On Mon, 2016-10-17 at 15:48 +0200, Christian Borntraeger wrote: > On 10/17/2016 03:40 PM, Radim Krčmář wrote: > > > > 2016-10-14 17:02+1100, Suraj Jitindar Singh: > > > > > > Various kvm vm and vcpu stats are provided via debugfs entries. > > > Currently there is no way to reset these stats back to zero for > > > the system > > > except by stopping all vms. > > It is not really resetting, just a special case of current behavior > > -- > > top-level stats are a sum of all existing VMs/VCPUs, so we would > > see 0 > > if they all had 0. > > > > > > > > Add the ability to clear (reset back to zero) these stats on a > > > per stat > > > basis by writing to the debugfs files. > > I can see this being used before a test, but the userspace could > > also > > read the value instead of writing and subtract it from the value > > after > > the test to get a similar result. > > > > The userspace-only solution is insufficient? > > > > When and what do you zero? > > > > > > > > The stats are just reset > > > to zero > > > irrespective of what is actually written to the file. > > Sounds good. Setting the val would be as simple, but there > > shouldn't be > > a reason to and it is a debug interface, so we can change it > > anytime. > > > > > > > > Signed-off-by: Suraj Jitindar Singh <sjitindarsingh@xxxxxxxxx> > > > --- > > I have applied it to kvm/queue for the moment, thanks. > FWIW, I was thinking about another change: provide the accumulated > values of > the counters of all VMs that have been destroyed (basically on > vm_destroy read > the values and add them on top of the "forever" counters. > Did not find the time to do something like this. This doesn't seem like it would be particularly hard to implement, I'm just not sure of the use case. > > > -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
