Messages in this thread | | | Subject | Re: [PATCH 2/5] perf: generic intel uncore support | From | Peter Zijlstra <> | Date | Tue, 03 Apr 2012 00:16:48 +0200 |
| |
On Sun, 2012-04-01 at 11:11 +0800, Yan, Zheng wrote: > > So what's up with all this box->lock business.. why does that lock > > exist? > > If user doesn't provide the "-C x" option to the perf tool, multiple cpus > will try adding/deleting events at the same time.
Right, however:
> > Aside from all this, there's still the problem that you don't place all > > events for a particular phys_id onto a single cpu. It doesn't matter > > which cpu in that package it is, but all events should go to the same. > > > > This means that on unplug of that cpu, you have to migrate all these > > events etc.. > > Any hints how to do this. I'm afraid it requires big changes to perf core.
Yes it'll need some core changes, but I don't think I'll be too bad, I'll try and write up something soonish..
| |