On 06/03/2016 02:26 PM, Torvald Riegel wrote:
On Fri, 2016-06-03 at 12:03 +0200, Jakub Jelinek wrote:
I guess it is a tough decision. If you don't have HW instruction to read
say double word aligned integer atomically, if you don't implement atomic
load on it through compare and swap (which indeed won't work with read-only
memory), then you probably need to fall back to locks in libatomic.
And that would be fine, IMO. If you can't even load atomically, doing
something useful with this type will be hard except in special cases.
It's possible to do this (in an ABI-compatible fashion even) with kernel
support. I doubt it makes sense to try to fix this with lock fallback
(which breaks in so many unexpected ways).
Also, doing a CAS (compare-and-swap) and thus potentially bringing in
the cache line in exclusive mode can be a lot more costly than what
users might expect from a load. A short critical section might not be
much slower.
The performance problem is difficult to address, of course. Kernel
emulation likely isn't cheap at all.
Florian
