Hi Pascal
On 16.03.2015 11:29, Pascal Brand wrote:
Hello,
There are lots of uint64_t in the proposal, and I wonder how it will behaves on 32bits platforms, in terms of performances, ease to use,...
This interface is not intended to be used directly by client code, instead it's supposed to be wrapped in a client lib. I'm trying to follow the guide lines in https://www.kernel.org/doc/Documentation/ioctl/botching-up-ioctls.txt to avoid the need of having one 64-bit and one 32-bit api.
Thanks Jens for the doc. However, it does not really answer: do you know if there are any performance drop using u64 types instead of native type on 32bits platforms? Or we simply don't care?
Pascal.
While a read to u64 might end up being a double load on 32bit platforms, the performance penalty will probably be minimum, since the they are in the same cache line and the read is performed seldom. The context switch(from the syscall) will have a much bigger performance penalty than reading a 64bit integer on a 32bit platform.
Also I think having the same interface between 32bit and 64bit kernels is a very good thing, since the same library will probably need to work with both.
Valentin