linux/arch/arm/mm/init.c
<<
>>
Prefs
   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
  45unsigned 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
  53static 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
  64static 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
  74static 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
  84phys_addr_t arm_dma_zone_size __read_mostly;
  85EXPORT_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 */
  93phys_addr_t arm_dma_limit;
  94unsigned long arm_dma_pfn_limit;
  95#endif
  96
  97void __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
 109static 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
 125int pfn_valid(unsigned long pfn)
 126{
 127        phys_addr_t addr = __pfn_to_phys(pfn);
 128        unsigned long pageblock_size = PAGE_SIZE * pageblock_nr_pages;
 129
 130        if (__phys_to_pfn(addr) != pfn)
 131                return 0;
 132
 133        /*
 134         * If address less than pageblock_size bytes away from a present
 135         * memory chunk there still will be a memory map entry for it
 136         * because we round freed memory map to the pageblock boundaries.
 137         */
 138        if (memblock_overlaps_region(&memblock.memory,
 139                                     ALIGN_DOWN(addr, pageblock_size),
 140                                     pageblock_size))
 141                return 1;
 142
 143        return 0;
 144}
 145EXPORT_SYMBOL(pfn_valid);
 146#endif
 147
 148static bool arm_memblock_steal_permitted = true;
 149
 150phys_addr_t __init arm_memblock_steal(phys_addr_t size, phys_addr_t align)
 151{
 152        phys_addr_t phys;
 153
 154        BUG_ON(!arm_memblock_steal_permitted);
 155
 156        phys = memblock_phys_alloc(size, align);
 157        if (!phys)
 158                panic("Failed to steal %pa bytes at %pS\n",
 159                      &size, (void *)_RET_IP_);
 160
 161        memblock_free(phys, size);
 162        memblock_remove(phys, size);
 163
 164        return phys;
 165}
 166
 167static void __init arm_initrd_init(void)
 168{
 169#ifdef CONFIG_BLK_DEV_INITRD
 170        phys_addr_t start;
 171        unsigned long size;
 172
 173        initrd_start = initrd_end = 0;
 174
 175        if (!phys_initrd_size)
 176                return;
 177
 178        /*
 179         * Round the memory region to page boundaries as per free_initrd_mem()
 180         * This allows us to detect whether the pages overlapping the initrd
 181         * are in use, but more importantly, reserves the entire set of pages
 182         * as we don't want these pages allocated for other purposes.
 183         */
 184        start = round_down(phys_initrd_start, PAGE_SIZE);
 185        size = phys_initrd_size + (phys_initrd_start - start);
 186        size = round_up(size, PAGE_SIZE);
 187
 188        if (!memblock_is_region_memory(start, size)) {
 189                pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region - disabling initrd\n",
 190                       (u64)start, size);
 191                return;
 192        }
 193
 194        if (memblock_is_region_reserved(start, size)) {
 195                pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region - disabling initrd\n",
 196                       (u64)start, size);
 197                return;
 198        }
 199
 200        memblock_reserve(start, size);
 201
 202        /* Now convert initrd to virtual addresses */
 203        initrd_start = __phys_to_virt(phys_initrd_start);
 204        initrd_end = initrd_start + phys_initrd_size;
 205#endif
 206}
 207
 208#ifdef CONFIG_CPU_ICACHE_MISMATCH_WORKAROUND
 209void check_cpu_icache_size(int cpuid)
 210{
 211        u32 size, ctr;
 212
 213        asm("mrc p15, 0, %0, c0, c0, 1" : "=r" (ctr));
 214
 215        size = 1 << ((ctr & 0xf) + 2);
 216        if (cpuid != 0 && icache_size != size)
 217                pr_info("CPU%u: detected I-Cache line size mismatch, workaround enabled\n",
 218                        cpuid);
 219        if (icache_size > size)
 220                icache_size = size;
 221}
 222#endif
 223
 224void __init arm_memblock_init(const struct machine_desc *mdesc)
 225{
 226        /* Register the kernel text, kernel data and initrd with memblock. */
 227        memblock_reserve(__pa(KERNEL_START), KERNEL_END - KERNEL_START);
 228
 229        arm_initrd_init();
 230
 231        arm_mm_memblock_reserve();
 232
 233        /* reserve any platform specific memblock areas */
 234        if (mdesc->reserve)
 235                mdesc->reserve();
 236
 237        early_init_fdt_scan_reserved_mem();
 238
 239        /* reserve memory for DMA contiguous allocations */
 240        dma_contiguous_reserve(arm_dma_limit);
 241
 242        arm_memblock_steal_permitted = false;
 243        memblock_dump_all();
 244}
 245
 246void __init bootmem_init(void)
 247{
 248        memblock_allow_resize();
 249
 250        find_limits(&min_low_pfn, &max_low_pfn, &max_pfn);
 251
 252        early_memtest((phys_addr_t)min_low_pfn << PAGE_SHIFT,
 253                      (phys_addr_t)max_low_pfn << PAGE_SHIFT);
 254
 255        /*
 256         * sparse_init() tries to allocate memory from memblock, so must be
 257         * done after the fixed reservations
 258         */
 259        sparse_init();
 260
 261        /*
 262         * Now free the memory - free_area_init needs
 263         * the sparse mem_map arrays initialized by sparse_init()
 264         * for memmap_init_zone(), otherwise all PFNs are invalid.
 265         */
 266        zone_sizes_init(min_low_pfn, max_low_pfn, max_pfn);
 267}
 268
 269/*
 270 * Poison init memory with an undefined instruction (ARM) or a branch to an
 271 * undefined instruction (Thumb).
 272 */
 273static inline void poison_init_mem(void *s, size_t count)
 274{
 275        u32 *p = (u32 *)s;
 276        for (; count != 0; count -= 4)
 277                *p++ = 0xe7fddef0;
 278}
 279
 280static void __init free_highpages(void)
 281{
 282#ifdef CONFIG_HIGHMEM
 283        unsigned long max_low = max_low_pfn;
 284        phys_addr_t range_start, range_end;
 285        u64 i;
 286
 287        /* set highmem page free */
 288        for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
 289                                &range_start, &range_end, NULL) {
 290                unsigned long start = PFN_UP(range_start);
 291                unsigned long end = PFN_DOWN(range_end);
 292
 293                /* Ignore complete lowmem entries */
 294                if (end <= max_low)
 295                        continue;
 296
 297                /* Truncate partial highmem entries */
 298                if (start < max_low)
 299                        start = max_low;
 300
 301                for (; start < end; start++)
 302                        free_highmem_page(pfn_to_page(start));
 303        }
 304#endif
 305}
 306
 307/*
 308 * mem_init() marks the free areas in the mem_map and tells us how much
 309 * memory is free.  This is done after various parts of the system have
 310 * claimed their memory after the kernel image.
 311 */
 312void __init mem_init(void)
 313{
 314#ifdef CONFIG_ARM_LPAE
 315        if (swiotlb_force == SWIOTLB_FORCE ||
 316            max_pfn > arm_dma_pfn_limit)
 317                swiotlb_init(1);
 318        else
 319                swiotlb_force = SWIOTLB_NO_FORCE;
 320#endif
 321
 322        set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
 323
 324        /* this will put all unused low memory onto the freelists */
 325        memblock_free_all();
 326
 327#ifdef CONFIG_SA1111
 328        /* now that our DMA memory is actually so designated, we can free it */
 329        free_reserved_area(__va(PHYS_OFFSET), swapper_pg_dir, -1, NULL);
 330#endif
 331
 332        free_highpages();
 333
 334        /*
 335         * Check boundaries twice: Some fundamental inconsistencies can
 336         * be detected at build time already.
 337         */
 338#ifdef CONFIG_MMU
 339        BUILD_BUG_ON(TASK_SIZE                          > MODULES_VADDR);
 340        BUG_ON(TASK_SIZE                                > MODULES_VADDR);
 341#endif
 342
 343#ifdef CONFIG_HIGHMEM
 344        BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
 345        BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE      > PAGE_OFFSET);
 346#endif
 347}
 348
 349#ifdef CONFIG_STRICT_KERNEL_RWX
 350struct section_perm {
 351        const char *name;
 352        unsigned long start;
 353        unsigned long end;
 354        pmdval_t mask;
 355        pmdval_t prot;
 356        pmdval_t clear;
 357};
 358
 359/* First section-aligned location at or after __start_rodata. */
 360extern char __start_rodata_section_aligned[];
 361
 362static struct section_perm nx_perms[] = {
 363        /* Make pages tables, etc before _stext RW (set NX). */
 364        {
 365                .name   = "pre-text NX",
 366                .start  = PAGE_OFFSET,
 367                .end    = (unsigned long)_stext,
 368                .mask   = ~PMD_SECT_XN,
 369                .prot   = PMD_SECT_XN,
 370        },
 371        /* Make init RW (set NX). */
 372        {
 373                .name   = "init NX",
 374                .start  = (unsigned long)__init_begin,
 375                .end    = (unsigned long)_sdata,
 376                .mask   = ~PMD_SECT_XN,
 377                .prot   = PMD_SECT_XN,
 378        },
 379        /* Make rodata NX (set RO in ro_perms below). */
 380        {
 381                .name   = "rodata NX",
 382                .start  = (unsigned long)__start_rodata_section_aligned,
 383                .end    = (unsigned long)__init_begin,
 384                .mask   = ~PMD_SECT_XN,
 385                .prot   = PMD_SECT_XN,
 386        },
 387};
 388
 389static struct section_perm ro_perms[] = {
 390        /* Make kernel code and rodata RX (set RO). */
 391        {
 392                .name   = "text/rodata RO",
 393                .start  = (unsigned long)_stext,
 394                .end    = (unsigned long)__init_begin,
 395#ifdef CONFIG_ARM_LPAE
 396                .mask   = ~(L_PMD_SECT_RDONLY | PMD_SECT_AP2),
 397                .prot   = L_PMD_SECT_RDONLY | PMD_SECT_AP2,
 398#else
 399                .mask   = ~(PMD_SECT_APX | PMD_SECT_AP_WRITE),
 400                .prot   = PMD_SECT_APX | PMD_SECT_AP_WRITE,
 401                .clear  = PMD_SECT_AP_WRITE,
 402#endif
 403        },
 404};
 405
 406/*
 407 * Updates section permissions only for the current mm (sections are
 408 * copied into each mm). During startup, this is the init_mm. Is only
 409 * safe to be called with preemption disabled, as under stop_machine().
 410 */
 411static inline void section_update(unsigned long addr, pmdval_t mask,
 412                                  pmdval_t prot, struct mm_struct *mm)
 413{
 414        pmd_t *pmd;
 415
 416        pmd = pmd_offset(pud_offset(p4d_offset(pgd_offset(mm, addr), addr), addr), addr);
 417
 418#ifdef CONFIG_ARM_LPAE
 419        pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
 420#else
 421        if (addr & SECTION_SIZE)
 422                pmd[1] = __pmd((pmd_val(pmd[1]) & mask) | prot);
 423        else
 424                pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
 425#endif
 426        flush_pmd_entry(pmd);
 427        local_flush_tlb_kernel_range(addr, addr + SECTION_SIZE);
 428}
 429
 430/* Make sure extended page tables are in use. */
 431static inline bool arch_has_strict_perms(void)
 432{
 433        if (cpu_architecture() < CPU_ARCH_ARMv6)
 434                return false;
 435
 436        return !!(get_cr() & CR_XP);
 437}
 438
 439static void set_section_perms(struct section_perm *perms, int n, bool set,
 440                              struct mm_struct *mm)
 441{
 442        size_t i;
 443        unsigned long addr;
 444
 445        if (!arch_has_strict_perms())
 446                return;
 447
 448        for (i = 0; i < n; i++) {
 449                if (!IS_ALIGNED(perms[i].start, SECTION_SIZE) ||
 450                    !IS_ALIGNED(perms[i].end, SECTION_SIZE)) {
 451                        pr_err("BUG: %s section %lx-%lx not aligned to %lx\n",
 452                                perms[i].name, perms[i].start, perms[i].end,
 453                                SECTION_SIZE);
 454                        continue;
 455                }
 456
 457                for (addr = perms[i].start;
 458                     addr < perms[i].end;
 459                     addr += SECTION_SIZE)
 460                        section_update(addr, perms[i].mask,
 461                                set ? perms[i].prot : perms[i].clear, mm);
 462        }
 463
 464}
 465
 466/**
 467 * update_sections_early intended to be called only through stop_machine
 468 * framework and executed by only one CPU while all other CPUs will spin and
 469 * wait, so no locking is required in this function.
 470 */
 471static void update_sections_early(struct section_perm perms[], int n)
 472{
 473        struct task_struct *t, *s;
 474
 475        for_each_process(t) {
 476                if (t->flags & PF_KTHREAD)
 477                        continue;
 478                for_each_thread(t, s)
 479                        if (s->mm)
 480                                set_section_perms(perms, n, true, s->mm);
 481        }
 482        set_section_perms(perms, n, true, current->active_mm);
 483        set_section_perms(perms, n, true, &init_mm);
 484}
 485
 486static int __fix_kernmem_perms(void *unused)
 487{
 488        update_sections_early(nx_perms, ARRAY_SIZE(nx_perms));
 489        return 0;
 490}
 491
 492static void fix_kernmem_perms(void)
 493{
 494        stop_machine(__fix_kernmem_perms, NULL, NULL);
 495}
 496
 497static int __mark_rodata_ro(void *unused)
 498{
 499        update_sections_early(ro_perms, ARRAY_SIZE(ro_perms));
 500        return 0;
 501}
 502
 503void mark_rodata_ro(void)
 504{
 505        stop_machine(__mark_rodata_ro, NULL, NULL);
 506        debug_checkwx();
 507}
 508
 509#else
 510static inline void fix_kernmem_perms(void) { }
 511#endif /* CONFIG_STRICT_KERNEL_RWX */
 512
 513void free_initmem(void)
 514{
 515        fix_kernmem_perms();
 516
 517        poison_init_mem(__init_begin, __init_end - __init_begin);
 518        if (!machine_is_integrator() && !machine_is_cintegrator())
 519                free_initmem_default(-1);
 520}
 521
 522#ifdef CONFIG_BLK_DEV_INITRD
 523void free_initrd_mem(unsigned long start, unsigned long end)
 524{
 525        if (start == initrd_start)
 526                start = round_down(start, PAGE_SIZE);
 527        if (end == initrd_end)
 528                end = round_up(end, PAGE_SIZE);
 529
 530        poison_init_mem((void *)start, PAGE_ALIGN(end) - start);
 531        free_reserved_area((void *)start, (void *)end, -1, "initrd");
 532}
 533#endif
 534
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.