Rose Thompson
Rose Thompson
Fixed with --cpu/aligned_uncached_PBMT=F. Thank you Lee!
If I am understanding` fdt_open_into()` correctly, buf is the destination array and fdt is the source. Zeroing the destination shouldn't erase the existing FDT. I've confirmed this hack does indeed...
I must say I'm baffled how CVW could have booted with that hack then. Your patch worked! Lockstep successfully made it through Linux's crc32 body function. We are now mismatching...
Hold on a second. I reran without lockstep and we are stuck during Linux boot. I'll need to confirm what's happening to make sure this isn't hardware bug.
I found what I think the root cause is. OpenSBI makes two calls to fdt_open_into with the length of the device tree + 32 and then again with the length...
I wouldn't be surprised if Linux just ignores the crc32 error.
This definitely fixed the issue and lets the Linux boot continue. I found another issue immediately after. The MCAUSE, MEPC, and several other CSRs are continuously enabled while StallW is...
Yes we should do open an issue with OpenSBI. I'm going to be busy for the next few hours, but when I get back I'll separate out the code for...
My solution is much uglier than I like and is probably not the correct solution. ftd_open_into(const void *fdt, void *buf, int bufsize) does not initialize buf and has multiple path...
Closing as OpenSBI has resolved.