x86: make NUMA work on 32-bit
The DISCONTIG memory model on x86 32 bit uses a remap allocator early
in boot. The objective is that portions of every node are mapped in to
the kernel virtual area (KVA) in place of ZONE_NORMAL so that node-local
allocations can be made for pgdat and mem_map structures.
With SPARSEMEM, the amount that is set aside is insufficient for all the
mem_maps to be allocated. During the boot process, it falls back to using
the bootmem allocator. This breaks assumptions that SPARSEMEM makes about
the layout of the mem_map in memory and results in a VM_BUG_ON triggering
due to pfn_to_page() returning garbage values.
This patch only enables the remap allocator for use with DISCONTIG.
Without SRAT support, a compile-error occurs because ACPI table parsing
functions are only available in x86-64. This patch also adds no-op stubs
and prints a warning message. What likely needs to be done is sharing
the table parsing functions between 32 and 64 bit if they are
compatible.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
diff --git a/arch/x86/mm/discontig_32.c b/arch/x86/mm/discontig_32.c
index 9f1d02cf..04b1d20 100644
--- a/arch/x86/mm/discontig_32.c
+++ b/arch/x86/mm/discontig_32.c
@@ -32,6 +32,7 @@
#include <linux/kexec.h>
#include <linux/pfn.h>
#include <linux/swap.h>
+#include <linux/acpi.h>
#include <asm/e820.h>
#include <asm/setup.h>
@@ -103,14 +104,10 @@
#define LARGE_PAGE_BYTES (PTRS_PER_PTE * PAGE_SIZE)
-static unsigned long node_remap_start_pfn[MAX_NUMNODES];
unsigned long node_remap_size[MAX_NUMNODES];
-static unsigned long node_remap_offset[MAX_NUMNODES];
static void *node_remap_start_vaddr[MAX_NUMNODES];
void set_pmd_pfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags);
-static void *node_remap_end_vaddr[MAX_NUMNODES];
-static void *node_remap_alloc_vaddr[MAX_NUMNODES];
static unsigned long kva_start_pfn;
static unsigned long kva_pages;
/*
@@ -167,6 +164,22 @@
}
}
+#ifdef CONFIG_DISCONTIGMEM
+/*
+ * In the discontig memory model, a portion of the kernel virtual area (KVA)
+ * is reserved and portions of nodes are mapped using it. This is to allow
+ * node-local memory to be allocated for structures that would normally require
+ * ZONE_NORMAL. The memory is allocated with alloc_remap() and callers
+ * should be prepared to allocate from the bootmem allocator instead. This KVA
+ * mechanism is incompatible with SPARSEMEM as it makes assumptions about the
+ * layout of memory that are broken if alloc_remap() succeeds for some of the
+ * map and fails for others
+ */
+static unsigned long node_remap_start_pfn[MAX_NUMNODES];
+static void *node_remap_end_vaddr[MAX_NUMNODES];
+static void *node_remap_alloc_vaddr[MAX_NUMNODES];
+static unsigned long node_remap_offset[MAX_NUMNODES];
+
void *alloc_remap(int nid, unsigned long size)
{
void *allocation = node_remap_alloc_vaddr[nid];
@@ -263,6 +276,40 @@
return reserve_pages;
}
+static void init_remap_allocator(int nid)
+{
+ node_remap_start_vaddr[nid] = pfn_to_kaddr(
+ kva_start_pfn + node_remap_offset[nid]);
+ node_remap_end_vaddr[nid] = node_remap_start_vaddr[nid] +
+ (node_remap_size[nid] * PAGE_SIZE);
+ node_remap_alloc_vaddr[nid] = node_remap_start_vaddr[nid] +
+ ALIGN(sizeof(pg_data_t), PAGE_SIZE);
+
+ printk ("node %d will remap to vaddr %08lx - %08lx\n", nid,
+ (ulong) node_remap_start_vaddr[nid],
+ (ulong) pfn_to_kaddr(highstart_pfn
+ + node_remap_offset[nid] + node_remap_size[nid]));
+}
+#else
+void *alloc_remap(int nid, unsigned long size)
+{
+ return NULL;
+}
+
+static unsigned long calculate_numa_remap_pages(void)
+{
+ return 0;
+}
+
+static void init_remap_allocator(int nid)
+{
+}
+
+void __init remap_numa_kva(void)
+{
+}
+#endif /* CONFIG_DISCONTIGMEM */
+
extern void setup_bootmem_allocator(void);
unsigned long __init setup_memory(void)
{
@@ -326,19 +373,9 @@
printk("Low memory ends at vaddr %08lx\n",
(ulong) pfn_to_kaddr(max_low_pfn));
for_each_online_node(nid) {
- node_remap_start_vaddr[nid] = pfn_to_kaddr(
- kva_start_pfn + node_remap_offset[nid]);
- /* Init the node remap allocator */
- node_remap_end_vaddr[nid] = node_remap_start_vaddr[nid] +
- (node_remap_size[nid] * PAGE_SIZE);
- node_remap_alloc_vaddr[nid] = node_remap_start_vaddr[nid] +
- ALIGN(sizeof(pg_data_t), PAGE_SIZE);
+ init_remap_allocator(nid);
allocate_pgdat(nid);
- printk ("node %d will remap to vaddr %08lx - %08lx\n", nid,
- (ulong) node_remap_start_vaddr[nid],
- (ulong) pfn_to_kaddr(highstart_pfn
- + node_remap_offset[nid] + node_remap_size[nid]));
}
printk("High memory starts at vaddr %08lx\n",
(ulong) pfn_to_kaddr(highstart_pfn));
@@ -439,3 +476,29 @@
EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
#endif
+
+#ifndef CONFIG_HAVE_ARCH_PARSE_SRAT
+/*
+ * XXX FIXME: Make SLIT table parsing available to 32-bit NUMA
+ *
+ * These stub functions are needed to compile 32-bit NUMA when SRAT is
+ * not set. There are functions in srat_64.c for parsing this table
+ * and it may be possible to make them common functions.
+ */
+void acpi_numa_slit_init (struct acpi_table_slit *slit)
+{
+ printk(KERN_INFO "ACPI: No support for parsing SLIT table\n");
+}
+
+void acpi_numa_processor_affinity_init (struct acpi_srat_cpu_affinity *pa)
+{
+}
+
+void acpi_numa_memory_affinity_init (struct acpi_srat_mem_affinity *ma)
+{
+}
+
+void acpi_numa_arch_fixup(void)
+{
+}
+#endif /* CONFIG_HAVE_ARCH_PARSE_SRAT */