From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Dan Magenheimer" Subject: RE: [PATCH] Add a timer mode that disables pending missed ticks Date: Wed, 20 Feb 2008 16:40:47 -0700 Message-ID: <20080220164047500.00000002392@djm-pc> References: <20080219163806078.00000002392@djm-pc> Reply-To: "dan.magenheimer@oracle.com" Mime-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: quoted-printable Return-path: In-Reply-To: <20080219163806078.00000002392@djm-pc> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: "dan.magenheimer@oracle.com" , Dave Winchell Cc: "xen-devel@lists.xensource.com" , Deepak Patel List-Id: xen-devel@lists.xenproject.org > > >percentage. However, our 32-bit clock skew seems to > > >show a measureable problem now. > > > > > For the 32 bit guest, which timesource did it pick? > = > The dmesg output is hard to interpret on a 32-bit guest, > but based on what we've seen, I think it was selecting > hpet as timesource (because we specified clocksource=3Dpit > which would have been ignored on RHEL4-32). We are > running another test with "clock=3Dpit" to see if the > skew goes away. Yep, that was the problem. No clock skew with clock=3Dpit on RHEL4-32 anymore. > > >For Xen RHEL5 HVM guests: > > >- I *think* clock=3Dpit is sufficient for RHEL5-32 > > > > > But still poor accuracy, right? > = > Unproven yet but I hope not. The nohpet and nopmtimer > parameters are ignored on RHEL5-32 so the clock=3Dpit > (or clocksource=3Dpit) is the only way to choose the > clock source, and thus the only way to get good > accuracy on RHEL5-32. We had an interesting firedrill with RHEL5-32. It seems that the RHEL-specific "divider=3D10" doesn't get along very well with "clock=3Dpit" and causes the system clock to go whacko, gaining literally HOURS every second. Until Deepak discovered this, it was impossible to boot. So we haven't tested this hypothesis yet, but the dmesg output looks promising. Dan