>>> On 01.09.17 at 11:55, <jbeul...@suse.com> wrote:
>>>> On 01.09.17 at 10:37, <chao....@intel.com> wrote:
>> it seems add_sized() won't be a LOCKed instruction.
>> #define build_add_sized(name, size, type, reg) \
>> static inline void name(volatile type *addr, type val) \
>> { \
>> asm volatile("add" size " %1,%0" \
>> : "=m" (*addr) \
>> : reg (val)); \
>> }
>
> Oh, you're right. But then I'd still like you to not add a new
> user, as I don't see why it was introduced in the first place:
> Independent of architecture it is equivalent to
>
> write_atomic(p, read_atomic(p) + c)
>
> and hence I'd like to get rid of it as misleading/redundant.
Actually, on x86 it still is a bit better than the generic replacement,
i.e. it would only be worthwhile dropping the custom ARM variant
in favor of a generic one. Keep using it here then.
Jan
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel
