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/cache.h>
26 #include <linux/mman.h>
27 #include <linux/nodemask.h>
28 #include <linux/initrd.h>
29 #include <linux/gfp.h>
30 #include <linux/memblock.h>
31 #include <linux/sort.h>
33 #include <linux/of_fdt.h>
34 #include <linux/dma-mapping.h>
35 #include <linux/dma-contiguous.h>
36 #include <linux/efi.h>
37 #include <linux/swiotlb.h>
38 #include <linux/vmalloc.h>
40 #include <linux/kexec.h>
41 #include <linux/crash_dump.h>
44 #include <asm/fixmap.h>
45 #include <asm/kasan.h>
46 #include <asm/kernel-pgtable.h>
47 #include <asm/memory.h>
49 #include <asm/sections.h>
50 #include <asm/setup.h>
51 #include <asm/sizes.h>
53 #include <asm/alternative.h>
56 * We need to be able to catch inadvertent references to memstart_addr
57 * that occur (potentially in generic code) before arm64_memblock_init()
58 * executes, which assigns it its actual value. So use a default value
59 * that cannot be mistaken for a real physical address.
61 s64 memstart_addr __ro_after_init = -1;
62 EXPORT_SYMBOL(memstart_addr);
64 phys_addr_t arm64_dma_phys_limit __ro_after_init;
66 #ifdef CONFIG_KEXEC_CORE
68 * reserve_crashkernel() - reserves memory for crash kernel
70 * This function reserves memory area given in "crashkernel=" kernel command
71 * line parameter. The memory reserved is used by dump capture kernel when
72 * primary kernel is crashing.
74 static void __init reserve_crashkernel(void)
76 unsigned long long crash_base, crash_size;
79 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
80 &crash_size, &crash_base);
81 /* no crashkernel= or invalid value specified */
82 if (ret || !crash_size)
85 crash_size = PAGE_ALIGN(crash_size);
87 if (crash_base == 0) {
88 /* Current arm64 boot protocol requires 2MB alignment */
89 crash_base = memblock_find_in_range(0, ARCH_LOW_ADDRESS_LIMIT,
91 if (crash_base == 0) {
92 pr_warn("cannot allocate crashkernel (size:0x%llx)\n",
97 /* User specifies base address explicitly. */
98 if (!memblock_is_region_memory(crash_base, crash_size)) {
99 pr_warn("cannot reserve crashkernel: region is not memory\n");
103 if (memblock_is_region_reserved(crash_base, crash_size)) {
104 pr_warn("cannot reserve crashkernel: region overlaps reserved memory\n");
108 if (!IS_ALIGNED(crash_base, SZ_2M)) {
109 pr_warn("cannot reserve crashkernel: base address is not 2MB aligned\n");
113 memblock_reserve(crash_base, crash_size);
115 pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n",
116 crash_base, crash_base + crash_size, crash_size >> 20);
118 crashk_res.start = crash_base;
119 crashk_res.end = crash_base + crash_size - 1;
122 static void __init kexec_reserve_crashkres_pages(void)
124 #ifdef CONFIG_HIBERNATION
132 * To reduce the size of hibernation image, all the pages are
133 * marked as Reserved initially.
135 for (addr = crashk_res.start; addr < (crashk_res.end + 1);
137 page = phys_to_page(addr);
138 SetPageReserved(page);
143 static void __init reserve_crashkernel(void)
147 static void __init kexec_reserve_crashkres_pages(void)
150 #endif /* CONFIG_KEXEC_CORE */
152 #ifdef CONFIG_CRASH_DUMP
153 static int __init early_init_dt_scan_elfcorehdr(unsigned long node,
154 const char *uname, int depth, void *data)
159 if (depth != 1 || strcmp(uname, "chosen") != 0)
162 reg = of_get_flat_dt_prop(node, "linux,elfcorehdr", &len);
163 if (!reg || (len < (dt_root_addr_cells + dt_root_size_cells)))
166 elfcorehdr_addr = dt_mem_next_cell(dt_root_addr_cells, ®);
167 elfcorehdr_size = dt_mem_next_cell(dt_root_size_cells, ®);
173 * reserve_elfcorehdr() - reserves memory for elf core header
175 * This function reserves the memory occupied by an elf core header
176 * described in the device tree. This region contains all the
177 * information about primary kernel's core image and is used by a dump
178 * capture kernel to access the system memory on primary kernel.
180 static void __init reserve_elfcorehdr(void)
182 of_scan_flat_dt(early_init_dt_scan_elfcorehdr, NULL);
184 if (!elfcorehdr_size)
187 if (memblock_is_region_reserved(elfcorehdr_addr, elfcorehdr_size)) {
188 pr_warn("elfcorehdr is overlapped\n");
192 memblock_reserve(elfcorehdr_addr, elfcorehdr_size);
194 pr_info("Reserving %lldKB of memory at 0x%llx for elfcorehdr\n",
195 elfcorehdr_size >> 10, elfcorehdr_addr);
198 static void __init reserve_elfcorehdr(void)
201 #endif /* CONFIG_CRASH_DUMP */
203 * Return the maximum physical address for ZONE_DMA32 (DMA_BIT_MASK(32)). It
204 * currently assumes that for memory starting above 4G, 32-bit devices will
207 static phys_addr_t __init max_zone_dma_phys(void)
209 phys_addr_t offset = memblock_start_of_DRAM() & GENMASK_ULL(63, 32);
210 return min(offset + (1ULL << 32), memblock_end_of_DRAM());
215 static void __init zone_sizes_init(unsigned long min, unsigned long max)
217 unsigned long max_zone_pfns[MAX_NR_ZONES] = {0};
219 if (IS_ENABLED(CONFIG_ZONE_DMA32))
220 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(max_zone_dma_phys());
221 max_zone_pfns[ZONE_NORMAL] = max;
223 free_area_init_nodes(max_zone_pfns);
228 static void __init zone_sizes_init(unsigned long min, unsigned long max)
230 struct memblock_region *reg;
231 unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES];
232 unsigned long max_dma = min;
234 memset(zone_size, 0, sizeof(zone_size));
236 /* 4GB maximum for 32-bit only capable devices */
237 #ifdef CONFIG_ZONE_DMA32
238 max_dma = PFN_DOWN(arm64_dma_phys_limit);
239 zone_size[ZONE_DMA32] = max_dma - min;
241 zone_size[ZONE_NORMAL] = max - max_dma;
243 memcpy(zhole_size, zone_size, sizeof(zhole_size));
245 for_each_memblock(memory, reg) {
246 unsigned long start = memblock_region_memory_base_pfn(reg);
247 unsigned long end = memblock_region_memory_end_pfn(reg);
252 #ifdef CONFIG_ZONE_DMA32
253 if (start < max_dma) {
254 unsigned long dma_end = min(end, max_dma);
255 zhole_size[ZONE_DMA32] -= dma_end - start;
259 unsigned long normal_end = min(end, max);
260 unsigned long normal_start = max(start, max_dma);
261 zhole_size[ZONE_NORMAL] -= normal_end - normal_start;
265 free_area_init_node(0, zone_size, min, zhole_size);
268 #endif /* CONFIG_NUMA */
270 int pfn_valid(unsigned long pfn)
272 phys_addr_t addr = pfn << PAGE_SHIFT;
274 if ((addr >> PAGE_SHIFT) != pfn)
277 #ifdef CONFIG_SPARSEMEM
278 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
281 if (!valid_section(__nr_to_section(pfn_to_section_nr(pfn))))
284 return memblock_is_map_memory(addr);
286 EXPORT_SYMBOL(pfn_valid);
288 #ifndef CONFIG_SPARSEMEM
289 static void __init arm64_memory_present(void)
293 static void __init arm64_memory_present(void)
295 struct memblock_region *reg;
297 for_each_memblock(memory, reg) {
298 int nid = memblock_get_region_node(reg);
300 memory_present(nid, memblock_region_memory_base_pfn(reg),
301 memblock_region_memory_end_pfn(reg));
306 static phys_addr_t memory_limit = PHYS_ADDR_MAX;
309 * Limit the memory size that was specified via FDT.
311 static int __init early_mem(char *p)
316 memory_limit = memparse(p, &p) & PAGE_MASK;
317 pr_notice("Memory limited to %lldMB\n", memory_limit >> 20);
321 early_param("mem", early_mem);
323 static int __init early_init_dt_scan_usablemem(unsigned long node,
324 const char *uname, int depth, void *data)
326 struct memblock_region *usablemem = data;
330 if (depth != 1 || strcmp(uname, "chosen") != 0)
333 reg = of_get_flat_dt_prop(node, "linux,usable-memory-range", &len);
334 if (!reg || (len < (dt_root_addr_cells + dt_root_size_cells)))
337 usablemem->base = dt_mem_next_cell(dt_root_addr_cells, ®);
338 usablemem->size = dt_mem_next_cell(dt_root_size_cells, ®);
343 static void __init fdt_enforce_memory_region(void)
345 struct memblock_region reg = {
349 of_scan_flat_dt(early_init_dt_scan_usablemem, ®);
352 memblock_cap_memory_range(reg.base, reg.size);
355 void __init arm64_memblock_init(void)
357 const s64 linear_region_size = -(s64)PAGE_OFFSET;
359 /* Handle linux,usable-memory-range property */
360 fdt_enforce_memory_region();
362 /* Remove memory above our supported physical address size */
363 memblock_remove(1ULL << PHYS_MASK_SHIFT, ULLONG_MAX);
366 * Ensure that the linear region takes up exactly half of the kernel
367 * virtual address space. This way, we can distinguish a linear address
368 * from a kernel/module/vmalloc address by testing a single bit.
370 BUILD_BUG_ON(linear_region_size != BIT(VA_BITS - 1));
373 * Select a suitable value for the base of physical memory.
375 memstart_addr = round_down(memblock_start_of_DRAM(),
376 ARM64_MEMSTART_ALIGN);
379 * Remove the memory that we will not be able to cover with the
380 * linear mapping. Take care not to clip the kernel which may be
383 memblock_remove(max_t(u64, memstart_addr + linear_region_size,
384 __pa_symbol(_end)), ULLONG_MAX);
385 if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {
386 /* ensure that memstart_addr remains sufficiently aligned */
387 memstart_addr = round_up(memblock_end_of_DRAM() - linear_region_size,
388 ARM64_MEMSTART_ALIGN);
389 memblock_remove(0, memstart_addr);
393 * Apply the memory limit if it was set. Since the kernel may be loaded
394 * high up in memory, add back the kernel region that must be accessible
395 * via the linear mapping.
397 if (memory_limit != PHYS_ADDR_MAX) {
398 memblock_mem_limit_remove_map(memory_limit);
399 memblock_add(__pa_symbol(_text), (u64)(_end - _text));
402 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
404 * Add back the memory we just removed if it results in the
405 * initrd to become inaccessible via the linear mapping.
406 * Otherwise, this is a no-op
408 u64 base = phys_initrd_start & PAGE_MASK;
409 u64 size = PAGE_ALIGN(phys_initrd_size);
412 * We can only add back the initrd memory if we don't end up
413 * with more memory than we can address via the linear mapping.
414 * It is up to the bootloader to position the kernel and the
415 * initrd reasonably close to each other (i.e., within 32 GB of
416 * each other) so that all granule/#levels combinations can
417 * always access both.
419 if (WARN(base < memblock_start_of_DRAM() ||
420 base + size > memblock_start_of_DRAM() +
422 "initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
425 memblock_remove(base, size); /* clear MEMBLOCK_ flags */
426 memblock_add(base, size);
427 memblock_reserve(base, size);
431 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
432 extern u16 memstart_offset_seed;
433 u64 range = linear_region_size -
434 (memblock_end_of_DRAM() - memblock_start_of_DRAM());
437 * If the size of the linear region exceeds, by a sufficient
438 * margin, the size of the region that the available physical
439 * memory spans, randomize the linear region as well.
441 if (memstart_offset_seed > 0 && range >= ARM64_MEMSTART_ALIGN) {
442 range /= ARM64_MEMSTART_ALIGN;
443 memstart_addr -= ARM64_MEMSTART_ALIGN *
444 ((range * memstart_offset_seed) >> 16);
449 * Register the kernel text, kernel data, initrd, and initial
450 * pagetables with memblock.
452 memblock_reserve(__pa_symbol(_text), _end - _text);
453 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
454 /* the generic initrd code expects virtual addresses */
455 initrd_start = __phys_to_virt(phys_initrd_start);
456 initrd_end = initrd_start + phys_initrd_size;
459 early_init_fdt_scan_reserved_mem();
461 /* 4GB maximum for 32-bit only capable devices */
462 if (IS_ENABLED(CONFIG_ZONE_DMA32))
463 arm64_dma_phys_limit = max_zone_dma_phys();
465 arm64_dma_phys_limit = PHYS_MASK + 1;
467 reserve_crashkernel();
469 reserve_elfcorehdr();
471 high_memory = __va(memblock_end_of_DRAM() - 1) + 1;
473 dma_contiguous_reserve(arm64_dma_phys_limit);
476 void __init bootmem_init(void)
478 unsigned long min, max;
480 min = PFN_UP(memblock_start_of_DRAM());
481 max = PFN_DOWN(memblock_end_of_DRAM());
483 early_memtest(min << PAGE_SHIFT, max << PAGE_SHIFT);
485 max_pfn = max_low_pfn = max;
489 * Sparsemem tries to allocate bootmem in memory_present(), so must be
490 * done after the fixed reservations.
492 arm64_memory_present();
495 zone_sizes_init(min, max);
500 #ifndef CONFIG_SPARSEMEM_VMEMMAP
501 static inline void free_memmap(unsigned long start_pfn, unsigned long end_pfn)
503 struct page *start_pg, *end_pg;
504 unsigned long pg, pgend;
507 * Convert start_pfn/end_pfn to a struct page pointer.
509 start_pg = pfn_to_page(start_pfn - 1) + 1;
510 end_pg = pfn_to_page(end_pfn - 1) + 1;
513 * Convert to physical addresses, and round start upwards and end
516 pg = (unsigned long)PAGE_ALIGN(__pa(start_pg));
517 pgend = (unsigned long)__pa(end_pg) & PAGE_MASK;
520 * If there are free pages between these, free the section of the
524 memblock_free(pg, pgend - pg);
528 * The mem_map array can get very big. Free the unused area of the memory map.
530 static void __init free_unused_memmap(void)
532 unsigned long start, prev_end = 0;
533 struct memblock_region *reg;
535 for_each_memblock(memory, reg) {
536 start = __phys_to_pfn(reg->base);
538 #ifdef CONFIG_SPARSEMEM
540 * Take care not to free memmap entries that don't exist due
541 * to SPARSEMEM sections which aren't present.
543 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
546 * If we had a previous bank, and there is a space between the
547 * current bank and the previous, free it.
549 if (prev_end && prev_end < start)
550 free_memmap(prev_end, start);
553 * Align up here since the VM subsystem insists that the
554 * memmap entries are valid from the bank end aligned to
555 * MAX_ORDER_NR_PAGES.
557 prev_end = ALIGN(__phys_to_pfn(reg->base + reg->size),
561 #ifdef CONFIG_SPARSEMEM
562 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
563 free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
566 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
569 * mem_init() marks the free areas in the mem_map and tells us how much memory
570 * is free. This is done after various parts of the system have claimed their
571 * memory after the kernel image.
573 void __init mem_init(void)
575 if (swiotlb_force == SWIOTLB_FORCE ||
576 max_pfn > (arm64_dma_phys_limit >> PAGE_SHIFT))
579 swiotlb_force = SWIOTLB_NO_FORCE;
581 set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
583 #ifndef CONFIG_SPARSEMEM_VMEMMAP
584 free_unused_memmap();
586 /* this will put all unused low memory onto the freelists */
589 kexec_reserve_crashkres_pages();
591 mem_init_print_info(NULL);
594 * Check boundaries twice: Some fundamental inconsistencies can be
595 * detected at build time already.
598 BUILD_BUG_ON(TASK_SIZE_32 > DEFAULT_MAP_WINDOW_64);
601 if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) {
602 extern int sysctl_overcommit_memory;
604 * On a machine this small we won't get anywhere without
605 * overcommit, so turn it on by default.
607 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
611 void free_initmem(void)
613 free_reserved_area(lm_alias(__init_begin),
614 lm_alias(__init_end),
617 * Unmap the __init region but leave the VM area in place. This
618 * prevents the region from being reused for kernel modules, which
619 * is not supported by kallsyms.
621 unmap_kernel_range((u64)__init_begin, (u64)(__init_end - __init_begin));
624 #ifdef CONFIG_BLK_DEV_INITRD
626 static int keep_initrd __initdata;
628 void __init free_initrd_mem(unsigned long start, unsigned long end)
631 free_reserved_area((void *)start, (void *)end, 0, "initrd");
632 memblock_free(__virt_to_phys(start), end - start);
636 static int __init keepinitrd_setup(char *__unused)
642 __setup("keepinitrd", keepinitrd_setup);
646 * Dump out memory limit information on panic.
648 static int dump_mem_limit(struct notifier_block *self, unsigned long v, void *p)
650 if (memory_limit != PHYS_ADDR_MAX) {
651 pr_emerg("Memory Limit: %llu MB\n", memory_limit >> 20);
653 pr_emerg("Memory Limit: none\n");
658 static struct notifier_block mem_limit_notifier = {
659 .notifier_call = dump_mem_limit,
662 static int __init register_mem_limit_dumper(void)
664 atomic_notifier_chain_register(&panic_notifier_list,
665 &mem_limit_notifier);
668 __initcall(register_mem_limit_dumper);