From mboxrd@z Thu Jan 1 00:00:00 1970 From: Boris Ostrovsky Subject: Re: [PATCH v10 11/20] x86/VPMU: Interface for setting PMU mode and flags Date: Fri, 12 Sep 2014 11:54:41 -0400 Message-ID: <54131741.90407@oracle.com> References: <1409802080-6160-1-git-send-email-boris.ostrovsky@oracle.com> <1409802080-6160-12-git-send-email-boris.ostrovsky@oracle.com> <541084C8020000780003366A@mail.emea.novell.com> <54108C50.7030500@oracle.com> <541160E90200007800033A7B@mail.emea.novell.com> <5411ADDD.2040003@oracle.com> <5411D4DF0200007800034005@mail.emea.novell.com> <5411C963.2000002@oracle.com> <5412B383020000780003449D@mail.emea.novell.com> <5412FF55.1090605@oracle.com> <541321CD02000078000348EB@mail.emea.novell.com> <54130B33.3020907@oracle.com> <54132DA5020000780003493C@mail.emea.novell.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii"; Format="flowed" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <54132DA5020000780003493C@mail.emea.novell.com> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xen.org Errors-To: xen-devel-bounces@lists.xen.org To: Jan Beulich Cc: tim@xen.org, kevin.tian@intel.com, keir@xen.org, suravee.suthikulpanit@amd.com, andrew.cooper3@citrix.com, eddie.dong@intel.com, xen-devel@lists.xen.org, Aravind.Gopalakrishnan@amd.com, jun.nakajima@intel.com List-Id: xen-devel@lists.xenproject.org On 09/12/2014 11:30 AM, Jan Beulich wrote: >>>> On 12.09.14 at 17:03, wrote: >> My concern is that on a very large system we may be getting lots of >> preemptions and thus this will be failing for the sysadmin somewhat >> often and without clear reason. (I am basing this assumption on the fact >> that I am currently looking at a ~200CPU box with gobs of memory and >> some operations that use contuations get preempted very often. Hundreds >> of times. I don't remember seeing this on "regular" systems). > Yeah, quite likely due to IPI counts growing approximately with the > square of the CPU count. > > We've been considering adding a "minimum amount of progress" > mechanism to various places where preemption can (at least under > some conditions) be observed on every iteration. Perhaps along > with having the 5s timeout you also want a minimum amount of > time during which you don't check for preemption (yet)? Yes, I can do that. Something like half second? This is a very infrequent operation that can only be done in dom0 by a sysadmin so even if this seems a bit too long I think it's safe enough. -boris