Re: copy_from_user() overhead

[Tim Hockin]

> We are interested in finding out the overhead in kernel->user buffer copy and
> vice-versa for disk reads and network writes resp. in copy_from_user() and
> copy_to_user() calls. Are there any quantitative data available for this? We
> plan to eliminate this overhead, if significant, in our integrated disk-read and
> network-write architecture. Any pointers/references related to this type of
> work?

Perhaps you are thinking of something like IO-Lite?  Has anyone in the
linux world ever implemented IO-Lite?  Sure would be interesting to have
essentially zero-copy file operations..