srat, x86: add support for nodes spanning other nodes

For example, If the physical address layout on a two node system with 8 GB memory is something like: node 0: 0-2GB, 4-6GB node 1: 2-4GB, 6-8GB Current kernels fail to boot/detect this NUMA topology. ACPI SRAT tables can expose such a topology which needs to be supported. Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

srat, x86: add support for nodes spanning other nodes
For example, If the physical address layout on a two node system with 8 GB memory is something like: node 0: 0-2GB, 4-6GB node 1: 2-4GB, 6-8GB Current kernels fail to boot/detect this NUMA topology. ACPI SRAT tables can expose such a topology which needs to be supported. Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Suresh Siddha · Ingo Molnar
1 parent 8705a49c35
Showing 5 changed files with 44 additions and 18 deletions Side-by-side Diff
arch/x86/Kconfig
arch/x86/mm/k8topology_64.c
arch/x86/mm/numa_64.c
arch/x86/mm/srat_64.c
include/asm-x86/numa_64.h
@@ -903,6 +903,15 @@
 	help
 	  Enable ACPI SRAT based node topology detection.
  
+# Some NUMA nodes have memory ranges that span
+# other nodes.  Even though a pfn is valid and
+# between a node's start and end pfns, it may not
+# reside on that node.  See memmap_init_zone()
+# for details.
+config NODES_SPAN_OTHER_NODES
+	def_bool y
+	depends on X86_64_ACPI_NUMA
+
 config NUMA_EMU
 	bool "NUMA emulation"
 	depends on X86_64 && NUMA
@@ -164,7 +164,7 @@
 	if (!found)
 		return -1;
  
-	memnode_shift = compute_hash_shift(nodes, 8);
+	memnode_shift = compute_hash_shift(nodes, 8, NULL);
 	if (memnode_shift < 0) {
 		printk(KERN_ERR "No NUMA node hash function found. Contact maintainer\n");
 		return -1;
@@ -60,7 +60,7 @@
  * -1 if node overlap or lost ram (shift too big)
  */
 static int __init populate_memnodemap(const struct bootnode *nodes,
-				      int numnodes, int shift)
+				      int numnodes, int shift, int *nodeids)
 {
 	unsigned long addr, end;
 	int i, res = -1;
@@ -76,7 +76,12 @@
 		do {
 			if (memnodemap[addr >> shift] != NUMA_NO_NODE)
 				return -1;
-			memnodemap[addr >> shift] = i;
+
+			if (!nodeids)
+				memnodemap[addr >> shift] = i;
+			else
+				memnodemap[addr >> shift] = nodeids[i];
+
 			addr += (1UL << shift);
 		} while (addr < end);
 		res = 1;
@@ -139,7 +144,8 @@
 	return i;
 }
  
-int __init compute_hash_shift(struct bootnode *nodes, int numnodes)
+int __init compute_hash_shift(struct bootnode *nodes, int numnodes,
+			      int *nodeids)
 {
 	int shift;
  
@@ -149,7 +155,7 @@
 	printk(KERN_DEBUG "NUMA: Using %d for the hash shift.\n",
 		shift);
  
-	if (populate_memnodemap(nodes, numnodes, shift) != 1) {
+	if (populate_memnodemap(nodes, numnodes, shift, nodeids) != 1) {
 		printk(KERN_INFO "Your memory is not aligned you need to "
 		       "rebuild your kernel with a bigger NODEMAPSIZE "
 		       "shift=%d\n", shift);
@@ -462,7 +468,7 @@
 		}
 	}
 out:
-	memnode_shift = compute_hash_shift(nodes, num_nodes);
+	memnode_shift = compute_hash_shift(nodes, num_nodes, NULL);
 	if (memnode_shift < 0) {
 		memnode_shift = 0;
 		printk(KERN_ERR "No NUMA hash function found.  NUMA emulation "
@@ -32,6 +32,10 @@
 static int found_add_area __initdata;
 int hotadd_percent __initdata = 0;
  
+static int num_node_memblks __initdata;
+static struct bootnode node_memblk_range[NR_NODE_MEMBLKS] __initdata;
+static int memblk_nodeid[NR_NODE_MEMBLKS] __initdata;
+
 /* Too small nodes confuse the VM badly. Usually they result
    from BIOS bugs. */
 #define NODE_MIN_SIZE (4*1024*1024)
  
  
  
@@ -41,17 +45,17 @@
 	return acpi_map_pxm_to_node(pxm);
 }
  
-static __init int conflicting_nodes(unsigned long start, unsigned long end)
+static __init int conflicting_memblks(unsigned long start, unsigned long end)
 {
 	int i;
-	for_each_node_mask(i, nodes_parsed) {
-		struct bootnode *nd = &nodes[i];
+	for (i = 0; i < num_node_memblks; i++) {
+		struct bootnode *nd = &node_memblk_range[i];
 		if (nd->start == nd->end)
 			continue;
 		if (nd->end > start && nd->start < end)
-			return i;
+			return memblk_nodeid[i];
 		if (nd->end == end && nd->start == start)
-			return i;
+			return memblk_nodeid[i];
 	}
 	return -1;
 }
@@ -258,7 +262,7 @@
 		bad_srat();
 		return;
 	}
-	i = conflicting_nodes(start, end);
+	i = conflicting_memblks(start, end);
 	if (i == node) {
 		printk(KERN_WARNING
 		"SRAT: Warning: PXM %d (%lx-%lx) overlaps with itself (%Lx-%Lx)\n",
@@ -283,10 +287,10 @@
 			nd->end = end;
 	}
  
-	printk(KERN_INFO "SRAT: Node %u PXM %u %Lx-%Lx\n", node, pxm,
-	       nd->start, nd->end);
-	e820_register_active_regions(node, nd->start >> PAGE_SHIFT,
-						nd->end >> PAGE_SHIFT);
+	printk(KERN_INFO "SRAT: Node %u PXM %u %lx-%lx\n", node, pxm,
+	       start, end);
+	e820_register_active_regions(node, start >> PAGE_SHIFT,
+				     end >> PAGE_SHIFT);
 	push_node_boundaries(node, nd->start >> PAGE_SHIFT,
 						nd->end >> PAGE_SHIFT);
  
@@ -298,6 +302,11 @@
 		if ((nd->start | nd->end) == 0)
 			node_clear(node, nodes_parsed);
 	}
+
+	node_memblk_range[num_node_memblks].start = start;
+	node_memblk_range[num_node_memblks].end = end;
+	memblk_nodeid[num_node_memblks] = node;
+	num_node_memblks++;
 }
  
 /* Sanity check to catch more bad SRATs (they are amazingly common).
@@ -368,7 +377,8 @@
 		return -1;
 	}
  
-	memnode_shift = compute_hash_shift(nodes, MAX_NUMNODES);
+	memnode_shift = compute_hash_shift(node_memblk_range, num_node_memblks,
+					   memblk_nodeid);
 	if (memnode_shift < 0) {
 		printk(KERN_ERR
 		     "SRAT: No NUMA node hash function found. Contact maintainer\n");
@@ -9,7 +9,8 @@
 	u64 end;
 };
  
-extern int compute_hash_shift(struct bootnode *nodes, int numnodes);
+extern int compute_hash_shift(struct bootnode *nodes, int numblks,
+			      int *nodeids);
  
 #define ZONE_ALIGN (1UL << (MAX_ORDER+PAGE_SHIFT))
...	...	@@ -903,6 +903,15 @@
903	903	help
904	904	Enable ACPI SRAT based node topology detection.
905	905
	906	+# Some NUMA nodes have memory ranges that span
	907	+# other nodes. Even though a pfn is valid and
	908	+# between a node's start and end pfns, it may not
	909	+# reside on that node. See memmap_init_zone()
	910	+# for details.
	911	+config NODES_SPAN_OTHER_NODES
	912	+ def_bool y
	913	+ depends on X86_64_ACPI_NUMA
	914	+
906	915	config NUMA_EMU
907	916	bool "NUMA emulation"
908	917	depends on X86_64 && NUMA
...	...	@@ -164,7 +164,7 @@
164	164	if (!found)
165	165	return -1;
166	166
167		- memnode_shift = compute_hash_shift(nodes, 8);
	167	+ memnode_shift = compute_hash_shift(nodes, 8, NULL);
168	168	if (memnode_shift < 0) {
169	169	printk(KERN_ERR "No NUMA node hash function found. Contact maintainer\n");
170	170	return -1;
...	...	@@ -60,7 +60,7 @@
60	60	* -1 if node overlap or lost ram (shift too big)
61	61	*/
62	62	static int __init populate_memnodemap(const struct bootnode *nodes,
63		- int numnodes, int shift)
	63	+ int numnodes, int shift, int *nodeids)
64	64	{
65	65	unsigned long addr, end;
66	66	int i, res = -1;
...	...	@@ -76,7 +76,12 @@
76	76	do {
77	77	if (memnodemap[addr >> shift] != NUMA_NO_NODE)
78	78	return -1;
79		- memnodemap[addr >> shift] = i;
	79	+
	80	+ if (!nodeids)
	81	+ memnodemap[addr >> shift] = i;
	82	+ else
	83	+ memnodemap[addr >> shift] = nodeids[i];
	84	+
80	85	addr += (1UL << shift);
81	86	} while (addr < end);
82	87	res = 1;
...	...	@@ -139,7 +144,8 @@
139	144	return i;
140	145	}
141	146
142		-int __init compute_hash_shift(struct bootnode *nodes, int numnodes)
	147	+int __init compute_hash_shift(struct bootnode *nodes, int numnodes,
	148	+ int *nodeids)
143	149	{
144	150	int shift;
145	151
...	...	@@ -149,7 +155,7 @@
149	155	printk(KERN_DEBUG "NUMA: Using %d for the hash shift.\n",
150	156	shift);
151	157
152		- if (populate_memnodemap(nodes, numnodes, shift) != 1) {
	158	+ if (populate_memnodemap(nodes, numnodes, shift, nodeids) != 1) {
153	159	printk(KERN_INFO "Your memory is not aligned you need to "
154	160	"rebuild your kernel with a bigger NODEMAPSIZE "
155	161	"shift=%d\n", shift);
...	...	@@ -462,7 +468,7 @@
462	468	}
463	469	}
464	470	out:
465		- memnode_shift = compute_hash_shift(nodes, num_nodes);
	471	+ memnode_shift = compute_hash_shift(nodes, num_nodes, NULL);
466	472	if (memnode_shift < 0) {
467	473	memnode_shift = 0;
468	474	printk(KERN_ERR "No NUMA hash function found. NUMA emulation "
...	...	@@ -32,6 +32,10 @@
32	32	static int found_add_area __initdata;
33	33	int hotadd_percent __initdata = 0;
34	34
	35	+static int num_node_memblks __initdata;
	36	+static struct bootnode node_memblk_range[NR_NODE_MEMBLKS] __initdata;
	37	+static int memblk_nodeid[NR_NODE_MEMBLKS] __initdata;
	38	+
35	39	/* Too small nodes confuse the VM badly. Usually they result
36	40	from BIOS bugs. */
37	41	#define NODE_MIN_SIZE (410241024)
38	42
39	43
40	44
...	...	@@ -41,17 +45,17 @@
41	45	return acpi_map_pxm_to_node(pxm);
42	46	}
43	47
44		-static __init int conflicting_nodes(unsigned long start, unsigned long end)
	48	+static __init int conflicting_memblks(unsigned long start, unsigned long end)
45	49	{
46	50	int i;
47		- for_each_node_mask(i, nodes_parsed) {
48		- struct bootnode *nd = &nodes[i];
	51	+ for (i = 0; i < num_node_memblks; i++) {
	52	+ struct bootnode *nd = &node_memblk_range[i];
49	53	if (nd->start == nd->end)
50	54	continue;
51	55	if (nd->end > start && nd->start < end)
52		- return i;
	56	+ return memblk_nodeid[i];
53	57	if (nd->end == end && nd->start == start)
54		- return i;
	58	+ return memblk_nodeid[i];
55	59	}
56	60	return -1;
57	61	}
...	...	@@ -258,7 +262,7 @@
258	262	bad_srat();
259	263	return;
260	264	}
261		- i = conflicting_nodes(start, end);
	265	+ i = conflicting_memblks(start, end);
262	266	if (i == node) {
263	267	printk(KERN_WARNING
264	268	"SRAT: Warning: PXM %d (%lx-%lx) overlaps with itself (%Lx-%Lx)\n",
...	...	@@ -283,10 +287,10 @@
283	287	nd->end = end;
284	288	}
285	289
286		- printk(KERN_INFO "SRAT: Node %u PXM %u %Lx-%Lx\n", node, pxm,
287		- nd->start, nd->end);
288		- e820_register_active_regions(node, nd->start >> PAGE_SHIFT,
289		- nd->end >> PAGE_SHIFT);
	290	+ printk(KERN_INFO "SRAT: Node %u PXM %u %lx-%lx\n", node, pxm,
	291	+ start, end);
	292	+ e820_register_active_regions(node, start >> PAGE_SHIFT,
	293	+ end >> PAGE_SHIFT);
290	294	push_node_boundaries(node, nd->start >> PAGE_SHIFT,
291	295	nd->end >> PAGE_SHIFT);
292	296
...	...	@@ -298,6 +302,11 @@
298	302	if ((nd->start \| nd->end) == 0)
299	303	node_clear(node, nodes_parsed);
300	304	}
	305	+
	306	+ node_memblk_range[num_node_memblks].start = start;
	307	+ node_memblk_range[num_node_memblks].end = end;
	308	+ memblk_nodeid[num_node_memblks] = node;
	309	+ num_node_memblks++;
301	310	}
302	311
303	312	/* Sanity check to catch more bad SRATs (they are amazingly common).
...	...	@@ -368,7 +377,8 @@
368	377	return -1;
369	378	}
370	379
371		- memnode_shift = compute_hash_shift(nodes, MAX_NUMNODES);
	380	+ memnode_shift = compute_hash_shift(node_memblk_range, num_node_memblks,
	381	+ memblk_nodeid);
372	382	if (memnode_shift < 0) {
373	383	printk(KERN_ERR
374	384	"SRAT: No NUMA node hash function found. Contact maintainer\n");
...	...	@@ -9,7 +9,8 @@
9	9	u64 end;
10	10	};
11	11
12		-extern int compute_hash_shift(struct bootnode *nodes, int numnodes);
	12	+extern int compute_hash_shift(struct bootnode *nodes, int numblks,
	13	+ int *nodeids);
13	14
14	15	#define ZONE_ALIGN (1UL << (MAX_ORDER+PAGE_SHIFT))
15	16