arch/arm/mm/init.c

   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/export.h>
  17 #include <linux/nodemask.h>
  18 #include <linux/initrd.h>
  19 #include <linux/of_fdt.h>
  20 #include <linux/highmem.h>
  21 #include <linux/gfp.h>
  22 #include <linux/memblock.h>
  23 #include <linux/dma-contiguous.h>
  24 #include <linux/sizes.h>
  25 #include <linux/stop_machine.h>
  26
  27 #include <asm/cp15.h>
  28 #include <asm/mach-types.h>
  29 #include <asm/memblock.h>
  30 #include <asm/prom.h>
  31 #include <asm/sections.h>
  32 #include <asm/setup.h>
  33 #include <asm/system_info.h>
  34 #include <asm/tlb.h>
  35 #include <asm/fixmap.h>
  36
  37 #include <asm/mach/arch.h>
  38 #include <asm/mach/map.h>
  39
  40 #include "mm.h"
  41
  42 #ifdef CONFIG_CPU_CP15_MMU
  43 unsigned long __init __clear_cr(unsigned long mask)
  44 {
  45         cr_alignment = cr_alignment & ~mask;
  46         return cr_alignment;
  47 }
  48 #endif
  49
  50 static phys_addr_t phys_initrd_start __initdata = 0;
  51 static unsigned long phys_initrd_size __initdata = 0;
  52
  53 static int __init early_initrd(char *p)
  54 {
  55         phys_addr_t start;
  56         unsigned long size;
  57         char *endp;
  58
  59         start = memparse(p, &endp);
  60         if (*endp == ',') {
  61                 size = memparse(endp + 1, NULL);
  62
  63                 phys_initrd_start = start;
  64                 phys_initrd_size = size;
  65         }
  66         return 0;
  67 }
  68 early_param("initrd", early_initrd);
  69
  70 static int __init parse_tag_initrd(const struct tag *tag)
  71 {
  72         pr_warn("ATAG_INITRD is deprecated; "
  73                 "please update your bootloader.\n");
  74         phys_initrd_start = __virt_to_phys(tag->u.initrd.start);
  75         phys_initrd_size = tag->u.initrd.size;
  76         return 0;
  77 }
  78
  79 __tagtable(ATAG_INITRD, parse_tag_initrd);
  80
  81 static int __init parse_tag_initrd2(const struct tag *tag)
  82 {
  83         phys_initrd_start = tag->u.initrd.start;
  84         phys_initrd_size = tag->u.initrd.size;
  85         return 0;
  86 }
  87
  88 __tagtable(ATAG_INITRD2, parse_tag_initrd2);
  89
  90 static void __init find_limits(unsigned long *min, unsigned long *max_low,
  91                                unsigned long *max_high)
  92 {
  93         *max_low = PFN_DOWN(memblock_get_current_limit());
  94         *min = PFN_UP(memblock_start_of_DRAM());
  95         *max_high = PFN_DOWN(memblock_end_of_DRAM());
  96 }
  97
  98 #ifdef CONFIG_ZONE_DMA
  99
 100 phys_addr_t arm_dma_zone_size __read_mostly;
 101 EXPORT_SYMBOL(arm_dma_zone_size);
 102
 103 /*
 104  * The DMA mask corresponding to the maximum bus address allocatable
 105  * using GFP_DMA.  The default here places no restriction on DMA
 106  * allocations.  This must be the smallest DMA mask in the system,
 107  * so a successful GFP_DMA allocation will always satisfy this.
 108  */
 109 phys_addr_t arm_dma_limit;
 110 unsigned long arm_dma_pfn_limit;
 111
 112 static void __init arm_adjust_dma_zone(unsigned long *size, unsigned long *hole,
 113         unsigned long dma_size)
 114 {
 115         if (size[0] <= dma_size)
 116                 return;
 117
 118         size[ZONE_NORMAL] = size[0] - dma_size;
 119         size[ZONE_DMA] = dma_size;
 120         hole[ZONE_NORMAL] = hole[0];
 121         hole[ZONE_DMA] = 0;
 122 }
 123 #endif
 124
 125 void __init setup_dma_zone(const struct machine_desc *mdesc)
 126 {
 127 #ifdef CONFIG_ZONE_DMA
 128         if (mdesc->dma_zone_size) {
 129                 arm_dma_zone_size = mdesc->dma_zone_size;
 130                 arm_dma_limit = PHYS_OFFSET + arm_dma_zone_size - 1;
 131         } else
 132                 arm_dma_limit = 0xffffffff;
 133         arm_dma_pfn_limit = arm_dma_limit >> PAGE_SHIFT;
 134 #endif
 135 }
 136
 137 static void __init zone_sizes_init(unsigned long min, unsigned long max_low,
 138         unsigned long max_high)
 139 {
 140         unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES];
 141         struct memblock_region *reg;
 142
 143         /*
 144          * initialise the zones.
 145          */
 146         memset(zone_size, 0, sizeof(zone_size));
 147
 148         /*
 149          * The memory size has already been determined.  If we need
 150          * to do anything fancy with the allocation of this memory
 151          * to the zones, now is the time to do it.
 152          */
 153         zone_size[0] = max_low - min;
 154 #ifdef CONFIG_HIGHMEM
 155         zone_size[ZONE_HIGHMEM] = max_high - max_low;
 156 #endif
 157
 158         /*
 159          * Calculate the size of the holes.
 160          *  holes = node_size - sum(bank_sizes)
 161          */
 162         memcpy(zhole_size, zone_size, sizeof(zhole_size));
 163         for_each_memblock(memory, reg) {
 164                 unsigned long start = memblock_region_memory_base_pfn(reg);
 165                 unsigned long end = memblock_region_memory_end_pfn(reg);
 166
 167                 if (start < max_low) {
 168                         unsigned long low_end = min(end, max_low);
 169                         zhole_size[0] -= low_end - start;
 170                 }
 171 #ifdef CONFIG_HIGHMEM
 172                 if (end > max_low) {
 173                         unsigned long high_start = max(start, max_low);
 174                         zhole_size[ZONE_HIGHMEM] -= end - high_start;
 175                 }
 176 #endif
 177         }
 178
 179 #ifdef CONFIG_ZONE_DMA
 180         /*
 181          * Adjust the sizes according to any special requirements for
 182          * this machine type.
 183          */
 184         if (arm_dma_zone_size)
 185                 arm_adjust_dma_zone(zone_size, zhole_size,
 186                         arm_dma_zone_size >> PAGE_SHIFT);
 187 #endif
 188
 189         free_area_init_node(0, zone_size, min, zhole_size);
 190 }
 191
 192 #ifdef CONFIG_HAVE_ARCH_PFN_VALID
 193 int pfn_valid(unsigned long pfn)
 194 {
 195         return memblock_is_memory(__pfn_to_phys(pfn));
 196 }
 197 EXPORT_SYMBOL(pfn_valid);
 198 #endif
 199
 200 #ifndef CONFIG_SPARSEMEM
 201 static void __init arm_memory_present(void)
 202 {
 203 }
 204 #else
 205 static void __init arm_memory_present(void)
 206 {
 207         struct memblock_region *reg;
 208
 209         for_each_memblock(memory, reg)
 210                 memory_present(0, memblock_region_memory_base_pfn(reg),
 211                                memblock_region_memory_end_pfn(reg));
 212 }
 213 #endif
 214
 215 static bool arm_memblock_steal_permitted = true;
 216
 217 phys_addr_t __init arm_memblock_steal(phys_addr_t size, phys_addr_t align)
 218 {
 219         phys_addr_t phys;
 220
 221         BUG_ON(!arm_memblock_steal_permitted);
 222
 223         phys = memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ANYWHERE);
 224         memblock_free(phys, size);
 225         memblock_remove(phys, size);
 226
 227         return phys;
 228 }
 229
 230 void __init arm_memblock_init(const struct machine_desc *mdesc)
 231 {
 232         /* Register the kernel text, kernel data and initrd with memblock. */
 233 #ifdef CONFIG_XIP_KERNEL
 234         memblock_reserve(__pa(_sdata), _end - _sdata);
 235 #else
 236         memblock_reserve(__pa(_stext), _end - _stext);
 237 #endif
 238 #ifdef CONFIG_BLK_DEV_INITRD
 239         /* FDT scan will populate initrd_start */
 240         if (initrd_start && !phys_initrd_size) {
 241                 phys_initrd_start = __virt_to_phys(initrd_start);
 242                 phys_initrd_size = initrd_end - initrd_start;
 243         }
 244         initrd_start = initrd_end = 0;
 245         if (phys_initrd_size &&
 246             !memblock_is_region_memory(phys_initrd_start, phys_initrd_size)) {
 247                 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region - disabling initrd\n",
 248                        (u64)phys_initrd_start, phys_initrd_size);
 249                 phys_initrd_start = phys_initrd_size = 0;
 250         }
 251         if (phys_initrd_size &&
 252             memblock_is_region_reserved(phys_initrd_start, phys_initrd_size)) {
 253                 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region - disabling initrd\n",
 254                        (u64)phys_initrd_start, phys_initrd_size);
 255                 phys_initrd_start = phys_initrd_size = 0;
 256         }
 257         if (phys_initrd_size) {
 258                 memblock_reserve(phys_initrd_start, phys_initrd_size);
 259
 260                 /* Now convert initrd to virtual addresses */
 261                 initrd_start = __phys_to_virt(phys_initrd_start);
 262                 initrd_end = initrd_start + phys_initrd_size;
 263         }
 264 #endif
 265
 266         arm_mm_memblock_reserve();
 267
 268         /* reserve any platform specific memblock areas */
 269         if (mdesc->reserve)
 270                 mdesc->reserve();
 271
 272         early_init_fdt_reserve_self();
 273         early_init_fdt_scan_reserved_mem();
 274
 275         /* reserve memory for DMA contiguous allocations */
 276         dma_contiguous_reserve(arm_dma_limit);
 277
 278         arm_memblock_steal_permitted = false;
 279         memblock_dump_all();
 280 }
 281
 282 void __init bootmem_init(void)
 283 {
 284         unsigned long min, max_low, max_high;
 285
 286         memblock_allow_resize();
 287         max_low = max_high = 0;
 288
 289         find_limits(&min, &max_low, &max_high);
 290
 291         early_memtest((phys_addr_t)min << PAGE_SHIFT,
 292                       (phys_addr_t)max_low << PAGE_SHIFT);
 293
 294         /*
 295          * Sparsemem tries to allocate bootmem in memory_present(),
 296          * so must be done after the fixed reservations
 297          */
 298         arm_memory_present();
 299
 300         /*
 301          * sparse_init() needs the bootmem allocator up and running.
 302          */
 303         sparse_init();
 304
 305         /*
 306          * Now free the memory - free_area_init_node needs
 307          * the sparse mem_map arrays initialized by sparse_init()
 308          * for memmap_init_zone(), otherwise all PFNs are invalid.
 309          */
 310         zone_sizes_init(min, max_low, max_high);
 311
 312         /*
 313          * This doesn't seem to be used by the Linux memory manager any
 314          * more, but is used by ll_rw_block.  If we can get rid of it, we
 315          * also get rid of some of the stuff above as well.
 316          */
 317         min_low_pfn = min;
 318         max_low_pfn = max_low;
 319         max_pfn = max_high;
 320 }
 321
 322 /*
 323  * Poison init memory with an undefined instruction (ARM) or a branch to an
 324  * undefined instruction (Thumb).
 325  */
 326 static inline void poison_init_mem(void *s, size_t count)
 327 {
 328         u32 *p = (u32 *)s;
 329         for (; count != 0; count -= 4)
 330                 *p++ = 0xe7fddef0;
 331 }
 332
 333 static inline void
 334 free_memmap(unsigned long start_pfn, unsigned long end_pfn)
 335 {
 336         struct page *start_pg, *end_pg;
 337         phys_addr_t pg, pgend;
 338
 339         /*
 340          * Convert start_pfn/end_pfn to a struct page pointer.
 341          */
 342         start_pg = pfn_to_page(start_pfn - 1) + 1;
 343         end_pg = pfn_to_page(end_pfn - 1) + 1;
 344
 345         /*
 346          * Convert to physical addresses, and
 347          * round start upwards and end downwards.
 348          */
 349         pg = PAGE_ALIGN(__pa(start_pg));
 350         pgend = __pa(end_pg) & PAGE_MASK;
 351
 352         /*
 353          * If there are free pages between these,
 354          * free the section of the memmap array.
 355          */
 356         if (pg < pgend)
 357                 memblock_free_early(pg, pgend - pg);
 358 }
 359
 360 /*
 361  * The mem_map array can get very big.  Free the unused area of the memory map.
 362  */
 363 static void __init free_unused_memmap(void)
 364 {
 365         unsigned long start, prev_end = 0;
 366         struct memblock_region *reg;
 367
 368         /*
 369          * This relies on each bank being in address order.
 370          * The banks are sorted previously in bootmem_init().
 371          */
 372         for_each_memblock(memory, reg) {
 373                 start = memblock_region_memory_base_pfn(reg);
 374
 375 #ifdef CONFIG_SPARSEMEM
 376                 /*
 377                  * Take care not to free memmap entries that don't exist
 378                  * due to SPARSEMEM sections which aren't present.
 379                  */
 380                 start = min(start,
 381                                  ALIGN(prev_end, PAGES_PER_SECTION));
 382 #else
 383                 /*
 384                  * Align down here since the VM subsystem insists that the
 385                  * memmap entries are valid from the bank start aligned to
 386                  * MAX_ORDER_NR_PAGES.
 387                  */
 388                 start = round_down(start, MAX_ORDER_NR_PAGES);
 389 #endif
 390                 /*
 391                  * If we had a previous bank, and there is a space
 392                  * between the current bank and the previous, free it.
 393                  */
 394                 if (prev_end && prev_end < start)
 395                         free_memmap(prev_end, start);
 396
 397                 /*
 398                  * Align up here since the VM subsystem insists that the
 399                  * memmap entries are valid from the bank end aligned to
 400                  * MAX_ORDER_NR_PAGES.
 401                  */
 402                 prev_end = ALIGN(memblock_region_memory_end_pfn(reg),
 403                                  MAX_ORDER_NR_PAGES);
 404         }
 405
 406 #ifdef CONFIG_SPARSEMEM
 407         if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
 408                 free_memmap(prev_end,
 409                             ALIGN(prev_end, PAGES_PER_SECTION));
 410 #endif
 411 }
 412
 413 #ifdef CONFIG_HIGHMEM
 414 static inline void free_area_high(unsigned long pfn, unsigned long end)
 415 {
 416         for (; pfn < end; pfn++)
 417                 free_highmem_page(pfn_to_page(pfn));
 418 }
 419 #endif
 420
 421 static void __init free_highpages(void)
 422 {
 423 #ifdef CONFIG_HIGHMEM
 424         unsigned long max_low = max_low_pfn;
 425         struct memblock_region *mem, *res;
 426
 427         /* set highmem page free */
 428         for_each_memblock(memory, mem) {
 429                 unsigned long start = memblock_region_memory_base_pfn(mem);
 430                 unsigned long end = memblock_region_memory_end_pfn(mem);
 431
 432                 /* Ignore complete lowmem entries */
 433                 if (end <= max_low)
 434                         continue;
 435
 436                 /* Truncate partial highmem entries */
 437                 if (start < max_low)
 438                         start = max_low;
 439
 440                 /* Find and exclude any reserved regions */
 441                 for_each_memblock(reserved, res) {
 442                         unsigned long res_start, res_end;
 443
 444                         res_start = memblock_region_reserved_base_pfn(res);
 445                         res_end = memblock_region_reserved_end_pfn(res);
 446
 447                         if (res_end < start)
 448                                 continue;
 449                         if (res_start < start)
 450                                 res_start = start;
 451                         if (res_start > end)
 452                                 res_start = end;
 453                         if (res_end > end)
 454                                 res_end = end;
 455                         if (res_start != start)
 456                                 free_area_high(start, res_start);
 457                         start = res_end;
 458                         if (start == end)
 459                                 break;
 460                 }
 461
 462                 /* And now free anything which remains */
 463                 if (start < end)
 464                         free_area_high(start, end);
 465         }
 466 #endif
 467 }
 468
 469 #define MLK(b, t) (b), (t), (((t) - (b)) >> 10)
 470 #define MLM(b, t) (b), (t), (((t) - (b)) >> 20)
 471 #define MLK_ROUNDUP(b, t) (b), (t), (DIV_ROUND_UP(((t) - (b)), SZ_1K))
 472
 473 #ifdef CONFIG_ENABLE_VMALLOC_SAVING
 474 static void print_vmalloc_lowmem_info(void)
 475 {
 476         struct memblock_region *reg, *prev_reg = NULL;
 477
 478         pr_notice(
 479                 "          vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n",
 480                 MLM((unsigned long)high_memory, VMALLOC_END));
 481
 482         for_each_memblock_rev(memory, reg) {
 483                 phys_addr_t start_phys = reg->base;
 484                 phys_addr_t end_phys = reg->base + reg->size;
 485
 486                 if (start_phys > arm_lowmem_limit)
 487                         continue;
 488
 489                 if (end_phys > arm_lowmem_limit)
 490                         end_phys = arm_lowmem_limit;
 491
 492                 if (prev_reg == NULL) {
 493                         prev_reg = reg;
 494
 495                         pr_notice(
 496                         "          lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n",
 497                         MLM((unsigned long)__va(start_phys),
 498                         (unsigned long)__va(end_phys)));
 499
 500                         continue;
 501                 }
 502
 503                 pr_notice(
 504                 "          vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n",
 505                 MLM((unsigned long)__va(end_phys),
 506                 (unsigned long)__va(prev_reg->base)));
 507
 508
 509                 pr_notice(
 510                 "          lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n",
 511                 MLM((unsigned long)__va(start_phys),
 512                 (unsigned long)__va(end_phys)));
 513         }
 514 }
 515 #endif
 516
 517 /*
 518  * mem_init() marks the free areas in the mem_map and tells us how much
 519  * memory is free.  This is done after various parts of the system have
 520  * claimed their memory after the kernel image.
 521  */
 522 void __init mem_init(void)
 523 {
 524 #ifdef CONFIG_HAVE_TCM
 525         /* These pointers are filled in on TCM detection */
 526         extern u32 dtcm_end;
 527         extern u32 itcm_end;
 528 #endif
 529
 530         set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
 531
 532         /* this will put all unused low memory onto the freelists */
 533         free_unused_memmap();
 534         free_all_bootmem();
 535
 536 #ifdef CONFIG_SA1111
 537         /* now that our DMA memory is actually so designated, we can free it */
 538         free_reserved_area(__va(PHYS_OFFSET), swapper_pg_dir, -1, NULL);
 539 #endif
 540
 541         free_highpages();
 542
 543         mem_init_print_info(NULL);
 544
 545         pr_notice("Virtual kernel memory layout:\n"
 546                         "    vector  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 547 #ifdef CONFIG_HAVE_TCM
 548                         "    DTCM    : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 549                         "    ITCM    : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 550 #endif
 551                         "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n",
 552                         MLK(UL(CONFIG_VECTORS_BASE), UL(CONFIG_VECTORS_BASE) +
 553                                 (PAGE_SIZE)),
 554 #ifdef CONFIG_HAVE_TCM
 555                         MLK(DTCM_OFFSET, (unsigned long) dtcm_end),
 556                         MLK(ITCM_OFFSET, (unsigned long) itcm_end),
 557 #endif
 558                         MLK(FIXADDR_START, FIXADDR_END));
 559 #ifdef CONFIG_ENABLE_VMALLOC_SAVING
 560         print_vmalloc_lowmem_info();
 561 #else
 562         printk(KERN_NOTICE
 563                    "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 564                    "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n",
 565                         MLM(VMALLOC_START, VMALLOC_END),
 566                         MLM(PAGE_OFFSET, (unsigned long)high_memory));
 567 #endif
 568         printk(KERN_NOTICE
 569 #ifdef CONFIG_HIGHMEM
 570                    "    pkmap   : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 571 #endif
 572 #ifdef CONFIG_MODULES
 573                    "    modules : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 574 #endif
 575                    "      .text : 0x%p" " - 0x%p" "   (%4d kB)\n"
 576                    "      .init : 0x%p" " - 0x%p" "   (%4d kB)\n"
 577                    "      .data : 0x%p" " - 0x%p" "   (%4d kB)\n"
 578                    "       .bss : 0x%p" " - 0x%p" "   (%4d kB)\n",
 579 #ifdef CONFIG_HIGHMEM
 580                         MLM(PKMAP_BASE, (PKMAP_BASE) + (LAST_PKMAP) *
 581                                 (PAGE_SIZE)),
 582 #endif
 583 #ifdef CONFIG_MODULES
 584                         MLM(MODULES_VADDR, MODULES_END),
 585 #endif
 586
 587                         MLK_ROUNDUP(_text, _etext),
 588                         MLK_ROUNDUP(__init_begin, __init_end),
 589                         MLK_ROUNDUP(_sdata, _edata),
 590                         MLK_ROUNDUP(__bss_start, __bss_stop));
 591
 592         /*
 593          * Check boundaries twice: Some fundamental inconsistencies can
 594          * be detected at build time already.
 595          */
 596 #ifdef CONFIG_MMU
 597         BUILD_BUG_ON(TASK_SIZE                          > MODULES_VADDR);
 598         BUG_ON(TASK_SIZE                                > MODULES_VADDR);
 599 #endif
 600
 601 #ifdef CONFIG_HIGHMEM
 602         BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
 603         BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE      > PAGE_OFFSET);
 604 #endif
 605
 606         if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) {
 607                 extern int sysctl_overcommit_memory;
 608                 /*
 609                  * On a machine this small we won't get
 610                  * anywhere without overcommit, so turn
 611                  * it on by default.
 612                  */
 613                 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
 614         }
 615 }
 616
 617 #undef MLK
 618 #undef MLM
 619 #undef MLK_ROUNDUP
 620
 621 #ifdef CONFIG_ARM_KERNMEM_PERMS
 622 struct section_perm {
 623         unsigned long start;
 624         unsigned long end;
 625         pmdval_t mask;
 626         pmdval_t prot;
 627         pmdval_t clear;
 628         pteval_t ptemask;
 629         pteval_t pteprot;
 630         pteval_t pteclear;
 631 };
 632
 633 static struct section_perm nx_perms[] = {
 634         /* Make pages tables, etc before _stext RW (set NX). */
 635         {
 636                 .start  = PAGE_OFFSET,
 637                 .end    = (unsigned long)_stext,
 638                 .mask   = ~PMD_SECT_XN,
 639                 .prot   = PMD_SECT_XN,
 640                 .ptemask = ~L_PTE_XN,
 641                 .pteprot = L_PTE_XN,
 642         },
 643         /* Make init RW (set NX). */
 644         {
 645                 .start  = (unsigned long)__init_begin,
 646                 .end    = (unsigned long)_sdata,
 647                 .mask   = ~PMD_SECT_XN,
 648                 .prot   = PMD_SECT_XN,
 649                 .ptemask = ~L_PTE_XN,
 650                 .pteprot = L_PTE_XN,
 651         },
 652 #ifdef CONFIG_DEBUG_RODATA
 653         /* Make rodata NX (set RO in ro_perms below). */
 654         {
 655                 .start  = (unsigned long)__start_rodata,
 656                 .end    = (unsigned long)__init_begin,
 657                 .mask   = ~PMD_SECT_XN,
 658                 .prot   = PMD_SECT_XN,
 659                 .ptemask = ~L_PTE_XN,
 660                 .pteprot = L_PTE_XN,
 661         },
 662 #endif
 663 };
 664
 665 #ifdef CONFIG_DEBUG_RODATA
 666 static struct section_perm ro_perms[] = {
 667         /* Make kernel code and rodata RX (set RO). */
 668         {
 669                 .start  = (unsigned long)_stext,
 670                 .end    = (unsigned long)__init_begin,
 671 #ifdef CONFIG_ARM_LPAE
 672                 .mask   = ~(L_PMD_SECT_RDONLY | PMD_SECT_AP2),
 673                 .prot   = L_PMD_SECT_RDONLY | PMD_SECT_AP2,
 674 #else
 675                 .mask   = ~(PMD_SECT_APX | PMD_SECT_AP_WRITE),
 676                 .prot   = PMD_SECT_APX | PMD_SECT_AP_WRITE,
 677                 .clear  = PMD_SECT_AP_WRITE,
 678 #endif
 679                 .ptemask = ~L_PTE_RDONLY,
 680                 .pteprot = L_PTE_RDONLY,
 681         },
 682 };
 683 #endif
 684
 685 /*
 686  * Updates section permissions only for the current mm (sections are
 687  * copied into each mm). During startup, this is the init_mm. Is only
 688  * safe to be called with preemption disabled, as under stop_machine().
 689  */
 690 struct pte_data {
 691         pteval_t mask;
 692         pteval_t val;
 693 };
 694
 695 static int __pte_update(pte_t *ptep, pgtable_t token, unsigned long addr,
 696                         void *d)
 697 {
 698         struct pte_data *data = d;
 699         pte_t pte = *ptep;
 700
 701         pte = __pte((pte_val(*ptep) & data->mask) | data->val);
 702         set_pte_ext(ptep, pte, 0);
 703
 704         return 0;
 705 }
 706
 707 static inline void pte_update(unsigned long addr, pteval_t mask,
 708                                   pteval_t prot, struct mm_struct *mm)
 709 {
 710         struct pte_data data;
 711
 712         data.mask = mask;
 713         data.val = prot;
 714
 715         apply_to_page_range(mm, addr, SECTION_SIZE, __pte_update, &data);
 716         flush_tlb_kernel_range(addr, addr + SECTION_SIZE);
 717 }
 718
 719 static inline void section_update(unsigned long addr, pmdval_t mask,
 720                                   pmdval_t prot, struct mm_struct *mm)
 721 {
 722         pmd_t *pmd;
 723
 724         pmd = pmd_offset(pud_offset(pgd_offset(mm, addr), addr), addr);
 725
 726 #ifdef CONFIG_ARM_LPAE
 727         pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
 728 #else
 729         if (addr & SECTION_SIZE)
 730                 pmd[1] = __pmd((pmd_val(pmd[1]) & mask) | prot);
 731         else
 732                 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
 733 #endif
 734         flush_pmd_entry(pmd);
 735         local_flush_tlb_kernel_range(addr, addr + SECTION_SIZE);
 736 }
 737
 738 /* Make sure extended page tables are in use. */
 739 static inline bool arch_has_strict_perms(void)
 740 {
 741         if (cpu_architecture() < CPU_ARCH_ARMv6)
 742                 return false;
 743
 744         return !!(get_cr() & CR_XP);
 745 }
 746
 747 void set_section_perms(struct section_perm *perms, int n, bool set,
 748                         struct mm_struct *mm)
 749 {
 750         size_t i;
 751         unsigned long addr;
 752
 753         if (!arch_has_strict_perms())
 754                 return;
 755
 756         for (i = 0; i < n; i++) {
 757                 if (!IS_ALIGNED(perms[i].start, SECTION_SIZE) ||
 758                     !IS_ALIGNED(perms[i].end, SECTION_SIZE)) {
 759                         pr_err("BUG: section %lx-%lx not aligned to %lx\n",
 760                                 perms[i].start, perms[i].end,
 761                                 SECTION_SIZE);
 762                         continue;
 763                 }
 764
 765                 for (addr = perms[i].start;
 766                      addr < perms[i].end;
 767                      addr += SECTION_SIZE) {
 768                         pmd_t *pmd;
 769
 770                         pmd = pmd_offset(pud_offset(pgd_offset(mm, addr),
 771                                                 addr), addr);
 772                         if (pmd_bad(*pmd))
 773                                 section_update(addr, perms[i].mask,
 774                                                set ? perms[i].prot : perms[i].clear,
 775                                                mm);
 776                         else
 777                                 pte_update(addr, perms[i].ptemask,
 778                                                set ? perms[i].pteprot : perms[i].pteclear,
 779                                                mm);
 780                 }
 781         }
 782 }
 783
 784 static void update_sections_early(struct section_perm perms[], int n)
 785 {
 786         struct task_struct *t, *s;
 787
 788         read_lock(&tasklist_lock);
 789         for_each_process(t) {
 790                 if (t->flags & PF_KTHREAD)
 791                         continue;
 792                 for_each_thread(t, s)
 793                         set_section_perms(perms, n, true, s->mm);
 794         }
 795         read_unlock(&tasklist_lock);
 796         set_section_perms(perms, n, true, current->active_mm);
 797         set_section_perms(perms, n, true, &init_mm);
 798 }
 799
 800 int __fix_kernmem_perms(void *unused)
 801 {
 802         update_sections_early(nx_perms, ARRAY_SIZE(nx_perms));
 803         return 0;
 804 }
 805
 806 void fix_kernmem_perms(void)
 807 {
 808         stop_machine(__fix_kernmem_perms, NULL, NULL);
 809 }
 810
 811 #ifdef CONFIG_DEBUG_RODATA
 812 int __mark_rodata_ro(void *unused)
 813 {
 814         update_sections_early(ro_perms, ARRAY_SIZE(ro_perms));
 815         return 0;
 816 }
 817
 818 static int kernel_set_to_readonly __read_mostly;
 819
 820 void mark_rodata_ro(void)
 821 {
 822         kernel_set_to_readonly = 1;
 823         stop_machine(__mark_rodata_ro, NULL, NULL);
 824 }
 825
 826 void set_kernel_text_rw(void)
 827 {
 828         if (!kernel_set_to_readonly)
 829                 return;
 830
 831         set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), false,
 832                                 current->active_mm);
 833 }
 834
 835 void set_kernel_text_ro(void)
 836 {
 837         if (!kernel_set_to_readonly)
 838                 return;
 839
 840         set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), true,
 841                                 current->active_mm);
 842 }
 843 #endif /* CONFIG_DEBUG_RODATA */
 844
 845 #else
 846 static inline void fix_kernmem_perms(void) { }
 847 #endif /* CONFIG_ARM_KERNMEM_PERMS */
 848
 849 void free_tcmmem(void)
 850 {
 851 #ifdef CONFIG_HAVE_TCM
 852         extern char __tcm_start, __tcm_end;
 853
 854         poison_init_mem(&__tcm_start, &__tcm_end - &__tcm_start);
 855         free_reserved_area(&__tcm_start, &__tcm_end, -1, "TCM link");
 856 #endif
 857 }
 858
 859 void free_initmem(void)
 860 {
 861         fix_kernmem_perms();
 862         free_tcmmem();
 863
 864         poison_init_mem(__init_begin, __init_end - __init_begin);
 865         if (!machine_is_integrator() && !machine_is_cintegrator())
 866                 free_initmem_default(-1);
 867 }
 868
 869 #ifdef CONFIG_BLK_DEV_INITRD
 870
 871 static int keep_initrd;
 872
 873 void free_initrd_mem(unsigned long start, unsigned long end)
 874 {
 875         if (!keep_initrd) {
 876                 if (start == initrd_start)
 877                         start = round_down(start, PAGE_SIZE);
 878                 if (end == initrd_end)
 879                         end = round_up(end, PAGE_SIZE);
 880
 881                 poison_init_mem((void *)start, PAGE_ALIGN(end) - start);
 882                 free_reserved_area((void *)start, (void *)end, -1, "initrd");
 883         }
 884 }
 885
 886 static int __init keepinitrd_setup(char *__unused)
 887 {
 888         keep_initrd = 1;
 889         return 1;
 890 }
 891
 892 __setup("keepinitrd", keepinitrd_setup);
 893 #endif