linux/arch/x86/boot/compressed/kaslr.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * kaslr.c
   4 *
   5 * This contains the routines needed to generate a reasonable level of
   6 * entropy to choose a randomized kernel base address offset in support
   7 * of Kernel Address Space Layout Randomization (KASLR). Additionally
   8 * handles walking the physical memory maps (and tracking memory regions
   9 * to avoid) in order to select a physical memory location that can
  10 * contain the entire properly aligned running kernel image.
  11 *
  12 */
  13
  14/*
  15 * isspace() in linux/ctype.h is expected by next_args() to filter
  16 * out "space/lf/tab". While boot/ctype.h conflicts with linux/ctype.h,
  17 * since isdigit() is implemented in both of them. Hence disable it
  18 * here.
  19 */
  20#define BOOT_CTYPE_H
  21
  22/*
  23 * _ctype[] in lib/ctype.c is needed by isspace() of linux/ctype.h.
  24 * While both lib/ctype.c and lib/cmdline.c will bring EXPORT_SYMBOL
  25 * which is meaningless and will cause compiling error in some cases.
  26 * So do not include linux/export.h and define EXPORT_SYMBOL(sym)
  27 * as empty.
  28 */
  29#define _LINUX_EXPORT_H
  30#define EXPORT_SYMBOL(sym)
  31
  32#include "misc.h"
  33#include "error.h"
  34#include "../string.h"
  35
  36#include <generated/compile.h>
  37#include <linux/module.h>
  38#include <linux/uts.h>
  39#include <linux/utsname.h>
  40#include <linux/ctype.h>
  41#include <linux/efi.h>
  42#include <generated/utsrelease.h>
  43#include <asm/efi.h>
  44
  45/* Macros used by the included decompressor code below. */
  46#define STATIC
  47#include <linux/decompress/mm.h>
  48
  49#ifdef CONFIG_X86_5LEVEL
  50unsigned int pgtable_l5_enabled __ro_after_init;
  51unsigned int pgdir_shift __ro_after_init = 39;
  52unsigned int ptrs_per_p4d __ro_after_init = 1;
  53#endif
  54
  55extern unsigned long get_cmd_line_ptr(void);
  56
  57/* Used by PAGE_KERN* macros: */
  58pteval_t __default_kernel_pte_mask __read_mostly = ~0;
  59
  60/* Simplified build-specific string for starting entropy. */
  61static const char build_str[] = UTS_RELEASE " (" LINUX_COMPILE_BY "@"
  62                LINUX_COMPILE_HOST ") (" LINUX_COMPILER ") " UTS_VERSION;
  63
  64static unsigned long rotate_xor(unsigned long hash, const void *area,
  65                                size_t size)
  66{
  67        size_t i;
  68        unsigned long *ptr = (unsigned long *)area;
  69
  70        for (i = 0; i < size / sizeof(hash); i++) {
  71                /* Rotate by odd number of bits and XOR. */
  72                hash = (hash << ((sizeof(hash) * 8) - 7)) | (hash >> 7);
  73                hash ^= ptr[i];
  74        }
  75
  76        return hash;
  77}
  78
  79/* Attempt to create a simple but unpredictable starting entropy. */
  80static unsigned long get_boot_seed(void)
  81{
  82        unsigned long hash = 0;
  83
  84        hash = rotate_xor(hash, build_str, sizeof(build_str));
  85        hash = rotate_xor(hash, boot_params, sizeof(*boot_params));
  86
  87        return hash;
  88}
  89
  90#define KASLR_COMPRESSED_BOOT
  91#include "../../lib/kaslr.c"
  92
  93struct mem_vector {
  94        unsigned long long start;
  95        unsigned long long size;
  96};
  97
  98/* Only supporting at most 4 unusable memmap regions with kaslr */
  99#define MAX_MEMMAP_REGIONS      4
 100
 101static bool memmap_too_large;
 102
 103
 104/* Store memory limit specified by "mem=nn[KMG]" or "memmap=nn[KMG]" */
 105unsigned long long mem_limit = ULLONG_MAX;
 106
 107
 108enum mem_avoid_index {
 109        MEM_AVOID_ZO_RANGE = 0,
 110        MEM_AVOID_INITRD,
 111        MEM_AVOID_CMDLINE,
 112        MEM_AVOID_BOOTPARAMS,
 113        MEM_AVOID_MEMMAP_BEGIN,
 114        MEM_AVOID_MEMMAP_END = MEM_AVOID_MEMMAP_BEGIN + MAX_MEMMAP_REGIONS - 1,
 115        MEM_AVOID_MAX,
 116};
 117
 118static struct mem_vector mem_avoid[MEM_AVOID_MAX];
 119
 120static bool mem_overlaps(struct mem_vector *one, struct mem_vector *two)
 121{
 122        /* Item one is entirely before item two. */
 123        if (one->start + one->size <= two->start)
 124                return false;
 125        /* Item one is entirely after item two. */
 126        if (one->start >= two->start + two->size)
 127                return false;
 128        return true;
 129}
 130
 131char *skip_spaces(const char *str)
 132{
 133        while (isspace(*str))
 134                ++str;
 135        return (char *)str;
 136}
 137#include "../../../../lib/ctype.c"
 138#include "../../../../lib/cmdline.c"
 139
 140static int
 141parse_memmap(char *p, unsigned long long *start, unsigned long long *size)
 142{
 143        char *oldp;
 144
 145        if (!p)
 146                return -EINVAL;
 147
 148        /* We don't care about this option here */
 149        if (!strncmp(p, "exactmap", 8))
 150                return -EINVAL;
 151
 152        oldp = p;
 153        *size = memparse(p, &p);
 154        if (p == oldp)
 155                return -EINVAL;
 156
 157        switch (*p) {
 158        case '#':
 159        case '$':
 160        case '!':
 161                *start = memparse(p + 1, &p);
 162                return 0;
 163        case '@':
 164                /* memmap=nn@ss specifies usable region, should be skipped */
 165                *size = 0;
 166                /* Fall through */
 167        default:
 168                /*
 169                 * If w/o offset, only size specified, memmap=nn[KMG] has the
 170                 * same behaviour as mem=nn[KMG]. It limits the max address
 171                 * system can use. Region above the limit should be avoided.
 172                 */
 173                *start = 0;
 174                return 0;
 175        }
 176
 177        return -EINVAL;
 178}
 179
 180static void mem_avoid_memmap(char *str)
 181{
 182        static int i;
 183
 184        if (i >= MAX_MEMMAP_REGIONS)
 185                return;
 186
 187        while (str && (i < MAX_MEMMAP_REGIONS)) {
 188                int rc;
 189                unsigned long long start, size;
 190                char *k = strchr(str, ',');
 191
 192                if (k)
 193                        *k++ = 0;
 194
 195                rc = parse_memmap(str, &start, &size);
 196                if (rc < 0)
 197                        break;
 198                str = k;
 199
 200                if (start == 0) {
 201                        /* Store the specified memory limit if size > 0 */
 202                        if (size > 0)
 203                                mem_limit = size;
 204
 205                        continue;
 206                }
 207
 208                mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].start = start;
 209                mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].size = size;
 210                i++;
 211        }
 212
 213        /* More than 4 memmaps, fail kaslr */
 214        if ((i >= MAX_MEMMAP_REGIONS) && str)
 215                memmap_too_large = true;
 216}
 217
 218static int handle_mem_memmap(void)
 219{
 220        char *args = (char *)get_cmd_line_ptr();
 221        size_t len = strlen((char *)args);
 222        char *tmp_cmdline;
 223        char *param, *val;
 224        u64 mem_size;
 225
 226        if (!strstr(args, "memmap=") && !strstr(args, "mem="))
 227                return 0;
 228
 229        tmp_cmdline = malloc(len + 1);
 230        if (!tmp_cmdline)
 231                error("Failed to allocate space for tmp_cmdline");
 232
 233        memcpy(tmp_cmdline, args, len);
 234        tmp_cmdline[len] = 0;
 235        args = tmp_cmdline;
 236
 237        /* Chew leading spaces */
 238        args = skip_spaces(args);
 239
 240        while (*args) {
 241                args = next_arg(args, &param, &val);
 242                /* Stop at -- */
 243                if (!val && strcmp(param, "--") == 0) {
 244                        warn("Only '--' specified in cmdline");
 245                        free(tmp_cmdline);
 246                        return -1;
 247                }
 248
 249                if (!strcmp(param, "memmap")) {
 250                        mem_avoid_memmap(val);
 251                } else if (!strcmp(param, "mem")) {
 252                        char *p = val;
 253
 254                        if (!strcmp(p, "nopentium"))
 255                                continue;
 256                        mem_size = memparse(p, &p);
 257                        if (mem_size == 0) {
 258                                free(tmp_cmdline);
 259                                return -EINVAL;
 260                        }
 261                        mem_limit = mem_size;
 262                }
 263        }
 264
 265        free(tmp_cmdline);
 266        return 0;
 267}
 268
 269/*
 270 * In theory, KASLR can put the kernel anywhere in the range of [16M, 64T).
 271 * The mem_avoid array is used to store the ranges that need to be avoided
 272 * when KASLR searches for an appropriate random address. We must avoid any
 273 * regions that are unsafe to overlap with during decompression, and other
 274 * things like the initrd, cmdline and boot_params. This comment seeks to
 275 * explain mem_avoid as clearly as possible since incorrect mem_avoid
 276 * memory ranges lead to really hard to debug boot failures.
 277 *
 278 * The initrd, cmdline, and boot_params are trivial to identify for
 279 * avoiding. They are MEM_AVOID_INITRD, MEM_AVOID_CMDLINE, and
 280 * MEM_AVOID_BOOTPARAMS respectively below.
 281 *
 282 * What is not obvious how to avoid is the range of memory that is used
 283 * during decompression (MEM_AVOID_ZO_RANGE below). This range must cover
 284 * the compressed kernel (ZO) and its run space, which is used to extract
 285 * the uncompressed kernel (VO) and relocs.
 286 *
 287 * ZO's full run size sits against the end of the decompression buffer, so
 288 * we can calculate where text, data, bss, etc of ZO are positioned more
 289 * easily.
 290 *
 291 * For additional background, the decompression calculations can be found
 292 * in header.S, and the memory diagram is based on the one found in misc.c.
 293 *
 294 * The following conditions are already enforced by the image layouts and
 295 * associated code:
 296 *  - input + input_size >= output + output_size
 297 *  - kernel_total_size <= init_size
 298 *  - kernel_total_size <= output_size (see Note below)
 299 *  - output + init_size >= output + output_size
 300 *
 301 * (Note that kernel_total_size and output_size have no fundamental
 302 * relationship, but output_size is passed to choose_random_location
 303 * as a maximum of the two. The diagram is showing a case where
 304 * kernel_total_size is larger than output_size, but this case is
 305 * handled by bumping output_size.)
 306 *
 307 * The above conditions can be illustrated by a diagram:
 308 *
 309 * 0   output            input            input+input_size    output+init_size
 310 * |     |                 |                             |             |
 311 * |     |                 |                             |             |
 312 * |-----|--------|--------|--------------|-----------|--|-------------|
 313 *                |                       |           |
 314 *                |                       |           |
 315 * output+init_size-ZO_INIT_SIZE  output+output_size  output+kernel_total_size
 316 *
 317 * [output, output+init_size) is the entire memory range used for
 318 * extracting the compressed image.
 319 *
 320 * [output, output+kernel_total_size) is the range needed for the
 321 * uncompressed kernel (VO) and its run size (bss, brk, etc).
 322 *
 323 * [output, output+output_size) is VO plus relocs (i.e. the entire
 324 * uncompressed payload contained by ZO). This is the area of the buffer
 325 * written to during decompression.
 326 *
 327 * [output+init_size-ZO_INIT_SIZE, output+init_size) is the worst-case
 328 * range of the copied ZO and decompression code. (i.e. the range
 329 * covered backwards of size ZO_INIT_SIZE, starting from output+init_size.)
 330 *
 331 * [input, input+input_size) is the original copied compressed image (ZO)
 332 * (i.e. it does not include its run size). This range must be avoided
 333 * because it contains the data used for decompression.
 334 *
 335 * [input+input_size, output+init_size) is [_text, _end) for ZO. This
 336 * range includes ZO's heap and stack, and must be avoided since it
 337 * performs the decompression.
 338 *
 339 * Since the above two ranges need to be avoided and they are adjacent,
 340 * they can be merged, resulting in: [input, output+init_size) which
 341 * becomes the MEM_AVOID_ZO_RANGE below.
 342 */
 343static void mem_avoid_init(unsigned long input, unsigned long input_size,
 344                           unsigned long output)
 345{
 346        unsigned long init_size = boot_params->hdr.init_size;
 347        u64 initrd_start, initrd_size;
 348        u64 cmd_line, cmd_line_size;
 349        char *ptr;
 350
 351        /*
 352         * Avoid the region that is unsafe to overlap during
 353         * decompression.
 354         */
 355        mem_avoid[MEM_AVOID_ZO_RANGE].start = input;
 356        mem_avoid[MEM_AVOID_ZO_RANGE].size = (output + init_size) - input;
 357        add_identity_map(mem_avoid[MEM_AVOID_ZO_RANGE].start,
 358                         mem_avoid[MEM_AVOID_ZO_RANGE].size);
 359
 360        /* Avoid initrd. */
 361        initrd_start  = (u64)boot_params->ext_ramdisk_image << 32;
 362        initrd_start |= boot_params->hdr.ramdisk_image;
 363        initrd_size  = (u64)boot_params->ext_ramdisk_size << 32;
 364        initrd_size |= boot_params->hdr.ramdisk_size;
 365        mem_avoid[MEM_AVOID_INITRD].start = initrd_start;
 366        mem_avoid[MEM_AVOID_INITRD].size = initrd_size;
 367        /* No need to set mapping for initrd, it will be handled in VO. */
 368
 369        /* Avoid kernel command line. */
 370        cmd_line  = (u64)boot_params->ext_cmd_line_ptr << 32;
 371        cmd_line |= boot_params->hdr.cmd_line_ptr;
 372        /* Calculate size of cmd_line. */
 373        ptr = (char *)(unsigned long)cmd_line;
 374        for (cmd_line_size = 0; ptr[cmd_line_size++];)
 375                ;
 376        mem_avoid[MEM_AVOID_CMDLINE].start = cmd_line;
 377        mem_avoid[MEM_AVOID_CMDLINE].size = cmd_line_size;
 378        add_identity_map(mem_avoid[MEM_AVOID_CMDLINE].start,
 379                         mem_avoid[MEM_AVOID_CMDLINE].size);
 380
 381        /* Avoid boot parameters. */
 382        mem_avoid[MEM_AVOID_BOOTPARAMS].start = (unsigned long)boot_params;
 383        mem_avoid[MEM_AVOID_BOOTPARAMS].size = sizeof(*boot_params);
 384        add_identity_map(mem_avoid[MEM_AVOID_BOOTPARAMS].start,
 385                         mem_avoid[MEM_AVOID_BOOTPARAMS].size);
 386
 387        /* We don't need to set a mapping for setup_data. */
 388
 389        /* Mark the memmap regions we need to avoid */
 390        handle_mem_memmap();
 391
 392#ifdef CONFIG_X86_VERBOSE_BOOTUP
 393        /* Make sure video RAM can be used. */
 394        add_identity_map(0, PMD_SIZE);
 395#endif
 396}
 397
 398/*
 399 * Does this memory vector overlap a known avoided area? If so, record the
 400 * overlap region with the lowest address.
 401 */
 402static bool mem_avoid_overlap(struct mem_vector *img,
 403                              struct mem_vector *overlap)
 404{
 405        int i;
 406        struct setup_data *ptr;
 407        unsigned long earliest = img->start + img->size;
 408        bool is_overlapping = false;
 409
 410        for (i = 0; i < MEM_AVOID_MAX; i++) {
 411                if (mem_overlaps(img, &mem_avoid[i]) &&
 412                    mem_avoid[i].start < earliest) {
 413                        *overlap = mem_avoid[i];
 414                        earliest = overlap->start;
 415                        is_overlapping = true;
 416                }
 417        }
 418
 419        /* Avoid all entries in the setup_data linked list. */
 420        ptr = (struct setup_data *)(unsigned long)boot_params->hdr.setup_data;
 421        while (ptr) {
 422                struct mem_vector avoid;
 423
 424                avoid.start = (unsigned long)ptr;
 425                avoid.size = sizeof(*ptr) + ptr->len;
 426
 427                if (mem_overlaps(img, &avoid) && (avoid.start < earliest)) {
 428                        *overlap = avoid;
 429                        earliest = overlap->start;
 430                        is_overlapping = true;
 431                }
 432
 433                ptr = (struct setup_data *)(unsigned long)ptr->next;
 434        }
 435
 436        return is_overlapping;
 437}
 438
 439struct slot_area {
 440        unsigned long addr;
 441        int num;
 442};
 443
 444#define MAX_SLOT_AREA 100
 445
 446static struct slot_area slot_areas[MAX_SLOT_AREA];
 447
 448static unsigned long slot_max;
 449
 450static unsigned long slot_area_index;
 451
 452static void store_slot_info(struct mem_vector *region, unsigned long image_size)
 453{
 454        struct slot_area slot_area;
 455
 456        if (slot_area_index == MAX_SLOT_AREA)
 457                return;
 458
 459        slot_area.addr = region->start;
 460        slot_area.num = (region->size - image_size) /
 461                        CONFIG_PHYSICAL_ALIGN + 1;
 462
 463        if (slot_area.num > 0) {
 464                slot_areas[slot_area_index++] = slot_area;
 465                slot_max += slot_area.num;
 466        }
 467}
 468
 469static unsigned long slots_fetch_random(void)
 470{
 471        unsigned long slot;
 472        int i;
 473
 474        /* Handle case of no slots stored. */
 475        if (slot_max == 0)
 476                return 0;
 477
 478        slot = kaslr_get_random_long("Physical") % slot_max;
 479
 480        for (i = 0; i < slot_area_index; i++) {
 481                if (slot >= slot_areas[i].num) {
 482                        slot -= slot_areas[i].num;
 483                        continue;
 484                }
 485                return slot_areas[i].addr + slot * CONFIG_PHYSICAL_ALIGN;
 486        }
 487
 488        if (i == slot_area_index)
 489                debug_putstr("slots_fetch_random() failed!?\n");
 490        return 0;
 491}
 492
 493static void process_mem_region(struct mem_vector *entry,
 494                               unsigned long minimum,
 495                               unsigned long image_size)
 496{
 497        struct mem_vector region, overlap;
 498        struct slot_area slot_area;
 499        unsigned long start_orig, end;
 500        struct mem_vector cur_entry;
 501
 502        /* On 32-bit, ignore entries entirely above our maximum. */
 503        if (IS_ENABLED(CONFIG_X86_32) && entry->start >= KERNEL_IMAGE_SIZE)
 504                return;
 505
 506        /* Ignore entries entirely below our minimum. */
 507        if (entry->start + entry->size < minimum)
 508                return;
 509
 510        /* Ignore entries above memory limit */
 511        end = min(entry->size + entry->start, mem_limit);
 512        if (entry->start >= end)
 513                return;
 514        cur_entry.start = entry->start;
 515        cur_entry.size = end - entry->start;
 516
 517        region.start = cur_entry.start;
 518        region.size = cur_entry.size;
 519
 520        /* Give up if slot area array is full. */
 521        while (slot_area_index < MAX_SLOT_AREA) {
 522                start_orig = region.start;
 523
 524                /* Potentially raise address to minimum location. */
 525                if (region.start < minimum)
 526                        region.start = minimum;
 527
 528                /* Potentially raise address to meet alignment needs. */
 529                region.start = ALIGN(region.start, CONFIG_PHYSICAL_ALIGN);
 530
 531                /* Did we raise the address above the passed in memory entry? */
 532                if (region.start > cur_entry.start + cur_entry.size)
 533                        return;
 534
 535                /* Reduce size by any delta from the original address. */
 536                region.size -= region.start - start_orig;
 537
 538                /* On 32-bit, reduce region size to fit within max size. */
 539                if (IS_ENABLED(CONFIG_X86_32) &&
 540                    region.start + region.size > KERNEL_IMAGE_SIZE)
 541                        region.size = KERNEL_IMAGE_SIZE - region.start;
 542
 543                /* Return if region can't contain decompressed kernel */
 544                if (region.size < image_size)
 545                        return;
 546
 547                /* If nothing overlaps, store the region and return. */
 548                if (!mem_avoid_overlap(&region, &overlap)) {
 549                        store_slot_info(&region, image_size);
 550                        return;
 551                }
 552
 553                /* Store beginning of region if holds at least image_size. */
 554                if (overlap.start > region.start + image_size) {
 555                        struct mem_vector beginning;
 556
 557                        beginning.start = region.start;
 558                        beginning.size = overlap.start - region.start;
 559                        store_slot_info(&beginning, image_size);
 560                }
 561
 562                /* Return if overlap extends to or past end of region. */
 563                if (overlap.start + overlap.size >= region.start + region.size)
 564                        return;
 565
 566                /* Clip off the overlapping region and start over. */
 567                region.size -= overlap.start - region.start + overlap.size;
 568                region.start = overlap.start + overlap.size;
 569        }
 570}
 571
 572#ifdef CONFIG_EFI
 573/*
 574 * Returns true if mirror region found (and must have been processed
 575 * for slots adding)
 576 */
 577static bool
 578process_efi_entries(unsigned long minimum, unsigned long image_size)
 579{
 580        struct efi_info *e = &boot_params->efi_info;
 581        bool efi_mirror_found = false;
 582        struct mem_vector region;
 583        efi_memory_desc_t *md;
 584        unsigned long pmap;
 585        char *signature;
 586        u32 nr_desc;
 587        int i;
 588
 589        signature = (char *)&e->efi_loader_signature;
 590        if (strncmp(signature, EFI32_LOADER_SIGNATURE, 4) &&
 591            strncmp(signature, EFI64_LOADER_SIGNATURE, 4))
 592                return false;
 593
 594#ifdef CONFIG_X86_32
 595        /* Can't handle data above 4GB at this time */
 596        if (e->efi_memmap_hi) {
 597                warn("EFI memmap is above 4GB, can't be handled now on x86_32. EFI should be disabled.\n");
 598                return false;
 599        }
 600        pmap =  e->efi_memmap;
 601#else
 602        pmap = (e->efi_memmap | ((__u64)e->efi_memmap_hi << 32));
 603#endif
 604
 605        nr_desc = e->efi_memmap_size / e->efi_memdesc_size;
 606        for (i = 0; i < nr_desc; i++) {
 607                md = efi_early_memdesc_ptr(pmap, e->efi_memdesc_size, i);
 608                if (md->attribute & EFI_MEMORY_MORE_RELIABLE) {
 609                        efi_mirror_found = true;
 610                        break;
 611                }
 612        }
 613
 614        for (i = 0; i < nr_desc; i++) {
 615                md = efi_early_memdesc_ptr(pmap, e->efi_memdesc_size, i);
 616
 617                /*
 618                 * Here we are more conservative in picking free memory than
 619                 * the EFI spec allows:
 620                 *
 621                 * According to the spec, EFI_BOOT_SERVICES_{CODE|DATA} are also
 622                 * free memory and thus available to place the kernel image into,
 623                 * but in practice there's firmware where using that memory leads
 624                 * to crashes.
 625                 *
 626                 * Only EFI_CONVENTIONAL_MEMORY is guaranteed to be free.
 627                 */
 628                if (md->type != EFI_CONVENTIONAL_MEMORY)
 629                        continue;
 630
 631                if (efi_mirror_found &&
 632                    !(md->attribute & EFI_MEMORY_MORE_RELIABLE))
 633                        continue;
 634
 635                region.start = md->phys_addr;
 636                region.size = md->num_pages << EFI_PAGE_SHIFT;
 637                process_mem_region(&region, minimum, image_size);
 638                if (slot_area_index == MAX_SLOT_AREA) {
 639                        debug_putstr("Aborted EFI scan (slot_areas full)!\n");
 640                        break;
 641                }
 642        }
 643        return true;
 644}
 645#else
 646static inline bool
 647process_efi_entries(unsigned long minimum, unsigned long image_size)
 648{
 649        return false;
 650}
 651#endif
 652
 653static void process_e820_entries(unsigned long minimum,
 654                                 unsigned long image_size)
 655{
 656        int i;
 657        struct mem_vector region;
 658        struct boot_e820_entry *entry;
 659
 660        /* Verify potential e820 positions, appending to slots list. */
 661        for (i = 0; i < boot_params->e820_entries; i++) {
 662                entry = &boot_params->e820_table[i];
 663                /* Skip non-RAM entries. */
 664                if (entry->type != E820_TYPE_RAM)
 665                        continue;
 666                region.start = entry->addr;
 667                region.size = entry->size;
 668                process_mem_region(&region, minimum, image_size);
 669                if (slot_area_index == MAX_SLOT_AREA) {
 670                        debug_putstr("Aborted e820 scan (slot_areas full)!\n");
 671                        break;
 672                }
 673        }
 674}
 675
 676static unsigned long find_random_phys_addr(unsigned long minimum,
 677                                           unsigned long image_size)
 678{
 679        /* Check if we had too many memmaps. */
 680        if (memmap_too_large) {
 681                debug_putstr("Aborted memory entries scan (more than 4 memmap= args)!\n");
 682                return 0;
 683        }
 684
 685        /* Make sure minimum is aligned. */
 686        minimum = ALIGN(minimum, CONFIG_PHYSICAL_ALIGN);
 687
 688        if (process_efi_entries(minimum, image_size))
 689                return slots_fetch_random();
 690
 691        process_e820_entries(minimum, image_size);
 692        return slots_fetch_random();
 693}
 694
 695static unsigned long find_random_virt_addr(unsigned long minimum,
 696                                           unsigned long image_size)
 697{
 698        unsigned long slots, random_addr;
 699
 700        /* Make sure minimum is aligned. */
 701        minimum = ALIGN(minimum, CONFIG_PHYSICAL_ALIGN);
 702        /* Align image_size for easy slot calculations. */
 703        image_size = ALIGN(image_size, CONFIG_PHYSICAL_ALIGN);
 704
 705        /*
 706         * There are how many CONFIG_PHYSICAL_ALIGN-sized slots
 707         * that can hold image_size within the range of minimum to
 708         * KERNEL_IMAGE_SIZE?
 709         */
 710        slots = (KERNEL_IMAGE_SIZE - minimum - image_size) /
 711                 CONFIG_PHYSICAL_ALIGN + 1;
 712
 713        random_addr = kaslr_get_random_long("Virtual") % slots;
 714
 715        return random_addr * CONFIG_PHYSICAL_ALIGN + minimum;
 716}
 717
 718/*
 719 * Since this function examines addresses much more numerically,
 720 * it takes the input and output pointers as 'unsigned long'.
 721 */
 722void choose_random_location(unsigned long input,
 723                            unsigned long input_size,
 724                            unsigned long *output,
 725                            unsigned long output_size,
 726                            unsigned long *virt_addr)
 727{
 728        unsigned long random_addr, min_addr;
 729
 730        if (cmdline_find_option_bool("nokaslr")) {
 731                warn("KASLR disabled: 'nokaslr' on cmdline.");
 732                return;
 733        }
 734
 735#ifdef CONFIG_X86_5LEVEL
 736        if (__read_cr4() & X86_CR4_LA57) {
 737                pgtable_l5_enabled = 1;
 738                pgdir_shift = 48;
 739                ptrs_per_p4d = 512;
 740        }
 741#endif
 742
 743        boot_params->hdr.loadflags |= KASLR_FLAG;
 744
 745        /* Prepare to add new identity pagetables on demand. */
 746        initialize_identity_maps();
 747
 748        /* Record the various known unsafe memory ranges. */
 749        mem_avoid_init(input, input_size, *output);
 750
 751        /*
 752         * Low end of the randomization range should be the
 753         * smaller of 512M or the initial kernel image
 754         * location:
 755         */
 756        min_addr = min(*output, 512UL << 20);
 757
 758        /* Walk available memory entries to find a random address. */
 759        random_addr = find_random_phys_addr(min_addr, output_size);
 760        if (!random_addr) {
 761                warn("Physical KASLR disabled: no suitable memory region!");
 762        } else {
 763                /* Update the new physical address location. */
 764                if (*output != random_addr) {
 765                        add_identity_map(random_addr, output_size);
 766                        *output = random_addr;
 767                }
 768
 769                /*
 770                 * This loads the identity mapping page table.
 771                 * This should only be done if a new physical address
 772                 * is found for the kernel, otherwise we should keep
 773                 * the old page table to make it be like the "nokaslr"
 774                 * case.
 775                 */
 776                finalize_identity_maps();
 777        }
 778
 779
 780        /* Pick random virtual address starting from LOAD_PHYSICAL_ADDR. */
 781        if (IS_ENABLED(CONFIG_X86_64))
 782                random_addr = find_random_virt_addr(LOAD_PHYSICAL_ADDR, output_size);
 783        *virt_addr = random_addr;
 784}
 785