[PATCH v3 0/6] rcutorture: Rewrite RCU priority-boost test around a CPU-hungry reader

From: Joel Fernandes

Date: Sat Nov 21 2026 - 10:36:01 EST


Hi,

This is v3 of the series that rewrites rcutorture's RCU priority-boost test
around a CPU-hungry reader, so that it neither depends on disabling RT
throttling nor infers boost failure from a grace period merely taking too
long. See the v1 and v2 cover letters for the full description:

v1: https://lore.kernel.org/all/20261111213826.2035045-1-joelagnelf@xxxxxxxxxx/
v2: https://lore.kernel.org/all/20261116194438.306618-1-jfern@xxxxxxxxxx/

Changes since v2 (from Sebastian's review of 5/7 and 6/7, thanks!):

- 5/6: boost_disable is now an rcutorture parameter,
rcutorture.boost_disable, instead of rcutree.boost_disable. It makes
rcu_spawn_one_boost_kthread() skip creating the boost kthreads.
- 6/6: BUSTED-BOOST boots with rcutorture.boost_disable=1.
- Added Sebastian's Reviewed-by (1/6, 3/6, 4/6) and Paul's Tested-by.
(note: Paul gave his Tested-by on v2, before the changes to 5/6 and 6/6 above,
so Paul, please let me know if the T-b still does not stand for those 2).
- Rebased onto the current rcu/next.

Paul also noticed that TREE03 with rcutorture.stall_cpu=25 reports a boost
failure. He considers that expected, so nothing here changes it.

Joel Fernandes (5):
rcutorture: Rewrite RCU priority-boost test around a CPU-hungry reader
rcutorture: Remove unused check_boost_failed hook
doc: Document rcutorture.test_boost_reader_ms
rcutorture: Add boost_disable to disable RCU priority boosting
rcutorture: Make BUSTED-BOOST disable RCU priority boosting

Sebastian Andrzej Siewior (1):
rcutorture: Don't try to disable to RT throttling.

.../admin-guide/kernel-parameters.txt | 16 ++
kernel/rcu/rcu.h | 5 +-
kernel/rcu/rcutorture.c | 202 +++++++++---------
kernel/rcu/tree_plugin.h | 4 +
kernel/rcu/tree_stall.h | 57 -----
.../rcutorture/configs/rcu/BUSTED-BOOST.boot | 1 +
6 files changed, 128 insertions(+), 157 deletions(-)

Range-diff against v2:
1: 307c7ad81e84 ! 1: f73cdf3db202 rcutorture: Rewrite RCU priority-boost test around a CPU-hungry reader
@@ Commit message
[ jfern: Apply feedback from Sebastian to use div64_ul() ]

Signed-off-by: Joel Fernandes <jfern@xxxxxxxxxx>
+ Reviewed-by: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx>
+ Tested-by: Paul E. McKenney <paulmck@xxxxxxxxxx>

## kernel/rcu/rcutorture.c ##
@@ kernel/rcu/rcutorture.c: torture_param(int, test_boost, 1, "Test RCU prio boost: 0=no, 1=maybe, 2=yes.");
2: 8037cd6fe691 ! 2: dd690a7332dc rcutorture: Don't try to disable to RT throttling.
@@ Commit message
[ jfern: Rebase onto the rewritten priority-boost test ]
[ jfern: Remove the now-unneeded sched/sysctl.h include ]
Signed-off-by: Joel Fernandes <jfern@xxxxxxxxxx>
+ Tested-by: Paul E. McKenney <paulmck@xxxxxxxxxx>

## kernel/rcu/rcu.h ##
@@
3: 121b55687de4 ! 3: c9bfbebcbc04 rcutorture: Remove unused check_boost_failed hook
@@ Commit message
[ jfern: Apply feedback from Sebastian to remove rcu_check_boost_fail() ]

Signed-off-by: Joel Fernandes <jfern@xxxxxxxxxx>
+ Reviewed-by: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx>
+ Tested-by: Paul E. McKenney <paulmck@xxxxxxxxxx>

## kernel/rcu/rcu.h ##
@@ kernel/rcu/rcu.h: static inline bool rcu_watching_zero_in_eqs(int cpu, int *vp) { return false; }
4: 66bf2bb52aa0 ! 4: 89de9f2ec120 doc: Document rcutorture.test_boost_reader_ms
@@ Commit message
kernel-parameters.txt.

Signed-off-by: Joel Fernandes <jfern@xxxxxxxxxx>
+ Reviewed-by: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx>
+ Tested-by: Paul E. McKenney <paulmck@xxxxxxxxxx>

## Documentation/admin-guide/kernel-parameters.txt ##
@@ Documentation/admin-guide/kernel-parameters.txt: Kernel parameters
5: 88332b3b02c5 < -: ------------ rcu: Add rcutree.boost_disable to disable priority boosting at runtime
6: 02867b20df51 ! 5: 67d1712c26ec doc: Document rcutree.boost_disable
@@ Metadata
Author: Joel Fernandes <jfern@xxxxxxxxxx>

## Commit message ##
- doc: Document rcutree.boost_disable
+ rcutorture: Add boost_disable to disable RCU priority boosting

- Add the new rcutree.boost_disable parameter to kernel-parameters.txt.
+ CONFIG_RCU_BOOST is now enabled automatically for preemptible RCU and
+ cannot be turned off, so rcutorture's BUSTED-BOOST scenario, which is
+ meant to show that rcutorture detects broken priority boosting, can no
+ longer break it. Add an rcutorture.boost_disable parameter that keeps
+ rcu_spawn_one_boost_kthread() from creating the boost kthreads, which
+ gives that scenario a way to break boosting.
+
+ The parameter only has an effect when rcutorture is built in, because
+ the kthreads are created during boot. Document it in
+ kernel-parameters.txt.
+
+ [ jfern: Apply feedback from Sebastian to not create the boost kthread ]
+ [ jfern: Apply feedback from Sebastian to limit this to rcutorture ]
+ [ jfern: Apply feedback from Sebastian to fold the docs into this patch ]

Signed-off-by: Joel Fernandes <jfern@xxxxxxxxxx>
+ Tested-by: Paul E. McKenney <paulmck@xxxxxxxxxx>

## Documentation/admin-guide/kernel-parameters.txt ##
@@ Documentation/admin-guide/kernel-parameters.txt: Kernel parameters
- Set maximum number of finished RCU callbacks to
- process in one batch.
-
-+ rcutree.boost_disable= [KNL]
-+ Disable RCU priority boosting at runtime, even in
-+ kernels built with CONFIG_RCU_BOOST=y. This is
-+ for testing the boost tests, for example by the
-+ rcutorture BUSTED-BOOST scenario, and should not
-+ be used otherwise.
+ periods, but in jiffies. The default of zero
+ says no holdoff.
+
++ rcutorture.boost_disable= [KNL]
++ Do not create the RCU priority-boost kthreads, so
++ that RCU priority boosting never happens. This
++ is for testing that rcutorture detects broken
++ boosting, as in the BUSTED-BOOST scenario, and
++ should not be used otherwise. It has an effect
++ only if rcutorture is built in, because the
++ kthreads are created during boot.
++
+ rcutorture.fqs_duration= [KNL]
+ Set duration of force_quiescent_state bursts
+ in microseconds.
+
+ ## kernel/rcu/rcu.h ##
+@@ kernel/rcu/rcu.h: bool rcu_is_task_rcu_boosted(void);
+ static inline bool rcu_is_task_rcu_boosted(void) { return false; }
+ #endif // #else // #if defined(CONFIG_RCU_TORTURE_TEST) && defined(CONFIG_RCU_BOOST)
+
++extern bool rcu_torture_boost_disable;
+
+ #endif /* __LINUX_RCU_H */
+
+ ## kernel/rcu/rcutorture.c ##
+@@ kernel/rcu/rcutorture.c: static char *torture_type = "rcu";
+ module_param(torture_type, charp, 0444);
+ MODULE_PARM_DESC(torture_type, "Type of RCU to torture (rcu, srcu, ...)");
+
++// Global because tree_plugin.h reads it during boot.
++bool rcu_torture_boost_disable;
++module_param_named(boost_disable, rcu_torture_boost_disable, bool, 0444);
++MODULE_PARM_DESC(boost_disable, "Do not create RCU boost kthreads, to test boost-failure detection.");
++
+ static int nrealnocbers;
+ static int nrealreaders;
+ static int nrealfakewriters;
+
+ ## kernel/rcu/tree_plugin.h ##
+@@ kernel/rcu/tree_plugin.h: static void rcu_spawn_one_boost_kthread(struct rcu_node *rnp)
+ if (rnp->boost_kthread_task)
+ return;
+
++ // With no boost kthread, rcu_initiate_boost() never boosts.
++ if (IS_BUILTIN(CONFIG_RCU_TORTURE_TEST) && rcu_torture_boost_disable)
++ return;
+
- rcutree.csd_lock_suppress_rcu_stall= [KNL]
- Do only a one-line RCU CPU stall warning when
- there is an ongoing too-long CSD-lock wait.
+ t = kthread_create(rcu_boost_kthread, (void *)rnp,
+ "rcub/%d", rnp_index);
+ if (WARN_ON_ONCE(IS_ERR(t)))
7: b172a791f0a1 ! 6: 99a7a7cfe376 rcutorture: Make BUSTED-BOOST disable RCU priority boosting
@@ Commit message

BUSTED-BOOST is supposed to fail because priority boosting is broken,
but boosting is now always built in for preemptible RCU, so the
- scenario passed. Boot it with rcutree.boost_disable=1 so that it again
+ scenario passed. Boot it with rcutorture.boost_disable=1 so that it again
exercises rcutorture's ability to detect missing priority boosting.

Testing: kvm.sh --configs BUSTED-BOOST --duration 10 now ends FAILURE,
- with 81 of 84 boost intervals reporting:
+ with 82 of 84 boost intervals reporting:

rcu_torture_boost boosting failed
Boost inversion: reader got only 203 of 2000 ms CPU in 4003 jiffies

Without the parameter the same scenario passed with no boost failures.

+ [ jfern: Follow the rename to rcutorture.boost_disable ]
+
Signed-off-by: Joel Fernandes <jfern@xxxxxxxxxx>
+ Tested-by: Paul E. McKenney <paulmck@xxxxxxxxxx>

## tools/testing/selftests/rcutorture/configs/rcu/BUSTED-BOOST.boot ##
@@ tools/testing/selftests/rcutorture/configs/rcu/BUSTED-BOOST.boot: rcutree.gp_cleanup_delay=3
rcutree.kthread_prio=2
threadirqs
rcutree.use_softirq=0
-+rcutree.boost_disable=1
++rcutorture.boost_disable=1

base-commit: 83984be2e4ae1aee94a0f6a65284cde82c2dc0eb
--
2.54.0