On Thu, 2015-04-16 at 15:07 +0100, Ian Campbell wrote: > On Thu, 2015-04-16 at 14:19 +0100, Jan Beulich wrote: > > >>> On 16.04.15 at 14:55, <ian.campb...@citrix.com> wrote: > > > On Thu, 2015-04-16 at 13:49 +0100, Tim Deegan wrote: > > >> At 12:49 +0100 on 16 Apr (1429188559), Jan Beulich wrote: > > >> > >>> On 16.04.15 at 13:25, <t...@xen.org> wrote: > > >> > > But also: AFAICS the GCC builtin __sync_fetch_and_add() does almost > > >> > > exactly this (the difference being that those are also compiler > > >> > > barriers where this is only a CPU barrier). Should we be using it > > >> > > instead? > > >> > > > >> > I'm afraid that's useful only from gcc 4.5.x onwards; earlier versions > > >> > (on x86 at least) simply generate a function call relying on a library > > >> > to > > >> > implement it. > > >> > > >> Darn. :( > > > > > > Might it still be a better generic fallback than the C/cmpxchg one? > > > > > > On x86 you'd get the asm version, and on ARM you need 4.5.x onwards > > > > Is that the case even for ARM32? > > Erm, I thought so when I wrote it but now I think again I'm not actually > so sure. > > > > anyway, plus in general the ARM folks seem to recommend using the > > > builtins more anyway. > > > > I assume you checked that 4.5.x actually expands said builtin for > > ARM (as I only checked x86)? > > No, I hadn't, I should though, you are right.
If nm on libgcc.a is any indication (it should be?) then it expands on arm64 but is an out of line function on arm32, so like Tim said: Darn. Ian. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xen.org http://lists.xen.org/xen-devel
