x86/numa: Correct assumption that each NUMA node has memory

It is now quite easy to buy servers with incorrectly populated DIMMs,
especially with AMD Magny-Cours and Interlagos systems which have two
NUMA nodes per socket.

Currently, Xen will assign all CPUs on nodes without memory to node 0,
which leads to interestingly wrong NUMA information, causing numa
aware functionality such as alloc_domheap_pages() to get things very
wrong.

This patch splits the current logic to accept NUMA nodes without
memory, which corrects the accounting of CPUs to online NUMA nodes.

Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Committed-by: Keir Fraser <keir@xen.org>

1 files changed, 19 insertions, 7 deletions

diff --git a/xen/arch/x86/srat.c b/xen/arch/x86/srat.c
index f46e05380c..cee665ae90 100644
--- a/xen/arch/x86/srat.c
+++ b/xen/arch/x86/srat.c
@@ -23,7 +23,8 @@
 
 static struct acpi_table_slit *__read_mostly acpi_slit;
 
-static nodemask_t nodes_parsed __initdata;
+static nodemask_t memory_nodes_parsed __initdata;
+static nodemask_t processor_nodes_parsed __initdata;
 static nodemask_t nodes_found __initdata;
 static struct node nodes[MAX_NUMNODES] __initdata;
 static u8 __read_mostly pxm2node[256] = { [0 ... 255] = NUMA_NO_NODE };
@@ -221,6 +222,7 @@ acpi_numa_processor_affinity_init(struct acpi_srat_cpu_affinity *pa)
 		return;
 	}
 	apicid_to_node[pa->apic_id] = node;
+	node_set(node, processor_nodes_parsed);
 	acpi_numa = 1;
 	printk(KERN_INFO "SRAT: PXM %u -> APIC %u -> Node %u\n",
 	       pxm, pa->apic_id, node);
@@ -287,7 +289,7 @@ acpi_numa_memory_affinity_init(struct acpi_srat_mem_affinity *ma)
 		return;
 	}
 	nd = &nodes[node];
-	if (!node_test_and_set(node, nodes_parsed)) {
+	if (!node_test_and_set(node, memory_nodes_parsed)) {
 		nd->start = start;
 		nd->end = end;
 	} else {
@@ -324,7 +326,7 @@ static int nodes_cover_memory(void)
 
 		do {
 			found = 0;
-			for_each_node_mask(j, nodes_parsed)
+			for_each_node_mask(j, memory_nodes_parsed)
 				if (start < nodes[j].end
 				    && end > nodes[j].start) {
 					if (start >= nodes[j].start) {
@@ -418,6 +420,7 @@ void __init srat_parse_regions(u64 addr)
 int __init acpi_scan_nodes(u64 start, u64 end)
 {
 	int i;
+	nodemask_t all_nodes_parsed;
 
 	/* First clean up the node list */
 	for (i = 0; i < MAX_NUMNODES; i++)
@@ -441,17 +444,26 @@ int __init acpi_scan_nodes(u64 start, u64 end)
 		return -1;
 	}
 
+	nodes_or(all_nodes_parsed, memory_nodes_parsed, processor_nodes_parsed);
+
 	/* Finally register nodes */
-	for_each_node_mask(i, nodes_parsed)
+	for_each_node_mask(i, all_nodes_parsed)
 	{
-		if ((nodes[i].end - nodes[i].start) < NODE_MIN_SIZE)
-			continue;
+		u64 size = nodes[i].end - nodes[i].start;
+		if ( size == 0 )
+			printk(KERN_WARNING "SRAT: Node %u has no memory. "
+			       "BIOS Bug or mis-configured hardware?\n", i);
+
+		else if (size < NODE_MIN_SIZE)
+			printk(KERN_WARNING "SRAT: Node %u has only %"PRIu64
+			       " bytes of memory. BIOS Bug?\n", i, size);
+
 		setup_node_bootmem(i, nodes[i].start, nodes[i].end);
 	}
 	for (i = 0; i < nr_cpu_ids; i++) {
 		if (cpu_to_node[i] == NUMA_NO_NODE)
 			continue;
-		if (!node_isset(cpu_to_node[i], nodes_parsed))
+		if (!node_isset(cpu_to_node[i], processor_nodes_parsed))
 			numa_set_node(i, NUMA_NO_NODE);
 	}
 	numa_init_array();