RISC-V ACLINT Support

[ninolomata]

Features:

• IPIs sent via ACLINT instead of SBI ECALLs

Notes:
Linux -> https://github.com/avpatel/linux/tree/riscv_aclint_v5
QEMU -> 7.2 with flag -M aclint=on

[josecm]

I agree we should support the ACLINT to take advantage of faster IPIs in the platforms thave have one. However, I don't think virtualizing the ACLINT is useful because the cost of trap-and-emulate is larger than the direct hypercall. I'd say it would be better we don't support a virtual ACLINT, until maybe in the future the ACLINT adds support for virtualization or we find some guest that only supports IPIs through the ACLINT.

[josecm]

Please also, squash the two commits in a single one.

[ninolomata]

@josecm I addressed all your comments in this recent update.

[josecm]

Actually, since the information about the clint being present or not is already present in the platform description I was thinking about adding to the platform define scripts the automatic generation of the macro. In a scripts/arch/riscv/platform_defs_gen.c

...
    if (platform.arch.aclint_sswi.base != 0) {
        printf("#define ACLINT_PRESENT");
    }
...

I'll be happy to do this myself if you don't have the time as it will involve a little more work for supporting arch-specific macro definition scripts.

[ninolomata]

@josecm I did as you ask and created a platform specific script. If you want, I can create a separate commit just for that.

[josecm]

@josecm I did as you ask and created a platform specific script. If you want, I can create a separate commit just for that.

Thanks! Yes, a separate commit for that would be great!

[josecm]

@ninolomata Can you resolve the conflicts in src/platform/qemu-riscv64-virt/virt_desc.c ?

[ninolomata]

@josecm all good now.

[josecm]

Now the code seems fine, but I think we need to add also a build check for compiling with IPIC == IPIC_ACLINT. I'll add this later today.

[josecm]

@DavidMCerdeira @danielRep @sandro2pinto

[sandro2pinto]

All good to me.