[PATCH 0/5] locking/rtmutex: priority inheritance fixes and tests

[Mathieu Desnoyers]

Hi Peter, Thomas,

Three fixes to rt_mutex priority inheritance on v7.4-rc1, and two tests
that catch them. They came out of v4 of "sched/deadline: bandwidth
accounting fixes and selftests", posted alongside, which applies on top
of this series: its patch 8 takes a deadline boost from the waiter's
rt_mutex node, which patch 3 here keeps current.

The three fixes, by patch number:

 1  the top waiter spins on a running owner past its own wakeups:
    behind the owner of a PI futex spinning in userspace, a 1ms timed
    lock returns up to 1.4s late and a signal waits as long, and with
    PREEMPT_RT two ww_mutex contexts in an ABBA spin on each other. The
    spin stops when the waiter is woken.
 2  with PREEMPT_RT, the chain walk reads the ww_mutex context that
    rt_mutex_init_waiter() leaves uninitialized, the debug poison on the
    waiters of PI futexes, and reports neither a deadlock on PI futexes
    nor a chain deeper than max_lock_depth. The context is initialized
    with the rest of the waiter.
 3  a waiter whose priority changes while another is ahead of it enters
    its owner's pi_waiters with stale sort keys once promoted, and the
    owner inherits from the wrong waiter. The keys are copied as it
    enters.

Patch 4 adds futex_lock_pi_inherit to the futex selftests. It checks the
priority an owner runs at as its waiters block, time out, get a signal,
change their priority or are promoted to the top of their lock, along
chains of owners; that FUTEX_LOCK_PI refuses a lock cycle, and a chain
deeper than max_lock_depth, with EDEADLK; and how late a timeout or a
signal ends a wait behind an owner that runs. Patch 5 adds to
test-ww_mutex an ABBA with both owners running. v7.4-rc1 fails the cases
of each fix, and the series passes them, with and without PREEMPT_RT,
along with the locking API selftests, the ww_mutex selftest,
locktorture's rtmutex_lock and the futex selftests. Each message has its
numbers.

These patches, their messages and the tests were written by an LLM
coding assistant under my direction and review. Patch 1 came out of a
stress test of the deadline boost, patch 3 out of a review of the
deadline fix by a second model, and patch 2 out of futex_lock_pi_inherit
on its first run with PREEMPT_RT. Each patch went through a local run of
Sashiko AI before posting, which changed the two tests only.

Based on v7.4-rc1, 6628e7d4c6b2.

Mathieu Desnoyers (5):
  locking/rtmutex: Stop spinning on the owner when the waiter is woken
  locking/rtmutex: Initialize the ww_mutex context of a waiter
  locking/rtmutex: Clone a waiter's sort keys as it enters pi_waiters
  selftests/futex: Test priority inheritance on PI futexes
  locking/ww_mutex/test: Time an ABBA with both owners running

 kernel/locking/rtmutex.c                      |  15 +-
 kernel/locking/rtmutex_common.h               |   1 +
 kernel/locking/test-ww_mutex.c                | 213 +++++
 .../selftests/futex/functional/.gitignore     |   1 +
 .../selftests/futex/functional/Makefile       |   3 +-
 .../futex/functional/futex_lock_pi_inherit.c  | 744 ++++++++++++++++++
 .../testing/selftests/futex/functional/run.sh |   3 +
 7 files changed, 976 insertions(+), 4 deletions(-)
 create mode 100644 tools/testing/selftests/futex/functional/futex_lock_pi_inherit.c


base-commit: 6628e7d4c6b25c8d18f1345e3a20d21829aae7e4
-- 
2.47.3