On 04/24/2014 06:23 PM, Tobias Brunner wrote:
Hi Christoph,
Thanks for testing the code.
The check is disabled completely if these built-ins are not available,
because fallbacks (like those for ref_*) would not improve the
performance. But since even the atomic implementation allows races an
alternative approach would be to use __atomic* if available and just
directly read and write if not (thus avoiding the __sync penalty). We
could make max_[v]level an int32_t array to be reasonably safe.
This sounds reasonable to simply read and write when atomic macros are
not available. I may be wrong, but I think that modern processors always
read or write an int32_t atomically, provided it is aligned on a 32 bit
boundary.
I think so too, but I feel safer not relying on assumptions :) So I
merged the branch as is.
Regards,
Tobias
Hi Tobias,
Great, thanks!
Christophe
_______________________________________________
Dev mailing list
[email protected]
https://lists.strongswan.org/mailman/listinfo/dev
