On Fri, Sep 29, 2017 at 01:01:12AM +1000, Nicholas Piggin wrote: > That's fine. If a user is not bound to a subset of CPUs, they could > also cause disturbances with other syscalls and faults, taking locks, > causing tlb flushes and IPIs and things. So on the big SGI class machines we've had trouble with for_each_cpu() loops before, and IIRC the biggest Power box is not too far from that 1-2K CPUs IIRC. Bouncing that lock across the machine is *painful*, I have vague memories of cases where the lock ping-pong was most the time spend. But only Power needs this, all the other architectures are fine with the lockless approach for MEMBAR_EXPEDITED_PRIVATE. The ISYNC variant of the same however appears to want TIF flags or something to aid a number of archs, the rq->lock will not help there.
