which would be a huge mess.
0
1
1
@tommythorn Yep, I agree with you again here. That was the argument against allowing T-Head to call their page table attribute stuff compatible with the ISA, it didn't work then and I don't see why it'd work now.
0
0
0
@cr1901 @brouhaha There's been a bunch of posts, I think I had it at some point but couldn't find it a few weeks ago when I looked. IIRC it's just ldconfig from some distro, maybe Debian? Prabhakar went through a pretty long public debugging process on this one, he's usually in #riscv on libera (not sure if he's around here somewhere).
0
0
2