Hi all, So this is my 2nd attempt (more successful I hope also with CI) at catching the mmu notifier deadlock using lockdep. Ingredients: - cross-release patches resurrect (so that lockdep see through flush_workqueue). - fake lock context for mmu notifier callbacks (similar trick to fs_reclaim) - bunch of sanity checks because hey! paranoia is good Intentionally not yet submitting to core/mm since I want to figure out whether this is indeed useful for us first. Cheers, Daniel Daniel Vetter (6): locking/lockdep: restore cross-release checks kthread: finer-grained lockdep/cross-release completion lockdep: Remove GFP_NOLOCKDEP annotation mm: Check if mmut notifier callbacks are allowed to fail mm, notifier: Catch sleeping/blocking for !blockable mm, notifier: Add a lockdep map for invalidate_range_start .../admin-guide/kernel-parameters.txt | 3 + Documentation/locking/crossrelease.txt | 874 ++++++++++++++++++ include/linux/completion.h | 46 + include/linux/gfp.h | 10 +- include/linux/irqflags.h | 4 + include/linux/kthread.h | 48 +- include/linux/lockdep.h | 127 +++ include/linux/mmu_notifier.h | 7 + include/linux/sched.h | 11 + kernel/kthread.c | 68 +- kernel/locking/lockdep.c | 652 ++++++++++++- kernel/sched/completion.c | 5 + lib/Kconfig.debug | 33 + mm/mmu_notifier.c | 17 +- mm/page_alloc.c | 3 - 15 files changed, 1832 insertions(+), 76 deletions(-) create mode 100644 Documentation/locking/crossrelease.txt -- 2.19.1 _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
