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>
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 phys_addr_t arm64_dma_phys_limit __ro_after_init
;
64 #ifdef CONFIG_BLK_DEV_INITRD
65 static int __init
early_initrd(char *p
)
67 unsigned long start
, size
;
70 start
= memparse(p
, &endp
);
72 size
= memparse(endp
+ 1, NULL
);
75 initrd_end
= start
+ size
;
79 early_param("initrd", early_initrd
);
82 #ifdef CONFIG_KEXEC_CORE
84 * reserve_crashkernel() - reserves memory for crash kernel
86 * This function reserves memory area given in "crashkernel=" kernel command
87 * line parameter. The memory reserved is used by dump capture kernel when
88 * primary kernel is crashing.
90 static void __init
reserve_crashkernel(void)
92 unsigned long long crash_base
, crash_size
;
95 ret
= parse_crashkernel(boot_command_line
, memblock_phys_mem_size(),
96 &crash_size
, &crash_base
);
97 /* no crashkernel= or invalid value specified */
98 if (ret
|| !crash_size
)
101 crash_size
= PAGE_ALIGN(crash_size
);
103 if (crash_base
== 0) {
104 /* Current arm64 boot protocol requires 2MB alignment */
105 crash_base
= memblock_find_in_range(0, ARCH_LOW_ADDRESS_LIMIT
,
107 if (crash_base
== 0) {
108 pr_warn("cannot allocate crashkernel (size:0x%llx)\n",
113 /* User specifies base address explicitly. */
114 if (!memblock_is_region_memory(crash_base
, crash_size
)) {
115 pr_warn("cannot reserve crashkernel: region is not memory\n");
119 if (memblock_is_region_reserved(crash_base
, crash_size
)) {
120 pr_warn("cannot reserve crashkernel: region overlaps reserved memory\n");
124 if (!IS_ALIGNED(crash_base
, SZ_2M
)) {
125 pr_warn("cannot reserve crashkernel: base address is not 2MB aligned\n");
129 memblock_reserve(crash_base
, crash_size
);
131 pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n",
132 crash_base
, crash_base
+ crash_size
, crash_size
>> 20);
134 crashk_res
.start
= crash_base
;
135 crashk_res
.end
= crash_base
+ crash_size
- 1;
138 static void __init
kexec_reserve_crashkres_pages(void)
140 #ifdef CONFIG_HIBERNATION
148 * To reduce the size of hibernation image, all the pages are
149 * marked as Reserved initially.
151 for (addr
= crashk_res
.start
; addr
< (crashk_res
.end
+ 1);
153 page
= phys_to_page(addr
);
154 SetPageReserved(page
);
159 static void __init
reserve_crashkernel(void)
163 static void __init
kexec_reserve_crashkres_pages(void)
166 #endif /* CONFIG_KEXEC_CORE */
168 #ifdef CONFIG_CRASH_DUMP
169 static int __init
early_init_dt_scan_elfcorehdr(unsigned long node
,
170 const char *uname
, int depth
, void *data
)
175 if (depth
!= 1 || strcmp(uname
, "chosen") != 0)
178 reg
= of_get_flat_dt_prop(node
, "linux,elfcorehdr", &len
);
179 if (!reg
|| (len
< (dt_root_addr_cells
+ dt_root_size_cells
)))
182 elfcorehdr_addr
= dt_mem_next_cell(dt_root_addr_cells
, ®
);
183 elfcorehdr_size
= dt_mem_next_cell(dt_root_size_cells
, ®
);
189 * reserve_elfcorehdr() - reserves memory for elf core header
191 * This function reserves the memory occupied by an elf core header
192 * described in the device tree. This region contains all the
193 * information about primary kernel's core image and is used by a dump
194 * capture kernel to access the system memory on primary kernel.
196 static void __init
reserve_elfcorehdr(void)
198 of_scan_flat_dt(early_init_dt_scan_elfcorehdr
, NULL
);
200 if (!elfcorehdr_size
)
203 if (memblock_is_region_reserved(elfcorehdr_addr
, elfcorehdr_size
)) {
204 pr_warn("elfcorehdr is overlapped\n");
208 memblock_reserve(elfcorehdr_addr
, elfcorehdr_size
);
210 pr_info("Reserving %lldKB of memory at 0x%llx for elfcorehdr\n",
211 elfcorehdr_size
>> 10, elfcorehdr_addr
);
214 static void __init
reserve_elfcorehdr(void)
217 #endif /* CONFIG_CRASH_DUMP */
219 * Return the maximum physical address for ZONE_DMA (DMA_BIT_MASK(32)). It
220 * currently assumes that for memory starting above 4G, 32-bit devices will
223 static phys_addr_t __init
max_zone_dma_phys(void)
225 phys_addr_t offset
= memblock_start_of_DRAM() & GENMASK_ULL(63, 32);
226 return min(offset
+ (1ULL << 32), memblock_end_of_DRAM());
231 static void __init
zone_sizes_init(unsigned long min
, unsigned long max
)
233 unsigned long max_zone_pfns
[MAX_NR_ZONES
] = {0};
235 if (IS_ENABLED(CONFIG_ZONE_DMA
))
236 max_zone_pfns
[ZONE_DMA
] = PFN_DOWN(max_zone_dma_phys());
237 max_zone_pfns
[ZONE_NORMAL
] = max
;
239 free_area_init_nodes(max_zone_pfns
);
244 static void __init
zone_sizes_init(unsigned long min
, unsigned long max
)
246 struct memblock_region
*reg
;
247 unsigned long zone_size
[MAX_NR_ZONES
], zhole_size
[MAX_NR_ZONES
];
248 unsigned long max_dma
= min
;
250 memset(zone_size
, 0, sizeof(zone_size
));
252 /* 4GB maximum for 32-bit only capable devices */
253 #ifdef CONFIG_ZONE_DMA
254 max_dma
= PFN_DOWN(arm64_dma_phys_limit
);
255 zone_size
[ZONE_DMA
] = max_dma
- min
;
257 zone_size
[ZONE_NORMAL
] = max
- max_dma
;
259 memcpy(zhole_size
, zone_size
, sizeof(zhole_size
));
261 for_each_memblock(memory
, reg
) {
262 unsigned long start
= memblock_region_memory_base_pfn(reg
);
263 unsigned long end
= memblock_region_memory_end_pfn(reg
);
268 #ifdef CONFIG_ZONE_DMA
269 if (start
< max_dma
) {
270 unsigned long dma_end
= min(end
, max_dma
);
271 zhole_size
[ZONE_DMA
] -= dma_end
- start
;
275 unsigned long normal_end
= min(end
, max
);
276 unsigned long normal_start
= max(start
, max_dma
);
277 zhole_size
[ZONE_NORMAL
] -= normal_end
- normal_start
;
281 free_area_init_node(0, zone_size
, min
, zhole_size
);
284 #endif /* CONFIG_NUMA */
286 #ifdef CONFIG_HAVE_ARCH_PFN_VALID
287 int pfn_valid(unsigned long pfn
)
289 return memblock_is_map_memory(pfn
<< PAGE_SHIFT
);
291 EXPORT_SYMBOL(pfn_valid
);
294 #ifndef CONFIG_SPARSEMEM
295 static void __init
arm64_memory_present(void)
299 static void __init
arm64_memory_present(void)
301 struct memblock_region
*reg
;
303 for_each_memblock(memory
, reg
) {
304 int nid
= memblock_get_region_node(reg
);
306 memory_present(nid
, memblock_region_memory_base_pfn(reg
),
307 memblock_region_memory_end_pfn(reg
));
312 static phys_addr_t memory_limit
= (phys_addr_t
)ULLONG_MAX
;
315 * Limit the memory size that was specified via FDT.
317 static int __init
early_mem(char *p
)
322 memory_limit
= memparse(p
, &p
) & PAGE_MASK
;
323 pr_notice("Memory limited to %lldMB\n", memory_limit
>> 20);
327 early_param("mem", early_mem
);
329 static int __init
early_init_dt_scan_usablemem(unsigned long node
,
330 const char *uname
, int depth
, void *data
)
332 struct memblock_region
*usablemem
= data
;
336 if (depth
!= 1 || strcmp(uname
, "chosen") != 0)
339 reg
= of_get_flat_dt_prop(node
, "linux,usable-memory-range", &len
);
340 if (!reg
|| (len
< (dt_root_addr_cells
+ dt_root_size_cells
)))
343 usablemem
->base
= dt_mem_next_cell(dt_root_addr_cells
, ®
);
344 usablemem
->size
= dt_mem_next_cell(dt_root_size_cells
, ®
);
349 static void __init
fdt_enforce_memory_region(void)
351 struct memblock_region reg
= {
355 of_scan_flat_dt(early_init_dt_scan_usablemem
, ®
);
358 memblock_cap_memory_range(reg
.base
, reg
.size
);
361 void __init
arm64_memblock_init(void)
363 const s64 linear_region_size
= -(s64
)PAGE_OFFSET
;
365 /* Handle linux,usable-memory-range property */
366 fdt_enforce_memory_region();
369 * Ensure that the linear region takes up exactly half of the kernel
370 * virtual address space. This way, we can distinguish a linear address
371 * from a kernel/module/vmalloc address by testing a single bit.
373 BUILD_BUG_ON(linear_region_size
!= BIT(VA_BITS
- 1));
376 * Select a suitable value for the base of physical memory.
378 memstart_addr
= round_down(memblock_start_of_DRAM(),
379 ARM64_MEMSTART_ALIGN
);
382 * Remove the memory that we will not be able to cover with the
383 * linear mapping. Take care not to clip the kernel which may be
386 memblock_remove(max_t(u64
, memstart_addr
+ linear_region_size
, __pa(_end
)),
388 if (memstart_addr
+ linear_region_size
< memblock_end_of_DRAM()) {
389 /* ensure that memstart_addr remains sufficiently aligned */
390 memstart_addr
= round_up(memblock_end_of_DRAM() - linear_region_size
,
391 ARM64_MEMSTART_ALIGN
);
392 memblock_remove(0, memstart_addr
);
396 * Apply the memory limit if it was set. Since the kernel may be loaded
397 * high up in memory, add back the kernel region that must be accessible
398 * via the linear mapping.
400 if (memory_limit
!= (phys_addr_t
)ULLONG_MAX
) {
401 memblock_mem_limit_remove_map(memory_limit
);
402 memblock_add(__pa(_text
), (u64
)(_end
- _text
));
405 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD
) && initrd_start
) {
407 * Add back the memory we just removed if it results in the
408 * initrd to become inaccessible via the linear mapping.
409 * Otherwise, this is a no-op
411 u64 base
= initrd_start
& PAGE_MASK
;
412 u64 size
= PAGE_ALIGN(initrd_end
) - base
;
415 * We can only add back the initrd memory if we don't end up
416 * with more memory than we can address via the linear mapping.
417 * It is up to the bootloader to position the kernel and the
418 * initrd reasonably close to each other (i.e., within 32 GB of
419 * each other) so that all granule/#levels combinations can
420 * always access both.
422 if (WARN(base
< memblock_start_of_DRAM() ||
423 base
+ size
> memblock_start_of_DRAM() +
425 "initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
428 memblock_remove(base
, size
); /* clear MEMBLOCK_ flags */
429 memblock_add(base
, size
);
430 memblock_reserve(base
, size
);
434 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE
)) {
435 extern u16 memstart_offset_seed
;
436 u64 range
= linear_region_size
-
437 (memblock_end_of_DRAM() - memblock_start_of_DRAM());
440 * If the size of the linear region exceeds, by a sufficient
441 * margin, the size of the region that the available physical
442 * memory spans, randomize the linear region as well.
444 if (memstart_offset_seed
> 0 && range
>= ARM64_MEMSTART_ALIGN
) {
445 range
= range
/ ARM64_MEMSTART_ALIGN
+ 1;
446 memstart_addr
-= ARM64_MEMSTART_ALIGN
*
447 ((range
* memstart_offset_seed
) >> 16);
452 * Register the kernel text, kernel data, initrd, and initial
453 * pagetables with memblock.
455 memblock_reserve(__pa(_text
), _end
- _text
);
456 #ifdef CONFIG_BLK_DEV_INITRD
458 memblock_reserve(initrd_start
, initrd_end
- initrd_start
);
460 /* the generic initrd code expects virtual addresses */
461 initrd_start
= __phys_to_virt(initrd_start
);
462 initrd_end
= __phys_to_virt(initrd_end
);
466 early_init_fdt_scan_reserved_mem();
468 /* 4GB maximum for 32-bit only capable devices */
469 if (IS_ENABLED(CONFIG_ZONE_DMA
))
470 arm64_dma_phys_limit
= max_zone_dma_phys();
472 arm64_dma_phys_limit
= PHYS_MASK
+ 1;
474 reserve_crashkernel();
476 reserve_elfcorehdr();
478 dma_contiguous_reserve(arm64_dma_phys_limit
);
480 memblock_allow_resize();
483 void __init
bootmem_init(void)
485 unsigned long min
, max
;
487 min
= PFN_UP(memblock_start_of_DRAM());
488 max
= PFN_DOWN(memblock_end_of_DRAM());
490 early_memtest(min
<< PAGE_SHIFT
, max
<< PAGE_SHIFT
);
492 max_pfn
= max_low_pfn
= max
;
496 * Sparsemem tries to allocate bootmem in memory_present(), so must be
497 * done after the fixed reservations.
499 arm64_memory_present();
502 zone_sizes_init(min
, max
);
504 high_memory
= __va((max
<< PAGE_SHIFT
) - 1) + 1;
508 #ifndef CONFIG_SPARSEMEM_VMEMMAP
509 static inline void free_memmap(unsigned long start_pfn
, unsigned long end_pfn
)
511 struct page
*start_pg
, *end_pg
;
512 unsigned long pg
, pgend
;
515 * Convert start_pfn/end_pfn to a struct page pointer.
517 start_pg
= pfn_to_page(start_pfn
- 1) + 1;
518 end_pg
= pfn_to_page(end_pfn
- 1) + 1;
521 * Convert to physical addresses, and round start upwards and end
524 pg
= (unsigned long)PAGE_ALIGN(__pa(start_pg
));
525 pgend
= (unsigned long)__pa(end_pg
) & PAGE_MASK
;
528 * If there are free pages between these, free the section of the
532 free_bootmem(pg
, pgend
- pg
);
536 * The mem_map array can get very big. Free the unused area of the memory map.
538 static void __init
free_unused_memmap(void)
540 unsigned long start
, prev_end
= 0;
541 struct memblock_region
*reg
;
543 for_each_memblock(memory
, reg
) {
544 start
= __phys_to_pfn(reg
->base
);
546 #ifdef CONFIG_SPARSEMEM
548 * Take care not to free memmap entries that don't exist due
549 * to SPARSEMEM sections which aren't present.
551 start
= min(start
, ALIGN(prev_end
, PAGES_PER_SECTION
));
554 * If we had a previous bank, and there is a space between the
555 * current bank and the previous, free it.
557 if (prev_end
&& prev_end
< start
)
558 free_memmap(prev_end
, start
);
561 * Align up here since the VM subsystem insists that the
562 * memmap entries are valid from the bank end aligned to
563 * MAX_ORDER_NR_PAGES.
565 prev_end
= ALIGN(__phys_to_pfn(reg
->base
+ reg
->size
),
569 #ifdef CONFIG_SPARSEMEM
570 if (!IS_ALIGNED(prev_end
, PAGES_PER_SECTION
))
571 free_memmap(prev_end
, ALIGN(prev_end
, PAGES_PER_SECTION
));
574 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
577 * mem_init() marks the free areas in the mem_map and tells us how much memory
578 * is free. This is done after various parts of the system have claimed their
579 * memory after the kernel image.
581 void __init
mem_init(void)
583 if (swiotlb_force
== SWIOTLB_FORCE
||
584 max_pfn
> (arm64_dma_phys_limit
>> PAGE_SHIFT
))
587 swiotlb_force
= SWIOTLB_NO_FORCE
;
589 set_max_mapnr(pfn_to_page(max_pfn
) - mem_map
);
591 #ifndef CONFIG_SPARSEMEM_VMEMMAP
592 free_unused_memmap();
594 /* this will put all unused low memory onto the freelists */
597 kexec_reserve_crashkres_pages();
599 mem_init_print_info(NULL
);
601 #define MLK(b, t) b, t, ((t) - (b)) >> 10
602 #define MLM(b, t) b, t, ((t) - (b)) >> 20
603 #define MLG(b, t) b, t, ((t) - (b)) >> 30
604 #define MLK_ROUNDUP(b, t) b, t, DIV_ROUND_UP(((t) - (b)), SZ_1K)
606 pr_notice("Virtual kernel memory layout:\n");
608 pr_notice(" kasan : 0x%16lx - 0x%16lx (%6ld GB)\n",
609 MLG(KASAN_SHADOW_START
, KASAN_SHADOW_END
));
611 pr_notice(" modules : 0x%16lx - 0x%16lx (%6ld MB)\n",
612 MLM(MODULES_VADDR
, MODULES_END
));
613 pr_notice(" vmalloc : 0x%16lx - 0x%16lx (%6ld GB)\n",
614 MLG(VMALLOC_START
, VMALLOC_END
));
615 pr_notice(" .text : 0x%p" " - 0x%p" " (%6ld KB)\n",
616 MLK_ROUNDUP(_text
, _etext
));
617 pr_notice(" .rodata : 0x%p" " - 0x%p" " (%6ld KB)\n",
618 MLK_ROUNDUP(__start_rodata
, __init_begin
));
619 pr_notice(" .init : 0x%p" " - 0x%p" " (%6ld KB)\n",
620 MLK_ROUNDUP(__init_begin
, __init_end
));
621 pr_notice(" .data : 0x%p" " - 0x%p" " (%6ld KB)\n",
622 MLK_ROUNDUP(_sdata
, _edata
));
623 pr_notice(" .bss : 0x%p" " - 0x%p" " (%6ld KB)\n",
624 MLK_ROUNDUP(__bss_start
, __bss_stop
));
625 pr_notice(" fixed : 0x%16lx - 0x%16lx (%6ld KB)\n",
626 MLK(FIXADDR_START
, FIXADDR_TOP
));
627 pr_notice(" PCI I/O : 0x%16lx - 0x%16lx (%6ld MB)\n",
628 MLM(PCI_IO_START
, PCI_IO_END
));
629 #ifdef CONFIG_SPARSEMEM_VMEMMAP
630 pr_notice(" vmemmap : 0x%16lx - 0x%16lx (%6ld GB maximum)\n",
631 MLG(VMEMMAP_START
, VMEMMAP_START
+ VMEMMAP_SIZE
));
632 pr_notice(" 0x%16lx - 0x%16lx (%6ld MB actual)\n",
633 MLM((unsigned long)phys_to_page(memblock_start_of_DRAM()),
634 (unsigned long)virt_to_page(high_memory
)));
636 pr_notice(" memory : 0x%16lx - 0x%16lx (%6ld MB)\n",
637 MLM(__phys_to_virt(memblock_start_of_DRAM()),
638 (unsigned long)high_memory
));
645 * Check boundaries twice: Some fundamental inconsistencies can be
646 * detected at build time already.
649 BUILD_BUG_ON(TASK_SIZE_32
> TASK_SIZE_64
);
653 * Make sure we chose the upper bound of sizeof(struct page)
656 BUILD_BUG_ON(sizeof(struct page
) > (1 << STRUCT_PAGE_MAX_SHIFT
));
658 if (PAGE_SIZE
>= 16384 && get_num_physpages() <= 128) {
659 extern int sysctl_overcommit_memory
;
661 * On a machine this small we won't get anywhere without
662 * overcommit, so turn it on by default.
664 sysctl_overcommit_memory
= OVERCOMMIT_ALWAYS
;
668 void free_initmem(void)
670 free_reserved_area(__va(__pa(__init_begin
)), __va(__pa(__init_end
)),
673 * Unmap the __init region but leave the VM area in place. This
674 * prevents the region from being reused for kernel modules, which
675 * is not supported by kallsyms.
677 unmap_kernel_range((u64
)__init_begin
, (u64
)(__init_end
- __init_begin
));
680 #ifdef CONFIG_BLK_DEV_INITRD
682 static int keep_initrd __initdata
;
684 void __init
free_initrd_mem(unsigned long start
, unsigned long end
)
687 free_reserved_area((void *)start
, (void *)end
, 0, "initrd");
690 static int __init
keepinitrd_setup(char *__unused
)
696 __setup("keepinitrd", keepinitrd_setup
);
700 * Dump out memory limit information on panic.
702 static int dump_mem_limit(struct notifier_block
*self
, unsigned long v
, void *p
)
704 if (memory_limit
!= (phys_addr_t
)ULLONG_MAX
) {
705 pr_emerg("Memory Limit: %llu MB\n", memory_limit
>> 20);
707 pr_emerg("Memory Limit: none\n");
712 static struct notifier_block mem_limit_notifier
= {
713 .notifier_call
= dump_mem_limit
,
716 static int __init
register_mem_limit_dumper(void)
718 atomic_notifier_chain_register(&panic_notifier_list
,
719 &mem_limit_notifier
);
722 __initcall(register_mem_limit_dumper
);