diff options
author | kfraser@localhost.localdomain <kfraser@localhost.localdomain> | 2006-10-25 12:25:54 +0100 |
---|---|---|
committer | kfraser@localhost.localdomain <kfraser@localhost.localdomain> | 2006-10-25 12:25:54 +0100 |
commit | 582724b0ea12d27f3a18c8c3c15cccf7d655dae6 (patch) | |
tree | 7ab46ccde55050f9725b8446edfe48d2ae5ea316 /xen/arch/x86/srat.c | |
parent | 67ce9409e141bb4949f2f037e937ef3fc18a8608 (diff) | |
download | xen-582724b0ea12d27f3a18c8c3c15cccf7d655dae6.tar.gz xen-582724b0ea12d27f3a18c8c3c15cccf7d655dae6.tar.bz2 xen-582724b0ea12d27f3a18c8c3c15cccf7d655dae6.zip |
[XEN] Add basic NUMA/SRAT support to Xen from Linux 2.6.16.29.
Signed-off-by: Ryan Harper <ryanh@us.ibm.com>
Diffstat (limited to 'xen/arch/x86/srat.c')
-rw-r--r-- | xen/arch/x86/srat.c | 325 |
1 files changed, 325 insertions, 0 deletions
diff --git a/xen/arch/x86/srat.c b/xen/arch/x86/srat.c new file mode 100644 index 0000000000..dfa85b2539 --- /dev/null +++ b/xen/arch/x86/srat.c @@ -0,0 +1,325 @@ +/* + * ACPI 3.0 based NUMA setup + * Copyright 2004 Andi Kleen, SuSE Labs. + * + * Reads the ACPI SRAT table to figure out what memory belongs to which CPUs. + * + * Called from acpi_numa_init while reading the SRAT and SLIT tables. + * Assumes all memory regions belonging to a single proximity domain + * are in one chunk. Holes between them will be included in the node. + * + * Adapted for Xen: Ryan Harper <ryanh@us.ibm.com> + */ + +#if 0 +#include <linux/kernel.h> +#include <linux/module.h> +#include <asm/proto.h> +#include <xen/bitmap.h> +#include <xen/numa.h> +#include <xen/topology.h> +#include <asm/e820.h> +#endif +#include <xen/init.h> +#include <xen/mm.h> +#include <xen/inttypes.h> +#include <xen/nodemask.h> +#include <xen/acpi.h> + +#include <asm/numa.h> +#include <asm/page.h> + +static struct acpi_table_slit *acpi_slit; + +static nodemask_t nodes_parsed __initdata; +static nodemask_t nodes_found __initdata; +static struct node nodes[MAX_NUMNODES] __initdata; +static u8 pxm2node[256] = { [0 ... 255] = 0xff }; + +/* Too small nodes confuse the VM badly. Usually they result + from BIOS bugs. */ +#define NODE_MIN_SIZE (4*1024*1024) + +static int node_to_pxm(int n); + +int pxm_to_node(int pxm) +{ + if ((unsigned)pxm >= 256) + return -1; + /* Extend 0xff to (int)-1 */ + return (signed char)pxm2node[pxm]; +} + +static __init int setup_node(int pxm) +{ + unsigned node = pxm2node[pxm]; + if (node == 0xff) { + if (nodes_weight(nodes_found) >= MAX_NUMNODES) + return -1; + node = first_unset_node(nodes_found); + node_set(node, nodes_found); + pxm2node[pxm] = node; + } + return pxm2node[pxm]; +} + +static __init int conflicting_nodes(u64 start, u64 end) +{ + int i; + for_each_node_mask(i, nodes_parsed) { + struct node *nd = &nodes[i]; + if (nd->start == nd->end) + continue; + if (nd->end > start && nd->start < end) + return i; + if (nd->end == end && nd->start == start) + return i; + } + return -1; +} + +static __init void cutoff_node(int i, u64 start, u64 end) +{ + struct node *nd = &nodes[i]; + if (nd->start < start) { + nd->start = start; + if (nd->end < nd->start) + nd->start = nd->end; + } + if (nd->end > end) { + nd->end = end; + if (nd->start > nd->end) + nd->start = nd->end; + } +} + +static __init void bad_srat(void) +{ + int i; + printk(KERN_ERR "SRAT: SRAT not used.\n"); + acpi_numa = -1; + for (i = 0; i < MAX_LOCAL_APIC; i++) + apicid_to_node[i] = NUMA_NO_NODE; +} + +static __init inline int srat_disabled(void) +{ + return numa_off || acpi_numa < 0; +} + +/* + * A lot of BIOS fill in 10 (= no distance) everywhere. This messes + * up the NUMA heuristics which wants the local node to have a smaller + * distance than the others. + * Do some quick checks here and only use the SLIT if it passes. + */ +static __init int slit_valid(struct acpi_table_slit *slit) +{ + int i, j; + int d = slit->localities; + for (i = 0; i < d; i++) { + for (j = 0; j < d; j++) { + u8 val = slit->entry[d*i + j]; + if (i == j) { + if (val != 10) + return 0; + } else if (val <= 10) + return 0; + } + } + return 1; +} + +/* Callback for SLIT parsing */ +void __init acpi_numa_slit_init(struct acpi_table_slit *slit) +{ + if (!slit_valid(slit)) { + printk(KERN_INFO "ACPI: SLIT table looks invalid. Not used.\n"); + return; + } + acpi_slit = slit; +} + +/* Callback for Proximity Domain -> LAPIC mapping */ +void __init +acpi_numa_processor_affinity_init(struct acpi_table_processor_affinity *pa) +{ + int pxm, node; + if (srat_disabled()) + return; + if (pa->header.length != sizeof(struct acpi_table_processor_affinity)) { bad_srat(); + return; + } + if (pa->flags.enabled == 0) + return; + pxm = pa->proximity_domain; + node = setup_node(pxm); + if (node < 0) { + printk(KERN_ERR "SRAT: Too many proximity domains %x\n", pxm); + bad_srat(); + return; + } + apicid_to_node[pa->apic_id] = node; + acpi_numa = 1; + printk(KERN_INFO "SRAT: PXM %u -> APIC %u -> Node %u\n", + pxm, pa->apic_id, node); +} + +/* Callback for parsing of the Proximity Domain <-> Memory Area mappings */ +void __init +acpi_numa_memory_affinity_init(struct acpi_table_memory_affinity *ma) +{ + struct node *nd; + u64 start, end; + int node, pxm; + int i; + + if (srat_disabled()) + return; + if (ma->header.length != sizeof(struct acpi_table_memory_affinity)) { + bad_srat(); + return; + } + if (ma->flags.enabled == 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; + node = setup_node(pxm); + if (node < 0) { + printk(KERN_ERR "SRAT: Too many proximity domains.\n"); + 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 %"PRIx64" - %"PRIx64" \n", + start, end); + i = conflicting_nodes(start, end); + if (i == node) { + printk(KERN_WARNING + "SRAT: Warning: PXM %d (%"PRIx64"-%"PRIx64") overlaps with itself (%" + PRIx64"-%"PRIx64")\n", pxm, start, end, nodes[i].start, nodes[i].end); + } else if (i >= 0) { + printk(KERN_ERR + "SRAT: PXM %d (%"PRIx64"-%"PRIx64") overlaps with PXM %d (%" + PRIx64"-%"PRIx64")\n", pxm, start, end, node_to_pxm(i), + nodes[i].start, nodes[i].end); + bad_srat(); + return; + } + nd = &nodes[node]; + if (!node_test_and_set(node, nodes_parsed)) { + nd->start = start; + nd->end = end; + } else { + if (start < nd->start) + nd->start = start; + if (nd->end < end) + nd->end = end; + } + printk(KERN_INFO "SRAT: Node %u PXM %u %"PRIx64"-%"PRIx64"\n", node, pxm, + nd->start, nd->end); +} + +/* Sanity check to catch more bad SRATs (they are amazingly common). + Make sure the PXMs cover all memory. */ +static int nodes_cover_memory(void) +{ + int i; + u64 pxmram, e820ram; + + pxmram = 0; + for_each_node_mask(i, nodes_parsed) { + u64 s = nodes[i].start >> PAGE_SHIFT; + u64 e = nodes[i].end >> PAGE_SHIFT; + pxmram += e - s; + } + + e820ram = max_page; + /* We seem to lose 3 pages somewhere. Allow a bit of slack. */ + if ((long)(e820ram - pxmram) >= 1*1024*1024) { + printk(KERN_ERR "SRAT: PXMs only cover %"PRIu64"MB of your %" + PRIu64"MB e820 RAM. Not used.\n", + (pxmram << PAGE_SHIFT) >> 20, + (e820ram << PAGE_SHIFT) >> 20); + return 0; + } + return 1; +} + +static void unparse_node(int node) +{ + int i; + node_clear(node, nodes_parsed); + for (i = 0; i < MAX_LOCAL_APIC; i++) { + if (apicid_to_node[i] == node) + apicid_to_node[i] = NUMA_NO_NODE; + } +} + +void __init acpi_numa_arch_fixup(void) {} + +/* Use the information discovered above to actually set up the nodes. */ +int __init acpi_scan_nodes(u64 start, u64 end) +{ + int i; + + /* First clean up the node list */ + for (i = 0; i < MAX_NUMNODES; i++) { + cutoff_node(i, start, end); + if ((nodes[i].end - nodes[i].start) < NODE_MIN_SIZE) + unparse_node(i); + } + + if (acpi_numa <= 0) + return -1; + + if (!nodes_cover_memory()) { + bad_srat(); + return -1; + } + + memnode_shift = compute_hash_shift(nodes, MAX_NUMNODES); + if (memnode_shift < 0) { + printk(KERN_ERR + "SRAT: No NUMA node hash function found. Contact maintainer\n"); + bad_srat(); + return -1; + } + + /* Finally register nodes */ + for_each_node_mask(i, nodes_parsed) + setup_node_bootmem(i, nodes[i].start, nodes[i].end); + for (i = 0; i < NR_CPUS; i++) { + if (cpu_to_node[i] == NUMA_NO_NODE) + continue; + if (!node_isset(cpu_to_node[i], nodes_parsed)) + numa_set_node(i, NUMA_NO_NODE); + } + numa_init_array(); + return 0; +} + +static int node_to_pxm(int n) +{ + int i; + if (pxm2node[n] == n) + return n; + for (i = 0; i < 256; i++) + if (pxm2node[i] == n) + return i; + return 0; +} + +int __node_distance(int a, int b) +{ + int index; + + if (!acpi_slit) + return a == b ? 10 : 20; + index = acpi_slit->localities * node_to_pxm(a); + return acpi_slit->entry[index + node_to_pxm(b)]; +} + +EXPORT_SYMBOL(__node_distance); |