]> www.infradead.org Git - users/hch/misc.git/commitdiff
rcutorture: Suppress "Writer stall state" reports during boot
authorPaul E. McKenney <paulmck@kernel.org>
Wed, 6 Aug 2025 23:07:33 +0000 (16:07 -0700)
committerPaul E. McKenney <paulmck@kernel.org>
Thu, 14 Aug 2025 22:26:30 +0000 (15:26 -0700)
When rcutorture is running on only the one boot-time CPU while that CPU
is busy invoking initcall() functions, the added load is quite likely to
unduly delay the RCU grace-period kthread, rcutorture readers, and much
else besides.  This can result in rcu_torture_stats_print() reporting
rcutorture writer stalls, which are not really a bug in that environment.
After all, one CPU can only do so much.

This commit therefore suppresses rcutorture writer stalls while the
kernel is booting, that is, while rcu_inkernel_boot_has_ended() continues
returning false.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
kernel/rcu/rcutorture.c

index 7a893d51d02b6afc2142b04309edec2bc05d5073..49e048da4f6810fac324dc06ad00cea027cbc207 100644 (file)
@@ -2756,7 +2756,8 @@ rcu_torture_stats_print(void)
                cur_ops->stats();
        if (rtcv_snap == rcu_torture_current_version &&
            rcu_access_pointer(rcu_torture_current) &&
-           !rcu_stall_is_suppressed()) {
+           !rcu_stall_is_suppressed() &&
+           rcu_inkernel_boot_has_ended()) {
                int __maybe_unused flags = 0;
                unsigned long __maybe_unused gp_seq = 0;