Re: [GIT PULL] First batch of KVM changes for 4.1

[Paolo Bonzini <pbonzini@redhat.com>]

On 17/04/2015 11:17, Peter Zijlstra wrote:
> On Fri, Apr 17, 2015 at 10:52:38AM +0200, Peter Zijlstra wrote:
>> On Fri, Apr 10, 2015 at 05:01:29PM +0200, Paolo Bonzini wrote:
>>>  include/linux/sched.h                              |    8 +
>>>  kernel/sched/core.c                                |   15 +
>>
>> Can you please not puke over the scheduler without Acks from at least
>> one maintainer?

Sorry, this was done while I was not handling the KVM tree.  At the very
least the commit message should have included the original hashes of the
commit and the revert.  This way one could have found the original Acks:

    commit 582b336ec2c0f0076f5650a029fcc9abd4a906f7
    Author: Marcelo Tosatti <mtosatti@redhat.com>
    Date:   Tue Nov 27 23:28:54 2012 -0200

    sched: add notifier for cross-cpu migrations

    Originally from Jeremy Fitzhardinge.

    Acked-by: Ingo Molnar <mingo@redhat.com>
    Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>


>> I complained about this very thing two years ago:
>>
>>   http://marc.info/?l=linux-kernel&m=137345253916751
>>
>> And now it magically re-appears WTF!
> 
> And I really don't understand _why_ you need that extra callback in the
> first place. You already have preempt notifiers, just track if you came
> in on another cpu than you went out on and voila!

Then you pay for _all_ preemptions of _all_ processes in the guest,
instead of the hopefully rare ones that do a CPU migration.

Preempt notifiers are registered on current only, this one is global.

Of course, adding a static key is a good idea.  I can also add a config
symbol, selected by paravirt, if you want.

Paolo