Messages in this thread | | | Date | Fri, 25 Mar 2011 20:21:45 +0100 | From | Ingo Molnar <> | Subject | Re: [PATCH RFC] x86: avoid atomic operation in test_and_set_bit_lock if possible |
| |
* Andi Kleen <andi@firstfloor.org> wrote:
> > Also seriously complicated by the kexec case where the previous kernel > > didn't clean up PMU state. There is simply no sane way to detect if its > > That's a good point, but we can easily stop the PMU before kexec.
Wrong - there's lots of existing Linux versions out there that will kexec with PMU state active. We cannot change them retroactively.
> > actually used and by whoem. > > You check if the counter is enabled. If it's already enabled it's used by > someone else.
Wrong - or it can be enabled if it was left enabled accidentally. We treat PMU state like we treat other CPU state.
> > The whole PMU 'sharing' concept championed by Andi is utter crap. > > Why? It's the same thing as having some less counters.
Wrong again - 25% or 50% of the events stolen with no user choice is a pretty big deal.
Consider the example in this thread: cachemiss profiling done via perf, which needs two events. If one of the generic counters is 'stolen' by a BIOS and Linux accepts this silently then it means the loss of real functionality.
In other words, '25% of a specific hardware functionality stolen by the BIOS' (or more) is absolutely not acceptable.
> [...] You need to already support that for architectural perfmon with > variable counters anyways or for sharing with oprofile.
Wrong, that's different - multiplexing the PMU is obviously done within the OS. It's evidently user configurable - users can use oprofile or perf - and the user can still fully utilise the PMU to the extent he wishes to - it's his hardware.
It is not possible for the kernel to stop the BIOS from using the PMU though, so it's not 'sharing' no matter what 'protocol' is used, it's 'stealing'.
> > As for simply using it despite the BIOS corrupting it, that might not > > always work, the BIOS might simply over-write your state because it > > one-sidedly declares to own the MSRs (observed behaviour). > > Yes, that doesn't work. If someone else is active you have to step back. > > > Its all a big clusterfuck and really the best way (IMO) is what we have > > now to put pressure on and force the BIOS vendors to play nice. > > It just results in users like Eric being screwed. I doubt that any > BIOS writer ever heard about it. Congratulations for a great plan.
I'd encourage you to think through this topic instead of making derisive comments about others ...
Thanks,
Ingo
| |