arch/mips/sgi-ip27/ip27-memory.c

   1 /*
   2  * This file is subject to the terms and conditions of the GNU General Public
   3  * License.  See the file "COPYING" in the main directory of this archive
   4  * for more details.
   5  *
   6  * Copyright (C) 2000, 05 by Ralf Baechle (ralf@linux-mips.org)
   7  * Copyright (C) 2000 by Silicon Graphics, Inc.
   8  * Copyright (C) 2004 by Christoph Hellwig
   9  *
  10  * On SGI IP27 the ARC memory configuration data is completly bogus but
  11  * alternate easier to use mechanisms are available.
  12  */
  13 #include <linux/init.h>
  14 #include <linux/kernel.h>
  15 #include <linux/mm.h>
  16 #include <linux/mmzone.h>
  17 #include <linux/module.h>
  18 #include <linux/nodemask.h>
  19 #include <linux/swap.h>
  20 #include <linux/bootmem.h>
  21 #include <linux/pfn.h>
  22 #include <linux/highmem.h>
  23 #include <asm/page.h>
  24 #include <asm/sections.h>
  25
  26 #include <asm/sn/arch.h>
  27 #include <asm/sn/hub.h>
  28 #include <asm/sn/klconfig.h>
  29 #include <asm/sn/sn_private.h>
  30
  31
  32 #define SLOT_PFNSHIFT           (SLOT_SHIFT - PAGE_SHIFT)
  33 #define PFN_NASIDSHFT           (NASID_SHFT - PAGE_SHIFT)
  34
  35 #define SLOT_IGNORED            0xffff
  36
  37 static short __initdata slot_lastfilled_cache[MAX_COMPACT_NODES];
  38 static unsigned short __initdata slot_psize_cache[MAX_COMPACT_NODES][MAX_MEM_SLOTS];
  39 static struct bootmem_data __initdata plat_node_bdata[MAX_COMPACT_NODES];
  40
  41 struct node_data *__node_data[MAX_COMPACT_NODES];
  42
  43 EXPORT_SYMBOL(__node_data);
  44
  45 static int fine_mode;
  46
  47 static int is_fine_dirmode(void)
  48 {
  49         return (((LOCAL_HUB_L(NI_STATUS_REV_ID) & NSRI_REGIONSIZE_MASK)
  50                 >> NSRI_REGIONSIZE_SHFT) & REGIONSIZE_FINE);
  51 }
  52
  53 static hubreg_t get_region(cnodeid_t cnode)
  54 {
  55         if (fine_mode)
  56                 return COMPACT_TO_NASID_NODEID(cnode) >> NASID_TO_FINEREG_SHFT;
  57         else
  58                 return COMPACT_TO_NASID_NODEID(cnode) >> NASID_TO_COARSEREG_SHFT;
  59 }
  60
  61 static hubreg_t region_mask;
  62
  63 static void gen_region_mask(hubreg_t *region_mask)
  64 {
  65         cnodeid_t cnode;
  66
  67         (*region_mask) = 0;
  68         for_each_online_node(cnode) {
  69                 (*region_mask) |= 1ULL << get_region(cnode);
  70         }
  71 }
  72
  73 #define rou_rflag       rou_flags
  74
  75 static int router_distance;
  76
  77 static void router_recurse(klrou_t *router_a, klrou_t *router_b, int depth)
  78 {
  79         klrou_t *router;
  80         lboard_t *brd;
  81         int     port;
  82
  83         if (router_a->rou_rflag == 1)
  84                 return;
  85
  86         if (depth >= router_distance)
  87                 return;
  88
  89         router_a->rou_rflag = 1;
  90
  91         for (port = 1; port <= MAX_ROUTER_PORTS; port++) {
  92                 if (router_a->rou_port[port].port_nasid == INVALID_NASID)
  93                         continue;
  94
  95                 brd = (lboard_t *)NODE_OFFSET_TO_K0(
  96                         router_a->rou_port[port].port_nasid,
  97                         router_a->rou_port[port].port_offset);
  98
  99                 if (brd->brd_type == KLTYPE_ROUTER) {
 100                         router = (klrou_t *)NODE_OFFSET_TO_K0(NASID_GET(brd), brd->brd_compts[0]);
 101                         if (router == router_b) {
 102                                 if (depth < router_distance)
 103                                         router_distance = depth;
 104                         }
 105                         else
 106                                 router_recurse(router, router_b, depth + 1);
 107                 }
 108         }
 109
 110         router_a->rou_rflag = 0;
 111 }
 112
 113 unsigned char __node_distances[MAX_COMPACT_NODES][MAX_COMPACT_NODES];
 114
 115 static int __init compute_node_distance(nasid_t nasid_a, nasid_t nasid_b)
 116 {
 117         klrou_t *router, *router_a = NULL, *router_b = NULL;
 118         lboard_t *brd, *dest_brd;
 119         cnodeid_t cnode;
 120         nasid_t nasid;
 121         int port;
 122
 123         /* Figure out which routers nodes in question are connected to */
 124         for_each_online_node(cnode) {
 125                 nasid = COMPACT_TO_NASID_NODEID(cnode);
 126
 127                 if (nasid == -1) continue;
 128
 129                 brd = find_lboard_class((lboard_t *)KL_CONFIG_INFO(nasid),
 130                                         KLTYPE_ROUTER);
 131
 132                 if (!brd)
 133                         continue;
 134
 135                 do {
 136                         if (brd->brd_flags & DUPLICATE_BOARD)
 137                                 continue;
 138
 139                         router = (klrou_t *)NODE_OFFSET_TO_K0(NASID_GET(brd), brd->brd_compts[0]);
 140                         router->rou_rflag = 0;
 141
 142                         for (port = 1; port <= MAX_ROUTER_PORTS; port++) {
 143                                 if (router->rou_port[port].port_nasid == INVALID_NASID)
 144                                         continue;
 145
 146                                 dest_brd = (lboard_t *)NODE_OFFSET_TO_K0(
 147                                         router->rou_port[port].port_nasid,
 148                                         router->rou_port[port].port_offset);
 149
 150                                 if (dest_brd->brd_type == KLTYPE_IP27) {
 151                                         if (dest_brd->brd_nasid == nasid_a)
 152                                                 router_a = router;
 153                                         if (dest_brd->brd_nasid == nasid_b)
 154                                                 router_b = router;
 155                                 }
 156                         }
 157
 158                 } while ((brd = find_lboard_class(KLCF_NEXT(brd), KLTYPE_ROUTER)));
 159         }
 160
 161         if (router_a == NULL) {
 162                 printk("node_distance: router_a NULL\n");
 163                 return -1;
 164         }
 165         if (router_b == NULL) {
 166                 printk("node_distance: router_b NULL\n");
 167                 return -1;
 168         }
 169
 170         if (nasid_a == nasid_b)
 171                 return 0;
 172
 173         if (router_a == router_b)
 174                 return 1;
 175
 176         router_distance = 100;
 177         router_recurse(router_a, router_b, 2);
 178
 179         return router_distance;
 180 }
 181
 182 static void __init init_topology_matrix(void)
 183 {
 184         nasid_t nasid, nasid2;
 185         cnodeid_t row, col;
 186
 187         for (row = 0; row < MAX_COMPACT_NODES; row++)
 188                 for (col = 0; col < MAX_COMPACT_NODES; col++)
 189                         __node_distances[row][col] = -1;
 190
 191         for_each_online_node(row) {
 192                 nasid = COMPACT_TO_NASID_NODEID(row);
 193                 for_each_online_node(col) {
 194                         nasid2 = COMPACT_TO_NASID_NODEID(col);
 195                         __node_distances[row][col] =
 196                                 compute_node_distance(nasid, nasid2);
 197                 }
 198         }
 199 }
 200
 201 static void __init dump_topology(void)
 202 {
 203         nasid_t nasid;
 204         cnodeid_t cnode;
 205         lboard_t *brd, *dest_brd;
 206         int port;
 207         int router_num = 0;
 208         klrou_t *router;
 209         cnodeid_t row, col;
 210
 211         printk("************** Topology ********************\n");
 212
 213         printk("    ");
 214         for_each_online_node(col)
 215                 printk("%02d ", col);
 216         printk("\n");
 217         for_each_online_node(row) {
 218                 printk("%02d  ", row);
 219                 for_each_online_node(col)
 220                         printk("%2d ", node_distance(row, col));
 221                 printk("\n");
 222         }
 223
 224         for_each_online_node(cnode) {
 225                 nasid = COMPACT_TO_NASID_NODEID(cnode);
 226
 227                 if (nasid == -1) continue;
 228
 229                 brd = find_lboard_class((lboard_t *)KL_CONFIG_INFO(nasid),
 230                                         KLTYPE_ROUTER);
 231
 232                 if (!brd)
 233                         continue;
 234
 235                 do {
 236                         if (brd->brd_flags & DUPLICATE_BOARD)
 237                                 continue;
 238                         printk("Router %d:", router_num);
 239                         router_num++;
 240
 241                         router = (klrou_t *)NODE_OFFSET_TO_K0(NASID_GET(brd), brd->brd_compts[0]);
 242
 243                         for (port = 1; port <= MAX_ROUTER_PORTS; port++) {
 244                                 if (router->rou_port[port].port_nasid == INVALID_NASID)
 245                                         continue;
 246
 247                                 dest_brd = (lboard_t *)NODE_OFFSET_TO_K0(
 248                                         router->rou_port[port].port_nasid,
 249                                         router->rou_port[port].port_offset);
 250
 251                                 if (dest_brd->brd_type == KLTYPE_IP27)
 252                                         printk(" %d", dest_brd->brd_nasid);
 253                                 if (dest_brd->brd_type == KLTYPE_ROUTER)
 254                                         printk(" r");
 255                         }
 256                         printk("\n");
 257
 258                 } while ( (brd = find_lboard_class(KLCF_NEXT(brd), KLTYPE_ROUTER)) );
 259         }
 260 }
 261
 262 static pfn_t __init slot_getbasepfn(cnodeid_t cnode, int slot)
 263 {
 264         nasid_t nasid = COMPACT_TO_NASID_NODEID(cnode);
 265
 266         return ((pfn_t)nasid << PFN_NASIDSHFT) | (slot << SLOT_PFNSHIFT);
 267 }
 268
 269 /*
 270  * Return the number of pages of memory provided by the given slot
 271  * on the specified node.
 272  */
 273 static pfn_t __init slot_getsize(cnodeid_t node, int slot)
 274 {
 275         return (pfn_t) slot_psize_cache[node][slot];
 276 }
 277
 278 /*
 279  * Return highest slot filled
 280  */
 281 static int __init node_getlastslot(cnodeid_t node)
 282 {
 283         return (int) slot_lastfilled_cache[node];
 284 }
 285
 286 /*
 287  * Return the pfn of the last free page of memory on a node.
 288  */
 289 static pfn_t __init node_getmaxclick(cnodeid_t node)
 290 {
 291         pfn_t   slot_psize;
 292         int     slot;
 293
 294         /*
 295          * Start at the top slot. When we find a slot with memory in it,
 296          * that's the winner.
 297          */
 298         for (slot = (MAX_MEM_SLOTS - 1); slot >= 0; slot--) {
 299                 if ((slot_psize = slot_getsize(node, slot))) {
 300                         if (slot_psize == SLOT_IGNORED)
 301                                 continue;
 302                         /* Return the basepfn + the slot size, minus 1. */
 303                         return slot_getbasepfn(node, slot) + slot_psize - 1;
 304                 }
 305         }
 306
 307         /*
 308          * If there's no memory on the node, return 0. This is likely
 309          * to cause problems.
 310          */
 311         return 0;
 312 }
 313
 314 static pfn_t __init slot_psize_compute(cnodeid_t node, int slot)
 315 {
 316         nasid_t nasid;
 317         lboard_t *brd;
 318         klmembnk_t *banks;
 319         unsigned long size;
 320
 321         nasid = COMPACT_TO_NASID_NODEID(node);
 322         /* Find the node board */
 323         brd = find_lboard((lboard_t *)KL_CONFIG_INFO(nasid), KLTYPE_IP27);
 324         if (!brd)
 325                 return 0;
 326
 327         /* Get the memory bank structure */
 328         banks = (klmembnk_t *) find_first_component(brd, KLSTRUCT_MEMBNK);
 329         if (!banks)
 330                 return 0;
 331
 332         /* Size in _Megabytes_ */
 333         size = (unsigned long)banks->membnk_bnksz[slot/4];
 334
 335         /* hack for 128 dimm banks */
 336         if (size <= 128) {
 337                 if (slot % 4 == 0) {
 338                         size <<= 20;            /* size in bytes */
 339                         return(size >> PAGE_SHIFT);
 340                 } else
 341                         return 0;
 342         } else {
 343                 size /= 4;
 344                 size <<= 20;
 345                 return size >> PAGE_SHIFT;
 346         }
 347 }
 348
 349 static void __init mlreset(void)
 350 {
 351         int i;
 352
 353         master_nasid = get_nasid();
 354         fine_mode = is_fine_dirmode();
 355
 356         /*
 357          * Probe for all CPUs - this creates the cpumask and sets up the
 358          * mapping tables.  We need to do this as early as possible.
 359          */
 360 #ifdef CONFIG_SMP
 361         cpu_node_probe();
 362 #endif
 363
 364         init_topology_matrix();
 365         dump_topology();
 366
 367         gen_region_mask(&region_mask);
 368
 369         setup_replication_mask();
 370
 371         /*
 372          * Set all nodes' calias sizes to 8k
 373          */
 374         for_each_online_node(i) {
 375                 nasid_t nasid;
 376
 377                 nasid = COMPACT_TO_NASID_NODEID(i);
 378
 379                 /*
 380                  * Always have node 0 in the region mask, otherwise
 381                  * CALIAS accesses get exceptions since the hub
 382                  * thinks it is a node 0 address.
 383                  */
 384                 REMOTE_HUB_S(nasid, PI_REGION_PRESENT, (region_mask | 1));
 385 #ifdef CONFIG_REPLICATE_EXHANDLERS
 386                 REMOTE_HUB_S(nasid, PI_CALIAS_SIZE, PI_CALIAS_SIZE_8K);
 387 #else
 388                 REMOTE_HUB_S(nasid, PI_CALIAS_SIZE, PI_CALIAS_SIZE_0);
 389 #endif
 390
 391 #ifdef LATER
 392                 /*
 393                  * Set up all hubs to have a big window pointing at
 394                  * widget 0. Memory mode, widget 0, offset 0
 395                  */
 396                 REMOTE_HUB_S(nasid, IIO_ITTE(SWIN0_BIGWIN),
 397                         ((HUB_PIO_MAP_TO_MEM << IIO_ITTE_IOSP_SHIFT) |
 398                         (0 << IIO_ITTE_WIDGET_SHIFT)));
 399 #endif
 400         }
 401 }
 402
 403 static void __init szmem(void)
 404 {
 405         pfn_t slot_psize, slot0sz = 0, nodebytes;       /* Hack to detect problem configs */
 406         int slot, ignore;
 407         cnodeid_t node;
 408
 409         num_physpages = 0;
 410
 411         for_each_online_node(node) {
 412                 ignore = nodebytes = 0;
 413                 for (slot = 0; slot < MAX_MEM_SLOTS; slot++) {
 414                         slot_psize = slot_psize_compute(node, slot);
 415                         if (slot == 0)
 416                                 slot0sz = slot_psize;
 417                         /*
 418                          * We need to refine the hack when we have replicated
 419                          * kernel text.
 420                          */
 421                         nodebytes += (1LL << SLOT_SHIFT);
 422                         if ((nodebytes >> PAGE_SHIFT) * (sizeof(struct page)) >
 423                                                 (slot0sz << PAGE_SHIFT))
 424                                 ignore = 1;
 425                         if (ignore && slot_psize) {
 426                                 printk("Ignoring slot %d onwards on node %d\n",
 427                                                                 slot, node);
 428                                 slot_psize_cache[node][slot] = SLOT_IGNORED;
 429                                 slot = MAX_MEM_SLOTS;
 430                                 continue;
 431                         }
 432                         num_physpages += slot_psize;
 433                         slot_psize_cache[node][slot] =
 434                                         (unsigned short) slot_psize;
 435                         if (slot_psize)
 436                                 slot_lastfilled_cache[node] = slot;
 437                 }
 438         }
 439 }
 440
 441 static void __init node_mem_init(cnodeid_t node)
 442 {
 443         pfn_t slot_firstpfn = slot_getbasepfn(node, 0);
 444         pfn_t slot_lastpfn = slot_firstpfn + slot_getsize(node, 0);
 445         pfn_t slot_freepfn = node_getfirstfree(node);
 446         struct pglist_data *pd;
 447         unsigned long bootmap_size;
 448
 449         /*
 450          * Allocate the node data structures on the node first.
 451          */
 452         __node_data[node] = __va(slot_freepfn << PAGE_SHIFT);
 453
 454         pd = NODE_DATA(node);
 455         pd->bdata = &plat_node_bdata[node];
 456
 457         cpus_clear(hub_data(node)->h_cpus);
 458
 459         slot_freepfn += PFN_UP(sizeof(struct pglist_data) +
 460                                sizeof(struct hub_data));
 461
 462         bootmap_size = init_bootmem_node(NODE_DATA(node), slot_freepfn,
 463                                         slot_firstpfn, slot_lastpfn);
 464         free_bootmem_node(NODE_DATA(node), slot_firstpfn << PAGE_SHIFT,
 465                         (slot_lastpfn - slot_firstpfn) << PAGE_SHIFT);
 466         reserve_bootmem_node(NODE_DATA(node), slot_firstpfn << PAGE_SHIFT,
 467                 ((slot_freepfn - slot_firstpfn) << PAGE_SHIFT) + bootmap_size);
 468 }
 469
 470 /*
 471  * A node with nothing.  We use it to avoid any special casing in
 472  * node_to_cpumask
 473  */
 474 static struct node_data null_node = {
 475         .hub = {
 476                 .h_cpus = CPU_MASK_NONE
 477         }
 478 };
 479
 480 /*
 481  * Currently, the intranode memory hole support assumes that each slot
 482  * contains at least 32 MBytes of memory. We assume all bootmem data
 483  * fits on the first slot.
 484  */
 485 void __init prom_meminit(void)
 486 {
 487         cnodeid_t node;
 488
 489         mlreset();
 490         szmem();
 491
 492         for (node = 0; node < MAX_COMPACT_NODES; node++) {
 493                 if (node_online(node)) {
 494                         node_mem_init(node);
 495                         continue;
 496                 }
 497                 __node_data[node] = &null_node;
 498         }
 499 }
 500
 501 unsigned long __init prom_free_prom_memory(void)
 502 {
 503         /* We got nothing to free here ...  */
 504         return 0;
 505 }
 506
 507 extern void pagetable_init(void);
 508 extern unsigned long setup_zero_pages(void);
 509
 510 void __init paging_init(void)
 511 {
 512         unsigned long zones_size[MAX_NR_ZONES] = {0, };
 513         unsigned node;
 514
 515         pagetable_init();
 516
 517         for_each_online_node(node) {
 518                 pfn_t start_pfn = slot_getbasepfn(node, 0);
 519                 pfn_t end_pfn = node_getmaxclick(node) + 1;
 520
 521                 zones_size[ZONE_DMA] = end_pfn - start_pfn;
 522                 free_area_init_node(node, NODE_DATA(node),
 523                                 zones_size, start_pfn, NULL);
 524
 525                 if (end_pfn > max_low_pfn)
 526                         max_low_pfn = end_pfn;
 527         }
 528 }
 529
 530 void __init mem_init(void)
 531 {
 532         unsigned long codesize, datasize, initsize, tmp;
 533         unsigned node;
 534
 535         high_memory = (void *) __va(num_physpages << PAGE_SHIFT);
 536
 537         for_each_online_node(node) {
 538                 unsigned slot, numslots;
 539                 struct page *end, *p;
 540
 541                 /*
 542                  * This will free up the bootmem, ie, slot 0 memory.
 543                  */
 544                 totalram_pages += free_all_bootmem_node(NODE_DATA(node));
 545
 546                 /*
 547                  * We need to manually do the other slots.
 548                  */
 549                 numslots = node_getlastslot(node);
 550                 for (slot = 1; slot <= numslots; slot++) {
 551                         p = nid_page_nr(node, slot_getbasepfn(node, slot) -
 552                                               slot_getbasepfn(node, 0));
 553
 554                         /*
 555                          * Free valid memory in current slot.
 556                          */
 557                         for (end = p + slot_getsize(node, slot); p < end; p++) {
 558                                 /* if (!page_is_ram(pgnr)) continue; */
 559                                 /* commented out until page_is_ram works */
 560                                 ClearPageReserved(p);
 561                                 init_page_count(p);
 562                                 __free_page(p);
 563                                 totalram_pages++;
 564                         }
 565                 }
 566         }
 567
 568         totalram_pages -= setup_zero_pages();   /* This comes from node 0 */
 569
 570         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 571         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 572         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 573
 574         tmp = nr_free_pages();
 575         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
 576                "%ldk reserved, %ldk data, %ldk init, %ldk highmem)\n",
 577                tmp << (PAGE_SHIFT-10),
 578                num_physpages << (PAGE_SHIFT-10),
 579                codesize >> 10,
 580                (num_physpages - tmp) << (PAGE_SHIFT-10),
 581                datasize >> 10,
 582                initsize >> 10,
 583                (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10)));
 584 }