forked from ~ljy/RK356X_SDK_RELEASE
blame | history | patch | commit | commitdiff
hc
2024-05-10 23fa18eaa71266feff7ba8d83022d9e1cc83c65a 
 kernel/tools/perf/bench/numa.c
....@@ -9,9 +9,6 @@
99 /* For the CLR_() macros */
1010 #include <pthread.h>
1111 
12
-#include "../perf.h"
13
-#include "../builtin.h"
14
-#include "../util/util.h"
1512 #include <subcmd/parse-options.h>
1613 #include "../util/cloexec.h"
1714 
....@@ -34,6 +31,8 @@
3431 #include <sys/types.h>
3532 #include <linux/kernel.h>
3633 #include <linux/time64.h>
34
+#include <linux/numa.h>
35
+#include <linux/zalloc.h>
3736 
3837 #include <numa.h>
3938 #include <numaif.h>
....@@ -138,12 +137,13 @@
138137 	u8			*data;
139138 
140139 	pthread_mutex_t		startup_mutex;
140
+	pthread_cond_t		startup_cond;
141141 	int			nr_tasks_started;
142142 
143
-	pthread_mutex_t		startup_done_mutex;
144
-
145143 	pthread_mutex_t		start_work_mutex;
144
+	pthread_cond_t		start_work_cond;
146145 	int			nr_tasks_working;
146
+	bool			start_work;
147147 
148148 	pthread_mutex_t		stop_work_mutex;
149149 	u64			bytes_done;
....@@ -248,17 +248,22 @@
248248  */
249249 static bool node_has_cpus(int node)
250250 {
251
-	struct bitmask *cpu = numa_allocate_cpumask();
252
-	unsigned int i;
251
+	struct bitmask *cpumask = numa_allocate_cpumask();
252
+	bool ret = false; /* fall back to nocpus */
253
+	int cpu;
253254 
254
-	if (cpu && !numa_node_to_cpus(node, cpu)) {
255
-		for (i = 0; i < cpu->size; i++) {
256
-			if (numa_bitmask_isbitset(cpu, i))
257
-				return true;
255
+	BUG_ON(!cpumask);
256
+	if (!numa_node_to_cpus(node, cpumask)) {
257
+		for (cpu = 0; cpu < (int)cpumask->size; cpu++) {
258
+			if (numa_bitmask_isbitset(cpumask, cpu)) {
259
+				ret = true;
260
+				break;
261
+			}
258262 		}
259263 	}
264
+	numa_free_cpumask(cpumask);
260265 
261
-	return false; /* lets fall back to nocpus safely */
266
+	return ret;
262267 }
263268 
264269 static cpu_set_t bind_to_cpu(int target_cpu)
....@@ -289,30 +294,29 @@
289294 
290295 static cpu_set_t bind_to_node(int target_node)
291296 {
292
-	int cpus_per_node = g->p.nr_cpus / nr_numa_nodes();
293297 	cpu_set_t orig_mask, mask;
294298 	int cpu;
295299 	int ret;
296
-
297
-	BUG_ON(cpus_per_node * nr_numa_nodes() != g->p.nr_cpus);
298
-	BUG_ON(!cpus_per_node);
299300 
300301 	ret = sched_getaffinity(0, sizeof(orig_mask), &orig_mask);
301302 	BUG_ON(ret);
302303 
303304 	CPU_ZERO(&mask);
304305 
305
-	if (target_node == -1) {
306
+	if (target_node == NUMA_NO_NODE) {
306307 		for (cpu = 0; cpu < g->p.nr_cpus; cpu++)
307308 			CPU_SET(cpu, &mask);
308309 	} else {
309
-		int cpu_start = (target_node + 0) * cpus_per_node;
310
-		int cpu_stop  = (target_node + 1) * cpus_per_node;
310
+		struct bitmask *cpumask = numa_allocate_cpumask();
311311 
312
-		BUG_ON(cpu_stop > g->p.nr_cpus);
313
-
314
-		for (cpu = cpu_start; cpu < cpu_stop; cpu++)
315
-			CPU_SET(cpu, &mask);
312
+		BUG_ON(!cpumask);
313
+		if (!numa_node_to_cpus(target_node, cpumask)) {
314
+			for (cpu = 0; cpu < (int)cpumask->size; cpu++) {
315
+				if (numa_bitmask_isbitset(cpumask, cpu))
316
+					CPU_SET(cpu, &mask);
317
+			}
318
+		}
319
+		numa_free_cpumask(cpumask);
316320 	}
317321 
318322 	ret = sched_setaffinity(0, sizeof(mask), &mask);
....@@ -343,7 +347,7 @@
343347 	unsigned long nodemask;
344348 	int ret;
345349 
346
-	if (node == -1)
350
+	if (node == NUMA_NO_NODE)
347351 		return;
348352 
349353 	BUG_ON(g->p.nr_nodes > (int)sizeof(nodemask)*8);
....@@ -478,6 +482,18 @@
478482 	pthread_mutexattr_init(&attr);
479483 	pthread_mutexattr_setpshared(&attr, PTHREAD_PROCESS_SHARED);
480484 	pthread_mutex_init(mutex, &attr);
485
+}
486
+
487
+/*
488
+ * Return a process-shared (global) condition variable:
489
+ */
490
+static void init_global_cond(pthread_cond_t *cond)
491
+{
492
+	pthread_condattr_t attr;
493
+
494
+	pthread_condattr_init(&attr);
495
+	pthread_condattr_setpshared(&attr, PTHREAD_PROCESS_SHARED);
496
+	pthread_cond_init(cond, &attr);
481497 }
482498 
483499 static int parse_cpu_list(const char *arg)
....@@ -730,8 +746,6 @@
730746 		return -1;
731747 
732748 	return parse_node_list(arg);
733
-
734
-	return 0;
735749 }
736750 
737751 #define BIT(x) (1ul << x)
....@@ -814,12 +828,12 @@
814828 			}
815829 		}
816830 	} else if (!g->p.data_backwards || (nr + loop) & 1) {
831
+		/* Process data forwards: */
817832 
818833 		d0 = data + off;
819834 		d  = data + off + 1;
820835 		d1 = data + words;
821836 
822
-		/* Process data forwards: */
823837 		for (;;) {
824838 			if (unlikely(d >= d1))
825839 				d = data;
....@@ -837,7 +851,6 @@
837851 		d  = data + off - 1;
838852 		d1 = data + words;
839853 
840
-		/* Process data forwards: */
841854 		for (;;) {
842855 			if (unlikely(d < data))
843856 				d = data + words-1;
....@@ -1136,15 +1149,18 @@
11361149 	if (g->p.serialize_startup) {
11371150 		pthread_mutex_lock(&g->startup_mutex);
11381151 		g->nr_tasks_started++;
1152
+		/* The last thread wakes the main process. */
1153
+		if (g->nr_tasks_started == g->p.nr_tasks)
1154
+			pthread_cond_signal(&g->startup_cond);
1155
+
11391156 		pthread_mutex_unlock(&g->startup_mutex);
11401157 
11411158 		/* Here we will wait for the main process to start us all at once: */
11421159 		pthread_mutex_lock(&g->start_work_mutex);
1160
+		g->start_work = false;
11431161 		g->nr_tasks_working++;
1144
-
1145
-		/* Last one wake the main process: */
1146
-		if (g->nr_tasks_working == g->p.nr_tasks)
1147
-			pthread_mutex_unlock(&g->startup_done_mutex);
1162
+		while (!g->start_work)
1163
+			pthread_cond_wait(&g->start_work_cond, &g->start_work_mutex);
11481164 
11491165 		pthread_mutex_unlock(&g->start_work_mutex);
11501166 	}
....@@ -1369,7 +1385,7 @@
13691385 		int cpu;
13701386 
13711387 		/* Allow all nodes by default: */
1372
-		td->bind_node = -1;
1388
+		td->bind_node = NUMA_NO_NODE;
13731389 
13741390 		/* Allow all CPUs by default: */
13751391 		CPU_ZERO(&td->bind_cpumask);
....@@ -1441,8 +1457,9 @@
14411457 
14421458 	/* Startup serialization: */
14431459 	init_global_mutex(&g->start_work_mutex);
1460
+	init_global_cond(&g->start_work_cond);
14441461 	init_global_mutex(&g->startup_mutex);
1445
-	init_global_mutex(&g->startup_done_mutex);
1462
+	init_global_cond(&g->startup_cond);
14461463 	init_global_mutex(&g->stop_work_mutex);
14471464 
14481465 	init_thread_data();
....@@ -1502,9 +1519,6 @@
15021519 	pids = zalloc(g->p.nr_proc * sizeof(*pids));
15031520 	pid = -1;
15041521 
1505
-	/* All threads try to acquire it, this way we can wait for them to start up: */
1506
-	pthread_mutex_lock(&g->start_work_mutex);
1507
-
15081522 	if (g->p.serialize_startup) {
15091523 		tprintf(" #\n");
15101524 		tprintf(" # Startup synchronization: ..."); fflush(stdout);
....@@ -1526,22 +1540,29 @@
15261540 		pids[i] = pid;
15271541 
15281542 	}
1529
-	/* Wait for all the threads to start up: */
1530
-	while (g->nr_tasks_started != g->p.nr_tasks)
1531
-		usleep(USEC_PER_MSEC);
1532
-
1533
-	BUG_ON(g->nr_tasks_started != g->p.nr_tasks);
15341543 
15351544 	if (g->p.serialize_startup) {
1545
+		bool threads_ready = false;
15361546 		double startup_sec;
15371547 
1538
-		pthread_mutex_lock(&g->startup_done_mutex);
1548
+		/*
1549
+		 * Wait for all the threads to start up. The last thread will
1550
+		 * signal this process.
1551
+		 */
1552
+		pthread_mutex_lock(&g->startup_mutex);
1553
+		while (g->nr_tasks_started != g->p.nr_tasks)
1554
+			pthread_cond_wait(&g->startup_cond, &g->startup_mutex);
15391555 
1540
-		/* This will start all threads: */
1541
-		pthread_mutex_unlock(&g->start_work_mutex);
1556
+		pthread_mutex_unlock(&g->startup_mutex);
15421557 
1543
-		/* This mutex is locked - the last started thread will wake us: */
1544
-		pthread_mutex_lock(&g->startup_done_mutex);
1558
+		/* Wait for all threads to be at the start_work_cond. */
1559
+		while (!threads_ready) {
1560
+			pthread_mutex_lock(&g->start_work_mutex);
1561
+			threads_ready = (g->nr_tasks_working == g->p.nr_tasks);
1562
+			pthread_mutex_unlock(&g->start_work_mutex);
1563
+			if (!threads_ready)
1564
+				usleep(1);
1565
+		}
15451566 
15461567 		gettimeofday(&stop, NULL);
15471568 
....@@ -1555,7 +1576,11 @@
15551576 		tprintf(" #\n");
15561577 
15571578 		start = stop;
1558
-		pthread_mutex_unlock(&g->startup_done_mutex);
1579
+		/* Start all threads running. */
1580
+		pthread_mutex_lock(&g->start_work_mutex);
1581
+		g->start_work = true;
1582
+		pthread_mutex_unlock(&g->start_work_mutex);
1583
+		pthread_cond_broadcast(&g->start_work_cond);
15591584 	} else {
15601585 		gettimeofday(&start, NULL);
15611586 	}
....@@ -1631,7 +1656,7 @@
16311656 		"GB/sec,", "total-speed",	"GB/sec total speed");
16321657 
16331658 	if (g->p.show_details >= 2) {
1634
-		char tname[14 + 2 * 10 + 1];
1659
+		char tname[14 + 2 * 11 + 1];
16351660 		struct thread_data *td;
16361661 		for (p = 0; p < g->p.nr_proc; p++) {
16371662 			for (t = 0; t < g->p.nr_threads; t++) {
....@@ -1734,12 +1759,12 @@
17341759  */
17351760 static const char *tests[][MAX_ARGS] = {
17361761    /* Basic single-stream NUMA bandwidth measurements: */
1737
-   { "RAM-bw-local,",	  "mem",  "-p",  "1",  "-t",  "1", "-P", "1024",
1762
+   { "RAM-bw-local,",     "mem",  "-p",  "1",  "-t",  "1", "-P", "1024",
17381763 			  "-C" ,   "0", "-M",   "0", OPT_BW_RAM },
17391764    { "RAM-bw-local-NOTHP,",
17401765 			  "mem",  "-p",  "1",  "-t",  "1", "-P", "1024",
17411766 			  "-C" ,   "0", "-M",   "0", OPT_BW_RAM_NOTHP },
1742
-   { "RAM-bw-remote,",	  "mem",  "-p",  "1",  "-t",  "1", "-P", "1024",
1767
+   { "RAM-bw-remote,",    "mem",  "-p",  "1",  "-t",  "1", "-P", "1024",
17431768 			  "-C" ,   "0", "-M",   "1", OPT_BW_RAM },
17441769 
17451770    /* 2-stream NUMA bandwidth measurements: */
....@@ -1756,7 +1781,7 @@
17561781    { " 1x3-convergence,", "mem",  "-p",  "1", "-t",  "3", "-P",  "512", OPT_CONV },
17571782    { " 1x4-convergence,", "mem",  "-p",  "1", "-t",  "4", "-P",  "512", OPT_CONV },
17581783    { " 1x6-convergence,", "mem",  "-p",  "1", "-t",  "6", "-P", "1020", OPT_CONV },
1759
-   { " 2x3-convergence,", "mem",  "-p",  "3", "-t",  "3", "-P", "1020", OPT_CONV },
1784
+   { " 2x3-convergence,", "mem",  "-p",  "2", "-t",  "3", "-P", "1020", OPT_CONV },
17601785    { " 3x3-convergence,", "mem",  "-p",  "3", "-t",  "3", "-P", "1020", OPT_CONV },
17611786    { " 4x4-convergence,", "mem",  "-p",  "4", "-t",  "4", "-P",  "512", OPT_CONV },
17621787    { " 4x4-convergence-NOTHP,",
....@@ -1781,24 +1806,24 @@
17811806 			  "mem",  "-p",  "8", "-t",  "1", "-P", " 512", OPT_BW_NOTHP },
17821807    { "16x1-bw-process,",  "mem",  "-p", "16", "-t",  "1", "-P",  "256", OPT_BW },
17831808 
1784
-   { " 4x1-bw-thread,",	  "mem",  "-p",  "1", "-t",  "4", "-T",  "256", OPT_BW },
1785
-   { " 8x1-bw-thread,",	  "mem",  "-p",  "1", "-t",  "8", "-T",  "256", OPT_BW },
1786
-   { "16x1-bw-thread,",   "mem",  "-p",  "1", "-t", "16", "-T",  "128", OPT_BW },
1787
-   { "32x1-bw-thread,",   "mem",  "-p",  "1", "-t", "32", "-T",   "64", OPT_BW },
1809
+   { " 1x4-bw-thread,",   "mem",  "-p",  "1", "-t",  "4", "-T",  "256", OPT_BW },
1810
+   { " 1x8-bw-thread,",   "mem",  "-p",  "1", "-t",  "8", "-T",  "256", OPT_BW },
1811
+   { "1x16-bw-thread,",   "mem",  "-p",  "1", "-t", "16", "-T",  "128", OPT_BW },
1812
+   { "1x32-bw-thread,",   "mem",  "-p",  "1", "-t", "32", "-T",   "64", OPT_BW },
17881813 
1789
-   { " 2x3-bw-thread,",	  "mem",  "-p",  "2", "-t",  "3", "-P",  "512", OPT_BW },
1790
-   { " 4x4-bw-thread,",	  "mem",  "-p",  "4", "-t",  "4", "-P",  "512", OPT_BW },
1791
-   { " 4x6-bw-thread,",	  "mem",  "-p",  "4", "-t",  "6", "-P",  "512", OPT_BW },
1792
-   { " 4x8-bw-thread,",	  "mem",  "-p",  "4", "-t",  "8", "-P",  "512", OPT_BW },
1793
-   { " 4x8-bw-thread-NOTHP,",
1814
+   { " 2x3-bw-process,",  "mem",  "-p",  "2", "-t",  "3", "-P",  "512", OPT_BW },
1815
+   { " 4x4-bw-process,",  "mem",  "-p",  "4", "-t",  "4", "-P",  "512", OPT_BW },
1816
+   { " 4x6-bw-process,",  "mem",  "-p",  "4", "-t",  "6", "-P",  "512", OPT_BW },
1817
+   { " 4x8-bw-process,",  "mem",  "-p",  "4", "-t",  "8", "-P",  "512", OPT_BW },
1818
+   { " 4x8-bw-process-NOTHP,",
17941819 			  "mem",  "-p",  "4", "-t",  "8", "-P",  "512", OPT_BW_NOTHP },
1795
-   { " 3x3-bw-thread,",	  "mem",  "-p",  "3", "-t",  "3", "-P",  "512", OPT_BW },
1796
-   { " 5x5-bw-thread,",	  "mem",  "-p",  "5", "-t",  "5", "-P",  "512", OPT_BW },
1820
+   { " 3x3-bw-process,",  "mem",  "-p",  "3", "-t",  "3", "-P",  "512", OPT_BW },
1821
+   { " 5x5-bw-process,",  "mem",  "-p",  "5", "-t",  "5", "-P",  "512", OPT_BW },
17971822 
1798
-   { "2x16-bw-thread,",   "mem",  "-p",  "2", "-t", "16", "-P",  "512", OPT_BW },
1799
-   { "1x32-bw-thread,",   "mem",  "-p",  "1", "-t", "32", "-P", "2048", OPT_BW },
1823
+   { "2x16-bw-process,",  "mem",  "-p",  "2", "-t", "16", "-P",  "512", OPT_BW },
1824
+   { "1x32-bw-process,",  "mem",  "-p",  "1", "-t", "32", "-P", "2048", OPT_BW },
18001825 
1801
-   { "numa02-bw,",	  "mem",  "-p",  "1", "-t", "32", "-T",   "32", OPT_BW },
1826
+   { "numa02-bw,",        "mem",  "-p",  "1", "-t", "32", "-T",   "32", OPT_BW },
18021827    { "numa02-bw-NOTHP,",  "mem",  "-p",  "1", "-t", "32", "-T",   "32", OPT_BW_NOTHP },
18031828    { "numa01-bw-thread,", "mem",  "-p",  "2", "-t", "16", "-T",  "192", OPT_BW },
18041829    { "numa01-bw-thread-NOTHP,",