On Fri, Apr 22, 2016 at 07:57:11PM +0100, Maciej W. Rozycki wrote: > On Fri, 22 Apr 2016, James Hogan wrote: > > > The PIT timer is slow, especially under virtualisation, compared with > > the r4k timer, and doesn't really provide any advantage on Malta since > > it doesn't support clock scaling (which is where the PIT has an > > advantage). > > I'm not sure I'm able to parse this correctly: are you saying that the > i8254-compatible PIT in Malta's southbridge does not support clock scaling > for some reason, unlike the same implementation on other platforms? > What's the reason then, the southbridge is the same as in many x86 PCs? Sorry, that wasn't written very well. the "it" on 3rd line is referring to our malta support. I mean something more like this: The PIT timer is slow under KVM+QEMU virtualisation compared with the r4k timer, and doesn't really provide any advantage on Malta since we don't support clock scaling on Malta, which is where the PIT would have an advantage over the CPU clock based r4k timer. > > > Drop the use of the PIT timer on Malta so that the r4k or GIC timer will > > be used in preference. > > Not everyone uses virtualisation, so it's a functional regression for > them. Can't you lower the priority for the timer instead so that it is > not selected by default, just as it's done with other platforms providing > a choice of timers? I'll look into that. Looking back at my IRC logs I suspect I meant to check why the PIT was taking priority before submitting upstream, but forgot. Thanks! James
Attachment:
signature.asc
Description: Digital signature
