rcutorture: Make in-kernel-loop testing more brutal
authorPaul E. McKenney <paulmck@kernel.org>
Thu, 15 Aug 2019 18:43:53 +0000 (11:43 -0700)
committerPaul E. McKenney <paulmck@kernel.org>
Sat, 5 Oct 2019 18:50:18 +0000 (11:50 -0700)
commitfbbd5e358cecb5fa490550ace66463517a7577e8
treec251b136d8d2b2f28b351d6bbbd3d31c97f2ce06
parent8b5ddf8b99dc42241d1d413c6685bce18275c40e
rcutorture: Make in-kernel-loop testing more brutal

The rcu_torture_fwd_prog_nr() tests the ability of RCU to tolerate
in-kernel busy loops.  It invokes rcu_torture_fwd_prog_cond_resched()
within its delay loop, which, in PREEMPT && NO_HZ_FULL kernels results
in the occasional direct call to schedule().  Now, this direct call to
schedule() is appropriate for call_rcu() flood testing, in which either
the kernel should restrain itself or userspace transitions will supply
the needed restraint.  But in pure in-kernel loops, the occasional
cond_resched() should do the job.

This commit therefore makes rcu_torture_fwd_prog_nr() use cond_resched()
instead of rcu_torture_fwd_prog_cond_resched() in order to increase the
brutality of this aspect of rcutorture testing.

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