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_DMA (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 if (IS_ENABLED(CONFIG_ZONE_DMA
))
237 max_zone_pfns
[ZONE_DMA
] = PFN_DOWN(max_zone_dma_phys());
238 max_zone_pfns
[ZONE_NORMAL
] = max
;
240 free_area_init_nodes(max_zone_pfns
);
245 static void __init
zone_sizes_init(unsigned long min
, unsigned long max
)
247 struct memblock_region
*reg
;
248 unsigned long zone_size
[MAX_NR_ZONES
], zhole_size
[MAX_NR_ZONES
];
249 unsigned long max_dma
= min
;
251 memset(zone_size
, 0, sizeof(zone_size
));
253 /* 4GB maximum for 32-bit only capable devices */
254 #ifdef CONFIG_ZONE_DMA
255 max_dma
= PFN_DOWN(arm64_dma_phys_limit
);
256 zone_size
[ZONE_DMA
] = max_dma
- min
;
258 zone_size
[ZONE_NORMAL
] = max
- max_dma
;
260 memcpy(zhole_size
, zone_size
, sizeof(zhole_size
));
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
);
269 #ifdef CONFIG_ZONE_DMA
270 if (start
< max_dma
) {
271 unsigned long dma_end
= min(end
, max_dma
);
272 zhole_size
[ZONE_DMA
] -= dma_end
- start
;
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
;
282 free_area_init_node(0, zone_size
, min
, zhole_size
);
285 #endif /* CONFIG_NUMA */
287 #ifdef CONFIG_HAVE_ARCH_PFN_VALID
288 int pfn_valid(unsigned long pfn
)
290 return memblock_is_map_memory(pfn
<< PAGE_SHIFT
);
292 EXPORT_SYMBOL(pfn_valid
);
295 #ifndef CONFIG_SPARSEMEM
296 static void __init
arm64_memory_present(void)
300 static void __init
arm64_memory_present(void)
302 struct memblock_region
*reg
;
304 for_each_memblock(memory
, reg
) {
305 int nid
= memblock_get_region_node(reg
);
307 memory_present(nid
, memblock_region_memory_base_pfn(reg
),
308 memblock_region_memory_end_pfn(reg
));
313 static phys_addr_t memory_limit
= (phys_addr_t
)ULLONG_MAX
;
316 * Limit the memory size that was specified via FDT.
318 static int __init
early_mem(char *p
)
323 memory_limit
= memparse(p
, &p
) & PAGE_MASK
;
324 pr_notice("Memory limited to %lldMB\n", memory_limit
>> 20);
328 early_param("mem", early_mem
);
330 static int __init
early_init_dt_scan_usablemem(unsigned long node
,
331 const char *uname
, int depth
, void *data
)
333 struct memblock_region
*usablemem
= data
;
337 if (depth
!= 1 || strcmp(uname
, "chosen") != 0)
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
)))
344 usablemem
->base
= dt_mem_next_cell(dt_root_addr_cells
, ®
);
345 usablemem
->size
= dt_mem_next_cell(dt_root_size_cells
, ®
);
350 static void __init
fdt_enforce_memory_region(void)
352 struct memblock_region reg
= {
356 of_scan_flat_dt(early_init_dt_scan_usablemem
, ®
);
359 memblock_cap_memory_range(reg
.base
, reg
.size
);
362 void __init
arm64_memblock_init(void)
364 const s64 linear_region_size
= -(s64
)PAGE_OFFSET
;
366 /* Handle linux,usable-memory-range property */
367 fdt_enforce_memory_region();
370 * Ensure that the linear region takes up exactly half of the kernel
371 * virtual address space. This way, we can distinguish a linear address
372 * from a kernel/module/vmalloc address by testing a single bit.
374 BUILD_BUG_ON(linear_region_size
!= BIT(VA_BITS
- 1));
377 * Select a suitable value for the base of physical memory.
379 memstart_addr
= round_down(memblock_start_of_DRAM(),
380 ARM64_MEMSTART_ALIGN
);
383 * Remove the memory that we will not be able to cover with the
384 * linear mapping. Take care not to clip the kernel which may be
387 memblock_remove(max_t(u64
, memstart_addr
+ linear_region_size
,
388 __pa_symbol(_end
)), ULLONG_MAX
);
389 if (memstart_addr
+ linear_region_size
< memblock_end_of_DRAM()) {
390 /* ensure that memstart_addr remains sufficiently aligned */
391 memstart_addr
= round_up(memblock_end_of_DRAM() - linear_region_size
,
392 ARM64_MEMSTART_ALIGN
);
393 memblock_remove(0, memstart_addr
);
397 * Apply the memory limit if it was set. Since the kernel may be loaded
398 * high up in memory, add back the kernel region that must be accessible
399 * via the linear mapping.
401 if (memory_limit
!= (phys_addr_t
)ULLONG_MAX
) {
402 memblock_mem_limit_remove_map(memory_limit
);
403 memblock_add(__pa_symbol(_text
), (u64
)(_end
- _text
));
406 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD
) && initrd_start
) {
408 * Add back the memory we just removed if it results in the
409 * initrd to become inaccessible via the linear mapping.
410 * Otherwise, this is a no-op
412 u64 base
= initrd_start
& PAGE_MASK
;
413 u64 size
= PAGE_ALIGN(initrd_end
) - base
;
416 * We can only add back the initrd memory if we don't end up
417 * with more memory than we can address via the linear mapping.
418 * It is up to the bootloader to position the kernel and the
419 * initrd reasonably close to each other (i.e., within 32 GB of
420 * each other) so that all granule/#levels combinations can
421 * always access both.
423 if (WARN(base
< memblock_start_of_DRAM() ||
424 base
+ size
> memblock_start_of_DRAM() +
426 "initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
429 memblock_remove(base
, size
); /* clear MEMBLOCK_ flags */
430 memblock_add(base
, size
);
431 memblock_reserve(base
, size
);
435 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE
)) {
436 extern u16 memstart_offset_seed
;
437 u64 range
= linear_region_size
-
438 (memblock_end_of_DRAM() - memblock_start_of_DRAM());
441 * If the size of the linear region exceeds, by a sufficient
442 * margin, the size of the region that the available physical
443 * memory spans, randomize the linear region as well.
445 if (memstart_offset_seed
> 0 && range
>= ARM64_MEMSTART_ALIGN
) {
446 range
= range
/ ARM64_MEMSTART_ALIGN
+ 1;
447 memstart_addr
-= ARM64_MEMSTART_ALIGN
*
448 ((range
* memstart_offset_seed
) >> 16);
453 * Register the kernel text, kernel data, initrd, and initial
454 * pagetables with memblock.
456 memblock_reserve(__pa_symbol(_text
), _end
- _text
);
457 #ifdef CONFIG_BLK_DEV_INITRD
459 memblock_reserve(initrd_start
, initrd_end
- initrd_start
);
461 /* the generic initrd code expects virtual addresses */
462 initrd_start
= __phys_to_virt(initrd_start
);
463 initrd_end
= __phys_to_virt(initrd_end
);
467 early_init_fdt_scan_reserved_mem();
469 /* 4GB maximum for 32-bit only capable devices */
470 if (IS_ENABLED(CONFIG_ZONE_DMA
))
471 arm64_dma_phys_limit
= max_zone_dma_phys();
473 arm64_dma_phys_limit
= PHYS_MASK
+ 1;
475 reserve_crashkernel();
477 reserve_elfcorehdr();
479 high_memory
= __va(memblock_end_of_DRAM() - 1) + 1;
481 dma_contiguous_reserve(arm64_dma_phys_limit
);
483 memblock_allow_resize();
486 void __init
bootmem_init(void)
488 unsigned long min
, max
;
490 min
= PFN_UP(memblock_start_of_DRAM());
491 max
= PFN_DOWN(memblock_end_of_DRAM());
493 early_memtest(min
<< PAGE_SHIFT
, max
<< PAGE_SHIFT
);
495 max_pfn
= max_low_pfn
= max
;
499 * Sparsemem tries to allocate bootmem in memory_present(), so must be
500 * done after the fixed reservations.
502 arm64_memory_present();
505 zone_sizes_init(min
, max
);
510 #ifndef CONFIG_SPARSEMEM_VMEMMAP
511 static inline void free_memmap(unsigned long start_pfn
, unsigned long end_pfn
)
513 struct page
*start_pg
, *end_pg
;
514 unsigned long pg
, pgend
;
517 * Convert start_pfn/end_pfn to a struct page pointer.
519 start_pg
= pfn_to_page(start_pfn
- 1) + 1;
520 end_pg
= pfn_to_page(end_pfn
- 1) + 1;
523 * Convert to physical addresses, and round start upwards and end
526 pg
= (unsigned long)PAGE_ALIGN(__pa(start_pg
));
527 pgend
= (unsigned long)__pa(end_pg
) & PAGE_MASK
;
530 * If there are free pages between these, free the section of the
534 free_bootmem(pg
, pgend
- pg
);
538 * The mem_map array can get very big. Free the unused area of the memory map.
540 static void __init
free_unused_memmap(void)
542 unsigned long start
, prev_end
= 0;
543 struct memblock_region
*reg
;
545 for_each_memblock(memory
, reg
) {
546 start
= __phys_to_pfn(reg
->base
);
548 #ifdef CONFIG_SPARSEMEM
550 * Take care not to free memmap entries that don't exist due
551 * to SPARSEMEM sections which aren't present.
553 start
= min(start
, ALIGN(prev_end
, PAGES_PER_SECTION
));
556 * If we had a previous bank, and there is a space between the
557 * current bank and the previous, free it.
559 if (prev_end
&& prev_end
< start
)
560 free_memmap(prev_end
, start
);
563 * Align up here since the VM subsystem insists that the
564 * memmap entries are valid from the bank end aligned to
565 * MAX_ORDER_NR_PAGES.
567 prev_end
= ALIGN(__phys_to_pfn(reg
->base
+ reg
->size
),
571 #ifdef CONFIG_SPARSEMEM
572 if (!IS_ALIGNED(prev_end
, PAGES_PER_SECTION
))
573 free_memmap(prev_end
, ALIGN(prev_end
, PAGES_PER_SECTION
));
576 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
579 * mem_init() marks the free areas in the mem_map and tells us how much memory
580 * is free. This is done after various parts of the system have claimed their
581 * memory after the kernel image.
583 void __init
mem_init(void)
585 if (swiotlb_force
== SWIOTLB_FORCE
||
586 max_pfn
> (arm64_dma_phys_limit
>> PAGE_SHIFT
))
589 swiotlb_force
= SWIOTLB_NO_FORCE
;
591 set_max_mapnr(pfn_to_page(max_pfn
) - mem_map
);
593 #ifndef CONFIG_SPARSEMEM_VMEMMAP
594 free_unused_memmap();
596 /* this will put all unused low memory onto the freelists */
599 kexec_reserve_crashkres_pages();
601 mem_init_print_info(NULL
);
603 #define MLK(b, t) b, t, ((t) - (b)) >> 10
604 #define MLM(b, t) b, t, ((t) - (b)) >> 20
605 #define MLG(b, t) b, t, ((t) - (b)) >> 30
606 #define MLK_ROUNDUP(b, t) b, t, DIV_ROUND_UP(((t) - (b)), SZ_1K)
608 pr_notice("Virtual kernel memory layout:\n");
610 pr_notice(" kasan : 0x%16lx - 0x%16lx (%6ld GB)\n",
611 MLG(KASAN_SHADOW_START
, KASAN_SHADOW_END
));
613 pr_notice(" modules : 0x%16lx - 0x%16lx (%6ld MB)\n",
614 MLM(MODULES_VADDR
, MODULES_END
));
615 pr_notice(" vmalloc : 0x%16lx - 0x%16lx (%6ld GB)\n",
616 MLG(VMALLOC_START
, VMALLOC_END
));
617 pr_notice(" .text : 0x%p" " - 0x%p" " (%6ld KB)\n",
618 MLK_ROUNDUP(_text
, _etext
));
619 pr_notice(" .rodata : 0x%p" " - 0x%p" " (%6ld KB)\n",
620 MLK_ROUNDUP(__start_rodata
, __init_begin
));
621 pr_notice(" .init : 0x%p" " - 0x%p" " (%6ld KB)\n",
622 MLK_ROUNDUP(__init_begin
, __init_end
));
623 pr_notice(" .data : 0x%p" " - 0x%p" " (%6ld KB)\n",
624 MLK_ROUNDUP(_sdata
, _edata
));
625 pr_notice(" .bss : 0x%p" " - 0x%p" " (%6ld KB)\n",
626 MLK_ROUNDUP(__bss_start
, __bss_stop
));
627 pr_notice(" fixed : 0x%16lx - 0x%16lx (%6ld KB)\n",
628 MLK(FIXADDR_START
, FIXADDR_TOP
));
629 pr_notice(" PCI I/O : 0x%16lx - 0x%16lx (%6ld MB)\n",
630 MLM(PCI_IO_START
, PCI_IO_END
));
631 #ifdef CONFIG_SPARSEMEM_VMEMMAP
632 pr_notice(" vmemmap : 0x%16lx - 0x%16lx (%6ld GB maximum)\n",
633 MLG(VMEMMAP_START
, VMEMMAP_START
+ VMEMMAP_SIZE
));
634 pr_notice(" 0x%16lx - 0x%16lx (%6ld MB actual)\n",
635 MLM((unsigned long)phys_to_page(memblock_start_of_DRAM()),
636 (unsigned long)virt_to_page(high_memory
)));
638 pr_notice(" memory : 0x%16lx - 0x%16lx (%6ld MB)\n",
639 MLM(__phys_to_virt(memblock_start_of_DRAM()),
640 (unsigned long)high_memory
));
647 * Check boundaries twice: Some fundamental inconsistencies can be
648 * detected at build time already.
651 BUILD_BUG_ON(TASK_SIZE_32
> TASK_SIZE_64
);
655 * Make sure we chose the upper bound of sizeof(struct page)
658 BUILD_BUG_ON(sizeof(struct page
) > (1 << STRUCT_PAGE_MAX_SHIFT
));
660 if (PAGE_SIZE
>= 16384 && get_num_physpages() <= 128) {
661 extern int sysctl_overcommit_memory
;
663 * On a machine this small we won't get anywhere without
664 * overcommit, so turn it on by default.
666 sysctl_overcommit_memory
= OVERCOMMIT_ALWAYS
;
670 void free_initmem(void)
672 free_reserved_area(lm_alias(__init_begin
),
673 lm_alias(__init_end
),
676 * Unmap the __init region but leave the VM area in place. This
677 * prevents the region from being reused for kernel modules, which
678 * is not supported by kallsyms.
680 unmap_kernel_range((u64
)__init_begin
, (u64
)(__init_end
- __init_begin
));
683 #ifdef CONFIG_BLK_DEV_INITRD
685 static int keep_initrd __initdata
;
687 void __init
free_initrd_mem(unsigned long start
, unsigned long end
)
690 free_reserved_area((void *)start
, (void *)end
, 0, "initrd");
693 static int __init
keepinitrd_setup(char *__unused
)
699 __setup("keepinitrd", keepinitrd_setup
);
703 * Dump out memory limit information on panic.
705 static int dump_mem_limit(struct notifier_block
*self
, unsigned long v
, void *p
)
707 if (memory_limit
!= (phys_addr_t
)ULLONG_MAX
) {
708 pr_emerg("Memory Limit: %llu MB\n", memory_limit
>> 20);
710 pr_emerg("Memory Limit: none\n");
715 static struct notifier_block mem_limit_notifier
= {
716 .notifier_call
= dump_mem_limit
,
719 static int __init
register_mem_limit_dumper(void)
721 atomic_notifier_chain_register(&panic_notifier_list
,
722 &mem_limit_notifier
);
725 __initcall(register_mem_limit_dumper
);