test/mcslock: add MCS queued lock unit test
authorPhil Yang <phil.yang@arm.com>
Fri, 5 Jul 2019 10:27:08 +0000 (18:27 +0800)
committerThomas Monjalon <thomas@monjalon.net>
Mon, 8 Jul 2019 08:57:14 +0000 (10:57 +0200)
Unit test and perf test for MCS queued lock.

Signed-off-by: Phil Yang <phil.yang@arm.com>
Reviewed-by: Gavin Hu <gavin.hu@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
MAINTAINERS
app/test/Makefile
app/test/autotest_data.py
app/test/autotest_test_funcs.py
app/test/meson.build
app/test/test_mcslock.c [new file with mode: 0644]

index 49f2aa9..4534e3e 100644 (file)
@@ -226,6 +226,7 @@ F: app/test/test_bitmap.c
 MCSlock - EXPERIMENTAL
 M: Phil Yang <phil.yang@arm.com>
 F: lib/librte_eal/common/include/generic/rte_mcslock.h
+F: app/test/test_mcslock.c
 
 Ticketlock
 M: Joyce Kong <joyce.kong@arm.com>
index 495b2fd..26ba6fe 100644 (file)
@@ -64,6 +64,7 @@ SRCS-y += test_atomic.c
 SRCS-y += test_barrier.c
 SRCS-y += test_malloc.c
 SRCS-y += test_cycles.c
+SRCS-y += test_mcslock.c
 SRCS-y += test_spinlock.c
 SRCS-y += test_ticketlock.c
 SRCS-y += test_memory.c
index 7e67e98..7405149 100644 (file)
@@ -176,6 +176,12 @@ parallel_test_list = [
         "Func":    ticketlock_autotest,
         "Report":  None,
     },
+    {
+        "Name":    "MCSlock autotest",
+        "Command": "mcslock_autotest",
+        "Func":    mcslock_autotest,
+        "Report":  None,
+    },
     {
         "Name":    "Byte order autotest",
         "Command": "byteorder_autotest",
index 31cc0f5..26688b7 100644 (file)
@@ -164,6 +164,38 @@ def ticketlock_autotest(child, test_name):
 
     return 0, "Success"
 
+def mcslock_autotest(child, test_name):
+    i = 0
+    ir = 0
+    child.sendline(test_name)
+    while True:
+        index = child.expect(["Test OK",
+                              "Test Failed",
+                              "lcore ([0-9]*) state: ([0-1])"
+                              "MCS lock taken on core ([0-9]*)",
+                              "MCS lock released on core ([0-9]*)",
+                              pexpect.TIMEOUT], timeout=5)
+        # ok
+        if index == 0:
+            break
+
+        # message, check ordering
+        elif index == 2:
+            if int(child.match.groups()[0]) < i:
+                return -1, "Fail [Bad order]"
+            i = int(child.match.groups()[0])
+        elif index == 3:
+            if int(child.match.groups()[0]) < ir:
+                return -1, "Fail [Bad order]"
+            ir = int(child.match.groups()[0])
+
+        # fail
+        elif index == 4:
+            return -1, "Fail [Timeout]"
+        elif index == 1:
+            return -1, "Fail"
+
+    return 0, "Success"
 
 def logs_autotest(child, test_name):
     child.sendline(test_name)
index 6dffae8..1834897 100644 (file)
@@ -80,6 +80,7 @@ test_sources = files('commands.c',
        'test_memzone.c',
        'test_meter.c',
        'test_metrics.c',
+       'test_mcslock.c',
        'test_mp_secondary.c',
        'test_pdump.c',
        'test_per_lcore.c',
@@ -185,6 +186,7 @@ fast_test_names = [
         'lpm6_autotest',
         'malloc_autotest',
         'mbuf_autotest',
+        'mcslock_autotest',
         'memcpy_autotest',
         'memory_autotest',
         'mempool_autotest',
diff --git a/app/test/test_mcslock.c b/app/test/test_mcslock.c
new file mode 100644 (file)
index 0000000..ed384b8
--- /dev/null
@@ -0,0 +1,250 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2019 Arm Limited
+ */
+
+#include <stdio.h>
+#include <stdint.h>
+#include <inttypes.h>
+#include <string.h>
+#include <unistd.h>
+#include <sys/queue.h>
+
+#include <rte_common.h>
+#include <rte_memory.h>
+#include <rte_per_lcore.h>
+#include <rte_launch.h>
+#include <rte_eal.h>
+#include <rte_lcore.h>
+#include <rte_cycles.h>
+#include <rte_mcslock.h>
+#include <rte_atomic.h>
+
+#include "test.h"
+
+/*
+ * RTE MCS lock test
+ * =================
+ *
+ * These tests are derived from spin lock test cases.
+ *
+ * - The functional test takes all of these locks and launches the
+ *   ''test_mcslock_per_core()'' function on each core (except the master).
+ *
+ *   - The function takes the global lock, display something, then releases
+ *     the global lock on each core.
+ *
+ * - A load test is carried out, with all cores attempting to lock a single
+ *   lock multiple times.
+ */
+
+RTE_DEFINE_PER_LCORE(rte_mcslock_t, _ml_me);
+RTE_DEFINE_PER_LCORE(rte_mcslock_t, _ml_try_me);
+RTE_DEFINE_PER_LCORE(rte_mcslock_t, _ml_perf_me);
+
+rte_mcslock_t *p_ml;
+rte_mcslock_t *p_ml_try;
+rte_mcslock_t *p_ml_perf;
+
+static unsigned int count;
+
+static rte_atomic32_t synchro;
+
+static int
+test_mcslock_per_core(__attribute__((unused)) void *arg)
+{
+       /* Per core me node. */
+       rte_mcslock_t ml_me = RTE_PER_LCORE(_ml_me);
+
+       rte_mcslock_lock(&p_ml, &ml_me);
+       printf("MCS lock taken on core %u\n", rte_lcore_id());
+       rte_mcslock_unlock(&p_ml, &ml_me);
+       printf("MCS lock released on core %u\n", rte_lcore_id());
+
+       return 0;
+}
+
+static uint64_t time_count[RTE_MAX_LCORE] = {0};
+
+#define MAX_LOOP 1000000
+
+static int
+load_loop_fn(void *func_param)
+{
+       uint64_t time_diff = 0, begin;
+       uint64_t hz = rte_get_timer_hz();
+       volatile uint64_t lcount = 0;
+       const int use_lock = *(int *)func_param;
+       const unsigned int lcore = rte_lcore_id();
+
+       /**< Per core me node. */
+       rte_mcslock_t ml_perf_me = RTE_PER_LCORE(_ml_perf_me);
+
+       /* wait synchro */
+       while (rte_atomic32_read(&synchro) == 0)
+               ;
+
+       begin = rte_get_timer_cycles();
+       while (lcount < MAX_LOOP) {
+               if (use_lock)
+                       rte_mcslock_lock(&p_ml_perf, &ml_perf_me);
+
+               lcount++;
+               if (use_lock)
+                       rte_mcslock_unlock(&p_ml_perf, &ml_perf_me);
+       }
+       time_diff = rte_get_timer_cycles() - begin;
+       time_count[lcore] = time_diff * 1000000 / hz;
+       return 0;
+}
+
+static int
+test_mcslock_perf(void)
+{
+       unsigned int i;
+       uint64_t total = 0;
+       int lock = 0;
+       const unsigned int lcore = rte_lcore_id();
+
+       printf("\nTest with no lock on single core...\n");
+       rte_atomic32_set(&synchro, 1);
+       load_loop_fn(&lock);
+       printf("Core [%u] Cost Time = %"PRIu64" us\n",
+                       lcore, time_count[lcore]);
+       memset(time_count, 0, sizeof(time_count));
+
+       printf("\nTest with lock on single core...\n");
+       lock = 1;
+       rte_atomic32_set(&synchro, 1);
+       load_loop_fn(&lock);
+       printf("Core [%u] Cost Time = %"PRIu64" us\n",
+                       lcore, time_count[lcore]);
+       memset(time_count, 0, sizeof(time_count));
+
+       printf("\nTest with lock on %u cores...\n", (rte_lcore_count()));
+
+       rte_atomic32_set(&synchro, 0);
+       rte_eal_mp_remote_launch(load_loop_fn, &lock, SKIP_MASTER);
+
+       /* start synchro and launch test on master */
+       rte_atomic32_set(&synchro, 1);
+       load_loop_fn(&lock);
+
+       rte_eal_mp_wait_lcore();
+
+       RTE_LCORE_FOREACH(i) {
+               printf("Core [%u] Cost Time = %"PRIu64" us\n",
+                               i, time_count[i]);
+               total += time_count[i];
+       }
+
+       printf("Total Cost Time = %"PRIu64" us\n", total);
+
+       return 0;
+}
+
+/*
+ * Use rte_mcslock_trylock() to trylock a mcs lock object,
+ * If it could not lock the object successfully, it would
+ * return immediately.
+ */
+static int
+test_mcslock_try(__attribute__((unused)) void *arg)
+{
+       /**< Per core me node. */
+       rte_mcslock_t ml_me     = RTE_PER_LCORE(_ml_me);
+       rte_mcslock_t ml_try_me = RTE_PER_LCORE(_ml_try_me);
+
+       /* Locked ml_try in the master lcore, so it should fail
+        * when trying to lock it in the slave lcore.
+        */
+       if (rte_mcslock_trylock(&p_ml_try, &ml_try_me) == 0) {
+               rte_mcslock_lock(&p_ml, &ml_me);
+               count++;
+               rte_mcslock_unlock(&p_ml, &ml_me);
+       }
+
+       return 0;
+}
+
+
+/*
+ * Test rte_eal_get_lcore_state() in addition to mcs locks
+ * as we have "waiting" then "running" lcores.
+ */
+static int
+test_mcslock(void)
+{
+       int ret = 0;
+       int i;
+
+       /* Define per core me node. */
+       rte_mcslock_t ml_me     = RTE_PER_LCORE(_ml_me);
+       rte_mcslock_t ml_try_me = RTE_PER_LCORE(_ml_try_me);
+
+       /*
+        * Test mcs lock & unlock on each core
+        */
+
+       /* slave cores should be waiting: print it */
+       RTE_LCORE_FOREACH_SLAVE(i) {
+               printf("lcore %d state: %d\n", i,
+                               (int) rte_eal_get_lcore_state(i));
+       }
+
+       rte_mcslock_lock(&p_ml, &ml_me);
+
+       RTE_LCORE_FOREACH_SLAVE(i) {
+               rte_eal_remote_launch(test_mcslock_per_core, NULL, i);
+       }
+
+       /* slave cores should be busy: print it */
+       RTE_LCORE_FOREACH_SLAVE(i) {
+               printf("lcore %d state: %d\n", i,
+                               (int) rte_eal_get_lcore_state(i));
+       }
+
+       rte_mcslock_unlock(&p_ml, &ml_me);
+
+       rte_eal_mp_wait_lcore();
+
+       /*
+        * Test if it could return immediately from try-locking a locked object.
+        * Here it will lock the mcs lock object first, then launch all the
+        * slave lcores to trylock the same mcs lock object.
+        * All the slave lcores should give up try-locking a locked object and
+        * return immediately, and then increase the "count" initialized with
+        * zero by one per times.
+        * We can check if the "count" is finally equal to the number of all
+        * slave lcores to see if the behavior of try-locking a locked
+        * mcslock object is correct.
+        */
+       if (rte_mcslock_trylock(&p_ml_try, &ml_try_me) == 0)
+               return -1;
+
+       count = 0;
+       RTE_LCORE_FOREACH_SLAVE(i) {
+               rte_eal_remote_launch(test_mcslock_try, NULL, i);
+       }
+       rte_mcslock_unlock(&p_ml_try, &ml_try_me);
+       rte_eal_mp_wait_lcore();
+
+       /* Test is_locked API */
+       if (rte_mcslock_is_locked(p_ml)) {
+               printf("mcslock is locked but it should not be\n");
+               return -1;
+       }
+
+       /* Counting the locked times in each core */
+       rte_mcslock_lock(&p_ml, &ml_me);
+       if (count != (rte_lcore_count() - 1))
+               ret = -1;
+       rte_mcslock_unlock(&p_ml, &ml_me);
+
+       /* mcs lock perf test */
+       if (test_mcslock_perf() < 0)
+               return -1;
+
+       return ret;
+}
+
+REGISTER_TEST_COMMAND(mcslock_autotest, test_mcslock);