On 03/26/2018 11:02 AM, Cornelia Huck wrote: > On Wed, 21 Mar 2018 03:08:18 +0100 > Dong Jia Shi <bjsdj...@linux.vnet.ibm.com> wrote: > >> Hi Conny, >> >> Halil reported a host crash when using vfio-ccw. The root cause of the >> problem >> is that vfio_pin_pages fails with EINVAL for reasons unknown. He has >> experienced such failures after online-ing a dasd in the guest (the dasd has >> 3 >> partitions, hat may or may not have any significance). The problem isn't >> experienced on every attempt to online the dasd, and breaking at css_do_ssch >> seems to make things work. >> >> One thing is sure: the host kernel should not crash under the described >> circumstances. >> >> To fix the problem, the first patch of this series fixes the cleanup when >> cp_prefetch fails in the higher level. The 2nd and the 3rd patches provide >> correctness and denfensive actions for the interfaces in the lower level. > > So, is the first patch stable material?
I think it should be. > >> >> The 4th patch is trying to add tracepoints for vfio-ccw, so that we can debug >> such issue easier in future. > > Tracepoints are nice :) > >> >> For details see the commit message portions of the inividual patches. > > Still digging through the post-vacation mail pile, will do more looking > later. > > >> Thanks. >> >> Dong Jia Shi (2): >> vfio: ccw: refactor and improve pfn_array_alloc_pin() >> vfio: ccw: set ccw->cda to NULL defensively >> >> Halil Pasic (2): >> vfio: ccw: fix cleanup if cp_prefetch fails >> vfio: ccw: add traceponits for interesting error paths >> >> drivers/s390/cio/Makefile | 1 + >> drivers/s390/cio/vfio_ccw_cp.c | 121 >> ++++++++++++++++++++------------------ >> drivers/s390/cio/vfio_ccw_fsm.c | 13 ++++ >> drivers/s390/cio/vfio_ccw_trace.h | 86 +++++++++++++++++++++++++++ >> 4 files changed, 163 insertions(+), 58 deletions(-) >> create mode 100644 drivers/s390/cio/vfio_ccw_trace.h >> >
