[PATCH] x86_64: Reserve SRAT hotadd memory on x86-64

From: Keith Mannthey, Andi Kleen

Implement memory hotadd without sparsemem. The memory in the SRAT
hotadd area is just preserved instead and can be activated later.

There are a few restrictions:
- Only one continuous hotadd area allowed per node

The main problem is dealing with the many buggy SRAT tables
that are out there. The strategy here is to reject anything
suspicious.

Originally from Keith Mannthey, with several hacks and changes by AK
and also contributions from Andrew Morton

[ TBD: Problems pointed out by KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>:

 1) Goto's rebuild_zonelist patch will not work if CONFIG_MEMORY_HOTPLUG=n.

    Rebuilding zonelist is necessary when the system has just memory <
    4G at boot, and hot add memory > 4G.  because x86_64 has DMA32,
    ZONE_NORAML is not included into zonelist at boot time if system
    doesn't have memory >4G at boot.

    [AK: should just force the higher zones at boot time when SRAT tells us]

 2) zone and node's spanned_pages and present_pages are not incremented.
    They should be.

    For example, our server (ia64/Fujitsu PrimeQuest) can equip memory
    from 4G to 1T(maybe 2T in future), and SRAT will *always* say we have
    possible 1T +memory.  (Microsoft requires "write all possible memory
    in SRAT") When we reserve memmap for possible 1T memory, Linux will
    not work well in +minimum 4G configuraion ;)

    [AK: needs limiting to 5-10% of max memory]
 ]

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Documentation/x86_64/boot-options.txt

@@ -151,6 +151,11 @@ NUMA
 
   numa=fake=X   Fake X nodes and ignore NUMA setup of the actual machine.
 
+  numa=hotadd=percent
+		Only allow hotadd memory to preallocate page structures upto
+		percent of already available memory.
+		numa=hotadd=0 will disable hotadd memory.
+
 ACPI
 
   acpi=off	Don't enable ACPI

arch/x86_64/mm/init.c

@@ -530,8 +530,7 @@ int __add_pages(struct zone *z, unsigned long start_pfn, unsigned long nr_pages)
 	unsigned long pfn;
 	unsigned long total = 0, mem = 0;
 	for (pfn = start_pfn; pfn < start_pfn + nr_pages; pfn++) {
-		unsigned long addr = pfn << PAGE_SHIFT;
-		if (pfn_valid(pfn) && e820_mapped(addr, addr+1, E820_RAM)) {
+		if (pfn_valid(pfn)) {
 			online_page(pfn_to_page(pfn));
 			err = 0;
 			mem++;

arch/x86_64/mm/numa.c

@@ -142,6 +142,9 @@ void __init setup_node_bootmem(int nodeid, unsigned long start, unsigned long en
 
 	reserve_bootmem_node(NODE_DATA(nodeid), nodedata_phys, pgdat_size); 
 	reserve_bootmem_node(NODE_DATA(nodeid), bootmap_start, bootmap_pages<<PAGE_SHIFT);
+#ifdef CONFIG_ACPI_NUMA
+	srat_reserve_add_area(nodeid);
+#endif
 	node_set_online(nodeid);
 } 
 
@@ -335,6 +338,8 @@ __init int numa_setup(char *opt)
 #ifdef CONFIG_ACPI_NUMA
  	if (!strncmp(opt,"noacpi",6))
  		acpi_numa = -1;
+	if (!strncmp(opt,"hotadd=", 7))
+		hotadd_percent = simple_strtoul(opt+7, NULL, 10);
 #endif
 	return 1;
 } 

arch/x86_64/mm/srat.c

@@ -15,15 +15,26 @@
 #include <linux/bitmap.h>
 #include <linux/module.h>
 #include <linux/topology.h>
+#include <linux/bootmem.h>
+#include <linux/mm.h>
 #include <asm/proto.h>
 #include <asm/numa.h>
 #include <asm/e820.h>
 
+#if (defined(CONFIG_ACPI_HOTPLUG_MEMORY) || \
+	defined(CONFIG_ACPI_HOTPLUG_MEMORY_MODULE)) \
+		&& !defined(CONFIG_MEMORY_HOTPLUG)
+#define RESERVE_HOTADD 1
+#endif
+
 static struct acpi_table_slit *acpi_slit;
 
 static nodemask_t nodes_parsed __initdata;
 static nodemask_t nodes_found __initdata;
 static struct bootnode nodes[MAX_NUMNODES] __initdata;
+static struct bootnode nodes_add[MAX_NUMNODES] __initdata;
+static int found_add_area __initdata;
+int hotadd_percent __initdata = 10;
 static u8 pxm2node[256] = { [0 ... 255] = 0xff };
 
 /* Too small nodes confuse the VM badly. Usually they result
@@ -71,6 +82,10 @@ static __init int conflicting_nodes(unsigned long start, unsigned long end)
 static __init void cutoff_node(int i, unsigned long start, unsigned long end)
 {
 	struct bootnode *nd = &nodes[i];
+
+	if (found_add_area)
+		return;
+
 	if (nd->start < start) {
 		nd->start = start;
 		if (nd->end < nd->start)
@@ -90,6 +105,8 @@ static __init void bad_srat(void)
 	acpi_numa = -1;
 	for (i = 0; i < MAX_LOCAL_APIC; i++)
 		apicid_to_node[i] = NUMA_NO_NODE;
+	for (i = 0; i < MAX_NUMNODES; i++)
+		nodes_add[i].start = nodes[i].end = 0;
 }
 
 static __init inline int srat_disabled(void)
@@ -155,11 +172,114 @@ acpi_numa_processor_affinity_init(struct acpi_table_processor_affinity *pa)
 	       pxm, pa->apic_id, node);
 }
 
+#ifdef RESERVE_HOTADD
+/*
+ * Protect against too large hotadd areas that would fill up memory.
+ */
+static int hotadd_enough_memory(struct bootnode *nd)
+{
+	static unsigned long allocated;
+	static unsigned long last_area_end;
+	unsigned long pages = (nd->end - nd->start) >> PAGE_SHIFT;
+	long mem = pages * sizeof(struct page);
+	unsigned long addr;
+	unsigned long allowed;
+	unsigned long oldpages = pages;
+
+	if (mem < 0)
+		return 0;
+	allowed = (end_pfn - e820_hole_size(0, end_pfn)) * PAGE_SIZE;
+	allowed = (allowed / 100) * hotadd_percent;
+	if (allocated + mem > allowed) {
+		/* Give them at least part of their hotadd memory upto hotadd_percent
+		   It would be better to spread the limit out
+		   over multiple hotplug areas, but that is too complicated
+		   right now */
+		if (allocated >= allowed)
+			return 0;
+		pages = (allowed - allocated + mem) / sizeof(struct page);
+		mem = pages * sizeof(struct page);
+		nd->end = nd->start + pages*PAGE_SIZE;
+	}
+	/* Not completely fool proof, but a good sanity check */
+	addr = find_e820_area(last_area_end, end_pfn<<PAGE_SHIFT, mem);
+	if (addr == -1UL)
+		return 0;
+	if (pages != oldpages)
+		printk(KERN_NOTICE "SRAT: Hotadd area limited to %lu bytes\n",
+			pages << PAGE_SHIFT);
+	last_area_end = addr + mem;
+	allocated += mem;
+	return 1;
+}
+
+/*
+ * It is fine to add this area to the nodes data it will be used later
+ * This code supports one contigious hot add area per node.
+ */
+static int reserve_hotadd(int node, unsigned long start, unsigned long end)
+{
+	unsigned long s_pfn = start >> PAGE_SHIFT;
+	unsigned long e_pfn = end >> PAGE_SHIFT;
+	int changed = 0;
+	struct bootnode *nd = &nodes_add[node];
+
+	/* I had some trouble with strange memory hotadd regions breaking
+	   the boot. Be very strict here and reject anything unexpected.
+	   If you want working memory hotadd write correct SRATs.
+
+	   The node size check is a basic sanity check to guard against
+	   mistakes */
+	if ((signed long)(end - start) < NODE_MIN_SIZE) {
+		printk(KERN_ERR "SRAT: Hotplug area too small\n");
+		return -1;
+	}
+
+	/* This check might be a bit too strict, but I'm keeping it for now. */
+	if (e820_hole_size(s_pfn, e_pfn) != e_pfn - s_pfn) {
+		printk(KERN_ERR "SRAT: Hotplug area has existing memory\n");
+		return -1;
+	}
+
+	if (!hotadd_enough_memory(&nodes_add[node]))  {
+		printk(KERN_ERR "SRAT: Hotplug area too large\n");
+		return -1;
+	}
+
+	/* Looks good */
+
+ 	found_add_area = 1;
+	if (nd->start == nd->end) {
+ 		nd->start = start;
+ 		nd->end = end;
+		changed = 1;
+ 	} else {
+ 		if (nd->start == end) {
+ 			nd->start = start;
+			changed = 1;
+		}
+ 		if (nd->end == start) {
+ 			nd->end = end;
+			changed = 1;
+		}
+		if (!changed)
+			printk(KERN_ERR "SRAT: Hotplug zone not continuous. Partly ignored\n");
+ 	}
+
+ 	if ((nd->end >> PAGE_SHIFT) > end_pfn)
+ 		end_pfn = nd->end >> PAGE_SHIFT;
+
+	if (changed)
+	 	printk(KERN_INFO "SRAT: hot plug zone found %Lx - %Lx\n", nd->start, nd->end);
+	return 0;
+}
+#endif
+
 /* Callback for parsing of the Proximity Domain <-> Memory Area mappings */
 void __init
 acpi_numa_memory_affinity_init(struct acpi_table_memory_affinity *ma)
 {
-	struct bootnode *nd;
+	struct bootnode *nd, oldnode;
 	unsigned long start, end;
 	int node, pxm;
 	int i;
@@ -172,6 +292,8 @@ acpi_numa_memory_affinity_init(struct acpi_table_memory_affinity *ma)
 	}
 	if (ma->flags.enabled == 0)
 		return;
+ 	if (ma->flags.hot_pluggable && hotadd_percent == 0)
+		return;
 	start = ma->base_addr_lo | ((u64)ma->base_addr_hi << 32);
 	end = start + (ma->length_lo | ((u64)ma->length_hi << 32));
 	pxm = ma->proximity_domain;
@@ -181,10 +303,6 @@ acpi_numa_memory_affinity_init(struct acpi_table_memory_affinity *ma)
 		bad_srat();
 		return;
 	}
-	/* It is fine to add this area to the nodes data it will be used later*/
-	if (ma->flags.hot_pluggable == 1)
-		printk(KERN_INFO "SRAT: hot plug zone found %lx - %lx \n",
-				start, end);
 	i = conflicting_nodes(start, end);
 	if (i == node) {
 		printk(KERN_WARNING
@@ -199,6 +317,7 @@ acpi_numa_memory_affinity_init(struct acpi_table_memory_affinity *ma)
 		return;
 	}
 	nd = &nodes[node];
+	oldnode = *nd;
 	if (!node_test_and_set(node, nodes_parsed)) {
 		nd->start = start;
 		nd->end = end;
@@ -208,8 +327,19 @@ acpi_numa_memory_affinity_init(struct acpi_table_memory_affinity *ma)
 		if (nd->end < end)
 			nd->end = end;
 	}
+
 	printk(KERN_INFO "SRAT: Node %u PXM %u %Lx-%Lx\n", node, pxm,
 	       nd->start, nd->end);
+
+#ifdef RESERVE_HOTADD
+ 	if (ma->flags.hot_pluggable && reserve_hotadd(node, start, end) < 0) {
+		/* Ignore hotadd region. Undo damage */
+		printk(KERN_NOTICE "SRAT: Hotplug region ignored\n");
+		*nd = oldnode;
+		if ((nd->start | nd->end) == 0)
+			node_clear(node, nodes_parsed);
+	}
+#endif
 }
 
 /* Sanity check to catch more bad SRATs (they are amazingly common).
@@ -225,6 +355,9 @@ static int nodes_cover_memory(void)
 		unsigned long e = nodes[i].end >> PAGE_SHIFT;
 		pxmram += e - s;
 		pxmram -= e820_hole_size(s, e);
+		pxmram -= nodes_add[i].end - nodes_add[i].start;
+		if ((long)pxmram < 0)
+			pxmram = 0;
 	}
 
 	e820ram = end_pfn - e820_hole_size(0, end_pfn);
@@ -303,6 +436,25 @@ static int node_to_pxm(int n)
        return 0;
 }
 
+void __init srat_reserve_add_area(int nodeid)
+{
+	if (found_add_area && nodes_add[nodeid].end) {
+		u64 total_mb;
+
+		printk(KERN_INFO "SRAT: Reserving hot-add memory space "
+				"for node %d at %Lx-%Lx\n",
+			nodeid, nodes_add[nodeid].start, nodes_add[nodeid].end);
+		total_mb = (nodes_add[nodeid].end - nodes_add[nodeid].start)
+					>> PAGE_SHIFT;
+		total_mb *= sizeof(struct page);
+		total_mb >>= 20;
+		printk(KERN_INFO "SRAT: This will cost you %Lu MB of "
+				"pre-allocated memory.\n", (unsigned long long)total_mb);
+		reserve_bootmem_node(NODE_DATA(nodeid), nodes_add[nodeid].start,
+			       nodes_add[nodeid].end - nodes_add[nodeid].start);
+	}
+}
+
 int __node_distance(int a, int b)
 {
 	int index;

include/asm-x86_64/numa.h

@@ -18,6 +18,8 @@ extern void numa_init_array(void);
 extern int numa_off;
 
 extern void numa_set_node(int cpu, int node);
+extern void srat_reserve_add_area(int nodeid);
+extern int hotadd_percent;
 
 extern unsigned char apicid_to_node[256];
 #ifdef CONFIG_NUMA