Public bug reported: The current implementation of Machine Check handler and HMI handler in Linux, goes down kernel panic path for unrecoverable errors. On FSP based system FSP also gets notified about these errors which then forwards it to PRD (that runs on FSP) for error analysis and gard record creation.
On OpenPower (BMC based system e.g. Habanero from TYAN) where PRD runs in Linux host, it never gets a chance to do error analysis at the time of Linux crash and no gard record is created for such errors. Since the faulty component never gets de-configured, the system is vulnerable to get hit by same HW error again. To fix this issue, a new OPAL call 'opal_cec_reboot2()' has been introduced to trigger a checkstop on BMC based system to inform BMC/OCC about this error, so that BMC can collect relevant data for error analysis and decide what component to de-configure before rebooting. Linux kernel should invoke this opal call for unrecoverable MCE and HMI instead before calling kernel panic so that OCC is informed about the error. The kernel changes has already been posted to upstream and are listed below: https://lists.ozlabs.org/pipermail/linuxppc-dev/2015-May/128341.html https://lists.ozlabs.org/pipermail/linuxppc-dev/2015-May/128342.html https://lists.ozlabs.org/pipermail/linuxppc-dev/2015-August/132045.html https://lists.ozlabs.org/pipermail/linuxppc-dev/2015-August/132114.html Above patches needs to be included in ubuntu 14.04.3+ We will update this bug with commit ids, once the above patches are accepted upstream. Contact Information = [email protected] ---uname output--- Linux rcx2d403 3.19.0-26-generic #27 SMP Tue Aug 4 01:38:15 CDT 2015 ppc64le ppc64le ppc64le GNU/Linux ---Additional Hardware Info--- Habanero pass2 system Machine Type = OpenPower, Habanero ---System Hang--- If system is hung, it can be recovered by sending ipmi power off/on command. $ ipmitool -H <BMC> -I lanplus -U <user> -P <passwd> power off $ ipmitool -H <BMC> -I lanplus -U <user> -P <passwd> power on ** Affects: ubuntu Importance: Undecided Status: New ** Tags: architecture-ppc64le bugnameltc-128601 severity-high targetmilestone-inin--- ** Tags added: architecture-ppc64le bugnameltc-128601 severity-high targetmilestone-inin--- -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/1482343 Title: Trigger a checkstop on unrecoverable MCE/HMI errors to inform BMC/OCC about the error. To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+bug/1482343/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
