[v5 PATCH] arm64: mm: force write fault for atomic RMW instructions

[Will Deacon]

On Wed, Aug 21, 2024 at 11:18:42AM +0100, Catalin Marinas wrote:
> My view (as kernel maintainer) is that there aren't strong enough
> arguments for merging this patch. It does show an improvement in
> specific cases but the downside is the risk of affecting other cases
> (plus additional maintenance cost, though I'd expect the opcode checking
> to remain the same). I do think the architecture should be changed so
> that the kernel gets the right information in the fault handler but it
> would be a while before such feature appears in hardware.

I agree. This patch is a kludge to work around a bad assumption in OpenJDK
which can be (has been?) easily fixed in userspace.

Will