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
  34 /*
  35  * proximity macros and definitions
  36  */
  37 #define NODE_ARRAY_INDEX(x)     ((x) / 8)       /* 8 bits/char */
  38 #define NODE_ARRAY_OFFSET(x)    ((x) % 8)       /* 8 bits/char */
  39 #define BMAP_SET(bmap, bit)     ((bmap)[NODE_ARRAY_INDEX(bit)] |= 1 << NODE_ARRAY_OFFSET(bit))
  40 #define BMAP_TEST(bmap, bit)    ((bmap)[NODE_ARRAY_INDEX(bit)] & (1 << NODE_ARRAY_OFFSET(bit)))
  41 /* bitmap length; _PXM is at most 255 */
  42 #define PXM_BITMAP_LEN (MAX_PXM_DOMAINS / 8)
  43 static u8 pxm_bitmap[PXM_BITMAP_LEN];   /* bitmap of proximity domains */
  44
  45 #define MAX_CHUNKS_PER_NODE     4
  46 #define MAXCHUNKS               (MAX_CHUNKS_PER_NODE * MAX_NUMNODES)
  47 struct node_memory_chunk_s {
  48         unsigned long   start_pfn;
  49         unsigned long   end_pfn;
  50         u8      pxm;            // proximity domain of node
  51         u8      nid;            // which cnode contains this chunk?
  52         u8      bank;           // which mem bank on this node
  53 };
  54 static struct node_memory_chunk_s node_memory_chunk[MAXCHUNKS];
  55
  56 static int num_memory_chunks;           /* total number of memory chunks */
  57 static int zholes_size_init;
  58 static unsigned long zholes_size[MAX_NUMNODES * MAX_NR_ZONES];
  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         printk("CPU 0x%02X in proximity domain 0x%02X\n",
  75                 cpu_affinity->apic_id, cpu_affinity->proximity_domain);
  76 }
  77
  78 /*
  79  * Identify memory proximity domains and hot-remove capabilities.
  80  * Fill node memory chunk list structure.
  81  */
  82 static void __init parse_memory_affinity_structure (char *sratp)
  83 {
  84         unsigned long long paddr, size;
  85         unsigned long start_pfn, end_pfn;
  86         u8 pxm;
  87         struct node_memory_chunk_s *p, *q, *pend;
  88         struct acpi_table_memory_affinity *memory_affinity =
  89                         (struct acpi_table_memory_affinity *) sratp;
  90
  91         if (!memory_affinity->flags.enabled)
  92                 return;         /* empty entry */
  93
  94         /* mark this node as "seen" in node bitmap */
  95         BMAP_SET(pxm_bitmap, memory_affinity->proximity_domain);
  96
  97         /* calculate info for memory chunk structure */
  98         paddr = memory_affinity->base_addr_hi;
  99         paddr = (paddr << 32) | memory_affinity->base_addr_lo;
 100         size = memory_affinity->length_hi;
 101         size = (size << 32) | memory_affinity->length_lo;
 102
 103         start_pfn = paddr >> PAGE_SHIFT;
 104         end_pfn = (paddr + size) >> PAGE_SHIFT;
 105
 106         pxm = memory_affinity->proximity_domain;
 107
 108         if (num_memory_chunks >= MAXCHUNKS) {
 109                 printk("Too many mem chunks in SRAT. Ignoring %lld MBytes at %llx\n",
 110                         size/(1024*1024), paddr);
 111                 return;
 112         }
 113
 114         /* Insertion sort based on base address */
 115         pend = &node_memory_chunk[num_memory_chunks];
 116         for (p = &node_memory_chunk[0]; p < pend; p++) {
 117                 if (start_pfn < p->start_pfn)
 118                         break;
 119         }
 120         if (p < pend) {
 121                 for (q = pend; q >= p; q--)
 122                         *(q + 1) = *q;
 123         }
 124         p->start_pfn = start_pfn;
 125         p->end_pfn = end_pfn;
 126         p->pxm = pxm;
 127
 128         num_memory_chunks++;
 129
 130         printk("Memory range 0x%lX to 0x%lX (type 0x%X) in proximity domain 0x%02X %s\n",
 131                 start_pfn, end_pfn,
 132                 memory_affinity->memory_type,
 133                 memory_affinity->proximity_domain,
 134                 (memory_affinity->flags.hot_pluggable ?
 135                  "enabled and removable" : "enabled" ) );
 136 }
 137
 138 #if MAX_NR_ZONES != 4
 139 #error "MAX_NR_ZONES != 4, chunk_to_zone requires review"
 140 #endif
 141 /* Take a chunk of pages from page frame cstart to cend and count the number
 142  * of pages in each zone, returned via zones[].
 143  */
 144 static __init void chunk_to_zones(unsigned long cstart, unsigned long cend,
 145                 unsigned long *zones)
 146 {
 147         unsigned long max_dma;
 148         extern unsigned long max_low_pfn;
 149
 150         int z;
 151         unsigned long rend;
 152
 153         /* FIXME: MAX_DMA_ADDRESS and max_low_pfn are trying to provide
 154          * similarly scoped information and should be handled in a consistant
 155          * manner.
 156          */
 157         max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 158
 159         /* Split the hole into the zones in which it falls.  Repeatedly
 160          * take the segment in which the remaining hole starts, round it
 161          * to the end of that zone.
 162          */
 163         memset(zones, 0, MAX_NR_ZONES * sizeof(long));
 164         while (cstart < cend) {
 165                 if (cstart < max_dma) {
 166                         z = ZONE_DMA;
 167                         rend = (cend < max_dma)? cend : max_dma;
 168
 169                 } else if (cstart < max_low_pfn) {
 170                         z = ZONE_NORMAL;
 171                         rend = (cend < max_low_pfn)? cend : max_low_pfn;
 172
 173                 } else {
 174                         z = ZONE_HIGHMEM;
 175                         rend = cend;
 176                 }
 177                 zones[z] += rend - cstart;
 178                 cstart = rend;
 179         }
 180 }
 181
 182 /*
 183  * The SRAT table always lists ascending addresses, so can always
 184  * assume that the first "start" address that you see is the real
 185  * start of the node, and that the current "end" address is after
 186  * the previous one.
 187  */
 188 static __init void node_read_chunk(int nid, struct node_memory_chunk_s *memory_chunk)
 189 {
 190         /*
 191          * Only add present memory as told by the e820.
 192          * There is no guarantee from the SRAT that the memory it
 193          * enumerates is present at boot time because it represents
 194          * *possible* memory hotplug areas the same as normal RAM.
 195          */
 196         if (memory_chunk->start_pfn >= max_pfn) {
 197                 printk (KERN_INFO "Ignoring SRAT pfns: 0x%08lx -> %08lx\n",
 198                         memory_chunk->start_pfn, memory_chunk->end_pfn);
 199                 return;
 200         }
 201         if (memory_chunk->nid != nid)
 202                 return;
 203
 204         if (!node_has_online_mem(nid))
 205                 node_start_pfn[nid] = memory_chunk->start_pfn;
 206
 207         if (node_start_pfn[nid] > memory_chunk->start_pfn)
 208                 node_start_pfn[nid] = memory_chunk->start_pfn;
 209
 210         if (node_end_pfn[nid] < memory_chunk->end_pfn)
 211                 node_end_pfn[nid] = memory_chunk->end_pfn;
 212 }
 213
 214 /* Parse the ACPI Static Resource Affinity Table */
 215 static int __init acpi20_parse_srat(struct acpi_table_srat *sratp)
 216 {
 217         u8 *start, *end, *p;
 218         int i, j, nid;
 219
 220         start = (u8 *)(&(sratp->reserved) + 1); /* skip header */
 221         p = start;
 222         end = (u8 *)sratp + sratp->header.length;
 223
 224         memset(pxm_bitmap, 0, sizeof(pxm_bitmap));      /* init proximity domain bitmap */
 225         memset(node_memory_chunk, 0, sizeof(node_memory_chunk));
 226         memset(zholes_size, 0, sizeof(zholes_size));
 227
 228         num_memory_chunks = 0;
 229         while (p < end) {
 230                 switch (*p) {
 231                 case ACPI_SRAT_PROCESSOR_AFFINITY:
 232                         parse_cpu_affinity_structure(p);
 233                         break;
 234                 case ACPI_SRAT_MEMORY_AFFINITY:
 235                         parse_memory_affinity_structure(p);
 236                         break;
 237                 default:
 238                         printk("ACPI 2.0 SRAT: unknown entry skipped: type=0x%02X, len=%d\n", p[0], p[1]);
 239                         break;
 240                 }
 241                 p += p[1];
 242                 if (p[1] == 0) {
 243                         printk("acpi20_parse_srat: Entry length value is zero;"
 244                                 " can't parse any further!\n");
 245                         break;
 246                 }
 247         }
 248
 249         if (num_memory_chunks == 0) {
 250                 printk("could not finy any ACPI SRAT memory areas.\n");
 251                 goto out_fail;
 252         }
 253
 254         /* Calculate total number of nodes in system from PXM bitmap and create
 255          * a set of sequential node IDs starting at zero.  (ACPI doesn't seem
 256          * to specify the range of _PXM values.)
 257          */
 258         /*
 259          * MCD - we no longer HAVE to number nodes sequentially.  PXM domain
 260          * numbers could go as high as 256, and MAX_NUMNODES for i386 is typically
 261          * 32, so we will continue numbering them in this manner until MAX_NUMNODES
 262          * approaches MAX_PXM_DOMAINS for i386.
 263          */
 264         nodes_clear(node_online_map);
 265         for (i = 0; i < MAX_PXM_DOMAINS; i++) {
 266                 if (BMAP_TEST(pxm_bitmap, i)) {
 267                         int nid = acpi_map_pxm_to_node(i);
 268                         node_set_online(nid);
 269                 }
 270         }
 271         BUG_ON(num_online_nodes() == 0);
 272
 273         /* set cnode id in memory chunk structure */
 274         for (i = 0; i < num_memory_chunks; i++)
 275                 node_memory_chunk[i].nid = pxm_to_node(node_memory_chunk[i].pxm);
 276
 277         printk("pxm bitmap: ");
 278         for (i = 0; i < sizeof(pxm_bitmap); i++) {
 279                 printk("%02X ", pxm_bitmap[i]);
 280         }
 281         printk("\n");
 282         printk("Number of logical nodes in system = %d\n", num_online_nodes());
 283         printk("Number of memory chunks in system = %d\n", num_memory_chunks);
 284
 285         for (j = 0; j < num_memory_chunks; j++){
 286                 struct node_memory_chunk_s * chunk = &node_memory_chunk[j];
 287                 printk("chunk %d nid %d start_pfn %08lx end_pfn %08lx\n",
 288                        j, chunk->nid, chunk->start_pfn, chunk->end_pfn);
 289                 node_read_chunk(chunk->nid, chunk);
 290         }
 291
 292         for_each_online_node(nid) {
 293                 unsigned long start = node_start_pfn[nid];
 294                 unsigned long end = node_end_pfn[nid];
 295
 296                 memory_present(nid, start, end);
 297                 node_remap_size[nid] = node_memmap_size_bytes(nid, start, end);
 298         }
 299         return 1;
 300 out_fail:
 301         return 0;
 302 }
 303
 304 int __init get_memcfg_from_srat(void)
 305 {
 306         struct acpi_table_header *header = NULL;
 307         struct acpi_table_rsdp *rsdp = NULL;
 308         struct acpi_table_rsdt *rsdt = NULL;
 309         struct acpi_pointer *rsdp_address = NULL;
 310         struct acpi_table_rsdt saved_rsdt;
 311         int tables = 0;
 312         int i = 0;
 313
 314         if (ACPI_FAILURE(acpi_find_root_pointer(ACPI_PHYSICAL_ADDRESSING,
 315                                                 rsdp_address))) {
 316                 printk("%s: System description tables not found\n",
 317                        __FUNCTION__);
 318                 goto out_err;
 319         }
 320
 321         if (rsdp_address->pointer_type == ACPI_PHYSICAL_POINTER) {
 322                 printk("%s: assigning address to rsdp\n", __FUNCTION__);
 323                 rsdp = (struct acpi_table_rsdp *)
 324                                 (u32)rsdp_address->pointer.physical;
 325         } else {
 326                 printk("%s: rsdp_address is not a physical pointer\n", __FUNCTION__);
 327                 goto out_err;
 328         }
 329         if (!rsdp) {
 330                 printk("%s: Didn't find ACPI root!\n", __FUNCTION__);
 331                 goto out_err;
 332         }
 333
 334         printk(KERN_INFO "%.8s v%d [%.6s]\n", rsdp->signature, rsdp->revision,
 335                 rsdp->oem_id);
 336
 337         if (strncmp(rsdp->signature, RSDP_SIG,strlen(RSDP_SIG))) {
 338                 printk(KERN_WARNING "%s: RSDP table signature incorrect\n", __FUNCTION__);
 339                 goto out_err;
 340         }
 341
 342         rsdt = (struct acpi_table_rsdt *)
 343             boot_ioremap(rsdp->rsdt_address, sizeof(struct acpi_table_rsdt));
 344
 345         if (!rsdt) {
 346                 printk(KERN_WARNING
 347                        "%s: ACPI: Invalid root system description tables (RSDT)\n",
 348                        __FUNCTION__);
 349                 goto out_err;
 350         }
 351
 352         header = & rsdt->header;
 353
 354         if (strncmp(header->signature, RSDT_SIG, strlen(RSDT_SIG))) {
 355                 printk(KERN_WARNING "ACPI: RSDT signature incorrect\n");
 356                 goto out_err;
 357         }
 358
 359         /*
 360          * The number of tables is computed by taking the
 361          * size of all entries (header size minus total
 362          * size of RSDT) divided by the size of each entry
 363          * (4-byte table pointers).
 364          */
 365         tables = (header->length - sizeof(struct acpi_table_header)) / 4;
 366
 367         if (!tables)
 368                 goto out_err;
 369
 370         memcpy(&saved_rsdt, rsdt, sizeof(saved_rsdt));
 371
 372         if (saved_rsdt.header.length > sizeof(saved_rsdt)) {
 373                 printk(KERN_WARNING "ACPI: Too big length in RSDT: %d\n",
 374                        saved_rsdt.header.length);
 375                 goto out_err;
 376         }
 377
 378         printk("Begin SRAT table scan....\n");
 379
 380         for (i = 0; i < tables; i++) {
 381                 /* Map in header, then map in full table length. */
 382                 header = (struct acpi_table_header *)
 383                         boot_ioremap(saved_rsdt.entry[i], sizeof(struct acpi_table_header));
 384                 if (!header)
 385                         break;
 386                 header = (struct acpi_table_header *)
 387                         boot_ioremap(saved_rsdt.entry[i], header->length);
 388                 if (!header)
 389                         break;
 390
 391                 if (strncmp((char *) &header->signature, "SRAT", 4))
 392                         continue;
 393
 394                 /* we've found the srat table. don't need to look at any more tables */
 395                 return acpi20_parse_srat((struct acpi_table_srat *)header);
 396         }
 397 out_err:
 398         printk("failed to get NUMA memory information from SRAT table\n");
 399         return 0;
 400 }
 401
 402 /* For each node run the memory list to determine whether there are
 403  * any memory holes.  For each hole determine which ZONE they fall
 404  * into.
 405  *
 406  * NOTE#1: this requires knowledge of the zone boundries and so
 407  * _cannot_ be performed before those are calculated in setup_memory.
 408  *
 409  * NOTE#2: we rely on the fact that the memory chunks are ordered by
 410  * start pfn number during setup.
 411  */
 412 static void __init get_zholes_init(void)
 413 {
 414         int nid;
 415         int c;
 416         int first;
 417         unsigned long end = 0;
 418
 419         for_each_online_node(nid) {
 420                 first = 1;
 421                 for (c = 0; c < num_memory_chunks; c++){
 422                         if (node_memory_chunk[c].nid == nid) {
 423                                 if (first) {
 424                                         end = node_memory_chunk[c].end_pfn;
 425                                         first = 0;
 426
 427                                 } else {
 428                                         /* Record any gap between this chunk
 429                                          * and the previous chunk on this node
 430                                          * against the zones it spans.
 431                                          */
 432                                         chunk_to_zones(end,
 433                                                 node_memory_chunk[c].start_pfn,
 434                                                 &zholes_size[nid * MAX_NR_ZONES]);
 435                                 }
 436                         }
 437                 }
 438         }
 439 }
 440
 441 unsigned long * __init get_zholes_size(int nid)
 442 {
 443         if (!zholes_size_init) {
 444                 zholes_size_init++;
 445                 get_zholes_init();
 446         }
 447         if (nid >= MAX_NUMNODES || !node_online(nid))
 448                 printk("%s: nid = %d is invalid/offline. num_online_nodes = %d",
 449                        __FUNCTION__, nid, num_online_nodes());
 450         return &zholes_size[nid * MAX_NR_ZONES];
 451 }