1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/arch/arm/mm/init.c
5 * Copyright (C) 1995-2005 Russell King
7 #include <linux/kernel.h>
8 #include <linux/errno.h>
9 #include <linux/swap.h>
10 #include <linux/init.h>
11 #include <linux/mman.h>
12 #include <linux/sched/signal.h>
13 #include <linux/sched/task.h>
14 #include <linux/export.h>
15 #include <linux/nodemask.h>
16 #include <linux/initrd.h>
17 #include <linux/of_fdt.h>
18 #include <linux/highmem.h>
19 #include <linux/gfp.h>
20 #include <linux/memblock.h>
21 #include <linux/dma-contiguous.h>
22 #include <linux/sizes.h>
23 #include <linux/stop_machine.h>
24 #include <linux/swiotlb.h>
27 #include <asm/mach-types.h>
28 #include <asm/memblock.h>
29 #include <asm/memory.h>
31 #include <asm/sections.h>
32 #include <asm/setup.h>
33 #include <asm/set_memory.h>
34 #include <asm/system_info.h>
36 #include <asm/fixmap.h>
37 #include <asm/ptdump.h>
39 #include <asm/mach/arch.h>
40 #include <asm/mach/map.h>
44 #ifdef CONFIG_CPU_CP15_MMU
45 unsigned long __init
__clear_cr(unsigned long mask
)
47 cr_alignment
= cr_alignment
& ~mask
;
52 #ifdef CONFIG_BLK_DEV_INITRD
53 static int __init
parse_tag_initrd(const struct tag
*tag
)
55 pr_warn("ATAG_INITRD is deprecated; "
56 "please update your bootloader.\n");
57 phys_initrd_start
= __virt_to_phys(tag
->u
.initrd
.start
);
58 phys_initrd_size
= tag
->u
.initrd
.size
;
62 __tagtable(ATAG_INITRD
, parse_tag_initrd
);
64 static int __init
parse_tag_initrd2(const struct tag
*tag
)
66 phys_initrd_start
= tag
->u
.initrd
.start
;
67 phys_initrd_size
= tag
->u
.initrd
.size
;
71 __tagtable(ATAG_INITRD2
, parse_tag_initrd2
);
74 static void __init
find_limits(unsigned long *min
, unsigned long *max_low
,
75 unsigned long *max_high
)
77 *max_low
= PFN_DOWN(memblock_get_current_limit());
78 *min
= PFN_UP(memblock_start_of_DRAM());
79 *max_high
= PFN_DOWN(memblock_end_of_DRAM());
82 #ifdef CONFIG_ZONE_DMA
84 phys_addr_t arm_dma_zone_size __read_mostly
;
85 EXPORT_SYMBOL(arm_dma_zone_size
);
88 * The DMA mask corresponding to the maximum bus address allocatable
89 * using GFP_DMA. The default here places no restriction on DMA
90 * allocations. This must be the smallest DMA mask in the system,
91 * so a successful GFP_DMA allocation will always satisfy this.
93 phys_addr_t arm_dma_limit
;
94 unsigned long arm_dma_pfn_limit
;
96 static void __init
arm_adjust_dma_zone(unsigned long *size
, unsigned long *hole
,
97 unsigned long dma_size
)
99 if (size
[0] <= dma_size
)
102 size
[ZONE_NORMAL
] = size
[0] - dma_size
;
103 size
[ZONE_DMA
] = dma_size
;
104 hole
[ZONE_NORMAL
] = hole
[0];
109 void __init
setup_dma_zone(const struct machine_desc
*mdesc
)
111 #ifdef CONFIG_ZONE_DMA
112 if (mdesc
->dma_zone_size
) {
113 arm_dma_zone_size
= mdesc
->dma_zone_size
;
114 arm_dma_limit
= PHYS_OFFSET
+ arm_dma_zone_size
- 1;
116 arm_dma_limit
= 0xffffffff;
117 arm_dma_pfn_limit
= arm_dma_limit
>> PAGE_SHIFT
;
121 static void __init
zone_sizes_init(unsigned long min
, unsigned long max_low
,
122 unsigned long max_high
)
124 unsigned long zone_size
[MAX_NR_ZONES
], zhole_size
[MAX_NR_ZONES
];
125 struct memblock_region
*reg
;
128 * initialise the zones.
130 memset(zone_size
, 0, sizeof(zone_size
));
133 * The memory size has already been determined. If we need
134 * to do anything fancy with the allocation of this memory
135 * to the zones, now is the time to do it.
137 zone_size
[0] = max_low
- min
;
138 #ifdef CONFIG_HIGHMEM
139 zone_size
[ZONE_HIGHMEM
] = max_high
- max_low
;
143 * Calculate the size of the holes.
144 * holes = node_size - sum(bank_sizes)
146 memcpy(zhole_size
, zone_size
, sizeof(zhole_size
));
147 for_each_memblock(memory
, reg
) {
148 unsigned long start
= memblock_region_memory_base_pfn(reg
);
149 unsigned long end
= memblock_region_memory_end_pfn(reg
);
151 if (start
< max_low
) {
152 unsigned long low_end
= min(end
, max_low
);
153 zhole_size
[0] -= low_end
- start
;
155 #ifdef CONFIG_HIGHMEM
157 unsigned long high_start
= max(start
, max_low
);
158 zhole_size
[ZONE_HIGHMEM
] -= end
- high_start
;
163 #ifdef CONFIG_ZONE_DMA
165 * Adjust the sizes according to any special requirements for
168 if (arm_dma_zone_size
)
169 arm_adjust_dma_zone(zone_size
, zhole_size
,
170 arm_dma_zone_size
>> PAGE_SHIFT
);
173 free_area_init_node(0, zone_size
, min
, zhole_size
);
176 #ifdef CONFIG_HAVE_ARCH_PFN_VALID
177 int pfn_valid(unsigned long pfn
)
179 phys_addr_t addr
= __pfn_to_phys(pfn
);
181 if (__phys_to_pfn(addr
) != pfn
)
184 return memblock_is_map_memory(addr
);
186 EXPORT_SYMBOL(pfn_valid
);
189 static bool arm_memblock_steal_permitted
= true;
191 phys_addr_t __init
arm_memblock_steal(phys_addr_t size
, phys_addr_t align
)
195 BUG_ON(!arm_memblock_steal_permitted
);
197 phys
= memblock_phys_alloc(size
, align
);
199 panic("Failed to steal %pa bytes at %pS\n",
200 &size
, (void *)_RET_IP_
);
202 memblock_free(phys
, size
);
203 memblock_remove(phys
, size
);
208 static void __init
arm_initrd_init(void)
210 #ifdef CONFIG_BLK_DEV_INITRD
214 initrd_start
= initrd_end
= 0;
216 if (!phys_initrd_size
)
220 * Round the memory region to page boundaries as per free_initrd_mem()
221 * This allows us to detect whether the pages overlapping the initrd
222 * are in use, but more importantly, reserves the entire set of pages
223 * as we don't want these pages allocated for other purposes.
225 start
= round_down(phys_initrd_start
, PAGE_SIZE
);
226 size
= phys_initrd_size
+ (phys_initrd_start
- start
);
227 size
= round_up(size
, PAGE_SIZE
);
229 if (!memblock_is_region_memory(start
, size
)) {
230 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region - disabling initrd\n",
235 if (memblock_is_region_reserved(start
, size
)) {
236 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region - disabling initrd\n",
241 memblock_reserve(start
, size
);
243 /* Now convert initrd to virtual addresses */
244 initrd_start
= __phys_to_virt(phys_initrd_start
);
245 initrd_end
= initrd_start
+ phys_initrd_size
;
249 #ifdef CONFIG_CPU_ICACHE_MISMATCH_WORKAROUND
250 void check_cpu_icache_size(int cpuid
)
254 asm("mrc p15, 0, %0, c0, c0, 1" : "=r" (ctr
));
256 size
= 1 << ((ctr
& 0xf) + 2);
257 if (cpuid
!= 0 && icache_size
!= size
)
258 pr_info("CPU%u: detected I-Cache line size mismatch, workaround enabled\n",
260 if (icache_size
> size
)
265 void __init
arm_memblock_init(const struct machine_desc
*mdesc
)
267 /* Register the kernel text, kernel data and initrd with memblock. */
268 memblock_reserve(__pa(KERNEL_START
), KERNEL_END
- KERNEL_START
);
272 arm_mm_memblock_reserve();
274 /* reserve any platform specific memblock areas */
278 early_init_fdt_reserve_self();
279 early_init_fdt_scan_reserved_mem();
281 /* reserve memory for DMA contiguous allocations */
282 dma_contiguous_reserve(arm_dma_limit
);
284 arm_memblock_steal_permitted
= false;
288 void __init
bootmem_init(void)
290 memblock_allow_resize();
292 find_limits(&min_low_pfn
, &max_low_pfn
, &max_pfn
);
294 early_memtest((phys_addr_t
)min_low_pfn
<< PAGE_SHIFT
,
295 (phys_addr_t
)max_low_pfn
<< PAGE_SHIFT
);
298 * Sparsemem tries to allocate bootmem in memory_present(),
299 * so must be done after the fixed reservations
304 * sparse_init() needs the bootmem allocator up and running.
309 * Now free the memory - free_area_init_node needs
310 * the sparse mem_map arrays initialized by sparse_init()
311 * for memmap_init_zone(), otherwise all PFNs are invalid.
313 zone_sizes_init(min_low_pfn
, max_low_pfn
, max_pfn
);
317 * Poison init memory with an undefined instruction (ARM) or a branch to an
318 * undefined instruction (Thumb).
320 static inline void poison_init_mem(void *s
, size_t count
)
323 for (; count
!= 0; count
-= 4)
327 static inline void __init
328 free_memmap(unsigned long start_pfn
, unsigned long end_pfn
)
330 struct page
*start_pg
, *end_pg
;
331 phys_addr_t pg
, pgend
;
334 * Convert start_pfn/end_pfn to a struct page pointer.
336 start_pg
= pfn_to_page(start_pfn
- 1) + 1;
337 end_pg
= pfn_to_page(end_pfn
- 1) + 1;
340 * Convert to physical addresses, and
341 * round start upwards and end downwards.
343 pg
= PAGE_ALIGN(__pa(start_pg
));
344 pgend
= __pa(end_pg
) & PAGE_MASK
;
347 * If there are free pages between these,
348 * free the section of the memmap array.
351 memblock_free_early(pg
, pgend
- pg
);
355 * The mem_map array can get very big. Free the unused area of the memory map.
357 static void __init
free_unused_memmap(void)
359 unsigned long start
, prev_end
= 0;
360 struct memblock_region
*reg
;
363 * This relies on each bank being in address order.
364 * The banks are sorted previously in bootmem_init().
366 for_each_memblock(memory
, reg
) {
367 start
= memblock_region_memory_base_pfn(reg
);
369 #ifdef CONFIG_SPARSEMEM
371 * Take care not to free memmap entries that don't exist
372 * due to SPARSEMEM sections which aren't present.
375 ALIGN(prev_end
, PAGES_PER_SECTION
));
378 * Align down here since the VM subsystem insists that the
379 * memmap entries are valid from the bank start aligned to
380 * MAX_ORDER_NR_PAGES.
382 start
= round_down(start
, MAX_ORDER_NR_PAGES
);
385 * If we had a previous bank, and there is a space
386 * between the current bank and the previous, free it.
388 if (prev_end
&& prev_end
< start
)
389 free_memmap(prev_end
, start
);
392 * Align up here since the VM subsystem insists that the
393 * memmap entries are valid from the bank end aligned to
394 * MAX_ORDER_NR_PAGES.
396 prev_end
= ALIGN(memblock_region_memory_end_pfn(reg
),
400 #ifdef CONFIG_SPARSEMEM
401 if (!IS_ALIGNED(prev_end
, PAGES_PER_SECTION
))
402 free_memmap(prev_end
,
403 ALIGN(prev_end
, PAGES_PER_SECTION
));
407 #ifdef CONFIG_HIGHMEM
408 static inline void free_area_high(unsigned long pfn
, unsigned long end
)
410 for (; pfn
< end
; pfn
++)
411 free_highmem_page(pfn_to_page(pfn
));
415 static void __init
free_highpages(void)
417 #ifdef CONFIG_HIGHMEM
418 unsigned long max_low
= max_low_pfn
;
419 struct memblock_region
*mem
, *res
;
421 /* set highmem page free */
422 for_each_memblock(memory
, mem
) {
423 unsigned long start
= memblock_region_memory_base_pfn(mem
);
424 unsigned long end
= memblock_region_memory_end_pfn(mem
);
426 /* Ignore complete lowmem entries */
430 if (memblock_is_nomap(mem
))
433 /* Truncate partial highmem entries */
437 /* Find and exclude any reserved regions */
438 for_each_memblock(reserved
, res
) {
439 unsigned long res_start
, res_end
;
441 res_start
= memblock_region_reserved_base_pfn(res
);
442 res_end
= memblock_region_reserved_end_pfn(res
);
446 if (res_start
< start
)
452 if (res_start
!= start
)
453 free_area_high(start
, res_start
);
459 /* And now free anything which remains */
461 free_area_high(start
, end
);
467 * mem_init() marks the free areas in the mem_map and tells us how much
468 * memory is free. This is done after various parts of the system have
469 * claimed their memory after the kernel image.
471 void __init
mem_init(void)
473 #ifdef CONFIG_ARM_LPAE
477 set_max_mapnr(pfn_to_page(max_pfn
) - mem_map
);
479 /* this will put all unused low memory onto the freelists */
480 free_unused_memmap();
484 /* now that our DMA memory is actually so designated, we can free it */
485 free_reserved_area(__va(PHYS_OFFSET
), swapper_pg_dir
, -1, NULL
);
490 mem_init_print_info(NULL
);
493 * Check boundaries twice: Some fundamental inconsistencies can
494 * be detected at build time already.
497 BUILD_BUG_ON(TASK_SIZE
> MODULES_VADDR
);
498 BUG_ON(TASK_SIZE
> MODULES_VADDR
);
501 #ifdef CONFIG_HIGHMEM
502 BUILD_BUG_ON(PKMAP_BASE
+ LAST_PKMAP
* PAGE_SIZE
> PAGE_OFFSET
);
503 BUG_ON(PKMAP_BASE
+ LAST_PKMAP
* PAGE_SIZE
> PAGE_OFFSET
);
507 #ifdef CONFIG_STRICT_KERNEL_RWX
508 struct section_perm
{
517 /* First section-aligned location at or after __start_rodata. */
518 extern char __start_rodata_section_aligned
[];
520 static struct section_perm nx_perms
[] = {
521 /* Make pages tables, etc before _stext RW (set NX). */
523 .name
= "pre-text NX",
524 .start
= PAGE_OFFSET
,
525 .end
= (unsigned long)_stext
,
526 .mask
= ~PMD_SECT_XN
,
529 /* Make init RW (set NX). */
532 .start
= (unsigned long)__init_begin
,
533 .end
= (unsigned long)_sdata
,
534 .mask
= ~PMD_SECT_XN
,
537 /* Make rodata NX (set RO in ro_perms below). */
540 .start
= (unsigned long)__start_rodata_section_aligned
,
541 .end
= (unsigned long)__init_begin
,
542 .mask
= ~PMD_SECT_XN
,
547 static struct section_perm ro_perms
[] = {
548 /* Make kernel code and rodata RX (set RO). */
550 .name
= "text/rodata RO",
551 .start
= (unsigned long)_stext
,
552 .end
= (unsigned long)__init_begin
,
553 #ifdef CONFIG_ARM_LPAE
554 .mask
= ~(L_PMD_SECT_RDONLY
| PMD_SECT_AP2
),
555 .prot
= L_PMD_SECT_RDONLY
| PMD_SECT_AP2
,
557 .mask
= ~(PMD_SECT_APX
| PMD_SECT_AP_WRITE
),
558 .prot
= PMD_SECT_APX
| PMD_SECT_AP_WRITE
,
559 .clear
= PMD_SECT_AP_WRITE
,
565 * Updates section permissions only for the current mm (sections are
566 * copied into each mm). During startup, this is the init_mm. Is only
567 * safe to be called with preemption disabled, as under stop_machine().
569 static inline void section_update(unsigned long addr
, pmdval_t mask
,
570 pmdval_t prot
, struct mm_struct
*mm
)
574 pmd
= pmd_offset(pud_offset(pgd_offset(mm
, addr
), addr
), addr
);
576 #ifdef CONFIG_ARM_LPAE
577 pmd
[0] = __pmd((pmd_val(pmd
[0]) & mask
) | prot
);
579 if (addr
& SECTION_SIZE
)
580 pmd
[1] = __pmd((pmd_val(pmd
[1]) & mask
) | prot
);
582 pmd
[0] = __pmd((pmd_val(pmd
[0]) & mask
) | prot
);
584 flush_pmd_entry(pmd
);
585 local_flush_tlb_kernel_range(addr
, addr
+ SECTION_SIZE
);
588 /* Make sure extended page tables are in use. */
589 static inline bool arch_has_strict_perms(void)
591 if (cpu_architecture() < CPU_ARCH_ARMv6
)
594 return !!(get_cr() & CR_XP
);
597 static void set_section_perms(struct section_perm
*perms
, int n
, bool set
,
598 struct mm_struct
*mm
)
603 if (!arch_has_strict_perms())
606 for (i
= 0; i
< n
; i
++) {
607 if (!IS_ALIGNED(perms
[i
].start
, SECTION_SIZE
) ||
608 !IS_ALIGNED(perms
[i
].end
, SECTION_SIZE
)) {
609 pr_err("BUG: %s section %lx-%lx not aligned to %lx\n",
610 perms
[i
].name
, perms
[i
].start
, perms
[i
].end
,
615 for (addr
= perms
[i
].start
;
617 addr
+= SECTION_SIZE
)
618 section_update(addr
, perms
[i
].mask
,
619 set
? perms
[i
].prot
: perms
[i
].clear
, mm
);
625 * update_sections_early intended to be called only through stop_machine
626 * framework and executed by only one CPU while all other CPUs will spin and
627 * wait, so no locking is required in this function.
629 static void update_sections_early(struct section_perm perms
[], int n
)
631 struct task_struct
*t
, *s
;
633 for_each_process(t
) {
634 if (t
->flags
& PF_KTHREAD
)
636 for_each_thread(t
, s
)
638 set_section_perms(perms
, n
, true, s
->mm
);
640 set_section_perms(perms
, n
, true, current
->active_mm
);
641 set_section_perms(perms
, n
, true, &init_mm
);
644 static int __fix_kernmem_perms(void *unused
)
646 update_sections_early(nx_perms
, ARRAY_SIZE(nx_perms
));
650 static void fix_kernmem_perms(void)
652 stop_machine(__fix_kernmem_perms
, NULL
, NULL
);
655 static int __mark_rodata_ro(void *unused
)
657 update_sections_early(ro_perms
, ARRAY_SIZE(ro_perms
));
661 static int kernel_set_to_readonly __read_mostly
;
663 void mark_rodata_ro(void)
665 kernel_set_to_readonly
= 1;
666 stop_machine(__mark_rodata_ro
, NULL
, NULL
);
670 void set_kernel_text_rw(void)
672 if (!kernel_set_to_readonly
)
675 set_section_perms(ro_perms
, ARRAY_SIZE(ro_perms
), false,
679 void set_kernel_text_ro(void)
681 if (!kernel_set_to_readonly
)
684 set_section_perms(ro_perms
, ARRAY_SIZE(ro_perms
), true,
689 static inline void fix_kernmem_perms(void) { }
690 #endif /* CONFIG_STRICT_KERNEL_RWX */
692 void free_initmem(void)
696 poison_init_mem(__init_begin
, __init_end
- __init_begin
);
697 if (!machine_is_integrator() && !machine_is_cintegrator())
698 free_initmem_default(-1);
701 #ifdef CONFIG_BLK_DEV_INITRD
702 void free_initrd_mem(unsigned long start
, unsigned long end
)
704 if (start
== initrd_start
)
705 start
= round_down(start
, PAGE_SIZE
);
706 if (end
== initrd_end
)
707 end
= round_up(end
, PAGE_SIZE
);
709 poison_init_mem((void *)start
, PAGE_ALIGN(end
) - start
);
710 free_reserved_area((void *)start
, (void *)end
, -1, "initrd");