linux/arch/i386/kernel/srat.c
<<
>>
Prefs
   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/config.h>
  27#include <linux/mm.h>
  28#include <linux/bootmem.h>
  29#include <linux/mmzone.h>
  30#include <linux/acpi.h>
  31#include <linux/nodemask.h>
  32#include <asm/srat.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#define MAX_PXM_DOMAINS         256     /* 1 byte and no promises about values */
  42/* bitmap length; _PXM is at most 255 */
  43#define PXM_BITMAP_LEN (MAX_PXM_DOMAINS / 8) 
  44static u8 pxm_bitmap[PXM_BITMAP_LEN];   /* bitmap of proximity domains */
  45
  46#define MAX_CHUNKS_PER_NODE     4
  47#define MAXCHUNKS               (MAX_CHUNKS_PER_NODE * MAX_NUMNODES)
  48struct 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};
  55static struct node_memory_chunk_s node_memory_chunk[MAXCHUNKS];
  56
  57static int num_memory_chunks;           /* total number of memory chunks */
  58static int zholes_size_init;
  59static unsigned long zholes_size[MAX_NUMNODES * MAX_NR_ZONES];
  60
  61extern unsigned long node_start_pfn[], node_end_pfn[];
  62
  63extern void * boot_ioremap(unsigned long, unsigned long);
  64
  65/* Identify CPU proximity domains */
  66static void __init parse_cpu_affinity_structure(char *p)
  67{
  68        struct acpi_table_processor_affinity *cpu_affinity = 
  69                                (struct acpi_table_processor_affinity *) p;
  70
  71        if (!cpu_affinity->flags.enabled)
  72                return;         /* empty entry */
  73
  74        /* mark this node as "seen" in node bitmap */
  75        BMAP_SET(pxm_bitmap, cpu_affinity->proximity_domain);
  76
  77        printk("CPU 0x%02X in proximity domain 0x%02X\n",
  78                cpu_affinity->apic_id, cpu_affinity->proximity_domain);
  79}
  80
  81/*
  82 * Identify memory proximity domains and hot-remove capabilities.
  83 * Fill node memory chunk list structure.
  84 */
  85static void __init parse_memory_affinity_structure (char *sratp)
  86{
  87        unsigned long long paddr, size;
  88        unsigned long start_pfn, end_pfn; 
  89        u8 pxm;
  90        struct node_memory_chunk_s *p, *q, *pend;
  91        struct acpi_table_memory_affinity *memory_affinity =
  92                        (struct acpi_table_memory_affinity *) sratp;
  93
  94        if (!memory_affinity->flags.enabled)
  95                return;         /* empty entry */
  96
  97        /* mark this node as "seen" in node bitmap */
  98        BMAP_SET(pxm_bitmap, memory_affinity->proximity_domain);
  99
 100        /* calculate info for memory chunk structure */
 101        paddr = memory_affinity->base_addr_hi;
 102        paddr = (paddr << 32) | memory_affinity->base_addr_lo;
 103        size = memory_affinity->length_hi;
 104        size = (size << 32) | memory_affinity->length_lo;
 105        
 106        start_pfn = paddr >> PAGE_SHIFT;
 107        end_pfn = (paddr + size) >> PAGE_SHIFT;
 108        
 109        pxm = memory_affinity->proximity_domain;
 110
 111        if (num_memory_chunks >= MAXCHUNKS) {
 112                printk("Too many mem chunks in SRAT. Ignoring %lld MBytes at %llx\n",
 113                        size/(1024*1024), paddr);
 114                return;
 115        }
 116
 117        /* Insertion sort based on base address */
 118        pend = &node_memory_chunk[num_memory_chunks];
 119        for (p = &node_memory_chunk[0]; p < pend; p++) {
 120                if (start_pfn < p->start_pfn)
 121                        break;
 122        }
 123        if (p < pend) {
 124                for (q = pend; q >= p; q--)
 125                        *(q + 1) = *q;
 126        }
 127        p->start_pfn = start_pfn;
 128        p->end_pfn = end_pfn;
 129        p->pxm = pxm;
 130
 131        num_memory_chunks++;
 132
 133        printk("Memory range 0x%lX to 0x%lX (type 0x%X) in proximity domain 0x%02X %s\n",
 134                start_pfn, end_pfn,
 135                memory_affinity->memory_type,
 136                memory_affinity->proximity_domain,
 137                (memory_affinity->flags.hot_pluggable ?
 138                 "enabled and removable" : "enabled" ) );
 139}
 140
 141#if MAX_NR_ZONES != 3
 142#error "MAX_NR_ZONES != 3, chunk_to_zone requires review"
 143#endif
 144/* Take a chunk of pages from page frame cstart to cend and count the number
 145 * of pages in each zone, returned via zones[].
 146 */
 147static __init void chunk_to_zones(unsigned long cstart, unsigned long cend, 
 148                unsigned long *zones)
 149{
 150        unsigned long max_dma;
 151        extern unsigned long max_low_pfn;
 152
 153        int z;
 154        unsigned long rend;
 155
 156        /* FIXME: MAX_DMA_ADDRESS and max_low_pfn are trying to provide
 157         * similarly scoped information and should be handled in a consistant
 158         * manner.
 159         */
 160        max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 161
 162        /* Split the hole into the zones in which it falls.  Repeatedly
 163         * take the segment in which the remaining hole starts, round it
 164         * to the end of that zone.
 165         */
 166        memset(zones, 0, MAX_NR_ZONES * sizeof(long));
 167        while (cstart < cend) {
 168                if (cstart < max_dma) {
 169                        z = ZONE_DMA;
 170                        rend = (cend < max_dma)? cend : max_dma;
 171
 172                } else if (cstart < max_low_pfn) {
 173                        z = ZONE_NORMAL;
 174                        rend = (cend < max_low_pfn)? cend : max_low_pfn;
 175
 176                } else {
 177                        z = ZONE_HIGHMEM;
 178                        rend = cend;
 179                }
 180                zones[z] += rend - cstart;
 181                cstart = rend;
 182        }
 183}
 184
 185/* Parse the ACPI Static Resource Affinity Table */
 186static int __init acpi20_parse_srat(struct acpi_table_srat *sratp)
 187{
 188        u8 *start, *end, *p;
 189        int i, j, nid;
 190        u8 pxm_to_nid_map[MAX_PXM_DOMAINS];/* _PXM to logical node ID map */
 191        u8 nid_to_pxm_map[MAX_NUMNODES];/* logical node ID to _PXM map */
 192
 193        start = (u8 *)(&(sratp->reserved) + 1); /* skip header */
 194        p = start;
 195        end = (u8 *)sratp + sratp->header.length;
 196
 197        memset(pxm_bitmap, 0, sizeof(pxm_bitmap));      /* init proximity domain bitmap */
 198        memset(node_memory_chunk, 0, sizeof(node_memory_chunk));
 199        memset(zholes_size, 0, sizeof(zholes_size));
 200
 201        /* -1 in these maps means not available */
 202        memset(pxm_to_nid_map, -1, sizeof(pxm_to_nid_map));
 203        memset(nid_to_pxm_map, -1, sizeof(nid_to_pxm_map));
 204
 205        num_memory_chunks = 0;
 206        while (p < end) {
 207                switch (*p) {
 208                case ACPI_SRAT_PROCESSOR_AFFINITY:
 209                        parse_cpu_affinity_structure(p);
 210                        break;
 211                case ACPI_SRAT_MEMORY_AFFINITY:
 212                        parse_memory_affinity_structure(p);
 213                        break;
 214                default:
 215                        printk("ACPI 2.0 SRAT: unknown entry skipped: type=0x%02X, len=%d\n", p[0], p[1]);
 216                        break;
 217                }
 218                p += p[1];
 219                if (p[1] == 0) {
 220                        printk("acpi20_parse_srat: Entry length value is zero;"
 221                                " can't parse any further!\n");
 222                        break;
 223                }
 224        }
 225
 226        if (num_memory_chunks == 0) {
 227                printk("could not finy any ACPI SRAT memory areas.\n");
 228                goto out_fail;
 229        }
 230
 231        /* Calculate total number of nodes in system from PXM bitmap and create
 232         * a set of sequential node IDs starting at zero.  (ACPI doesn't seem
 233         * to specify the range of _PXM values.)
 234         */
 235        /*
 236         * MCD - we no longer HAVE to number nodes sequentially.  PXM domain
 237         * numbers could go as high as 256, and MAX_NUMNODES for i386 is typically
 238         * 32, so we will continue numbering them in this manner until MAX_NUMNODES
 239         * approaches MAX_PXM_DOMAINS for i386.
 240         */
 241        nodes_clear(node_online_map);
 242        for (i = 0; i < MAX_PXM_DOMAINS; i++) {
 243                if (BMAP_TEST(pxm_bitmap, i)) {
 244                        nid = num_online_nodes();
 245                        pxm_to_nid_map[i] = nid;
 246                        nid_to_pxm_map[nid] = i;
 247                        node_set_online(nid);
 248                }
 249        }
 250        BUG_ON(num_online_nodes() == 0);
 251
 252        /* set cnode id in memory chunk structure */
 253        for (i = 0; i < num_memory_chunks; i++)
 254                node_memory_chunk[i].nid = pxm_to_nid_map[node_memory_chunk[i].pxm];
 255
 256        printk("pxm bitmap: ");
 257        for (i = 0; i < sizeof(pxm_bitmap); i++) {
 258                printk("%02X ", pxm_bitmap[i]);
 259        }
 260        printk("\n");
 261        printk("Number of logical nodes in system = %d\n", num_online_nodes());
 262        printk("Number of memory chunks in system = %d\n", num_memory_chunks);
 263
 264        for (j = 0; j < num_memory_chunks; j++){
 265                printk("chunk %d nid %d start_pfn %08lx end_pfn %08lx\n",
 266                       j, node_memory_chunk[j].nid,
 267                       node_memory_chunk[j].start_pfn,
 268                       node_memory_chunk[j].end_pfn);
 269        }
 270 
 271        /*calculate node_start_pfn/node_end_pfn arrays*/
 272        for_each_online_node(nid) {
 273                int been_here_before = 0;
 274
 275                for (j = 0; j < num_memory_chunks; j++){
 276                        if (node_memory_chunk[j].nid == nid) {
 277                                if (been_here_before == 0) {
 278                                        node_start_pfn[nid] = node_memory_chunk[j].start_pfn;
 279                                        node_end_pfn[nid] = node_memory_chunk[j].end_pfn;
 280                                        been_here_before = 1;
 281                                } else { /* We've found another chunk of memory for the node */
 282                                        if (node_start_pfn[nid] < node_memory_chunk[j].start_pfn) {
 283                                                node_end_pfn[nid] = node_memory_chunk[j].end_pfn;
 284                                        }
 285                                }
 286                        }
 287                }
 288        }
 289        return 1;
 290out_fail:
 291        return 0;
 292}
 293
 294int __init get_memcfg_from_srat(void)
 295{
 296        struct acpi_table_header *header = NULL;
 297        struct acpi_table_rsdp *rsdp = NULL;
 298        struct acpi_table_rsdt *rsdt = NULL;
 299        struct acpi_pointer *rsdp_address = NULL;
 300        struct acpi_table_rsdt saved_rsdt;
 301        int tables = 0;
 302        int i = 0;
 303
 304        acpi_find_root_pointer(ACPI_PHYSICAL_ADDRESSING, rsdp_address);
 305
 306        if (rsdp_address->pointer_type == ACPI_PHYSICAL_POINTER) {
 307                printk("%s: assigning address to rsdp\n", __FUNCTION__);
 308                rsdp = (struct acpi_table_rsdp *)
 309                                (u32)rsdp_address->pointer.physical;
 310        } else {
 311                printk("%s: rsdp_address is not a physical pointer\n", __FUNCTION__);
 312                goto out_err;
 313        }
 314        if (!rsdp) {
 315                printk("%s: Didn't find ACPI root!\n", __FUNCTION__);
 316                goto out_err;
 317        }
 318
 319        printk(KERN_INFO "%.8s v%d [%.6s]\n", rsdp->signature, rsdp->revision,
 320                rsdp->oem_id);
 321
 322        if (strncmp(rsdp->signature, RSDP_SIG,strlen(RSDP_SIG))) {
 323                printk(KERN_WARNING "%s: RSDP table signature incorrect\n", __FUNCTION__);
 324                goto out_err;
 325        }
 326
 327        rsdt = (struct acpi_table_rsdt *)
 328            boot_ioremap(rsdp->rsdt_address, sizeof(struct acpi_table_rsdt));
 329
 330        if (!rsdt) {
 331                printk(KERN_WARNING
 332                       "%s: ACPI: Invalid root system description tables (RSDT)\n",
 333                       __FUNCTION__);
 334                goto out_err;
 335        }
 336
 337        header = & rsdt->header;
 338
 339        if (strncmp(header->signature, RSDT_SIG, strlen(RSDT_SIG))) {
 340                printk(KERN_WARNING "ACPI: RSDT signature incorrect\n");
 341                goto out_err;
 342        }
 343
 344        /* 
 345         * The number of tables is computed by taking the 
 346         * size of all entries (header size minus total 
 347         * size of RSDT) divided by the size of each entry
 348         * (4-byte table pointers).
 349         */
 350        tables = (header->length - sizeof(struct acpi_table_header)) / 4;
 351
 352        if (!tables)
 353                goto out_err;
 354
 355        memcpy(&saved_rsdt, rsdt, sizeof(saved_rsdt));
 356
 357        if (saved_rsdt.header.length > sizeof(saved_rsdt)) {
 358                printk(KERN_WARNING "ACPI: Too big length in RSDT: %d\n",
 359                       saved_rsdt.header.length);
 360                goto out_err;
 361        }
 362
 363        printk("Begin SRAT table scan....\n");
 364
 365        for (i = 0; i < tables; i++) {
 366                /* Map in header, then map in full table length. */
 367                header = (struct acpi_table_header *)
 368                        boot_ioremap(saved_rsdt.entry[i], sizeof(struct acpi_table_header));
 369                if (!header)
 370                        break;
 371                header = (struct acpi_table_header *)
 372                        boot_ioremap(saved_rsdt.entry[i], header->length);
 373                if (!header)
 374                        break;
 375
 376                if (strncmp((char *) &header->signature, "SRAT", 4))
 377                        continue;
 378
 379                /* we've found the srat table. don't need to look at any more tables */
 380                return acpi20_parse_srat((struct acpi_table_srat *)header);
 381        }
 382out_err:
 383        printk("failed to get NUMA memory information from SRAT table\n");
 384        return 0;
 385}
 386
 387/* For each node run the memory list to determine whether there are
 388 * any memory holes.  For each hole determine which ZONE they fall
 389 * into.
 390 *
 391 * NOTE#1: this requires knowledge of the zone boundries and so
 392 * _cannot_ be performed before those are calculated in setup_memory.
 393 * 
 394 * NOTE#2: we rely on the fact that the memory chunks are ordered by
 395 * start pfn number during setup.
 396 */
 397static void __init get_zholes_init(void)
 398{
 399        int nid;
 400        int c;
 401        int first;
 402        unsigned long end = 0;
 403
 404        for_each_online_node(nid) {
 405                first = 1;
 406                for (c = 0; c < num_memory_chunks; c++){
 407                        if (node_memory_chunk[c].nid == nid) {
 408                                if (first) {
 409                                        end = node_memory_chunk[c].end_pfn;
 410                                        first = 0;
 411
 412                                } else {
 413                                        /* Record any gap between this chunk
 414                                         * and the previous chunk on this node
 415                                         * against the zones it spans.
 416                                         */
 417                                        chunk_to_zones(end,
 418                                                node_memory_chunk[c].start_pfn,
 419                                                &zholes_size[nid * MAX_NR_ZONES]);
 420                                }
 421                        }
 422                }
 423        }
 424}
 425
 426unsigned long * __init get_zholes_size(int nid)
 427{
 428        if (!zholes_size_init) {
 429                zholes_size_init++;
 430                get_zholes_init();
 431        }
 432        if (nid >= MAX_NUMNODES || !node_online(nid))
 433                printk("%s: nid = %d is invalid/offline. num_online_nodes = %d",
 434                       __FUNCTION__, nid, num_online_nodes());
 435        return &zholes_size[nid * MAX_NR_ZONES];
 436}
 437
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.