Hi, This is version 4 of patch series that provides a framework to choose the state a pseries CPU must be put to when it is offlined.
Previous versions can be found here: Version 3: http://lkml.org/lkml/2009/9/15/164 Version 2: http://lkml.org/lkml/2009/8/28/102 Version 1: http://lkml.org/lkml/2009/8/6/236 This patch series differs considerably from the previous iterations in that: - It is based on top of Nathan Fontenot's "pseries kernel handling of dynamic logical partitioning v2" patches found here: http://lkml.org/lkml/2009/9/18/173 - It does away with the sysfs tunables "available_hotplug_states" and "currrent_hotplug_state", thereby refraining from exposing any platform specific option to the userspace. - It provides a framework that discovers the support for ceding the vcpu to the hypervisor with a latency specifier value. When such a support is present, it would put the offlined CPUs into a ceded state while setting the cede latency specifier value in the VPA indicating the latency expectation of the guest OS. Hypervisor may use this for better energy savings. - On platforms which don't have support for ceding with a latency specifier hint, it puts the offlined CPUs into the rtas_stop_self() state. - Currently, Nathan's patches have provided sysfs interfaces /sys/devices/system/cpu/{probe, release} to dynamically allocate and deallocate resources respectively. The current process of allocating and deallocating CPUs would require two steps, namely: - Write to the "probe/release" file - Write to the "online" file. This patch-series combines these two operations such that when the user writes to the "probe", the CPUs would not only get allocated to the partition but also onlined. Similarly, when the user writes to "release", the CPUs would be offlined before they are deallocated. Thus, the user will now have to play with only one set of sysfs tunables for performing deallocate and the sysfs tunable "online" can be used for the purpose of deactivating the CPU while still letting the LPAR own it. - It provides serializations for handling the writes the the "online" file against the writes to "probe/release" pair so that at any given instant, the user could either be deallocating the CPUs or deactivating it, but not both. With this approach, I hope that the objection regarding the layering violation that was raised by Peter for previous approaches is addressed, while also providing clean interfaces for the userspace tools to distinguish between the deallocation and deactivation without exposing any platform specific details. The patchset has been tested on the available pseries platforms and it works as per the expectations. Comments are very much appreciated. --- Gautham R Shenoy (4): pseries: Serialize cpu hotplug operations during deactivate Vs deallocate pseries: Add code to online/offline CPUs of a DLPAR node. pSeries: Add hooks to put the CPU into an appropriate offline state pSeries: extended_cede_processor() helper function. arch/powerpc/include/asm/lppaca.h | 9 + arch/powerpc/platforms/pseries/dlpar.c | 129 ++++++++++++++++++- arch/powerpc/platforms/pseries/hotplug-cpu.c | 157 ++++++++++++++++++++++- arch/powerpc/platforms/pseries/offline_states.h | 18 +++ arch/powerpc/platforms/pseries/plpar_wrappers.h | 22 +++ arch/powerpc/platforms/pseries/smp.c | 19 +++ arch/powerpc/xmon/xmon.c | 3 drivers/base/cpu.c | 2 include/linux/cpu.h | 13 ++ 9 files changed, 356 insertions(+), 16 deletions(-) create mode 100644 arch/powerpc/platforms/pseries/offline_states.h -- Thanks and Regards gautham. _______________________________________________ Linuxppc-dev mailing list Linuxppc-dev@lists.ozlabs.org https://lists.ozlabs.org/listinfo/linuxppc-dev
