arch/i386/kernel/srat.c

   1 /*
   2  * Some of the code in this file has been gleaned from the 64 bit
   3  * discontigmem support code base.
   4  *
   5  * Copyright (C) 2002, IBM Corp.
   6  *
   7  * All rights reserved.
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * This program is distributed in the hope that it will be useful, but
  15  * WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  17  * NON INFRINGEMENT.  See the GNU General Public License for more
  18  * details.
  19  *
  20  * You should have received a copy of the GNU General Public License
  21  * along with this program; if not, write to the Free Software
  22  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  23  *
  24  * Send feedback to Pat Gaughen <gone@us.ibm.com>
  25  */
  26 #include <linux/mm.h>
  27 #include <linux/bootmem.h>
  28 #include <linux/mmzone.h>
  29 #include <linux/acpi.h>
  30 #include <linux/nodemask.h>
  31 #include <asm/srat.h>
  32 #include <asm/topology.h>
  33 #include <asm/smp.h>
  34
  35 /*
  36  * proximity macros and definitions
  37  */
  38 #define NODE_ARRAY_INDEX(x)     ((x) / 8)       /* 8 bits/char */
  39 #define NODE_ARRAY_OFFSET(x)    ((x) % 8)       /* 8 bits/char */
  40 #define BMAP_SET(bmap, bit)     ((bmap)[NODE_ARRAY_INDEX(bit)] |= 1 << NODE_ARRAY_OFFSET(bit))
  41 #define BMAP_TEST(bmap, bit)    ((bmap)[NODE_ARRAY_INDEX(bit)] & (1 << NODE_ARRAY_OFFSET(bit)))
  42 /* bitmap length; _PXM is at most 255 */
  43 #define PXM_BITMAP_LEN (MAX_PXM_DOMAINS / 8)
  44 static u8 pxm_bitmap[PXM_BITMAP_LEN];   /* bitmap of proximity domains */
  45
  46 #define MAX_CHUNKS_PER_NODE     3
  47 #define MAXCHUNKS               (MAX_CHUNKS_PER_NODE * MAX_NUMNODES)
  48 struct node_memory_chunk_s {
  49         unsigned long   start_pfn;
  50         unsigned long   end_pfn;
  51         u8      pxm;            // proximity domain of node
  52         u8      nid;            // which cnode contains this chunk?
  53         u8      bank;           // which mem bank on this node
  54 };
  55 static struct node_memory_chunk_s node_memory_chunk[MAXCHUNKS];
  56
  57 static int num_memory_chunks;           /* total number of memory chunks */
  58 static u8 __initdata apicid_to_pxm[MAX_APICID];
  59
  60 extern void * boot_ioremap(unsigned long, unsigned long);
  61
  62 /* Identify CPU proximity domains */
  63 static void __init parse_cpu_affinity_structure(char *p)
  64 {
  65         struct acpi_table_processor_affinity *cpu_affinity =
  66                                 (struct acpi_table_processor_affinity *) p;
  67
  68         if (!cpu_affinity->flags.enabled)
  69                 return;         /* empty entry */
  70
  71         /* mark this node as "seen" in node bitmap */
  72         BMAP_SET(pxm_bitmap, cpu_affinity->proximity_domain);
  73
  74         apicid_to_pxm[cpu_affinity->apic_id] = cpu_affinity->proximity_domain;
  75
  76         printk("CPU 0x%02X in proximity domain 0x%02X\n",
  77                 cpu_affinity->apic_id, cpu_affinity->proximity_domain);
  78 }
  79
  80 /*
  81  * Identify memory proximity domains and hot-remove capabilities.
  82  * Fill node memory chunk list structure.
  83  */
  84 static void __init parse_memory_affinity_structure (char *sratp)
  85 {
  86         unsigned long long paddr, size;
  87         unsigned long start_pfn, end_pfn;
  88         u8 pxm;
  89         struct node_memory_chunk_s *p, *q, *pend;
  90         struct acpi_table_memory_affinity *memory_affinity =
  91                         (struct acpi_table_memory_affinity *) sratp;
  92
  93         if (!memory_affinity->flags.enabled)
  94                 return;         /* empty entry */
  95
  96         /* mark this node as "seen" in node bitmap */
  97         BMAP_SET(pxm_bitmap, memory_affinity->proximity_domain);
  98
  99         /* calculate info for memory chunk structure */
 100         paddr = memory_affinity->base_addr_hi;
 101         paddr = (paddr << 32) | memory_affinity->base_addr_lo;
 102         size = memory_affinity->length_hi;
 103         size = (size << 32) | memory_affinity->length_lo;
 104
 105         start_pfn = paddr >> PAGE_SHIFT;
 106         end_pfn = (paddr + size) >> PAGE_SHIFT;
 107
 108         pxm = memory_affinity->proximity_domain;
 109
 110         if (num_memory_chunks >= MAXCHUNKS) {
 111                 printk("Too many mem chunks in SRAT. Ignoring %lld MBytes at %llx\n",
 112                         size/(1024*1024), paddr);
 113                 return;
 114         }
 115
 116         /* Insertion sort based on base address */
 117         pend = &node_memory_chunk[num_memory_chunks];
 118         for (p = &node_memory_chunk[0]; p < pend; p++) {
 119                 if (start_pfn < p->start_pfn)
 120                         break;
 121         }
 122         if (p < pend) {
 123                 for (q = pend; q >= p; q--)
 124                         *(q + 1) = *q;
 125         }
 126         p->start_pfn = start_pfn;
 127         p->end_pfn = end_pfn;
 128         p->pxm = pxm;
 129
 130         num_memory_chunks++;
 131
 132         printk("Memory range 0x%lX to 0x%lX (type 0x%X) in proximity domain 0x%02X %s\n",
 133                 start_pfn, end_pfn,
 134                 memory_affinity->memory_type,
 135                 memory_affinity->proximity_domain,
 136                 (memory_affinity->flags.hot_pluggable ?
 137                  "enabled and removable" : "enabled" ) );
 138 }
 139
 140 /*
 141  * The SRAT table always lists ascending addresses, so can always
 142  * assume that the first "start" address that you see is the real
 143  * start of the node, and that the current "end" address is after
 144  * the previous one.
 145  */
 146 static __init void node_read_chunk(int nid, struct node_memory_chunk_s *memory_chunk)
 147 {
 148         /*
 149          * Only add present memory as told by the e820.
 150          * There is no guarantee from the SRAT that the memory it
 151          * enumerates is present at boot time because it represents
 152          * *possible* memory hotplug areas the same as normal RAM.
 153          */
 154         if (memory_chunk->start_pfn >= max_pfn) {
 155                 printk (KERN_INFO "Ignoring SRAT pfns: 0x%08lx -> %08lx\n",
 156                         memory_chunk->start_pfn, memory_chunk->end_pfn);
 157                 return;
 158         }
 159         if (memory_chunk->nid != nid)
 160                 return;
 161
 162         if (!node_has_online_mem(nid))
 163                 node_start_pfn[nid] = memory_chunk->start_pfn;
 164
 165         if (node_start_pfn[nid] > memory_chunk->start_pfn)
 166                 node_start_pfn[nid] = memory_chunk->start_pfn;
 167
 168         if (node_end_pfn[nid] < memory_chunk->end_pfn)
 169                 node_end_pfn[nid] = memory_chunk->end_pfn;
 170 }
 171
 172 /* Parse the ACPI Static Resource Affinity Table */
 173 static int __init acpi20_parse_srat(struct acpi_table_srat *sratp)
 174 {
 175         u8 *start, *end, *p;
 176         int i, j, nid;
 177
 178         start = (u8 *)(&(sratp->reserved) + 1); /* skip header */
 179         p = start;
 180         end = (u8 *)sratp + sratp->header.length;
 181
 182         memset(pxm_bitmap, 0, sizeof(pxm_bitmap));      /* init proximity domain bitmap */
 183         memset(node_memory_chunk, 0, sizeof(node_memory_chunk));
 184
 185         num_memory_chunks = 0;
 186         while (p < end) {
 187                 switch (*p) {
 188                 case ACPI_SRAT_PROCESSOR_AFFINITY:
 189                         parse_cpu_affinity_structure(p);
 190                         break;
 191                 case ACPI_SRAT_MEMORY_AFFINITY:
 192                         parse_memory_affinity_structure(p);
 193                         break;
 194                 default:
 195                         printk("ACPI 2.0 SRAT: unknown entry skipped: type=0x%02X, len=%d\n", p[0], p[1]);
 196                         break;
 197                 }
 198                 p += p[1];
 199                 if (p[1] == 0) {
 200                         printk("acpi20_parse_srat: Entry length value is zero;"
 201                                 " can't parse any further!\n");
 202                         break;
 203                 }
 204         }
 205
 206         if (num_memory_chunks == 0) {
 207                 printk("could not finy any ACPI SRAT memory areas.\n");
 208                 goto out_fail;
 209         }
 210
 211         /* Calculate total number of nodes in system from PXM bitmap and create
 212          * a set of sequential node IDs starting at zero.  (ACPI doesn't seem
 213          * to specify the range of _PXM values.)
 214          */
 215         /*
 216          * MCD - we no longer HAVE to number nodes sequentially.  PXM domain
 217          * numbers could go as high as 256, and MAX_NUMNODES for i386 is typically
 218          * 32, so we will continue numbering them in this manner until MAX_NUMNODES
 219          * approaches MAX_PXM_DOMAINS for i386.
 220          */
 221         nodes_clear(node_online_map);
 222         for (i = 0; i < MAX_PXM_DOMAINS; i++) {
 223                 if (BMAP_TEST(pxm_bitmap, i)) {
 224                         int nid = acpi_map_pxm_to_node(i);
 225                         node_set_online(nid);
 226                 }
 227         }
 228         BUG_ON(num_online_nodes() == 0);
 229
 230         /* set cnode id in memory chunk structure */
 231         for (i = 0; i < num_memory_chunks; i++)
 232                 node_memory_chunk[i].nid = pxm_to_node(node_memory_chunk[i].pxm);
 233
 234         printk("pxm bitmap: ");
 235         for (i = 0; i < sizeof(pxm_bitmap); i++) {
 236                 printk("%02X ", pxm_bitmap[i]);
 237         }
 238         printk("\n");
 239         printk("Number of logical nodes in system = %d\n", num_online_nodes());
 240         printk("Number of memory chunks in system = %d\n", num_memory_chunks);
 241
 242         for (i = 0; i < MAX_APICID; i++)
 243                 apicid_2_node[i] = pxm_to_node(apicid_to_pxm[i]);
 244
 245         for (j = 0; j < num_memory_chunks; j++){
 246                 struct node_memory_chunk_s * chunk = &node_memory_chunk[j];
 247                 printk("chunk %d nid %d start_pfn %08lx end_pfn %08lx\n",
 248                        j, chunk->nid, chunk->start_pfn, chunk->end_pfn);
 249                 node_read_chunk(chunk->nid, chunk);
 250                 add_active_range(chunk->nid, chunk->start_pfn, chunk->end_pfn);
 251         }
 252
 253         for_each_online_node(nid) {
 254                 unsigned long start = node_start_pfn[nid];
 255                 unsigned long end = node_end_pfn[nid];
 256
 257                 memory_present(nid, start, end);
 258                 node_remap_size[nid] = node_memmap_size_bytes(nid, start, end);
 259         }
 260         return 1;
 261 out_fail:
 262         return 0;
 263 }
 264
 265 int __init get_memcfg_from_srat(void)
 266 {
 267         struct acpi_table_header *header = NULL;
 268         struct acpi_table_rsdp *rsdp = NULL;
 269         struct acpi_table_rsdt *rsdt = NULL;
 270         struct acpi_pointer *rsdp_address = NULL;
 271         struct acpi_table_rsdt saved_rsdt;
 272         int tables = 0;
 273         int i = 0;
 274
 275         if (ACPI_FAILURE(acpi_find_root_pointer(ACPI_PHYSICAL_ADDRESSING,
 276                                                 rsdp_address))) {
 277                 printk("%s: System description tables not found\n",
 278                        __FUNCTION__);
 279                 goto out_err;
 280         }
 281
 282         if (rsdp_address->pointer_type == ACPI_PHYSICAL_POINTER) {
 283                 printk("%s: assigning address to rsdp\n", __FUNCTION__);
 284                 rsdp = (struct acpi_table_rsdp *)
 285                                 (u32)rsdp_address->pointer.physical;
 286         } else {
 287                 printk("%s: rsdp_address is not a physical pointer\n", __FUNCTION__);
 288                 goto out_err;
 289         }
 290         if (!rsdp) {
 291                 printk("%s: Didn't find ACPI root!\n", __FUNCTION__);
 292                 goto out_err;
 293         }
 294
 295         printk(KERN_INFO "%.8s v%d [%.6s]\n", rsdp->signature, rsdp->revision,
 296                 rsdp->oem_id);
 297
 298         if (strncmp(rsdp->signature, RSDP_SIG,strlen(RSDP_SIG))) {
 299                 printk(KERN_WARNING "%s: RSDP table signature incorrect\n", __FUNCTION__);
 300                 goto out_err;
 301         }
 302
 303         rsdt = (struct acpi_table_rsdt *)
 304             boot_ioremap(rsdp->rsdt_address, sizeof(struct acpi_table_rsdt));
 305
 306         if (!rsdt) {
 307                 printk(KERN_WARNING
 308                        "%s: ACPI: Invalid root system description tables (RSDT)\n",
 309                        __FUNCTION__);
 310                 goto out_err;
 311         }
 312
 313         header = & rsdt->header;
 314
 315         if (strncmp(header->signature, RSDT_SIG, strlen(RSDT_SIG))) {
 316                 printk(KERN_WARNING "ACPI: RSDT signature incorrect\n");
 317                 goto out_err;
 318         }
 319
 320         /*
 321          * The number of tables is computed by taking the
 322          * size of all entries (header size minus total
 323          * size of RSDT) divided by the size of each entry
 324          * (4-byte table pointers).
 325          */
 326         tables = (header->length - sizeof(struct acpi_table_header)) / 4;
 327
 328         if (!tables)
 329                 goto out_err;
 330
 331         memcpy(&saved_rsdt, rsdt, sizeof(saved_rsdt));
 332
 333         if (saved_rsdt.header.length > sizeof(saved_rsdt)) {
 334                 printk(KERN_WARNING "ACPI: Too big length in RSDT: %d\n",
 335                        saved_rsdt.header.length);
 336                 goto out_err;
 337         }
 338
 339         printk("Begin SRAT table scan....\n");
 340
 341         for (i = 0; i < tables; i++) {
 342                 /* Map in header, then map in full table length. */
 343                 header = (struct acpi_table_header *)
 344                         boot_ioremap(saved_rsdt.entry[i], sizeof(struct acpi_table_header));
 345                 if (!header)
 346                         break;
 347                 header = (struct acpi_table_header *)
 348                         boot_ioremap(saved_rsdt.entry[i], header->length);
 349                 if (!header)
 350                         break;
 351
 352                 if (strncmp((char *) &header->signature, "SRAT", 4))
 353                         continue;
 354
 355                 /* we've found the srat table. don't need to look at any more tables */
 356                 return acpi20_parse_srat((struct acpi_table_srat *)header);
 357         }
 358 out_err:
 359         remove_all_active_ranges();
 360         printk("failed to get NUMA memory information from SRAT table\n");
 361         return 0;
 362 }