kaslr.c

  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#include "misc.h"
 23#include "error.h"
 24#include "../string.h"
 25#include "efi.h"
 26
 27#include <generated/compile.h>
 28#include <generated/utsversion.h>
 29#include <generated/utsrelease.h>
 30
 31#define _SETUP
 32#include <asm/setup.h>	/* For COMMAND_LINE_SIZE */
 33#undef _SETUP
 34
 35extern unsigned long get_cmd_line_ptr(void);
 36
 37/* Simplified build-specific string for starting entropy. */
 38static const char build_str[] = UTS_RELEASE " (" LINUX_COMPILE_BY "@"
 39		LINUX_COMPILE_HOST ") (" LINUX_COMPILER ") " UTS_VERSION;
 40
 41static unsigned long rotate_xor(unsigned long hash, const void *area,
 42				size_t size)
 43{
 44	size_t i;
 45	unsigned long *ptr = (unsigned long *)area;
 46
 47	for (i = 0; i < size / sizeof(hash); i++) {
 48		/* Rotate by odd number of bits and XOR. */
 49		hash = (hash << ((sizeof(hash) * 8) - 7)) | (hash >> 7);
 50		hash ^= ptr[i];
 51	}
 52
 53	return hash;
 54}
 55
 56/* Attempt to create a simple but unpredictable starting entropy. */
 57static unsigned long get_boot_seed(void)
 58{
 59	unsigned long hash = 0;
 60
 61	hash = rotate_xor(hash, build_str, sizeof(build_str));
 62	hash = rotate_xor(hash, boot_params_ptr, sizeof(*boot_params_ptr));
 63
 64	return hash;
 65}
 66
 67#define KASLR_COMPRESSED_BOOT
 68#include "../../lib/kaslr.c"
 69
 70
 71/* Only supporting at most 4 unusable memmap regions with kaslr */
 72#define MAX_MEMMAP_REGIONS	4
 73
 74static bool memmap_too_large;
 75
 76
 77/*
 78 * Store memory limit: MAXMEM on 64-bit and KERNEL_IMAGE_SIZE on 32-bit.
 79 * It may be reduced by "mem=nn[KMG]" or "memmap=nn[KMG]" command line options.
 80 */
 81static u64 mem_limit;
 82
 83/* Number of immovable memory regions */
 84static int num_immovable_mem;
 85
 86enum mem_avoid_index {
 87	MEM_AVOID_ZO_RANGE = 0,
 88	MEM_AVOID_INITRD,
 89	MEM_AVOID_CMDLINE,
 90	MEM_AVOID_BOOTPARAMS,
 91	MEM_AVOID_MEMMAP_BEGIN,
 92	MEM_AVOID_MEMMAP_END = MEM_AVOID_MEMMAP_BEGIN + MAX_MEMMAP_REGIONS - 1,
 93	MEM_AVOID_MAX,
 94};
 95
 96static struct mem_vector mem_avoid[MEM_AVOID_MAX];
 97
 98static bool mem_overlaps(struct mem_vector *one, struct mem_vector *two)
 99{
100	/* Item one is entirely before item two. */
101	if (one->start + one->size <= two->start)
102		return false;
103	/* Item one is entirely after item two. */
104	if (one->start >= two->start + two->size)
105		return false;
106	return true;
107}
108
109char *skip_spaces(const char *str)
110{
111	while (isspace(*str))
112		++str;
113	return (char *)str;
114}
115#include "../../../../lib/ctype.c"
116#include "../../../../lib/cmdline.c"
117
118static int
119parse_memmap(char *p, u64 *start, u64 *size)
120{
121	char *oldp;
122
123	if (!p)
124		return -EINVAL;
125
126	/* We don't care about this option here */
127	if (!strncmp(p, "exactmap", 8))
128		return -EINVAL;
129
130	oldp = p;
131	*size = memparse(p, &p);
132	if (p == oldp)
133		return -EINVAL;
134
135	switch (*p) {
136	case '#':
137	case '$':
138	case '!':
139		*start = memparse(p + 1, &p);
140		return 0;
141	case '@':
142		/*
143		 * memmap=nn@ss specifies usable region, should
144		 * be skipped
145		 */
146		*size = 0;
147		fallthrough;
148	default:
149		/*
150		 * If w/o offset, only size specified, memmap=nn[KMG] has the
151		 * same behaviour as mem=nn[KMG]. It limits the max address
152		 * system can use. Region above the limit should be avoided.
153		 */
154		*start = 0;
155		return 0;
156	}
157
158	return -EINVAL;
159}
160
161static void mem_avoid_memmap(char *str)
162{
163	static int i;
164
165	if (i >= MAX_MEMMAP_REGIONS)
166		return;
167
168	while (str && (i < MAX_MEMMAP_REGIONS)) {
169		int rc;
170		u64 start, size;
171		char *k = strchr(str, ',');
172
173		if (k)
174			*k++ = 0;
175
176		rc = parse_memmap(str, &start, &size);
177		if (rc < 0)
178			break;
179		str = k;
180
181		if (start == 0) {
182			/* Store the specified memory limit if size > 0 */
183			if (size > 0 && size < mem_limit)
184				mem_limit = size;
185
186			continue;
187		}
188
189		mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].start = start;
190		mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].size = size;
191		i++;
192	}
193
194	/* More than 4 memmaps, fail kaslr */
195	if ((i >= MAX_MEMMAP_REGIONS) && str)
196		memmap_too_large = true;
197}
198
199/* Store the number of 1GB huge pages which users specified: */
200static unsigned long max_gb_huge_pages;
201
202static void parse_gb_huge_pages(char *param, char *val)
203{
204	static bool gbpage_sz;
205	char *p;
206
207	if (!strcmp(param, "hugepagesz")) {
208		p = val;
209		if (memparse(p, &p) != PUD_SIZE) {
210			gbpage_sz = false;
211			return;
212		}
213
214		if (gbpage_sz)
215			warn("Repeatedly set hugeTLB page size of 1G!\n");
216		gbpage_sz = true;
217		return;
218	}
219
220	if (!strcmp(param, "hugepages") && gbpage_sz) {
221		p = val;
222		max_gb_huge_pages = simple_strtoull(p, &p, 0);
223		return;
224	}
225}
226
227static void handle_mem_options(void)
228{
229	char *args = (char *)get_cmd_line_ptr();
230	size_t len;
231	char *tmp_cmdline;
232	char *param, *val;
233	u64 mem_size;
234
235	if (!args)
236		return;
237
238	len = strnlen(args, COMMAND_LINE_SIZE-1);
239	tmp_cmdline = malloc(len + 1);
240	if (!tmp_cmdline)
241		error("Failed to allocate space for tmp_cmdline");
242
243	memcpy(tmp_cmdline, args, len);
244	tmp_cmdline[len] = 0;
245	args = tmp_cmdline;
246
247	/* Chew leading spaces */
248	args = skip_spaces(args);
249
250	while (*args) {
251		args = next_arg(args, &param, &val);
252		/* Stop at -- */
253		if (!val && strcmp(param, "--") == 0)
254			break;
255
256		if (!strcmp(param, "memmap")) {
257			mem_avoid_memmap(val);
258		} else if (IS_ENABLED(CONFIG_X86_64) && strstr(param, "hugepages")) {
259			parse_gb_huge_pages(param, val);
260		} else if (!strcmp(param, "mem")) {
261			char *p = val;
262
263			if (!strcmp(p, "nopentium"))
264				continue;
265			mem_size = memparse(p, &p);
266			if (mem_size == 0)
267				break;
268
269			if (mem_size < mem_limit)
270				mem_limit = mem_size;
271		}
272	}
273
274	free(tmp_cmdline);
275	return;
276}
277
278/*
279 * In theory, KASLR can put the kernel anywhere in the range of [16M, MAXMEM)
280 * on 64-bit, and [16M, KERNEL_IMAGE_SIZE) on 32-bit.
281 *
282 * The mem_avoid array is used to store the ranges that need to be avoided
283 * when KASLR searches for an appropriate random address. We must avoid any
284 * regions that are unsafe to overlap with during decompression, and other
285 * things like the initrd, cmdline and boot_params. This comment seeks to
286 * explain mem_avoid as clearly as possible since incorrect mem_avoid
287 * memory ranges lead to really hard to debug boot failures.
288 *
289 * The initrd, cmdline, and boot_params are trivial to identify for
290 * avoiding. They are MEM_AVOID_INITRD, MEM_AVOID_CMDLINE, and
291 * MEM_AVOID_BOOTPARAMS respectively below.
292 *
293 * What is not obvious how to avoid is the range of memory that is used
294 * during decompression (MEM_AVOID_ZO_RANGE below). This range must cover
295 * the compressed kernel (ZO) and its run space, which is used to extract
296 * the uncompressed kernel (VO) and relocs.
297 *
298 * ZO's full run size sits against the end of the decompression buffer, so
299 * we can calculate where text, data, bss, etc of ZO are positioned more
300 * easily.
301 *
302 * For additional background, the decompression calculations can be found
303 * in header.S, and the memory diagram is based on the one found in misc.c.
304 *
305 * The following conditions are already enforced by the image layouts and
306 * associated code:
307 *  - input + input_size >= output + output_size
308 *  - kernel_total_size <= init_size
309 *  - kernel_total_size <= output_size (see Note below)
310 *  - output + init_size >= output + output_size
311 *
312 * (Note that kernel_total_size and output_size have no fundamental
313 * relationship, but output_size is passed to choose_random_location
314 * as a maximum of the two. The diagram is showing a case where
315 * kernel_total_size is larger than output_size, but this case is
316 * handled by bumping output_size.)
317 *
318 * The above conditions can be illustrated by a diagram:
319 *
320 * 0   output            input            input+input_size    output+init_size
321 * |     |                 |                             |             |
322 * |     |                 |                             |             |
323 * |-----|--------|--------|--------------|-----------|--|-------------|
324 *                |                       |           |
325 *                |                       |           |
326 * output+init_size-ZO_INIT_SIZE  output+output_size  output+kernel_total_size
327 *
328 * [output, output+init_size) is the entire memory range used for
329 * extracting the compressed image.
330 *
331 * [output, output+kernel_total_size) is the range needed for the
332 * uncompressed kernel (VO) and its run size (bss, brk, etc).
333 *
334 * [output, output+output_size) is VO plus relocs (i.e. the entire
335 * uncompressed payload contained by ZO). This is the area of the buffer
336 * written to during decompression.
337 *
338 * [output+init_size-ZO_INIT_SIZE, output+init_size) is the worst-case
339 * range of the copied ZO and decompression code. (i.e. the range
340 * covered backwards of size ZO_INIT_SIZE, starting from output+init_size.)
341 *
342 * [input, input+input_size) is the original copied compressed image (ZO)
343 * (i.e. it does not include its run size). This range must be avoided
344 * because it contains the data used for decompression.
345 *
346 * [input+input_size, output+init_size) is [_text, _end) for ZO. This
347 * range includes ZO's heap and stack, and must be avoided since it
348 * performs the decompression.
349 *
350 * Since the above two ranges need to be avoided and they are adjacent,
351 * they can be merged, resulting in: [input, output+init_size) which
352 * becomes the MEM_AVOID_ZO_RANGE below.
353 */
354static void mem_avoid_init(unsigned long input, unsigned long input_size,
355			   unsigned long output)
356{
357	unsigned long init_size = boot_params_ptr->hdr.init_size;
358	u64 initrd_start, initrd_size;
359	unsigned long cmd_line, cmd_line_size;
360
361	/*
362	 * Avoid the region that is unsafe to overlap during
363	 * decompression.
364	 */
365	mem_avoid[MEM_AVOID_ZO_RANGE].start = input;
366	mem_avoid[MEM_AVOID_ZO_RANGE].size = (output + init_size) - input;
367
368	/* Avoid initrd. */
369	initrd_start  = (u64)boot_params_ptr->ext_ramdisk_image << 32;
370	initrd_start |= boot_params_ptr->hdr.ramdisk_image;
371	initrd_size  = (u64)boot_params_ptr->ext_ramdisk_size << 32;
372	initrd_size |= boot_params_ptr->hdr.ramdisk_size;
373	mem_avoid[MEM_AVOID_INITRD].start = initrd_start;
374	mem_avoid[MEM_AVOID_INITRD].size = initrd_size;
375	/* No need to set mapping for initrd, it will be handled in VO. */
376
377	/* Avoid kernel command line. */
378	cmd_line = get_cmd_line_ptr();
379	/* Calculate size of cmd_line. */
380	if (cmd_line) {
381		cmd_line_size = strnlen((char *)cmd_line, COMMAND_LINE_SIZE-1) + 1;
382		mem_avoid[MEM_AVOID_CMDLINE].start = cmd_line;
383		mem_avoid[MEM_AVOID_CMDLINE].size = cmd_line_size;
384	}
385
386	/* Avoid boot parameters. */
387	mem_avoid[MEM_AVOID_BOOTPARAMS].start = (unsigned long)boot_params_ptr;
388	mem_avoid[MEM_AVOID_BOOTPARAMS].size = sizeof(*boot_params_ptr);
389
390	/* We don't need to set a mapping for setup_data. */
391
392	/* Mark the memmap regions we need to avoid */
393	handle_mem_options();
394
395	/* Enumerate the immovable memory regions */
396	num_immovable_mem = count_immovable_mem_regions();
397}
398
399/*
400 * Does this memory vector overlap a known avoided area? If so, record the
401 * overlap region with the lowest address.
402 */
403static bool mem_avoid_overlap(struct mem_vector *img,
404			      struct mem_vector *overlap)
405{
406	int i;
407	struct setup_data *ptr;
408	u64 earliest = img->start + img->size;
409	bool is_overlapping = false;
410
411	for (i = 0; i < MEM_AVOID_MAX; i++) {
412		if (mem_overlaps(img, &mem_avoid[i]) &&
413		    mem_avoid[i].start < earliest) {
414			*overlap = mem_avoid[i];
415			earliest = overlap->start;
416			is_overlapping = true;
417		}
418	}
419
420	/* Avoid all entries in the setup_data linked list. */
421	ptr = (struct setup_data *)(unsigned long)boot_params_ptr->hdr.setup_data;
422	while (ptr) {
423		struct mem_vector avoid;
424
425		avoid.start = (unsigned long)ptr;
426		avoid.size = sizeof(*ptr) + ptr->len;
427
428		if (mem_overlaps(img, &avoid) && (avoid.start < earliest)) {
429			*overlap = avoid;
430			earliest = overlap->start;
431			is_overlapping = true;
432		}
433
434		if (ptr->type == SETUP_INDIRECT &&
435		    ((struct setup_indirect *)ptr->data)->type != SETUP_INDIRECT) {
436			avoid.start = ((struct setup_indirect *)ptr->data)->addr;
437			avoid.size = ((struct setup_indirect *)ptr->data)->len;
438
439			if (mem_overlaps(img, &avoid) && (avoid.start < earliest)) {
440				*overlap = avoid;
441				earliest = overlap->start;
442				is_overlapping = true;
443			}
444		}
445
446		ptr = (struct setup_data *)(unsigned long)ptr->next;
447	}
448
449	return is_overlapping;
450}
451
452struct slot_area {
453	u64 addr;
454	unsigned long num;
455};
456
457#define MAX_SLOT_AREA 100
458
459static struct slot_area slot_areas[MAX_SLOT_AREA];
460static unsigned int slot_area_index;
461static unsigned long slot_max;
462
463static void store_slot_info(struct mem_vector *region, unsigned long image_size)
464{
465	struct slot_area slot_area;
466
467	if (slot_area_index == MAX_SLOT_AREA)
468		return;
469
470	slot_area.addr = region->start;
471	slot_area.num = 1 + (region->size - image_size) / CONFIG_PHYSICAL_ALIGN;
472
473	slot_areas[slot_area_index++] = slot_area;
474	slot_max += slot_area.num;
475}
476
477/*
478 * Skip as many 1GB huge pages as possible in the passed region
479 * according to the number which users specified:
480 */
481static void
482process_gb_huge_pages(struct mem_vector *region, unsigned long image_size)
483{
484	u64 pud_start, pud_end;
485	unsigned long gb_huge_pages;
486	struct mem_vector tmp;
487
488	if (!IS_ENABLED(CONFIG_X86_64) || !max_gb_huge_pages) {
489		store_slot_info(region, image_size);
490		return;
491	}
492
493	/* Are there any 1GB pages in the region? */
494	pud_start = ALIGN(region->start, PUD_SIZE);
495	pud_end = ALIGN_DOWN(region->start + region->size, PUD_SIZE);
496
497	/* No good 1GB huge pages found: */
498	if (pud_start >= pud_end) {
499		store_slot_info(region, image_size);
500		return;
501	}
502
503	/* Check if the head part of the region is usable. */
504	if (pud_start >= region->start + image_size) {
505		tmp.start = region->start;
506		tmp.size = pud_start - region->start;
507		store_slot_info(&tmp, image_size);
508	}
509
510	/* Skip the good 1GB pages. */
511	gb_huge_pages = (pud_end - pud_start) >> PUD_SHIFT;
512	if (gb_huge_pages > max_gb_huge_pages) {
513		pud_end = pud_start + (max_gb_huge_pages << PUD_SHIFT);
514		max_gb_huge_pages = 0;
515	} else {
516		max_gb_huge_pages -= gb_huge_pages;
517	}
518
519	/* Check if the tail part of the region is usable. */
520	if (region->start + region->size >= pud_end + image_size) {
521		tmp.start = pud_end;
522		tmp.size = region->start + region->size - pud_end;
523		store_slot_info(&tmp, image_size);
524	}
525}
526
527static u64 slots_fetch_random(void)
528{
529	unsigned long slot;
530	unsigned int i;
531
532	/* Handle case of no slots stored. */
533	if (slot_max == 0)
534		return 0;
535
536	slot = kaslr_get_random_long("Physical") % slot_max;
537
538	for (i = 0; i < slot_area_index; i++) {
539		if (slot >= slot_areas[i].num) {
540			slot -= slot_areas[i].num;
541			continue;
542		}
543		return slot_areas[i].addr + ((u64)slot * CONFIG_PHYSICAL_ALIGN);
544	}
545
546	if (i == slot_area_index)
547		debug_putstr("slots_fetch_random() failed!?\n");
548	return 0;
549}
550
551static void __process_mem_region(struct mem_vector *entry,
552				 unsigned long minimum,
553				 unsigned long image_size)
554{
555	struct mem_vector region, overlap;
556	u64 region_end;
557
558	/* Enforce minimum and memory limit. */
559	region.start = max_t(u64, entry->start, minimum);
560	region_end = min(entry->start + entry->size, mem_limit);
561
562	/* Give up if slot area array is full. */
563	while (slot_area_index < MAX_SLOT_AREA) {
564		/* Potentially raise address to meet alignment needs. */
565		region.start = ALIGN(region.start, CONFIG_PHYSICAL_ALIGN);
566
567		/* Did we raise the address above the passed in memory entry? */
568		if (region.start > region_end)
569			return;
570
571		/* Reduce size by any delta from the original address. */
572		region.size = region_end - region.start;
573
574		/* Return if region can't contain decompressed kernel */
575		if (region.size < image_size)
576			return;
577
578		/* If nothing overlaps, store the region and return. */
579		if (!mem_avoid_overlap(&region, &overlap)) {
580			process_gb_huge_pages(&region, image_size);
581			return;
582		}
583
584		/* Store beginning of region if holds at least image_size. */
585		if (overlap.start >= region.start + image_size) {
586			region.size = overlap.start - region.start;
587			process_gb_huge_pages(&region, image_size);
588		}
589
590		/* Clip off the overlapping region and start over. */
591		region.start = overlap.start + overlap.size;
592	}
593}
594
595static bool process_mem_region(struct mem_vector *region,
596			       unsigned long minimum,
597			       unsigned long image_size)
598{
599	int i;
600	/*
601	 * If no immovable memory found, or MEMORY_HOTREMOVE disabled,
602	 * use @region directly.
603	 */
604	if (!num_immovable_mem) {
605		__process_mem_region(region, minimum, image_size);
606
607		if (slot_area_index == MAX_SLOT_AREA) {
608			debug_putstr("Aborted e820/efi memmap scan (slot_areas full)!\n");
609			return true;
610		}
611		return false;
612	}
613
614#if defined(CONFIG_MEMORY_HOTREMOVE) && defined(CONFIG_ACPI)
615	/*
616	 * If immovable memory found, filter the intersection between
617	 * immovable memory and @region.
618	 */
619	for (i = 0; i < num_immovable_mem; i++) {
620		u64 start, end, entry_end, region_end;
621		struct mem_vector entry;
622
623		if (!mem_overlaps(region, &immovable_mem[i]))
624			continue;
625
626		start = immovable_mem[i].start;
627		end = start + immovable_mem[i].size;
628		region_end = region->start + region->size;
629
630		entry.start = clamp(region->start, start, end);
631		entry_end = clamp(region_end, start, end);
632		entry.size = entry_end - entry.start;
633
634		__process_mem_region(&entry, minimum, image_size);
635
636		if (slot_area_index == MAX_SLOT_AREA) {
637			debug_putstr("Aborted e820/efi memmap scan when walking immovable regions(slot_areas full)!\n");
638			return true;
639		}
640	}
641#endif
642	return false;
643}
644
645#ifdef CONFIG_EFI
646
647/*
648 * Only EFI_CONVENTIONAL_MEMORY and EFI_UNACCEPTED_MEMORY (if supported) are
649 * guaranteed to be free.
650 *
651 * Pick free memory more conservatively than the EFI spec allows: according to
652 * the spec, EFI_BOOT_SERVICES_{CODE|DATA} are also free memory and thus
653 * available to place the kernel image into, but in practice there's firmware
654 * where using that memory leads to crashes. Buggy vendor EFI code registers
655 * for an event that triggers on SetVirtualAddressMap(). The handler assumes
656 * that EFI_BOOT_SERVICES_DATA memory has not been touched by loader yet, which
657 * is probably true for Windows.
658 *
659 * Preserve EFI_BOOT_SERVICES_* regions until after SetVirtualAddressMap().
660 */
661static inline bool memory_type_is_free(efi_memory_desc_t *md)
662{
663	if (md->type == EFI_CONVENTIONAL_MEMORY)
664		return true;
665
666	if (IS_ENABLED(CONFIG_UNACCEPTED_MEMORY) &&
667	    md->type == EFI_UNACCEPTED_MEMORY)
668		    return true;
669
670	return false;
671}
672
673/*
674 * Returns true if we processed the EFI memmap, which we prefer over the E820
675 * table if it is available.
676 */
677static bool
678process_efi_entries(unsigned long minimum, unsigned long image_size)
679{
680	struct efi_info *e = &boot_params_ptr->efi_info;
681	bool efi_mirror_found = false;
682	struct mem_vector region;
683	efi_memory_desc_t *md;
684	unsigned long pmap;
685	char *signature;
686	u32 nr_desc;
687	int i;
688
689	signature = (char *)&e->efi_loader_signature;
690	if (strncmp(signature, EFI32_LOADER_SIGNATURE, 4) &&
691	    strncmp(signature, EFI64_LOADER_SIGNATURE, 4))
692		return false;
693
694#ifdef CONFIG_X86_32
695	/* Can't handle data above 4GB at this time */
696	if (e->efi_memmap_hi) {
697		warn("EFI memmap is above 4GB, can't be handled now on x86_32. EFI should be disabled.\n");
698		return false;
699	}
700	pmap =  e->efi_memmap;
701#else
702	pmap = (e->efi_memmap | ((__u64)e->efi_memmap_hi << 32));
703#endif
704
705	nr_desc = e->efi_memmap_size / e->efi_memdesc_size;
706	for (i = 0; i < nr_desc; i++) {
707		md = efi_early_memdesc_ptr(pmap, e->efi_memdesc_size, i);
708		if (md->attribute & EFI_MEMORY_MORE_RELIABLE) {
709			efi_mirror_found = true;
710			break;
711		}
712	}
713
714	for (i = 0; i < nr_desc; i++) {
715		md = efi_early_memdesc_ptr(pmap, e->efi_memdesc_size, i);
716
717		if (!memory_type_is_free(md))
718			continue;
719
720		if (efi_soft_reserve_enabled() &&
721		    (md->attribute & EFI_MEMORY_SP))
722			continue;
723
724		if (efi_mirror_found &&
725		    !(md->attribute & EFI_MEMORY_MORE_RELIABLE))
726			continue;
727
728		region.start = md->phys_addr;
729		region.size = md->num_pages << EFI_PAGE_SHIFT;
730		if (process_mem_region(&region, minimum, image_size))
731			break;
732	}
733	return true;
734}
735#else
736static inline bool
737process_efi_entries(unsigned long minimum, unsigned long image_size)
738{
739	return false;
740}
741#endif
742
743static void process_e820_entries(unsigned long minimum,
744				 unsigned long image_size)
745{
746	int i;
747	struct mem_vector region;
748	struct boot_e820_entry *entry;
749
750	/* Verify potential e820 positions, appending to slots list. */
751	for (i = 0; i < boot_params_ptr->e820_entries; i++) {
752		entry = &boot_params_ptr->e820_table[i];
753		/* Skip non-RAM entries. */
754		if (entry->type != E820_TYPE_RAM)
755			continue;
756		region.start = entry->addr;
757		region.size = entry->size;
758		if (process_mem_region(&region, minimum, image_size))
759			break;
760	}
761}
762
763/*
764 * If KHO is active, only process its scratch areas to ensure we are not
765 * stepping onto preserved memory.
766 */
767static bool process_kho_entries(unsigned long minimum, unsigned long image_size)
768{
769	struct kho_scratch *kho_scratch;
770	struct setup_data *ptr;
771	struct kho_data *kho;
772	int i, nr_areas = 0;
773
774	if (!IS_ENABLED(CONFIG_KEXEC_HANDOVER))
775		return false;
776
777	ptr = (struct setup_data *)(unsigned long)boot_params_ptr->hdr.setup_data;
778	while (ptr) {
779		if (ptr->type == SETUP_KEXEC_KHO) {
780			kho = (struct kho_data *)(unsigned long)ptr->data;
781			kho_scratch = (void *)(unsigned long)kho->scratch_addr;
782			nr_areas = kho->scratch_size / sizeof(*kho_scratch);
783			break;
784		}
785
786		ptr = (struct setup_data *)(unsigned long)ptr->next;
787	}
788
789	if (!nr_areas)
790		return false;
791
792	for (i = 0; i < nr_areas; i++) {
793		struct kho_scratch *area = &kho_scratch[i];
794		struct mem_vector region = {
795			.start = area->addr,
796			.size = area->size,
797		};
798
799		if (process_mem_region(&region, minimum, image_size))
800			break;
801	}
802
803	return true;
804}
805
806static unsigned long find_random_phys_addr(unsigned long minimum,
807					   unsigned long image_size)
808{
809	u64 phys_addr;
810
811	/* Bail out early if it's impossible to succeed. */
812	if (minimum + image_size > mem_limit)
813		return 0;
814
815	/* Check if we had too many memmaps. */
816	if (memmap_too_large) {
817		debug_putstr("Aborted memory entries scan (more than 4 memmap= args)!\n");
818		return 0;
819	}
820
821	/*
822	 * During kexec handover only process KHO scratch areas that are known
823	 * not to contain any data that must be preserved.
824	 */
825	if (!process_kho_entries(minimum, image_size) &&
826	    !process_efi_entries(minimum, image_size))
827		process_e820_entries(minimum, image_size);
828
829	phys_addr = slots_fetch_random();
830
831	/* Perform a final check to make sure the address is in range. */
832	if (phys_addr < minimum || phys_addr + image_size > mem_limit) {
833		warn("Invalid physical address chosen!\n");
834		return 0;
835	}
836
837	return (unsigned long)phys_addr;
838}
839
840static unsigned long find_random_virt_addr(unsigned long minimum,
841					   unsigned long image_size)
842{
843	unsigned long slots, random_addr;
844
845	/*
846	 * There are how many CONFIG_PHYSICAL_ALIGN-sized slots
847	 * that can hold image_size within the range of minimum to
848	 * KERNEL_IMAGE_SIZE?
849	 */
850	slots = 1 + (KERNEL_IMAGE_SIZE - minimum - image_size) / CONFIG_PHYSICAL_ALIGN;
851
852	random_addr = kaslr_get_random_long("Virtual") % slots;
853
854	return random_addr * CONFIG_PHYSICAL_ALIGN + minimum;
855}
856
857/*
858 * Since this function examines addresses much more numerically,
859 * it takes the input and output pointers as 'unsigned long'.
860 */
861void choose_random_location(unsigned long input,
862			    unsigned long input_size,
863			    unsigned long *output,
864			    unsigned long output_size,
865			    unsigned long *virt_addr)
866{
867	unsigned long random_addr, min_addr;
868
869	if (cmdline_find_option_bool("nokaslr")) {
870		warn("KASLR disabled: 'nokaslr' on cmdline.");
871		return;
872	}
873
874	boot_params_ptr->hdr.loadflags |= KASLR_FLAG;
875
876	if (IS_ENABLED(CONFIG_X86_32))
877		mem_limit = KERNEL_IMAGE_SIZE;
878	else
879		mem_limit = MAXMEM;
880
881	/* Record the various known unsafe memory ranges. */
882	mem_avoid_init(input, input_size, *output);
883
884	/*
885	 * Low end of the randomization range should be the
886	 * smaller of 512M or the initial kernel image
887	 * location:
888	 */
889	min_addr = min(*output, 512UL << 20);
890	/* Make sure minimum is aligned. */
891	min_addr = ALIGN(min_addr, CONFIG_PHYSICAL_ALIGN);
892
893	/* Walk available memory entries to find a random address. */
894	random_addr = find_random_phys_addr(min_addr, output_size);
895	if (!random_addr) {
896		warn("Physical KASLR disabled: no suitable memory region!");
897	} else {
898		/* Update the new physical address location. */
899		if (*output != random_addr)
900			*output = random_addr;
901	}
902
903
904	/* Pick random virtual address starting from LOAD_PHYSICAL_ADDR. */
905	if (IS_ENABLED(CONFIG_X86_64))
906		random_addr = find_random_virt_addr(LOAD_PHYSICAL_ADDR, output_size);
907	*virt_addr = random_addr;
908}