On Tue, Dec 17, 2019 at 10:05:53AM -0800, Linus Torvalds wrote:

On Tue, Dec 17, 2019 at 10:04 AM Linus Torvalds
[off-list ref] wrote:

quoted

Let me think about it.

.. and in the short term, maybe for code generation, the right thing
is to just do the cast in the bitops, where we can just cast to
"unsigned long *" and remove the volatile that way.

Yeah, I think I'll spin that patch series tomorrow anyway, since I don't
think we need to hold it up.

I'm still hoping there's a trick, but..

Well, there's always Peter's awful hack [1] but it's really gross. FWIW,
I've pushed the handful of patches I have to [2], which drop the GCC 4.8
workaround and introduce a non-atomic version instead of the
'__builtin_memcpy()'.

Will

[1] https://lore.kernel.org/lkml/20191213125618.GD2844@hirez.programming.kicks-ass.net (local)
[2] https://git.kernel.org/pub/scm/linux/kernel/git/will/linux.git/log/?h=rwonce/cleanup