| 1 | |
| 2 | |
| 3 | * mm_init.c - Memory initialisation verification and debugging
|
| 4 | *
|
| 5 | * Copyright 2008 IBM Corporation, 2008
|
| 6 | * Author Mel Gorman <[email protected]>
|
| 7 | *
|
| 8 |
|
| 9 | #include <linux/kernel.h>
|
| 10 | #include <linux/init.h>
|
| 11 | #include <linux/kobject.h>
|
| 12 | #include <linux/export.h>
|
| 13 | #include <linux/memory.h>
|
| 14 | #include <linux/notifier.h>
|
| 15 | #include <linux/sched.h>
|
| 16 | #include <linux/mman.h>
|
| 17 | #include <linux/memblock.h>
|
| 18 | #include <linux/page-isolation.h>
|
| 19 | #include <linux/padata.h>
|
| 20 | #include <linux/nmi.h>
|
| 21 | #include <linux/buffer_head.h>
|
| 22 | #include <linux/kmemleak.h>
|
| 23 | #include <linux/kfence.h>
|
| 24 | #include <linux/page_ext.h>
|
| 25 | #include <linux/pti.h>
|
| 26 | #include <linux/pgtable.h>
|
| 27 | #include <linux/stackdepot.h>
|
| 28 | #include <linux/swap.h>
|
| 29 | #include <linux/cma.h>
|
| 30 | #include <linux/crash_dump.h>
|
| 31 | #include <linux/execmem.h>
|
| 32 | #include <linux/vmstat.h>
|
| 33 | #include <linux/kexec_handover.h>
|
| 34 | #include <linux/hugetlb.h>
|
| 35 | #include "internal.h"
|
| 36 | #include "mm_init.h"
|
| 37 | #include "page_alloc.h"
|
| 38 | #include "sparse.h"
|
| 39 | #include "slab.h"
|
| 40 | #include "shuffle.h"
|
| 41 | #include "vmalloc.h"
|
| 42 |
|
| 43 | #include <asm/setup.h>
|
| 44 |
|
| 45 | #ifndef CONFIG_NUMA
|
| 46 | unsigned long max_mapnr;
|
| 47 | EXPORT_SYMBOL(max_mapnr);
|
| 48 |
|
| 49 | struct page *mem_map;
|
| 50 | EXPORT_SYMBOL(mem_map);
|
| 51 | #endif
|
| 52 |
|
| 53 | |
| 54 | * high_memory defines the upper bound on direct map memory, then end
|
| 55 | * of ZONE_NORMAL.
|
| 56 |
|
| 57 | void *high_memory;
|
| 58 | EXPORT_SYMBOL(high_memory);
|
| 59 |
|
| 60 | unsigned long zero_page_pfn __ro_after_init;
|
| 61 | EXPORT_SYMBOL(zero_page_pfn);
|
| 62 |
|
| 63 | #ifndef __HAVE_COLOR_ZERO_PAGE
|
| 64 | const uint8_t empty_zero_page[PAGE_SIZE] __aligned(PAGE_SIZE);
|
| 65 | EXPORT_SYMBOL(empty_zero_page);
|
| 66 |
|
| 67 | struct page *__zero_page __ro_after_init;
|
| 68 | EXPORT_SYMBOL(__zero_page);
|
| 69 | #endif
|
| 70 |
|
| 71 | #ifdef CONFIG_DEBUG_MEMORY_INIT
|
| 72 | int __meminitdata mminit_loglevel;
|
| 73 |
|
| 74 |
|
| 75 | void __init mminit_verify_zonelist(void)
|
| 76 | {
|
| 77 | int nid;
|
| 78 |
|
| 79 | if (mminit_loglevel < MMINIT_VERIFY)
|
| 80 | return;
|
| 81 |
|
| 82 | for_each_online_node(nid) {
|
| 83 | pg_data_t *pgdat = NODE_DATA(nid);
|
| 84 | struct zone *zone;
|
| 85 | struct zoneref *z;
|
| 86 | struct zonelist *zonelist;
|
| 87 | int i, listid, zoneid;
|
| 88 |
|
| 89 | for (i = 0; i < MAX_ZONELISTS * MAX_NR_ZONES; i++) {
|
| 90 |
|
| 91 |
|
| 92 | zoneid = i % MAX_NR_ZONES;
|
| 93 | listid = i / MAX_NR_ZONES;
|
| 94 | zonelist = &pgdat->node_zonelists[listid];
|
| 95 | zone = &pgdat->node_zones[zoneid];
|
| 96 | if (!populated_zone(zone))
|
| 97 | continue;
|
| 98 |
|
| 99 |
|
| 100 | printk(KERN_DEBUG "mminit::zonelist %s %d:%s = ",
|
| 101 | listid > 0 ? "thisnode" : "general", nid,
|
| 102 | zone->name);
|
| 103 |
|
| 104 |
|
| 105 | for_each_zone_zonelist(zone, z, zonelist, zoneid)
|
| 106 | pr_cont("%d:%s ", zone_to_nid(zone), zone->name);
|
| 107 | pr_cont("\n");
|
| 108 | }
|
| 109 | }
|
| 110 | }
|
| 111 |
|
| 112 | void __init mminit_verify_pageflags_layout(void)
|
| 113 | {
|
| 114 | int shift, width;
|
| 115 | unsigned long or_mask, add_mask;
|
| 116 |
|
| 117 | shift = BITS_PER_LONG;
|
| 118 | width = shift - NR_NON_PAGEFLAG_BITS;
|
| 119 | mminit_dprintk(MMINIT_TRACE, "pageflags_layout_widths",
|
| 120 | "Section %d Node %d Zone %d Lastcpupid %d Kasantag %d Gen %d Tier %d Flags %d\n",
|
| 121 | SECTIONS_WIDTH,
|
| 122 | NODES_WIDTH,
|
| 123 | ZONES_WIDTH,
|
| 124 | LAST_CPUPID_WIDTH,
|
| 125 | KASAN_TAG_WIDTH,
|
| 126 | LRU_GEN_WIDTH,
|
| 127 | LRU_REFS_WIDTH,
|
| 128 | NR_PAGEFLAGS);
|
| 129 | mminit_dprintk(MMINIT_TRACE, "pageflags_layout_shifts",
|
| 130 | "Section %d Node %d Zone %d Lastcpupid %d Kasantag %d\n",
|
| 131 | SECTIONS_SHIFT,
|
| 132 | NODES_SHIFT,
|
| 133 | ZONES_SHIFT,
|
| 134 | LAST_CPUPID_SHIFT,
|
| 135 | KASAN_TAG_WIDTH);
|
| 136 | mminit_dprintk(MMINIT_TRACE, "pageflags_layout_pgshifts",
|
| 137 | "Section %lu Node %lu Zone %lu Lastcpupid %lu Kasantag %lu\n",
|
| 138 | (unsigned long)SECTIONS_PGSHIFT,
|
| 139 | (unsigned long)NODES_PGSHIFT,
|
| 140 | (unsigned long)ZONES_PGSHIFT,
|
| 141 | (unsigned long)LAST_CPUPID_PGSHIFT,
|
| 142 | (unsigned long)KASAN_TAG_PGSHIFT);
|
| 143 | mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodezoneid",
|
| 144 | "Node/Zone ID: %lu -> %lu\n",
|
| 145 | (unsigned long)(ZONEID_PGOFF + ZONEID_SHIFT),
|
| 146 | (unsigned long)ZONEID_PGOFF);
|
| 147 | mminit_dprintk(MMINIT_TRACE, "pageflags_layout_usage",
|
| 148 | "location: %d -> %d layout %d -> %d unused %d -> %d page-flags\n",
|
| 149 | shift, width, width, NR_PAGEFLAGS, NR_PAGEFLAGS, 0);
|
| 150 | #ifdef NODE_NOT_IN_PAGE_FLAGS
|
| 151 | mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodeflags",
|
| 152 | "Node not in page flags");
|
| 153 | #endif
|
| 154 | #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
|
| 155 | mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodeflags",
|
| 156 | "Last cpupid not in page flags");
|
| 157 | #endif
|
| 158 |
|
| 159 | if (SECTIONS_WIDTH) {
|
| 160 | shift -= SECTIONS_WIDTH;
|
| 161 | BUG_ON(shift != SECTIONS_PGSHIFT);
|
| 162 | }
|
| 163 | if (NODES_WIDTH) {
|
| 164 | shift -= NODES_WIDTH;
|
| 165 | BUG_ON(shift != NODES_PGSHIFT);
|
| 166 | }
|
| 167 | if (ZONES_WIDTH) {
|
| 168 | shift -= ZONES_WIDTH;
|
| 169 | BUG_ON(shift != ZONES_PGSHIFT);
|
| 170 | }
|
| 171 |
|
| 172 |
|
| 173 | or_mask = (ZONES_MASK << ZONES_PGSHIFT) |
|
| 174 | (NODES_MASK << NODES_PGSHIFT) |
|
| 175 | (SECTIONS_MASK << SECTIONS_PGSHIFT);
|
| 176 | add_mask = (ZONES_MASK << ZONES_PGSHIFT) +
|
| 177 | (NODES_MASK << NODES_PGSHIFT) +
|
| 178 | (SECTIONS_MASK << SECTIONS_PGSHIFT);
|
| 179 | BUG_ON(or_mask != add_mask);
|
| 180 | }
|
| 181 |
|
| 182 | static __init int set_mminit_loglevel(char *str)
|
| 183 | {
|
| 184 | get_option(&str, &mminit_loglevel);
|
| 185 | return 0;
|
| 186 | }
|
| 187 | early_param("mminit_loglevel", set_mminit_loglevel);
|
| 188 | #endif
|
| 189 |
|
| 190 | struct kobject *mm_kobj;
|
| 191 |
|
| 192 | #ifdef CONFIG_SMP
|
| 193 | s32 vm_committed_as_batch = 32;
|
| 194 |
|
| 195 | void mm_compute_batch(int overcommit_policy)
|
| 196 | {
|
| 197 | u64 memsized_batch;
|
| 198 | s32 nr = num_present_cpus();
|
| 199 | s32 batch = max_t(s32, nr*2, 32);
|
| 200 | unsigned long ram_pages = totalram_pages();
|
| 201 |
|
| 202 | |
| 203 | * For policy OVERCOMMIT_NEVER, set batch size to 0.4% of
|
| 204 | * (total memory/#cpus), and lift it to 25% for other policies
|
| 205 | * to ease the possible lock contention for percpu_counter
|
| 206 | * vm_committed_as, while the max limit is INT_MAX
|
| 207 |
|
| 208 | if (overcommit_policy == OVERCOMMIT_NEVER)
|
| 209 | memsized_batch = min_t(u64, ram_pages/nr/256, INT_MAX);
|
| 210 | else
|
| 211 | memsized_batch = min_t(u64, ram_pages/nr/4, INT_MAX);
|
| 212 |
|
| 213 | vm_committed_as_batch = max_t(s32, memsized_batch, batch);
|
| 214 | }
|
| 215 |
|
| 216 | static int __meminit mm_compute_batch_notifier(struct notifier_block *self,
|
| 217 | unsigned long action, void *arg)
|
| 218 | {
|
| 219 | switch (action) {
|
| 220 | case MEM_ONLINE:
|
| 221 | case MEM_OFFLINE:
|
| 222 | mm_compute_batch(sysctl_overcommit_memory);
|
| 223 | break;
|
| 224 | default:
|
| 225 | break;
|
| 226 | }
|
| 227 | return NOTIFY_OK;
|
| 228 | }
|
| 229 |
|
| 230 | static int __init mm_compute_batch_init(void)
|
| 231 | {
|
| 232 | mm_compute_batch(sysctl_overcommit_memory);
|
| 233 | hotplug_memory_notifier(mm_compute_batch_notifier, MM_COMPUTE_BATCH_PRI);
|
| 234 | return 0;
|
| 235 | }
|
| 236 |
|
| 237 | __initcall(mm_compute_batch_init);
|
| 238 |
|
| 239 | #endif
|
| 240 |
|
| 241 | static int __init mm_sysfs_init(void)
|
| 242 | {
|
| 243 | mm_kobj = kobject_create_and_add("mm", kernel_kobj);
|
| 244 | if (!mm_kobj)
|
| 245 | return -ENOMEM;
|
| 246 |
|
| 247 | return 0;
|
| 248 | }
|
| 249 | postcore_initcall(mm_sysfs_init);
|
| 250 |
|
| 251 | static unsigned long arch_zone_lowest_possible_pfn[MAX_NR_ZONES] __initdata;
|
| 252 | static unsigned long arch_zone_highest_possible_pfn[MAX_NR_ZONES] __initdata;
|
| 253 | static unsigned long zone_movable_pfn[MAX_NUMNODES] __initdata;
|
| 254 |
|
| 255 | static unsigned long required_kernelcore __initdata;
|
| 256 | static unsigned long required_kernelcore_percent __initdata;
|
| 257 | static unsigned long required_movablecore __initdata;
|
| 258 | static unsigned long required_movablecore_percent __initdata;
|
| 259 |
|
| 260 | static unsigned long nr_kernel_pages __initdata;
|
| 261 | static unsigned long nr_all_pages __initdata;
|
| 262 |
|
| 263 | static bool deferred_struct_pages __meminitdata;
|
| 264 |
|
| 265 | static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats);
|
| 266 |
|
| 267 | static int __init cmdline_parse_core(char *p, unsigned long *core,
|
| 268 | unsigned long *percent)
|
| 269 | {
|
| 270 | unsigned long long coremem;
|
| 271 | char *endptr;
|
| 272 |
|
| 273 | if (!p)
|
| 274 | return -EINVAL;
|
| 275 |
|
| 276 |
|
| 277 | coremem = simple_strtoull(p, &endptr, 0);
|
| 278 | if (*endptr == '%') {
|
| 279 |
|
| 280 | WARN_ON(coremem > 100);
|
| 281 |
|
| 282 | *percent = coremem;
|
| 283 | } else {
|
| 284 | coremem = memparse(p, &p);
|
| 285 |
|
| 286 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
|
| 287 |
|
| 288 | *core = coremem >> PAGE_SHIFT;
|
| 289 | *percent = 0UL;
|
| 290 | }
|
| 291 | return 0;
|
| 292 | }
|
| 293 |
|
| 294 | bool mirrored_kernelcore __initdata_memblock;
|
| 295 |
|
| 296 | |
| 297 | * kernelcore=size sets the amount of memory for use for allocations that
|
| 298 | * cannot be reclaimed or migrated.
|
| 299 |
|
| 300 | static int __init cmdline_parse_kernelcore(char *p)
|
| 301 | {
|
| 302 |
|
| 303 | if (parse_option_str(p, "mirror")) {
|
| 304 | mirrored_kernelcore = true;
|
| 305 | return 0;
|
| 306 | }
|
| 307 |
|
| 308 | return cmdline_parse_core(p, &required_kernelcore,
|
| 309 | &required_kernelcore_percent);
|
| 310 | }
|
| 311 | early_param("kernelcore", cmdline_parse_kernelcore);
|
| 312 |
|
| 313 | |
| 314 | * movablecore=size sets the amount of memory for use for allocations that
|
| 315 | * can be reclaimed or migrated.
|
| 316 |
|
| 317 | static int __init cmdline_parse_movablecore(char *p)
|
| 318 | {
|
| 319 | return cmdline_parse_core(p, &required_movablecore,
|
| 320 | &required_movablecore_percent);
|
| 321 | }
|
| 322 | early_param("movablecore", cmdline_parse_movablecore);
|
| 323 |
|
| 324 | |
| 325 | * early_calculate_totalpages()
|
| 326 | * Sum pages in active regions for movable zone.
|
| 327 | * Populate N_MEMORY for calculating usable_nodes.
|
| 328 |
|
| 329 | static unsigned long __init early_calculate_totalpages(void)
|
| 330 | {
|
| 331 | unsigned long totalpages = 0;
|
| 332 | unsigned long start_pfn, end_pfn;
|
| 333 | int i, nid;
|
| 334 |
|
| 335 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
|
| 336 | unsigned long pages = end_pfn - start_pfn;
|
| 337 |
|
| 338 | totalpages += pages;
|
| 339 | node_set_state(nid, N_MEMORY);
|
| 340 | }
|
| 341 | return totalpages;
|
| 342 | }
|
| 343 |
|
| 344 | |
| 345 | * This finds a zone that can be used for ZONE_MOVABLE pages. The
|
| 346 | * assumption is made that zones within a node are ordered in monotonic
|
| 347 | * increasing memory addresses so that the "highest" populated zone is used
|
| 348 |
|
| 349 | static void __init find_usable_zone_for_movable(void)
|
| 350 | {
|
| 351 | int zone_index;
|
| 352 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
|
| 353 | if (zone_index == ZONE_MOVABLE)
|
| 354 | continue;
|
| 355 |
|
| 356 | if (arch_zone_highest_possible_pfn[zone_index] >
|
| 357 | arch_zone_lowest_possible_pfn[zone_index])
|
| 358 | break;
|
| 359 | }
|
| 360 |
|
| 361 | VM_BUG_ON(zone_index == -1);
|
| 362 | movable_zone = zone_index;
|
| 363 | }
|
| 364 |
|
| 365 | |
| 366 | * Find the PFN the Movable zone begins in each node. Kernel memory
|
| 367 | * is spread evenly between nodes as long as the nodes have enough
|
| 368 | * memory. When they don't, some nodes will have more kernelcore than
|
| 369 | * others
|
| 370 |
|
| 371 | static void __init find_zone_movable_pfns_for_nodes(void)
|
| 372 | {
|
| 373 | int i, nid;
|
| 374 | unsigned long usable_startpfn;
|
| 375 | unsigned long kernelcore_node, kernelcore_remaining;
|
| 376 |
|
| 377 | nodemask_t saved_node_state = node_states[N_MEMORY];
|
| 378 | unsigned long totalpages = early_calculate_totalpages();
|
| 379 | int usable_nodes = nodes_weight(node_states[N_MEMORY]);
|
| 380 | struct memblock_region *r;
|
| 381 |
|
| 382 |
|
| 383 | find_usable_zone_for_movable();
|
| 384 |
|
| 385 | |
| 386 | * If movable_node is specified, ignore kernelcore and movablecore
|
| 387 | * options.
|
| 388 |
|
| 389 | if (movable_node_is_enabled()) {
|
| 390 | for_each_mem_region(r) {
|
| 391 | if (!memblock_is_hotpluggable(r))
|
| 392 | continue;
|
| 393 |
|
| 394 | nid = memblock_get_region_node(r);
|
| 395 |
|
| 396 | usable_startpfn = memblock_region_memory_base_pfn(r);
|
| 397 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
|
| 398 | min(usable_startpfn, zone_movable_pfn[nid]) :
|
| 399 | usable_startpfn;
|
| 400 | }
|
| 401 |
|
| 402 | goto out2;
|
| 403 | }
|
| 404 |
|
| 405 | |
| 406 | * If kernelcore=mirror is specified, ignore movablecore option
|
| 407 |
|
| 408 | if (mirrored_kernelcore) {
|
| 409 | bool mem_below_4gb_not_mirrored = false;
|
| 410 |
|
| 411 | if (!memblock_has_mirror()) {
|
| 412 | pr_warn("The system has no mirror memory, ignore kernelcore=mirror.\n");
|
| 413 | goto out;
|
| 414 | }
|
| 415 |
|
| 416 | if (is_kdump_kernel()) {
|
| 417 | pr_warn("The system is under kdump, ignore kernelcore=mirror.\n");
|
| 418 | goto out;
|
| 419 | }
|
| 420 |
|
| 421 | for_each_mem_region(r) {
|
| 422 | if (memblock_is_mirror(r))
|
| 423 | continue;
|
| 424 |
|
| 425 | nid = memblock_get_region_node(r);
|
| 426 |
|
| 427 | usable_startpfn = memblock_region_memory_base_pfn(r);
|
| 428 |
|
| 429 | if (usable_startpfn < PHYS_PFN(SZ_4G)) {
|
| 430 | mem_below_4gb_not_mirrored = true;
|
| 431 | continue;
|
| 432 | }
|
| 433 |
|
| 434 | zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
|
| 435 | min(usable_startpfn, zone_movable_pfn[nid]) :
|
| 436 | usable_startpfn;
|
| 437 | }
|
| 438 |
|
| 439 | if (mem_below_4gb_not_mirrored)
|
| 440 | pr_warn("This configuration results in unmirrored kernel memory.\n");
|
| 441 |
|
| 442 | goto out2;
|
| 443 | }
|
| 444 |
|
| 445 | |
| 446 | * If kernelcore=nn% or movablecore=nn% was specified, calculate the
|
| 447 | * amount of necessary memory.
|
| 448 |
|
| 449 | if (required_kernelcore_percent)
|
| 450 | required_kernelcore = (totalpages * 100 * required_kernelcore_percent) /
|
| 451 | 10000UL;
|
| 452 | if (required_movablecore_percent)
|
| 453 | required_movablecore = (totalpages * 100 * required_movablecore_percent) /
|
| 454 | 10000UL;
|
| 455 |
|
| 456 | |
| 457 | * If movablecore= was specified, calculate what size of
|
| 458 | * kernelcore that corresponds so that memory usable for
|
| 459 | * any allocation type is evenly spread. If both kernelcore
|
| 460 | * and movablecore are specified, then the value of kernelcore
|
| 461 | * will be used for required_kernelcore if it's greater than
|
| 462 | * what movablecore would have allowed.
|
| 463 |
|
| 464 | if (required_movablecore) {
|
| 465 | unsigned long corepages;
|
| 466 |
|
| 467 | |
| 468 | * Round-up so that ZONE_MOVABLE is at least as large as what
|
| 469 | * was requested by the user
|
| 470 |
|
| 471 | required_movablecore =
|
| 472 | round_up(required_movablecore, MAX_ORDER_NR_PAGES);
|
| 473 | required_movablecore = min(totalpages, required_movablecore);
|
| 474 | corepages = totalpages - required_movablecore;
|
| 475 |
|
| 476 | required_kernelcore = max(required_kernelcore, corepages);
|
| 477 | }
|
| 478 |
|
| 479 | |
| 480 | * If kernelcore was not specified or kernelcore size is larger
|
| 481 | * than totalpages, there is no ZONE_MOVABLE.
|
| 482 |
|
| 483 | if (!required_kernelcore || required_kernelcore >= totalpages)
|
| 484 | goto out;
|
| 485 |
|
| 486 |
|
| 487 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
|
| 488 |
|
| 489 | restart:
|
| 490 |
|
| 491 | kernelcore_node = required_kernelcore / usable_nodes;
|
| 492 | for_each_node_state(nid, N_MEMORY) {
|
| 493 | unsigned long start_pfn, end_pfn;
|
| 494 |
|
| 495 | |
| 496 | * Recalculate kernelcore_node if the division per node
|
| 497 | * now exceeds what is necessary to satisfy the requested
|
| 498 | * amount of memory for the kernel
|
| 499 |
|
| 500 | if (required_kernelcore < kernelcore_node)
|
| 501 | kernelcore_node = required_kernelcore / usable_nodes;
|
| 502 |
|
| 503 | |
| 504 | * As the map is walked, we track how much memory is usable
|
| 505 | * by the kernel using kernelcore_remaining. When it is
|
| 506 | * 0, the rest of the node is usable by ZONE_MOVABLE
|
| 507 |
|
| 508 | kernelcore_remaining = kernelcore_node;
|
| 509 |
|
| 510 |
|
| 511 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
|
| 512 | unsigned long size_pages;
|
| 513 |
|
| 514 | start_pfn = max(start_pfn, zone_movable_pfn[nid]);
|
| 515 | if (start_pfn >= end_pfn)
|
| 516 | continue;
|
| 517 |
|
| 518 |
|
| 519 | if (start_pfn < usable_startpfn) {
|
| 520 | unsigned long kernel_pages;
|
| 521 | kernel_pages = min(end_pfn, usable_startpfn)
|
| 522 | - start_pfn;
|
| 523 |
|
| 524 | kernelcore_remaining -= min(kernel_pages,
|
| 525 | kernelcore_remaining);
|
| 526 | required_kernelcore -= min(kernel_pages,
|
| 527 | required_kernelcore);
|
| 528 |
|
| 529 |
|
| 530 | if (end_pfn <= usable_startpfn) {
|
| 531 |
|
| 532 | |
| 533 | * Push zone_movable_pfn to the end so
|
| 534 | * that if we have to rebalance
|
| 535 | * kernelcore across nodes, we will
|
| 536 | * not double account here
|
| 537 |
|
| 538 | zone_movable_pfn[nid] = end_pfn;
|
| 539 | continue;
|
| 540 | }
|
| 541 | start_pfn = usable_startpfn;
|
| 542 | }
|
| 543 |
|
| 544 | |
| 545 | * The usable PFN range for ZONE_MOVABLE is from
|
| 546 | * start_pfn->end_pfn. Calculate size_pages as the
|
| 547 | * number of pages used as kernelcore
|
| 548 |
|
| 549 | size_pages = end_pfn - start_pfn;
|
| 550 | if (size_pages > kernelcore_remaining)
|
| 551 | size_pages = kernelcore_remaining;
|
| 552 | zone_movable_pfn[nid] = start_pfn + size_pages;
|
| 553 |
|
| 554 | |
| 555 | * Some kernelcore has been met, update counts and
|
| 556 | * break if the kernelcore for this node has been
|
| 557 | * satisfied
|
| 558 |
|
| 559 | required_kernelcore -= min(required_kernelcore,
|
| 560 | size_pages);
|
| 561 | kernelcore_remaining -= size_pages;
|
| 562 | if (!kernelcore_remaining)
|
| 563 | break;
|
| 564 | }
|
| 565 | }
|
| 566 |
|
| 567 | |
| 568 | * If there is still required_kernelcore, we do another pass with one
|
| 569 | * less node in the count. This will push zone_movable_pfn[nid] further
|
| 570 | * along on the nodes that still have memory until kernelcore is
|
| 571 | * satisfied
|
| 572 |
|
| 573 | usable_nodes--;
|
| 574 | if (usable_nodes && required_kernelcore > usable_nodes)
|
| 575 | goto restart;
|
| 576 |
|
| 577 | out2:
|
| 578 |
|
| 579 | for_each_node_state(nid, N_MEMORY) {
|
| 580 | unsigned long start_pfn, end_pfn;
|
| 581 |
|
| 582 | zone_movable_pfn[nid] =
|
| 583 | round_up(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
|
| 584 |
|
| 585 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
|
| 586 | if (zone_movable_pfn[nid] >= end_pfn)
|
| 587 | zone_movable_pfn[nid] = 0;
|
| 588 | }
|
| 589 |
|
| 590 | out:
|
| 591 |
|
| 592 | node_states[N_MEMORY] = saved_node_state;
|
| 593 | }
|
| 594 |
|
| 595 | void __meminit __init_single_page(struct page *page, unsigned long pfn,
|
| 596 | unsigned long zone, int nid)
|
| 597 | {
|
| 598 | mm_zero_struct_page(page);
|
| 599 | set_page_links(page, zone, nid, pfn);
|
| 600 | init_page_count(page);
|
| 601 | atomic_set(&page->_mapcount, -1);
|
| 602 | page_cpupid_reset_last(page);
|
| 603 | page_kasan_tag_reset(page);
|
| 604 |
|
| 605 | INIT_LIST_HEAD(&page->lru);
|
| 606 | #ifdef WANT_PAGE_VIRTUAL
|
| 607 |
|
| 608 | if (!is_highmem_idx(zone))
|
| 609 | set_page_address(page, __va(pfn << PAGE_SHIFT));
|
| 610 | #endif
|
| 611 | }
|
| 612 |
|
| 613 | #ifdef CONFIG_NUMA
|
| 614 | |
| 615 | * During memory init memblocks map pfns to nids. The search is expensive and
|
| 616 | * this caches recent lookups. The implementation of __early_pfn_to_nid
|
| 617 | * treats start/end as pfns.
|
| 618 |
|
| 619 | struct mminit_pfnnid_cache {
|
| 620 | unsigned long last_start;
|
| 621 | unsigned long last_end;
|
| 622 | int last_nid;
|
| 623 | };
|
| 624 |
|
| 625 | static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
|
| 626 |
|
| 627 | |
| 628 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
|
| 629 |
|
| 630 | static int __meminit __early_pfn_to_nid(unsigned long pfn,
|
| 631 | struct mminit_pfnnid_cache *state)
|
| 632 | {
|
| 633 | unsigned long start_pfn, end_pfn;
|
| 634 | int nid;
|
| 635 |
|
| 636 | if (state->last_start <= pfn && pfn < state->last_end)
|
| 637 | return state->last_nid;
|
| 638 |
|
| 639 | nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
|
| 640 | if (nid != NUMA_NO_NODE) {
|
| 641 | state->last_start = start_pfn;
|
| 642 | state->last_end = end_pfn;
|
| 643 | state->last_nid = nid;
|
| 644 | }
|
| 645 |
|
| 646 | return nid;
|
| 647 | }
|
| 648 |
|
| 649 | int __meminit early_pfn_to_nid(unsigned long pfn)
|
| 650 | {
|
| 651 | static DEFINE_SPINLOCK(early_pfn_lock);
|
| 652 | int nid;
|
| 653 |
|
| 654 | spin_lock(&early_pfn_lock);
|
| 655 | nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
|
| 656 | if (nid < 0)
|
| 657 | nid = first_online_node;
|
| 658 | spin_unlock(&early_pfn_lock);
|
| 659 |
|
| 660 | return nid;
|
| 661 | }
|
| 662 |
|
| 663 | bool hashdist = HASHDIST_DEFAULT;
|
| 664 |
|
| 665 | static int __init set_hashdist(char *str)
|
| 666 | {
|
| 667 | return kstrtobool(str, &hashdist) == 0;
|
| 668 | }
|
| 669 | __setup("hashdist=", set_hashdist);
|
| 670 |
|
| 671 | static inline void fixup_hashdist(void)
|
| 672 | {
|
| 673 | if (num_node_state(N_MEMORY) == 1)
|
| 674 | hashdist = false;
|
| 675 | }
|
| 676 | #else
|
| 677 | static inline void fixup_hashdist(void) {}
|
| 678 | #endif
|
| 679 |
|
| 680 | #if defined(CONFIG_ZONE_DEVICE) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
|
| 681 | static __meminit void pageblock_migratetype_init_range(unsigned long pfn,
|
| 682 | unsigned long nr_pages, int migratetype, bool atomic)
|
| 683 | {
|
| 684 | const unsigned long end = pfn + nr_pages;
|
| 685 |
|
| 686 | for (pfn = pageblock_align(pfn); pfn < end; pfn += pageblock_nr_pages) {
|
| 687 | enum migratetype mt = kho_scratch_migratetype(pfn, migratetype);
|
| 688 |
|
| 689 | init_pageblock_migratetype(pfn_to_page(pfn), mt, false);
|
| 690 | if (!atomic && IS_ALIGNED(pfn, PAGES_PER_SECTION))
|
| 691 | cond_resched();
|
| 692 | }
|
| 693 | }
|
| 694 | #endif
|
| 695 |
|
| 696 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
|
| 697 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat)
|
| 698 | {
|
| 699 | pgdat->first_deferred_pfn = ULONG_MAX;
|
| 700 | }
|
| 701 |
|
| 702 |
|
| 703 | static inline bool __meminit early_page_initialised(unsigned long pfn, int nid)
|
| 704 | {
|
| 705 | if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn)
|
| 706 | return false;
|
| 707 |
|
| 708 | return true;
|
| 709 | }
|
| 710 |
|
| 711 | |
| 712 | * Returns true when the remaining initialisation should be deferred until
|
| 713 | * later in the boot cycle when it can be parallelised.
|
| 714 |
|
| 715 | static bool __meminit
|
| 716 | defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
|
| 717 | {
|
| 718 | static unsigned long prev_end_pfn, nr_initialised;
|
| 719 |
|
| 720 | if (early_page_ext_enabled())
|
| 721 | return false;
|
| 722 |
|
| 723 |
|
| 724 | if (end_pfn < pgdat_end_pfn(NODE_DATA(nid)))
|
| 725 | return false;
|
| 726 |
|
| 727 | if (NODE_DATA(nid)->first_deferred_pfn != ULONG_MAX)
|
| 728 | return true;
|
| 729 |
|
| 730 | |
| 731 | * prev_end_pfn static that contains the end of previous zone
|
| 732 | * No need to protect because called very early in boot before smp_init.
|
| 733 |
|
| 734 | if (prev_end_pfn != end_pfn) {
|
| 735 | prev_end_pfn = end_pfn;
|
| 736 | nr_initialised = 0;
|
| 737 | }
|
| 738 |
|
| 739 | |
| 740 | * We start only with one section of pages, more pages are added as
|
| 741 | * needed until the rest of deferred pages are initialized.
|
| 742 |
|
| 743 | nr_initialised++;
|
| 744 | if ((nr_initialised > PAGES_PER_SECTION) &&
|
| 745 | (pfn & (PAGES_PER_SECTION - 1)) == 0) {
|
| 746 | NODE_DATA(nid)->first_deferred_pfn = pfn;
|
| 747 | return true;
|
| 748 | }
|
| 749 | return false;
|
| 750 | }
|
| 751 |
|
| 752 | static void __meminit __init_deferred_page(unsigned long pfn, int nid)
|
| 753 | {
|
| 754 | pg_data_t *pgdat = NODE_DATA(nid);
|
| 755 | int zid;
|
| 756 |
|
| 757 | if (early_page_initialised(pfn, nid))
|
| 758 | return;
|
| 759 |
|
| 760 | for (zid = 0; zid < MAX_NR_ZONES; zid++) {
|
| 761 | struct zone *zone = &pgdat->node_zones[zid];
|
| 762 |
|
| 763 | if (zone_spans_pfn(zone, pfn))
|
| 764 | break;
|
| 765 | }
|
| 766 | __init_single_page(pfn_to_page(pfn), pfn, zid, nid);
|
| 767 |
|
| 768 | if (pageblock_aligned(pfn)) {
|
| 769 | enum migratetype mt =
|
| 770 | kho_scratch_migratetype(pfn, MIGRATE_MOVABLE);
|
| 771 | init_pageblock_migratetype(pfn_to_page(pfn), mt, false);
|
| 772 | }
|
| 773 | }
|
| 774 | #else
|
| 775 | static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {}
|
| 776 |
|
| 777 | static inline bool early_page_initialised(unsigned long pfn, int nid)
|
| 778 | {
|
| 779 | return true;
|
| 780 | }
|
| 781 |
|
| 782 | static inline bool defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
|
| 783 | {
|
| 784 | return false;
|
| 785 | }
|
| 786 |
|
| 787 | static inline void __init_deferred_page(unsigned long pfn, int nid)
|
| 788 | {
|
| 789 | }
|
| 790 | #endif
|
| 791 |
|
| 792 | void __meminit init_deferred_page(unsigned long pfn, int nid)
|
| 793 | {
|
| 794 | __init_deferred_page(pfn, nid);
|
| 795 | }
|
| 796 |
|
| 797 | |
| 798 | * Only struct pages that correspond to ranges defined by memblock.memory
|
| 799 | * are zeroed and initialized by going through __init_single_page() during
|
| 800 | * memmap_init_zone_range().
|
| 801 | *
|
| 802 | * But, there could be struct pages that correspond to holes in
|
| 803 | * memblock.memory. This can happen because of the following reasons:
|
| 804 | * - physical memory bank size is not necessarily the exact multiple of the
|
| 805 | * arbitrary section size
|
| 806 | * - early reserved memory may not be listed in memblock.memory
|
| 807 | * - non-memory regions covered by the contiguous flatmem mapping
|
| 808 | * - memory layouts defined with memmap= kernel parameter may not align
|
| 809 | * nicely with memmap sections
|
| 810 | *
|
| 811 | * Explicitly initialize those struct pages so that:
|
| 812 | * - PG_Reserved is set
|
| 813 | * - zone and node links point to zone and node that span the page if the
|
| 814 | * hole is in the middle of a zone
|
| 815 | * - zone and node links point to adjacent zone/node if the hole falls on
|
| 816 | * the zone boundary; the pages in such holes will be prepended to the
|
| 817 | * zone/node above the hole except for the trailing pages in the last
|
| 818 | * section that will be appended to the zone/node below.
|
| 819 |
|
| 820 | static void __init init_unavailable_range(unsigned long spfn,
|
| 821 | unsigned long epfn,
|
| 822 | int zone, int node)
|
| 823 | {
|
| 824 | unsigned long pfn;
|
| 825 | u64 pgcnt = 0;
|
| 826 |
|
| 827 | for_each_valid_pfn(pfn, spfn, epfn) {
|
| 828 | __init_single_page(pfn_to_page(pfn), pfn, zone, node);
|
| 829 | __SetPageReserved(pfn_to_page(pfn));
|
| 830 | pgcnt++;
|
| 831 | }
|
| 832 |
|
| 833 | if (pgcnt)
|
| 834 | pr_info("On node %d, zone %s: %lld pages in unavailable ranges\n",
|
| 835 | node, zone_names[zone], pgcnt);
|
| 836 | }
|
| 837 |
|
| 838 | |
| 839 | * Initially all pages are reserved - free ones are freed
|
| 840 | * up by memblock_free_all() once the early boot process is
|
| 841 | * done. Non-atomic initialization, single-pass.
|
| 842 | *
|
| 843 | * All aligned pageblocks are initialized to the specified migratetype
|
| 844 | * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related
|
| 845 | * zone stats (e.g., nr_isolate_pageblock) are touched.
|
| 846 |
|
| 847 | void __meminit memmap_init_range(unsigned long size, int nid, unsigned long zone,
|
| 848 | unsigned long start_pfn, unsigned long zone_end_pfn,
|
| 849 | enum meminit_context context,
|
| 850 | struct vmem_altmap *altmap, int migratetype,
|
| 851 | bool isolate_pageblock)
|
| 852 | {
|
| 853 | unsigned long pfn, end_pfn = start_pfn + size;
|
| 854 | struct page *page;
|
| 855 |
|
| 856 | if (highest_memmap_pfn < end_pfn - 1)
|
| 857 | highest_memmap_pfn = end_pfn - 1;
|
| 858 |
|
| 859 | #ifdef CONFIG_ZONE_DEVICE
|
| 860 | |
| 861 | * Honor reservation requested by the driver for this ZONE_DEVICE
|
| 862 | * memory. We limit the total number of pages to initialize to just
|
| 863 | * those that might contain the memory mapping. We will defer the
|
| 864 | * ZONE_DEVICE page initialization until after we have released
|
| 865 | * the hotplug lock.
|
| 866 |
|
| 867 | if (zone == ZONE_DEVICE) {
|
| 868 | if (!altmap)
|
| 869 | return;
|
| 870 |
|
| 871 | if (start_pfn == altmap->base_pfn)
|
| 872 | start_pfn += altmap->reserve;
|
| 873 | end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
|
| 874 | }
|
| 875 | #endif
|
| 876 |
|
| 877 | for (pfn = start_pfn; pfn < end_pfn; ) {
|
| 878 | |
| 879 | * There can be holes in boot-time mem_map[]s handed to this
|
| 880 | * function. They do not exist on hotplugged memory.
|
| 881 |
|
| 882 | if (context == MEMINIT_EARLY) {
|
| 883 | if (defer_init(nid, pfn, zone_end_pfn)) {
|
| 884 | deferred_struct_pages = true;
|
| 885 | break;
|
| 886 | }
|
| 887 | }
|
| 888 |
|
| 889 | page = pfn_to_page(pfn);
|
| 890 | __init_single_page(page, pfn, zone, nid);
|
| 891 | if (context == MEMINIT_HOTPLUG) {
|
| 892 | #ifdef CONFIG_ZONE_DEVICE
|
| 893 | if (zone == ZONE_DEVICE)
|
| 894 | __SetPageReserved(page);
|
| 895 | else
|
| 896 | #endif
|
| 897 | __SetPageOffline(page);
|
| 898 | }
|
| 899 |
|
| 900 | |
| 901 | * Usually, we want to mark the pageblock MIGRATE_MOVABLE,
|
| 902 | * such that unmovable allocations won't be scattered all
|
| 903 | * over the place during system boot.
|
| 904 |
|
| 905 | if (pageblock_aligned(pfn)) {
|
| 906 | enum migratetype mt = kho_scratch_migratetype(pfn, migratetype);
|
| 907 |
|
| 908 | init_pageblock_migratetype(page, mt, isolate_pageblock);
|
| 909 | cond_resched();
|
| 910 | }
|
| 911 | pfn++;
|
| 912 | }
|
| 913 | }
|
| 914 |
|
| 915 | static void __init memmap_init_zone_range(struct zone *zone,
|
| 916 | unsigned long start_pfn,
|
| 917 | unsigned long end_pfn,
|
| 918 | unsigned long *hole_pfn)
|
| 919 | {
|
| 920 | unsigned long zone_start_pfn = zone->zone_start_pfn;
|
| 921 | unsigned long zone_end_pfn = zone_start_pfn + zone->spanned_pages;
|
| 922 | int nid = zone_to_nid(zone), zone_id = zone_idx(zone);
|
| 923 |
|
| 924 | start_pfn = clamp(start_pfn, zone_start_pfn, zone_end_pfn);
|
| 925 | end_pfn = clamp(end_pfn, zone_start_pfn, zone_end_pfn);
|
| 926 |
|
| 927 | if (start_pfn >= end_pfn)
|
| 928 | return;
|
| 929 |
|
| 930 | memmap_init_range(end_pfn - start_pfn, nid, zone_id, start_pfn,
|
| 931 | zone_end_pfn, MEMINIT_EARLY, NULL, MIGRATE_MOVABLE,
|
| 932 | false);
|
| 933 |
|
| 934 | if (*hole_pfn < start_pfn)
|
| 935 | init_unavailable_range(*hole_pfn, start_pfn, zone_id, nid);
|
| 936 |
|
| 937 | *hole_pfn = end_pfn;
|
| 938 | }
|
| 939 |
|
| 940 | static void __init memmap_init(void)
|
| 941 | {
|
| 942 | unsigned long start_pfn, end_pfn;
|
| 943 | unsigned long hole_pfn = 0;
|
| 944 | int i, j, zone_id = 0, nid;
|
| 945 |
|
| 946 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
|
| 947 | struct pglist_data *node = NODE_DATA(nid);
|
| 948 |
|
| 949 | for (j = 0; j < MAX_NR_ZONES; j++) {
|
| 950 | struct zone *zone = node->node_zones + j;
|
| 951 |
|
| 952 | if (!populated_zone(zone))
|
| 953 | continue;
|
| 954 |
|
| 955 | memmap_init_zone_range(zone, start_pfn, end_pfn,
|
| 956 | &hole_pfn);
|
| 957 | zone_id = j;
|
| 958 | }
|
| 959 | }
|
| 960 |
|
| 961 | |
| 962 | * Initialize the memory map for hole in the range [memory_end,
|
| 963 | * section_end] for SPARSEMEM and in the range [memory_end, memmap_end]
|
| 964 | * for FLATMEM.
|
| 965 | * Append the pages in this hole to the highest zone in the last
|
| 966 | * node.
|
| 967 |
|
| 968 | #ifdef CONFIG_SPARSEMEM
|
| 969 | end_pfn = round_up(end_pfn, PAGES_PER_SECTION);
|
| 970 | #else
|
| 971 | end_pfn = round_up(end_pfn, MAX_ORDER_NR_PAGES);
|
| 972 | #endif
|
| 973 | if (hole_pfn < end_pfn)
|
| 974 | init_unavailable_range(hole_pfn, end_pfn, zone_id, nid);
|
| 975 | }
|
| 976 |
|
| 977 | #ifdef CONFIG_ZONE_DEVICE
|
| 978 | static void __ref __init_zone_device_page(struct page *page, unsigned long pfn,
|
| 979 | unsigned long zone_idx, int nid,
|
| 980 | struct dev_pagemap *pgmap)
|
| 981 | {
|
| 982 |
|
| 983 | __init_single_page(page, pfn, zone_idx, nid);
|
| 984 |
|
| 985 | |
| 986 | * Mark page reserved as it will need to wait for onlining
|
| 987 | * phase for it to be fully associated with a zone.
|
| 988 | *
|
| 989 | * We can use the non-atomic __set_bit operation for setting
|
| 990 | * the flag as we are still initializing the pages.
|
| 991 |
|
| 992 | __SetPageReserved(page);
|
| 993 |
|
| 994 | |
| 995 | * ZONE_DEVICE pages union ->lru with a ->pgmap back pointer
|
| 996 | * and zone_device_data. It is a bug if a ZONE_DEVICE page is
|
| 997 | * ever freed or placed on a driver-private list.
|
| 998 |
|
| 999 | page_folio(page)->pgmap = pgmap;
|
| 1000 | page->zone_device_data = NULL;
|
| 1001 |
|
| 1002 | |
| 1003 | * ZONE_DEVICE pages other than MEMORY_TYPE_GENERIC are released
|
| 1004 | * directly to the driver page allocator which will set the page count
|
| 1005 | * to 1 when allocating the page.
|
| 1006 | *
|
| 1007 | * MEMORY_TYPE_GENERIC and MEMORY_TYPE_FS_DAX pages automatically have
|
| 1008 | * their refcount reset to one whenever they are freed (ie. after
|
| 1009 | * their refcount drops to 0).
|
| 1010 |
|
| 1011 | switch (pgmap->type) {
|
| 1012 | case MEMORY_DEVICE_FS_DAX:
|
| 1013 | case MEMORY_DEVICE_PRIVATE:
|
| 1014 | case MEMORY_DEVICE_COHERENT:
|
| 1015 | case MEMORY_DEVICE_PCI_P2PDMA:
|
| 1016 | set_page_count(page, 0);
|
| 1017 | break;
|
| 1018 |
|
| 1019 | case MEMORY_DEVICE_GENERIC:
|
| 1020 | break;
|
| 1021 | }
|
| 1022 | }
|
| 1023 |
|
| 1024 | |
| 1025 | * With compound page geometry and when struct pages are stored in ram most
|
| 1026 | * tail pages are reused. Consequently, the amount of unique struct pages to
|
| 1027 | * initialize is a lot smaller that the total amount of struct pages being
|
| 1028 | * mapped. This is a paired / mild layering violation with explicit knowledge
|
| 1029 | * of how the sparse_vmemmap internals handle compound pages in the lack
|
| 1030 | * of an altmap. See vmemmap_populate_compound_pages().
|
| 1031 |
|
| 1032 | static inline unsigned long compound_nr_pages(unsigned long pfn,
|
| 1033 | struct vmem_altmap *altmap,
|
| 1034 | struct dev_pagemap *pgmap)
|
| 1035 | {
|
| 1036 | |
| 1037 | * If DAX memory is hot-plugged into an unoccupied subsection
|
| 1038 | * of an early section, the unoptimized boot memmap is reused.
|
| 1039 | * See section_activate().
|
| 1040 |
|
| 1041 | if (early_section(__pfn_to_section(pfn)) ||
|
| 1042 | !vmemmap_can_optimize(altmap, pgmap))
|
| 1043 | return pgmap_vmemmap_nr(pgmap);
|
| 1044 |
|
| 1045 | return VMEMMAP_RESERVE_NR * (PAGE_SIZE / sizeof(struct page));
|
| 1046 | }
|
| 1047 |
|
| 1048 | static void __ref memmap_init_compound(struct page *head,
|
| 1049 | unsigned long head_pfn,
|
| 1050 | unsigned long zone_idx, int nid,
|
| 1051 | struct dev_pagemap *pgmap,
|
| 1052 | unsigned long nr_pages)
|
| 1053 | {
|
| 1054 | unsigned long pfn, end_pfn = head_pfn + nr_pages;
|
| 1055 | unsigned int order = pgmap->vmemmap_shift;
|
| 1056 |
|
| 1057 | |
| 1058 | * We have to initialize the pages, including setting up page links.
|
| 1059 | * prep_compound_page() does not take care of that, so instead we
|
| 1060 | * open-code prep_compound_page() so we can take care of initializing
|
| 1061 | * the pages in the same go.
|
| 1062 |
|
| 1063 | __SetPageHead(head);
|
| 1064 | for (pfn = head_pfn + 1; pfn < end_pfn; pfn++) {
|
| 1065 | struct page *page = pfn_to_page(pfn);
|
| 1066 |
|
| 1067 | __init_zone_device_page(page, pfn, zone_idx, nid, pgmap);
|
| 1068 | prep_compound_tail(page, head, order);
|
| 1069 | set_page_count(page, 0);
|
| 1070 | }
|
| 1071 | prep_compound_head(head, order);
|
| 1072 | }
|
| 1073 |
|
| 1074 | void __ref memmap_init_zone_device(struct zone *zone,
|
| 1075 | unsigned long start_pfn,
|
| 1076 | unsigned long nr_pages,
|
| 1077 | struct dev_pagemap *pgmap)
|
| 1078 | {
|
| 1079 | unsigned long pfn, end_pfn = start_pfn + nr_pages;
|
| 1080 | struct pglist_data *pgdat = zone->zone_pgdat;
|
| 1081 | struct vmem_altmap *altmap = pgmap_altmap(pgmap);
|
| 1082 | unsigned int pfns_per_compound = pgmap_vmemmap_nr(pgmap);
|
| 1083 | unsigned long zone_idx = zone_idx(zone);
|
| 1084 | unsigned long start = jiffies;
|
| 1085 | int nid = pgdat->node_id;
|
| 1086 |
|
| 1087 | if (WARN_ON_ONCE(!pgmap || zone_idx != ZONE_DEVICE))
|
| 1088 | return;
|
| 1089 |
|
| 1090 | |
| 1091 | * The call to memmap_init should have already taken care
|
| 1092 | * of the pages reserved for the memmap, so we can just jump to
|
| 1093 | * the end of that region and start processing the device pages.
|
| 1094 |
|
| 1095 | if (altmap) {
|
| 1096 | start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
|
| 1097 | nr_pages = end_pfn - start_pfn;
|
| 1098 | }
|
| 1099 |
|
| 1100 | for (pfn = start_pfn; pfn < end_pfn; pfn += pfns_per_compound) {
|
| 1101 | struct page *page = pfn_to_page(pfn);
|
| 1102 |
|
| 1103 | __init_zone_device_page(page, pfn, zone_idx, nid, pgmap);
|
| 1104 |
|
| 1105 | if (IS_ALIGNED(pfn, PAGES_PER_SECTION))
|
| 1106 | cond_resched();
|
| 1107 |
|
| 1108 | if (pfns_per_compound == 1)
|
| 1109 | continue;
|
| 1110 |
|
| 1111 | memmap_init_compound(page, pfn, zone_idx, nid, pgmap,
|
| 1112 | compound_nr_pages(pfn, altmap, pgmap));
|
| 1113 | }
|
| 1114 |
|
| 1115 | pageblock_migratetype_init_range(start_pfn, nr_pages, MIGRATE_MOVABLE, false);
|
| 1116 |
|
| 1117 | pr_debug("%s initialised %lu pages in %ums\n", __func__,
|
| 1118 | nr_pages, jiffies_to_msecs(jiffies - start));
|
| 1119 | }
|
| 1120 | #endif
|
| 1121 |
|
| 1122 | |
| 1123 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE
|
| 1124 | * because it is sized independent of architecture. Unlike the other zones,
|
| 1125 | * the starting point for ZONE_MOVABLE is not fixed. It may be different
|
| 1126 | * in each node depending on the size of each node and how evenly kernelcore
|
| 1127 | * is distributed. This helper function adjusts the zone ranges
|
| 1128 | * provided by the architecture for a given node by using the end of the
|
| 1129 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
|
| 1130 | * zones within a node are in order of monotonic increases memory addresses
|
| 1131 |
|
| 1132 | static void __init adjust_zone_range_for_zone_movable(int nid,
|
| 1133 | unsigned long zone_type,
|
| 1134 | unsigned long node_end_pfn,
|
| 1135 | unsigned long *zone_start_pfn,
|
| 1136 | unsigned long *zone_end_pfn)
|
| 1137 | {
|
| 1138 |
|
| 1139 | if (zone_movable_pfn[nid]) {
|
| 1140 |
|
| 1141 | if (zone_type == ZONE_MOVABLE) {
|
| 1142 | *zone_start_pfn = zone_movable_pfn[nid];
|
| 1143 | *zone_end_pfn = min(node_end_pfn,
|
| 1144 | arch_zone_highest_possible_pfn[movable_zone]);
|
| 1145 |
|
| 1146 |
|
| 1147 | } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
|
| 1148 | *zone_end_pfn > zone_movable_pfn[nid]) {
|
| 1149 | *zone_end_pfn = zone_movable_pfn[nid];
|
| 1150 |
|
| 1151 |
|
| 1152 | } else if (*zone_start_pfn >= zone_movable_pfn[nid])
|
| 1153 | *zone_start_pfn = *zone_end_pfn;
|
| 1154 | }
|
| 1155 | }
|
| 1156 |
|
| 1157 | |
| 1158 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
|
| 1159 | * then all holes in the requested range will be accounted for.
|
| 1160 |
|
| 1161 | static unsigned long __init __absent_pages_in_range(int nid,
|
| 1162 | unsigned long range_start_pfn,
|
| 1163 | unsigned long range_end_pfn)
|
| 1164 | {
|
| 1165 | unsigned long nr_absent = range_end_pfn - range_start_pfn;
|
| 1166 | unsigned long start_pfn, end_pfn;
|
| 1167 | int i;
|
| 1168 |
|
| 1169 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
|
| 1170 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
|
| 1171 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
|
| 1172 | nr_absent -= end_pfn - start_pfn;
|
| 1173 | }
|
| 1174 | return nr_absent;
|
| 1175 | }
|
| 1176 |
|
| 1177 | |
| 1178 | * absent_pages_in_range - Return number of page frames in holes within a range
|
| 1179 | * @start_pfn: The start PFN to start searching for holes
|
| 1180 | * @end_pfn: The end PFN to stop searching for holes
|
| 1181 | *
|
| 1182 | * Return: the number of pages frames in memory holes within a range.
|
| 1183 |
|
| 1184 | unsigned long __init absent_pages_in_range(unsigned long start_pfn,
|
| 1185 | unsigned long end_pfn)
|
| 1186 | {
|
| 1187 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
|
| 1188 | }
|
| 1189 |
|
| 1190 |
|
| 1191 | static unsigned long __init zone_absent_pages_in_node(int nid,
|
| 1192 | unsigned long zone_type,
|
| 1193 | unsigned long zone_start_pfn,
|
| 1194 | unsigned long zone_end_pfn)
|
| 1195 | {
|
| 1196 |
|
| 1197 | if (zone_start_pfn == zone_end_pfn)
|
| 1198 | return 0;
|
| 1199 |
|
| 1200 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
|
| 1201 | }
|
| 1202 |
|
| 1203 | |
| 1204 | * Return the number of pages a zone spans in a node, including holes
|
| 1205 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
|
| 1206 |
|
| 1207 | static unsigned long __init zone_spanned_pages_in_node(int nid,
|
| 1208 | unsigned long zone_type,
|
| 1209 | unsigned long node_start_pfn,
|
| 1210 | unsigned long node_end_pfn,
|
| 1211 | unsigned long *zone_start_pfn,
|
| 1212 | unsigned long *zone_end_pfn)
|
| 1213 | {
|
| 1214 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
|
| 1215 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
|
| 1216 |
|
| 1217 |
|
| 1218 | *zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
|
| 1219 | *zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
|
| 1220 | adjust_zone_range_for_zone_movable(nid, zone_type, node_end_pfn,
|
| 1221 | zone_start_pfn, zone_end_pfn);
|
| 1222 |
|
| 1223 |
|
| 1224 | if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
|
| 1225 | return 0;
|
| 1226 |
|
| 1227 |
|
| 1228 | *zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
|
| 1229 | *zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
|
| 1230 |
|
| 1231 |
|
| 1232 | return *zone_end_pfn - *zone_start_pfn;
|
| 1233 | }
|
| 1234 |
|
| 1235 | static void __init reset_memoryless_node_totalpages(struct pglist_data *pgdat)
|
| 1236 | {
|
| 1237 | struct zone *z;
|
| 1238 |
|
| 1239 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) {
|
| 1240 | z->zone_start_pfn = 0;
|
| 1241 | z->spanned_pages = 0;
|
| 1242 | z->present_pages = 0;
|
| 1243 | #if defined(CONFIG_MEMORY_HOTPLUG)
|
| 1244 | z->present_early_pages = 0;
|
| 1245 | #endif
|
| 1246 | }
|
| 1247 |
|
| 1248 | pgdat->node_spanned_pages = 0;
|
| 1249 | pgdat->node_present_pages = 0;
|
| 1250 | pr_debug("On node %d totalpages: 0\n", pgdat->node_id);
|
| 1251 | }
|
| 1252 |
|
| 1253 | static void __init calc_nr_kernel_pages(void)
|
| 1254 | {
|
| 1255 | unsigned long start_pfn, end_pfn;
|
| 1256 | phys_addr_t start_addr, end_addr;
|
| 1257 | u64 u;
|
| 1258 | #ifdef CONFIG_HIGHMEM
|
| 1259 | unsigned long high_zone_low = arch_zone_lowest_possible_pfn[ZONE_HIGHMEM];
|
| 1260 | #endif
|
| 1261 |
|
| 1262 | for_each_free_mem_range(u, NUMA_NO_NODE, MEMBLOCK_NONE, &start_addr, &end_addr, NULL) {
|
| 1263 | start_pfn = PFN_UP(start_addr);
|
| 1264 | end_pfn = PFN_DOWN(end_addr);
|
| 1265 |
|
| 1266 | if (start_pfn < end_pfn) {
|
| 1267 | nr_all_pages += end_pfn - start_pfn;
|
| 1268 | #ifdef CONFIG_HIGHMEM
|
| 1269 | start_pfn = clamp(start_pfn, 0, high_zone_low);
|
| 1270 | end_pfn = clamp(end_pfn, 0, high_zone_low);
|
| 1271 | #endif
|
| 1272 | nr_kernel_pages += end_pfn - start_pfn;
|
| 1273 | }
|
| 1274 | }
|
| 1275 | }
|
| 1276 |
|
| 1277 | static void __init calculate_node_totalpages(struct pglist_data *pgdat,
|
| 1278 | unsigned long node_start_pfn,
|
| 1279 | unsigned long node_end_pfn)
|
| 1280 | {
|
| 1281 | unsigned long realtotalpages = 0;
|
| 1282 | enum zone_type i;
|
| 1283 |
|
| 1284 | for (i = 0; i < MAX_NR_ZONES; i++) {
|
| 1285 | struct zone *zone = pgdat->node_zones + i;
|
| 1286 | unsigned long zone_start_pfn, zone_end_pfn;
|
| 1287 | unsigned long spanned, absent;
|
| 1288 | unsigned long real_size;
|
| 1289 |
|
| 1290 | spanned = zone_spanned_pages_in_node(pgdat->node_id, i,
|
| 1291 | node_start_pfn,
|
| 1292 | node_end_pfn,
|
| 1293 | &zone_start_pfn,
|
| 1294 | &zone_end_pfn);
|
| 1295 | absent = zone_absent_pages_in_node(pgdat->node_id, i,
|
| 1296 | zone_start_pfn,
|
| 1297 | zone_end_pfn);
|
| 1298 |
|
| 1299 | real_size = spanned - absent;
|
| 1300 |
|
| 1301 | if (spanned)
|
| 1302 | zone->zone_start_pfn = zone_start_pfn;
|
| 1303 | else
|
| 1304 | zone->zone_start_pfn = 0;
|
| 1305 | zone->spanned_pages = spanned;
|
| 1306 | zone->present_pages = real_size;
|
| 1307 | #if defined(CONFIG_MEMORY_HOTPLUG)
|
| 1308 | zone->present_early_pages = real_size;
|
| 1309 | #endif
|
| 1310 |
|
| 1311 | realtotalpages += real_size;
|
| 1312 | }
|
| 1313 |
|
| 1314 | pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
|
| 1315 | pgdat->node_present_pages = realtotalpages;
|
| 1316 | pr_debug("On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
|
| 1317 | }
|
| 1318 |
|
| 1319 | #ifdef CONFIG_COMPACTION
|
| 1320 | static void pgdat_init_kcompactd(struct pglist_data *pgdat)
|
| 1321 | {
|
| 1322 | init_waitqueue_head(&pgdat->kcompactd_wait);
|
| 1323 | }
|
| 1324 | #else
|
| 1325 | static void pgdat_init_kcompactd(struct pglist_data *pgdat) {}
|
| 1326 | #endif
|
| 1327 |
|
| 1328 | static void __meminit pgdat_init_internals(struct pglist_data *pgdat)
|
| 1329 | {
|
| 1330 | int i;
|
| 1331 |
|
| 1332 | pgdat_resize_init(pgdat);
|
| 1333 | pgdat_kswapd_lock_init(pgdat);
|
| 1334 | pgdat_init_kcompactd(pgdat);
|
| 1335 |
|
| 1336 | init_waitqueue_head(&pgdat->kswapd_wait);
|
| 1337 | init_waitqueue_head(&pgdat->pfmemalloc_wait);
|
| 1338 |
|
| 1339 | for (i = 0; i < NR_VMSCAN_THROTTLE; i++)
|
| 1340 | init_waitqueue_head(&pgdat->reclaim_wait[i]);
|
| 1341 |
|
| 1342 | lruvec_init(&pgdat->__lruvec);
|
| 1343 | }
|
| 1344 |
|
| 1345 | static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid,
|
| 1346 | unsigned long remaining_pages)
|
| 1347 | {
|
| 1348 | atomic_long_set(&zone->managed_pages, remaining_pages);
|
| 1349 | zone_set_nid(zone, nid);
|
| 1350 | zone->name = zone_names[idx];
|
| 1351 | zone->zone_pgdat = NODE_DATA(nid);
|
| 1352 | spin_lock_init(&zone->lock);
|
| 1353 | zone_seqlock_init(zone);
|
| 1354 | zone_pcp_init(zone);
|
| 1355 | }
|
| 1356 |
|
| 1357 | static void __meminit zone_init_free_lists(struct zone *zone)
|
| 1358 | {
|
| 1359 | struct list_head *list;
|
| 1360 | unsigned int order;
|
| 1361 |
|
| 1362 | for_each_free_list(list, zone, order)
|
| 1363 | INIT_LIST_HEAD(list);
|
| 1364 |
|
| 1365 | for (order = 0; order < NR_PAGE_ORDERS; order++)
|
| 1366 | zone->free_area[order].nr_free = 0;
|
| 1367 |
|
| 1368 | #ifdef CONFIG_UNACCEPTED_MEMORY
|
| 1369 | INIT_LIST_HEAD(&zone->unaccepted_pages);
|
| 1370 | #endif
|
| 1371 | }
|
| 1372 |
|
| 1373 | void __meminit init_currently_empty_zone(struct zone *zone,
|
| 1374 | unsigned long zone_start_pfn,
|
| 1375 | unsigned long size)
|
| 1376 | {
|
| 1377 | struct pglist_data *pgdat = zone->zone_pgdat;
|
| 1378 | int zone_idx = zone_idx(zone) + 1;
|
| 1379 |
|
| 1380 | if (zone_idx > pgdat->nr_zones)
|
| 1381 | pgdat->nr_zones = zone_idx;
|
| 1382 |
|
| 1383 | zone->zone_start_pfn = zone_start_pfn;
|
| 1384 |
|
| 1385 | mminit_dprintk(MMINIT_TRACE, "memmap_init",
|
| 1386 | "Initialising map node %d zone %lu pfns %lu -> %lu\n",
|
| 1387 | pgdat->node_id,
|
| 1388 | (unsigned long)zone_idx(zone),
|
| 1389 | zone_start_pfn, (zone_start_pfn + size));
|
| 1390 |
|
| 1391 | zone_init_free_lists(zone);
|
| 1392 | zone->initialized = 1;
|
| 1393 | }
|
| 1394 |
|
| 1395 | #ifndef CONFIG_SPARSEMEM
|
| 1396 | |
| 1397 | * Calculate the size of the zone->pageblock_flags rounded to an unsigned long
|
| 1398 | * Start by making sure zonesize is a multiple of pageblock_order by rounding
|
| 1399 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
|
| 1400 | * round what is now in bits to nearest long in bits, then return it in
|
| 1401 | * bytes.
|
| 1402 |
|
| 1403 | static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
|
| 1404 | {
|
| 1405 | unsigned long usemapsize;
|
| 1406 |
|
| 1407 | zonesize += zone_start_pfn & (pageblock_nr_pages-1);
|
| 1408 | usemapsize = round_up(zonesize, pageblock_nr_pages);
|
| 1409 | usemapsize = usemapsize >> pageblock_order;
|
| 1410 | usemapsize *= NR_PAGEBLOCK_BITS;
|
| 1411 | usemapsize = round_up(usemapsize, BITS_PER_LONG);
|
| 1412 |
|
| 1413 | return usemapsize / BITS_PER_BYTE;
|
| 1414 | }
|
| 1415 |
|
| 1416 | static void __ref setup_usemap(struct zone *zone)
|
| 1417 | {
|
| 1418 | unsigned long usemapsize = usemap_size(zone->zone_start_pfn,
|
| 1419 | zone->spanned_pages);
|
| 1420 | zone->pageblock_flags = NULL;
|
| 1421 | if (usemapsize) {
|
| 1422 | zone->pageblock_flags =
|
| 1423 | memblock_alloc_node(usemapsize, SMP_CACHE_BYTES,
|
| 1424 | zone_to_nid(zone));
|
| 1425 | if (!zone->pageblock_flags)
|
| 1426 | panic("Failed to allocate %ld bytes for zone %s pageblock flags on node %d\n",
|
| 1427 | usemapsize, zone->name, zone_to_nid(zone));
|
| 1428 | }
|
| 1429 | }
|
| 1430 | #else
|
| 1431 | static inline void setup_usemap(struct zone *zone) {}
|
| 1432 | #endif
|
| 1433 |
|
| 1434 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
|
| 1435 |
|
| 1436 |
|
| 1437 | static void __init set_pageblock_order(void)
|
| 1438 | {
|
| 1439 | unsigned int order = PAGE_BLOCK_MAX_ORDER;
|
| 1440 |
|
| 1441 |
|
| 1442 | if (pageblock_order)
|
| 1443 | return;
|
| 1444 |
|
| 1445 |
|
| 1446 | if (HPAGE_SHIFT > PAGE_SHIFT && HUGETLB_PAGE_ORDER < order)
|
| 1447 | order = HUGETLB_PAGE_ORDER;
|
| 1448 |
|
| 1449 | |
| 1450 | * Assume the largest contiguous order of interest is a huge page.
|
| 1451 | * This value may be variable depending on boot parameters on powerpc.
|
| 1452 |
|
| 1453 | pageblock_order = order;
|
| 1454 | }
|
| 1455 | #else
|
| 1456 |
|
| 1457 | |
| 1458 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
|
| 1459 | * is unused as pageblock_order is set at compile-time. See
|
| 1460 | * include/linux/pageblock-flags.h for the values of pageblock_order based on
|
| 1461 | * the kernel config
|
| 1462 |
|
| 1463 | static inline void __init set_pageblock_order(void)
|
| 1464 | {
|
| 1465 | }
|
| 1466 |
|
| 1467 | #endif
|
| 1468 |
|
| 1469 | |
| 1470 | * Set up the zone data structures
|
| 1471 | * - init pgdat internals
|
| 1472 | * - init all zones belonging to this node
|
| 1473 | *
|
| 1474 | * NOTE: this function is only called during memory hotplug
|
| 1475 |
|
| 1476 | #ifdef CONFIG_MEMORY_HOTPLUG
|
| 1477 | int __ref free_area_init_core_hotplug(struct pglist_data *pgdat)
|
| 1478 | {
|
| 1479 | int nid = pgdat->node_id;
|
| 1480 | enum zone_type z;
|
| 1481 | int cpu, i;
|
| 1482 |
|
| 1483 | pgdat_init_internals(pgdat);
|
| 1484 |
|
| 1485 | if (pgdat->per_cpu_nodestats == &boot_nodestats) {
|
| 1486 | struct per_cpu_nodestat __percpu *p;
|
| 1487 |
|
| 1488 | p = alloc_percpu(struct per_cpu_nodestat);
|
| 1489 | if (!p)
|
| 1490 | return -ENOMEM;
|
| 1491 | pgdat->per_cpu_nodestats = p;
|
| 1492 | }
|
| 1493 |
|
| 1494 | |
| 1495 | * Reset the nr_zones, order and highest_zoneidx before reuse.
|
| 1496 | * Note that kswapd will init kswapd_highest_zoneidx properly
|
| 1497 | * when it starts in the near future.
|
| 1498 |
|
| 1499 | pgdat->nr_zones = 0;
|
| 1500 | pgdat->kswapd_order = 0;
|
| 1501 | pgdat->kswapd_highest_zoneidx = 0;
|
| 1502 | pgdat->node_start_pfn = 0;
|
| 1503 | pgdat->node_present_pages = 0;
|
| 1504 |
|
| 1505 | |
| 1506 | * Hot-unplug can leave per-cpu vmstat deltas unfolded (folders skip
|
| 1507 | * offline nodes) - reconcile this at online. Foreign access to counters
|
| 1508 | * is safe: the node is not online yet and we hold the hotplug lock.
|
| 1509 |
|
| 1510 | for_each_possible_cpu(cpu) {
|
| 1511 | struct per_cpu_nodestat *p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
|
| 1512 |
|
| 1513 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
|
| 1514 | if (p->vm_node_stat_diff[i])
|
| 1515 | node_page_state_add(p->vm_node_stat_diff[i], pgdat, i);
|
| 1516 | memset(p, 0, sizeof(*p));
|
| 1517 | }
|
| 1518 |
|
| 1519 | |
| 1520 | * When memory is hot-added, all the memory is in offline state. So
|
| 1521 | * clear all zones' present_pages and managed_pages because they will
|
| 1522 | * be updated in online_pages() and offline_pages().
|
| 1523 |
|
| 1524 | for (z = 0; z < MAX_NR_ZONES; z++) {
|
| 1525 | struct zone *zone = pgdat->node_zones + z;
|
| 1526 |
|
| 1527 | zone->present_pages = 0;
|
| 1528 | zone_init_internals(zone, z, nid, 0);
|
| 1529 | }
|
| 1530 |
|
| 1531 | return 0;
|
| 1532 | }
|
| 1533 | #endif
|
| 1534 |
|
| 1535 | static void __init free_area_init_core(struct pglist_data *pgdat)
|
| 1536 | {
|
| 1537 | enum zone_type j;
|
| 1538 | int nid = pgdat->node_id;
|
| 1539 |
|
| 1540 | pgdat_init_internals(pgdat);
|
| 1541 | pgdat->per_cpu_nodestats = &boot_nodestats;
|
| 1542 |
|
| 1543 | for (j = 0; j < MAX_NR_ZONES; j++) {
|
| 1544 | struct zone *zone = pgdat->node_zones + j;
|
| 1545 | unsigned long size = zone->spanned_pages;
|
| 1546 |
|
| 1547 | |
| 1548 | * Initialize zone->managed_pages as 0 , it will be reset
|
| 1549 | * when memblock allocator frees pages into buddy system.
|
| 1550 |
|
| 1551 | zone_init_internals(zone, j, nid, zone->present_pages);
|
| 1552 |
|
| 1553 | if (!size)
|
| 1554 | continue;
|
| 1555 |
|
| 1556 | setup_usemap(zone);
|
| 1557 | init_currently_empty_zone(zone, zone->zone_start_pfn, size);
|
| 1558 | }
|
| 1559 | }
|
| 1560 |
|
| 1561 | void __init *memmap_alloc(phys_addr_t size, phys_addr_t align,
|
| 1562 | phys_addr_t min_addr, int nid, bool exact_nid)
|
| 1563 | {
|
| 1564 | void *ptr;
|
| 1565 |
|
| 1566 | |
| 1567 | * Kmemleak will explicitly scan mem_map by traversing all valid
|
| 1568 | * `struct *page`,so memblock does not need to be added to the scan list.
|
| 1569 |
|
| 1570 | if (exact_nid)
|
| 1571 | ptr = memblock_alloc_exact_nid_raw(size, align, min_addr,
|
| 1572 | MEMBLOCK_ALLOC_NOLEAKTRACE,
|
| 1573 | nid);
|
| 1574 | else
|
| 1575 | ptr = memblock_alloc_try_nid_raw(size, align, min_addr,
|
| 1576 | MEMBLOCK_ALLOC_NOLEAKTRACE,
|
| 1577 | nid);
|
| 1578 |
|
| 1579 | if (ptr && size > 0)
|
| 1580 | page_init_poison(ptr, size);
|
| 1581 |
|
| 1582 | return ptr;
|
| 1583 | }
|
| 1584 |
|
| 1585 | #ifdef CONFIG_FLATMEM
|
| 1586 | static void __init alloc_node_mem_map(struct pglist_data *pgdat)
|
| 1587 | {
|
| 1588 | unsigned long start, offset, size, end;
|
| 1589 | struct page *map;
|
| 1590 |
|
| 1591 |
|
| 1592 | if (!pgdat->node_spanned_pages)
|
| 1593 | return;
|
| 1594 |
|
| 1595 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
|
| 1596 | offset = pgdat->node_start_pfn - start;
|
| 1597 | |
| 1598 | * The zone's endpoints aren't required to be MAX_PAGE_ORDER
|
| 1599 | * aligned but the node_mem_map endpoints must be in order
|
| 1600 | * for the buddy allocator to function correctly.
|
| 1601 |
|
| 1602 | end = ALIGN(pgdat_end_pfn(pgdat), MAX_ORDER_NR_PAGES);
|
| 1603 | size = (end - start) * sizeof(struct page);
|
| 1604 | map = memmap_alloc(size, SMP_CACHE_BYTES, MEMBLOCK_LOW_LIMIT,
|
| 1605 | pgdat->node_id, false);
|
| 1606 | if (!map)
|
| 1607 | panic("Failed to allocate %ld bytes for node %d memory map\n",
|
| 1608 | size, pgdat->node_id);
|
| 1609 | pgdat->node_mem_map = map + offset;
|
| 1610 | memmap_boot_pages_add(DIV_ROUND_UP(size, PAGE_SIZE));
|
| 1611 | pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n",
|
| 1612 | __func__, pgdat->node_id, (unsigned long)pgdat,
|
| 1613 | (unsigned long)pgdat->node_mem_map);
|
| 1614 |
|
| 1615 |
|
| 1616 | WARN_ON(pgdat != NODE_DATA(0));
|
| 1617 |
|
| 1618 | mem_map = pgdat->node_mem_map;
|
| 1619 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
|
| 1620 | mem_map -= offset;
|
| 1621 |
|
| 1622 | max_mapnr = end - start;
|
| 1623 | }
|
| 1624 | #else
|
| 1625 | static inline void alloc_node_mem_map(struct pglist_data *pgdat) { }
|
| 1626 | #endif
|
| 1627 |
|
| 1628 | |
| 1629 | * get_pfn_range_for_nid - Return the start and end page frames for a node
|
| 1630 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
|
| 1631 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
|
| 1632 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
|
| 1633 | *
|
| 1634 | * It returns the start and end page frame of a node based on information
|
| 1635 | * provided by memblock_set_node(). If called for a node
|
| 1636 | * with no available memory, the start and end PFNs will be 0.
|
| 1637 |
|
| 1638 | void __init get_pfn_range_for_nid(unsigned int nid,
|
| 1639 | unsigned long *start_pfn, unsigned long *end_pfn)
|
| 1640 | {
|
| 1641 | unsigned long this_start_pfn, this_end_pfn;
|
| 1642 | int i;
|
| 1643 |
|
| 1644 | *start_pfn = -1UL;
|
| 1645 | *end_pfn = 0;
|
| 1646 |
|
| 1647 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
|
| 1648 | *start_pfn = min(*start_pfn, this_start_pfn);
|
| 1649 | *end_pfn = max(*end_pfn, this_end_pfn);
|
| 1650 | }
|
| 1651 |
|
| 1652 | if (*start_pfn == -1UL)
|
| 1653 | *start_pfn = 0;
|
| 1654 | }
|
| 1655 |
|
| 1656 | static void __init free_area_init_node(int nid)
|
| 1657 | {
|
| 1658 | pg_data_t *pgdat = NODE_DATA(nid);
|
| 1659 | unsigned long start_pfn = 0;
|
| 1660 | unsigned long end_pfn = 0;
|
| 1661 |
|
| 1662 |
|
| 1663 | WARN_ON(pgdat->nr_zones || pgdat->kswapd_highest_zoneidx);
|
| 1664 |
|
| 1665 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
|
| 1666 |
|
| 1667 | pgdat->node_id = nid;
|
| 1668 | pgdat->node_start_pfn = start_pfn;
|
| 1669 |
|
| 1670 | if (start_pfn != end_pfn) {
|
| 1671 | pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
|
| 1672 | (u64)start_pfn << PAGE_SHIFT,
|
| 1673 | end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
|
| 1674 |
|
| 1675 | calculate_node_totalpages(pgdat, start_pfn, end_pfn);
|
| 1676 | } else {
|
| 1677 | pr_info("Initmem setup node %d as memoryless\n", nid);
|
| 1678 |
|
| 1679 | reset_memoryless_node_totalpages(pgdat);
|
| 1680 | }
|
| 1681 |
|
| 1682 | alloc_node_mem_map(pgdat);
|
| 1683 | pgdat_set_deferred_range(pgdat);
|
| 1684 |
|
| 1685 | free_area_init_core(pgdat);
|
| 1686 | lru_gen_init_pgdat(pgdat);
|
| 1687 | }
|
| 1688 |
|
| 1689 |
|
| 1690 | static void __init check_for_memory(pg_data_t *pgdat)
|
| 1691 | {
|
| 1692 | enum zone_type zone_type;
|
| 1693 |
|
| 1694 | for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
|
| 1695 | struct zone *zone = &pgdat->node_zones[zone_type];
|
| 1696 | if (populated_zone(zone)) {
|
| 1697 | if (IS_ENABLED(CONFIG_HIGHMEM))
|
| 1698 | node_set_state(pgdat->node_id, N_HIGH_MEMORY);
|
| 1699 | if (zone_type <= ZONE_NORMAL)
|
| 1700 | node_set_state(pgdat->node_id, N_NORMAL_MEMORY);
|
| 1701 | break;
|
| 1702 | }
|
| 1703 | }
|
| 1704 | }
|
| 1705 |
|
| 1706 | #if MAX_NUMNODES > 1
|
| 1707 | |
| 1708 | * Figure out the number of possible node ids.
|
| 1709 |
|
| 1710 | void __init setup_nr_node_ids(void)
|
| 1711 | {
|
| 1712 | unsigned int highest;
|
| 1713 |
|
| 1714 | highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
|
| 1715 | nr_node_ids = highest + 1;
|
| 1716 | }
|
| 1717 | #endif
|
| 1718 |
|
| 1719 | |
| 1720 | * Some architectures, e.g. ARC may have ZONE_HIGHMEM below ZONE_NORMAL. For
|
| 1721 | * such cases we allow max_zone_pfn sorted in the descending order
|
| 1722 |
|
| 1723 | static bool arch_has_descending_max_zone_pfns(void)
|
| 1724 | {
|
| 1725 | return IS_ENABLED(CONFIG_ARC) && !IS_ENABLED(CONFIG_ARC_HAS_PAE40);
|
| 1726 | }
|
| 1727 |
|
| 1728 | static void __init set_high_memory(void)
|
| 1729 | {
|
| 1730 | phys_addr_t highmem = memblock_end_of_DRAM();
|
| 1731 |
|
| 1732 | |
| 1733 | * Some architectures (e.g. ARM) set high_memory very early and
|
| 1734 | * use it in arch setup code.
|
| 1735 | * If an architecture already set high_memory don't overwrite it
|
| 1736 |
|
| 1737 | if (high_memory)
|
| 1738 | return;
|
| 1739 |
|
| 1740 | #ifdef CONFIG_HIGHMEM
|
| 1741 | if (arch_has_descending_max_zone_pfns() ||
|
| 1742 | highmem > PFN_PHYS(arch_zone_lowest_possible_pfn[ZONE_HIGHMEM]))
|
| 1743 | highmem = PFN_PHYS(arch_zone_lowest_possible_pfn[ZONE_HIGHMEM]);
|
| 1744 | #endif
|
| 1745 |
|
| 1746 | high_memory = phys_to_virt(highmem - 1) + 1;
|
| 1747 | }
|
| 1748 |
|
| 1749 | |
| 1750 | * free_area_init - Initialise all pg_data_t and zone data
|
| 1751 | *
|
| 1752 | * This will call free_area_init_node() for each active node in the system.
|
| 1753 | * Using the page ranges provided by memblock_set_node(), the size of each
|
| 1754 | * zone in each node and their holes is calculated. If the maximum PFN
|
| 1755 | * between two adjacent zones match, it is assumed that the zone is empty.
|
| 1756 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
|
| 1757 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
|
| 1758 | * starts where the previous one ended. For example, ZONE_DMA32 starts
|
| 1759 | * at arch_max_dma_pfn.
|
| 1760 |
|
| 1761 | static void __init free_area_init(void)
|
| 1762 | {
|
| 1763 | unsigned long max_zone_pfn[MAX_NR_ZONES] = { 0 };
|
| 1764 | unsigned long start_pfn, end_pfn;
|
| 1765 | int i, nid, zone;
|
| 1766 | bool descending;
|
| 1767 |
|
| 1768 | arch_zone_limits_init(max_zone_pfn);
|
| 1769 |
|
| 1770 | start_pfn = PHYS_PFN(memblock_start_of_DRAM());
|
| 1771 | descending = arch_has_descending_max_zone_pfns();
|
| 1772 |
|
| 1773 | for (i = 0; i < MAX_NR_ZONES; i++) {
|
| 1774 | if (descending)
|
| 1775 | zone = MAX_NR_ZONES - i - 1;
|
| 1776 | else
|
| 1777 | zone = i;
|
| 1778 |
|
| 1779 | if (zone == ZONE_MOVABLE)
|
| 1780 | continue;
|
| 1781 |
|
| 1782 | end_pfn = max(max_zone_pfn[zone], start_pfn);
|
| 1783 | arch_zone_lowest_possible_pfn[zone] = start_pfn;
|
| 1784 | arch_zone_highest_possible_pfn[zone] = end_pfn;
|
| 1785 |
|
| 1786 | start_pfn = end_pfn;
|
| 1787 | }
|
| 1788 |
|
| 1789 |
|
| 1790 | find_zone_movable_pfns_for_nodes();
|
| 1791 |
|
| 1792 |
|
| 1793 | pr_info("Zone ranges:\n");
|
| 1794 | for (i = 0; i < MAX_NR_ZONES; i++) {
|
| 1795 | if (i == ZONE_MOVABLE)
|
| 1796 | continue;
|
| 1797 | pr_info(" %-8s ", zone_names[i]);
|
| 1798 | if (arch_zone_lowest_possible_pfn[i] ==
|
| 1799 | arch_zone_highest_possible_pfn[i])
|
| 1800 | pr_cont("empty\n");
|
| 1801 | else
|
| 1802 | pr_cont("[mem %#018Lx-%#018Lx]\n",
|
| 1803 | (u64)arch_zone_lowest_possible_pfn[i]
|
| 1804 | << PAGE_SHIFT,
|
| 1805 | ((u64)arch_zone_highest_possible_pfn[i]
|
| 1806 | << PAGE_SHIFT) - 1);
|
| 1807 | }
|
| 1808 |
|
| 1809 |
|
| 1810 | pr_info("Movable zone start for each node\n");
|
| 1811 | for (i = 0; i < MAX_NUMNODES; i++) {
|
| 1812 | if (zone_movable_pfn[i])
|
| 1813 | pr_info(" Node %d: %#018Lx\n", i,
|
| 1814 | (u64)zone_movable_pfn[i] << PAGE_SHIFT);
|
| 1815 | }
|
| 1816 |
|
| 1817 |
|
| 1818 | pr_info("Early memory node ranges\n");
|
| 1819 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
|
| 1820 | pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
|
| 1821 | (u64)start_pfn << PAGE_SHIFT,
|
| 1822 | ((u64)end_pfn << PAGE_SHIFT) - 1);
|
| 1823 |
|
| 1824 |
|
| 1825 | mminit_verify_pageflags_layout();
|
| 1826 | setup_nr_node_ids();
|
| 1827 | set_pageblock_order();
|
| 1828 |
|
| 1829 | for_each_node(nid) {
|
| 1830 | pg_data_t *pgdat;
|
| 1831 |
|
| 1832 | |
| 1833 | * If an architecture has not allocated node data for
|
| 1834 | * this node, presume the node is memoryless or offline.
|
| 1835 |
|
| 1836 | if (!NODE_DATA(nid))
|
| 1837 | alloc_offline_node_data(nid);
|
| 1838 |
|
| 1839 | pgdat = NODE_DATA(nid);
|
| 1840 | free_area_init_node(nid);
|
| 1841 |
|
| 1842 | |
| 1843 | * No sysfs hierarchy will be created via register_node()
|
| 1844 | *for memory-less node because here it's not marked as N_MEMORY
|
| 1845 | *and won't be set online later. The benefit is userspace
|
| 1846 | *program won't be confused by sysfs files/directories of
|
| 1847 | *memory-less node. The pgdat will get fully initialized by
|
| 1848 | *hotadd_init_pgdat() when memory is hotplugged into this node.
|
| 1849 |
|
| 1850 | if (pgdat->node_present_pages) {
|
| 1851 | node_set_state(nid, N_MEMORY);
|
| 1852 | check_for_memory(pgdat);
|
| 1853 | }
|
| 1854 | }
|
| 1855 |
|
| 1856 | calc_nr_kernel_pages();
|
| 1857 |
|
| 1858 |
|
| 1859 | fixup_hashdist();
|
| 1860 |
|
| 1861 | set_high_memory();
|
| 1862 | }
|
| 1863 |
|
| 1864 | |
| 1865 | * node_map_pfn_alignment - determine the maximum internode alignment
|
| 1866 | *
|
| 1867 | * This function should be called after node map is populated and sorted.
|
| 1868 | * It calculates the maximum power of two alignment which can distinguish
|
| 1869 | * all the nodes.
|
| 1870 | *
|
| 1871 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
|
| 1872 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
|
| 1873 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
|
| 1874 | * shifted, 1GiB is enough and this function will indicate so.
|
| 1875 | *
|
| 1876 | * This is used to test whether pfn -> nid mapping of the chosen memory
|
| 1877 | * model has fine enough granularity to avoid incorrect mapping for the
|
| 1878 | * populated node map.
|
| 1879 | *
|
| 1880 | * Return: the determined alignment in pfn's. 0 if there is no alignment
|
| 1881 | * requirement (single node).
|
| 1882 |
|
| 1883 | unsigned long __init node_map_pfn_alignment(void)
|
| 1884 | {
|
| 1885 | unsigned long accl_mask = 0, last_end = 0;
|
| 1886 | unsigned long start, end, mask;
|
| 1887 | int last_nid = NUMA_NO_NODE;
|
| 1888 | int i, nid;
|
| 1889 |
|
| 1890 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
|
| 1891 | if (!start || last_nid < 0 || last_nid == nid) {
|
| 1892 | last_nid = nid;
|
| 1893 | last_end = end;
|
| 1894 | continue;
|
| 1895 | }
|
| 1896 |
|
| 1897 | |
| 1898 | * Start with a mask granular enough to pin-point to the
|
| 1899 | * start pfn and tick off bits one-by-one until it becomes
|
| 1900 | * too coarse to separate the current node from the last.
|
| 1901 |
|
| 1902 | mask = ~((1 << __ffs(start)) - 1);
|
| 1903 | while (mask && last_end <= (start & (mask << 1)))
|
| 1904 | mask <<= 1;
|
| 1905 |
|
| 1906 |
|
| 1907 | accl_mask |= mask;
|
| 1908 | }
|
| 1909 |
|
| 1910 |
|
| 1911 | return ~accl_mask + 1;
|
| 1912 | }
|
| 1913 |
|
| 1914 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
|
| 1915 | static void __init deferred_free_pages(unsigned long pfn,
|
| 1916 | unsigned long nr_pages)
|
| 1917 | {
|
| 1918 | struct page *page;
|
| 1919 | unsigned long i;
|
| 1920 |
|
| 1921 | if (!nr_pages)
|
| 1922 | return;
|
| 1923 |
|
| 1924 | pageblock_migratetype_init_range(pfn, nr_pages, MIGRATE_MOVABLE, true);
|
| 1925 |
|
| 1926 | page = pfn_to_page(pfn);
|
| 1927 |
|
| 1928 |
|
| 1929 | if (nr_pages == MAX_ORDER_NR_PAGES && IS_MAX_ORDER_ALIGNED(pfn)) {
|
| 1930 | __free_pages_core(page, MAX_PAGE_ORDER, MEMINIT_EARLY);
|
| 1931 | return;
|
| 1932 | }
|
| 1933 |
|
| 1934 |
|
| 1935 | accept_memory(PFN_PHYS(pfn), nr_pages * PAGE_SIZE);
|
| 1936 |
|
| 1937 | for (i = 0; i < nr_pages; i++)
|
| 1938 | __free_pages_core(page + i, 0, MEMINIT_EARLY);
|
| 1939 | }
|
| 1940 |
|
| 1941 |
|
| 1942 | static atomic_t pgdat_init_n_undone __initdata;
|
| 1943 | static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
|
| 1944 |
|
| 1945 | static inline void __init pgdat_init_report_one_done(void)
|
| 1946 | {
|
| 1947 | if (atomic_dec_and_test(&pgdat_init_n_undone))
|
| 1948 | complete(&pgdat_init_all_done_comp);
|
| 1949 | }
|
| 1950 |
|
| 1951 | |
| 1952 | * Initialize struct pages. We minimize pfn page lookups and scheduler checks
|
| 1953 | * by performing it only once every MAX_ORDER_NR_PAGES.
|
| 1954 | * Return number of pages initialized.
|
| 1955 |
|
| 1956 | static unsigned long __init deferred_init_pages(struct zone *zone,
|
| 1957 | unsigned long pfn, unsigned long end_pfn)
|
| 1958 | {
|
| 1959 | int nid = zone_to_nid(zone);
|
| 1960 | unsigned long nr_pages = end_pfn - pfn;
|
| 1961 | int zid = zone_idx(zone);
|
| 1962 | struct page *page = pfn_to_page(pfn);
|
| 1963 |
|
| 1964 | for (; pfn < end_pfn; pfn++, page++)
|
| 1965 | __init_single_page(page, pfn, zid, nid);
|
| 1966 | return nr_pages;
|
| 1967 | }
|
| 1968 |
|
| 1969 | |
| 1970 | * Initialize and free pages.
|
| 1971 | *
|
| 1972 | * At this point reserved pages and struct pages that correspond to holes in
|
| 1973 | * memblock.memory are already initialized so every free range has a valid
|
| 1974 | * memory map around it.
|
| 1975 | * This ensures that access of pages that are ahead of the range being
|
| 1976 | * initialized (computing buddy page in __free_one_page()) always reads a valid
|
| 1977 | * struct page.
|
| 1978 | *
|
| 1979 | * In order to try and improve CPU cache locality we have the loop broken along
|
| 1980 | * max page order boundaries.
|
| 1981 |
|
| 1982 | static unsigned long __init
|
| 1983 | deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn,
|
| 1984 | struct zone *zone, bool can_resched)
|
| 1985 | {
|
| 1986 | int nid = zone_to_nid(zone);
|
| 1987 | unsigned long nr_pages = 0;
|
| 1988 | phys_addr_t start, end;
|
| 1989 | u64 i = 0;
|
| 1990 |
|
| 1991 | for_each_free_mem_range(i, nid, 0, &start, &end, NULL) {
|
| 1992 | unsigned long spfn = PFN_UP(start);
|
| 1993 | unsigned long epfn = PFN_DOWN(end);
|
| 1994 |
|
| 1995 | if (spfn >= end_pfn)
|
| 1996 | break;
|
| 1997 |
|
| 1998 | spfn = max(spfn, start_pfn);
|
| 1999 | epfn = min(epfn, end_pfn);
|
| 2000 |
|
| 2001 | while (spfn < epfn) {
|
| 2002 | unsigned long mo_pfn = ALIGN(spfn + 1, MAX_ORDER_NR_PAGES);
|
| 2003 | unsigned long chunk_end = min(mo_pfn, epfn);
|
| 2004 |
|
| 2005 | nr_pages += deferred_init_pages(zone, spfn, chunk_end);
|
| 2006 | deferred_free_pages(spfn, chunk_end - spfn);
|
| 2007 |
|
| 2008 | spfn = chunk_end;
|
| 2009 |
|
| 2010 | if (can_resched)
|
| 2011 | cond_resched();
|
| 2012 | else
|
| 2013 | touch_nmi_watchdog();
|
| 2014 | }
|
| 2015 | }
|
| 2016 |
|
| 2017 | return nr_pages;
|
| 2018 | }
|
| 2019 |
|
| 2020 | static void __init
|
| 2021 | deferred_init_memmap_job(unsigned long start_pfn, unsigned long end_pfn,
|
| 2022 | void *arg)
|
| 2023 | {
|
| 2024 | struct zone *zone = arg;
|
| 2025 |
|
| 2026 | deferred_init_memmap_chunk(start_pfn, end_pfn, zone, true);
|
| 2027 | }
|
| 2028 |
|
| 2029 | static unsigned int __init
|
| 2030 | deferred_page_init_max_threads(const struct cpumask *node_cpumask)
|
| 2031 | {
|
| 2032 | return max(cpumask_weight(node_cpumask), 1U);
|
| 2033 | }
|
| 2034 |
|
| 2035 |
|
| 2036 | static int __init deferred_init_memmap(void *data)
|
| 2037 | {
|
| 2038 | pg_data_t *pgdat = data;
|
| 2039 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
|
| 2040 | int max_threads = deferred_page_init_max_threads(cpumask);
|
| 2041 | unsigned long first_init_pfn, last_pfn, flags;
|
| 2042 | unsigned long start = jiffies;
|
| 2043 | struct zone *zone;
|
| 2044 |
|
| 2045 |
|
| 2046 | if (!cpumask_empty(cpumask))
|
| 2047 | set_cpus_allowed_ptr(current, cpumask);
|
| 2048 |
|
| 2049 | pgdat_resize_lock(pgdat, &flags);
|
| 2050 | first_init_pfn = pgdat->first_deferred_pfn;
|
| 2051 | if (first_init_pfn == ULONG_MAX) {
|
| 2052 | pgdat_resize_unlock(pgdat, &flags);
|
| 2053 | pgdat_init_report_one_done();
|
| 2054 | return 0;
|
| 2055 | }
|
| 2056 |
|
| 2057 |
|
| 2058 | BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
|
| 2059 | BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
|
| 2060 | pgdat->first_deferred_pfn = ULONG_MAX;
|
| 2061 |
|
| 2062 | |
| 2063 | * Once we unlock here, the zone cannot be grown anymore, thus if an
|
| 2064 | * interrupt thread must allocate this early in boot, zone must be
|
| 2065 | * pre-grown prior to start of deferred page initialization.
|
| 2066 |
|
| 2067 | pgdat_resize_unlock(pgdat, &flags);
|
| 2068 |
|
| 2069 |
|
| 2070 | zone = pgdat->node_zones + pgdat->nr_zones - 1;
|
| 2071 | last_pfn = SECTION_ALIGN_UP(zone_end_pfn(zone));
|
| 2072 |
|
| 2073 | struct padata_mt_job job = {
|
| 2074 | .thread_fn = deferred_init_memmap_job,
|
| 2075 | .fn_arg = zone,
|
| 2076 | .start = first_init_pfn,
|
| 2077 | .size = last_pfn - first_init_pfn,
|
| 2078 | .align = PAGES_PER_SECTION,
|
| 2079 | .min_chunk = PAGES_PER_SECTION,
|
| 2080 | .max_threads = max_threads,
|
| 2081 | .numa_aware = false,
|
| 2082 | };
|
| 2083 |
|
| 2084 | padata_do_multithreaded(&job);
|
| 2085 |
|
| 2086 |
|
| 2087 | WARN_ON(pgdat->nr_zones < MAX_NR_ZONES && populated_zone(++zone));
|
| 2088 |
|
| 2089 | pr_info("node %d deferred pages initialised in %ums\n",
|
| 2090 | pgdat->node_id, jiffies_to_msecs(jiffies - start));
|
| 2091 |
|
| 2092 | pgdat_init_report_one_done();
|
| 2093 | return 0;
|
| 2094 | }
|
| 2095 |
|
| 2096 | |
| 2097 | * If this zone has deferred pages, try to grow it by initializing enough
|
| 2098 | * deferred pages to satisfy the allocation specified by order, rounded up to
|
| 2099 | * the nearest PAGES_PER_SECTION boundary. So we're adding memory in increments
|
| 2100 | * of SECTION_SIZE bytes by initializing struct pages in increments of
|
| 2101 | * PAGES_PER_SECTION * sizeof(struct page) bytes.
|
| 2102 | *
|
| 2103 | * Return true when zone was grown, otherwise return false. We return true even
|
| 2104 | * when we grow less than requested, to let the caller decide if there are
|
| 2105 | * enough pages to satisfy the allocation.
|
| 2106 |
|
| 2107 | bool __init deferred_grow_zone(struct zone *zone, unsigned int order)
|
| 2108 | {
|
| 2109 | unsigned long nr_pages_needed = SECTION_ALIGN_UP(1 << order);
|
| 2110 | pg_data_t *pgdat = zone->zone_pgdat;
|
| 2111 | unsigned long first_deferred_pfn = pgdat->first_deferred_pfn;
|
| 2112 | unsigned long spfn, epfn, flags;
|
| 2113 | unsigned long nr_pages = 0;
|
| 2114 |
|
| 2115 |
|
| 2116 | if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat))
|
| 2117 | return false;
|
| 2118 |
|
| 2119 | pgdat_resize_lock(pgdat, &flags);
|
| 2120 |
|
| 2121 | |
| 2122 | * If someone grew this zone while we were waiting for spinlock, return
|
| 2123 | * true, as there might be enough pages already.
|
| 2124 |
|
| 2125 | if (first_deferred_pfn != pgdat->first_deferred_pfn) {
|
| 2126 | pgdat_resize_unlock(pgdat, &flags);
|
| 2127 | return true;
|
| 2128 | }
|
| 2129 |
|
| 2130 | |
| 2131 | * Initialize at least nr_pages_needed in section chunks.
|
| 2132 | * If a section has less free memory than nr_pages_needed, the next
|
| 2133 | * section will be also initialized.
|
| 2134 | * Note, that it still does not guarantee that allocation of order can
|
| 2135 | * be satisfied if the sections are fragmented because of memblock
|
| 2136 | * allocations.
|
| 2137 |
|
| 2138 | for (spfn = first_deferred_pfn, epfn = SECTION_ALIGN_UP(spfn + 1);
|
| 2139 | nr_pages < nr_pages_needed && spfn < zone_end_pfn(zone);
|
| 2140 | spfn = epfn, epfn += PAGES_PER_SECTION) {
|
| 2141 | nr_pages += deferred_init_memmap_chunk(spfn, epfn, zone, false);
|
| 2142 | }
|
| 2143 |
|
| 2144 | |
| 2145 | * The loop only tests spfn before entering an iteration, so on exit it
|
| 2146 | * may point up to a section past the end of the zone. When it does,
|
| 2147 | * the rest of the zone has already been handed to
|
| 2148 | * deferred_init_memmap_chunk() and nothing is left to initialize.
|
| 2149 |
|
| 2150 | pgdat->first_deferred_pfn =
|
| 2151 | spfn < zone_end_pfn(zone) ? spfn : ULONG_MAX;
|
| 2152 |
|
| 2153 | pgdat_resize_unlock(pgdat, &flags);
|
| 2154 |
|
| 2155 | return nr_pages > 0;
|
| 2156 | }
|
| 2157 |
|
| 2158 | #endif
|
| 2159 |
|
| 2160 | #ifdef CONFIG_CMA
|
| 2161 | void __init init_cma_reserved_pageblock(struct page *page)
|
| 2162 | {
|
| 2163 | unsigned i = pageblock_nr_pages;
|
| 2164 | struct page *p = page;
|
| 2165 |
|
| 2166 | do {
|
| 2167 | __ClearPageReserved(p);
|
| 2168 | set_page_count(p, 0);
|
| 2169 | } while (++p, --i);
|
| 2170 |
|
| 2171 | init_pageblock_migratetype(page, MIGRATE_CMA, false);
|
| 2172 | set_page_refcounted(page);
|
| 2173 |
|
| 2174 | clear_page_tag_ref(page);
|
| 2175 | __free_pages(page, pageblock_order);
|
| 2176 |
|
| 2177 | adjust_managed_page_count(page, pageblock_nr_pages);
|
| 2178 | page_zone(page)->cma_pages += pageblock_nr_pages;
|
| 2179 | }
|
| 2180 | |
| 2181 | * Similar to above, but only set the migrate type and stats.
|
| 2182 |
|
| 2183 | void __init init_cma_pageblock(struct page *page)
|
| 2184 | {
|
| 2185 | init_pageblock_migratetype(page, MIGRATE_CMA, false);
|
| 2186 | adjust_managed_page_count(page, pageblock_nr_pages);
|
| 2187 | page_zone(page)->cma_pages += pageblock_nr_pages;
|
| 2188 | }
|
| 2189 | #endif
|
| 2190 |
|
| 2191 | void set_zone_contiguous(struct zone *zone)
|
| 2192 | {
|
| 2193 | unsigned long block_start_pfn = zone->zone_start_pfn;
|
| 2194 | unsigned long block_end_pfn;
|
| 2195 |
|
| 2196 | block_end_pfn = pageblock_end_pfn(block_start_pfn);
|
| 2197 | for (; block_start_pfn < zone_end_pfn(zone);
|
| 2198 | block_start_pfn = block_end_pfn,
|
| 2199 | block_end_pfn += pageblock_nr_pages) {
|
| 2200 |
|
| 2201 | block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
|
| 2202 |
|
| 2203 | if (!__pageblock_pfn_to_page(block_start_pfn,
|
| 2204 | block_end_pfn, zone))
|
| 2205 | return;
|
| 2206 | cond_resched();
|
| 2207 | }
|
| 2208 |
|
| 2209 |
|
| 2210 | zone->contiguous = true;
|
| 2211 | }
|
| 2212 |
|
| 2213 | |
| 2214 | * Check if a PFN range intersects multiple zones on one or more
|
| 2215 | * NUMA nodes. Specify the @nid argument if it is known that this
|
| 2216 | * PFN range is on one node, NUMA_NO_NODE otherwise.
|
| 2217 |
|
| 2218 | bool pfn_range_intersects_zones(int nid, unsigned long start_pfn,
|
| 2219 | unsigned long nr_pages)
|
| 2220 | {
|
| 2221 | struct zone *zone, *izone = NULL;
|
| 2222 |
|
| 2223 | for_each_zone(zone) {
|
| 2224 | if (nid != NUMA_NO_NODE && zone_to_nid(zone) != nid)
|
| 2225 | continue;
|
| 2226 |
|
| 2227 | if (zone_intersects(zone, start_pfn, nr_pages)) {
|
| 2228 | if (izone != NULL)
|
| 2229 | return true;
|
| 2230 | izone = zone;
|
| 2231 | }
|
| 2232 |
|
| 2233 | }
|
| 2234 |
|
| 2235 | return false;
|
| 2236 | }
|
| 2237 |
|
| 2238 | static void __init mem_init_print_info(void);
|
| 2239 | void __init page_alloc_init_late(void)
|
| 2240 | {
|
| 2241 | struct zone *zone;
|
| 2242 | int nid;
|
| 2243 |
|
| 2244 | #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
|
| 2245 |
|
| 2246 |
|
| 2247 | atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
|
| 2248 | for_each_node_state(nid, N_MEMORY) {
|
| 2249 | kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
|
| 2250 | }
|
| 2251 |
|
| 2252 |
|
| 2253 | wait_for_completion(&pgdat_init_all_done_comp);
|
| 2254 |
|
| 2255 | |
| 2256 | * We initialized the rest of the deferred pages. Permanently disable
|
| 2257 | * on-demand struct page initialization.
|
| 2258 |
|
| 2259 | static_branch_disable(&deferred_pages);
|
| 2260 |
|
| 2261 |
|
| 2262 | files_maxfiles_init();
|
| 2263 | #endif
|
| 2264 | hugetlb_bootmem_struct_page_init();
|
| 2265 |
|
| 2266 |
|
| 2267 | mem_init_print_info();
|
| 2268 | buffer_init();
|
| 2269 |
|
| 2270 |
|
| 2271 | memblock_discard();
|
| 2272 |
|
| 2273 | for_each_node_state(nid, N_MEMORY)
|
| 2274 | shuffle_free_memory(NODE_DATA(nid));
|
| 2275 |
|
| 2276 | for_each_populated_zone(zone)
|
| 2277 | set_zone_contiguous(zone);
|
| 2278 |
|
| 2279 |
|
| 2280 | if (deferred_struct_pages)
|
| 2281 | page_ext_init();
|
| 2282 |
|
| 2283 | page_alloc_sysctl_init();
|
| 2284 | }
|
| 2285 |
|
| 2286 | |
| 2287 | * Adaptive scale is meant to reduce sizes of hash tables on large memory
|
| 2288 | * machines. As memory size is increased the scale is also increased but at
|
| 2289 | * slower pace. Starting from ADAPT_SCALE_BASE (64G), every time memory
|
| 2290 | * quadruples the scale is increased by one, which means the size of hash table
|
| 2291 | * only doubles, instead of quadrupling as well.
|
| 2292 | * Because 32-bit systems cannot have large physical memory, where this scaling
|
| 2293 | * makes sense, it is disabled on such platforms.
|
| 2294 |
|
| 2295 | #if __BITS_PER_LONG > 32
|
| 2296 | #define ADAPT_SCALE_BASE (64ul << 30)
|
| 2297 | #define ADAPT_SCALE_SHIFT 2
|
| 2298 | #define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT)
|
| 2299 | #endif
|
| 2300 |
|
| 2301 | |
| 2302 | * allocate a large system hash table from bootmem
|
| 2303 | * - it is assumed that the hash table must contain an exact power-of-2
|
| 2304 | * quantity of entries
|
| 2305 | * - limit is the number of hash buckets, not the total allocation size
|
| 2306 |
|
| 2307 | void *__init alloc_large_system_hash(const char *tablename,
|
| 2308 | unsigned long bucketsize,
|
| 2309 | unsigned long numentries,
|
| 2310 | int scale,
|
| 2311 | int flags,
|
| 2312 | unsigned int *_hash_shift,
|
| 2313 | unsigned int *_hash_mask,
|
| 2314 | unsigned long low_limit,
|
| 2315 | unsigned long high_limit)
|
| 2316 | {
|
| 2317 | unsigned long long max = high_limit;
|
| 2318 | unsigned long log2qty, size;
|
| 2319 | void *table;
|
| 2320 | gfp_t gfp_flags;
|
| 2321 | bool virt;
|
| 2322 | bool huge;
|
| 2323 |
|
| 2324 |
|
| 2325 | if (!numentries) {
|
| 2326 |
|
| 2327 | numentries = nr_kernel_pages;
|
| 2328 |
|
| 2329 |
|
| 2330 | if (PAGE_SIZE < SZ_1M)
|
| 2331 | numentries = round_up(numentries, SZ_1M / PAGE_SIZE);
|
| 2332 |
|
| 2333 | #if __BITS_PER_LONG > 32
|
| 2334 | if (!high_limit) {
|
| 2335 | unsigned long adapt;
|
| 2336 |
|
| 2337 | for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries;
|
| 2338 | adapt <<= ADAPT_SCALE_SHIFT)
|
| 2339 | scale++;
|
| 2340 | }
|
| 2341 | #endif
|
| 2342 |
|
| 2343 |
|
| 2344 | if (scale > PAGE_SHIFT)
|
| 2345 | numentries >>= (scale - PAGE_SHIFT);
|
| 2346 | else
|
| 2347 | numentries <<= (PAGE_SHIFT - scale);
|
| 2348 |
|
| 2349 | if (unlikely((numentries * bucketsize) < PAGE_SIZE))
|
| 2350 | numentries = PAGE_SIZE / bucketsize;
|
| 2351 | }
|
| 2352 | numentries = roundup_pow_of_two(numentries);
|
| 2353 |
|
| 2354 |
|
| 2355 | if (max == 0) {
|
| 2356 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
|
| 2357 | max = div64_ul(max, bucketsize);
|
| 2358 | }
|
| 2359 | max = min(max, 0x80000000ULL);
|
| 2360 |
|
| 2361 | if (numentries < low_limit)
|
| 2362 | numentries = low_limit;
|
| 2363 | if (numentries > max)
|
| 2364 | numentries = max;
|
| 2365 |
|
| 2366 | log2qty = ilog2(numentries);
|
| 2367 |
|
| 2368 | gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC;
|
| 2369 | do {
|
| 2370 | virt = false;
|
| 2371 | size = bucketsize << log2qty;
|
| 2372 | if (flags & HASH_EARLY) {
|
| 2373 | if (flags & HASH_ZERO)
|
| 2374 | table = memblock_alloc(size, SMP_CACHE_BYTES);
|
| 2375 | else
|
| 2376 | table = memblock_alloc_raw(size,
|
| 2377 | SMP_CACHE_BYTES);
|
| 2378 | } else if (get_order(size) > MAX_PAGE_ORDER || hashdist) {
|
| 2379 | table = vmalloc_huge(size, gfp_flags);
|
| 2380 | virt = true;
|
| 2381 | if (table)
|
| 2382 | huge = is_vm_area_hugepages(table);
|
| 2383 | } else {
|
| 2384 | |
| 2385 | * If bucketsize is not a power-of-two, we may free
|
| 2386 | * some pages at the end of hash table which
|
| 2387 | * alloc_pages_exact() automatically does
|
| 2388 |
|
| 2389 | table = alloc_pages_exact(size, gfp_flags);
|
| 2390 | kmemleak_alloc(table, size, 1, gfp_flags);
|
| 2391 | }
|
| 2392 | } while (!table && size > PAGE_SIZE && --log2qty);
|
| 2393 |
|
| 2394 | if (!table)
|
| 2395 | panic("Failed to allocate %s hash table\n", tablename);
|
| 2396 |
|
| 2397 | pr_info("%s hash table entries: %ld (order: %d, %lu bytes, %s)\n",
|
| 2398 | tablename, 1UL << log2qty, get_order(size), size,
|
| 2399 | virt ? (huge ? "vmalloc hugepage" : "vmalloc") : "linear");
|
| 2400 |
|
| 2401 | if (_hash_shift)
|
| 2402 | *_hash_shift = log2qty;
|
| 2403 | if (_hash_mask)
|
| 2404 | *_hash_mask = (1 << log2qty) - 1;
|
| 2405 |
|
| 2406 | return table;
|
| 2407 | }
|
| 2408 |
|
| 2409 | void __init memblock_free_pages(unsigned long pfn, unsigned int order)
|
| 2410 | {
|
| 2411 | struct page *page = pfn_to_page(pfn);
|
| 2412 |
|
| 2413 | if (IS_ENABLED(CONFIG_DEFERRED_STRUCT_PAGE_INIT)) {
|
| 2414 | int nid = early_pfn_to_nid(pfn);
|
| 2415 |
|
| 2416 | if (!early_page_initialised(pfn, nid))
|
| 2417 | return;
|
| 2418 | }
|
| 2419 |
|
| 2420 | if (!kmsan_memblock_free_pages(page, order)) {
|
| 2421 |
|
| 2422 | return;
|
| 2423 | }
|
| 2424 |
|
| 2425 |
|
| 2426 | clear_page_tag_ref(page);
|
| 2427 | __free_pages_core(page, order, MEMINIT_EARLY);
|
| 2428 | }
|
| 2429 |
|
| 2430 | DEFINE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_ALLOC_DEFAULT_ON, init_on_alloc);
|
| 2431 | EXPORT_SYMBOL(init_on_alloc);
|
| 2432 |
|
| 2433 | DEFINE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_FREE_DEFAULT_ON, init_on_free);
|
| 2434 | EXPORT_SYMBOL(init_on_free);
|
| 2435 |
|
| 2436 | static bool _init_on_alloc_enabled_early __read_mostly
|
| 2437 | = IS_ENABLED(CONFIG_INIT_ON_ALLOC_DEFAULT_ON);
|
| 2438 | static int __init early_init_on_alloc(char *buf)
|
| 2439 | {
|
| 2440 |
|
| 2441 | return kstrtobool(buf, &_init_on_alloc_enabled_early);
|
| 2442 | }
|
| 2443 | early_param("init_on_alloc", early_init_on_alloc);
|
| 2444 |
|
| 2445 | static bool _init_on_free_enabled_early __read_mostly
|
| 2446 | = IS_ENABLED(CONFIG_INIT_ON_FREE_DEFAULT_ON);
|
| 2447 | static int __init early_init_on_free(char *buf)
|
| 2448 | {
|
| 2449 | return kstrtobool(buf, &_init_on_free_enabled_early);
|
| 2450 | }
|
| 2451 | early_param("init_on_free", early_init_on_free);
|
| 2452 |
|
| 2453 | DEFINE_STATIC_KEY_MAYBE(CONFIG_DEBUG_VM, check_pages_enabled);
|
| 2454 |
|
| 2455 | static bool check_pages_enabled_early __initdata;
|
| 2456 |
|
| 2457 | static int __init early_check_pages(char *buf)
|
| 2458 | {
|
| 2459 | return kstrtobool(buf, &check_pages_enabled_early);
|
| 2460 | }
|
| 2461 | early_param("check_pages", early_check_pages);
|
| 2462 |
|
| 2463 | |
| 2464 | * Enable static keys related to various memory debugging and hardening options.
|
| 2465 | * Some override others, and depend on early params that are evaluated in the
|
| 2466 | * order of appearance. So we need to first gather the full picture of what was
|
| 2467 | * enabled, and then make decisions.
|
| 2468 |
|
| 2469 | static void __init mem_debugging_and_hardening_init(void)
|
| 2470 | {
|
| 2471 | bool page_poisoning_requested = false;
|
| 2472 | bool want_check_pages = check_pages_enabled_early;
|
| 2473 |
|
| 2474 | #ifdef CONFIG_PAGE_POISONING
|
| 2475 | |
| 2476 | * Page poisoning is debug page alloc for some arches. If
|
| 2477 | * either of those options are enabled, enable poisoning.
|
| 2478 |
|
| 2479 | if (page_poisoning_enabled() ||
|
| 2480 | (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_DEBUG_PAGEALLOC) &&
|
| 2481 | debug_pagealloc_enabled())) {
|
| 2482 | static_branch_enable(&_page_poisoning_enabled);
|
| 2483 | page_poisoning_requested = true;
|
| 2484 | want_check_pages = true;
|
| 2485 | }
|
| 2486 | #endif
|
| 2487 |
|
| 2488 | if ((_init_on_alloc_enabled_early || _init_on_free_enabled_early) &&
|
| 2489 | page_poisoning_requested) {
|
| 2490 | pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, "
|
| 2491 | "will take precedence over init_on_alloc and init_on_free\n");
|
| 2492 | _init_on_alloc_enabled_early = false;
|
| 2493 | _init_on_free_enabled_early = false;
|
| 2494 | }
|
| 2495 |
|
| 2496 | if (_init_on_alloc_enabled_early) {
|
| 2497 | want_check_pages = true;
|
| 2498 | static_branch_enable(&init_on_alloc);
|
| 2499 | } else {
|
| 2500 | static_branch_disable(&init_on_alloc);
|
| 2501 | }
|
| 2502 |
|
| 2503 | if (_init_on_free_enabled_early) {
|
| 2504 | want_check_pages = true;
|
| 2505 | static_branch_enable(&init_on_free);
|
| 2506 | } else {
|
| 2507 | static_branch_disable(&init_on_free);
|
| 2508 | }
|
| 2509 |
|
| 2510 | if (IS_ENABLED(CONFIG_KMSAN) &&
|
| 2511 | (_init_on_alloc_enabled_early || _init_on_free_enabled_early))
|
| 2512 | pr_info("mem auto-init: please make sure init_on_alloc and init_on_free are disabled when running KMSAN\n");
|
| 2513 |
|
| 2514 | #ifdef CONFIG_DEBUG_PAGEALLOC
|
| 2515 | if (debug_pagealloc_enabled()) {
|
| 2516 | want_check_pages = true;
|
| 2517 | static_branch_enable(&_debug_pagealloc_enabled);
|
| 2518 |
|
| 2519 | if (debug_guardpage_minorder())
|
| 2520 | static_branch_enable(&_debug_guardpage_enabled);
|
| 2521 | }
|
| 2522 | #endif
|
| 2523 |
|
| 2524 | |
| 2525 | * Any page debugging or hardening option also enables sanity checking
|
| 2526 | * of struct pages being allocated or freed. With CONFIG_DEBUG_VM it's
|
| 2527 | * enabled already.
|
| 2528 |
|
| 2529 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && want_check_pages)
|
| 2530 | static_branch_enable(&check_pages_enabled);
|
| 2531 | }
|
| 2532 |
|
| 2533 |
|
| 2534 | static void __init report_meminit(void)
|
| 2535 | {
|
| 2536 | const char *stack;
|
| 2537 |
|
| 2538 | if (IS_ENABLED(CONFIG_INIT_STACK_ALL_PATTERN))
|
| 2539 | stack = "all(pattern)";
|
| 2540 | else if (IS_ENABLED(CONFIG_INIT_STACK_ALL_ZERO))
|
| 2541 | stack = "all(zero)";
|
| 2542 | else
|
| 2543 | stack = "off";
|
| 2544 |
|
| 2545 | pr_info("mem auto-init: stack:%s, heap alloc:%s, heap free:%s\n",
|
| 2546 | stack, str_on_off(want_init_on_alloc(GFP_KERNEL)),
|
| 2547 | str_on_off(want_init_on_free()));
|
| 2548 | if (want_init_on_free())
|
| 2549 | pr_info("mem auto-init: clearing system memory may take some time...\n");
|
| 2550 | }
|
| 2551 |
|
| 2552 | static void __init mem_init_print_info(void)
|
| 2553 | {
|
| 2554 | unsigned long physpages, codesize, datasize, rosize, bss_size;
|
| 2555 | unsigned long init_code_size, init_data_size;
|
| 2556 |
|
| 2557 | physpages = get_num_physpages();
|
| 2558 | codesize = _etext - _stext;
|
| 2559 | datasize = _edata - _sdata;
|
| 2560 | rosize = __end_rodata - __start_rodata;
|
| 2561 | bss_size = __bss_stop - __bss_start;
|
| 2562 | init_data_size = __init_end - __init_begin;
|
| 2563 | init_code_size = _einittext - _sinittext;
|
| 2564 |
|
| 2565 | |
| 2566 | * Detect special cases and adjust section sizes accordingly:
|
| 2567 | * 1) .init.* may be embedded into .data sections
|
| 2568 | * 2) .init.text.* may be out of [__init_begin, __init_end],
|
| 2569 | * please refer to arch/tile/kernel/vmlinux.lds.S.
|
| 2570 | * 3) .rodata.* may be embedded into .text or .data sections.
|
| 2571 |
|
| 2572 | #define adj_init_size(start, end, size, pos, adj) \
|
| 2573 | do { \
|
| 2574 | if (&start[0] <= &pos[0] && &pos[0] < &end[0] && size > adj) \
|
| 2575 | size -= adj; \
|
| 2576 | } while (0)
|
| 2577 |
|
| 2578 | adj_init_size(__init_begin, __init_end, init_data_size,
|
| 2579 | _sinittext, init_code_size);
|
| 2580 | adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
|
| 2581 | adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
|
| 2582 | adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
|
| 2583 | adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
|
| 2584 |
|
| 2585 | #undef adj_init_size
|
| 2586 |
|
| 2587 | pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
|
| 2588 | #ifdef CONFIG_HIGHMEM
|
| 2589 | ", %luK highmem"
|
| 2590 | #endif
|
| 2591 | ")\n",
|
| 2592 | K(nr_free_pages()), K(physpages),
|
| 2593 | codesize / SZ_1K, datasize / SZ_1K, rosize / SZ_1K,
|
| 2594 | (init_data_size + init_code_size) / SZ_1K, bss_size / SZ_1K,
|
| 2595 | K(physpages - totalram_pages() - totalcma_pages),
|
| 2596 | K(totalcma_pages)
|
| 2597 | #ifdef CONFIG_HIGHMEM
|
| 2598 | , K(totalhigh_pages())
|
| 2599 | #endif
|
| 2600 | );
|
| 2601 | }
|
| 2602 |
|
| 2603 | #ifndef __HAVE_COLOR_ZERO_PAGE
|
| 2604 | |
| 2605 | * architectures that __HAVE_COLOR_ZERO_PAGE must define this function
|
| 2606 |
|
| 2607 | void __init __weak arch_setup_zero_pages(void)
|
| 2608 | {
|
| 2609 | __zero_page = virt_to_page(empty_zero_page);
|
| 2610 | }
|
| 2611 | #endif
|
| 2612 |
|
| 2613 | static void __init init_zero_page_pfn(void)
|
| 2614 | {
|
| 2615 | arch_setup_zero_pages();
|
| 2616 | zero_page_pfn = page_to_pfn(ZERO_PAGE(0));
|
| 2617 | }
|
| 2618 |
|
| 2619 | void __init __weak arch_mm_preinit(void)
|
| 2620 | {
|
| 2621 | }
|
| 2622 |
|
| 2623 | void __init __weak mem_init(void)
|
| 2624 | {
|
| 2625 | }
|
| 2626 |
|
| 2627 | void __init mm_core_init_early(void)
|
| 2628 | {
|
| 2629 | kho_memory_init_early();
|
| 2630 |
|
| 2631 | free_area_init();
|
| 2632 |
|
| 2633 | hugetlb_cma_reserve();
|
| 2634 | hugetlb_bootmem_alloc();
|
| 2635 |
|
| 2636 | sparse_init();
|
| 2637 | memmap_init();
|
| 2638 | }
|
| 2639 |
|
| 2640 | |
| 2641 | * Set up kernel memory allocators
|
| 2642 |
|
| 2643 | void __init mm_core_init(void)
|
| 2644 | {
|
| 2645 | arch_mm_preinit();
|
| 2646 | init_zero_page_pfn();
|
| 2647 |
|
| 2648 |
|
| 2649 | BUILD_BUG_ON(MAX_ZONELISTS > 2);
|
| 2650 | build_all_zonelists(NULL);
|
| 2651 | page_alloc_init_cpuhp();
|
| 2652 | alloc_tag_sec_init();
|
| 2653 | |
| 2654 | * page_ext requires contiguous pages,
|
| 2655 | * bigger than MAX_PAGE_ORDER unless SPARSEMEM.
|
| 2656 |
|
| 2657 | page_ext_init_flatmem();
|
| 2658 | mem_debugging_and_hardening_init();
|
| 2659 | kfence_alloc_pool_and_metadata();
|
| 2660 | report_meminit();
|
| 2661 | kmsan_init_shadow();
|
| 2662 | stack_depot_early_init();
|
| 2663 |
|
| 2664 | |
| 2665 | * KHO memory setup must happen while memblock is still active, but
|
| 2666 | * as close as possible to buddy initialization
|
| 2667 |
|
| 2668 | kho_memory_init();
|
| 2669 |
|
| 2670 | memblock_free_all();
|
| 2671 | mem_init();
|
| 2672 | kmem_cache_init();
|
| 2673 | |
| 2674 | * page_owner must be initialized after buddy is ready, and also after
|
| 2675 | * slab is ready so that stack_depot_init() works properly
|
| 2676 |
|
| 2677 | page_ext_init_flatmem_late();
|
| 2678 | kmemleak_init();
|
| 2679 | ptlock_cache_init();
|
| 2680 | pgtable_cache_init();
|
| 2681 | debug_objects_mem_init();
|
| 2682 | vmalloc_init();
|
| 2683 |
|
| 2684 | if (!deferred_struct_pages)
|
| 2685 | page_ext_init();
|
| 2686 |
|
| 2687 | init_espfix_bsp();
|
| 2688 |
|
| 2689 | pti_init();
|
| 2690 | kmsan_init_runtime();
|
| 2691 | mm_cache_init();
|
| 2692 | execmem_init();
|
| 2693 | }
|