Content-Type: multipart/mixed; boundary="===============8245850828381411186==" MIME-Version: 1.0 From: Frederic Weisbecker To: lkp@lists.01.org Subject: Re: [nohz] 2a16fc93d2c: kernel lockup on idle injection Date: Wed, 17 Dec 2014 13:47:30 +0100 Message-ID: <20141217124724.GB20739@lerouge> In-Reply-To: List-Id: --===============8245850828381411186== Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable On Wed, Dec 17, 2014 at 10:11:58AM +0100, Thomas Gleixner wrote: > On Wed, 17 Dec 2014, Frederic Weisbecker wrote: > > On Tue, Dec 16, 2014 at 10:21:27PM +0100, Thomas Gleixner wrote: > > > So instead of evaluating the whole nonsense a gazillion times in a row > > > and firing pointless self ipis why are you not looking at the obvious > > > solution of sane state change tracking? > > = > > Because I had to do it wrong first before somebody like you with a fres= h mind > > comes, look at the whole picture and propose a rethink :-) > = > Point taken. > = > > > I told all of you from the very beginning that remote accounting > > > is a key mechanism for this, but you keep insisting on hacking > > > around it. > > = > > I don't, and I think everybody has understood we are not going to accept > > hacks to solve the 1hz issue. > = > You wish. > = > > I'm pretty sure you know what this is about to maintain a project with a > > (variably) cruel lack of manpower ;-) > = > Oh yes! :( > = > > Moreover I'm not the kind of kernel developer who ignores reviews (whet= her pre or > > post merge). > = > I know that. > = > > So, no point in threatening a dependency to BROKEN or ripping out excep= t perhaps to > > push me out to other projects with more thankful managers. > = > This was not directed to you, really. > = > This was directed at the folks who randomly "fix" that code by voodoo > bandaids instead of analyzing the root causes in the first > place. You're the least one to blame for that. Well I was about to propose a bad fix for powerclamp and I sometimes take overengineering directions. I'm still to blame for several things 8-) Sorry I got too touchy :-) Anyway, I'll give a try to your proposition. Somehow I have always felt that something was wrong with this tick_nohz_can_stop_tick() called all over the= place, that plus the IPI that's often near the scheduler IPI. I'm pretty sure this is going to help fixing some overhead issues I got rep= orted. Thanks a lot! --===============8245850828381411186==--