Hi, I split this up into 3 pieces instead of the messy single patch, hope this helps with review.
Patch 1 adds task_sigpending(), which tests TIF_SIGPENDING. Core use cases that need to check for an actual signal pending are switched to using task_sigpending() instead of signal_pending(). This should fix Oleg's concern on signal_pending() == true, but no signals pending, for actual signal delivery. Patch 2 adds x86 and generic entry code support for TIF_TASKWORK. Patch 3 adds task_work support for TIF_TASKWORK, if the arch supports it. There's no need for any io_uring specific changes, so I've dropped those. If TIF_TASKWORK is used, then JOBCTL_TASK_WORK will never be true and hence we won't enter that case. If TIF_TASKWORK isn't available, then we still need that code. I've run this through my usual liburing test, and it passes. I also ran it through all the ltp signal testing, and no changes from mainline in terms of all tests passing. arch/x86/include/asm/thread_info.h | 2 ++ arch/x86/kernel/signal.c | 32 +++++++++++--------- include/linux/entry-common.h | 20 +++++++++++-- include/linux/sched/signal.h | 32 ++++++++++++++++---- kernel/entry/common.c | 14 +++++++-- kernel/events/uprobes.c | 2 +- kernel/ptrace.c | 2 +- kernel/signal.c | 12 ++++---- kernel/task_work.c | 48 ++++++++++++++++++++++-------- 9 files changed, 118 insertions(+), 46 deletions(-) Changes can also be viewed/pulled from this branch: git://git.kernel.dk/linux-block tif-task_work https://git.kernel.dk/cgit/linux-block/log/?h=tif-task_work -- Jens Axboe
