shartid CSR

[aousherovitch-tt]

Hello,
could you please elaborate on why this CSR was needed and how it is used by your OPTEE platform implementation in https://github.com/Nuclei-Software/optee_os/blob/34e68e2b1b27e5222aaee4d9956aa2e53bd98808/core/arch/riscv/plat-nuclei/main.c#L32 (itr_core_handler) ?
There are also other uses other than itr_core_handler(), the optee core code is relying on another wrapper for itr_core_handler() - get_core_pos() in https://github.com/Nuclei-Software/optee_os/blob/34e68e2b1b27e5222aaee4d9956aa2e53bd98808/core/arch/riscv/include/kernel/misc.h#L18. The latter is used extensively.
Have you considered the alternatives to having this CSR discussed here for example: riscv-non-isa/riscv-sbi-doc#25 ?
Thank you

[fanghuaqi]

We just add this csr to avoid ecall to m-mode to get hartid, @matthewgui could you provide more details

[matthewgui]

hello, @aousherovitch-tt

Hello, could you please elaborate on why this CSR was needed and how it is used by your OPTEE platform implementation in https://github.com/Nuclei-Software/optee_os/blob/34e68e2b1b27e5222aaee4d9956aa2e53bd98808/core/arch/riscv/plat-nuclei/main.c#L32 (itr_core_handler) ?

Because we use plic as interrupt controller，its claim register is percpu context(https://github.com/riscv/riscv-plic-spec/blob/master/riscv-plic.adoc#2-memory-map), so core hartid is needed.

There are also other uses other than itr_core_handler(), the optee core code is relying on another wrapper for itr_core_handler() - get_core_pos() in https://github.com/Nuclei-Software/optee_os/blob/34e68e2b1b27e5222aaee4d9956aa2e53bd98808/core/arch/riscv/include/kernel/misc.h#L18. The latter is used extensively. Have you considered the alternatives to having this CSR discussed here for example: riscv-non-isa/riscv-sbi-doc#25 ?

yes, optee core code is relying on core id, which is determined by the optee framework, in our current hardware environment, using hartid is convenient for us.
we cannot save hartid into thread_core_local, because this structure is determined by hartid, some case hartid is not in parameter. for example normal world return from RPC service. Maybe it is feasible to calculate hartid using the current stack value, because
stack tmp is bound to hartid, but this method requires consideration of multiple scenarios.

[aousherovitch-tt]

Thank you, @fanghuaqi and @matthewgui

[fanghuaqi]

Hi @aousherovitch-tt , we just pushed changes to support mhartid in optee, you can check our latest commit, it is configurable now, take a try.

[aousherovitch-tt]

Hi @fanghuaqi and @matthewgui, your changes work when CFG_SHART_FEATURE is disabled. Thank you. In the upstream this appears to be solved differently, without an ecall to opensbi, but I was not able to verify the upstream behavior because there's no yet available integrated build environment, so not sure how they are testing their RISCV patches, frankly. At some point, I assume, if you still wanted to upstream your solution, it would need to be merged.