Merge branch 'locking-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull locking updates from Ingo Molnar:
 "Here are the locking changes in this cycle:

   - rwsem unification and simpler micro-optimizations to prepare for
     more intrusive (and more lucrative) scalability improvements in
     v5.3 (Waiman Long)

   - Lockdep irq state tracking flag usage cleanups (Frederic
     Weisbecker)

   - static key improvements (Jakub Kicinski, Peter Zijlstra)

   - misc updates, cleanups and smaller fixes"

* 'locking-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (26 commits)
  locking/lockdep: Remove unnecessary unlikely()
  locking/static_key: Don't take sleeping locks in __static_key_slow_dec_deferred()
  locking/static_key: Factor out the fast path of static_key_slow_dec()
  locking/static_key: Add support for deferred static branches
  locking/lockdep: Test all incompatible scenarios at once in check_irq_usage()
  locking/lockdep: Avoid bogus Clang warning
  locking/lockdep: Generate LOCKF_ bit composites
  locking/lockdep: Use expanded masks on find_usage_*() functions
  locking/lockdep: Map remaining magic numbers to lock usage mask names
  locking/lockdep: Move valid_state() inside CONFIG_TRACE_IRQFLAGS && CONFIG_PROVE_LOCKING
  locking/rwsem: Prevent unneeded warning during locking selftest
  locking/rwsem: Optimize rwsem structure for uncontended lock acquisition
  locking/rwsem: Enable lock event counting
  locking/lock_events: Don't show pvqspinlock events on bare metal
  locking/lock_events: Make lock_events available for all archs & other locks
  locking/qspinlock_stat: Introduce generic lockevent_*() counting APIs
  locking/rwsem: Enhance DEBUG_RWSEMS_WARN_ON() macro
  locking/rwsem: Add debug check for __down_read*()
  locking/rwsem: Micro-optimize rwsem_try_read_lock_unqueued()
  locking/rwsem: Move rwsem internal function declarations to rwsem-xadd.h
  ...

1 files changed, 179 insertions, 0 deletions

diff --git a/kernel/locking/lock_events.c b/kernel/locking/lock_events.c
new file mode 100644
index 000000000000..fa2c2f951c6b
--- /dev/null
+++ b/kernel/locking/lock_events.c
@@ -0,0 +1,179 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * 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.
+ *
+ * Authors: Waiman Long <waiman.long@hpe.com>
+ */
+
+/*
+ * Collect locking event counts
+ */
+#include <linux/debugfs.h>
+#include <linux/sched.h>
+#include <linux/sched/clock.h>
+#include <linux/fs.h>
+
+#include "lock_events.h"
+
+#undef  LOCK_EVENT
+#define LOCK_EVENT(name)	[LOCKEVENT_ ## name] = #name,
+
+#define LOCK_EVENTS_DIR		"lock_event_counts"
+
+/*
+ * When CONFIG_LOCK_EVENT_COUNTS is enabled, event counts of different
+ * types of locks will be reported under the <debugfs>/lock_event_counts/
+ * directory. See lock_events_list.h for the list of available locking
+ * events.
+ *
+ * Writing to the special ".reset_counts" file will reset all the above
+ * locking event counts. This is a very slow operation and so should not
+ * be done frequently.
+ *
+ * These event counts are implemented as per-cpu variables which are
+ * summed and computed whenever the corresponding debugfs files are read. This
+ * minimizes added overhead making the counts usable even in a production
+ * environment.
+ */
+static const char * const lockevent_names[lockevent_num + 1] = {
+
+#include "lock_events_list.h"
+
+	[LOCKEVENT_reset_cnts] = ".reset_counts",
+};
+
+/*
+ * Per-cpu counts
+ */
+DEFINE_PER_CPU(unsigned long, lockevents[lockevent_num]);
+
+/*
+ * The lockevent_read() function can be overridden.
+ */
+ssize_t __weak lockevent_read(struct file *file, char __user *user_buf,
+			      size_t count, loff_t *ppos)
+{
+	char buf[64];
+	int cpu, id, len;
+	u64 sum = 0;
+
+	/*
+	 * Get the counter ID stored in file->f_inode->i_private
+	 */
+	id = (long)file_inode(file)->i_private;
+
+	if (id >= lockevent_num)
+		return -EBADF;
+
+	for_each_possible_cpu(cpu)
+		sum += per_cpu(lockevents[id], cpu);
+	len = snprintf(buf, sizeof(buf) - 1, "%llu\n", sum);
+
+	return simple_read_from_buffer(user_buf, count, ppos, buf, len);
+}
+
+/*
+ * Function to handle write request
+ *
+ * When idx = reset_cnts, reset all the counts.
+ */
+static ssize_t lockevent_write(struct file *file, const char __user *user_buf,
+			   size_t count, loff_t *ppos)
+{
+	int cpu;
+
+	/*
+	 * Get the counter ID stored in file->f_inode->i_private
+	 */
+	if ((long)file_inode(file)->i_private != LOCKEVENT_reset_cnts)
+		return count;
+
+	for_each_possible_cpu(cpu) {
+		int i;
+		unsigned long *ptr = per_cpu_ptr(lockevents, cpu);
+
+		for (i = 0 ; i < lockevent_num; i++)
+			WRITE_ONCE(ptr[i], 0);
+	}
+	return count;
+}
+
+/*
+ * Debugfs data structures
+ */
+static const struct file_operations fops_lockevent = {
+	.read = lockevent_read,
+	.write = lockevent_write,
+	.llseek = default_llseek,
+};
+
+#ifdef CONFIG_PARAVIRT_SPINLOCKS
+#include <asm/paravirt.h>
+
+static bool __init skip_lockevent(const char *name)
+{
+	static int pv_on __initdata = -1;
+
+	if (pv_on < 0)
+		pv_on = !pv_is_native_spin_unlock();
+	/*
+	 * Skip PV qspinlock events on bare metal.
+	 */
+	if (!pv_on && !memcmp(name, "pv_", 3))
+		return true;
+	return false;
+}
+#else
+static inline bool skip_lockevent(const char *name)
+{
+	return false;
+}
+#endif
+
+/*
+ * Initialize debugfs for the locking event counts.
+ */
+static int __init init_lockevent_counts(void)
+{
+	struct dentry *d_counts = debugfs_create_dir(LOCK_EVENTS_DIR, NULL);
+	int i;
+
+	if (!d_counts)
+		goto out;
+
+	/*
+	 * Create the debugfs files
+	 *
+	 * As reading from and writing to the stat files can be slow, only
+	 * root is allowed to do the read/write to limit impact to system
+	 * performance.
+	 */
+	for (i = 0; i < lockevent_num; i++) {
+		if (skip_lockevent(lockevent_names[i]))
+			continue;
+		if (!debugfs_create_file(lockevent_names[i], 0400, d_counts,
+					 (void *)(long)i, &fops_lockevent))
+			goto fail_undo;
+	}
+
+	if (!debugfs_create_file(lockevent_names[LOCKEVENT_reset_cnts], 0200,
+				 d_counts, (void *)(long)LOCKEVENT_reset_cnts,
+				 &fops_lockevent))
+		goto fail_undo;
+
+	return 0;
+fail_undo:
+	debugfs_remove_recursive(d_counts);
+out:
+	pr_warn("Could not create '%s' debugfs entries\n", LOCK_EVENTS_DIR);
+	return -ENOMEM;
+}
+fs_initcall(init_lockevent_counts);