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