Alexandre Ghiti <alexgh...@rivosinc.com> writes: > The satp mode is set using the svXX properties, but that actually > restricts the satp mode to the minimum required by the profile and > prevents the use of higher satp modes. > > Fix this by not setting any svXX property and allow all satp mode to be > supported.
I figured I'll add some more findings. The RISC-V profile initialization seems a bit brittle; Without Alex' fix we're getting different supported satp on the first hart (0) than the other harts; Hart 0 has sv57, whereas the other harts has enforced sv39. This caused the smp bringup on Linux to fail. Some observations on, e.g., a "rv64,b=on,zbc=off,v=true,vlen=256,elen=64,sscofpmf=on,svade=on,svinval=on,svnapot=on,svpbmt=on,zcb=on,zcmop=on,zfhmin=on,zicond=on,zimop=on,zkt=on,zvbb=on,zvfhmin=on,zvkt=on,zkr=on" 8 hart machine: When realizing the cpus, the first cpu calls riscv_cpu_add_profiles() all profiles are disabled, whereas for the other cpu calls to riscv_cpu_add_profiles() have some profiles enabled. Having some profiles enabled, will issue a call to cpu_set_profile() that will enforce the satp code that Alex removes in this patch. The riscv_cpu_validate_profile() function is called after riscv_cpu_add_profiles(), which explains why the harts subsequent the first one will get the cpu_set_profile() call. The first hart will not have cpu_set_profile() called. IOW, there are more issues hiding here, in addition to Alex' fix. Björn
