BMC/Intel-BMC/linux.git/kernel/rcu/rcutorture.c, branch dev

rcutorture: Test SRCU cleanup code path

2015-05-27T19:59:58+00:00

The current rcutorture testing does not do any cleanup operations. This works because the srcu_struct is statically allocated, but it does represent a memory leak of the associated dynamically allocated ->per_cpu_ref per-CPU variables. However, rcutorture currently uses a statically allocated srcu_struct, which cannot legally be passed to cleanup_srcu_struct(). Therefore, this commit adds a second form of srcu (called srcud) that dynamically allocates and frees the associated per-CPU variables. This commit also adds a ->cleanup() member to rcu_torture_ops that is invoked at the end of the test, after ->cb_barriers(). This ->cleanup() pointer is NULL for all existing tests, and thus only used for scrud. Finally, the SRCU-P torture-test configuration selects scrud instead of srcu, with SRCU-N continuing to use srcu, thereby testing both static and dynamic srcu_struct structures. Reported-by: "Ahmed, Iftekhar" Signed-off-by: Paul E. McKenney Reviewed-by: Josh Triplett

rcutorture: Replace barriers with smp_store_release() and smp_load_acquire()

2015-05-27T19:59:58+00:00

The rcutorture.c file uses several explicit memory barriers that can easily be converted to smp_store_release() and smp_load_acquire(), which improves maintainability and also improves performance a bit. Signed-off-by: Paul E. McKenney Reviewed-by: Josh Triplett

rcutorture: Allow negative values of nreaders to oversubscribe

2015-05-27T19:59:57+00:00

By default, with rcutorture.nreaders equal to -1, rcutorture provisions N-1 reader kthreads, where N is the number of CPUs. This avoids rcutorture-induced stalls, but also avoids heavier levels of torture. This commit therefore allows negative values of rcutorture.nreaders to specify larger numbers of reader kthreads, so that for example rcutorture.nreaders=-2 provisions N kthreads and rcutorture.nreaders=-5 provisions N+3 kthreads. Signed-off-by: Paul E. McKenney [ paulmck: Update documentation, as suggested by Josh Triplett. ]

rcu: Convert ACCESS_ONCE() to READ_ONCE() and WRITE_ONCE()

2015-05-27T19:56:15+00:00

This commit moves from the old ACCESS_ONCE() API to the new READ_ONCE() and WRITE_ONCE() APIs. Signed-off-by: Paul E. McKenney [ paulmck: Updated to include kernel/torture.c as suggested by Jason Low. ]

rcutorture: Make consistent use of variables

2015-02-26T20:03:04+00:00

The "if" statement at the beginning of rcu_torture_writer() should use the same set of variables. In theory, this does not matter because the corresponding variables (gp_sync and gp_sync1) have the same value at this point in the code, but in practice such puzzles should be removed. This commit therefore makes the use of variables consistent. Signed-off-by: Paul E. McKenney

rcu: Add rcu_expedite_gp() and rcu_unexpedite_gp() to rcutorture

2015-02-26T20:03:00+00:00

Signed-off-by: Paul E. McKenney

rcutorture: Add more diagnostics in rcu_barrier() test failure case

2015-01-11T03:08:06+00:00

Signed-off-by: Paul E. McKenney

rcutorture: Check from beginning to end of grace period

2015-01-11T03:08:02+00:00

Currently, rcutorture's Reader Batch checks measure from the end of the previous grace period to the end of the current one. This commit tightens up these checks by measuring from the start and end of the same grace period. This involves adding rcu_batches_started() and friends corresponding to the existing rcu_batches_completed() and friends. We leave SRCU alone for the moment, as it does not yet have a way of tracking both ends of its grace periods. Signed-off-by: Paul E. McKenney

rcutorture: Drop rcu_torture_completed() and friends

2015-01-11T03:08:00+00:00

Now that the return type of rcu_batches_completed() and friends matches that of the rcu_torture_ops structure's ->completed field, the wrapper functions can be deleted. This commit carries out that deletion, while also wiring "sched"'s ->completed field to rcu_batches_completed_sched(). Signed-off-by: Paul E. McKenney

rcutorture: Use unsigned for Reader Batch computations

2015-01-11T03:07:58+00:00

The counter returned by the various ->completed functions is subject to overflow, which means that subtracting two such counters might result in overflow, which invokes undefined behavior in the C standard. This commit therefore changes these functions and variables to unsigned to avoid this undefined behavior. Signed-off-by: Paul E. McKenney