2 * Based on arch/arm/mm/init.c
4 * Copyright (C) 1995-2005 Russell King
5 * Copyright (C) 2012 ARM Ltd.
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.
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.
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/>.
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>
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>
41 #include <linux/kexec.h>
42 #include <linux/crash_dump.h>
45 #include <asm/fixmap.h>
46 #include <asm/kasan.h>
47 #include <asm/kernel-pgtable.h>
48 #include <asm/memory.h>
50 #include <asm/sections.h>
51 #include <asm/setup.h>
52 #include <asm/sizes.h>
54 #include <asm/alternative.h>
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.
62 s64 memstart_addr __ro_after_init = -1;
63 phys_addr_t arm64_dma_phys_limit __ro_after_init;
65 #ifdef CONFIG_BLK_DEV_INITRD
66 static int __init early_initrd(char *p)
68 unsigned long start, size;
71 start = memparse(p, &endp);
73 size = memparse(endp + 1, NULL);
76 initrd_end = start + size;
80 early_param("initrd", early_initrd);
83 #ifdef CONFIG_KEXEC_CORE
85 * reserve_crashkernel() - reserves memory for crash kernel
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.
91 static void __init reserve_crashkernel(void)
93 unsigned long long crash_base, crash_size;
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)
102 crash_size = PAGE_ALIGN(crash_size);
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,
108 if (crash_base == 0) {
109 pr_warn("cannot allocate crashkernel (size:0x%llx)\n",
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");
120 if (memblock_is_region_reserved(crash_base, crash_size)) {
121 pr_warn("cannot reserve crashkernel: region overlaps reserved memory\n");
125 if (!IS_ALIGNED(crash_base, SZ_2M)) {
126 pr_warn("cannot reserve crashkernel: base address is not 2MB aligned\n");
130 memblock_reserve(crash_base, crash_size);
132 pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n",
133 crash_base, crash_base + crash_size, crash_size >> 20);
135 crashk_res.start = crash_base;
136 crashk_res.end = crash_base + crash_size - 1;
139 static void __init kexec_reserve_crashkres_pages(void)
141 #ifdef CONFIG_HIBERNATION
149 * To reduce the size of hibernation image, all the pages are
150 * marked as Reserved initially.
152 for (addr = crashk_res.start; addr < (crashk_res.end + 1);
154 page = phys_to_page(addr);
155 SetPageReserved(page);
160 static void __init reserve_crashkernel(void)
164 static void __init kexec_reserve_crashkres_pages(void)
167 #endif /* CONFIG_KEXEC_CORE */
169 #ifdef CONFIG_CRASH_DUMP
170 static int __init early_init_dt_scan_elfcorehdr(unsigned long node,
171 const char *uname, int depth, void *data)
176 if (depth != 1 || strcmp(uname, "chosen") != 0)
179 reg = of_get_flat_dt_prop(node, "linux,elfcorehdr", &len);
180 if (!reg || (len < (dt_root_addr_cells + dt_root_size_cells)))
183 elfcorehdr_addr = dt_mem_next_cell(dt_root_addr_cells, ®);
184 elfcorehdr_size = dt_mem_next_cell(dt_root_size_cells, ®);
190 * reserve_elfcorehdr() - reserves memory for elf core header
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.
197 static void __init reserve_elfcorehdr(void)
199 of_scan_flat_dt(early_init_dt_scan_elfcorehdr, NULL);
201 if (!elfcorehdr_size)
204 if (memblock_is_region_reserved(elfcorehdr_addr, elfcorehdr_size)) {
205 pr_warn("elfcorehdr is overlapped\n");
209 memblock_reserve(elfcorehdr_addr, elfcorehdr_size);
211 pr_info("Reserving %lldKB of memory at 0x%llx for elfcorehdr\n",
212 elfcorehdr_size >> 10, elfcorehdr_addr);
215 static void __init reserve_elfcorehdr(void)
218 #endif /* CONFIG_CRASH_DUMP */
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
224 static phys_addr_t __init max_zone_dma_phys(void)
226 phys_addr_t offset = memblock_start_of_DRAM() & GENMASK_ULL(63, 32);
227 return min(offset + (1ULL << 32), memblock_end_of_DRAM());
232 static void __init zone_sizes_init(unsigned long min, unsigned long max)
234 unsigned long max_zone_pfns[MAX_NR_ZONES] = {0};
236 #ifdef CONFIG_ZONE_DMA32
237 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(max_zone_dma_phys());
239 max_zone_pfns[ZONE_NORMAL] = max;
241 free_area_init_nodes(max_zone_pfns);
246 static void __init zone_sizes_init(unsigned long min, unsigned long max)
248 struct memblock_region *reg;
249 unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES];
250 unsigned long max_dma = min;
252 memset(zone_size, 0, sizeof(zone_size));
254 /* 4GB maximum for 32-bit only capable devices */
255 #ifdef CONFIG_ZONE_DMA32
256 max_dma = PFN_DOWN(arm64_dma_phys_limit);
257 zone_size[ZONE_DMA32] = max_dma - min;
259 zone_size[ZONE_NORMAL] = max - max_dma;
261 memcpy(zhole_size, zone_size, sizeof(zhole_size));
263 for_each_memblock(memory, reg) {
264 unsigned long start = memblock_region_memory_base_pfn(reg);
265 unsigned long end = memblock_region_memory_end_pfn(reg);
270 #ifdef CONFIG_ZONE_DMA32
271 if (start < max_dma) {
272 unsigned long dma_end = min(end, max_dma);
273 zhole_size[ZONE_DMA32] -= dma_end - start;
277 unsigned long normal_end = min(end, max);
278 unsigned long normal_start = max(start, max_dma);
279 zhole_size[ZONE_NORMAL] -= normal_end - normal_start;
283 free_area_init_node(0, zone_size, min, zhole_size);
286 #endif /* CONFIG_NUMA */
288 #ifdef CONFIG_HAVE_ARCH_PFN_VALID
289 int pfn_valid(unsigned long pfn)
291 phys_addr_t addr = pfn << PAGE_SHIFT;
293 if ((addr >> PAGE_SHIFT) != pfn)
295 return memblock_is_map_memory(addr);
297 EXPORT_SYMBOL(pfn_valid);
300 #ifndef CONFIG_SPARSEMEM
301 static void __init arm64_memory_present(void)
305 static void __init arm64_memory_present(void)
307 struct memblock_region *reg;
309 for_each_memblock(memory, reg) {
310 int nid = memblock_get_region_node(reg);
312 memory_present(nid, memblock_region_memory_base_pfn(reg),
313 memblock_region_memory_end_pfn(reg));
318 static phys_addr_t memory_limit = PHYS_ADDR_MAX;
321 * Limit the memory size that was specified via FDT.
323 static int __init early_mem(char *p)
328 memory_limit = memparse(p, &p) & PAGE_MASK;
329 pr_notice("Memory limited to %lldMB\n", memory_limit >> 20);
333 early_param("mem", early_mem);
335 static int __init early_init_dt_scan_usablemem(unsigned long node,
336 const char *uname, int depth, void *data)
338 struct memblock_region *usablemem = data;
342 if (depth != 1 || strcmp(uname, "chosen") != 0)
345 reg = of_get_flat_dt_prop(node, "linux,usable-memory-range", &len);
346 if (!reg || (len < (dt_root_addr_cells + dt_root_size_cells)))
349 usablemem->base = dt_mem_next_cell(dt_root_addr_cells, ®);
350 usablemem->size = dt_mem_next_cell(dt_root_size_cells, ®);
355 static void __init fdt_enforce_memory_region(void)
357 struct memblock_region reg = {
361 of_scan_flat_dt(early_init_dt_scan_usablemem, ®);
364 memblock_cap_memory_range(reg.base, reg.size);
367 void __init arm64_memblock_init(void)
369 const s64 linear_region_size = -(s64)PAGE_OFFSET;
371 /* Handle linux,usable-memory-range property */
372 fdt_enforce_memory_region();
374 /* Remove memory above our supported physical address size */
375 memblock_remove(1ULL << PHYS_MASK_SHIFT, ULLONG_MAX);
378 * Ensure that the linear region takes up exactly half of the kernel
379 * virtual address space. This way, we can distinguish a linear address
380 * from a kernel/module/vmalloc address by testing a single bit.
382 BUILD_BUG_ON(linear_region_size != BIT(VA_BITS - 1));
385 * Select a suitable value for the base of physical memory.
387 memstart_addr = round_down(memblock_start_of_DRAM(),
388 ARM64_MEMSTART_ALIGN);
391 * Remove the memory that we will not be able to cover with the
392 * linear mapping. Take care not to clip the kernel which may be
395 memblock_remove(max_t(u64, memstart_addr + linear_region_size,
396 __pa_symbol(_end)), ULLONG_MAX);
397 if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {
398 /* ensure that memstart_addr remains sufficiently aligned */
399 memstart_addr = round_up(memblock_end_of_DRAM() - linear_region_size,
400 ARM64_MEMSTART_ALIGN);
401 memblock_remove(0, memstart_addr);
405 * Apply the memory limit if it was set. Since the kernel may be loaded
406 * high up in memory, add back the kernel region that must be accessible
407 * via the linear mapping.
409 if (memory_limit != PHYS_ADDR_MAX) {
410 memblock_mem_limit_remove_map(memory_limit);
411 memblock_add(__pa_symbol(_text), (u64)(_end - _text));
414 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && initrd_start) {
416 * Add back the memory we just removed if it results in the
417 * initrd to become inaccessible via the linear mapping.
418 * Otherwise, this is a no-op
420 u64 base = initrd_start & PAGE_MASK;
421 u64 size = PAGE_ALIGN(initrd_end) - base;
424 * We can only add back the initrd memory if we don't end up
425 * with more memory than we can address via the linear mapping.
426 * It is up to the bootloader to position the kernel and the
427 * initrd reasonably close to each other (i.e., within 32 GB of
428 * each other) so that all granule/#levels combinations can
429 * always access both.
431 if (WARN(base < memblock_start_of_DRAM() ||
432 base + size > memblock_start_of_DRAM() +
434 "initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
437 memblock_remove(base, size); /* clear MEMBLOCK_ flags */
438 memblock_add(base, size);
439 memblock_reserve(base, size);
443 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
444 extern u16 memstart_offset_seed;
445 u64 range = linear_region_size -
446 (memblock_end_of_DRAM() - memblock_start_of_DRAM());
449 * If the size of the linear region exceeds, by a sufficient
450 * margin, the size of the region that the available physical
451 * memory spans, randomize the linear region as well.
453 if (memstart_offset_seed > 0 && range >= ARM64_MEMSTART_ALIGN) {
454 range /= ARM64_MEMSTART_ALIGN;
455 memstart_addr -= ARM64_MEMSTART_ALIGN *
456 ((range * memstart_offset_seed) >> 16);
461 * Register the kernel text, kernel data, initrd, and initial
462 * pagetables with memblock.
464 memblock_reserve(__pa_symbol(_text), _end - _text);
465 #ifdef CONFIG_BLK_DEV_INITRD
467 memblock_reserve(initrd_start, initrd_end - initrd_start);
469 /* the generic initrd code expects virtual addresses */
470 initrd_start = __phys_to_virt(initrd_start);
471 initrd_end = __phys_to_virt(initrd_end);
475 early_init_fdt_scan_reserved_mem();
477 /* 4GB maximum for 32-bit only capable devices */
478 if (IS_ENABLED(CONFIG_ZONE_DMA32))
479 arm64_dma_phys_limit = max_zone_dma_phys();
481 arm64_dma_phys_limit = PHYS_MASK + 1;
483 reserve_crashkernel();
485 reserve_elfcorehdr();
487 high_memory = __va(memblock_end_of_DRAM() - 1) + 1;
489 dma_contiguous_reserve(arm64_dma_phys_limit);
491 memblock_allow_resize();
494 void __init bootmem_init(void)
496 unsigned long min, max;
498 min = PFN_UP(memblock_start_of_DRAM());
499 max = PFN_DOWN(memblock_end_of_DRAM());
501 early_memtest(min << PAGE_SHIFT, max << PAGE_SHIFT);
503 max_pfn = max_low_pfn = max;
507 * Sparsemem tries to allocate bootmem in memory_present(), so must be
508 * done after the fixed reservations.
510 arm64_memory_present();
513 zone_sizes_init(min, max);
518 #ifndef CONFIG_SPARSEMEM_VMEMMAP
519 static inline void free_memmap(unsigned long start_pfn, unsigned long end_pfn)
521 struct page *start_pg, *end_pg;
522 unsigned long pg, pgend;
525 * Convert start_pfn/end_pfn to a struct page pointer.
527 start_pg = pfn_to_page(start_pfn - 1) + 1;
528 end_pg = pfn_to_page(end_pfn - 1) + 1;
531 * Convert to physical addresses, and round start upwards and end
534 pg = (unsigned long)PAGE_ALIGN(__pa(start_pg));
535 pgend = (unsigned long)__pa(end_pg) & PAGE_MASK;
538 * If there are free pages between these, free the section of the
542 free_bootmem(pg, pgend - pg);
546 * The mem_map array can get very big. Free the unused area of the memory map.
548 static void __init free_unused_memmap(void)
550 unsigned long start, prev_end = 0;
551 struct memblock_region *reg;
553 for_each_memblock(memory, reg) {
554 start = __phys_to_pfn(reg->base);
556 #ifdef CONFIG_SPARSEMEM
558 * Take care not to free memmap entries that don't exist due
559 * to SPARSEMEM sections which aren't present.
561 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
564 * If we had a previous bank, and there is a space between the
565 * current bank and the previous, free it.
567 if (prev_end && prev_end < start)
568 free_memmap(prev_end, start);
571 * Align up here since the VM subsystem insists that the
572 * memmap entries are valid from the bank end aligned to
573 * MAX_ORDER_NR_PAGES.
575 prev_end = ALIGN(__phys_to_pfn(reg->base + reg->size),
579 #ifdef CONFIG_SPARSEMEM
580 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
581 free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
584 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
587 * mem_init() marks the free areas in the mem_map and tells us how much memory
588 * is free. This is done after various parts of the system have claimed their
589 * memory after the kernel image.
591 void __init mem_init(void)
593 if (swiotlb_force == SWIOTLB_FORCE ||
594 max_pfn > (arm64_dma_phys_limit >> PAGE_SHIFT))
597 swiotlb_force = SWIOTLB_NO_FORCE;
599 set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
601 #ifndef CONFIG_SPARSEMEM_VMEMMAP
602 free_unused_memmap();
604 /* this will put all unused low memory onto the freelists */
607 kexec_reserve_crashkres_pages();
609 mem_init_print_info(NULL);
612 * Check boundaries twice: Some fundamental inconsistencies can be
613 * detected at build time already.
616 BUILD_BUG_ON(TASK_SIZE_32 > TASK_SIZE_64);
619 #ifdef CONFIG_SPARSEMEM_VMEMMAP
621 * Make sure we chose the upper bound of sizeof(struct page)
622 * correctly when sizing the VMEMMAP array.
624 BUILD_BUG_ON(sizeof(struct page) > (1 << STRUCT_PAGE_MAX_SHIFT));
627 if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) {
628 extern int sysctl_overcommit_memory;
630 * On a machine this small we won't get anywhere without
631 * overcommit, so turn it on by default.
633 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
637 void free_initmem(void)
639 free_reserved_area(lm_alias(__init_begin),
640 lm_alias(__init_end),
643 * Unmap the __init region but leave the VM area in place. This
644 * prevents the region from being reused for kernel modules, which
645 * is not supported by kallsyms.
647 unmap_kernel_range((u64)__init_begin, (u64)(__init_end - __init_begin));
650 #ifdef CONFIG_BLK_DEV_INITRD
652 static int keep_initrd __initdata;
654 void __init free_initrd_mem(unsigned long start, unsigned long end)
657 free_reserved_area((void *)start, (void *)end, 0, "initrd");
658 memblock_free(__virt_to_phys(start), end - start);
662 static int __init keepinitrd_setup(char *__unused)
668 __setup("keepinitrd", keepinitrd_setup);
672 * Dump out memory limit information on panic.
674 static int dump_mem_limit(struct notifier_block *self, unsigned long v, void *p)
676 if (memory_limit != PHYS_ADDR_MAX) {
677 pr_emerg("Memory Limit: %llu MB\n", memory_limit >> 20);
679 pr_emerg("Memory Limit: none\n");
684 static struct notifier_block mem_limit_notifier = {
685 .notifier_call = dump_mem_limit,
688 static int __init register_mem_limit_dumper(void)
690 atomic_notifier_chain_register(&panic_notifier_list,
691 &mem_limit_notifier);
694 __initcall(register_mem_limit_dumper);