Loading...
1/*
2 * Based on arch/arm/mm/init.c
3 *
4 * Copyright (C) 1995-2005 Russell King
5 * Copyright (C) 2012 ARM Ltd.
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program. If not, see <http://www.gnu.org/licenses/>.
18 */
19
20#include <linux/kernel.h>
21#include <linux/export.h>
22#include <linux/errno.h>
23#include <linux/swap.h>
24#include <linux/init.h>
25#include <linux/bootmem.h>
26#include <linux/cache.h>
27#include <linux/mman.h>
28#include <linux/nodemask.h>
29#include <linux/initrd.h>
30#include <linux/gfp.h>
31#include <linux/memblock.h>
32#include <linux/sort.h>
33#include <linux/of.h>
34#include <linux/of_fdt.h>
35#include <linux/dma-mapping.h>
36#include <linux/dma-contiguous.h>
37#include <linux/efi.h>
38#include <linux/swiotlb.h>
39#include <linux/vmalloc.h>
40#include <linux/mm.h>
41#include <linux/kexec.h>
42#include <linux/crash_dump.h>
43
44#include <asm/boot.h>
45#include <asm/fixmap.h>
46#include <asm/kasan.h>
47#include <asm/kernel-pgtable.h>
48#include <asm/memory.h>
49#include <asm/numa.h>
50#include <asm/sections.h>
51#include <asm/setup.h>
52#include <asm/sizes.h>
53#include <asm/tlb.h>
54#include <asm/alternative.h>
55
56/*
57 * We need to be able to catch inadvertent references to memstart_addr
58 * that occur (potentially in generic code) before arm64_memblock_init()
59 * executes, which assigns it its actual value. So use a default value
60 * that cannot be mistaken for a real physical address.
61 */
62s64 memstart_addr __ro_after_init = -1;
63phys_addr_t arm64_dma_phys_limit __ro_after_init;
64
65#ifdef CONFIG_BLK_DEV_INITRD
66static int __init early_initrd(char *p)
67{
68 unsigned long start, size;
69 char *endp;
70
71 start = memparse(p, &endp);
72 if (*endp == ',') {
73 size = memparse(endp + 1, NULL);
74
75 initrd_start = start;
76 initrd_end = start + size;
77 }
78 return 0;
79}
80early_param("initrd", early_initrd);
81#endif
82
83#ifdef CONFIG_KEXEC_CORE
84/*
85 * reserve_crashkernel() - reserves memory for crash kernel
86 *
87 * This function reserves memory area given in "crashkernel=" kernel command
88 * line parameter. The memory reserved is used by dump capture kernel when
89 * primary kernel is crashing.
90 */
91static void __init reserve_crashkernel(void)
92{
93 unsigned long long crash_base, crash_size;
94 int ret;
95
96 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
97 &crash_size, &crash_base);
98 /* no crashkernel= or invalid value specified */
99 if (ret || !crash_size)
100 return;
101
102 crash_size = PAGE_ALIGN(crash_size);
103
104 if (crash_base == 0) {
105 /* Current arm64 boot protocol requires 2MB alignment */
106 crash_base = memblock_find_in_range(0, ARCH_LOW_ADDRESS_LIMIT,
107 crash_size, SZ_2M);
108 if (crash_base == 0) {
109 pr_warn("cannot allocate crashkernel (size:0x%llx)\n",
110 crash_size);
111 return;
112 }
113 } else {
114 /* User specifies base address explicitly. */
115 if (!memblock_is_region_memory(crash_base, crash_size)) {
116 pr_warn("cannot reserve crashkernel: region is not memory\n");
117 return;
118 }
119
120 if (memblock_is_region_reserved(crash_base, crash_size)) {
121 pr_warn("cannot reserve crashkernel: region overlaps reserved memory\n");
122 return;
123 }
124
125 if (!IS_ALIGNED(crash_base, SZ_2M)) {
126 pr_warn("cannot reserve crashkernel: base address is not 2MB aligned\n");
127 return;
128 }
129 }
130 memblock_reserve(crash_base, crash_size);
131
132 pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n",
133 crash_base, crash_base + crash_size, crash_size >> 20);
134
135 crashk_res.start = crash_base;
136 crashk_res.end = crash_base + crash_size - 1;
137}
138
139static void __init kexec_reserve_crashkres_pages(void)
140{
141#ifdef CONFIG_HIBERNATION
142 phys_addr_t addr;
143 struct page *page;
144
145 if (!crashk_res.end)
146 return;
147
148 /*
149 * To reduce the size of hibernation image, all the pages are
150 * marked as Reserved initially.
151 */
152 for (addr = crashk_res.start; addr < (crashk_res.end + 1);
153 addr += PAGE_SIZE) {
154 page = phys_to_page(addr);
155 SetPageReserved(page);
156 }
157#endif
158}
159#else
160static void __init reserve_crashkernel(void)
161{
162}
163
164static void __init kexec_reserve_crashkres_pages(void)
165{
166}
167#endif /* CONFIG_KEXEC_CORE */
168
169#ifdef CONFIG_CRASH_DUMP
170static int __init early_init_dt_scan_elfcorehdr(unsigned long node,
171 const char *uname, int depth, void *data)
172{
173 const __be32 *reg;
174 int len;
175
176 if (depth != 1 || strcmp(uname, "chosen") != 0)
177 return 0;
178
179 reg = of_get_flat_dt_prop(node, "linux,elfcorehdr", &len);
180 if (!reg || (len < (dt_root_addr_cells + dt_root_size_cells)))
181 return 1;
182
183 elfcorehdr_addr = dt_mem_next_cell(dt_root_addr_cells, ®);
184 elfcorehdr_size = dt_mem_next_cell(dt_root_size_cells, ®);
185
186 return 1;
187}
188
189/*
190 * reserve_elfcorehdr() - reserves memory for elf core header
191 *
192 * This function reserves the memory occupied by an elf core header
193 * described in the device tree. This region contains all the
194 * information about primary kernel's core image and is used by a dump
195 * capture kernel to access the system memory on primary kernel.
196 */
197static void __init reserve_elfcorehdr(void)
198{
199 of_scan_flat_dt(early_init_dt_scan_elfcorehdr, NULL);
200
201 if (!elfcorehdr_size)
202 return;
203
204 if (memblock_is_region_reserved(elfcorehdr_addr, elfcorehdr_size)) {
205 pr_warn("elfcorehdr is overlapped\n");
206 return;
207 }
208
209 memblock_reserve(elfcorehdr_addr, elfcorehdr_size);
210
211 pr_info("Reserving %lldKB of memory at 0x%llx for elfcorehdr\n",
212 elfcorehdr_size >> 10, elfcorehdr_addr);
213}
214#else
215static void __init reserve_elfcorehdr(void)
216{
217}
218#endif /* CONFIG_CRASH_DUMP */
219/*
220 * Return the maximum physical address for ZONE_DMA32 (DMA_BIT_MASK(32)). It
221 * currently assumes that for memory starting above 4G, 32-bit devices will
222 * use a DMA offset.
223 */
224static phys_addr_t __init max_zone_dma_phys(void)
225{
226 phys_addr_t offset = memblock_start_of_DRAM() & GENMASK_ULL(63, 32);
227 return min(offset + (1ULL << 32), memblock_end_of_DRAM());
228}
229
230#ifdef CONFIG_NUMA
231
232static void __init zone_sizes_init(unsigned long min, unsigned long max)
233{
234 unsigned long max_zone_pfns[MAX_NR_ZONES] = {0};
235
236 if (IS_ENABLED(CONFIG_ZONE_DMA32))
237 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(max_zone_dma_phys());
238 max_zone_pfns[ZONE_NORMAL] = max;
239
240 free_area_init_nodes(max_zone_pfns);
241}
242
243#else
244
245static void __init zone_sizes_init(unsigned long min, unsigned long max)
246{
247 struct memblock_region *reg;
248 unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES];
249 unsigned long max_dma = min;
250
251 memset(zone_size, 0, sizeof(zone_size));
252
253 /* 4GB maximum for 32-bit only capable devices */
254#ifdef CONFIG_ZONE_DMA32
255 max_dma = PFN_DOWN(arm64_dma_phys_limit);
256 zone_size[ZONE_DMA32] = max_dma - min;
257#endif
258 zone_size[ZONE_NORMAL] = max - max_dma;
259
260 memcpy(zhole_size, zone_size, sizeof(zhole_size));
261
262 for_each_memblock(memory, reg) {
263 unsigned long start = memblock_region_memory_base_pfn(reg);
264 unsigned long end = memblock_region_memory_end_pfn(reg);
265
266 if (start >= max)
267 continue;
268
269#ifdef CONFIG_ZONE_DMA32
270 if (start < max_dma) {
271 unsigned long dma_end = min(end, max_dma);
272 zhole_size[ZONE_DMA32] -= dma_end - start;
273 }
274#endif
275 if (end > max_dma) {
276 unsigned long normal_end = min(end, max);
277 unsigned long normal_start = max(start, max_dma);
278 zhole_size[ZONE_NORMAL] -= normal_end - normal_start;
279 }
280 }
281
282 free_area_init_node(0, zone_size, min, zhole_size);
283}
284
285#endif /* CONFIG_NUMA */
286
287#ifdef CONFIG_HAVE_ARCH_PFN_VALID
288int pfn_valid(unsigned long pfn)
289{
290 return memblock_is_map_memory(pfn << PAGE_SHIFT);
291}
292EXPORT_SYMBOL(pfn_valid);
293#endif
294
295#ifndef CONFIG_SPARSEMEM
296static void __init arm64_memory_present(void)
297{
298}
299#else
300static void __init arm64_memory_present(void)
301{
302 struct memblock_region *reg;
303
304 for_each_memblock(memory, reg) {
305 int nid = memblock_get_region_node(reg);
306
307 memory_present(nid, memblock_region_memory_base_pfn(reg),
308 memblock_region_memory_end_pfn(reg));
309 }
310}
311#endif
312
313static phys_addr_t memory_limit = (phys_addr_t)ULLONG_MAX;
314
315/*
316 * Limit the memory size that was specified via FDT.
317 */
318static int __init early_mem(char *p)
319{
320 if (!p)
321 return 1;
322
323 memory_limit = memparse(p, &p) & PAGE_MASK;
324 pr_notice("Memory limited to %lldMB\n", memory_limit >> 20);
325
326 return 0;
327}
328early_param("mem", early_mem);
329
330static int __init early_init_dt_scan_usablemem(unsigned long node,
331 const char *uname, int depth, void *data)
332{
333 struct memblock_region *usablemem = data;
334 const __be32 *reg;
335 int len;
336
337 if (depth != 1 || strcmp(uname, "chosen") != 0)
338 return 0;
339
340 reg = of_get_flat_dt_prop(node, "linux,usable-memory-range", &len);
341 if (!reg || (len < (dt_root_addr_cells + dt_root_size_cells)))
342 return 1;
343
344 usablemem->base = dt_mem_next_cell(dt_root_addr_cells, ®);
345 usablemem->size = dt_mem_next_cell(dt_root_size_cells, ®);
346
347 return 1;
348}
349
350static void __init fdt_enforce_memory_region(void)
351{
352 struct memblock_region reg = {
353 .size = 0,
354 };
355
356 of_scan_flat_dt(early_init_dt_scan_usablemem, ®);
357
358 if (reg.size)
359 memblock_cap_memory_range(reg.base, reg.size);
360}
361
362void __init arm64_memblock_init(void)
363{
364 const s64 linear_region_size = -(s64)PAGE_OFFSET;
365
366 /* Handle linux,usable-memory-range property */
367 fdt_enforce_memory_region();
368
369 /* Remove memory above our supported physical address size */
370 memblock_remove(1ULL << PHYS_MASK_SHIFT, ULLONG_MAX);
371
372 /*
373 * Ensure that the linear region takes up exactly half of the kernel
374 * virtual address space. This way, we can distinguish a linear address
375 * from a kernel/module/vmalloc address by testing a single bit.
376 */
377 BUILD_BUG_ON(linear_region_size != BIT(VA_BITS - 1));
378
379 /*
380 * Select a suitable value for the base of physical memory.
381 */
382 memstart_addr = round_down(memblock_start_of_DRAM(),
383 ARM64_MEMSTART_ALIGN);
384
385 /*
386 * Remove the memory that we will not be able to cover with the
387 * linear mapping. Take care not to clip the kernel which may be
388 * high in memory.
389 */
390 memblock_remove(max_t(u64, memstart_addr + linear_region_size,
391 __pa_symbol(_end)), ULLONG_MAX);
392 if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {
393 /* ensure that memstart_addr remains sufficiently aligned */
394 memstart_addr = round_up(memblock_end_of_DRAM() - linear_region_size,
395 ARM64_MEMSTART_ALIGN);
396 memblock_remove(0, memstart_addr);
397 }
398
399 /*
400 * Apply the memory limit if it was set. Since the kernel may be loaded
401 * high up in memory, add back the kernel region that must be accessible
402 * via the linear mapping.
403 */
404 if (memory_limit != (phys_addr_t)ULLONG_MAX) {
405 memblock_mem_limit_remove_map(memory_limit);
406 memblock_add(__pa_symbol(_text), (u64)(_end - _text));
407 }
408
409 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && initrd_start) {
410 /*
411 * Add back the memory we just removed if it results in the
412 * initrd to become inaccessible via the linear mapping.
413 * Otherwise, this is a no-op
414 */
415 u64 base = initrd_start & PAGE_MASK;
416 u64 size = PAGE_ALIGN(initrd_end) - base;
417
418 /*
419 * We can only add back the initrd memory if we don't end up
420 * with more memory than we can address via the linear mapping.
421 * It is up to the bootloader to position the kernel and the
422 * initrd reasonably close to each other (i.e., within 32 GB of
423 * each other) so that all granule/#levels combinations can
424 * always access both.
425 */
426 if (WARN(base < memblock_start_of_DRAM() ||
427 base + size > memblock_start_of_DRAM() +
428 linear_region_size,
429 "initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
430 initrd_start = 0;
431 } else {
432 memblock_remove(base, size); /* clear MEMBLOCK_ flags */
433 memblock_add(base, size);
434 memblock_reserve(base, size);
435 }
436 }
437
438 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
439 extern u16 memstart_offset_seed;
440 u64 range = linear_region_size -
441 (memblock_end_of_DRAM() - memblock_start_of_DRAM());
442
443 /*
444 * If the size of the linear region exceeds, by a sufficient
445 * margin, the size of the region that the available physical
446 * memory spans, randomize the linear region as well.
447 */
448 if (memstart_offset_seed > 0 && range >= ARM64_MEMSTART_ALIGN) {
449 range = range / ARM64_MEMSTART_ALIGN + 1;
450 memstart_addr -= ARM64_MEMSTART_ALIGN *
451 ((range * memstart_offset_seed) >> 16);
452 }
453 }
454
455 /*
456 * Register the kernel text, kernel data, initrd, and initial
457 * pagetables with memblock.
458 */
459 memblock_reserve(__pa_symbol(_text), _end - _text);
460#ifdef CONFIG_BLK_DEV_INITRD
461 if (initrd_start) {
462 memblock_reserve(initrd_start, initrd_end - initrd_start);
463
464 /* the generic initrd code expects virtual addresses */
465 initrd_start = __phys_to_virt(initrd_start);
466 initrd_end = __phys_to_virt(initrd_end);
467 }
468#endif
469
470 early_init_fdt_scan_reserved_mem();
471
472 /* 4GB maximum for 32-bit only capable devices */
473 if (IS_ENABLED(CONFIG_ZONE_DMA32))
474 arm64_dma_phys_limit = max_zone_dma_phys();
475 else
476 arm64_dma_phys_limit = PHYS_MASK + 1;
477
478 reserve_crashkernel();
479
480 reserve_elfcorehdr();
481
482 high_memory = __va(memblock_end_of_DRAM() - 1) + 1;
483
484 dma_contiguous_reserve(arm64_dma_phys_limit);
485
486 memblock_allow_resize();
487}
488
489void __init bootmem_init(void)
490{
491 unsigned long min, max;
492
493 min = PFN_UP(memblock_start_of_DRAM());
494 max = PFN_DOWN(memblock_end_of_DRAM());
495
496 early_memtest(min << PAGE_SHIFT, max << PAGE_SHIFT);
497
498 max_pfn = max_low_pfn = max;
499
500 arm64_numa_init();
501 /*
502 * Sparsemem tries to allocate bootmem in memory_present(), so must be
503 * done after the fixed reservations.
504 */
505 arm64_memory_present();
506
507 sparse_init();
508 zone_sizes_init(min, max);
509
510 memblock_dump_all();
511}
512
513#ifndef CONFIG_SPARSEMEM_VMEMMAP
514static inline void free_memmap(unsigned long start_pfn, unsigned long end_pfn)
515{
516 struct page *start_pg, *end_pg;
517 unsigned long pg, pgend;
518
519 /*
520 * Convert start_pfn/end_pfn to a struct page pointer.
521 */
522 start_pg = pfn_to_page(start_pfn - 1) + 1;
523 end_pg = pfn_to_page(end_pfn - 1) + 1;
524
525 /*
526 * Convert to physical addresses, and round start upwards and end
527 * downwards.
528 */
529 pg = (unsigned long)PAGE_ALIGN(__pa(start_pg));
530 pgend = (unsigned long)__pa(end_pg) & PAGE_MASK;
531
532 /*
533 * If there are free pages between these, free the section of the
534 * memmap array.
535 */
536 if (pg < pgend)
537 free_bootmem(pg, pgend - pg);
538}
539
540/*
541 * The mem_map array can get very big. Free the unused area of the memory map.
542 */
543static void __init free_unused_memmap(void)
544{
545 unsigned long start, prev_end = 0;
546 struct memblock_region *reg;
547
548 for_each_memblock(memory, reg) {
549 start = __phys_to_pfn(reg->base);
550
551#ifdef CONFIG_SPARSEMEM
552 /*
553 * Take care not to free memmap entries that don't exist due
554 * to SPARSEMEM sections which aren't present.
555 */
556 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
557#endif
558 /*
559 * If we had a previous bank, and there is a space between the
560 * current bank and the previous, free it.
561 */
562 if (prev_end && prev_end < start)
563 free_memmap(prev_end, start);
564
565 /*
566 * Align up here since the VM subsystem insists that the
567 * memmap entries are valid from the bank end aligned to
568 * MAX_ORDER_NR_PAGES.
569 */
570 prev_end = ALIGN(__phys_to_pfn(reg->base + reg->size),
571 MAX_ORDER_NR_PAGES);
572 }
573
574#ifdef CONFIG_SPARSEMEM
575 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
576 free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
577#endif
578}
579#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
580
581/*
582 * mem_init() marks the free areas in the mem_map and tells us how much memory
583 * is free. This is done after various parts of the system have claimed their
584 * memory after the kernel image.
585 */
586void __init mem_init(void)
587{
588 if (swiotlb_force == SWIOTLB_FORCE ||
589 max_pfn > (arm64_dma_phys_limit >> PAGE_SHIFT))
590 swiotlb_init(1);
591 else
592 swiotlb_force = SWIOTLB_NO_FORCE;
593
594 set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
595
596#ifndef CONFIG_SPARSEMEM_VMEMMAP
597 free_unused_memmap();
598#endif
599 /* this will put all unused low memory onto the freelists */
600 free_all_bootmem();
601
602 kexec_reserve_crashkres_pages();
603
604 mem_init_print_info(NULL);
605
606 /*
607 * Check boundaries twice: Some fundamental inconsistencies can be
608 * detected at build time already.
609 */
610#ifdef CONFIG_COMPAT
611 BUILD_BUG_ON(TASK_SIZE_32 > TASK_SIZE_64);
612#endif
613
614 /*
615 * Make sure we chose the upper bound of sizeof(struct page)
616 * correctly.
617 */
618 BUILD_BUG_ON(sizeof(struct page) > (1 << STRUCT_PAGE_MAX_SHIFT));
619
620 if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) {
621 extern int sysctl_overcommit_memory;
622 /*
623 * On a machine this small we won't get anywhere without
624 * overcommit, so turn it on by default.
625 */
626 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
627 }
628}
629
630void free_initmem(void)
631{
632 free_reserved_area(lm_alias(__init_begin),
633 lm_alias(__init_end),
634 0, "unused kernel");
635 /*
636 * Unmap the __init region but leave the VM area in place. This
637 * prevents the region from being reused for kernel modules, which
638 * is not supported by kallsyms.
639 */
640 unmap_kernel_range((u64)__init_begin, (u64)(__init_end - __init_begin));
641}
642
643#ifdef CONFIG_BLK_DEV_INITRD
644
645static int keep_initrd __initdata;
646
647void __init free_initrd_mem(unsigned long start, unsigned long end)
648{
649 if (!keep_initrd) {
650 free_reserved_area((void *)start, (void *)end, 0, "initrd");
651 memblock_free(__virt_to_phys(start), end - start);
652 }
653}
654
655static int __init keepinitrd_setup(char *__unused)
656{
657 keep_initrd = 1;
658 return 1;
659}
660
661__setup("keepinitrd", keepinitrd_setup);
662#endif
663
664/*
665 * Dump out memory limit information on panic.
666 */
667static int dump_mem_limit(struct notifier_block *self, unsigned long v, void *p)
668{
669 if (memory_limit != (phys_addr_t)ULLONG_MAX) {
670 pr_emerg("Memory Limit: %llu MB\n", memory_limit >> 20);
671 } else {
672 pr_emerg("Memory Limit: none\n");
673 }
674 return 0;
675}
676
677static struct notifier_block mem_limit_notifier = {
678 .notifier_call = dump_mem_limit,
679};
680
681static int __init register_mem_limit_dumper(void)
682{
683 atomic_notifier_chain_register(&panic_notifier_list,
684 &mem_limit_notifier);
685 return 0;
686}
687__initcall(register_mem_limit_dumper);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Based on arch/arm/mm/init.c
4 *
5 * Copyright (C) 1995-2005 Russell King
6 * Copyright (C) 2012 ARM Ltd.
7 */
8
9#include <linux/kernel.h>
10#include <linux/export.h>
11#include <linux/errno.h>
12#include <linux/swap.h>
13#include <linux/init.h>
14#include <linux/cache.h>
15#include <linux/mman.h>
16#include <linux/nodemask.h>
17#include <linux/initrd.h>
18#include <linux/gfp.h>
19#include <linux/math.h>
20#include <linux/memblock.h>
21#include <linux/sort.h>
22#include <linux/of.h>
23#include <linux/of_fdt.h>
24#include <linux/dma-direct.h>
25#include <linux/dma-map-ops.h>
26#include <linux/efi.h>
27#include <linux/swiotlb.h>
28#include <linux/vmalloc.h>
29#include <linux/mm.h>
30#include <linux/kexec.h>
31#include <linux/crash_dump.h>
32#include <linux/hugetlb.h>
33#include <linux/acpi_iort.h>
34#include <linux/kmemleak.h>
35
36#include <asm/boot.h>
37#include <asm/fixmap.h>
38#include <asm/kasan.h>
39#include <asm/kernel-pgtable.h>
40#include <asm/kvm_host.h>
41#include <asm/memory.h>
42#include <asm/numa.h>
43#include <asm/sections.h>
44#include <asm/setup.h>
45#include <linux/sizes.h>
46#include <asm/tlb.h>
47#include <asm/alternative.h>
48#include <asm/xen/swiotlb-xen.h>
49
50/*
51 * We need to be able to catch inadvertent references to memstart_addr
52 * that occur (potentially in generic code) before arm64_memblock_init()
53 * executes, which assigns it its actual value. So use a default value
54 * that cannot be mistaken for a real physical address.
55 */
56s64 memstart_addr __ro_after_init = -1;
57EXPORT_SYMBOL(memstart_addr);
58
59/*
60 * If the corresponding config options are enabled, we create both ZONE_DMA
61 * and ZONE_DMA32. By default ZONE_DMA covers the 32-bit addressable memory
62 * unless restricted on specific platforms (e.g. 30-bit on Raspberry Pi 4).
63 * In such case, ZONE_DMA32 covers the rest of the 32-bit addressable memory,
64 * otherwise it is empty.
65 */
66phys_addr_t __ro_after_init arm64_dma_phys_limit;
67
68/*
69 * To make optimal use of block mappings when laying out the linear
70 * mapping, round down the base of physical memory to a size that can
71 * be mapped efficiently, i.e., either PUD_SIZE (4k granule) or PMD_SIZE
72 * (64k granule), or a multiple that can be mapped using contiguous bits
73 * in the page tables: 32 * PMD_SIZE (16k granule)
74 */
75#if defined(CONFIG_ARM64_4K_PAGES)
76#define ARM64_MEMSTART_SHIFT PUD_SHIFT
77#elif defined(CONFIG_ARM64_16K_PAGES)
78#define ARM64_MEMSTART_SHIFT CONT_PMD_SHIFT
79#else
80#define ARM64_MEMSTART_SHIFT PMD_SHIFT
81#endif
82
83/*
84 * sparsemem vmemmap imposes an additional requirement on the alignment of
85 * memstart_addr, due to the fact that the base of the vmemmap region
86 * has a direct correspondence, and needs to appear sufficiently aligned
87 * in the virtual address space.
88 */
89#if ARM64_MEMSTART_SHIFT < SECTION_SIZE_BITS
90#define ARM64_MEMSTART_ALIGN (1UL << SECTION_SIZE_BITS)
91#else
92#define ARM64_MEMSTART_ALIGN (1UL << ARM64_MEMSTART_SHIFT)
93#endif
94
95static void __init arch_reserve_crashkernel(void)
96{
97 unsigned long long low_size = 0;
98 unsigned long long crash_base, crash_size;
99 char *cmdline = boot_command_line;
100 bool high = false;
101 int ret;
102
103 if (!IS_ENABLED(CONFIG_KEXEC_CORE))
104 return;
105
106 ret = parse_crashkernel(cmdline, memblock_phys_mem_size(),
107 &crash_size, &crash_base,
108 &low_size, &high);
109 if (ret)
110 return;
111
112 reserve_crashkernel_generic(cmdline, crash_size, crash_base,
113 low_size, high);
114}
115
116/*
117 * Return the maximum physical address for a zone accessible by the given bits
118 * limit. If DRAM starts above 32-bit, expand the zone to the maximum
119 * available memory, otherwise cap it at 32-bit.
120 */
121static phys_addr_t __init max_zone_phys(unsigned int zone_bits)
122{
123 phys_addr_t zone_mask = DMA_BIT_MASK(zone_bits);
124 phys_addr_t phys_start = memblock_start_of_DRAM();
125
126 if (phys_start > U32_MAX)
127 zone_mask = PHYS_ADDR_MAX;
128 else if (phys_start > zone_mask)
129 zone_mask = U32_MAX;
130
131 return min(zone_mask, memblock_end_of_DRAM() - 1) + 1;
132}
133
134static void __init zone_sizes_init(void)
135{
136 unsigned long max_zone_pfns[MAX_NR_ZONES] = {0};
137 unsigned int __maybe_unused acpi_zone_dma_bits;
138 unsigned int __maybe_unused dt_zone_dma_bits;
139 phys_addr_t __maybe_unused dma32_phys_limit = max_zone_phys(32);
140
141#ifdef CONFIG_ZONE_DMA
142 acpi_zone_dma_bits = fls64(acpi_iort_dma_get_max_cpu_address());
143 dt_zone_dma_bits = fls64(of_dma_get_max_cpu_address(NULL));
144 zone_dma_bits = min3(32U, dt_zone_dma_bits, acpi_zone_dma_bits);
145 arm64_dma_phys_limit = max_zone_phys(zone_dma_bits);
146 max_zone_pfns[ZONE_DMA] = PFN_DOWN(arm64_dma_phys_limit);
147#endif
148#ifdef CONFIG_ZONE_DMA32
149 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
150 if (!arm64_dma_phys_limit)
151 arm64_dma_phys_limit = dma32_phys_limit;
152#endif
153 if (!arm64_dma_phys_limit)
154 arm64_dma_phys_limit = PHYS_MASK + 1;
155 max_zone_pfns[ZONE_NORMAL] = max_pfn;
156
157 free_area_init(max_zone_pfns);
158}
159
160int pfn_is_map_memory(unsigned long pfn)
161{
162 phys_addr_t addr = PFN_PHYS(pfn);
163
164 /* avoid false positives for bogus PFNs, see comment in pfn_valid() */
165 if (PHYS_PFN(addr) != pfn)
166 return 0;
167
168 return memblock_is_map_memory(addr);
169}
170EXPORT_SYMBOL(pfn_is_map_memory);
171
172static phys_addr_t memory_limit __ro_after_init = PHYS_ADDR_MAX;
173
174/*
175 * Limit the memory size that was specified via FDT.
176 */
177static int __init early_mem(char *p)
178{
179 if (!p)
180 return 1;
181
182 memory_limit = memparse(p, &p) & PAGE_MASK;
183 pr_notice("Memory limited to %lldMB\n", memory_limit >> 20);
184
185 return 0;
186}
187early_param("mem", early_mem);
188
189void __init arm64_memblock_init(void)
190{
191 s64 linear_region_size = PAGE_END - _PAGE_OFFSET(vabits_actual);
192
193 /*
194 * Corner case: 52-bit VA capable systems running KVM in nVHE mode may
195 * be limited in their ability to support a linear map that exceeds 51
196 * bits of VA space, depending on the placement of the ID map. Given
197 * that the placement of the ID map may be randomized, let's simply
198 * limit the kernel's linear map to 51 bits as well if we detect this
199 * configuration.
200 */
201 if (IS_ENABLED(CONFIG_KVM) && vabits_actual == 52 &&
202 is_hyp_mode_available() && !is_kernel_in_hyp_mode()) {
203 pr_info("Capping linear region to 51 bits for KVM in nVHE mode on LVA capable hardware.\n");
204 linear_region_size = min_t(u64, linear_region_size, BIT(51));
205 }
206
207 /* Remove memory above our supported physical address size */
208 memblock_remove(1ULL << PHYS_MASK_SHIFT, ULLONG_MAX);
209
210 /*
211 * Select a suitable value for the base of physical memory.
212 */
213 memstart_addr = round_down(memblock_start_of_DRAM(),
214 ARM64_MEMSTART_ALIGN);
215
216 if ((memblock_end_of_DRAM() - memstart_addr) > linear_region_size)
217 pr_warn("Memory doesn't fit in the linear mapping, VA_BITS too small\n");
218
219 /*
220 * Remove the memory that we will not be able to cover with the
221 * linear mapping. Take care not to clip the kernel which may be
222 * high in memory.
223 */
224 memblock_remove(max_t(u64, memstart_addr + linear_region_size,
225 __pa_symbol(_end)), ULLONG_MAX);
226 if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {
227 /* ensure that memstart_addr remains sufficiently aligned */
228 memstart_addr = round_up(memblock_end_of_DRAM() - linear_region_size,
229 ARM64_MEMSTART_ALIGN);
230 memblock_remove(0, memstart_addr);
231 }
232
233 /*
234 * If we are running with a 52-bit kernel VA config on a system that
235 * does not support it, we have to place the available physical
236 * memory in the 48-bit addressable part of the linear region, i.e.,
237 * we have to move it upward. Since memstart_addr represents the
238 * physical address of PAGE_OFFSET, we have to *subtract* from it.
239 */
240 if (IS_ENABLED(CONFIG_ARM64_VA_BITS_52) && (vabits_actual != 52))
241 memstart_addr -= _PAGE_OFFSET(48) - _PAGE_OFFSET(52);
242
243 /*
244 * Apply the memory limit if it was set. Since the kernel may be loaded
245 * high up in memory, add back the kernel region that must be accessible
246 * via the linear mapping.
247 */
248 if (memory_limit != PHYS_ADDR_MAX) {
249 memblock_mem_limit_remove_map(memory_limit);
250 memblock_add(__pa_symbol(_text), (u64)(_end - _text));
251 }
252
253 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
254 /*
255 * Add back the memory we just removed if it results in the
256 * initrd to become inaccessible via the linear mapping.
257 * Otherwise, this is a no-op
258 */
259 u64 base = phys_initrd_start & PAGE_MASK;
260 u64 size = PAGE_ALIGN(phys_initrd_start + phys_initrd_size) - base;
261
262 /*
263 * We can only add back the initrd memory if we don't end up
264 * with more memory than we can address via the linear mapping.
265 * It is up to the bootloader to position the kernel and the
266 * initrd reasonably close to each other (i.e., within 32 GB of
267 * each other) so that all granule/#levels combinations can
268 * always access both.
269 */
270 if (WARN(base < memblock_start_of_DRAM() ||
271 base + size > memblock_start_of_DRAM() +
272 linear_region_size,
273 "initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
274 phys_initrd_size = 0;
275 } else {
276 memblock_add(base, size);
277 memblock_clear_nomap(base, size);
278 memblock_reserve(base, size);
279 }
280 }
281
282 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
283 extern u16 memstart_offset_seed;
284 u64 mmfr0 = read_cpuid(ID_AA64MMFR0_EL1);
285 int parange = cpuid_feature_extract_unsigned_field(
286 mmfr0, ID_AA64MMFR0_EL1_PARANGE_SHIFT);
287 s64 range = linear_region_size -
288 BIT(id_aa64mmfr0_parange_to_phys_shift(parange));
289
290 /*
291 * If the size of the linear region exceeds, by a sufficient
292 * margin, the size of the region that the physical memory can
293 * span, randomize the linear region as well.
294 */
295 if (memstart_offset_seed > 0 && range >= (s64)ARM64_MEMSTART_ALIGN) {
296 range /= ARM64_MEMSTART_ALIGN;
297 memstart_addr -= ARM64_MEMSTART_ALIGN *
298 ((range * memstart_offset_seed) >> 16);
299 }
300 }
301
302 /*
303 * Register the kernel text, kernel data, initrd, and initial
304 * pagetables with memblock.
305 */
306 memblock_reserve(__pa_symbol(_stext), _end - _stext);
307 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
308 /* the generic initrd code expects virtual addresses */
309 initrd_start = __phys_to_virt(phys_initrd_start);
310 initrd_end = initrd_start + phys_initrd_size;
311 }
312
313 early_init_fdt_scan_reserved_mem();
314
315 high_memory = __va(memblock_end_of_DRAM() - 1) + 1;
316}
317
318void __init bootmem_init(void)
319{
320 unsigned long min, max;
321
322 min = PFN_UP(memblock_start_of_DRAM());
323 max = PFN_DOWN(memblock_end_of_DRAM());
324
325 early_memtest(min << PAGE_SHIFT, max << PAGE_SHIFT);
326
327 max_pfn = max_low_pfn = max;
328 min_low_pfn = min;
329
330 arch_numa_init();
331
332 /*
333 * must be done after arch_numa_init() which calls numa_init() to
334 * initialize node_online_map that gets used in hugetlb_cma_reserve()
335 * while allocating required CMA size across online nodes.
336 */
337#if defined(CONFIG_HUGETLB_PAGE) && defined(CONFIG_CMA)
338 arm64_hugetlb_cma_reserve();
339#endif
340
341 kvm_hyp_reserve();
342
343 /*
344 * sparse_init() tries to allocate memory from memblock, so must be
345 * done after the fixed reservations
346 */
347 sparse_init();
348 zone_sizes_init();
349
350 /*
351 * Reserve the CMA area after arm64_dma_phys_limit was initialised.
352 */
353 dma_contiguous_reserve(arm64_dma_phys_limit);
354
355 /*
356 * request_standard_resources() depends on crashkernel's memory being
357 * reserved, so do it here.
358 */
359 arch_reserve_crashkernel();
360
361 memblock_dump_all();
362}
363
364/*
365 * mem_init() marks the free areas in the mem_map and tells us how much memory
366 * is free. This is done after various parts of the system have claimed their
367 * memory after the kernel image.
368 */
369void __init mem_init(void)
370{
371 bool swiotlb = max_pfn > PFN_DOWN(arm64_dma_phys_limit);
372
373 if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) && !swiotlb) {
374 /*
375 * If no bouncing needed for ZONE_DMA, reduce the swiotlb
376 * buffer for kmalloc() bouncing to 1MB per 1GB of RAM.
377 */
378 unsigned long size =
379 DIV_ROUND_UP(memblock_phys_mem_size(), 1024);
380 swiotlb_adjust_size(min(swiotlb_size_or_default(), size));
381 swiotlb = true;
382 }
383
384 swiotlb_init(swiotlb, SWIOTLB_VERBOSE);
385
386 /* this will put all unused low memory onto the freelists */
387 memblock_free_all();
388
389 /*
390 * Check boundaries twice: Some fundamental inconsistencies can be
391 * detected at build time already.
392 */
393#ifdef CONFIG_COMPAT
394 BUILD_BUG_ON(TASK_SIZE_32 > DEFAULT_MAP_WINDOW_64);
395#endif
396
397 /*
398 * Selected page table levels should match when derived from
399 * scratch using the virtual address range and page size.
400 */
401 BUILD_BUG_ON(ARM64_HW_PGTABLE_LEVELS(CONFIG_ARM64_VA_BITS) !=
402 CONFIG_PGTABLE_LEVELS);
403
404 if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) {
405 extern int sysctl_overcommit_memory;
406 /*
407 * On a machine this small we won't get anywhere without
408 * overcommit, so turn it on by default.
409 */
410 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
411 }
412}
413
414void free_initmem(void)
415{
416 free_reserved_area(lm_alias(__init_begin),
417 lm_alias(__init_end),
418 POISON_FREE_INITMEM, "unused kernel");
419 /*
420 * Unmap the __init region but leave the VM area in place. This
421 * prevents the region from being reused for kernel modules, which
422 * is not supported by kallsyms.
423 */
424 vunmap_range((u64)__init_begin, (u64)__init_end);
425}
426
427void dump_mem_limit(void)
428{
429 if (memory_limit != PHYS_ADDR_MAX) {
430 pr_emerg("Memory Limit: %llu MB\n", memory_limit >> 20);
431 } else {
432 pr_emerg("Memory Limit: none\n");
433 }
434}