Being silly I figured I would poke my nose in and see how much work it
is to never wake up kthreads until we are ready to use them.

This is my first draft at that and something that can hopefully shape
the conversation on how we want to fix things a little bit.

The big thing that needs to happen that I haven't implemented is that
kthread_run and kthread_run_on_cpu need to be uninlined and moved into
kthread.c.  This will allow them to call wake_up_new_task even from
modular code.

The handful of drivers that are using kthread_create_on_node by
extension need to be modified to use kthread_run or kthread_run_on_cpu.

Eric W. Biederman (3):
      kthread: Remove the flags argument from kernel_thread
      kthread: Replace kernel_thread with new_kthread
      kthread: Stop abusing TASK_UNINTERRUPTIBLE (INCOMPLETE)

 arch/arm/common/bL_switcher.c             |   2 +-
 arch/x86/kernel/cpu/resctrl/pseudo_lock.c |   4 +-
 drivers/block/mtip32xx/mtip32xx.c         |   2 +-
 drivers/firmware/psci/psci_checker.c      |   2 +-
 drivers/firmware/stratix10-svc.c          |   4 +-
 drivers/scsi/bnx2fc/bnx2fc_fcoe.c         |   2 +-
 drivers/scsi/bnx2i/bnx2i_init.c           |   2 +-
 drivers/scsi/qedi/qedi_main.c             |   2 +-
 include/linux/kthread.h                   |   4 +-
 include/linux/sched/task.h                |   2 +-
 init/main.c                               |   6 +-
 kernel/bpf/cpumap.c                       |   2 +-
 kernel/dma/map_benchmark.c                |   2 +-
 kernel/fork.c                             |   5 +-
 kernel/kthread.c                          | 120 +++++++++++++++---------------
 kernel/smpboot.c                          |   1 +
 kernel/workqueue.c                        |   2 +-
 net/core/pktgen.c                         |   2 +-
 net/sunrpc/svc.c                          |   2 +-
 19 files changed, 82 insertions(+), 86 deletions(-)

Eric