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

[Christoph Lameter (Ampere)]

On Fri, 28 Jun 2024, Catalin Marinas wrote:

> Linux-specific (e.g. madvise()), I agree, but arm64-specific definitely
> not. I'd argue that expecting the atomic_add(0) to only trigger a single
> write fault is arch specific. You can't do this on arm32 or arm64
> pre-LSE (I haven't checked other architectures).

The single write fault is x86 behavior. I am not sure how other 
architectures handle that.

> IIUC, OpenJDK added this feature about two years ago but the arm64
> behaviour hasn't changed in the meantime. So it's not like we broke the
> ABI and forcing user space to update.

The focus of OpenJDK may not be arm64 and they never saw the issue? We 
only know this because we have an insider on staff. AFACIT we get pushback 
from them as well. There are certainly numerous other open 
source applications that behave in a similar way. We just dont know about 
it.

> This patch does feel a bit like working around a non-optimal user choice
> in kernel space. Who knows, madvise() may even be quicker if you do a
> single call for a larger VA vs touching each page.

Looks to me like unexpected surprising behavior on ARM64. madvise is 
rather particular to Linux and its semantics are ever evolving.

> > A lot of these are proprietary.
>
> Are you aware of other (proprietary) software relying on such pattern to
> fault pages in as writeable?

I would not be told about such things by companies I did not work for and 
if I have gotten knowledge about this in some way in the past then I would 
not be allowed to talk about it.