Created attachment 354857 [details] kvm-svm-force-new-asid-on-vcpu-migration.patch KVM: SVM: force new asid on vcpu migration If a migrated vcpu matches the asid_generation value of the target pcpu, there will be no TLB flush via TLB_CONTROL_FLUSH_ALL_ASID. The check for vcpu.cpu in pre_svm_run is meaningless since svm_vcpu_load already updated it on schedule in. Such vcpu will VMRUN with stale TLB entries. Based on original patch from Joerg Roedel (http://patchwork.kernel.org/patch Signed-off-by: Marcelo Tosatti <mtosatti> Acked-by: Joerg Roedel <joerg.roedel>
Marcelo, how serious is the bug the patch solves? Do you think this should be considered a blocker?
Its possible, in rare conditions, that guest 2 runs with TLB entries from guest 1. So when that happens, both guests can have their memory corrupted. There is no report of the problem, and there is no testcase, but its a theoretical possibility.
Given that there are no reports of the such issues, i would say its not a blocker. Given the description of the bug, i would say its a blocker. So nice to have it included if possible.
*** Bug 511151 has been marked as a duplicate of this bug. ***
Bug 511151 has been marked a duplicate of this bug, but some things are not clear: 1. In bug 511151 we have a reproducer for this problem, contrary to comment 6 in this bug, so it's clearly a blocker. 2. Bug 511151 is high priority, high severity, but this bug is marked low/medium, so this bug probably needs to be modified appropriately. 3. This is marked ON_QA but I can't see if this is included in RHEL 5.4 already. Comments 7-15 are hidden.
(In reply to comment #17) > 3. This is marked ON_QA but I can't see if this is included in RHEL 5.4 > already. Comments 7-15 are hidden. The patch mentioned on comment #0 was included on kvm-83-96.el5, and this bug is marked as blocker.
Tested : 1 run 48 VMs ( 1 vcpu,4 GB RAM ) on a 16 CPU AMD host for 12 hours 2 run 24 VMs ( 1 vcpu,1 GB RAM ) on a 8 CPU Intel host for 72 hours 3 run 12 VMs ( 1 vcpu,1 GB RAM ) on a 4 CPU AMD host for 72 hours No Error was found so far. and no evidence show this patch introduce regression in the acceptance testing . setting to VERIFIED
An advisory has been issued which should help the problem described in this bug report. This report is therefore being closed with a resolution of ERRATA. For more information on therefore solution and/or where to find the updated files, please follow the link below. You may reopen this bug report if the solution does not work for you. http://rhn.redhat.com/errata/RHEA-2009-1272.html