On Mon, Jul 09, 2018 at 02:55:16PM +0200, Peter Zijlstra wrote: > On Mon, Jul 09, 2018 at 05:34:57AM -0700, Paul E. McKenney wrote: > > But KVM defeats this by checking need_resched() before invoking > > cond_resched(). > > That's not wrong or even uncommon I think.
In fact, I think we recently put that pattern in crypto code in order to break up very long kernel_fpu sections. Note that you also 'broke' cond_resched_lock() as that no longer matches cond_resched().