Create a cpuset mask for each NUMA domain that is available in the

kernel via the global cpuset_domain[] array. To export these to userland,
add a CPU_WHICH_DOMAIN level that can be used to fetch the mask for a
specific domain. Add a -d flag to cpuset(1) that can be used to fetch
the mask for a given domain.

Differential Revision:	https://reviews.freebsd.org/D1232
Submitted by:	jeff (kernel bits)
Reviewed by:	adrian, jeff

sys/kern/kern_cpuset.c

@@ -56,6 +56,10 @@ __FBSDID("$FreeBSD$");
 #include <sys/interrupt.h>
 
 #include <vm/uma.h>
+#include <vm/vm.h>
+#include <vm/vm_page.h>
+#include <vm/vm_param.h>
+#include <vm/vm_phys.h>
 
 #ifdef DDB
 #include <ddb/ddb.h>
@@ -113,6 +117,7 @@ SYSCTL_INT(_kern_sched, OID_AUTO, cpusetsize, CTLFLAG_RD,
     SYSCTL_NULL_INT_PTR, sizeof(cpuset_t), "sizeof(cpuset_t)");
 
 cpuset_t *cpuset_root;
+cpuset_t cpuset_domain[MAXMEMDOM];
 
 /*
  * Acquire a reference to a cpuset, all pointers must be tracked with refs.
@@ -457,6 +462,7 @@ cpuset_which(cpuwhich_t which, id_t id, struct proc **pp, struct thread **tdp,
 		return (0);
 	}
 	case CPU_WHICH_IRQ:
+	case CPU_WHICH_DOMAIN:
 		return (0);
 	default:
 		return (EINVAL);
@@ -810,7 +816,8 @@ out:
 
 
 /*
- * Creates the cpuset for thread0.  We make two sets:
+ * Creates system-wide cpusets and the cpuset for thread0 including two
+ * sets:
  * 
  * 0 - The root set which should represent all valid processors in the
  *     system.  It is initially created with a mask of all processors
@@ -856,6 +863,10 @@ cpuset_thread0(void)
 	 */
 	cpuset_unr = new_unrhdr(2, INT_MAX, NULL);
 
+	/* MD Code is responsible for initializing sets if vm_ndomains > 1. */
+	if (vm_ndomains == 1)
+		CPU_COPY(&all_cpus, &cpuset_domain[0]);
+
 	return (set);
 }
 
@@ -1010,6 +1021,7 @@ sys_cpuset_getid(struct thread *td, struct cpuset_getid_args *uap)
 	case CPU_WHICH_JAIL:
 		break;
 	case CPU_WHICH_IRQ:
+	case CPU_WHICH_DOMAIN:
 		return (EINVAL);
 	}
 	switch (uap->level) {
@@ -1073,6 +1085,7 @@ sys_cpuset_getaffinity(struct thread *td, struct cpuset_getaffinity_args *uap)
 		case CPU_WHICH_JAIL:
 			break;
 		case CPU_WHICH_IRQ:
+		case CPU_WHICH_DOMAIN:
 			error = EINVAL;
 			goto out;
 		}
@@ -1104,6 +1117,12 @@ sys_cpuset_getaffinity(struct thread *td, struct cpuset_getaffinity_args *uap)
 		case CPU_WHICH_IRQ:
 			error = intr_getaffinity(uap->id, mask);
 			break;
+		case CPU_WHICH_DOMAIN:
+			if (uap->id >= vm_ndomains)
+				error = ESRCH;
+			else
+				CPU_COPY(&cpuset_domain[uap->id], mask);
+			break;
 		}
 		break;
 	default:
@@ -1182,6 +1201,7 @@ sys_cpuset_setaffinity(struct thread *td, struct cpuset_setaffinity_args *uap)
 		case CPU_WHICH_JAIL:
 			break;
 		case CPU_WHICH_IRQ:
+		case CPU_WHICH_DOMAIN:
 			error = EINVAL;
 			goto out;
 		}

sys/sys/cpuset.h

@@ -76,6 +76,7 @@
 #define	CPU_WHICH_CPUSET	3	/* Specifies a set id. */
 #define	CPU_WHICH_IRQ		4	/* Specifies an irq #. */
 #define	CPU_WHICH_JAIL		5	/* Specifies a jail id. */
+#define	CPU_WHICH_DOMAIN	6	/* Specifies a NUMA domain id. */
 
 /*
  * Reserved cpuset identifiers.

sys/sys/smp.h

@@ -85,6 +85,7 @@ extern int mp_ncpus;
 extern volatile int smp_started;
 
 extern cpuset_t all_cpus;
+extern cpuset_t cpuset_domain[MAXMEMDOM]; 	/* CPUs in each NUMA domain. */
 
 /*
  * Macro allowing us to determine whether a CPU is absent at any given

sys/x86/acpica/srat.c

@@ -342,7 +342,7 @@ srat_walk_table(acpi_subtable_handler *handler, void *arg)
 }
 
 /*
- * Setup per-CPU ACPI IDs.
+ * Setup per-CPU domain IDs.
  */
 static void
 srat_set_cpus(void *dummy)
@@ -363,6 +363,7 @@ srat_set_cpus(void *dummy)
 			panic("SRAT: CPU with APIC ID %u is not known",
 			    pc->pc_apic_id);
 		pc->pc_domain = cpu->domain;
+		CPU_SET(i, &cpuset_domain[cpu->domain]);
 		if (bootverbose)
 			printf("SRAT: CPU %u has memory domain %d\n", i,
 			    cpu->domain);

usr.bin/cpuset/cpuset.1

@@ -46,12 +46,13 @@
 .Fl C
 .Fl p Ar pid
 .Nm
-.Op Fl cr
+.Op Fl c
 .Op Fl l Ar cpu-list
 .Op Fl j Ar jailid | Fl p Ar pid | Fl t Ar tid | Fl s Ar setid | Fl x Ar irq
 .Nm
-.Op Fl cgir
+.Fl g
-.Op Fl j Ar jailid | Fl p Ar pid | Fl t Ar tid | Fl s Ar setid | Fl x Ar irq
+.Op Fl cir
+.Op Fl d Ar domain | j Ar jailid | Fl p Ar pid | Fl t Ar tid | Fl s Ar setid | Fl x Ar irq
 .Sh DESCRIPTION
 The
 .Nm
@@ -62,7 +63,7 @@ about processor binding, sets, and available processors in the system.
 .Nm
 requires a target to modify or query.
 The target may be specified as a command, process id, thread id, a
-cpuset id, an irq or a jail id.
+cpuset id, an irq, a jail id, or a NUMA domain.
 Using
 .Fl g
 the target's set id or mask may be queried.
@@ -108,6 +109,8 @@ Create a new cpuset and assign the target process to that set.
 .It Fl c
 The requested operation should reference the cpuset available via the
 target specifier.
+.It Fl d Ar domain
+Specifies a NUMA domain id as the target of the operation.
 .It Fl g
 Causes
 .Nm

usr.bin/cpuset/cpuset.c

@@ -48,6 +48,7 @@ __FBSDID("$FreeBSD$");
 
 static int Cflag;
 static int cflag;
+static int dflag;
 static int gflag;
 static int iflag;
 static int jflag;
@@ -161,7 +162,8 @@ printset(cpuset_t *mask)
 	printf("\n");
 }
 
-static const char *whichnames[] = { NULL, "tid", "pid", "cpuset", "irq", "jail" };
+static const char *whichnames[] = { NULL, "tid", "pid", "cpuset", "irq", "jail",
+				    "domain" };
 static const char *levelnames[] = { NULL, " root", " cpuset", "" };
 
 static void
@@ -206,17 +208,20 @@ main(int argc, char *argv[])
 	level = CPU_LEVEL_WHICH;
 	which = CPU_WHICH_PID;
 	id = pid = tid = setid = -1;
-	while ((ch = getopt(argc, argv, "Ccgij:l:p:rs:t:x:")) != -1) {
+	while ((ch = getopt(argc, argv, "Ccd:gij:l:p:rs:t:x:")) != -1) {
 		switch (ch) {
 		case 'C':
 			Cflag = 1;
 			break;
 		case 'c':
-			if (rflag)
-				usage();
 			cflag = 1;
 			level = CPU_LEVEL_CPUSET;
 			break;
+		case 'd':
+			dflag = 1;
+			which = CPU_WHICH_DOMAIN;
+			id = atoi(optarg);
+			break;
 		case 'g':
 			gflag = 1;
 			break;
@@ -238,8 +243,6 @@ main(int argc, char *argv[])
 			id = pid = atoi(optarg);
 			break;
 		case 'r':
-			if (cflag)
-				usage();
 			level = CPU_LEVEL_ROOT;
 			rflag = 1;
 			break;
@@ -268,7 +271,7 @@ main(int argc, char *argv[])
 		if (argc || Cflag || lflag)
 			usage();
 		/* Only one identity specifier. */
-		if (jflag + xflag + sflag + pflag + tflag > 1)
+		if (dflag + jflag + xflag + sflag + pflag + tflag > 1)
 			usage();
 		if (iflag)
 			printsetid();
@@ -276,13 +279,13 @@ main(int argc, char *argv[])
 			printaffinity();
 		exit(EXIT_SUCCESS);
 	}
-	if (iflag)
+	if (dflag || iflag || rflag)
 		usage();
 	/*
 	 * The user wants to run a command with a set and possibly cpumask.
 	 */
 	if (argc) {
-		if (Cflag | pflag | rflag | tflag | xflag | jflag)
+		if (Cflag || pflag || tflag || xflag || jflag)
 			usage();
 		if (sflag) {
 			if (cpuset_setid(CPU_WHICH_PID, -1, setid))
@@ -303,9 +306,9 @@ main(int argc, char *argv[])
 	/*
 	 * We're modifying something that presently exists.
 	 */
-	if (Cflag && (sflag || rflag || !pflag || tflag || xflag || jflag))
+	if (Cflag && (jflag || !pflag || sflag || tflag || xflag))
 		usage();
-	if (!lflag && (cflag || rflag))
+	if (!lflag && cflag)
 		usage();
 	if (!lflag && !(Cflag || sflag))
 		usage();
@@ -354,8 +357,9 @@ usage(void)
 	fprintf(stderr,
 	    "       cpuset [-c] [-l cpu-list] -C -p pid\n");
 	fprintf(stderr,
-	    "       cpuset [-cr] [-l cpu-list] [-j jailid | -p pid | -t tid | -s setid | -x irq]\n");
+	    "       cpuset [-c] [-l cpu-list] [-j jailid | -p pid | -t tid | -s setid | -x irq]\n");
 	fprintf(stderr,
-	    "       cpuset [-cgir] [-j jailid | -p pid | -t tid | -s setid | -x irq]\n");
+	    "       cpuset -g [-cir] [-d domain | -j jailid | -p pid | -t tid | -s setid |\n"
+	    "              -x irq]\n");
 	exit(1);
 }