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 completely bogus but 11 * alternate easier to use mechanisms are available. 12 */ 13 #include <linux/init.h> 14 #include <linux/kernel.h> 15 #include <linux/memblock.h> 16 #include <linux/mm.h> 17 #include <linux/mmzone.h> 18 #include <linux/export.h> 19 #include <linux/nodemask.h> 20 #include <linux/swap.h> 21 #include <linux/pfn.h> 22 #include <linux/highmem.h> 23 #include <asm/page.h> 24 #include <asm/pgalloc.h> 25 #include <asm/sections.h> 26 27 #include <asm/sn/arch.h> 28 #include <asm/sn/hub.h> 29 #include <asm/sn/klconfig.h> 30 #include <asm/sn/sn_private.h> 31 32 33 #define SLOT_PFNSHIFT (SLOT_SHIFT - PAGE_SHIFT) 34 #define PFN_NASIDSHFT (NASID_SHFT - PAGE_SHIFT) 35 36 struct node_data *__node_data[MAX_NUMNODES]; 37 38 EXPORT_SYMBOL(__node_data); 39 40 static int fine_mode; 41 42 static int is_fine_dirmode(void) 43 { 44 return ((LOCAL_HUB_L(NI_STATUS_REV_ID) & NSRI_REGIONSIZE_MASK) >> NSRI_REGIONSIZE_SHFT) & REGIONSIZE_FINE; 45 } 46 47 static u64 get_region(nasid_t nasid) 48 { 49 if (fine_mode) 50 return nasid >> NASID_TO_FINEREG_SHFT; 51 else 52 return nasid >> NASID_TO_COARSEREG_SHFT; 53 } 54 55 static u64 region_mask; 56 57 static void gen_region_mask(u64 *region_mask) 58 { 59 nasid_t nasid; 60 61 (*region_mask) = 0; 62 for_each_online_node(nasid) { 63 (*region_mask) |= 1ULL << get_region(nasid); 64 } 65 } 66 67 #define rou_rflag rou_flags 68 69 static int router_distance; 70 71 static void router_recurse(klrou_t *router_a, klrou_t *router_b, int depth) 72 { 73 klrou_t *router; 74 lboard_t *brd; 75 int port; 76 77 if (router_a->rou_rflag == 1) 78 return; 79 80 if (depth >= router_distance) 81 return; 82 83 router_a->rou_rflag = 1; 84 85 for (port = 1; port <= MAX_ROUTER_PORTS; port++) { 86 if (router_a->rou_port[port].port_nasid == INVALID_NASID) 87 continue; 88 89 brd = (lboard_t *)NODE_OFFSET_TO_K0( 90 router_a->rou_port[port].port_nasid, 91 router_a->rou_port[port].port_offset); 92 93 if (brd->brd_type == KLTYPE_ROUTER) { 94 router = (klrou_t *)NODE_OFFSET_TO_K0(NASID_GET(brd), brd->brd_compts[0]); 95 if (router == router_b) { 96 if (depth < router_distance) 97 router_distance = depth; 98 } 99 else 100 router_recurse(router, router_b, depth + 1); 101 } 102 } 103 104 router_a->rou_rflag = 0; 105 } 106 107 unsigned char __node_distances[MAX_NUMNODES][MAX_NUMNODES]; 108 EXPORT_SYMBOL(__node_distances); 109 110 static int __init compute_node_distance(nasid_t nasid_a, nasid_t nasid_b) 111 { 112 klrou_t *router, *router_a = NULL, *router_b = NULL; 113 lboard_t *brd, *dest_brd; 114 nasid_t nasid; 115 int port; 116 117 /* Figure out which routers nodes in question are connected to */ 118 for_each_online_node(nasid) { 119 brd = find_lboard_class((lboard_t *)KL_CONFIG_INFO(nasid), 120 KLTYPE_ROUTER); 121 122 if (!brd) 123 continue; 124 125 do { 126 if (brd->brd_flags & DUPLICATE_BOARD) 127 continue; 128 129 router = (klrou_t *)NODE_OFFSET_TO_K0(NASID_GET(brd), brd->brd_compts[0]); 130 router->rou_rflag = 0; 131 132 for (port = 1; port <= MAX_ROUTER_PORTS; port++) { 133 if (router->rou_port[port].port_nasid == INVALID_NASID) 134 continue; 135 136 dest_brd = (lboard_t *)NODE_OFFSET_TO_K0( 137 router->rou_port[port].port_nasid, 138 router->rou_port[port].port_offset); 139 140 if (dest_brd->brd_type == KLTYPE_IP27) { 141 if (dest_brd->brd_nasid == nasid_a) 142 router_a = router; 143 if (dest_brd->brd_nasid == nasid_b) 144 router_b = router; 145 } 146 } 147 148 } while ((brd = find_lboard_class(KLCF_NEXT(brd), KLTYPE_ROUTER))); 149 } 150 151 if (router_a == NULL) { 152 pr_info("node_distance: router_a NULL\n"); 153 return -1; 154 } 155 if (router_b == NULL) { 156 pr_info("node_distance: router_b NULL\n"); 157 return -1; 158 } 159 160 if (nasid_a == nasid_b) 161 return 0; 162 163 if (router_a == router_b) 164 return 1; 165 166 router_distance = 100; 167 router_recurse(router_a, router_b, 2); 168 169 return router_distance; 170 } 171 172 static void __init init_topology_matrix(void) 173 { 174 nasid_t row, col; 175 176 for (row = 0; row < MAX_NUMNODES; row++) 177 for (col = 0; col < MAX_NUMNODES; col++) 178 __node_distances[row][col] = -1; 179 180 for_each_online_node(row) { 181 for_each_online_node(col) { 182 __node_distances[row][col] = 183 compute_node_distance(row, col); 184 } 185 } 186 } 187 188 static void __init dump_topology(void) 189 { 190 nasid_t nasid; 191 lboard_t *brd, *dest_brd; 192 int port; 193 int router_num = 0; 194 klrou_t *router; 195 nasid_t row, col; 196 197 pr_info("************** Topology ********************\n"); 198 199 pr_info(" "); 200 for_each_online_node(col) 201 pr_cont("%02d ", col); 202 pr_cont("\n"); 203 for_each_online_node(row) { 204 pr_info("%02d ", row); 205 for_each_online_node(col) 206 pr_cont("%2d ", node_distance(row, col)); 207 pr_cont("\n"); 208 } 209 210 for_each_online_node(nasid) { 211 brd = find_lboard_class((lboard_t *)KL_CONFIG_INFO(nasid), 212 KLTYPE_ROUTER); 213 214 if (!brd) 215 continue; 216 217 do { 218 if (brd->brd_flags & DUPLICATE_BOARD) 219 continue; 220 pr_cont("Router %d:", router_num); 221 router_num++; 222 223 router = (klrou_t *)NODE_OFFSET_TO_K0(NASID_GET(brd), brd->brd_compts[0]); 224 225 for (port = 1; port <= MAX_ROUTER_PORTS; port++) { 226 if (router->rou_port[port].port_nasid == INVALID_NASID) 227 continue; 228 229 dest_brd = (lboard_t *)NODE_OFFSET_TO_K0( 230 router->rou_port[port].port_nasid, 231 router->rou_port[port].port_offset); 232 233 if (dest_brd->brd_type == KLTYPE_IP27) 234 pr_cont(" %d", dest_brd->brd_nasid); 235 if (dest_brd->brd_type == KLTYPE_ROUTER) 236 pr_cont(" r"); 237 } 238 pr_cont("\n"); 239 240 } while ( (brd = find_lboard_class(KLCF_NEXT(brd), KLTYPE_ROUTER)) ); 241 } 242 } 243 244 static unsigned long __init slot_getbasepfn(nasid_t nasid, int slot) 245 { 246 return ((unsigned long)nasid << PFN_NASIDSHFT) | (slot << SLOT_PFNSHIFT); 247 } 248 249 static unsigned long __init slot_psize_compute(nasid_t nasid, int slot) 250 { 251 lboard_t *brd; 252 klmembnk_t *banks; 253 unsigned long size; 254 255 /* Find the node board */ 256 brd = find_lboard((lboard_t *)KL_CONFIG_INFO(nasid), KLTYPE_IP27); 257 if (!brd) 258 return 0; 259 260 /* Get the memory bank structure */ 261 banks = (klmembnk_t *) find_first_component(brd, KLSTRUCT_MEMBNK); 262 if (!banks) 263 return 0; 264 265 /* Size in _Megabytes_ */ 266 size = (unsigned long)banks->membnk_bnksz[slot/4]; 267 268 /* hack for 128 dimm banks */ 269 if (size <= 128) { 270 if (slot % 4 == 0) { 271 size <<= 20; /* size in bytes */ 272 return size >> PAGE_SHIFT; 273 } else 274 return 0; 275 } else { 276 size /= 4; 277 size <<= 20; 278 return size >> PAGE_SHIFT; 279 } 280 } 281 282 static void __init mlreset(void) 283 { 284 nasid_t nasid; 285 286 master_nasid = get_nasid(); 287 fine_mode = is_fine_dirmode(); 288 289 /* 290 * Probe for all CPUs - this creates the cpumask and sets up the 291 * mapping tables. We need to do this as early as possible. 292 */ 293 #ifdef CONFIG_SMP 294 cpu_node_probe(); 295 #endif 296 297 init_topology_matrix(); 298 dump_topology(); 299 300 gen_region_mask(®ion_mask); 301 302 setup_replication_mask(); 303 304 /* 305 * Set all nodes' calias sizes to 8k 306 */ 307 for_each_online_node(nasid) { 308 /* 309 * Always have node 0 in the region mask, otherwise 310 * CALIAS accesses get exceptions since the hub 311 * thinks it is a node 0 address. 312 */ 313 REMOTE_HUB_S(nasid, PI_REGION_PRESENT, (region_mask | 1)); 314 #ifdef CONFIG_REPLICATE_EXHANDLERS 315 REMOTE_HUB_S(nasid, PI_CALIAS_SIZE, PI_CALIAS_SIZE_8K); 316 #else 317 REMOTE_HUB_S(nasid, PI_CALIAS_SIZE, PI_CALIAS_SIZE_0); 318 #endif 319 320 #ifdef LATER 321 /* 322 * Set up all hubs to have a big window pointing at 323 * widget 0. Memory mode, widget 0, offset 0 324 */ 325 REMOTE_HUB_S(nasid, IIO_ITTE(SWIN0_BIGWIN), 326 ((HUB_PIO_MAP_TO_MEM << IIO_ITTE_IOSP_SHIFT) | 327 (0 << IIO_ITTE_WIDGET_SHIFT))); 328 #endif 329 } 330 } 331 332 static void __init szmem(void) 333 { 334 unsigned long slot_psize, slot0sz = 0, nodebytes; /* Hack to detect problem configs */ 335 int slot; 336 nasid_t node; 337 338 for_each_online_node(node) { 339 nodebytes = 0; 340 for (slot = 0; slot < MAX_MEM_SLOTS; slot++) { 341 slot_psize = slot_psize_compute(node, slot); 342 if (slot == 0) 343 slot0sz = slot_psize; 344 /* 345 * We need to refine the hack when we have replicated 346 * kernel text. 347 */ 348 nodebytes += (1LL << SLOT_SHIFT); 349 350 if (!slot_psize) 351 continue; 352 353 if ((nodebytes >> PAGE_SHIFT) * (sizeof(struct page)) > 354 (slot0sz << PAGE_SHIFT)) { 355 pr_info("Ignoring slot %d onwards on node %d\n", 356 slot, node); 357 slot = MAX_MEM_SLOTS; 358 continue; 359 } 360 memblock_add_node(PFN_PHYS(slot_getbasepfn(node, slot)), 361 PFN_PHYS(slot_psize), node); 362 } 363 } 364 } 365 366 static void __init node_mem_init(nasid_t node) 367 { 368 unsigned long slot_firstpfn = slot_getbasepfn(node, 0); 369 unsigned long slot_freepfn = node_getfirstfree(node); 370 unsigned long start_pfn, end_pfn; 371 372 get_pfn_range_for_nid(node, &start_pfn, &end_pfn); 373 374 /* 375 * Allocate the node data structures on the node first. 376 */ 377 __node_data[node] = __va(slot_freepfn << PAGE_SHIFT); 378 memset(__node_data[node], 0, PAGE_SIZE); 379 380 NODE_DATA(node)->node_start_pfn = start_pfn; 381 NODE_DATA(node)->node_spanned_pages = end_pfn - start_pfn; 382 383 cpumask_clear(&hub_data(node)->h_cpus); 384 385 slot_freepfn += PFN_UP(sizeof(struct pglist_data) + 386 sizeof(struct hub_data)); 387 388 memblock_reserve(slot_firstpfn << PAGE_SHIFT, 389 ((slot_freepfn - slot_firstpfn) << PAGE_SHIFT)); 390 } 391 392 /* 393 * A node with nothing. We use it to avoid any special casing in 394 * cpumask_of_node 395 */ 396 static struct node_data null_node = { 397 .hub = { 398 .h_cpus = CPU_MASK_NONE 399 } 400 }; 401 402 /* 403 * Currently, the intranode memory hole support assumes that each slot 404 * contains at least 32 MBytes of memory. We assume all bootmem data 405 * fits on the first slot. 406 */ 407 void __init prom_meminit(void) 408 { 409 nasid_t node; 410 411 mlreset(); 412 szmem(); 413 max_low_pfn = PHYS_PFN(memblock_end_of_DRAM()); 414 415 for (node = 0; node < MAX_NUMNODES; node++) { 416 if (node_online(node)) { 417 node_mem_init(node); 418 continue; 419 } 420 __node_data[node] = &null_node; 421 } 422 423 memblocks_present(); 424 } 425 426 void __init prom_free_prom_memory(void) 427 { 428 /* We got nothing to free here ... */ 429 } 430 431 extern void setup_zero_pages(void); 432 433 void __init paging_init(void) 434 { 435 unsigned long zones_size[MAX_NR_ZONES] = {0, }; 436 437 pagetable_init(); 438 zones_size[ZONE_NORMAL] = max_low_pfn; 439 free_area_init_nodes(zones_size); 440 } 441 442 void __init mem_init(void) 443 { 444 high_memory = (void *) __va(get_num_physpages() << PAGE_SHIFT); 445 memblock_free_all(); 446 setup_zero_pages(); /* This comes from node 0 */ 447 mem_init_print_info(NULL); 448 } 449