arch/arm/mm/init.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/arch/arm/mm/init.c
   4  *
   5  *  Copyright (C) 1995-2005 Russell King
   6  */
   7 #include <linux/kernel.h>
   8 #include <linux/errno.h>
   9 #include <linux/swap.h>
  10 #include <linux/init.h>
  11 #include <linux/mman.h>
  12 #include <linux/sched/signal.h>
  13 #include <linux/sched/task.h>
  14 #include <linux/export.h>
  15 #include <linux/nodemask.h>
  16 #include <linux/initrd.h>
  17 #include <linux/of_fdt.h>
  18 #include <linux/highmem.h>
  19 #include <linux/gfp.h>
  20 #include <linux/memblock.h>
  21 #include <linux/dma-map-ops.h>
  22 #include <linux/sizes.h>
  23 #include <linux/stop_machine.h>
  24 #include <linux/swiotlb.h>
  25
  26 #include <asm/cp15.h>
  27 #include <asm/mach-types.h>
  28 #include <asm/memblock.h>
  29 #include <asm/memory.h>
  30 #include <asm/prom.h>
  31 #include <asm/sections.h>
  32 #include <asm/setup.h>
  33 #include <asm/set_memory.h>
  34 #include <asm/system_info.h>
  35 #include <asm/tlb.h>
  36 #include <asm/fixmap.h>
  37 #include <asm/ptdump.h>
  38
  39 #include <asm/mach/arch.h>
  40 #include <asm/mach/map.h>
  41
  42 #include "mm.h"
  43
  44 #ifdef CONFIG_CPU_CP15_MMU
  45 unsigned long __init __clear_cr(unsigned long mask)
  46 {
  47         cr_alignment = cr_alignment & ~mask;
  48         return cr_alignment;
  49 }
  50 #endif
  51
  52 #ifdef CONFIG_BLK_DEV_INITRD
  53 static int __init parse_tag_initrd(const struct tag *tag)
  54 {
  55         pr_warn("ATAG_INITRD is deprecated; "
  56                 "please update your bootloader.\n");
  57         phys_initrd_start = __virt_to_phys(tag->u.initrd.start);
  58         phys_initrd_size = tag->u.initrd.size;
  59         return 0;
  60 }
  61
  62 __tagtable(ATAG_INITRD, parse_tag_initrd);
  63
  64 static int __init parse_tag_initrd2(const struct tag *tag)
  65 {
  66         phys_initrd_start = tag->u.initrd.start;
  67         phys_initrd_size = tag->u.initrd.size;
  68         return 0;
  69 }
  70
  71 __tagtable(ATAG_INITRD2, parse_tag_initrd2);
  72 #endif
  73
  74 static void __init find_limits(unsigned long *min, unsigned long *max_low,
  75                                unsigned long *max_high)
  76 {
  77         *max_low = PFN_DOWN(memblock_get_current_limit());
  78         *min = PFN_UP(memblock_start_of_DRAM());
  79         *max_high = PFN_DOWN(memblock_end_of_DRAM());
  80 }
  81
  82 #ifdef CONFIG_ZONE_DMA
  83
  84 phys_addr_t arm_dma_zone_size __read_mostly;
  85 EXPORT_SYMBOL(arm_dma_zone_size);
  86
  87 /*
  88  * The DMA mask corresponding to the maximum bus address allocatable
  89  * using GFP_DMA.  The default here places no restriction on DMA
  90  * allocations.  This must be the smallest DMA mask in the system,
  91  * so a successful GFP_DMA allocation will always satisfy this.
  92  */
  93 phys_addr_t arm_dma_limit;
  94 unsigned long arm_dma_pfn_limit;
  95 #endif
  96
  97 void __init setup_dma_zone(const struct machine_desc *mdesc)
  98 {
  99 #ifdef CONFIG_ZONE_DMA
 100         if (mdesc->dma_zone_size) {
 101                 arm_dma_zone_size = mdesc->dma_zone_size;
 102                 arm_dma_limit = PHYS_OFFSET + arm_dma_zone_size - 1;
 103         } else
 104                 arm_dma_limit = 0xffffffff;
 105         arm_dma_pfn_limit = arm_dma_limit >> PAGE_SHIFT;
 106 #endif
 107 }
 108
 109 static void __init zone_sizes_init(unsigned long min, unsigned long max_low,
 110         unsigned long max_high)
 111 {
 112         unsigned long max_zone_pfn[MAX_NR_ZONES] = { 0 };
 113
 114 #ifdef CONFIG_ZONE_DMA
 115         max_zone_pfn[ZONE_DMA] = min(arm_dma_pfn_limit, max_low);
 116 #endif
 117         max_zone_pfn[ZONE_NORMAL] = max_low;
 118 #ifdef CONFIG_HIGHMEM
 119         max_zone_pfn[ZONE_HIGHMEM] = max_high;
 120 #endif
 121         free_area_init(max_zone_pfn);
 122 }
 123
 124 #ifdef CONFIG_HAVE_ARCH_PFN_VALID
 125 int pfn_valid(unsigned long pfn)
 126 {
 127         phys_addr_t addr = __pfn_to_phys(pfn);
 128
 129         if (__phys_to_pfn(addr) != pfn)
 130                 return 0;
 131
 132         return memblock_is_map_memory(addr);
 133 }
 134 EXPORT_SYMBOL(pfn_valid);
 135 #endif
 136
 137 static bool arm_memblock_steal_permitted = true;
 138
 139 phys_addr_t __init arm_memblock_steal(phys_addr_t size, phys_addr_t align)
 140 {
 141         phys_addr_t phys;
 142
 143         BUG_ON(!arm_memblock_steal_permitted);
 144
 145         phys = memblock_phys_alloc(size, align);
 146         if (!phys)
 147                 panic("Failed to steal %pa bytes at %pS\n",
 148                       &size, (void *)_RET_IP_);
 149
 150         memblock_free(phys, size);
 151         memblock_remove(phys, size);
 152
 153         return phys;
 154 }
 155
 156 static void __init arm_initrd_init(void)
 157 {
 158 #ifdef CONFIG_BLK_DEV_INITRD
 159         phys_addr_t start;
 160         unsigned long size;
 161
 162         initrd_start = initrd_end = 0;
 163
 164         if (!phys_initrd_size)
 165                 return;
 166
 167         /*
 168          * Round the memory region to page boundaries as per free_initrd_mem()
 169          * This allows us to detect whether the pages overlapping the initrd
 170          * are in use, but more importantly, reserves the entire set of pages
 171          * as we don't want these pages allocated for other purposes.
 172          */
 173         start = round_down(phys_initrd_start, PAGE_SIZE);
 174         size = phys_initrd_size + (phys_initrd_start - start);
 175         size = round_up(size, PAGE_SIZE);
 176
 177         if (!memblock_is_region_memory(start, size)) {
 178                 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region - disabling initrd\n",
 179                        (u64)start, size);
 180                 return;
 181         }
 182
 183         if (memblock_is_region_reserved(start, size)) {
 184                 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region - disabling initrd\n",
 185                        (u64)start, size);
 186                 return;
 187         }
 188
 189         memblock_reserve(start, size);
 190
 191         /* Now convert initrd to virtual addresses */
 192         initrd_start = __phys_to_virt(phys_initrd_start);
 193         initrd_end = initrd_start + phys_initrd_size;
 194 #endif
 195 }
 196
 197 #ifdef CONFIG_CPU_ICACHE_MISMATCH_WORKAROUND
 198 void check_cpu_icache_size(int cpuid)
 199 {
 200         u32 size, ctr;
 201
 202         asm("mrc p15, 0, %0, c0, c0, 1" : "=r" (ctr));
 203
 204         size = 1 << ((ctr & 0xf) + 2);
 205         if (cpuid != 0 && icache_size != size)
 206                 pr_info("CPU%u: detected I-Cache line size mismatch, workaround enabled\n",
 207                         cpuid);
 208         if (icache_size > size)
 209                 icache_size = size;
 210 }
 211 #endif
 212
 213 void __init arm_memblock_init(const struct machine_desc *mdesc)
 214 {
 215         /* Register the kernel text, kernel data and initrd with memblock. */
 216         memblock_reserve(__pa(KERNEL_START), KERNEL_END - KERNEL_START);
 217
 218         arm_initrd_init();
 219
 220         arm_mm_memblock_reserve();
 221
 222         /* reserve any platform specific memblock areas */
 223         if (mdesc->reserve)
 224                 mdesc->reserve();
 225
 226         early_init_fdt_scan_reserved_mem();
 227
 228         /* reserve memory for DMA contiguous allocations */
 229         dma_contiguous_reserve(arm_dma_limit);
 230
 231         arm_memblock_steal_permitted = false;
 232         memblock_dump_all();
 233 }
 234
 235 void __init bootmem_init(void)
 236 {
 237         memblock_allow_resize();
 238
 239         find_limits(&min_low_pfn, &max_low_pfn, &max_pfn);
 240
 241         early_memtest((phys_addr_t)min_low_pfn << PAGE_SHIFT,
 242                       (phys_addr_t)max_low_pfn << PAGE_SHIFT);
 243
 244         /*
 245          * sparse_init() tries to allocate memory from memblock, so must be
 246          * done after the fixed reservations
 247          */
 248         sparse_init();
 249
 250         /*
 251          * Now free the memory - free_area_init needs
 252          * the sparse mem_map arrays initialized by sparse_init()
 253          * for memmap_init_zone(), otherwise all PFNs are invalid.
 254          */
 255         zone_sizes_init(min_low_pfn, max_low_pfn, max_pfn);
 256 }
 257
 258 /*
 259  * Poison init memory with an undefined instruction (ARM) or a branch to an
 260  * undefined instruction (Thumb).
 261  */
 262 static inline void poison_init_mem(void *s, size_t count)
 263 {
 264         u32 *p = (u32 *)s;
 265         for (; count != 0; count -= 4)
 266                 *p++ = 0xe7fddef0;
 267 }
 268
 269 static void __init free_highpages(void)
 270 {
 271 #ifdef CONFIG_HIGHMEM
 272         unsigned long max_low = max_low_pfn;
 273         phys_addr_t range_start, range_end;
 274         u64 i;
 275
 276         /* set highmem page free */
 277         for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
 278                                 &range_start, &range_end, NULL) {
 279                 unsigned long start = PFN_UP(range_start);
 280                 unsigned long end = PFN_DOWN(range_end);
 281
 282                 /* Ignore complete lowmem entries */
 283                 if (end <= max_low)
 284                         continue;
 285
 286                 /* Truncate partial highmem entries */
 287                 if (start < max_low)
 288                         start = max_low;
 289
 290                 for (; start < end; start++)
 291                         free_highmem_page(pfn_to_page(start));
 292         }
 293 #endif
 294 }
 295
 296 /*
 297  * mem_init() marks the free areas in the mem_map and tells us how much
 298  * memory is free.  This is done after various parts of the system have
 299  * claimed their memory after the kernel image.
 300  */
 301 void __init mem_init(void)
 302 {
 303 #ifdef CONFIG_ARM_LPAE
 304         swiotlb_init(1);
 305 #endif
 306
 307         set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
 308
 309         /* this will put all unused low memory onto the freelists */
 310         memblock_free_all();
 311
 312 #ifdef CONFIG_SA1111
 313         /* now that our DMA memory is actually so designated, we can free it */
 314         free_reserved_area(__va(PHYS_OFFSET), swapper_pg_dir, -1, NULL);
 315 #endif
 316
 317         free_highpages();
 318
 319         mem_init_print_info(NULL);
 320
 321         /*
 322          * Check boundaries twice: Some fundamental inconsistencies can
 323          * be detected at build time already.
 324          */
 325 #ifdef CONFIG_MMU
 326         BUILD_BUG_ON(TASK_SIZE                          > MODULES_VADDR);
 327         BUG_ON(TASK_SIZE                                > MODULES_VADDR);
 328 #endif
 329
 330 #ifdef CONFIG_HIGHMEM
 331         BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
 332         BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE      > PAGE_OFFSET);
 333 #endif
 334 }
 335
 336 #ifdef CONFIG_STRICT_KERNEL_RWX
 337 struct section_perm {
 338         const char *name;
 339         unsigned long start;
 340         unsigned long end;
 341         pmdval_t mask;
 342         pmdval_t prot;
 343         pmdval_t clear;
 344 };
 345
 346 /* First section-aligned location at or after __start_rodata. */
 347 extern char __start_rodata_section_aligned[];
 348
 349 static struct section_perm nx_perms[] = {
 350         /* Make pages tables, etc before _stext RW (set NX). */
 351         {
 352                 .name   = "pre-text NX",
 353                 .start  = PAGE_OFFSET,
 354                 .end    = (unsigned long)_stext,
 355                 .mask   = ~PMD_SECT_XN,
 356                 .prot   = PMD_SECT_XN,
 357         },
 358         /* Make init RW (set NX). */
 359         {
 360                 .name   = "init NX",
 361                 .start  = (unsigned long)__init_begin,
 362                 .end    = (unsigned long)_sdata,
 363                 .mask   = ~PMD_SECT_XN,
 364                 .prot   = PMD_SECT_XN,
 365         },
 366         /* Make rodata NX (set RO in ro_perms below). */
 367         {
 368                 .name   = "rodata NX",
 369                 .start  = (unsigned long)__start_rodata_section_aligned,
 370                 .end    = (unsigned long)__init_begin,
 371                 .mask   = ~PMD_SECT_XN,
 372                 .prot   = PMD_SECT_XN,
 373         },
 374 };
 375
 376 static struct section_perm ro_perms[] = {
 377         /* Make kernel code and rodata RX (set RO). */
 378         {
 379                 .name   = "text/rodata RO",
 380                 .start  = (unsigned long)_stext,
 381                 .end    = (unsigned long)__init_begin,
 382 #ifdef CONFIG_ARM_LPAE
 383                 .mask   = ~(L_PMD_SECT_RDONLY | PMD_SECT_AP2),
 384                 .prot   = L_PMD_SECT_RDONLY | PMD_SECT_AP2,
 385 #else
 386                 .mask   = ~(PMD_SECT_APX | PMD_SECT_AP_WRITE),
 387                 .prot   = PMD_SECT_APX | PMD_SECT_AP_WRITE,
 388                 .clear  = PMD_SECT_AP_WRITE,
 389 #endif
 390         },
 391 };
 392
 393 /*
 394  * Updates section permissions only for the current mm (sections are
 395  * copied into each mm). During startup, this is the init_mm. Is only
 396  * safe to be called with preemption disabled, as under stop_machine().
 397  */
 398 static inline void section_update(unsigned long addr, pmdval_t mask,
 399                                   pmdval_t prot, struct mm_struct *mm)
 400 {
 401         pmd_t *pmd;
 402
 403         pmd = pmd_offset(pud_offset(p4d_offset(pgd_offset(mm, addr), addr), addr), addr);
 404
 405 #ifdef CONFIG_ARM_LPAE
 406         pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
 407 #else
 408         if (addr & SECTION_SIZE)
 409                 pmd[1] = __pmd((pmd_val(pmd[1]) & mask) | prot);
 410         else
 411                 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
 412 #endif
 413         flush_pmd_entry(pmd);
 414         local_flush_tlb_kernel_range(addr, addr + SECTION_SIZE);
 415 }
 416
 417 /* Make sure extended page tables are in use. */
 418 static inline bool arch_has_strict_perms(void)
 419 {
 420         if (cpu_architecture() < CPU_ARCH_ARMv6)
 421                 return false;
 422
 423         return !!(get_cr() & CR_XP);
 424 }
 425
 426 static void set_section_perms(struct section_perm *perms, int n, bool set,
 427                               struct mm_struct *mm)
 428 {
 429         size_t i;
 430         unsigned long addr;
 431
 432         if (!arch_has_strict_perms())
 433                 return;
 434
 435         for (i = 0; i < n; i++) {
 436                 if (!IS_ALIGNED(perms[i].start, SECTION_SIZE) ||
 437                     !IS_ALIGNED(perms[i].end, SECTION_SIZE)) {
 438                         pr_err("BUG: %s section %lx-%lx not aligned to %lx\n",
 439                                 perms[i].name, perms[i].start, perms[i].end,
 440                                 SECTION_SIZE);
 441                         continue;
 442                 }
 443
 444                 for (addr = perms[i].start;
 445                      addr < perms[i].end;
 446                      addr += SECTION_SIZE)
 447                         section_update(addr, perms[i].mask,
 448                                 set ? perms[i].prot : perms[i].clear, mm);
 449         }
 450
 451 }
 452
 453 /**
 454  * update_sections_early intended to be called only through stop_machine
 455  * framework and executed by only one CPU while all other CPUs will spin and
 456  * wait, so no locking is required in this function.
 457  */
 458 static void update_sections_early(struct section_perm perms[], int n)
 459 {
 460         struct task_struct *t, *s;
 461
 462         for_each_process(t) {
 463                 if (t->flags & PF_KTHREAD)
 464                         continue;
 465                 for_each_thread(t, s)
 466                         if (s->mm)
 467                                 set_section_perms(perms, n, true, s->mm);
 468         }
 469         set_section_perms(perms, n, true, current->active_mm);
 470         set_section_perms(perms, n, true, &init_mm);
 471 }
 472
 473 static int __fix_kernmem_perms(void *unused)
 474 {
 475         update_sections_early(nx_perms, ARRAY_SIZE(nx_perms));
 476         return 0;
 477 }
 478
 479 static void fix_kernmem_perms(void)
 480 {
 481         stop_machine(__fix_kernmem_perms, NULL, NULL);
 482 }
 483
 484 static int __mark_rodata_ro(void *unused)
 485 {
 486         update_sections_early(ro_perms, ARRAY_SIZE(ro_perms));
 487         return 0;
 488 }
 489
 490 static int kernel_set_to_readonly __read_mostly;
 491
 492 void mark_rodata_ro(void)
 493 {
 494         kernel_set_to_readonly = 1;
 495         stop_machine(__mark_rodata_ro, NULL, NULL);
 496         debug_checkwx();
 497 }
 498
 499 void set_kernel_text_rw(void)
 500 {
 501         if (!kernel_set_to_readonly)
 502                 return;
 503
 504         set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), false,
 505                                 current->active_mm);
 506 }
 507
 508 void set_kernel_text_ro(void)
 509 {
 510         if (!kernel_set_to_readonly)
 511                 return;
 512
 513         set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), true,
 514                                 current->active_mm);
 515 }
 516
 517 #else
 518 static inline void fix_kernmem_perms(void) { }
 519 #endif /* CONFIG_STRICT_KERNEL_RWX */
 520
 521 void free_initmem(void)
 522 {
 523         fix_kernmem_perms();
 524
 525         poison_init_mem(__init_begin, __init_end - __init_begin);
 526         if (!machine_is_integrator() && !machine_is_cintegrator())
 527                 free_initmem_default(-1);
 528 }
 529
 530 #ifdef CONFIG_BLK_DEV_INITRD
 531 void free_initrd_mem(unsigned long start, unsigned long end)
 532 {
 533         if (start == initrd_start)
 534                 start = round_down(start, PAGE_SIZE);
 535         if (end == initrd_end)
 536                 end = round_up(end, PAGE_SIZE);
 537
 538         poison_init_mem((void *)start, PAGE_ALIGN(end) - start);
 539         free_reserved_area((void *)start, (void *)end, -1, "initrd");
 540 }
 541 #endif