Due to an early patch (backport of upstream changeset 13028), the PIT handlers for HVM guests ended up getting registered to dom0. This causes two problems: 1) The HVM guest doesn't work because there is no PIT IO handler. 2) dom0 eventually dies because all its IO handler slots are consumed (there are only 8 to start with). This is a reduced version of upstream changeset 12243 that fixes the problem.
Created attachment 145517 [details] Register PIT handlers to the correct domain
QE ack for RHEL5 and setting the proposed blocker flags, as there's really no way to ship without fixing this.
in 2.6.18-2.el5
Is this patch supposed to have some effect on paravirtualized guests? I've seen my dom0 reboot occasionally since upgrading to 2.6.18-1.3002.el5xen, and with 2.6.18-2.el5xen I had no crashes yet, but I wonder if it's just luck or if the code could have fixed paravirt as well.
Yes it does affect PV as well, albeit partially. It doesn't prevent PV guests from functioning (although the PIT handler won't be completely functional). The part where dom0's IO handlers fill up definitely will occur for all guests regardless of whether they're PV or FV.
*** Bug 223264 has been marked as a duplicate of this bug. ***
Herbert, thanks. The 2.6.18-2.el5xen is definitely more stable for PV then 2.6.18-1.3002.el5xen.
2.6.18-7.el5 included in 20070125.0.