Update to support multiple nodes

Probably should squash stuff in with this. Signed-off-by: Ben Widawsky <ben.widawsky@intel.com> Signed-off-by: Feng Tang <feng.tang@intel.com>
filbranden · Dec 9, 2021 · 0c844d8 · 0c844d8
1 parent 34a278a
commit 0c844d8
Show file tree

Hide file tree

Showing 4 changed files with 45 additions and 14 deletions.
diff --git a/libnuma.c b/libnuma.c
@@ -1806,7 +1806,9 @@ static struct bitmask *__numa_preferred(void)
 	numa_bitmask_clearall(bmp);
 	getpol(&policy, bmp);
 
-	if (policy != MPOL_PREFERRED && policy != MPOL_BIND)
+	if (policy != MPOL_PREFERRED &&
+			policy != MPOL_PREFERRED_MANY &&
+			policy != MPOL_BIND)
 		return bmp;
 
 	if (numa_bitmask_weight(bmp) > 1)

diff --git a/numactl.8 b/numactl.8
@@ -31,6 +31,8 @@ numactl \- Control NUMA policy for processes or shared memory
 ] [
 .B \-\-preferred node 
 ] [
+.B \-\-preferred-many nodes
+] [
 .B \-\-membind nodes
 ] [ 
 .B \-\-cpunodebind nodes
@@ -176,6 +178,14 @@ This should only be used with
 .I \-\-membind, \-m
 only, otherwise ignored.
 .TP
+.B \-\-preferred-many=node
+Preferably allocate memory on
+.I nodes,
+but if memory cannot be allocated there fall back to other nodes.
+This option takes a mask of preferred nodes where the closest node to local is
+considered most preferred.
+Relative notation may be used.
+.TP
 .B \-\-show, \-s
 Show NUMA policy settings of the current process. 
 .TP
@@ -195,6 +205,7 @@ above (
 .I \-\-interleave, 
 .I \-\-localalloc, 
 .I \-\-preferred,
+.I \-\-preferred-many,
 .I \-\-membind
 ).
 .TP
@@ -297,6 +308,9 @@ also in the same node.
 numactl \-\-preferred=1 numactl \-\-show
 Set preferred node 1 and show the resulting state.
 
+numactl \-\-preferred-many=0x3 numactl \-\-show
+Set preferred nodes 1 and 2, and show the resulting state.
+
 numactl --interleave=all --shm /tmp/shmkey 
 Interleave all of the sysv shared memory region specified by
 /tmp/shmkey over all nodes.

diff --git a/numactl.c b/numactl.c
@@ -39,6 +39,7 @@ struct option opts[] = {
 	{"all", 0, 0, 'a'},
 	{"interleave", 1, 0, 'i' },
 	{"preferred", 1, 0, 'p' },
+	{"preferred-many", 1, 0, 'P' },
 	{"cpubind", 1, 0, 'c' },
 	{"cpunodebind", 1, 0, 'N' },
 	{"physcpubind", 1, 0, 'C' },
@@ -67,8 +68,9 @@ void usage(void)
 {
 	fprintf(stderr,
 		"usage: numactl [--all | -a] [--balancing | -b] [--interleave= | -i <nodes>]\n"
-		"               [--preferred= | -p <node>] [--physcpubind= | -C <cpus>]\n"
-		"               [--cpunodebind= | -N <nodes>] [--membind= | -m <nodes>]\n"
+		"		[--preferred= | -p <node>] [--preferred-many= | -P <nodes>]\n"
+		"               [--physcpubind= | -C <cpus>] [--cpunodebind= | -N <nodes>]\n"
+		"               [--membind= | -m <nodes>] [--localalloc | -l] command args ...\n"
 		"               [--localalloc | -l] command args ...\n"
 		"       numactl [--show | -s]\n"
 		"       numactl [--hardware | -H]\n"
@@ -132,8 +134,7 @@ void show_physcpubind(void)
 
 void show(void)
 {
-	unsigned long prefnode;
-	struct bitmask *membind, *interleave, *cpubind;
+	struct bitmask *membind, *interleave, *cpubind, *preferred;
 	unsigned long cur;
 	int policy;
 
@@ -145,7 +146,7 @@ void show(void)
 
 	cpubind = numa_get_run_node_mask();
 
-	prefnode = numa_preferred();
+	preferred = numa_preferred_many();
 	interleave = numa_get_interleave_mask();
 	membind = numa_get_membind();
 	cur = numa_get_interleave_node();
@@ -159,8 +160,10 @@ void show(void)
 	printf("preferred node: ");
 	switch (policy) {
 	case MPOL_PREFERRED:
-		assert(prefnode != -1);
-		printf("%ld\n", prefnode);
+		if (numa_bitmask_weight(preferred))
+			printf("%d\n", find_first(preferred));
+		else
+			printf("%d\n", 0);
 		break;
 	case MPOL_DEFAULT:
 		printf("current\n");
@@ -171,6 +174,9 @@ void show(void)
 	case MPOL_BIND:
 		printf("%d\n", find_first(membind));
 		break;
+	case MPOL_PREFERRED_MANY:
+		printf("%ld (preferred-many)\n",cur);
+		break;
 	}
 	if (policy == MPOL_INTERLEAVE) {
 		printmask("interleavemask", interleave);
@@ -180,6 +186,7 @@ void show(void)
 	printmask("cpubind", cpubind);  // for compatibility
 	printmask("nodebind", cpubind);
 	printmask("membind", membind);
+	printmask("preferred", preferred);
 }
 
 char *fmt_mem(unsigned long long mem, char *buf)
@@ -524,9 +531,11 @@ int main(int ac, char **av)
 			numa_set_bind_policy(0);
 			checkerror("setting membind");
 			break;
+		case 'P': /* --preferred-many */
+			if (!numa_has_preferred_many())
+				complain("preferred-many requested without kernel support");
 		case 'p': /* --preferred */
 			checknuma();
-			setpolicy(MPOL_PREFERRED);
 			if (parse_all)
 				mask = numactl_parse_nodestring(optarg, ALL);
 			else
@@ -535,16 +544,21 @@ int main(int ac, char **av)
 				printf ("<%s> is invalid\n", optarg);
 				usage();
 			}
-			if (numa_bitmask_weight(mask) != 1)
-				usage();
 			errno = 0;
 			did_node_cpu_parse = 1;
 			numa_set_bind_policy(0);
-			if (shmfd >= 0)
+			if (shmfd >= 0) {
 				numa_tonode_memory(shmptr, shmlen, node);
-			else
+			} else if (c == 'p') {
+				if (numa_bitmask_weight(mask) != 1)
+					usage();
+
+				setpolicy(MPOL_PREFERRED);
 				numa_set_preferred(find_first(mask));
-			numa_bitmask_free(mask);
+			} else {
+				setpolicy(MPOL_PREFERRED_MANY);
+				numa_set_preferred_many(mask);
+			}
 			checkerror("setting preferred node");
 			break;
 		case 'l': /* --local */

diff --git a/shm.c b/shm.c
@@ -315,6 +315,7 @@ void verify_shm(int policy, struct bitmask *nodes)
 			}
 			ilnode = interleave_next(ilnode, nodes2);
 			break;
+		case MPOL_PREFERRED_MANY:
 		case MPOL_PREFERRED:
 		case MPOL_BIND:
 			if (!numa_bitmask_isbitset(nodes2, node)) {