Re: [PATCH 5/5] remove HAVE_ARCH_CMPXCHG

["David S. Miller" <davem@davemloft.net>]

From: Russell King <rmk+lkml@arm.linux.org.uk>
Date: Wed, 14 Sep 2005 23:03:53 +0100

> What business has userspace got of telling whether cmpxchg works on
> an architecture by looking at kernel headers?

Russell, please don't fly off the handle like this.

Nick is talking about something slightly different.

Things, for example, like DRM assume there is a cmpxchg()
they can use in the kernel and userland for interlocking.

DRM doesn't actually _check_ anything to see if this is the case, it
just so happens to only get enabled on platforms where cmpxchg() is
available in this fashion.

What Nick is suggesting is to actually move HAVE_ARCH_CMPXCHG or
something like it into the Kconfig so that things like DRM can
actually do the correct dependency check.  If you want something like
"HAVE_ARCH_CMPXCHG_WHICH_CAN_INTERFACE_WITH_USERSPACE" that DRM can
check too, all the better.

It's not about whether userspace can include some kernel header
and get cmpxchg(), it's whether there is some way that a cmpxchg()
shared semaphore between userspace and kernel is possible, which
things like DRM depend upon having available.