RFR: 8324580: SIGFPE on THP initialization on kernels < 4.10 [v4]

[Thomas Stuefe]

On Thu, 8 Feb 2024 11:04:28 GMT, Stefan Johansson <sjohanss at openjdk.org> wrote:

> For this reason it would have been good if `thp_pagesize()` returned a proper value, but again, I see the arguments for keeping it true to the configuration.

Yeah, I am almost convinced now that modifying `HugePages::thp_pagesize()` would have been better. Maybe with an extra boolean flag that means "We did guess the THP page size".

> Btw. there is one more use of HugePages::thp_pagesize for [THPStackMitigation](https://github.com/openjdk/jdk/blob/a9c6e87c6ac8f278e4e241a77aef3479f7b5f2e4/src/hotspot/os/linux/os_linux.cpp#L978). It currently only does something for known thp_pagesize (>0). Should I change it to make use of os::large_page_size to also do "mitigation" based on assumed page size?

No, definitely not; for the reason @kstefanj mentioned. We don't want 1GB stacks if someone uses 1GB pages :=)

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

PR Comment: https://git.openjdk.org/jdk/pull/17545#issuecomment-1935612916