locktorture: Add a lock-torture kernel module
authorPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Tue, 4 Feb 2014 23:51:41 +0000 (15:51 -0800)
committerPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Sun, 23 Feb 2014 17:04:29 +0000 (09:04 -0800)
This commit adds the locking counterpart to rcutorture.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
[ paulmck: Make n_lock_torture_errors and torture_spinlock static
  as suggested by Fengguang Wu. ]
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
kernel/locking/Makefile
kernel/locking/locktorture.c [new file with mode: 0644]
lib/Kconfig.debug

index baab8e5e7f66acd327f71afd40fbd7530803a449..a28ea6d9e6e8845819d782b66e01aa3c5f6859a8 100644 (file)
@@ -23,3 +23,4 @@ obj-$(CONFIG_DEBUG_SPINLOCK) += spinlock_debug.o
 obj-$(CONFIG_RWSEM_GENERIC_SPINLOCK) += rwsem-spinlock.o
 obj-$(CONFIG_RWSEM_XCHGADD_ALGORITHM) += rwsem-xadd.o
 obj-$(CONFIG_PERCPU_RWSEM) += percpu-rwsem.o
+obj-$(CONFIG_LOCK_TORTURE_TEST) += locktorture.o
diff --git a/kernel/locking/locktorture.c b/kernel/locking/locktorture.c
new file mode 100644 (file)
index 0000000..d69d20d
--- /dev/null
@@ -0,0 +1,421 @@
+/*
+ * Module-based torture test facility for locking
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, you can access it online at
+ * http://www.gnu.org/licenses/gpl-2.0.html.
+ *
+ * Copyright (C) IBM Corporation, 2014
+ *
+ * Author: Paul E. McKenney <paulmck@us.ibm.com>
+ *     Based on kernel/rcu/torture.c.
+ */
+#include <linux/types.h>
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/kthread.h>
+#include <linux/err.h>
+#include <linux/spinlock.h>
+#include <linux/smp.h>
+#include <linux/interrupt.h>
+#include <linux/sched.h>
+#include <linux/atomic.h>
+#include <linux/bitops.h>
+#include <linux/completion.h>
+#include <linux/moduleparam.h>
+#include <linux/percpu.h>
+#include <linux/notifier.h>
+#include <linux/reboot.h>
+#include <linux/freezer.h>
+#include <linux/cpu.h>
+#include <linux/delay.h>
+#include <linux/stat.h>
+#include <linux/slab.h>
+#include <linux/trace_clock.h>
+#include <asm/byteorder.h>
+#include <linux/torture.h>
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Paul E. McKenney <paulmck@us.ibm.com>");
+
+torture_param(int, nwriters_stress, -1,
+            "Number of write-locking stress-test threads");
+torture_param(int, onoff_holdoff, 0, "Time after boot before CPU hotplugs (s)");
+torture_param(int, onoff_interval, 0,
+            "Time between CPU hotplugs (s), 0=disable");
+torture_param(int, shuffle_interval, 3,
+            "Number of jiffies between shuffles, 0=disable");
+torture_param(int, shutdown_secs, 0, "Shutdown time (j), <= zero to disable.");
+torture_param(int, stat_interval, 60,
+            "Number of seconds between stats printk()s");
+torture_param(int, stutter, 5, "Number of jiffies to run/halt test, 0=disable");
+torture_param(bool, verbose, true,
+            "Enable verbose debugging printk()s");
+
+static char *torture_type = "spin_lock";
+module_param(torture_type, charp, 0444);
+MODULE_PARM_DESC(torture_type,
+                "Type of lock to torture (spin_lock, spin_lock_irq, ...)");
+
+static atomic_t n_lock_torture_errors;
+
+static struct task_struct *stats_task;
+static struct task_struct **writer_tasks;
+
+static int nrealwriters_stress;
+static bool lock_is_write_held;
+
+struct lock_writer_stress_stats {
+       long n_write_lock_fail;
+       long n_write_lock_acquired;
+};
+static struct lock_writer_stress_stats *lwsa;
+
+#if defined(MODULE) || defined(CONFIG_LOCK_TORTURE_TEST_RUNNABLE)
+#define LOCKTORTURE_RUNNABLE_INIT 1
+#else
+#define LOCKTORTURE_RUNNABLE_INIT 0
+#endif
+int locktorture_runnable = LOCKTORTURE_RUNNABLE_INIT;
+module_param(locktorture_runnable, int, 0444);
+MODULE_PARM_DESC(locktorture_runnable, "Start locktorture at boot");
+
+/* Forward reference. */
+static void lock_torture_cleanup(void);
+
+/*
+ * Operations vector for selecting different types of tests.
+ */
+struct lock_torture_ops {
+       void (*init)(void);
+       int (*writelock)(void);
+       void (*write_delay)(struct torture_random_state *trsp);
+       void (*writeunlock)(void);
+       unsigned long flags;
+       const char *name;
+};
+
+static struct lock_torture_ops *cur_ops;
+
+/*
+ * Definitions for lock torture testing.
+ */
+
+static DEFINE_SPINLOCK(torture_spinlock);
+
+static int torture_spin_lock_write_lock(void) __acquires(torture_spinlock)
+{
+       spin_lock(&torture_spinlock);
+       return 0;
+}
+
+static void torture_spin_lock_write_delay(struct torture_random_state *trsp)
+{
+       const unsigned long shortdelay_us = 2;
+       const unsigned long longdelay_us = 100;
+
+       /* We want a short delay mostly to emulate likely code, and
+        * we want a long delay occasionally to force massive contention.
+        */
+       if (!(torture_random(trsp) %
+             (nrealwriters_stress * 2000 * longdelay_us)))
+               mdelay(longdelay_us);
+       if (!(torture_random(trsp) %
+             (nrealwriters_stress * 2 * shortdelay_us)))
+               udelay(shortdelay_us);
+#ifdef CONFIG_PREEMPT
+       if (!(torture_random(trsp) % (nrealwriters_stress * 20000)))
+               preempt_schedule();  /* Allow test to be preempted. */
+#endif
+}
+
+static void torture_spin_lock_write_unlock(void) __releases(torture_spinlock)
+{
+       spin_unlock(&torture_spinlock);
+}
+
+static struct lock_torture_ops spin_lock_ops = {
+       .writelock      = torture_spin_lock_write_lock,
+       .write_delay    = torture_spin_lock_write_delay,
+       .writeunlock    = torture_spin_lock_write_unlock,
+       .name           = "spin_lock"
+};
+
+static int torture_spin_lock_write_lock_irq(void)
+__acquires(torture_spinlock_irq)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&torture_spinlock, flags);
+       cur_ops->flags = flags;
+       return 0;
+}
+
+static void torture_lock_spin_write_unlock_irq(void)
+__releases(torture_spinlock)
+{
+       spin_unlock_irqrestore(&torture_spinlock, cur_ops->flags);
+}
+
+static struct lock_torture_ops spin_lock_irq_ops = {
+       .writelock      = torture_spin_lock_write_lock_irq,
+       .write_delay    = torture_spin_lock_write_delay,
+       .writeunlock    = torture_lock_spin_write_unlock_irq,
+       .name           = "spin_lock_irq"
+};
+
+/*
+ * Lock torture writer kthread.  Repeatedly acquires and releases
+ * the lock, checking for duplicate acquisitions.
+ */
+static int lock_torture_writer(void *arg)
+{
+       struct lock_writer_stress_stats *lwsp = arg;
+       static DEFINE_TORTURE_RANDOM(rand);
+
+       VERBOSE_TOROUT_STRING("lock_torture_writer task started");
+       set_user_nice(current, 19);
+
+       do {
+               schedule_timeout_uninterruptible(1);
+               cur_ops->writelock();
+               if (WARN_ON_ONCE(lock_is_write_held))
+                       lwsp->n_write_lock_fail++;
+               lock_is_write_held = 1;
+               lwsp->n_write_lock_acquired++;
+               cur_ops->write_delay(&rand);
+               lock_is_write_held = 0;
+               cur_ops->writeunlock();
+               stutter_wait("lock_torture_writer");
+       } while (!torture_must_stop());
+       torture_kthread_stopping("lock_torture_writer");
+       return 0;
+}
+
+/*
+ * Create an lock-torture-statistics message in the specified buffer.
+ */
+static void lock_torture_printk(char *page)
+{
+       bool fail = 0;
+       int i;
+       long max = 0;
+       long min = lwsa[0].n_write_lock_acquired;
+       long long sum = 0;
+
+       for (i = 0; i < nrealwriters_stress; i++) {
+               if (lwsa[i].n_write_lock_fail)
+                       fail = true;
+               sum += lwsa[i].n_write_lock_acquired;
+               if (max < lwsa[i].n_write_lock_fail)
+                       max = lwsa[i].n_write_lock_fail;
+               if (min > lwsa[i].n_write_lock_fail)
+                       min = lwsa[i].n_write_lock_fail;
+       }
+       page += sprintf(page, "%s%s ", torture_type, TORTURE_FLAG);
+       page += sprintf(page,
+                       "Writes:  Total: %lld  Max/Min: %ld/%ld %s  Fail: %d %s\n",
+                       sum, max, min, max / 2 > min ? "???" : "",
+                       fail, fail ? "!!!" : "");
+       if (fail)
+               atomic_inc(&n_lock_torture_errors);
+}
+
+/*
+ * Print torture statistics.  Caller must ensure that there is only one
+ * call to this function at a given time!!!  This is normally accomplished
+ * by relying on the module system to only have one copy of the module
+ * loaded, and then by giving the lock_torture_stats kthread full control
+ * (or the init/cleanup functions when lock_torture_stats thread is not
+ * running).
+ */
+static void lock_torture_stats_print(void)
+{
+       int size = nrealwriters_stress * 200 + 8192;
+       char *buf;
+
+       buf = kmalloc(size, GFP_KERNEL);
+       if (!buf) {
+               pr_err("lock_torture_stats_print: Out of memory, need: %d",
+                      size);
+               return;
+       }
+       lock_torture_printk(buf);
+       pr_alert("%s", buf);
+       kfree(buf);
+}
+
+/*
+ * Periodically prints torture statistics, if periodic statistics printing
+ * was specified via the stat_interval module parameter.
+ *
+ * No need to worry about fullstop here, since this one doesn't reference
+ * volatile state or register callbacks.
+ */
+static int lock_torture_stats(void *arg)
+{
+       VERBOSE_TOROUT_STRING("lock_torture_stats task started");
+       do {
+               schedule_timeout_interruptible(stat_interval * HZ);
+               lock_torture_stats_print();
+               torture_shutdown_absorb("lock_torture_stats");
+       } while (!torture_must_stop());
+       torture_kthread_stopping("lock_torture_stats");
+       return 0;
+}
+
+static inline void
+lock_torture_print_module_parms(struct lock_torture_ops *cur_ops,
+                               const char *tag)
+{
+       pr_alert("%s" TORTURE_FLAG
+                "--- %s: nwriters_stress=%d stat_interval=%d verbose=%d shuffle_interval=%d stutter=%d shutdown_secs=%d onoff_interval=%d onoff_holdoff=%d\n",
+                torture_type, tag, nrealwriters_stress, stat_interval, verbose,
+                shuffle_interval, stutter, shutdown_secs,
+                onoff_interval, onoff_holdoff);
+}
+
+static void lock_torture_cleanup(void)
+{
+       int i;
+
+       if (torture_cleanup())
+               return;
+
+       if (writer_tasks) {
+               for (i = 0; i < nrealwriters_stress; i++)
+                       torture_stop_kthread(lock_torture_writer,
+                                            writer_tasks[i]);
+               kfree(writer_tasks);
+               writer_tasks = NULL;
+       }
+
+       torture_stop_kthread(lock_torture_stats, stats_task);
+       lock_torture_stats_print();  /* -After- the stats thread is stopped! */
+
+       if (atomic_read(&n_lock_torture_errors))
+               lock_torture_print_module_parms(cur_ops,
+                                               "End of test: FAILURE");
+       else if (torture_onoff_failures())
+               lock_torture_print_module_parms(cur_ops,
+                                               "End of test: LOCK_HOTPLUG");
+       else
+               lock_torture_print_module_parms(cur_ops,
+                                               "End of test: SUCCESS");
+}
+
+static int __init lock_torture_init(void)
+{
+       int i;
+       int firsterr = 0;
+       static struct lock_torture_ops *torture_ops[] = {
+               &spin_lock_ops, &spin_lock_irq_ops,
+       };
+
+       torture_init_begin(torture_type, verbose, &locktorture_runnable);
+
+       /* Process args and tell the world that the torturer is on the job. */
+       for (i = 0; i < ARRAY_SIZE(torture_ops); i++) {
+               cur_ops = torture_ops[i];
+               if (strcmp(torture_type, cur_ops->name) == 0)
+                       break;
+       }
+       if (i == ARRAY_SIZE(torture_ops)) {
+               pr_alert("lock-torture: invalid torture type: \"%s\"\n",
+                        torture_type);
+               pr_alert("lock-torture types:");
+               for (i = 0; i < ARRAY_SIZE(torture_ops); i++)
+                       pr_alert(" %s", torture_ops[i]->name);
+               pr_alert("\n");
+               torture_init_end();
+               return -EINVAL;
+       }
+       if (cur_ops->init)
+               cur_ops->init(); /* no "goto unwind" prior to this point!!! */
+
+       if (nwriters_stress >= 0)
+               nrealwriters_stress = nwriters_stress;
+       else
+               nrealwriters_stress = 2 * num_online_cpus();
+       lock_torture_print_module_parms(cur_ops, "Start of test");
+
+       /* Initialize the statistics so that each run gets its own numbers. */
+
+       lock_is_write_held = 0;
+       lwsa = kmalloc(sizeof(*lwsa) * nrealwriters_stress, GFP_KERNEL);
+       if (lwsa == NULL) {
+               VERBOSE_TOROUT_STRING("lwsa: Out of memory");
+               firsterr = -ENOMEM;
+               goto unwind;
+       }
+       for (i = 0; i < nrealwriters_stress; i++) {
+               lwsa[i].n_write_lock_fail = 0;
+               lwsa[i].n_write_lock_acquired = 0;
+       }
+
+       /* Start up the kthreads. */
+
+       if (onoff_interval > 0) {
+               firsterr = torture_onoff_init(onoff_holdoff * HZ,
+                                             onoff_interval * HZ);
+               if (firsterr)
+                       goto unwind;
+       }
+       if (shuffle_interval > 0) {
+               firsterr = torture_shuffle_init(shuffle_interval);
+               if (firsterr)
+                       goto unwind;
+       }
+       if (shutdown_secs > 0) {
+               firsterr = torture_shutdown_init(shutdown_secs,
+                                                lock_torture_cleanup);
+               if (firsterr)
+                       goto unwind;
+       }
+       if (stutter > 0) {
+               firsterr = torture_stutter_init(stutter);
+               if (firsterr)
+                       goto unwind;
+       }
+
+       writer_tasks = kzalloc(nrealwriters_stress * sizeof(writer_tasks[0]),
+                              GFP_KERNEL);
+       if (writer_tasks == NULL) {
+               VERBOSE_TOROUT_ERRSTRING("writer_tasks: Out of memory");
+               firsterr = -ENOMEM;
+               goto unwind;
+       }
+       for (i = 0; i < nrealwriters_stress; i++) {
+               firsterr = torture_create_kthread(lock_torture_writer, &lwsa[i],
+                                                 writer_tasks[i]);
+               if (firsterr)
+                       goto unwind;
+       }
+       if (stat_interval > 0) {
+               firsterr = torture_create_kthread(lock_torture_stats, NULL,
+                                                 stats_task);
+               if (firsterr)
+                       goto unwind;
+       }
+       torture_init_end();
+       return 0;
+
+unwind:
+       torture_init_end();
+       lock_torture_cleanup();
+       return firsterr;
+}
+
+module_init(lock_torture_init);
+module_exit(lock_torture_cleanup);
index 2bfb4e5cdf8cf325f9acb0e7aacc8f3e2a8c021c..dd7f8858188a6ac92ac19bb7ae032d62786ee612 100644 (file)
@@ -980,6 +980,21 @@ config DEBUG_LOCKING_API_SELFTESTS
          The following locking APIs are covered: spinlocks, rwlocks,
          mutexes and rwsems.
 
+config LOCK_TORTURE_TEST
+       tristate "torture tests for locking"
+       depends on DEBUG_KERNEL
+       select TORTURE_TEST
+       default n
+       help
+         This option provides a kernel module that runs torture tests
+         on kernel locking primitives.  The kernel module may be built
+         after the fact on the running kernel to be tested, if desired.
+
+         Say Y here if you want kernel locking-primitive torture tests
+         to be built into the kernel.
+         Say M if you want these torture tests to build as a module.
+         Say N if you are unsure.
+
 endmenu # lock debugging
 
 config TRACE_IRQFLAGS