[Lldb-commits] [PATCH] D117928: [lldb] Disable tests for x86 that uses write command on XMM registers

Tue Jan 25 04:37:01 PST 2022

labath added a comment.

Thanks for tracking this down. I'm looking forward to seeing how the kernel issue gets resolved.

> In D117928#3266895 <https://reviews.llvm.org/D117928#3266895>, @labath wrote:
>
>> If you want, I can try to create a patch for this, though it might take me a couple of days to get around to it.
>
> I'm probably not getting the full picture but my idea on this is: if we propose this change due to this specific kernel regression is not worth it, although we should consider it if there is any other benefit on using `PTRACE_POKEUSER`.

PTRACE_POKE/PEEKUSER might be theoretically faster for some use cases, as it can access only some bytes instead of the whole XSAVE block, but we're nowhere near doing those kinds of optimizations.

As for being worth it, we have added workarounds for kernel bugs in the past (try grepping for SingleStepWorkaround), so this wouldn't be the first one and, if done right, it also shouldn't be particularly burdensome.

I'd say it all comes down to how much you care about this particular configuration having a clean test run. My position is that I'd rather have a workaround in place than to have to figure out how to skip this test for this exact setup.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D117928/new/

https://reviews.llvm.org/D117928