rjw | 1f88458 | 2022-01-06 17:20:42 +0800 | [diff] [blame^] | 1 | /* |
| 2 | * linux/arch/arm/mm/init.c |
| 3 | * |
| 4 | * Copyright (C) 1995-2005 Russell King |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or modify |
| 7 | * it under the terms of the GNU General Public License version 2 as |
| 8 | * published by the Free Software Foundation. |
| 9 | */ |
| 10 | #include <linux/kernel.h> |
| 11 | #include <linux/errno.h> |
| 12 | #include <linux/swap.h> |
| 13 | #include <linux/init.h> |
| 14 | #include <linux/bootmem.h> |
| 15 | #include <linux/mman.h> |
| 16 | #include <linux/sched/signal.h> |
| 17 | #include <linux/sched/task.h> |
| 18 | #include <linux/export.h> |
| 19 | #include <linux/nodemask.h> |
| 20 | #include <linux/initrd.h> |
| 21 | #include <linux/of_fdt.h> |
| 22 | #include <linux/highmem.h> |
| 23 | #include <linux/gfp.h> |
| 24 | #include <linux/memblock.h> |
| 25 | #include <linux/dma-contiguous.h> |
| 26 | #include <linux/sizes.h> |
| 27 | #include <linux/stop_machine.h> |
| 28 | |
| 29 | #include <asm/cp15.h> |
| 30 | #include <asm/mach-types.h> |
| 31 | #include <asm/memblock.h> |
| 32 | #include <asm/memory.h> |
| 33 | #include <asm/prom.h> |
| 34 | #include <asm/sections.h> |
| 35 | #include <asm/setup.h> |
| 36 | #include <asm/system_info.h> |
| 37 | #include <asm/tlb.h> |
| 38 | #include <asm/fixmap.h> |
| 39 | |
| 40 | #include <asm/mach/arch.h> |
| 41 | #include <asm/mach/map.h> |
| 42 | |
| 43 | #include "mm.h" |
| 44 | |
| 45 | #ifdef CONFIG_CPU_CP15_MMU |
| 46 | unsigned long __init __clear_cr(unsigned long mask) |
| 47 | { |
| 48 | cr_alignment = cr_alignment & ~mask; |
| 49 | return cr_alignment; |
| 50 | } |
| 51 | #endif |
| 52 | |
| 53 | static phys_addr_t phys_initrd_start __initdata = 0; |
| 54 | static unsigned long phys_initrd_size __initdata = 0; |
| 55 | |
| 56 | static int __init early_initrd(char *p) |
| 57 | { |
| 58 | phys_addr_t start; |
| 59 | unsigned long size; |
| 60 | char *endp; |
| 61 | |
| 62 | start = memparse(p, &endp); |
| 63 | if (*endp == ',') { |
| 64 | size = memparse(endp + 1, NULL); |
| 65 | |
| 66 | phys_initrd_start = start; |
| 67 | phys_initrd_size = size; |
| 68 | } |
| 69 | return 0; |
| 70 | } |
| 71 | early_param("initrd", early_initrd); |
| 72 | |
| 73 | static int __init parse_tag_initrd(const struct tag *tag) |
| 74 | { |
| 75 | pr_warn("ATAG_INITRD is deprecated; " |
| 76 | "please update your bootloader.\n"); |
| 77 | phys_initrd_start = __virt_to_phys(tag->u.initrd.start); |
| 78 | phys_initrd_size = tag->u.initrd.size; |
| 79 | return 0; |
| 80 | } |
| 81 | |
| 82 | __tagtable(ATAG_INITRD, parse_tag_initrd); |
| 83 | |
| 84 | static int __init parse_tag_initrd2(const struct tag *tag) |
| 85 | { |
| 86 | phys_initrd_start = tag->u.initrd.start; |
| 87 | phys_initrd_size = tag->u.initrd.size; |
| 88 | return 0; |
| 89 | } |
| 90 | |
| 91 | __tagtable(ATAG_INITRD2, parse_tag_initrd2); |
| 92 | |
| 93 | static void __init find_limits(unsigned long *min, unsigned long *max_low, |
| 94 | unsigned long *max_high) |
| 95 | { |
| 96 | *max_low = PFN_DOWN(memblock_get_current_limit()); |
| 97 | *min = PFN_UP(memblock_start_of_DRAM()); |
| 98 | *max_high = PFN_DOWN(memblock_end_of_DRAM()); |
| 99 | } |
| 100 | |
| 101 | #ifdef CONFIG_ZONE_DMA |
| 102 | |
| 103 | phys_addr_t arm_dma_zone_size __read_mostly; |
| 104 | EXPORT_SYMBOL(arm_dma_zone_size); |
| 105 | |
| 106 | /* |
| 107 | * The DMA mask corresponding to the maximum bus address allocatable |
| 108 | * using GFP_DMA. The default here places no restriction on DMA |
| 109 | * allocations. This must be the smallest DMA mask in the system, |
| 110 | * so a successful GFP_DMA allocation will always satisfy this. |
| 111 | */ |
| 112 | phys_addr_t arm_dma_limit; |
| 113 | unsigned long arm_dma_pfn_limit; |
| 114 | |
| 115 | static void __init arm_adjust_dma_zone(unsigned long *size, unsigned long *hole, |
| 116 | unsigned long dma_size) |
| 117 | { |
| 118 | if (size[0] <= dma_size) |
| 119 | return; |
| 120 | |
| 121 | size[ZONE_NORMAL] = size[0] - dma_size; |
| 122 | size[ZONE_DMA] = dma_size; |
| 123 | hole[ZONE_NORMAL] = hole[0]; |
| 124 | hole[ZONE_DMA] = 0; |
| 125 | } |
| 126 | #endif |
| 127 | |
| 128 | void __init setup_dma_zone(const struct machine_desc *mdesc) |
| 129 | { |
| 130 | #ifdef CONFIG_ZONE_DMA |
| 131 | if (mdesc->dma_zone_size) { |
| 132 | arm_dma_zone_size = mdesc->dma_zone_size; |
| 133 | arm_dma_limit = PHYS_OFFSET + arm_dma_zone_size - 1; |
| 134 | } else |
| 135 | arm_dma_limit = 0xffffffff; |
| 136 | arm_dma_pfn_limit = arm_dma_limit >> PAGE_SHIFT; |
| 137 | #endif |
| 138 | } |
| 139 | |
| 140 | static void __init zone_sizes_init(unsigned long min, unsigned long max_low, |
| 141 | unsigned long max_high) |
| 142 | { |
| 143 | unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES]; |
| 144 | struct memblock_region *reg; |
| 145 | |
| 146 | /* |
| 147 | * initialise the zones. |
| 148 | */ |
| 149 | memset(zone_size, 0, sizeof(zone_size)); |
| 150 | |
| 151 | /* |
| 152 | * The memory size has already been determined. If we need |
| 153 | * to do anything fancy with the allocation of this memory |
| 154 | * to the zones, now is the time to do it. |
| 155 | */ |
| 156 | zone_size[0] = max_low - min; |
| 157 | #ifdef CONFIG_HIGHMEM |
| 158 | zone_size[ZONE_HIGHMEM] = max_high - max_low; |
| 159 | #endif |
| 160 | |
| 161 | /* |
| 162 | * Calculate the size of the holes. |
| 163 | * holes = node_size - sum(bank_sizes) |
| 164 | */ |
| 165 | memcpy(zhole_size, zone_size, sizeof(zhole_size)); |
| 166 | for_each_memblock(memory, reg) { |
| 167 | unsigned long start = memblock_region_memory_base_pfn(reg); |
| 168 | unsigned long end = memblock_region_memory_end_pfn(reg); |
| 169 | |
| 170 | if (start < max_low) { |
| 171 | unsigned long low_end = min(end, max_low); |
| 172 | zhole_size[0] -= low_end - start; |
| 173 | } |
| 174 | #ifdef CONFIG_HIGHMEM |
| 175 | if (end > max_low) { |
| 176 | unsigned long high_start = max(start, max_low); |
| 177 | zhole_size[ZONE_HIGHMEM] -= end - high_start; |
| 178 | } |
| 179 | #endif |
| 180 | } |
| 181 | |
| 182 | #ifdef CONFIG_ZONE_DMA |
| 183 | /* |
| 184 | * Adjust the sizes according to any special requirements for |
| 185 | * this machine type. |
| 186 | */ |
| 187 | if (arm_dma_zone_size) |
| 188 | arm_adjust_dma_zone(zone_size, zhole_size, |
| 189 | arm_dma_zone_size >> PAGE_SHIFT); |
| 190 | #endif |
| 191 | |
| 192 | free_area_init_node(0, zone_size, min, zhole_size); |
| 193 | } |
| 194 | |
| 195 | #ifdef CONFIG_HAVE_ARCH_PFN_VALID |
| 196 | int pfn_valid(unsigned long pfn) |
| 197 | { |
| 198 | phys_addr_t addr = __pfn_to_phys(pfn); |
| 199 | |
| 200 | if (__phys_to_pfn(addr) != pfn) |
| 201 | return 0; |
| 202 | |
| 203 | return memblock_is_map_memory(__pfn_to_phys(pfn)); |
| 204 | } |
| 205 | EXPORT_SYMBOL(pfn_valid); |
| 206 | #endif |
| 207 | |
| 208 | #ifndef CONFIG_SPARSEMEM |
| 209 | static void __init arm_memory_present(void) |
| 210 | { |
| 211 | } |
| 212 | #else |
| 213 | static void __init arm_memory_present(void) |
| 214 | { |
| 215 | struct memblock_region *reg; |
| 216 | |
| 217 | for_each_memblock(memory, reg) |
| 218 | memory_present(0, memblock_region_memory_base_pfn(reg), |
| 219 | memblock_region_memory_end_pfn(reg)); |
| 220 | } |
| 221 | #endif |
| 222 | |
| 223 | static bool arm_memblock_steal_permitted = true; |
| 224 | |
| 225 | phys_addr_t __init arm_memblock_steal(phys_addr_t size, phys_addr_t align) |
| 226 | { |
| 227 | phys_addr_t phys; |
| 228 | |
| 229 | BUG_ON(!arm_memblock_steal_permitted); |
| 230 | |
| 231 | phys = memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ANYWHERE); |
| 232 | memblock_free(phys, size); |
| 233 | memblock_remove(phys, size); |
| 234 | |
| 235 | return phys; |
| 236 | } |
| 237 | |
| 238 | static void __init arm_initrd_init(void) |
| 239 | { |
| 240 | #ifdef CONFIG_BLK_DEV_INITRD |
| 241 | phys_addr_t start; |
| 242 | unsigned long size; |
| 243 | |
| 244 | /* FDT scan will populate initrd_start */ |
| 245 | if (initrd_start && !phys_initrd_size) { |
| 246 | phys_initrd_start = __virt_to_phys(initrd_start); |
| 247 | phys_initrd_size = initrd_end - initrd_start; |
| 248 | } |
| 249 | |
| 250 | initrd_start = initrd_end = 0; |
| 251 | |
| 252 | if (!phys_initrd_size) |
| 253 | return; |
| 254 | |
| 255 | /* |
| 256 | * Round the memory region to page boundaries as per free_initrd_mem() |
| 257 | * This allows us to detect whether the pages overlapping the initrd |
| 258 | * are in use, but more importantly, reserves the entire set of pages |
| 259 | * as we don't want these pages allocated for other purposes. |
| 260 | */ |
| 261 | start = round_down(phys_initrd_start, PAGE_SIZE); |
| 262 | size = phys_initrd_size + (phys_initrd_start - start); |
| 263 | size = round_up(size, PAGE_SIZE); |
| 264 | |
| 265 | if (!memblock_is_region_memory(start, size)) { |
| 266 | pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region - disabling initrd\n", |
| 267 | (u64)start, size); |
| 268 | return; |
| 269 | } |
| 270 | |
| 271 | if (memblock_is_region_reserved(start, size)) { |
| 272 | pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region - disabling initrd\n", |
| 273 | (u64)start, size); |
| 274 | return; |
| 275 | } |
| 276 | |
| 277 | memblock_reserve(start, size); |
| 278 | |
| 279 | /* Now convert initrd to virtual addresses */ |
| 280 | initrd_start = __phys_to_virt(phys_initrd_start); |
| 281 | initrd_end = initrd_start + phys_initrd_size; |
| 282 | #endif |
| 283 | } |
| 284 | |
| 285 | void __init arm_memblock_init(const struct machine_desc *mdesc) |
| 286 | { |
| 287 | /* Register the kernel text, kernel data and initrd with memblock. */ |
| 288 | memblock_reserve(__pa(KERNEL_START), KERNEL_END - KERNEL_START); |
| 289 | |
| 290 | arm_initrd_init(); |
| 291 | |
| 292 | arm_mm_memblock_reserve(); |
| 293 | |
| 294 | /* reserve any platform specific memblock areas */ |
| 295 | if (mdesc->reserve) |
| 296 | mdesc->reserve(); |
| 297 | |
| 298 | early_init_fdt_reserve_self(); |
| 299 | early_init_fdt_scan_reserved_mem(); |
| 300 | |
| 301 | /* reserve memory for DMA contiguous allocations */ |
| 302 | dma_contiguous_reserve(arm_dma_limit); |
| 303 | |
| 304 | arm_memblock_steal_permitted = false; |
| 305 | memblock_dump_all(); |
| 306 | } |
| 307 | |
| 308 | void __init bootmem_init(void) |
| 309 | { |
| 310 | unsigned long min, max_low, max_high; |
| 311 | |
| 312 | memblock_allow_resize(); |
| 313 | max_low = max_high = 0; |
| 314 | |
| 315 | find_limits(&min, &max_low, &max_high); |
| 316 | |
| 317 | early_memtest((phys_addr_t)min << PAGE_SHIFT, |
| 318 | (phys_addr_t)max_low << PAGE_SHIFT); |
| 319 | |
| 320 | /* |
| 321 | * Sparsemem tries to allocate bootmem in memory_present(), |
| 322 | * so must be done after the fixed reservations |
| 323 | */ |
| 324 | arm_memory_present(); |
| 325 | |
| 326 | /* |
| 327 | * sparse_init() needs the bootmem allocator up and running. |
| 328 | */ |
| 329 | sparse_init(); |
| 330 | |
| 331 | /* |
| 332 | * Now free the memory - free_area_init_node needs |
| 333 | * the sparse mem_map arrays initialized by sparse_init() |
| 334 | * for memmap_init_zone(), otherwise all PFNs are invalid. |
| 335 | */ |
| 336 | zone_sizes_init(min, max_low, max_high); |
| 337 | |
| 338 | /* |
| 339 | * This doesn't seem to be used by the Linux memory manager any |
| 340 | * more, but is used by ll_rw_block. If we can get rid of it, we |
| 341 | * also get rid of some of the stuff above as well. |
| 342 | */ |
| 343 | min_low_pfn = min; |
| 344 | max_low_pfn = max_low; |
| 345 | max_pfn = max_high; |
| 346 | } |
| 347 | |
| 348 | /* |
| 349 | * Poison init memory with an undefined instruction (ARM) or a branch to an |
| 350 | * undefined instruction (Thumb). |
| 351 | */ |
| 352 | static inline void poison_init_mem(void *s, size_t count) |
| 353 | { |
| 354 | u32 *p = (u32 *)s; |
| 355 | for (; count != 0; count -= 4) |
| 356 | *p++ = 0xe7fddef0; |
| 357 | } |
| 358 | |
| 359 | static inline void __init |
| 360 | free_memmap(unsigned long start_pfn, unsigned long end_pfn) |
| 361 | { |
| 362 | struct page *start_pg, *end_pg; |
| 363 | phys_addr_t pg, pgend; |
| 364 | |
| 365 | /* |
| 366 | * Convert start_pfn/end_pfn to a struct page pointer. |
| 367 | */ |
| 368 | start_pg = pfn_to_page(start_pfn - 1) + 1; |
| 369 | end_pg = pfn_to_page(end_pfn - 1) + 1; |
| 370 | |
| 371 | /* |
| 372 | * Convert to physical addresses, and |
| 373 | * round start upwards and end downwards. |
| 374 | */ |
| 375 | pg = PAGE_ALIGN(__pa(start_pg)); |
| 376 | pgend = __pa(end_pg) & PAGE_MASK; |
| 377 | |
| 378 | /* |
| 379 | * If there are free pages between these, |
| 380 | * free the section of the memmap array. |
| 381 | */ |
| 382 | if (pg < pgend) |
| 383 | memblock_free_early(pg, pgend - pg); |
| 384 | } |
| 385 | |
| 386 | /* |
| 387 | * The mem_map array can get very big. Free the unused area of the memory map. |
| 388 | */ |
| 389 | static void __init free_unused_memmap(void) |
| 390 | { |
| 391 | unsigned long start, prev_end = 0; |
| 392 | struct memblock_region *reg; |
| 393 | |
| 394 | /* |
| 395 | * This relies on each bank being in address order. |
| 396 | * The banks are sorted previously in bootmem_init(). |
| 397 | */ |
| 398 | for_each_memblock(memory, reg) { |
| 399 | start = memblock_region_memory_base_pfn(reg); |
| 400 | |
| 401 | #ifdef CONFIG_SPARSEMEM |
| 402 | /* |
| 403 | * Take care not to free memmap entries that don't exist |
| 404 | * due to SPARSEMEM sections which aren't present. |
| 405 | */ |
| 406 | start = min(start, |
| 407 | ALIGN(prev_end, PAGES_PER_SECTION)); |
| 408 | #else |
| 409 | /* |
| 410 | * Align down here since the VM subsystem insists that the |
| 411 | * memmap entries are valid from the bank start aligned to |
| 412 | * MAX_ORDER_NR_PAGES. |
| 413 | */ |
| 414 | start = round_down(start, MAX_ORDER_NR_PAGES); |
| 415 | #endif |
| 416 | /* |
| 417 | * If we had a previous bank, and there is a space |
| 418 | * between the current bank and the previous, free it. |
| 419 | */ |
| 420 | if (prev_end && prev_end < start) |
| 421 | free_memmap(prev_end, start); |
| 422 | |
| 423 | /* |
| 424 | * Align up here since the VM subsystem insists that the |
| 425 | * memmap entries are valid from the bank end aligned to |
| 426 | * MAX_ORDER_NR_PAGES. |
| 427 | */ |
| 428 | prev_end = ALIGN(memblock_region_memory_end_pfn(reg), |
| 429 | MAX_ORDER_NR_PAGES); |
| 430 | } |
| 431 | |
| 432 | #ifdef CONFIG_SPARSEMEM |
| 433 | if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) |
| 434 | free_memmap(prev_end, |
| 435 | ALIGN(prev_end, PAGES_PER_SECTION)); |
| 436 | #endif |
| 437 | } |
| 438 | |
| 439 | #ifdef CONFIG_HIGHMEM |
| 440 | static inline void free_area_high(unsigned long pfn, unsigned long end) |
| 441 | { |
| 442 | for (; pfn < end; pfn++) |
| 443 | free_highmem_page(pfn_to_page(pfn)); |
| 444 | } |
| 445 | #endif |
| 446 | |
| 447 | static void __init free_highpages(void) |
| 448 | { |
| 449 | #ifdef CONFIG_HIGHMEM |
| 450 | unsigned long max_low = max_low_pfn; |
| 451 | struct memblock_region *mem, *res; |
| 452 | |
| 453 | /* set highmem page free */ |
| 454 | for_each_memblock(memory, mem) { |
| 455 | unsigned long start = memblock_region_memory_base_pfn(mem); |
| 456 | unsigned long end = memblock_region_memory_end_pfn(mem); |
| 457 | |
| 458 | /* Ignore complete lowmem entries */ |
| 459 | if (end <= max_low) |
| 460 | continue; |
| 461 | |
| 462 | if (memblock_is_nomap(mem)) |
| 463 | continue; |
| 464 | |
| 465 | /* Truncate partial highmem entries */ |
| 466 | if (start < max_low) |
| 467 | start = max_low; |
| 468 | |
| 469 | /* Find and exclude any reserved regions */ |
| 470 | for_each_memblock(reserved, res) { |
| 471 | unsigned long res_start, res_end; |
| 472 | |
| 473 | res_start = memblock_region_reserved_base_pfn(res); |
| 474 | res_end = memblock_region_reserved_end_pfn(res); |
| 475 | |
| 476 | if (res_end < start) |
| 477 | continue; |
| 478 | if (res_start < start) |
| 479 | res_start = start; |
| 480 | if (res_start > end) |
| 481 | res_start = end; |
| 482 | if (res_end > end) |
| 483 | res_end = end; |
| 484 | if (res_start != start) |
| 485 | free_area_high(start, res_start); |
| 486 | start = res_end; |
| 487 | if (start == end) |
| 488 | break; |
| 489 | } |
| 490 | |
| 491 | /* And now free anything which remains */ |
| 492 | if (start < end) |
| 493 | free_area_high(start, end); |
| 494 | } |
| 495 | #endif |
| 496 | } |
| 497 | |
| 498 | /* |
| 499 | * mem_init() marks the free areas in the mem_map and tells us how much |
| 500 | * memory is free. This is done after various parts of the system have |
| 501 | * claimed their memory after the kernel image. |
| 502 | */ |
| 503 | void __init mem_init(void) |
| 504 | { |
| 505 | #ifdef CONFIG_HAVE_TCM |
| 506 | /* These pointers are filled in on TCM detection */ |
| 507 | extern u32 dtcm_end; |
| 508 | extern u32 itcm_end; |
| 509 | #endif |
| 510 | |
| 511 | set_max_mapnr(pfn_to_page(max_pfn) - mem_map); |
| 512 | |
| 513 | /* this will put all unused low memory onto the freelists */ |
| 514 | free_unused_memmap(); |
| 515 | free_all_bootmem(); |
| 516 | |
| 517 | #ifdef CONFIG_SA1111 |
| 518 | /* now that our DMA memory is actually so designated, we can free it */ |
| 519 | free_reserved_area(__va(PHYS_OFFSET), swapper_pg_dir, -1, NULL); |
| 520 | #endif |
| 521 | |
| 522 | free_highpages(); |
| 523 | |
| 524 | mem_init_print_info(NULL); |
| 525 | |
| 526 | #define MLK(b, t) b, t, ((t) - (b)) >> 10 |
| 527 | #define MLM(b, t) b, t, ((t) - (b)) >> 20 |
| 528 | #define MLK_ROUNDUP(b, t) b, t, DIV_ROUND_UP(((t) - (b)), SZ_1K) |
| 529 | |
| 530 | pr_notice("Virtual kernel memory layout:\n" |
| 531 | " vector : 0x%08lx - 0x%08lx (%4ld kB)\n" |
| 532 | #ifdef CONFIG_HAVE_TCM |
| 533 | " DTCM : 0x%08lx - 0x%08lx (%4ld kB)\n" |
| 534 | " ITCM : 0x%08lx - 0x%08lx (%4ld kB)\n" |
| 535 | #endif |
| 536 | " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n" |
| 537 | " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n" |
| 538 | " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n" |
| 539 | #ifdef CONFIG_HIGHMEM |
| 540 | " pkmap : 0x%08lx - 0x%08lx (%4ld MB)\n" |
| 541 | #endif |
| 542 | #ifdef CONFIG_MODULES |
| 543 | " modules : 0x%08lx - 0x%08lx (%4ld MB)\n" |
| 544 | #endif |
| 545 | " .text : 0x%p" " - 0x%p" " (%4td kB)\n" |
| 546 | " .init : 0x%p" " - 0x%p" " (%4td kB)\n" |
| 547 | " .data : 0x%p" " - 0x%p" " (%4td kB)\n" |
| 548 | " .bss : 0x%p" " - 0x%p" " (%4td kB)\n", |
| 549 | |
| 550 | MLK(VECTORS_BASE, VECTORS_BASE + PAGE_SIZE), |
| 551 | #ifdef CONFIG_HAVE_TCM |
| 552 | MLK(DTCM_OFFSET, (unsigned long) dtcm_end), |
| 553 | MLK(ITCM_OFFSET, (unsigned long) itcm_end), |
| 554 | #endif |
| 555 | MLK(FIXADDR_START, FIXADDR_END), |
| 556 | MLM(VMALLOC_START, VMALLOC_END), |
| 557 | MLM(PAGE_OFFSET, (unsigned long)high_memory), |
| 558 | #ifdef CONFIG_HIGHMEM |
| 559 | MLM(PKMAP_BASE, (PKMAP_BASE) + (LAST_PKMAP) * |
| 560 | (PAGE_SIZE)), |
| 561 | #endif |
| 562 | #ifdef CONFIG_MODULES |
| 563 | MLM(MODULES_VADDR, MODULES_END), |
| 564 | #endif |
| 565 | |
| 566 | MLK_ROUNDUP(_text, _etext), |
| 567 | MLK_ROUNDUP(__init_begin, __init_end), |
| 568 | MLK_ROUNDUP(_sdata, _edata), |
| 569 | MLK_ROUNDUP(__bss_start, __bss_stop)); |
| 570 | |
| 571 | #undef MLK |
| 572 | #undef MLM |
| 573 | #undef MLK_ROUNDUP |
| 574 | |
| 575 | /* |
| 576 | * Check boundaries twice: Some fundamental inconsistencies can |
| 577 | * be detected at build time already. |
| 578 | */ |
| 579 | #ifdef CONFIG_MMU |
| 580 | BUILD_BUG_ON(TASK_SIZE > MODULES_VADDR); |
| 581 | BUG_ON(TASK_SIZE > MODULES_VADDR); |
| 582 | #endif |
| 583 | |
| 584 | #ifdef CONFIG_HIGHMEM |
| 585 | BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET); |
| 586 | BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET); |
| 587 | #endif |
| 588 | |
| 589 | if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) { |
| 590 | extern int sysctl_overcommit_memory; |
| 591 | /* |
| 592 | * On a machine this small we won't get |
| 593 | * anywhere without overcommit, so turn |
| 594 | * it on by default. |
| 595 | */ |
| 596 | sysctl_overcommit_memory = OVERCOMMIT_ALWAYS; |
| 597 | } |
| 598 | } |
| 599 | |
| 600 | #ifdef CONFIG_STRICT_KERNEL_RWX |
| 601 | struct section_perm { |
| 602 | const char *name; |
| 603 | unsigned long start; |
| 604 | unsigned long end; |
| 605 | pmdval_t mask; |
| 606 | pmdval_t prot; |
| 607 | pmdval_t clear; |
| 608 | }; |
| 609 | |
| 610 | /* First section-aligned location at or after __start_rodata. */ |
| 611 | extern char __start_rodata_section_aligned[]; |
| 612 | |
| 613 | static struct section_perm nx_perms[] = { |
| 614 | /* Make pages tables, etc before _stext RW (set NX). */ |
| 615 | { |
| 616 | .name = "pre-text NX", |
| 617 | .start = PAGE_OFFSET, |
| 618 | .end = (unsigned long)_stext, |
| 619 | .mask = ~PMD_SECT_XN, |
| 620 | .prot = PMD_SECT_XN, |
| 621 | }, |
| 622 | /* Make init RW (set NX). */ |
| 623 | { |
| 624 | .name = "init NX", |
| 625 | .start = (unsigned long)__init_begin, |
| 626 | .end = (unsigned long)_sdata, |
| 627 | .mask = ~PMD_SECT_XN, |
| 628 | .prot = PMD_SECT_XN, |
| 629 | }, |
| 630 | /* Make rodata NX (set RO in ro_perms below). */ |
| 631 | { |
| 632 | .name = "rodata NX", |
| 633 | .start = (unsigned long)__start_rodata_section_aligned, |
| 634 | .end = (unsigned long)__init_begin, |
| 635 | .mask = ~PMD_SECT_XN, |
| 636 | .prot = PMD_SECT_XN, |
| 637 | }, |
| 638 | }; |
| 639 | |
| 640 | static struct section_perm ro_perms[] = { |
| 641 | /* Make kernel code and rodata RX (set RO). */ |
| 642 | { |
| 643 | .name = "text/rodata RO", |
| 644 | .start = (unsigned long)_stext, |
| 645 | .end = (unsigned long)__init_begin, |
| 646 | #ifdef CONFIG_ARM_LPAE |
| 647 | .mask = ~(L_PMD_SECT_RDONLY | PMD_SECT_AP2), |
| 648 | .prot = L_PMD_SECT_RDONLY | PMD_SECT_AP2, |
| 649 | #else |
| 650 | .mask = ~(PMD_SECT_APX | PMD_SECT_AP_WRITE), |
| 651 | .prot = PMD_SECT_APX | PMD_SECT_AP_WRITE, |
| 652 | .clear = PMD_SECT_AP_WRITE, |
| 653 | #endif |
| 654 | }, |
| 655 | }; |
| 656 | |
| 657 | /* |
| 658 | * Updates section permissions only for the current mm (sections are |
| 659 | * copied into each mm). During startup, this is the init_mm. Is only |
| 660 | * safe to be called with preemption disabled, as under stop_machine(). |
| 661 | */ |
| 662 | static inline void section_update(unsigned long addr, pmdval_t mask, |
| 663 | pmdval_t prot, struct mm_struct *mm) |
| 664 | { |
| 665 | pmd_t *pmd; |
| 666 | |
| 667 | pmd = pmd_offset(pud_offset(pgd_offset(mm, addr), addr), addr); |
| 668 | |
| 669 | #ifdef CONFIG_ARM_LPAE |
| 670 | pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot); |
| 671 | #else |
| 672 | if (addr & SECTION_SIZE) |
| 673 | pmd[1] = __pmd((pmd_val(pmd[1]) & mask) | prot); |
| 674 | else |
| 675 | pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot); |
| 676 | #endif |
| 677 | flush_pmd_entry(pmd); |
| 678 | local_flush_tlb_kernel_range(addr, addr + SECTION_SIZE); |
| 679 | } |
| 680 | |
| 681 | /* Make sure extended page tables are in use. */ |
| 682 | static inline bool arch_has_strict_perms(void) |
| 683 | { |
| 684 | if (cpu_architecture() < CPU_ARCH_ARMv6) |
| 685 | return false; |
| 686 | |
| 687 | return !!(get_cr() & CR_XP); |
| 688 | } |
| 689 | |
| 690 | void set_section_perms(struct section_perm *perms, int n, bool set, |
| 691 | struct mm_struct *mm) |
| 692 | { |
| 693 | size_t i; |
| 694 | unsigned long addr; |
| 695 | |
| 696 | if (!arch_has_strict_perms()) |
| 697 | return; |
| 698 | |
| 699 | for (i = 0; i < n; i++) { |
| 700 | if (!IS_ALIGNED(perms[i].start, SECTION_SIZE) || |
| 701 | !IS_ALIGNED(perms[i].end, SECTION_SIZE)) { |
| 702 | pr_err("BUG: %s section %lx-%lx not aligned to %lx\n", |
| 703 | perms[i].name, perms[i].start, perms[i].end, |
| 704 | SECTION_SIZE); |
| 705 | continue; |
| 706 | } |
| 707 | |
| 708 | for (addr = perms[i].start; |
| 709 | addr < perms[i].end; |
| 710 | addr += SECTION_SIZE) |
| 711 | section_update(addr, perms[i].mask, |
| 712 | set ? perms[i].prot : perms[i].clear, mm); |
| 713 | } |
| 714 | |
| 715 | } |
| 716 | |
| 717 | /** |
| 718 | * update_sections_early intended to be called only through stop_machine |
| 719 | * framework and executed by only one CPU while all other CPUs will spin and |
| 720 | * wait, so no locking is required in this function. |
| 721 | */ |
| 722 | static void update_sections_early(struct section_perm perms[], int n) |
| 723 | { |
| 724 | struct task_struct *t, *s; |
| 725 | |
| 726 | for_each_process(t) { |
| 727 | if (t->flags & PF_KTHREAD) |
| 728 | continue; |
| 729 | for_each_thread(t, s) |
| 730 | if (s->mm) |
| 731 | set_section_perms(perms, n, true, s->mm); |
| 732 | } |
| 733 | set_section_perms(perms, n, true, current->active_mm); |
| 734 | set_section_perms(perms, n, true, &init_mm); |
| 735 | } |
| 736 | |
| 737 | static int __fix_kernmem_perms(void *unused) |
| 738 | { |
| 739 | update_sections_early(nx_perms, ARRAY_SIZE(nx_perms)); |
| 740 | return 0; |
| 741 | } |
| 742 | |
| 743 | static void fix_kernmem_perms(void) |
| 744 | { |
| 745 | stop_machine(__fix_kernmem_perms, NULL, NULL); |
| 746 | } |
| 747 | |
| 748 | static int __mark_rodata_ro(void *unused) |
| 749 | { |
| 750 | update_sections_early(ro_perms, ARRAY_SIZE(ro_perms)); |
| 751 | return 0; |
| 752 | } |
| 753 | |
| 754 | static int kernel_set_to_readonly __read_mostly; |
| 755 | |
| 756 | void mark_rodata_ro(void) |
| 757 | { |
| 758 | kernel_set_to_readonly = 1; |
| 759 | stop_machine(__mark_rodata_ro, NULL, NULL); |
| 760 | } |
| 761 | |
| 762 | void set_kernel_text_rw(void) |
| 763 | { |
| 764 | if (!kernel_set_to_readonly) |
| 765 | return; |
| 766 | |
| 767 | set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), false, |
| 768 | current->active_mm); |
| 769 | } |
| 770 | |
| 771 | void set_kernel_text_ro(void) |
| 772 | { |
| 773 | if (!kernel_set_to_readonly) |
| 774 | return; |
| 775 | |
| 776 | set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), true, |
| 777 | current->active_mm); |
| 778 | } |
| 779 | |
| 780 | #else |
| 781 | static inline void fix_kernmem_perms(void) { } |
| 782 | #endif /* CONFIG_STRICT_KERNEL_RWX */ |
| 783 | |
| 784 | void free_tcmmem(void) |
| 785 | { |
| 786 | #ifdef CONFIG_HAVE_TCM |
| 787 | extern char __tcm_start, __tcm_end; |
| 788 | |
| 789 | poison_init_mem(&__tcm_start, &__tcm_end - &__tcm_start); |
| 790 | free_reserved_area(&__tcm_start, &__tcm_end, -1, "TCM link"); |
| 791 | #endif |
| 792 | } |
| 793 | |
| 794 | void free_initmem(void) |
| 795 | { |
| 796 | fix_kernmem_perms(); |
| 797 | free_tcmmem(); |
| 798 | |
| 799 | poison_init_mem(__init_begin, __init_end - __init_begin); |
| 800 | if (!machine_is_integrator() && !machine_is_cintegrator()) |
| 801 | free_initmem_default(-1); |
| 802 | } |
| 803 | |
| 804 | #ifdef CONFIG_BLK_DEV_INITRD |
| 805 | |
| 806 | static int keep_initrd; |
| 807 | |
| 808 | void free_initrd_mem(unsigned long start, unsigned long end) |
| 809 | { |
| 810 | if (!keep_initrd) { |
| 811 | if (start == initrd_start) |
| 812 | start = round_down(start, PAGE_SIZE); |
| 813 | if (end == initrd_end) |
| 814 | end = round_up(end, PAGE_SIZE); |
| 815 | |
| 816 | poison_init_mem((void *)start, PAGE_ALIGN(end) - start); |
| 817 | free_reserved_area((void *)start, (void *)end, -1, "initrd"); |
| 818 | } |
| 819 | } |
| 820 | |
| 821 | static int __init keepinitrd_setup(char *__unused) |
| 822 | { |
| 823 | keep_initrd = 1; |
| 824 | return 1; |
| 825 | } |
| 826 | |
| 827 | __setup("keepinitrd", keepinitrd_setup); |
| 828 | #endif |