On Mon, Jan 29, 2018 at 10:30:05AM -0800, H. Peter Anvin wrote: > On 01/29/18 10:26, Andy Lutomirski wrote: > >>> > >>> That will utterly suck on non-UMIP machines that have > >>> hypervisor-provided UMIP emulation. > >> > >> Is that a valid thing to optimize for, especially given that paranoid > >> entries aren't the most common anyway? > > > > A bunch of people seem to care about NMI performance for perf. > > > > That wasn't really the question... > > > And the current patch set works without this trick. > > But I believe the tricks it uses are fragile. > > > FWIW, if we switch all entries to the entry text trampoline, we get direct > > percpu access for free. > > That might be a better option.
Has there been any conclusion to this thread? I can still reproduce the issue on mainline and next. Thanks, Dan
