Steam hit by major security breach

Re:Steam hit by major security breach by delinka on Monday July 27, 2015 @01:58PM (#9956845)

I was willing to concede your point, but I tested first. Using clang to compile, I did some profiling on my function optimized to various levels (-O0...-O3, -Os, -Ofast) with various inputs. The variation in timing is noise (I get the same variation on the same test inputs in different runs of the program.) I'd be curious to know whether you can provide a combination of compiler and compiler options that do indeed produce noticeable variations in timing as a function of the inputs.

"You cannot write a constant-time function in portable C / C++..." perhaps you mean that we can't outsmart compiler optimizations. Without optimizations ("Debug" configuration), the code runs exactly as written.

Ultimately, I'm going to prefer code with obvious intent regardless of how the compiler will optimize the thing. We'll just have to find a way around this timing attack business.

[ Parent | on HN ]

Re:Steam hit by major security breach by TheLoneWolfling on Tuesday July 28, 2015 @08:39AM (#9960985)

I would be very surprised if a compiler on a modern high-performance processor actually did this optimization - as deep pipelines and speculative execution mean that the additional conditional branch per loop may hurt more than it helps.

That being said, on a simple architecture - namely a shallow pipeline - this sort of thing is an "obvious" optimization.

And w.r.t. "without optimizations" - there is no such thing at the C / C++ level. There are computer architectures where optimizations pretty much have to be done, for instance. (Many architectures with compile-time scheduling, for instance.)

For instance, if you have a dataflow machine it may end up running this in non-constant time. And I'll note that modern x86 processors are looking more and more like dataflow machines.

[ Parent | on HN ]