Paolo Bonzini <pbonz...@redhat.com> writes:

> On 21/03/2016 10:35, Peter Maydell wrote:
>> > x86 can actually do 64-bit atomic loads and stores through the FPU, but
>> > I'm not sure about other 32-bit targets?
>>
>> As I say, ppc32 linux is one that can't -- see commit 8a5956ad6392f1155
>> for last time this came up.
>
> Yes, I remember that.  Loads and stores are different from RMW
> operations, though.
>
>> Ideally we should put a compile-time assert in the atomic ops that
>> the data being operated on isn't larger than sizeof(void*) so that
>> this is a compile time error on all 32-bit hosts rather than just
>> a small subset of them.
>
> Yes, I will do it if Alex doesn't beat me to it.

I'm working on it now.

 - fix build failures on 32bit hosts (Linux/BSD)
 - add compile-time assert for sizeof <= sizeof (void*)

Anything else?

>
> Paolo


--
Alex Bennée

Reply via email to