rcu: Restrict default jiffies_till_first_fqs for strict RCU GPs
authorPaul E. McKenney <paulmck@kernel.org>
Thu, 6 Aug 2020 00:25:23 +0000 (17:25 -0700)
committerPaul E. McKenney <paulmck@kernel.org>
Tue, 25 Aug 2020 01:40:24 +0000 (18:40 -0700)
commitaecd34b9765de3b58c98a1d75b982fc64becd1e9
treee24b28222c483c0fdcb19a4ef2d6e4cc291b5d9a
parentdc1269186bed3afc5a2018527516be84fe55d3e0
rcu: Restrict default jiffies_till_first_fqs for strict RCU GPs

If there are idle CPUs, RCU's grace-period kthread will wait several
jiffies before even thinking about polling them.  This promotes
efficiency, which is normally a good thing, but when the kernel
has been built with CONFIG_RCU_STRICT_GRACE_PERIOD=y, we care more
about short grace periods.  This commit therefore restricts the
default jiffies_till_first_fqs value to zero in kernels built with
CONFIG_RCU_STRICT_GRACE_PERIOD=y, which causes RCU's grace-period kthread
to poll for idle CPUs immediately after starting a grace period.

Reported-by Jann Horn <jannh@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
kernel/rcu/tree.c