Re: [PATCHv2] KVM: optimize apic interrupt delivery

["Michael S. Tsirkin" <mst@redhat.com>]

On Wed, Sep 12, 2012 at 08:13:54AM -0700, Paul E. McKenney wrote:
> On Wed, Sep 12, 2012 at 03:44:26PM +0300, Gleb Natapov wrote:
> > On Wed, Sep 12, 2012 at 03:36:57PM +0300, Avi Kivity wrote:
> > > On 09/12/2012 03:34 PM, Gleb Natapov wrote:
> > > > On Wed, Sep 12, 2012 at 10:45:22AM +0300, Avi Kivity wrote:
> > > >> On 09/12/2012 04:03 AM, Paul E. McKenney wrote:
> > > >> >> > > Paul, I'd like to check something with you here:
> > > >> >> > > this function can be triggered by userspace,
> > > >> >> > > any number of times; we allocate
> > > >> >> > > a 2K chunk of memory that is later freed by
> > > >> >> > > kfree_rcu.
> > > >> >> > > 
> > > >> >> > > Is there a risk of DOS if RCU is delayed while
> > > >> >> > > lots of memory is queued up in this way?
> > > >> >> > > If yes is this a generic problem with kfree_rcu
> > > >> >> > > that should be addressed in core kernel?
> > > >> >> > 
> > > >> >> > There is indeed a risk.
> > > >> >> 
> > > >> >> In our case it's a 2K object. Is it a practical risk?
> > > >> > 
> > > >> > How many kfree_rcu()s per second can a given user cause to happen?
> > > >> 
> > > >> Not much more than a few hundred thousand per second per process (normal
> > > >> operation is zero).
> > > >> 
> > > > I managed to do 21466 per second.
> > > 
> > > Strange, why so slow?
> > > 
> > Because ftrace buffer overflows :) With bigger buffer I get 169940.
> 
> Ah, good, should not be a problem.  In contrast, if you ran kfree_rcu() in
> a tight loop, you could probably do in excess of 100M per CPU per second.
> Now -that- might be a problem.
> 
> Well, it -might- be a problem if you somehow figured out how to allocate
> memory that quickly in a steady-state manner.  ;-)
> 
> > > >> Good idea.  Michael, is should be easy to modify kvm-unit-tests to write
> > > >> to the APIC ID register in a loop.
> > > >> 
> > > > I did. Memory consumption does not grow on otherwise idle host.
> 
> Very good -- the checks in __call_rcu(), which is common code invoked by
> kfree_rcu(), seem to be doing their job, then.  These do keep a per-CPU
> counter, which can be adjusted via rcutree.blimit, which defaults
> to taking evasive action if more than 10K callbacks are waiting on a
> given CPU.
> 
> My concern was that you might be overrunning that limit in way less
> than a grace period (as in about a hundred microseconds.  My concern
> was of course unfounded -- you take several grace periods in push 10K
> callbacks through.
> 
> 							Thanx, Paul

Gleb noted that Documentation/RCU/checklist.txt has this text:

        An especially important property of the synchronize_rcu()
        primitive is that it automatically self-limits: if grace periods
        are delayed for whatever reason, then the synchronize_rcu()
        primitive will correspondingly delay updates.  In contrast,
        code using call_rcu() should explicitly limit update rate in
        cases where grace periods are delayed, as failing to do so can
        result in excessive realtime latencies or even OOM conditions.

If call_rcu is self-limiting maybe this should be documented ...

> > > Ok, thanks.
> > > 
> > > 
> > > -- 
> > > error compiling committee.c: too many arguments to function
> > 
> > --
> > 			Gleb.
> >