search

sorear / riscv-specs

UNOFFICIAL bug tracker for the RISC-V specifications
4 stars 1 forks source link

sbi_hart_id versus the boot protocol #27

Closed sorear closed 7 years ago

sorear commented 7 years ago

Ron Minnich points out that sbi_hart_id through an upcall is a bit weird: https://groups.google.com/a/groups.riscv.org/d/topic/isa-dev/frrGFW95j1s/discussion

The simplest proposal at this point is to kill the SBI call and just provide the hart id in the bringup code, requiring the supervisor to track it in sscratch/tp if it cares.

neuschaefer commented 7 years ago

This seems to require that the firmware brings up all harts. Such a requirement should be made explicit once the boot protocol is specified. How will a hart find its hartid when it's started at runtime?

sorear commented 7 years ago

priv-1.10 passes hart id in a0 and device tree pointer in a1