** Description changed: + [Impact] + When panic_timeout is set, the user would expect the system to reboot. However, it is shutdown, possibly requiring user intervention to power on the system again. This may impact availability. This happens under certain versions of qemu (2.10+) emulating pseries. + + [Test Case] + Built kernels (trusty, xenial, zesty, artful) have been tested and they reboot after the fix, when running under an affected qemu. + + [Regression Potential] + fadump might require that RTAS ibm,os-term be called. The fix should still allow it to be called when fadump is registered. + + + ----------- + When panic_timeout is set, the user would expect that the system would reboot after some timeout after a panic. However, on powerpc architecture, a panic notifier may not ever return and even shutdown the system. The main example that impacts platforms we support is pseries calling RTAS ibm,os-term. That panic notifier and calling RTAS ibm,os-term is useful for fadump, so this should not be impacted. The Linux code has changed back and forth on considering the panic_timeout setting and checking whether ibm,extended-os-term was available. Unfortunately, recent changes on qemu led to the guest being shut down when calling ibm,os-term even when ibm,extended-os-term was available. Luckily, upstream Linux already has the change that does not call ibm ,os-term on panic, but only during fadump. So, we can use that commit for fixing this. Changing qemu itself is harder as: 1) qemu community already decided some qemu settings should override PAPR; 2) updating deployed hypervisor is harder than updating our guest kernels. Cascardo.
-- You received this bug notification because you are a member of Kernel Packages, which is subscribed to linux in Ubuntu. https://bugs.launchpad.net/bugs/1736954 Title: ppc64el: Do not call ibm,os-term on panic Status in linux package in Ubuntu: In Progress Bug description: [Impact] When panic_timeout is set, the user would expect the system to reboot. However, it is shutdown, possibly requiring user intervention to power on the system again. This may impact availability. This happens under certain versions of qemu (2.10+) emulating pseries. [Test Case] Built kernels (trusty, xenial, zesty, artful) have been tested and they reboot after the fix, when running under an affected qemu. [Regression Potential] fadump might require that RTAS ibm,os-term be called. The fix should still allow it to be called when fadump is registered. ----------- When panic_timeout is set, the user would expect that the system would reboot after some timeout after a panic. However, on powerpc architecture, a panic notifier may not ever return and even shutdown the system. The main example that impacts platforms we support is pseries calling RTAS ibm,os-term. That panic notifier and calling RTAS ibm,os-term is useful for fadump, so this should not be impacted. The Linux code has changed back and forth on considering the panic_timeout setting and checking whether ibm,extended-os-term was available. Unfortunately, recent changes on qemu led to the guest being shut down when calling ibm,os-term even when ibm,extended-os- term was available. Luckily, upstream Linux already has the change that does not call ibm ,os-term on panic, but only during fadump. So, we can use that commit for fixing this. Changing qemu itself is harder as: 1) qemu community already decided some qemu settings should override PAPR; 2) updating deployed hypervisor is harder than updating our guest kernels. Cascardo. To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1736954/+subscriptions -- Mailing list: https://launchpad.net/~kernel-packages Post to : kernel-packages@lists.launchpad.net Unsubscribe : https://launchpad.net/~kernel-packages More help : https://help.launchpad.net/ListHelp
