RFR: 8309258: RISC-V: Add riscv_hwprobe syscall [v2]

[Feilong Jiang]

On Tue, 13 Jun 2023 12:34:28 GMT, Robbin Ehn <rehn at openjdk.org> wrote:

>> Hi, please consider.
>> 
>> Linux kernel 6.4 will come with the new syscall https://docs.kernel.org/riscv/hwprobe.html to determine CPU and features.
>> RV cpus features/capabilities can vastly differ and it is not feasible for users to manually enable the correct feature set.
>> Today the VM uses the ELF aux vector and cpuinfo to gather some information about CPU capabilities.
>> 
>> Currently features are track with a bit field, this is insufficient.
>> There are many capabilities and these can have values attached to them.
>> CPU features should also be possible to turn if we can determine vendor (hwprobe).
>> 
>> This patchs adds the syscall, uses the syscall in combination with the aux and cpuinfo to enable features by default.
>> If there is a vendor specific path it calls that in addition.
>> Then we build the feature string(and bit field) and update flags accordingly.
>> 
>> Tested t1 and hwprobe with:
>> https://lore.kernel.org/qemu-devel/7f8d733df6e9b6151e9efb843d55441348805e70.camel@rivosinc.com/
>
> Robbin Ehn has updated the pull request incrementally with one additional commit since the last revision:
> 
>   free uarch string

src/hotspot/os_cpu/linux_riscv/vm_version_linux_riscv.cpp line 93:

> 91: 
> 92:   RiscvHwprobe::probe_features();
> 93:   os_aux_features();

Looks like `probe_features` and `os_aux_features` are doing the same thing (enabling supported extensions). Should we call them both? Maybe it's better to call `os_aux_features` when `probe_features` returns false? e.g.:


if (!RiscvHwprobe::probe_features()) {
  os_aux_features();
}

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/14445#discussion_r1229129540