Merge branch 'akpm' (Andrew's patch-bomb)

Merge misc patches from Andrew Morton:

 - the "misc" tree - stuff from all over the map

 - checkpatch updates

 - fatfs

 - kmod changes

 - procfs

 - cpumask

 - UML

 - kexec

 - mqueue

 - rapidio

 - pidns

 - some checkpoint-restore feature work.  Reluctantly.  Most of it
   delayed a release.  I'm still rather worried that we don't have a
   clear roadmap to completion for this work.

* emailed from Andrew Morton <akpm@linux-foundation.org>: (78 patches)
  kconfig: update compression algorithm info
  c/r: prctl: add ability to set new mm_struct::exe_file
  c/r: prctl: extend PR_SET_MM to set up more mm_struct entries
  c/r: procfs: add arg_start/end, env_start/end and exit_code members to /proc/$pid/stat
  syscalls, x86: add __NR_kcmp syscall
  fs, proc: introduce /proc/<pid>/task/<tid>/children entry
  sysctl: make kernel.ns_last_pid control dependent on CHECKPOINT_RESTORE
  aio/vfs: cleanup of rw_copy_check_uvector() and compat_rw_copy_check_uvector()
  eventfd: change int to __u64 in eventfd_signal()
  fs/nls: add Apple NLS
  pidns: make killed children autoreap
  pidns: use task_active_pid_ns in do_notify_parent
  rapidio/tsi721: add DMA engine support
  rapidio: add DMA engine support for RIO data transfers
  ipc/mqueue: add rbtree node caching support
  tools/selftests: add mq_perf_tests
  ipc/mqueue: strengthen checks on mqueue creation
  ipc/mqueue: correct mq_attr_ok test
  ipc/mqueue: improve performance of send/recv
  selftests: add mq_open_tests
  ...

7 files changed, 1369 insertions, 1 deletions

diff --git a/tools/testing/selftests/Makefile b/tools/testing/selftests/Makefile
index 28bc57ee757c..a4162e15c25f 100644
--- a/tools/testing/selftests/Makefile
+++ b/tools/testing/selftests/Makefile
@@ -1,4 +1,4 @@
-TARGETS = breakpoints vm
+TARGETS = breakpoints kcmp mqueue vm
 
 all:
 	for TARGET in $(TARGETS); do \
diff --git a/tools/testing/selftests/kcmp/Makefile b/tools/testing/selftests/kcmp/Makefile
new file mode 100644
index 000000000000..dc79b86ea65c
--- /dev/null
+++ b/tools/testing/selftests/kcmp/Makefile
@@ -0,0 +1,29 @@
+uname_M := $(shell uname -m 2>/dev/null || echo not)
+ARCH ?= $(shell echo $(uname_M) | sed -e s/i.86/i386/)
+ifeq ($(ARCH),i386)
+        ARCH := X86
+	CFLAGS := -DCONFIG_X86_32 -D__i386__
+endif
+ifeq ($(ARCH),x86_64)
+	ARCH := X86
+	CFLAGS := -DCONFIG_X86_64 -D__x86_64__
+endif
+
+CFLAGS += -I../../../../arch/x86/include/generated/
+CFLAGS += -I../../../../include/
+CFLAGS += -I../../../../usr/include/
+CFLAGS += -I../../../../arch/x86/include/
+
+all:
+ifeq ($(ARCH),X86)
+	gcc $(CFLAGS) kcmp_test.c -o run_test
+else
+	echo "Not an x86 target, can't build kcmp selftest"
+endif
+
+run-tests: all
+	./kcmp_test
+
+clean:
+	rm -fr ./run_test
+	rm -fr ./test-file
diff --git a/tools/testing/selftests/kcmp/kcmp_test.c b/tools/testing/selftests/kcmp/kcmp_test.c
new file mode 100644
index 000000000000..358cc6bfa35d
--- /dev/null
+++ b/tools/testing/selftests/kcmp/kcmp_test.c
@@ -0,0 +1,94 @@
+#define _GNU_SOURCE
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <signal.h>
+#include <limits.h>
+#include <unistd.h>
+#include <errno.h>
+#include <string.h>
+#include <fcntl.h>
+
+#include <linux/unistd.h>
+#include <linux/kcmp.h>
+
+#include <sys/syscall.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/wait.h>
+
+static long sys_kcmp(int pid1, int pid2, int type, int fd1, int fd2)
+{
+	return syscall(__NR_kcmp, pid1, pid2, type, fd1, fd2);
+}
+
+int main(int argc, char **argv)
+{
+	const char kpath[] = "kcmp-test-file";
+	int pid1, pid2;
+	int fd1, fd2;
+	int status;
+
+	fd1 = open(kpath, O_RDWR | O_CREAT | O_TRUNC, 0644);
+	pid1 = getpid();
+
+	if (fd1 < 0) {
+		perror("Can't create file");
+		exit(1);
+	}
+
+	pid2 = fork();
+	if (pid2 < 0) {
+		perror("fork failed");
+		exit(1);
+	}
+
+	if (!pid2) {
+		int pid2 = getpid();
+		int ret;
+
+		fd2 = open(kpath, O_RDWR, 0644);
+		if (fd2 < 0) {
+			perror("Can't open file");
+			exit(1);
+		}
+
+		/* An example of output and arguments */
+		printf("pid1: %6d pid2: %6d FD: %2ld FILES: %2ld VM: %2ld "
+		       "FS: %2ld SIGHAND: %2ld IO: %2ld SYSVSEM: %2ld "
+		       "INV: %2ld\n",
+		       pid1, pid2,
+		       sys_kcmp(pid1, pid2, KCMP_FILE,		fd1, fd2),
+		       sys_kcmp(pid1, pid2, KCMP_FILES,		0, 0),
+		       sys_kcmp(pid1, pid2, KCMP_VM,		0, 0),
+		       sys_kcmp(pid1, pid2, KCMP_FS,		0, 0),
+		       sys_kcmp(pid1, pid2, KCMP_SIGHAND,	0, 0),
+		       sys_kcmp(pid1, pid2, KCMP_IO,		0, 0),
+		       sys_kcmp(pid1, pid2, KCMP_SYSVSEM,	0, 0),
+
+			/* This one should fail */
+		       sys_kcmp(pid1, pid2, KCMP_TYPES + 1,	0, 0));
+
+		/* This one should return same fd */
+		ret = sys_kcmp(pid1, pid2, KCMP_FILE, fd1, fd1);
+		if (ret) {
+			printf("FAIL: 0 expected but %d returned\n", ret);
+			ret = -1;
+		} else
+			printf("PASS: 0 returned as expected\n");
+
+		/* Compare with self */
+		ret = sys_kcmp(pid1, pid1, KCMP_VM, 0, 0);
+		if (ret) {
+			printf("FAIL: 0 expected but %li returned\n", ret);
+			ret = -1;
+		} else
+			printf("PASS: 0 returned as expected\n");
+
+		exit(ret);
+	}
+
+	waitpid(pid2, &status, P_ALL);
+
+	return 0;
+}
diff --git a/tools/testing/selftests/mqueue/.gitignore b/tools/testing/selftests/mqueue/.gitignore
new file mode 100644
index 000000000000..d8d42377205a
--- /dev/null
+++ b/tools/testing/selftests/mqueue/.gitignore
@@ -0,0 +1,2 @@
+mq_open_tests
+mq_perf_tests
diff --git a/tools/testing/selftests/mqueue/Makefile b/tools/testing/selftests/mqueue/Makefile
new file mode 100644
index 000000000000..54c0aad2b47c
--- /dev/null
+++ b/tools/testing/selftests/mqueue/Makefile
@@ -0,0 +1,10 @@
+all:
+	gcc -O2 -lrt mq_open_tests.c -o mq_open_tests
+	gcc -O2 -lrt -lpthread -lpopt -o mq_perf_tests mq_perf_tests.c
+
+run_tests:
+	./mq_open_tests /test1
+	./mq_perf_tests
+
+clean:
+	rm -f mq_open_tests mq_perf_tests
diff --git a/tools/testing/selftests/mqueue/mq_open_tests.c b/tools/testing/selftests/mqueue/mq_open_tests.c
new file mode 100644
index 000000000000..711cc2923047
--- /dev/null
+++ b/tools/testing/selftests/mqueue/mq_open_tests.c
@@ -0,0 +1,492 @@
+/*
+ * This application is Copyright 2012 Red Hat, Inc.
+ *	Doug Ledford <dledford@redhat.com>
+ *
+ * mq_open_tests 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, version 3.
+ *
+ * mq_open_tests 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.
+ *
+ * For the full text of the license, see <http://www.gnu.org/licenses/>.
+ *
+ * mq_open_tests.c
+ *   Tests the various situations that should either succeed or fail to
+ *   open a posix message queue and then reports whether or not they
+ *   did as they were supposed to.
+ *
+ */
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <string.h>
+#include <limits.h>
+#include <errno.h>
+#include <sys/types.h>
+#include <sys/time.h>
+#include <sys/resource.h>
+#include <sys/stat.h>
+#include <mqueue.h>
+
+static char *usage =
+"Usage:\n"
+"  %s path\n"
+"\n"
+"	path	Path name of the message queue to create\n"
+"\n"
+"	Note: this program must be run as root in order to enable all tests\n"
+"\n";
+
+char *DEF_MSGS = "/proc/sys/fs/mqueue/msg_default";
+char *DEF_MSGSIZE = "/proc/sys/fs/mqueue/msgsize_default";
+char *MAX_MSGS = "/proc/sys/fs/mqueue/msg_max";
+char *MAX_MSGSIZE = "/proc/sys/fs/mqueue/msgsize_max";
+
+int default_settings;
+struct rlimit saved_limits, cur_limits;
+int saved_def_msgs, saved_def_msgsize, saved_max_msgs, saved_max_msgsize;
+int cur_def_msgs, cur_def_msgsize, cur_max_msgs, cur_max_msgsize;
+FILE *def_msgs, *def_msgsize, *max_msgs, *max_msgsize;
+char *queue_path;
+mqd_t queue = -1;
+
+static inline void __set(FILE *stream, int value, char *err_msg);
+void shutdown(int exit_val, char *err_cause, int line_no);
+static inline int get(FILE *stream);
+static inline void set(FILE *stream, int value);
+static inline void getr(int type, struct rlimit *rlim);
+static inline void setr(int type, struct rlimit *rlim);
+void validate_current_settings();
+static inline void test_queue(struct mq_attr *attr, struct mq_attr *result);
+static inline int test_queue_fail(struct mq_attr *attr, struct mq_attr *result);
+
+static inline void __set(FILE *stream, int value, char *err_msg)
+{
+	rewind(stream);
+	if (fprintf(stream, "%d", value) < 0)
+		perror(err_msg);
+}
+
+
+void shutdown(int exit_val, char *err_cause, int line_no)
+{
+	static int in_shutdown = 0;
+
+	/* In case we get called recursively by a set() call below */
+	if (in_shutdown++)
+		return;
+
+	seteuid(0);
+
+	if (queue != -1)
+		if (mq_close(queue))
+			perror("mq_close() during shutdown");
+	if (queue_path)
+		/*
+		 * Be silent if this fails, if we cleaned up already it's
+		 * expected to fail
+		 */
+		mq_unlink(queue_path);
+	if (default_settings) {
+		if (saved_def_msgs)
+			__set(def_msgs, saved_def_msgs,
+			      "failed to restore saved_def_msgs");
+		if (saved_def_msgsize)
+			__set(def_msgsize, saved_def_msgsize,
+			      "failed to restore saved_def_msgsize");
+	}
+	if (saved_max_msgs)
+		__set(max_msgs, saved_max_msgs,
+		      "failed to restore saved_max_msgs");
+	if (saved_max_msgsize)
+		__set(max_msgsize, saved_max_msgsize,
+		      "failed to restore saved_max_msgsize");
+	if (exit_val)
+		error(exit_val, errno, "%s at %d", err_cause, line_no);
+	exit(0);
+}
+
+static inline int get(FILE *stream)
+{
+	int value;
+	rewind(stream);
+	if (fscanf(stream, "%d", &value) != 1)
+		shutdown(4, "Error reading /proc entry", __LINE__ - 1);
+	return value;
+}
+
+static inline void set(FILE *stream, int value)
+{
+	int new_value;
+
+	rewind(stream);
+	if (fprintf(stream, "%d", value) < 0)
+		return shutdown(5, "Failed writing to /proc file",
+				__LINE__ - 1);
+	new_value = get(stream);
+	if (new_value != value)
+		return shutdown(5, "We didn't get what we wrote to /proc back",
+				__LINE__ - 1);
+}
+
+static inline void getr(int type, struct rlimit *rlim)
+{
+	if (getrlimit(type, rlim))
+		shutdown(6, "getrlimit()", __LINE__ - 1);
+}
+
+static inline void setr(int type, struct rlimit *rlim)
+{
+	if (setrlimit(type, rlim))
+		shutdown(7, "setrlimit()", __LINE__ - 1);
+}
+
+void validate_current_settings()
+{
+	int rlim_needed;
+
+	if (cur_limits.rlim_cur < 4096) {
+		printf("Current rlimit value for POSIX message queue bytes is "
+		       "unreasonably low,\nincreasing.\n\n");
+		cur_limits.rlim_cur = 8192;
+		cur_limits.rlim_max = 16384;
+		setr(RLIMIT_MSGQUEUE, &cur_limits);
+	}
+
+	if (default_settings) {
+		rlim_needed = (cur_def_msgs + 1) * (cur_def_msgsize + 1 +
+						    2 * sizeof(void *));
+		if (rlim_needed > cur_limits.rlim_cur) {
+			printf("Temporarily lowering default queue parameters "
+			       "to something that will work\n"
+			       "with the current rlimit values.\n\n");
+			set(def_msgs, 10);
+			cur_def_msgs = 10;
+			set(def_msgsize, 128);
+			cur_def_msgsize = 128;
+		}
+	} else {
+		rlim_needed = (cur_max_msgs + 1) * (cur_max_msgsize + 1 +
+						    2 * sizeof(void *));
+		if (rlim_needed > cur_limits.rlim_cur) {
+			printf("Temporarily lowering maximum queue parameters "
+			       "to something that will work\n"
+			       "with the current rlimit values in case this is "
+			       "a kernel that ties the default\n"
+			       "queue parameters to the maximum queue "
+			       "parameters.\n\n");
+			set(max_msgs, 10);
+			cur_max_msgs = 10;
+			set(max_msgsize, 128);
+			cur_max_msgsize = 128;
+		}
+	}
+}
+
+/*
+ * test_queue - Test opening a queue, shutdown if we fail.  This should
+ * only be called in situations that should never fail.  We clean up
+ * after ourselves and return the queue attributes in *result.
+ */
+static inline void test_queue(struct mq_attr *attr, struct mq_attr *result)
+{
+	int flags = O_RDWR | O_EXCL | O_CREAT;
+	int perms = DEFFILEMODE;
+
+	if ((queue = mq_open(queue_path, flags, perms, attr)) == -1)
+		shutdown(1, "mq_open()", __LINE__);
+	if (mq_getattr(queue, result))
+		shutdown(1, "mq_getattr()", __LINE__);
+	if (mq_close(queue))
+		shutdown(1, "mq_close()", __LINE__);
+	queue = -1;
+	if (mq_unlink(queue_path))
+		shutdown(1, "mq_unlink()", __LINE__);
+}
+
+/*
+ * Same as test_queue above, but failure is not fatal.
+ * Returns:
+ * 0 - Failed to create a queue
+ * 1 - Created a queue, attributes in *result
+ */
+static inline int test_queue_fail(struct mq_attr *attr, struct mq_attr *result)
+{
+	int flags = O_RDWR | O_EXCL | O_CREAT;
+	int perms = DEFFILEMODE;
+
+	if ((queue = mq_open(queue_path, flags, perms, attr)) == -1)
+		return 0;
+	if (mq_getattr(queue, result))
+		shutdown(1, "mq_getattr()", __LINE__);
+	if (mq_close(queue))
+		shutdown(1, "mq_close()", __LINE__);
+	queue = -1;
+	if (mq_unlink(queue_path))
+		shutdown(1, "mq_unlink()", __LINE__);
+	return 1;
+}
+
+int main(int argc, char *argv[])
+{
+	struct mq_attr attr, result;
+
+	if (argc != 2) {
+		fprintf(stderr, "Must pass a valid queue name\n\n");
+		fprintf(stderr, usage, argv[0]);
+		exit(1);
+	}
+
+	/*
+	 * Although we can create a msg queue with a non-absolute path name,
+	 * unlink will fail.  So, if the name doesn't start with a /, add one
+	 * when we save it.
+	 */
+	if (*argv[1] == '/')
+		queue_path = strdup(argv[1]);
+	else {
+		queue_path = malloc(strlen(argv[1]) + 2);
+		if (!queue_path) {
+			perror("malloc()");
+			exit(1);
+		}
+		queue_path[0] = '/';
+		queue_path[1] = 0;
+		strcat(queue_path, argv[1]);
+	}
+
+	if (getuid() != 0) {
+		fprintf(stderr, "Not running as root, but almost all tests "
+			"require root in order to modify\nsystem settings.  "
+			"Exiting.\n");
+		exit(1);
+	}
+
+	/* Find out what files there are for us to make tweaks in */
+	def_msgs = fopen(DEF_MSGS, "r+");
+	def_msgsize = fopen(DEF_MSGSIZE, "r+");
+	max_msgs = fopen(MAX_MSGS, "r+");
+	max_msgsize = fopen(MAX_MSGSIZE, "r+");
+
+	if (!max_msgs)
+		shutdown(2, "Failed to open msg_max", __LINE__);
+	if (!max_msgsize)
+		shutdown(2, "Failed to open msgsize_max", __LINE__);
+	if (def_msgs || def_msgsize)
+		default_settings = 1;
+
+	/* Load up the current system values for everything we can */
+	getr(RLIMIT_MSGQUEUE, &saved_limits);
+	cur_limits = saved_limits;
+	if (default_settings) {
+		saved_def_msgs = cur_def_msgs = get(def_msgs);
+		saved_def_msgsize = cur_def_msgsize = get(def_msgsize);
+	}
+	saved_max_msgs = cur_max_msgs = get(max_msgs);
+	saved_max_msgsize = cur_max_msgsize = get(max_msgsize);
+
+	/* Tell the user our initial state */
+	printf("\nInitial system state:\n");
+	printf("\tUsing queue path:\t\t%s\n", queue_path);
+	printf("\tRLIMIT_MSGQUEUE(soft):\t\t%d\n", saved_limits.rlim_cur);
+	printf("\tRLIMIT_MSGQUEUE(hard):\t\t%d\n", saved_limits.rlim_max);
+	printf("\tMaximum Message Size:\t\t%d\n", saved_max_msgsize);
+	printf("\tMaximum Queue Size:\t\t%d\n", saved_max_msgs);
+	if (default_settings) {
+		printf("\tDefault Message Size:\t\t%d\n", saved_def_msgsize);
+		printf("\tDefault Queue Size:\t\t%d\n", saved_def_msgs);
+	} else {
+		printf("\tDefault Message Size:\t\tNot Supported\n");
+		printf("\tDefault Queue Size:\t\tNot Supported\n");
+	}
+	printf("\n");
+
+	validate_current_settings();
+
+	printf("Adjusted system state for testing:\n");
+	printf("\tRLIMIT_MSGQUEUE(soft):\t\t%d\n", cur_limits.rlim_cur);
+	printf("\tRLIMIT_MSGQUEUE(hard):\t\t%d\n", cur_limits.rlim_max);
+	printf("\tMaximum Message Size:\t\t%d\n", cur_max_msgsize);
+	printf("\tMaximum Queue Size:\t\t%d\n", cur_max_msgs);
+	if (default_settings) {
+		printf("\tDefault Message Size:\t\t%d\n", cur_def_msgsize);
+		printf("\tDefault Queue Size:\t\t%d\n", cur_def_msgs);
+	}
+
+	printf("\n\nTest series 1, behavior when no attr struct "
+	       "passed to mq_open:\n");
+	if (!default_settings) {
+		test_queue(NULL, &result);
+		printf("Given sane system settings, mq_open without an attr "
+		       "struct succeeds:\tPASS\n");
+		if (result.mq_maxmsg != cur_max_msgs ||
+		    result.mq_msgsize != cur_max_msgsize) {
+			printf("Kernel does not support setting the default "
+			       "mq attributes,\nbut also doesn't tie the "
+			       "defaults to the maximums:\t\t\tPASS\n");
+		} else {
+			set(max_msgs, ++cur_max_msgs);
+			set(max_msgsize, ++cur_max_msgsize);
+			test_queue(NULL, &result);
+			if (result.mq_maxmsg == cur_max_msgs &&
+			    result.mq_msgsize == cur_max_msgsize)
+				printf("Kernel does not support setting the "
+				       "default mq attributes and\n"
+				       "also ties system wide defaults to "
+				       "the system wide maximums:\t\t"
+				       "FAIL\n");
+			else
+				printf("Kernel does not support setting the "
+				       "default mq attributes,\n"
+				       "but also doesn't tie the defaults to "
+				       "the maximums:\t\t\tPASS\n");
+		}
+	} else {
+		printf("Kernel supports setting defaults separately from "
+		       "maximums:\t\tPASS\n");
+		/*
+		 * While we are here, go ahead and test that the kernel
+		 * properly follows the default settings
+		 */
+		test_queue(NULL, &result);
+		printf("Given sane values, mq_open without an attr struct "
+		       "succeeds:\t\tPASS\n");
+		if (result.mq_maxmsg != cur_def_msgs ||
+		    result.mq_msgsize != cur_def_msgsize)
+			printf("Kernel supports setting defaults, but does "
+			       "not actually honor them:\tFAIL\n\n");
+		else {
+			set(def_msgs, ++cur_def_msgs);
+			set(def_msgsize, ++cur_def_msgsize);
+			/* In case max was the same as the default */
+			set(max_msgs, ++cur_max_msgs);
+			set(max_msgsize, ++cur_max_msgsize);
+			test_queue(NULL, &result);
+			if (result.mq_maxmsg != cur_def_msgs ||
+			    result.mq_msgsize != cur_def_msgsize)
+				printf("Kernel supports setting defaults, but "
+				       "does not actually honor them:\t"
+				       "FAIL\n");
+			else
+				printf("Kernel properly honors default setting "
+				       "knobs:\t\t\t\tPASS\n");
+		}
+		set(def_msgs, cur_max_msgs + 1);
+		cur_def_msgs = cur_max_msgs + 1;
+		set(def_msgsize, cur_max_msgsize + 1);
+		cur_def_msgsize = cur_max_msgsize + 1;
+		if (cur_def_msgs * (cur_def_msgsize + 2 * sizeof(void *)) >=
+		    cur_limits.rlim_cur) {
+			cur_limits.rlim_cur = (cur_def_msgs + 2) *
+				(cur_def_msgsize + 2 * sizeof(void *));
+			cur_limits.rlim_max = 2 * cur_limits.rlim_cur;
+			setr(RLIMIT_MSGQUEUE, &cur_limits);
+		}
+		if (test_queue_fail(NULL, &result)) {
+			if (result.mq_maxmsg == cur_max_msgs &&
+			    result.mq_msgsize == cur_max_msgsize)
+				printf("Kernel properly limits default values "
+				       "to lesser of default/max:\t\tPASS\n");
+			else
+				printf("Kernel does not properly set default "
+				       "queue parameters when\ndefaults > "
+				       "max:\t\t\t\t\t\t\t\tFAIL\n");
+		} else
+			printf("Kernel fails to open mq because defaults are "
+			       "greater than maximums:\tFAIL\n");
+		set(def_msgs, --cur_def_msgs);
+		set(def_msgsize, --cur_def_msgsize);
+		cur_limits.rlim_cur = cur_limits.rlim_max = cur_def_msgs *
+			cur_def_msgsize;
+		setr(RLIMIT_MSGQUEUE, &cur_limits);
+		if (test_queue_fail(NULL, &result))
+			printf("Kernel creates queue even though defaults "
+			       "would exceed\nrlimit setting:"
+			       "\t\t\t\t\t\t\t\tFAIL\n");
+		else
+			printf("Kernel properly fails to create queue when "
+			       "defaults would\nexceed rlimit:"
+			       "\t\t\t\t\t\t\t\tPASS\n");
+	}
+
+	/*
+	 * Test #2 - open with an attr struct that exceeds rlimit
+	 */
+	printf("\n\nTest series 2, behavior when attr struct is "
+	       "passed to mq_open:\n");
+	cur_max_msgs = 32;
+	cur_max_msgsize = cur_limits.rlim_max >> 4;
+	set(max_msgs, cur_max_msgs);
+	set(max_msgsize, cur_max_msgsize);
+	attr.mq_maxmsg = cur_max_msgs;
+	attr.mq_msgsize = cur_max_msgsize;
+	if (test_queue_fail(&attr, &result))
+		printf("Queue open in excess of rlimit max when euid = 0 "
+		       "succeeded:\t\tFAIL\n");
+	else
+		printf("Queue open in excess of rlimit max when euid = 0 "
+		       "failed:\t\tPASS\n");
+	attr.mq_maxmsg = cur_max_msgs + 1;
+	attr.mq_msgsize = 10;
+	if (test_queue_fail(&attr, &result))
+		printf("Queue open with mq_maxmsg > limit when euid = 0 "
+		       "succeeded:\t\tPASS\n");
+	else
+		printf("Queue open with mq_maxmsg > limit when euid = 0 "
+		       "failed:\t\tFAIL\n");
+	attr.mq_maxmsg = 1;
+	attr.mq_msgsize = cur_max_msgsize + 1;
+	if (test_queue_fail(&attr, &result))
+		printf("Queue open with mq_msgsize > limit when euid = 0 "
+		       "succeeded:\t\tPASS\n");
+	else
+		printf("Queue open with mq_msgsize > limit when euid = 0 "
+		       "failed:\t\tFAIL\n");
+	attr.mq_maxmsg = 65536;
+	attr.mq_msgsize = 65536;
+	if (test_queue_fail(&attr, &result))
+		printf("Queue open with total size > 2GB when euid = 0 "
+		       "succeeded:\t\tFAIL\n");
+	else
+		printf("Queue open with total size > 2GB when euid = 0 "
+		       "failed:\t\t\tPASS\n");
+	seteuid(99);
+	attr.mq_maxmsg = cur_max_msgs;
+	attr.mq_msgsize = cur_max_msgsize;
+	if (test_queue_fail(&attr, &result))
+		printf("Queue open in excess of rlimit max when euid = 99 "
+		       "succeeded:\t\tFAIL\n");
+	else
+		printf("Queue open in excess of rlimit max when euid = 99 "
+		       "failed:\t\tPASS\n");
+	attr.mq_maxmsg = cur_max_msgs + 1;
+	attr.mq_msgsize = 10;
+	if (test_queue_fail(&attr, &result))
+		printf("Queue open with mq_maxmsg > limit when euid = 99 "
+		       "succeeded:\t\tFAIL\n");
+	else
+		printf("Queue open with mq_maxmsg > limit when euid = 99 "
+		       "failed:\t\tPASS\n");
+	attr.mq_maxmsg = 1;
+	attr.mq_msgsize = cur_max_msgsize + 1;
+	if (test_queue_fail(&attr, &result))
+		printf("Queue open with mq_msgsize > limit when euid = 99 "
+		       "succeeded:\t\tFAIL\n");
+	else
+		printf("Queue open with mq_msgsize > limit when euid = 99 "
+		       "failed:\t\tPASS\n");
+	attr.mq_maxmsg = 65536;
+	attr.mq_msgsize = 65536;
+	if (test_queue_fail(&attr, &result))
+		printf("Queue open with total size > 2GB when euid = 99 "
+		       "succeeded:\t\tFAIL\n");
+	else
+		printf("Queue open with total size > 2GB when euid = 99 "
+		       "failed:\t\t\tPASS\n");
+
+	shutdown(0,"",0);
+}
diff --git a/tools/testing/selftests/mqueue/mq_perf_tests.c b/tools/testing/selftests/mqueue/mq_perf_tests.c
new file mode 100644
index 000000000000..2fadd4b97045
--- /dev/null
+++ b/tools/testing/selftests/mqueue/mq_perf_tests.c
@@ -0,0 +1,741 @@
+/*
+ * This application is Copyright 2012 Red Hat, Inc.
+ *	Doug Ledford <dledford@redhat.com>
+ *
+ * mq_perf_tests 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, version 3.
+ *
+ * mq_perf_tests 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.
+ *
+ * For the full text of the license, see <http://www.gnu.org/licenses/>.
+ *
+ * mq_perf_tests.c
+ *   Tests various types of message queue workloads, concentrating on those
+ *   situations that invole large message sizes, large message queue depths,
+ *   or both, and reports back useful metrics about kernel message queue
+ *   performance.
+ *
+ */
+#define _GNU_SOURCE
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <string.h>
+#include <limits.h>
+#include <errno.h>
+#include <signal.h>
+#include <pthread.h>
+#include <sched.h>
+#include <sys/types.h>
+#include <sys/time.h>
+#include <sys/resource.h>
+#include <sys/stat.h>
+#include <mqueue.h>
+#include <popt.h>
+
+static char *usage =
+"Usage:\n"
+"  %s [-c #[,#..] -f] path\n"
+"\n"
+"	-c #	Skip most tests and go straight to a high queue depth test\n"
+"		and then run that test continuously (useful for running at\n"
+"		the same time as some other workload to see how much the\n"
+"		cache thrashing caused by adding messages to a very deep\n"
+"		queue impacts the performance of other programs).  The number\n"
+"		indicates which CPU core we should bind the process to during\n"
+"		the run.  If you have more than one physical CPU, then you\n"
+"		will need one copy per physical CPU package, and you should\n"
+"		specify the CPU cores to pin ourself to via a comma separated\n"
+"		list of CPU values.\n"
+"	-f	Only usable with continuous mode.  Pin ourself to the CPUs\n"
+"		as requested, then instead of looping doing a high mq\n"
+"		workload, just busy loop.  This will allow us to lock up a\n"
+"		single CPU just like we normally would, but without actually\n"
+"		thrashing the CPU cache.  This is to make it easier to get\n"
+"		comparable numbers from some other workload running on the\n"
+"		other CPUs.  One set of numbers with # CPUs locked up running\n"
+"		an mq workload, and another set of numbers with those same\n"
+"		CPUs locked away from the test workload, but not doing\n"
+"		anything to trash the cache like the mq workload might.\n"
+"	path	Path name of the message queue to create\n"
+"\n"
+"	Note: this program must be run as root in order to enable all tests\n"
+"\n";
+
+char *MAX_MSGS = "/proc/sys/fs/mqueue/msg_max";
+char *MAX_MSGSIZE = "/proc/sys/fs/mqueue/msgsize_max";
+
+#define min(a, b) ((a) < (b) ? (a) : (b))
+#define MAX_CPUS 64
+char *cpu_option_string;
+int cpus_to_pin[MAX_CPUS];
+int num_cpus_to_pin;
+pthread_t cpu_threads[MAX_CPUS];
+pthread_t main_thread;
+cpu_set_t *cpu_set;
+int cpu_set_size;
+int cpus_online;
+
+#define MSG_SIZE 16
+#define TEST1_LOOPS 10000000
+#define TEST2_LOOPS 100000
+int continuous_mode;
+int continuous_mode_fake;
+
+struct rlimit saved_limits, cur_limits;
+int saved_max_msgs, saved_max_msgsize;
+int cur_max_msgs, cur_max_msgsize;
+FILE *max_msgs, *max_msgsize;
+int cur_nice;
+char *queue_path = "/mq_perf_tests";
+mqd_t queue = -1;
+struct mq_attr result;
+int mq_prio_max;
+
+const struct poptOption options[] = {
+	{
+		.longName = "continuous",
+		.shortName = 'c',
+		.argInfo = POPT_ARG_STRING,
+		.arg = &cpu_option_string,
+		.val = 'c',
+		.descrip = "Run continuous tests at a high queue depth in "
+			"order to test the effects of cache thrashing on "
+			"other tasks on the system.  This test is intended "
+			"to be run on one core of each physical CPU while "
+			"some other CPU intensive task is run on all the other "
+			"cores of that same physical CPU and the other task "
+			"is timed.  It is assumed that the process of adding "
+			"messages to the message queue in a tight loop will "
+			"impact that other task to some degree.  Once the "
+			"tests are performed in this way, you should then "
+			"re-run the tests using fake mode in order to check "
+			"the difference in time required to perform the CPU "
+			"intensive task",
+		.argDescrip = "cpu[,cpu]",
+	},
+	{
+		.longName = "fake",
+		.shortName = 'f',
+		.argInfo = POPT_ARG_NONE,
+		.arg = &continuous_mode_fake,
+		.val = 0,
+		.descrip = "Tie up the CPUs that we would normally tie up in"
+			"continuous mode, but don't actually do any mq stuff, "
+			"just keep the CPU busy so it can't be used to process "
+			"system level tasks as this would free up resources on "
+			"the other CPU cores and skew the comparison between "
+			"the no-mqueue work and mqueue work tests",
+		.argDescrip = NULL,
+	},
+	{
+		.longName = "path",
+		.shortName = 'p',
+		.argInfo = POPT_ARG_STRING | POPT_ARGFLAG_SHOW_DEFAULT,
+		.arg = &queue_path,
+		.val = 'p',
+		.descrip = "The name of the path to use in the mqueue "
+			"filesystem for our tests",
+		.argDescrip = "pathname",
+	},
+	POPT_AUTOHELP
+	POPT_TABLEEND
+};
+
+static inline void __set(FILE *stream, int value, char *err_msg);
+void shutdown(int exit_val, char *err_cause, int line_no);
+void sig_action_SIGUSR1(int signum, siginfo_t *info, void *context);
+void sig_action(int signum, siginfo_t *info, void *context);
+static inline int get(FILE *stream);
+static inline void set(FILE *stream, int value);
+static inline int try_set(FILE *stream, int value);
+static inline void getr(int type, struct rlimit *rlim);
+static inline void setr(int type, struct rlimit *rlim);
+static inline void open_queue(struct mq_attr *attr);
+void increase_limits(void);
+
+static inline void __set(FILE *stream, int value, char *err_msg)
+{
+	rewind(stream);
+	if (fprintf(stream, "%d", value) < 0)
+		perror(err_msg);
+}
+
+
+void shutdown(int exit_val, char *err_cause, int line_no)
+{
+	static int in_shutdown = 0;
+	int errno_at_shutdown = errno;
+	int i;
+
+	/* In case we get called by multiple threads or from an sighandler */
+	if (in_shutdown++)
+		return;
+
+	for (i = 0; i < num_cpus_to_pin; i++)
+		if (cpu_threads[i]) {
+			pthread_kill(cpu_threads[i], SIGUSR1);
+			pthread_join(cpu_threads[i], NULL);
+		}
+
+	if (queue != -1)
+		if (mq_close(queue))
+			perror("mq_close() during shutdown");
+	if (queue_path)
+		/*
+		 * Be silent if this fails, if we cleaned up already it's
+		 * expected to fail
+		 */
+		mq_unlink(queue_path);
+	if (saved_max_msgs)
+		__set(max_msgs, saved_max_msgs,
+		      "failed to restore saved_max_msgs");
+	if (saved_max_msgsize)
+		__set(max_msgsize, saved_max_msgsize,
+		      "failed to restore saved_max_msgsize");
+	if (exit_val)
+		error(exit_val, errno_at_shutdown, "%s at %d",
+		      err_cause, line_no);
+	exit(0);
+}
+
+void sig_action_SIGUSR1(int signum, siginfo_t *info, void *context)
+{
+	if (pthread_self() != main_thread)
+		pthread_exit(0);
+	else {
+		fprintf(stderr, "Caught signal %d in SIGUSR1 handler, "
+				"exiting\n", signum);
+		shutdown(0, "", 0);
+		fprintf(stderr, "\n\nReturned from shutdown?!?!\n\n");
+		exit(0);
+	}
+}
+
+void sig_action(int signum, siginfo_t *info, void *context)
+{
+	if (pthread_self() != main_thread)
+		pthread_kill(main_thread, signum);
+	else {
+		fprintf(stderr, "Caught signal %d, exiting\n", signum);
+		shutdown(0, "", 0);
+		fprintf(stderr, "\n\nReturned from shutdown?!?!\n\n");
+		exit(0);
+	}
+}
+
+static inline int get(FILE *stream)
+{
+	int value;
+	rewind(stream);
+	if (fscanf(stream, "%d", &value) != 1)
+		shutdown(4, "Error reading /proc entry", __LINE__);
+	return value;
+}
+
+static inline void set(FILE *stream, int value)
+{
+	int new_value;
+
+	rewind(stream);
+	if (fprintf(stream, "%d", value) < 0)
+		return shutdown(5, "Failed writing to /proc file", __LINE__);
+	new_value = get(stream);
+	if (new_value != value)
+		return shutdown(5, "We didn't get what we wrote to /proc back",
+				__LINE__);
+}
+
+static inline int try_set(FILE *stream, int value)
+{
+	int new_value;
+
+	rewind(stream);
+	fprintf(stream, "%d", value);
+	new_value = get(stream);
+	return new_value == value;
+}
+
+static inline void getr(int type, struct rlimit *rlim)
+{
+	if (getrlimit(type, rlim))
+		shutdown(6, "getrlimit()", __LINE__);
+}
+
+static inline void setr(int type, struct rlimit *rlim)
+{
+	if (setrlimit(type, rlim))
+		shutdown(7, "setrlimit()", __LINE__);
+}
+
+/**
+ * open_queue - open the global queue for testing
+ * @attr - An attr struct specifying the desired queue traits
+ * @result - An attr struct that lists the actual traits the queue has
+ *
+ * This open is not allowed to fail, failure will result in an orderly
+ * shutdown of the program.  The global queue_path is used to set what
+ * queue to open, the queue descriptor is saved in the global queue
+ * variable.
+ */
+static inline void open_queue(struct mq_attr *attr)
+{
+	int flags = O_RDWR | O_EXCL | O_CREAT | O_NONBLOCK;
+	int perms = DEFFILEMODE;
+
+	queue = mq_open(queue_path, flags, perms, attr);
+	if (queue == -1)
+		shutdown(1, "mq_open()", __LINE__);
+	if (mq_getattr(queue, &result))
+		shutdown(1, "mq_getattr()", __LINE__);
+	printf("\n\tQueue %s created:\n", queue_path);
+	printf("\t\tmq_flags:\t\t\t%s\n", result.mq_flags & O_NONBLOCK ?
+	       "O_NONBLOCK" : "(null)");
+	printf("\t\tmq_maxmsg:\t\t\t%d\n", result.mq_maxmsg);
+	printf("\t\tmq_msgsize:\t\t\t%d\n", result.mq_msgsize);
+	printf("\t\tmq_curmsgs:\t\t\t%d\n", result.mq_curmsgs);
+}
+
+void *fake_cont_thread(void *arg)
+{
+	int i;
+
+	for (i = 0; i < num_cpus_to_pin; i++)
+		if (cpu_threads[i] == pthread_self())
+			break;
+	printf("\tStarted fake continuous mode thread %d on CPU %d\n", i,
+	       cpus_to_pin[i]);
+	while (1)
+		;
+}
+
+void *cont_thread(void *arg)
+{
+	char buff[MSG_SIZE];
+	int i, priority;
+
+	for (i = 0; i < num_cpus_to_pin; i++)
+		if (cpu_threads[i] == pthread_self())
+			break;
+	printf("\tStarted continuous mode thread %d on CPU %d\n", i,
+	       cpus_to_pin[i]);
+	while (1) {
+		while (mq_send(queue, buff, sizeof(buff), 0) == 0)
+			;
+		mq_receive(queue, buff, sizeof(buff), &priority);
+	}
+}
+
+#define drain_queue() \
+	while (mq_receive(queue, buff, MSG_SIZE, &prio_in) == MSG_SIZE)
+
+#define do_untimed_send() \
+	do { \
+		if (mq_send(queue, buff, MSG_SIZE, prio_out)) \
+			shutdown(3, "Test send failure", __LINE__); \
+	} while (0)
+
+#define do_send_recv() \
+	do { \
+		clock_gettime(clock, &start); \
+		if (mq_send(queue, buff, MSG_SIZE, prio_out)) \
+			shutdown(3, "Test send failure", __LINE__); \
+		clock_gettime(clock, &middle); \
+		if (mq_receive(queue, buff, MSG_SIZE, &prio_in) != MSG_SIZE) \
+			shutdown(3, "Test receive failure", __LINE__); \
+		clock_gettime(clock, &end); \
+		nsec = ((middle.tv_sec - start.tv_sec) * 1000000000) + \
+			(middle.tv_nsec - start.tv_nsec); \
+		send_total.tv_nsec += nsec; \
+		if (send_total.tv_nsec >= 1000000000) { \
+			send_total.tv_sec++; \
+			send_total.tv_nsec -= 1000000000; \
+		} \
+		nsec = ((end.tv_sec - middle.tv_sec) * 1000000000) + \
+			(end.tv_nsec - middle.tv_nsec); \
+		recv_total.tv_nsec += nsec; \
+		if (recv_total.tv_nsec >= 1000000000) { \
+			recv_total.tv_sec++; \
+			recv_total.tv_nsec -= 1000000000; \
+		} \
+	} while (0)
+
+struct test {
+	char *desc;
+	void (*func)(int *);
+};
+
+void const_prio(int *prio)
+{
+	return;
+}
+
+void inc_prio(int *prio)
+{
+	if (++*prio == mq_prio_max)
+		*prio = 0;
+}
+
+void dec_prio(int *prio)
+{
+	if (--*prio < 0)
+		*prio = mq_prio_max - 1;
+}
+
+void random_prio(int *prio)
+{
+	*prio = random() % mq_prio_max;
+}
+
+struct test test2[] = {
+	{"\n\tTest #2a: Time send/recv message, queue full, constant prio\n",
+		const_prio},
+	{"\n\tTest #2b: Time send/recv message, queue full, increasing prio\n",
+		inc_prio},
+	{"\n\tTest #2c: Time send/recv message, queue full, decreasing prio\n",
+		dec_prio},
+	{"\n\tTest #2d: Time send/recv message, queue full, random prio\n",
+		random_prio},
+	{NULL, NULL}
+};
+
+/**
+ * Tests to perform (all done with MSG_SIZE messages):
+ *
+ * 1) Time to add/remove message with 0 messages on queue
+ * 1a) with constant prio
+ * 2) Time to add/remove message when queue close to capacity:
+ * 2a) with constant prio
+ * 2b) with increasing prio
+ * 2c) with decreasing prio
+ * 2d) with random prio
+ * 3) Test limits of priorities honored (double check _SC_MQ_PRIO_MAX)
+ */
+void *perf_test_thread(void *arg)
+{
+	char buff[MSG_SIZE];
+	int prio_out, prio_in;
+	int i;
+	clockid_t clock;
+	pthread_t *t;
+	struct timespec res, start, middle, end, send_total, recv_total;
+	unsigned long long nsec;
+	struct test *cur_test;
+
+	t = &cpu_threads[0];
+	printf("\n\tStarted mqueue performance test thread on CPU %d\n",
+	       cpus_to_pin[0]);
+	mq_prio_max = sysconf(_SC_MQ_PRIO_MAX);
+	if (mq_prio_max == -1)
+		shutdown(2, "sysconf(_SC_MQ_PRIO_MAX)", __LINE__);
+	if (pthread_getcpuclockid(cpu_threads[0], &clock) != 0)
+		shutdown(2, "pthread_getcpuclockid", __LINE__);
+
+	if (clock_getres(clock, &res))
+		shutdown(2, "clock_getres()", __LINE__);
+
+	printf("\t\tMax priorities:\t\t\t%d\n", mq_prio_max);
+	printf("\t\tClock resolution:\t\t%d nsec%s\n", res.tv_nsec,
+	       res.tv_nsec > 1 ? "s" : "");
+
+
+
+	printf("\n\tTest #1: Time send/recv message, queue empty\n");
+	printf("\t\t(%d iterations)\n", TEST1_LOOPS);
+	prio_out = 0;
+	send_total.tv_sec = 0;
+	send_total.tv_nsec = 0;
+	recv_total.tv_sec = 0;
+	recv_total.tv_nsec = 0;
+	for (i = 0; i < TEST1_LOOPS; i++)
+		do_send_recv();
+	printf("\t\tSend msg:\t\t\t%d.%ds total time\n",
+	       send_total.tv_sec, send_total.tv_nsec);
+	nsec = ((unsigned long long)send_total.tv_sec * 1000000000 +
+		 send_total.tv_nsec) / TEST1_LOOPS;
+	printf("\t\t\t\t\t\t%d nsec/msg\n", nsec);
+	printf("\t\tRecv msg:\t\t\t%d.%ds total time\n",
+	       recv_total.tv_sec, recv_total.tv_nsec);
+	nsec = ((unsigned long long)recv_total.tv_sec * 1000000000 +
+		recv_total.tv_nsec) / TEST1_LOOPS;
+	printf("\t\t\t\t\t\t%d nsec/msg\n", nsec);
+
+
+	for (cur_test = test2; cur_test->desc != NULL; cur_test++) {
+		printf(cur_test->desc);
+		printf("\t\t(%d iterations)\n", TEST2_LOOPS);
+		prio_out = 0;
+		send_total.tv_sec = 0;
+		send_total.tv_nsec = 0;
+		recv_total.tv_sec = 0;
+		recv_total.tv_nsec = 0;
+		printf("\t\tFilling queue...");
+		fflush(stdout);
+		clock_gettime(clock, &start);
+		for (i = 0; i < result.mq_maxmsg - 1; i++) {
+			do_untimed_send();
+			cur_test->func(&prio_out);
+		}
+		clock_gettime(clock, &end);
+		nsec = ((unsigned long long)(end.tv_sec - start.tv_sec) *
+			1000000000) + (end.tv_nsec - start.tv_nsec);
+		printf("done.\t\t%lld.%llds\n", nsec / 1000000000,
+		       nsec % 1000000000);
+		printf("\t\tTesting...");
+		fflush(stdout);
+		for (i = 0; i < TEST2_LOOPS; i++) {
+			do_send_recv();
+			cur_test->func(&prio_out);
+		}
+		printf("done.\n");
+		printf("\t\tSend msg:\t\t\t%d.%ds total time\n",
+		       send_total.tv_sec, send_total.tv_nsec);
+		nsec = ((unsigned long long)send_total.tv_sec * 1000000000 +
+			 send_total.tv_nsec) / TEST2_LOOPS;
+		printf("\t\t\t\t\t\t%d nsec/msg\n", nsec);
+		printf("\t\tRecv msg:\t\t\t%d.%ds total time\n",
+		       recv_total.tv_sec, recv_total.tv_nsec);
+		nsec = ((unsigned long long)recv_total.tv_sec * 1000000000 +
+			recv_total.tv_nsec) / TEST2_LOOPS;
+		printf("\t\t\t\t\t\t%d nsec/msg\n", nsec);
+		printf("\t\tDraining queue...");
+		fflush(stdout);
+		clock_gettime(clock, &start);
+		drain_queue();
+		clock_gettime(clock, &end);
+		nsec = ((unsigned long long)(end.tv_sec - start.tv_sec) *
+			1000000000) + (end.tv_nsec - start.tv_nsec);
+		printf("done.\t\t%lld.%llds\n", nsec / 1000000000,
+		       nsec % 1000000000);
+	}
+	return 0;
+}
+
+void increase_limits(void)
+{
+	cur_limits.rlim_cur = RLIM_INFINITY;
+	cur_limits.rlim_max = RLIM_INFINITY;
+	setr(RLIMIT_MSGQUEUE, &cur_limits);
+	while (try_set(max_msgs, cur_max_msgs += 10))
+		;
+	cur_max_msgs = get(max_msgs);
+	while (try_set(max_msgsize, cur_max_msgsize += 1024))
+		;
+	cur_max_msgsize = get(max_msgsize);
+	if (setpriority(PRIO_PROCESS, 0, -20) != 0)
+		shutdown(2, "setpriority()", __LINE__);
+	cur_nice = -20;
+}
+
+int main(int argc, char *argv[])
+{
+	struct mq_attr attr;
+	char *option, *next_option;
+	int i, cpu;
+	struct sigaction sa;
+	poptContext popt_context;
+	char rc;
+	void *retval;
+
+	main_thread = pthread_self();
+	num_cpus_to_pin = 0;
+
+	if (sysconf(_SC_NPROCESSORS_ONLN) == -1) {
+		perror("sysconf(_SC_NPROCESSORS_ONLN)");
+		exit(1);
+	}
+	cpus_online = min(MAX_CPUS, sysconf(_SC_NPROCESSORS_ONLN));
+	cpu_set = CPU_ALLOC(cpus_online);
+	if (cpu_set == NULL) {
+		perror("CPU_ALLOC()");
+		exit(1);
+	}
+	cpu_set_size = CPU_ALLOC_SIZE(cpus_online);
+	CPU_ZERO_S(cpu_set_size, cpu_set);
+
+	popt_context = poptGetContext(NULL, argc, (const char **)argv,
+				      options, 0);
+
+	while ((rc = poptGetNextOpt(popt_context)) > 0) {
+		switch (rc) {
+		case 'c':
+			continuous_mode = 1;
+			option = cpu_option_string;
+			do {
+				next_option = strchr(option, ',');
+				if (next_option)
+					*next_option = '\0';
+				cpu = atoi(option);
+				if (cpu >= cpus_online)
+					fprintf(stderr, "CPU %d exceeds "
+						"cpus online, ignoring.\n",
+						cpu);
+				else
+					cpus_to_pin[num_cpus_to_pin++] = cpu;
+				if (next_option)
+					option = ++next_option;
+			} while (next_option && num_cpus_to_pin < MAX_CPUS);
+			/* Double check that they didn't give us the same CPU
+			 * more than once */
+			for (cpu = 0; cpu < num_cpus_to_pin; cpu++) {
+				if (CPU_ISSET_S(cpus_to_pin[cpu], cpu_set_size,
+						cpu_set)) {
+					fprintf(stderr, "Any given CPU may "
+						"only be given once.\n");
+					exit(1);
+				} else
+					CPU_SET_S(cpus_to_pin[cpu],
+						  cpu_set_size, cpu_set);
+			}
+			break;
+		case 'p':
+			/*
+			 * Although we can create a msg queue with a
+			 * non-absolute path name, unlink will fail.  So,
+			 * if the name doesn't start with a /, add one
+			 * when we save it.
+			 */
+			option = queue_path;
+			if (*option != '/') {
+				queue_path = malloc(strlen(option) + 2);
+				if (!queue_path) {
+					perror("malloc()");
+					exit(1);
+				}
+				queue_path[0] = '/';
+				queue_path[1] = 0;
+				strcat(queue_path, option);
+				free(option);
+			}
+			break;
+		}
+	}
+
+	if (continuous_mode && num_cpus_to_pin == 0) {
+		fprintf(stderr, "Must pass at least one CPU to continuous "
+			"mode.\n");
+		poptPrintUsage(popt_context, stderr, 0);
+		exit(1);
+	} else if (!continuous_mode) {
+		num_cpus_to_pin = 1;
+		cpus_to_pin[0] = cpus_online - 1;
+	}
+
+	if (getuid() != 0) {
+		fprintf(stderr, "Not running as root, but almost all tests "
+			"require root in order to modify\nsystem settings.  "
+			"Exiting.\n");
+		exit(1);
+	}
+
+	max_msgs = fopen(MAX_MSGS, "r+");
+	max_msgsize = fopen(MAX_MSGSIZE, "r+");
+	if (!max_msgs)
+		shutdown(2, "Failed to open msg_max", __LINE__);
+	if (!max_msgsize)
+		shutdown(2, "Failed to open msgsize_max", __LINE__);
+
+	/* Load up the current system values for everything we can */
+	getr(RLIMIT_MSGQUEUE, &saved_limits);
+	cur_limits = saved_limits;
+	saved_max_msgs = cur_max_msgs = get(max_msgs);
+	saved_max_msgsize = cur_max_msgsize = get(max_msgsize);
+	errno = 0;
+	cur_nice = getpriority(PRIO_PROCESS, 0);
+	if (errno)
+		shutdown(2, "getpriority()", __LINE__);
+
+	/* Tell the user our initial state */
+	printf("\nInitial system state:\n");
+	printf("\tUsing queue path:\t\t\t%s\n", queue_path);
+	printf("\tRLIMIT_MSGQUEUE(soft):\t\t\t%d\n", saved_limits.rlim_cur);
+	printf("\tRLIMIT_MSGQUEUE(hard):\t\t\t%d\n", saved_limits.rlim_max);
+	printf("\tMaximum Message Size:\t\t\t%d\n", saved_max_msgsize);
+	printf("\tMaximum Queue Size:\t\t\t%d\n", saved_max_msgs);
+	printf("\tNice value:\t\t\t\t%d\n", cur_nice);
+	printf("\n");
+
+	increase_limits();
+
+	printf("Adjusted system state for testing:\n");
+	if (cur_limits.rlim_cur == RLIM_INFINITY) {
+		printf("\tRLIMIT_MSGQUEUE(soft):\t\t\t(unlimited)\n");
+		printf("\tRLIMIT_MSGQUEUE(hard):\t\t\t(unlimited)\n");
+	} else {
+		printf("\tRLIMIT_MSGQUEUE(soft):\t\t\t%d\n",
+		       cur_limits.rlim_cur);
+		printf("\tRLIMIT_MSGQUEUE(hard):\t\t\t%d\n",
+		       cur_limits.rlim_max);
+	}
+	printf("\tMaximum Message Size:\t\t\t%d\n", cur_max_msgsize);
+	printf("\tMaximum Queue Size:\t\t\t%d\n", cur_max_msgs);
+	printf("\tNice value:\t\t\t\t%d\n", cur_nice);
+	printf("\tContinuous mode:\t\t\t(%s)\n", continuous_mode ?
+	       (continuous_mode_fake ? "fake mode" : "enabled") :
+	       "disabled");
+	printf("\tCPUs to pin:\t\t\t\t%d", cpus_to_pin[0]);
+	for (cpu = 1; cpu < num_cpus_to_pin; cpu++)
+			printf(",%d", cpus_to_pin[cpu]);
+	printf("\n");
+
+	sa.sa_sigaction = sig_action_SIGUSR1;
+	sigemptyset(&sa.sa_mask);
+	sigaddset(&sa.sa_mask, SIGHUP);
+	sigaddset(&sa.sa_mask, SIGINT);
+	sigaddset(&sa.sa_mask, SIGQUIT);
+	sigaddset(&sa.sa_mask, SIGTERM);
+	sa.sa_flags = SA_SIGINFO;
+	if (sigaction(SIGUSR1, &sa, NULL) == -1)
+		shutdown(1, "sigaction(SIGUSR1)", __LINE__);
+	sa.sa_sigaction = sig_action;
+	if (sigaction(SIGHUP, &sa, NULL) == -1)
+		shutdown(1, "sigaction(SIGHUP)", __LINE__);
+	if (sigaction(SIGINT, &sa, NULL) == -1)
+		shutdown(1, "sigaction(SIGINT)", __LINE__);
+	if (sigaction(SIGQUIT, &sa, NULL) == -1)
+		shutdown(1, "sigaction(SIGQUIT)", __LINE__);
+	if (sigaction(SIGTERM, &sa, NULL) == -1)
+		shutdown(1, "sigaction(SIGTERM)", __LINE__);
+
+	if (!continuous_mode_fake) {
+		attr.mq_flags = O_NONBLOCK;
+		attr.mq_maxmsg = cur_max_msgs;
+		attr.mq_msgsize = MSG_SIZE;
+		open_queue(&attr);
+	}
+	for (i = 0; i < num_cpus_to_pin; i++) {
+		pthread_attr_t thread_attr;
+		void *thread_func;
+
+		if (continuous_mode_fake)
+			thread_func = &fake_cont_thread;
+		else if (continuous_mode)
+			thread_func = &cont_thread;
+		else
+			thread_func = &perf_test_thread;
+
+		CPU_ZERO_S(cpu_set_size, cpu_set);
+		CPU_SET_S(cpus_to_pin[i], cpu_set_size, cpu_set);
+		pthread_attr_init(&thread_attr);
+		pthread_attr_setaffinity_np(&thread_attr, cpu_set_size,
+					    cpu_set);
+		if (pthread_create(&cpu_threads[i], &thread_attr, thread_func,
+				   NULL))
+			shutdown(1, "pthread_create()", __LINE__);
+		pthread_attr_destroy(&thread_attr);
+	}
+
+	if (!continuous_mode) {
+		pthread_join(cpu_threads[0], &retval);
+		shutdown((long)retval, "perf_test_thread()", __LINE__);
+	} else {
+		while (1)
+			sleep(1);
+	}
+	shutdown(0, "", 0);
+}