Move VM_NUMA_ALLOC and DEVICE_NUMA under the single global config option NUMA.

Sponsored by:	Netflix, Dell/EMC Isilon
Discussed with:	jhb

11 files changed, 58 insertions, 30 deletions

diff --git a/sys/amd64/conf/GENERIC b/sys/amd64/conf/GENERIC
index 548d9adf794f..1af7b7b7863b 100644
--- a/sys/amd64/conf/GENERIC
+++ b/sys/amd64/conf/GENERIC
@@ -99,7 +99,6 @@ options 	MALLOC_DEBUG_MAXZONES=8	# Separate malloc(9) zones
 
 # Make an SMP-capable kernel by default
 options 	SMP			# Symmetric MultiProcessor Kernel
-options 	DEVICE_NUMA		# I/O Device Affinity
 options 	EARLY_AP_STARTUP
 
 # CPU frequency control
diff --git a/sys/amd64/conf/MINIMAL b/sys/amd64/conf/MINIMAL
index 3a3e9c500bf3..010148ff2d86 100644
--- a/sys/amd64/conf/MINIMAL
+++ b/sys/amd64/conf/MINIMAL
@@ -93,7 +93,6 @@ options 	MALLOC_DEBUG_MAXZONES=8	# Separate malloc(9) zones
 
 # Make an SMP-capable kernel by default
 options 	SMP			# Symmetric MultiProcessor Kernel
-options 	DEVICE_NUMA		# I/O Device Affinity
 options 	EARLY_AP_STARTUP
 
 # CPU frequency control
diff --git a/sys/conf/NOTES b/sys/conf/NOTES
index f7253873828c..6c7386992f19 100644
--- a/sys/conf/NOTES
+++ b/sys/conf/NOTES
@@ -231,18 +231,14 @@ options		EARLY_AP_STARTUP
 # A default value should be already present, for every architecture.
 options 	MAXCPU=32
 
+# NUMA enables use of Non-Uniform Memory Access policies in various kernel
+# subsystems.
+options 	NUMA
+
 # MAXMEMDOM defines the maximum number of memory domains that can boot in the
 # system.  A default value should already be defined by every architecture.
 options 	MAXMEMDOM=2
 
-# VM_NUMA_ALLOC enables use of memory domain-aware allocation in the VM
-# system.
-options 	VM_NUMA_ALLOC
-
-# DEVICE_NUMA enables reporting of domain affinity of I/O devices via
-# bus_get_domain(), etc.
-options 	DEVICE_NUMA
-
 # ADAPTIVE_MUTEXES changes the behavior of blocking mutexes to spin
 # if the thread that currently owns the mutex is executing on another
 # CPU.  This behavior is enabled by default, so this option can be used
diff --git a/sys/conf/options b/sys/conf/options
index 9cb925487119..e5abb344230a 100644
--- a/sys/conf/options
+++ b/sys/conf/options
@@ -95,7 +95,6 @@ COMPAT_LINUXKPI	opt_compat.h
 COMPILING_LINT	opt_global.h
 CY_PCI_FASTINTR
 DEADLKRES	opt_watchdog.h
-DEVICE_NUMA
 EXT_RESOURCES	opt_global.h
 DIRECTIO
 FILEMON		opt_dontuse.h
@@ -605,7 +604,6 @@ VM_KMEM_SIZE		opt_vm.h
 VM_KMEM_SIZE_SCALE	opt_vm.h
 VM_KMEM_SIZE_MAX	opt_vm.h
 VM_NRESERVLEVEL		opt_vm.h
-VM_NUMA_ALLOC		opt_vm.h
 VM_LEVEL_0_ORDER	opt_vm.h
 NO_SWAPPING		opt_vm.h
 MALLOC_MAKE_FAILURES	opt_vm.h
@@ -621,6 +619,7 @@ DEBUG_REDZONE		opt_vm.h
 # Standard SMP options
 EARLY_AP_STARTUP	opt_global.h
 SMP			opt_global.h
+NUMA			opt_global.h
 
 # Size of the kernel message buffer
 MSGBUF_SIZE		opt_msgbuf.h
diff --git a/sys/dev/acpica/acpi.c b/sys/dev/acpica/acpi.c
index c6559579f3a9..be05a76643c8 100644
--- a/sys/dev/acpica/acpi.c
+++ b/sys/dev/acpica/acpi.c
@@ -31,7 +31,6 @@
 __FBSDID("$FreeBSD$");
 
 #include "opt_acpi.h"
-#include "opt_device_numa.h"
 
 #include <sys/param.h>
 #include <sys/kernel.h>
@@ -1089,7 +1088,7 @@ acpi_hint_device_unit(device_t acdev, device_t child, const char *name,
 static int
 acpi_parse_pxm(device_t dev)
 {
-#ifdef DEVICE_NUMA
+#ifdef NUMA
 	ACPI_HANDLE handle;
 	ACPI_STATUS status;
 	int pxm;
diff --git a/sys/vm/uma_core.c b/sys/vm/uma_core.c
index 86b6030b1935..c2d6882f6418 100644
--- a/sys/vm/uma_core.c
+++ b/sys/vm/uma_core.c
@@ -2616,7 +2616,7 @@ zone_import(uma_zone_t zone, void **bucket, int max, int domain, int flags)
 			bucket[i++] = slab_alloc_item(keg, slab);
 			if (keg->uk_free <= keg->uk_reserve)
 				break;
-#if MAXMEMDOM > 1
+#ifdef NUMA
 			/*
 			 * If the zone is striped we pick a new slab for every
 			 * N allocations.  Eliminating this conditional will
diff --git a/sys/vm/vm_domainset.c b/sys/vm/vm_domainset.c
index 34a111e2c033..deb2f5162b38 100644
--- a/sys/vm/vm_domainset.c
+++ b/sys/vm/vm_domainset.c
@@ -49,6 +49,7 @@ __FBSDID("$FreeBSD$");
 #include <vm/vm_page.h>
 #include <vm/vm_phys.h>
 
+#ifdef NUMA
 /*
  * Iterators are written such that the first nowait pass has as short a
  * codepath as possible to eliminate bloat from the allocator.  It is
@@ -241,3 +242,36 @@ vm_domainset_iter_malloc(struct vm_domainset_iter *di, int *domain, int *flags)
 
 	return (0);
 }
+
+#else /* !NUMA */
+int
+vm_domainset_iter_page(struct vm_domainset_iter *di, int *domain, int *flags)
+{
+
+	return (EJUSTRETURN);
+}
+
+void
+vm_domainset_iter_page_init(struct vm_domainset_iter *di,
+            struct vm_object *obj, int *domain, int *flags)
+{
+
+	*domain = 0;
+}
+
+int
+vm_domainset_iter_malloc(struct vm_domainset_iter *di, int *domain, int *flags)
+{
+
+	return (EJUSTRETURN);
+}
+
+void
+vm_domainset_iter_malloc_init(struct vm_domainset_iter *di,
+            struct vm_object *obj, int *domain, int *flags)
+{
+
+	*domain = 0;
+}
+
+#endif
diff --git a/sys/vm/vm_pageout.c b/sys/vm/vm_pageout.c
index 543e5b144f57..51c7af7e1146 100644
--- a/sys/vm/vm_pageout.c
+++ b/sys/vm/vm_pageout.c
@@ -1919,16 +1919,13 @@ static void
 vm_pageout(void)
 {
 	int error;
-#ifdef VM_NUMA_ALLOC
 	int i;
-#endif
 
 	swap_pager_swap_init();
 	error = kthread_add(vm_pageout_laundry_worker, NULL, curproc, NULL,
 	    0, 0, "laundry: dom0");
 	if (error != 0)
 		panic("starting laundry for domain 0, error %d", error);
-#ifdef VM_NUMA_ALLOC
 	for (i = 1; i < vm_ndomains; i++) {
 		error = kthread_add(vm_pageout_worker, (void *)(uintptr_t)i,
 		    curproc, NULL, 0, 0, "dom%d", i);
@@ -1937,7 +1934,6 @@ vm_pageout(void)
 			    i, error);
 		}
 	}
-#endif
 	error = kthread_add(uma_reclaim_worker, NULL, curproc, NULL,
 	    0, 0, "uma");
 	if (error != 0)
diff --git a/sys/vm/vm_phys.c b/sys/vm/vm_phys.c
index 3da8c05b481a..3c0f42353c1a 100644
--- a/sys/vm/vm_phys.c
+++ b/sys/vm/vm_phys.c
@@ -71,7 +71,7 @@ __FBSDID("$FreeBSD$");
 _Static_assert(sizeof(long) * NBBY >= VM_PHYSSEG_MAX,
     "Too many physsegs.");
 
-#ifdef VM_NUMA_ALLOC
+#ifdef NUMA
 struct mem_affinity *mem_affinity;
 int *mem_locality;
 #endif
@@ -140,7 +140,7 @@ static int sysctl_vm_phys_segs(SYSCTL_HANDLER_ARGS);
 SYSCTL_OID(_vm, OID_AUTO, phys_segs, CTLTYPE_STRING | CTLFLAG_RD,
     NULL, 0, sysctl_vm_phys_segs, "A", "Phys Seg Info");
 
-#ifdef VM_NUMA_ALLOC
+#ifdef NUMA
 static int sysctl_vm_phys_locality(SYSCTL_HANDLER_ARGS);
 SYSCTL_OID(_vm, OID_AUTO, phys_locality, CTLTYPE_STRING | CTLFLAG_RD,
     NULL, 0, sysctl_vm_phys_locality, "A", "Phys Locality Info");
@@ -201,7 +201,7 @@ vm_phys_fictitious_cmp(struct vm_phys_fictitious_seg *p1,
 int
 vm_phys_domain_match(int prefer, vm_paddr_t low, vm_paddr_t high)
 {
-#ifdef VM_NUMA_ALLOC
+#ifdef NUMA
 	domainset_t mask;
 	int i;
 
@@ -306,7 +306,7 @@ int
 vm_phys_mem_affinity(int f, int t)
 {
 
-#ifdef VM_NUMA_ALLOC
+#ifdef NUMA
 	if (mem_locality == NULL)
 		return (-1);
 	if (f >= vm_ndomains || t >= vm_ndomains)
@@ -317,7 +317,7 @@ vm_phys_mem_affinity(int f, int t)
 #endif
 }
 
-#ifdef VM_NUMA_ALLOC
+#ifdef NUMA
 /*
  * Outputs the VM locality table.
  */
@@ -393,7 +393,7 @@ _vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end, int domain)
 static void
 vm_phys_create_seg(vm_paddr_t start, vm_paddr_t end)
 {
-#ifdef VM_NUMA_ALLOC
+#ifdef NUMA
 	int i;
 
 	if (mem_affinity == NULL) {
diff --git a/sys/vm/vm_phys.h b/sys/vm/vm_phys.h
index 9f14abaa73d2..b6a1402d4994 100644
--- a/sys/vm/vm_phys.h
+++ b/sys/vm/vm_phys.h
@@ -48,6 +48,11 @@ struct mem_affinity {
 	vm_paddr_t end;
 	int domain;
 };
+#ifdef NUMA
+extern struct mem_affinity *mem_affinity;
+extern int *mem_locality;
+#endif
+extern int vm_ndomains;
 
 struct vm_freelist {
 	struct pglist pl;
@@ -62,9 +67,6 @@ struct vm_phys_seg {
 	struct vm_freelist (*free_queues)[VM_NFREEPOOL][VM_NFREEORDER];
 };
 
-extern struct mem_affinity *mem_affinity;
-extern int *mem_locality;
-extern int vm_ndomains;
 extern struct vm_phys_seg vm_phys_segs[];
 extern int vm_phys_nsegs;
 
@@ -101,6 +103,7 @@ int vm_phys_mem_affinity(int f, int t);
 static inline int
 vm_phys_domidx(vm_page_t m)
 {
+#ifdef NUMA
 	int domn, segind;
 
 	/* XXXKIB try to assert that the page is managed */
@@ -109,6 +112,9 @@ vm_phys_domidx(vm_page_t m)
 	domn = vm_phys_segs[segind].domain;
 	KASSERT(domn < vm_ndomains, ("domain %d m %p", domn, m));
 	return (domn);
+#else
+	return (0);
+#endif
 }
 
 /*
diff --git a/sys/x86/acpica/srat.c b/sys/x86/acpica/srat.c
index df568cafcb26..d51189eb7174 100644
--- a/sys/x86/acpica/srat.c
+++ b/sys/x86/acpica/srat.c
@@ -153,7 +153,7 @@ parse_slit(void)
 	acpi_unmap_table(slit);
 	slit = NULL;
 
-#ifdef VM_NUMA_ALLOC
+#ifdef NUMA
 	/* Tell the VM about it! */
 	mem_locality = vm_locality_table;
 #endif
@@ -469,7 +469,7 @@ parse_srat(void)
 		return (-1);
 	}
 
-#ifdef VM_NUMA_ALLOC
+#ifdef NUMA
 	/* Point vm_phys at our memory affinity table. */
 	vm_ndomains = ndomain;
 	mem_affinity = mem_info;