RFR: 8348384: RISC-V: Disable auto-enable Vector on buggy kernels

[Robbin Ehn]

On Thu, 23 Jan 2025 12:43:44 GMT, Ludovic Henry <luhenry at openjdk.org> wrote:

>> Hi please consider!
>> 
>> Linux kernels pre-6.9 can copy a dirty v-state during signal return path:
>> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/arch/riscv/kernel/signal.c?h=v6.13-rc1&id=c27fa53b858b4ee6552a719aa599c250cf98a586
>> 
>> We should not auto-enable vector with such kernel, as we must be able to receive signals during the execution of vector code.
>> 
>> Manual tested pre and post 6.9 as expected.
>> 
>> Thanks, Robbin
>
> src/hotspot/os_cpu/linux_riscv/riscv_hwprobe.cpp line 176:
> 
>> 174:         log.info("Linux kernels before 6.9 (current %ld.%ld) have a known bug when using Vector and signals.", major, minor);
>> 175:         log.info("Vector not enabled automatically via hwprobe, but can be turned on with -XX:+UseRVV.");
>> 176:       }
> 
> Couldn't you simply have:
> Suggestion:
> 
>       log_info(
>             "Linux kernels before 6.9 (current %ld.%ld) have a known bug when using Vector and signals.\n"
>             "Vector not enabled automatically via hwprobe, but can be turned on with -XX:+UseRVV.",
>             major, minor);

No, the second line will come out without proper decorators which causes issues for folks grepping/tailing and such from log files.

[0.048s][info ][os                   ] Linux kernels before 6.9 (current 6.11) have a known bug when using Vector and signals.
[                                    ] Vector not enabled automatically via hwprobe, but can be turned on with -XX:+UseRVV.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/23256#discussion_r1926941483