2 * linux/arch/arm/mm/init.c
4 * Copyright (C) 1995-2005 Russell King
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
10 #include <linux/kernel.h>
11 #include <linux/errno.h>
12 #include <linux/swap.h>
13 #include <linux/init.h>
14 #include <linux/bootmem.h>
15 #include <linux/mman.h>
16 #include <linux/export.h>
17 #include <linux/nodemask.h>
18 #include <linux/initrd.h>
19 #include <linux/of_fdt.h>
20 #include <linux/highmem.h>
21 #include <linux/gfp.h>
22 #include <linux/memblock.h>
23 #include <linux/dma-contiguous.h>
24 #include <linux/sizes.h>
25 #include <linux/stop_machine.h>
28 #include <asm/mach-types.h>
29 #include <asm/memblock.h>
31 #include <asm/sections.h>
32 #include <asm/setup.h>
33 #include <asm/system_info.h>
35 #include <asm/fixmap.h>
37 #include <asm/mach/arch.h>
38 #include <asm/mach/map.h>
42 #ifdef CONFIG_CPU_CP15_MMU
43 unsigned long __init
__clear_cr(unsigned long mask
)
45 cr_alignment
= cr_alignment
& ~mask
;
50 static phys_addr_t phys_initrd_start __initdata
= 0;
51 static unsigned long phys_initrd_size __initdata
= 0;
53 static int __init
early_initrd(char *p
)
59 start
= memparse(p
, &endp
);
61 size
= memparse(endp
+ 1, NULL
);
63 phys_initrd_start
= start
;
64 phys_initrd_size
= size
;
68 early_param("initrd", early_initrd
);
70 static int __init
parse_tag_initrd(const struct tag
*tag
)
72 pr_warn("ATAG_INITRD is deprecated; "
73 "please update your bootloader.\n");
74 phys_initrd_start
= __virt_to_phys(tag
->u
.initrd
.start
);
75 phys_initrd_size
= tag
->u
.initrd
.size
;
79 __tagtable(ATAG_INITRD
, parse_tag_initrd
);
81 static int __init
parse_tag_initrd2(const struct tag
*tag
)
83 phys_initrd_start
= tag
->u
.initrd
.start
;
84 phys_initrd_size
= tag
->u
.initrd
.size
;
88 __tagtable(ATAG_INITRD2
, parse_tag_initrd2
);
90 static void __init
find_limits(unsigned long *min
, unsigned long *max_low
,
91 unsigned long *max_high
)
93 *max_low
= PFN_DOWN(memblock_get_current_limit());
94 *min
= PFN_UP(memblock_start_of_DRAM());
95 *max_high
= PFN_DOWN(memblock_end_of_DRAM());
98 #ifdef CONFIG_ZONE_DMA
100 phys_addr_t arm_dma_zone_size __read_mostly
;
101 EXPORT_SYMBOL(arm_dma_zone_size
);
104 * The DMA mask corresponding to the maximum bus address allocatable
105 * using GFP_DMA. The default here places no restriction on DMA
106 * allocations. This must be the smallest DMA mask in the system,
107 * so a successful GFP_DMA allocation will always satisfy this.
109 phys_addr_t arm_dma_limit
;
110 unsigned long arm_dma_pfn_limit
;
112 static void __init
arm_adjust_dma_zone(unsigned long *size
, unsigned long *hole
,
113 unsigned long dma_size
)
115 if (size
[0] <= dma_size
)
118 size
[ZONE_NORMAL
] = size
[0] - dma_size
;
119 size
[ZONE_DMA
] = dma_size
;
120 hole
[ZONE_NORMAL
] = hole
[0];
125 void __init
setup_dma_zone(const struct machine_desc
*mdesc
)
127 #ifdef CONFIG_ZONE_DMA
128 if (mdesc
->dma_zone_size
) {
129 arm_dma_zone_size
= mdesc
->dma_zone_size
;
130 arm_dma_limit
= PHYS_OFFSET
+ arm_dma_zone_size
- 1;
132 arm_dma_limit
= 0xffffffff;
133 arm_dma_pfn_limit
= arm_dma_limit
>> PAGE_SHIFT
;
137 static void __init
zone_sizes_init(unsigned long min
, unsigned long max_low
,
138 unsigned long max_high
)
140 unsigned long zone_size
[MAX_NR_ZONES
], zhole_size
[MAX_NR_ZONES
];
141 struct memblock_region
*reg
;
144 * initialise the zones.
146 memset(zone_size
, 0, sizeof(zone_size
));
149 * The memory size has already been determined. If we need
150 * to do anything fancy with the allocation of this memory
151 * to the zones, now is the time to do it.
153 zone_size
[0] = max_low
- min
;
154 #ifdef CONFIG_HIGHMEM
155 zone_size
[ZONE_HIGHMEM
] = max_high
- max_low
;
159 * Calculate the size of the holes.
160 * holes = node_size - sum(bank_sizes)
162 memcpy(zhole_size
, zone_size
, sizeof(zhole_size
));
163 for_each_memblock(memory
, reg
) {
164 unsigned long start
= memblock_region_memory_base_pfn(reg
);
165 unsigned long end
= memblock_region_memory_end_pfn(reg
);
167 if (start
< max_low
) {
168 unsigned long low_end
= min(end
, max_low
);
169 zhole_size
[0] -= low_end
- start
;
171 #ifdef CONFIG_HIGHMEM
173 unsigned long high_start
= max(start
, max_low
);
174 zhole_size
[ZONE_HIGHMEM
] -= end
- high_start
;
179 #ifdef CONFIG_ZONE_DMA
181 * Adjust the sizes according to any special requirements for
184 if (arm_dma_zone_size
)
185 arm_adjust_dma_zone(zone_size
, zhole_size
,
186 arm_dma_zone_size
>> PAGE_SHIFT
);
189 free_area_init_node(0, zone_size
, min
, zhole_size
);
192 #ifdef CONFIG_HAVE_ARCH_PFN_VALID
193 int pfn_valid(unsigned long pfn
)
195 return memblock_is_memory(__pfn_to_phys(pfn
));
197 EXPORT_SYMBOL(pfn_valid
);
200 #ifndef CONFIG_SPARSEMEM
201 static void __init
arm_memory_present(void)
205 static void __init
arm_memory_present(void)
207 struct memblock_region
*reg
;
209 for_each_memblock(memory
, reg
)
210 memory_present(0, memblock_region_memory_base_pfn(reg
),
211 memblock_region_memory_end_pfn(reg
));
215 static bool arm_memblock_steal_permitted
= true;
217 phys_addr_t __init
arm_memblock_steal(phys_addr_t size
, phys_addr_t align
)
221 BUG_ON(!arm_memblock_steal_permitted
);
223 phys
= memblock_alloc_base(size
, align
, MEMBLOCK_ALLOC_ANYWHERE
);
224 memblock_free(phys
, size
);
225 memblock_remove(phys
, size
);
230 void __init
arm_memblock_init(const struct machine_desc
*mdesc
)
232 /* Register the kernel text, kernel data and initrd with memblock. */
233 #ifdef CONFIG_XIP_KERNEL
234 memblock_reserve(__pa(_sdata
), _end
- _sdata
);
236 memblock_reserve(__pa(_stext
), _end
- _stext
);
238 #ifdef CONFIG_BLK_DEV_INITRD
239 /* FDT scan will populate initrd_start */
240 if (initrd_start
&& !phys_initrd_size
) {
241 phys_initrd_start
= __virt_to_phys(initrd_start
);
242 phys_initrd_size
= initrd_end
- initrd_start
;
244 initrd_start
= initrd_end
= 0;
245 if (phys_initrd_size
&&
246 !memblock_is_region_memory(phys_initrd_start
, phys_initrd_size
)) {
247 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region - disabling initrd\n",
248 (u64
)phys_initrd_start
, phys_initrd_size
);
249 phys_initrd_start
= phys_initrd_size
= 0;
251 if (phys_initrd_size
&&
252 memblock_is_region_reserved(phys_initrd_start
, phys_initrd_size
)) {
253 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region - disabling initrd\n",
254 (u64
)phys_initrd_start
, phys_initrd_size
);
255 phys_initrd_start
= phys_initrd_size
= 0;
257 if (phys_initrd_size
) {
258 memblock_reserve(phys_initrd_start
, phys_initrd_size
);
260 /* Now convert initrd to virtual addresses */
261 initrd_start
= __phys_to_virt(phys_initrd_start
);
262 initrd_end
= initrd_start
+ phys_initrd_size
;
266 arm_mm_memblock_reserve();
268 /* reserve any platform specific memblock areas */
272 early_init_fdt_reserve_self();
273 early_init_fdt_scan_reserved_mem();
275 /* reserve memory for DMA contiguous allocations */
276 dma_contiguous_reserve(arm_dma_limit
);
278 arm_memblock_steal_permitted
= false;
282 void __init
bootmem_init(void)
284 unsigned long min
, max_low
, max_high
;
286 memblock_allow_resize();
287 max_low
= max_high
= 0;
289 find_limits(&min
, &max_low
, &max_high
);
291 early_memtest((phys_addr_t
)min
<< PAGE_SHIFT
,
292 (phys_addr_t
)max_low
<< PAGE_SHIFT
);
295 * Sparsemem tries to allocate bootmem in memory_present(),
296 * so must be done after the fixed reservations
298 arm_memory_present();
301 * sparse_init() needs the bootmem allocator up and running.
306 * Now free the memory - free_area_init_node needs
307 * the sparse mem_map arrays initialized by sparse_init()
308 * for memmap_init_zone(), otherwise all PFNs are invalid.
310 zone_sizes_init(min
, max_low
, max_high
);
313 * This doesn't seem to be used by the Linux memory manager any
314 * more, but is used by ll_rw_block. If we can get rid of it, we
315 * also get rid of some of the stuff above as well.
318 max_low_pfn
= max_low
;
323 * Poison init memory with an undefined instruction (ARM) or a branch to an
324 * undefined instruction (Thumb).
326 static inline void poison_init_mem(void *s
, size_t count
)
329 for (; count
!= 0; count
-= 4)
334 free_memmap(unsigned long start_pfn
, unsigned long end_pfn
)
336 struct page
*start_pg
, *end_pg
;
337 phys_addr_t pg
, pgend
;
340 * Convert start_pfn/end_pfn to a struct page pointer.
342 start_pg
= pfn_to_page(start_pfn
- 1) + 1;
343 end_pg
= pfn_to_page(end_pfn
- 1) + 1;
346 * Convert to physical addresses, and
347 * round start upwards and end downwards.
349 pg
= PAGE_ALIGN(__pa(start_pg
));
350 pgend
= __pa(end_pg
) & PAGE_MASK
;
353 * If there are free pages between these,
354 * free the section of the memmap array.
357 memblock_free_early(pg
, pgend
- pg
);
361 * The mem_map array can get very big. Free the unused area of the memory map.
363 static void __init
free_unused_memmap(void)
365 unsigned long start
, prev_end
= 0;
366 struct memblock_region
*reg
;
369 * This relies on each bank being in address order.
370 * The banks are sorted previously in bootmem_init().
372 for_each_memblock(memory
, reg
) {
373 start
= memblock_region_memory_base_pfn(reg
);
375 #ifdef CONFIG_SPARSEMEM
377 * Take care not to free memmap entries that don't exist
378 * due to SPARSEMEM sections which aren't present.
381 ALIGN(prev_end
, PAGES_PER_SECTION
));
384 * Align down here since the VM subsystem insists that the
385 * memmap entries are valid from the bank start aligned to
386 * MAX_ORDER_NR_PAGES.
388 start
= round_down(start
, MAX_ORDER_NR_PAGES
);
391 * If we had a previous bank, and there is a space
392 * between the current bank and the previous, free it.
394 if (prev_end
&& prev_end
< start
)
395 free_memmap(prev_end
, start
);
398 * Align up here since the VM subsystem insists that the
399 * memmap entries are valid from the bank end aligned to
400 * MAX_ORDER_NR_PAGES.
402 prev_end
= ALIGN(memblock_region_memory_end_pfn(reg
),
406 #ifdef CONFIG_SPARSEMEM
407 if (!IS_ALIGNED(prev_end
, PAGES_PER_SECTION
))
408 free_memmap(prev_end
,
409 ALIGN(prev_end
, PAGES_PER_SECTION
));
413 #ifdef CONFIG_HIGHMEM
414 static inline void free_area_high(unsigned long pfn
, unsigned long end
)
416 for (; pfn
< end
; pfn
++)
417 free_highmem_page(pfn_to_page(pfn
));
421 static void __init
free_highpages(void)
423 #ifdef CONFIG_HIGHMEM
424 unsigned long max_low
= max_low_pfn
;
425 struct memblock_region
*mem
, *res
;
427 /* set highmem page free */
428 for_each_memblock(memory
, mem
) {
429 unsigned long start
= memblock_region_memory_base_pfn(mem
);
430 unsigned long end
= memblock_region_memory_end_pfn(mem
);
432 /* Ignore complete lowmem entries */
436 /* Truncate partial highmem entries */
440 /* Find and exclude any reserved regions */
441 for_each_memblock(reserved
, res
) {
442 unsigned long res_start
, res_end
;
444 res_start
= memblock_region_reserved_base_pfn(res
);
445 res_end
= memblock_region_reserved_end_pfn(res
);
449 if (res_start
< start
)
455 if (res_start
!= start
)
456 free_area_high(start
, res_start
);
462 /* And now free anything which remains */
464 free_area_high(start
, end
);
470 * mem_init() marks the free areas in the mem_map and tells us how much
471 * memory is free. This is done after various parts of the system have
472 * claimed their memory after the kernel image.
474 void __init
mem_init(void)
476 #ifdef CONFIG_HAVE_TCM
477 /* These pointers are filled in on TCM detection */
482 set_max_mapnr(pfn_to_page(max_pfn
) - mem_map
);
484 /* this will put all unused low memory onto the freelists */
485 free_unused_memmap();
489 /* now that our DMA memory is actually so designated, we can free it */
490 free_reserved_area(__va(PHYS_OFFSET
), swapper_pg_dir
, -1, NULL
);
495 mem_init_print_info(NULL
);
497 #define MLK(b, t) b, t, ((t) - (b)) >> 10
498 #define MLM(b, t) b, t, ((t) - (b)) >> 20
499 #define MLK_ROUNDUP(b, t) b, t, DIV_ROUND_UP(((t) - (b)), SZ_1K)
501 pr_notice("Virtual kernel memory layout:\n"
502 " vector : 0x%08lx - 0x%08lx (%4ld kB)\n"
503 #ifdef CONFIG_HAVE_TCM
504 " DTCM : 0x%08lx - 0x%08lx (%4ld kB)\n"
505 " ITCM : 0x%08lx - 0x%08lx (%4ld kB)\n"
507 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
508 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
509 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
510 #ifdef CONFIG_HIGHMEM
511 " pkmap : 0x%08lx - 0x%08lx (%4ld MB)\n"
513 #ifdef CONFIG_MODULES
514 " modules : 0x%08lx - 0x%08lx (%4ld MB)\n"
516 " .text : 0x%p" " - 0x%p" " (%4td kB)\n"
517 " .init : 0x%p" " - 0x%p" " (%4td kB)\n"
518 " .data : 0x%p" " - 0x%p" " (%4td kB)\n"
519 " .bss : 0x%p" " - 0x%p" " (%4td kB)\n",
521 MLK(UL(CONFIG_VECTORS_BASE
), UL(CONFIG_VECTORS_BASE
) +
523 #ifdef CONFIG_HAVE_TCM
524 MLK(DTCM_OFFSET
, (unsigned long) dtcm_end
),
525 MLK(ITCM_OFFSET
, (unsigned long) itcm_end
),
527 MLK(FIXADDR_START
, FIXADDR_END
),
528 MLM(VMALLOC_START
, VMALLOC_END
),
529 MLM(PAGE_OFFSET
, (unsigned long)high_memory
),
530 #ifdef CONFIG_HIGHMEM
531 MLM(PKMAP_BASE
, (PKMAP_BASE
) + (LAST_PKMAP
) *
534 #ifdef CONFIG_MODULES
535 MLM(MODULES_VADDR
, MODULES_END
),
538 MLK_ROUNDUP(_text
, _etext
),
539 MLK_ROUNDUP(__init_begin
, __init_end
),
540 MLK_ROUNDUP(_sdata
, _edata
),
541 MLK_ROUNDUP(__bss_start
, __bss_stop
));
548 * Check boundaries twice: Some fundamental inconsistencies can
549 * be detected at build time already.
552 BUILD_BUG_ON(TASK_SIZE
> MODULES_VADDR
);
553 BUG_ON(TASK_SIZE
> MODULES_VADDR
);
556 #ifdef CONFIG_HIGHMEM
557 BUILD_BUG_ON(PKMAP_BASE
+ LAST_PKMAP
* PAGE_SIZE
> PAGE_OFFSET
);
558 BUG_ON(PKMAP_BASE
+ LAST_PKMAP
* PAGE_SIZE
> PAGE_OFFSET
);
561 if (PAGE_SIZE
>= 16384 && get_num_physpages() <= 128) {
562 extern int sysctl_overcommit_memory
;
564 * On a machine this small we won't get
565 * anywhere without overcommit, so turn
568 sysctl_overcommit_memory
= OVERCOMMIT_ALWAYS
;
572 #ifdef CONFIG_ARM_KERNMEM_PERMS
573 struct section_perm
{
581 static struct section_perm nx_perms
[] = {
582 /* Make pages tables, etc before _stext RW (set NX). */
584 .start
= PAGE_OFFSET
,
585 .end
= (unsigned long)_stext
,
586 .mask
= ~PMD_SECT_XN
,
589 /* Make init RW (set NX). */
591 .start
= (unsigned long)__init_begin
,
592 .end
= (unsigned long)_sdata
,
593 .mask
= ~PMD_SECT_XN
,
596 #ifdef CONFIG_DEBUG_RODATA
597 /* Make rodata NX (set RO in ro_perms below). */
599 .start
= (unsigned long)__start_rodata
,
600 .end
= (unsigned long)__init_begin
,
601 .mask
= ~PMD_SECT_XN
,
607 #ifdef CONFIG_DEBUG_RODATA
608 static struct section_perm ro_perms
[] = {
609 /* Make kernel code and rodata RX (set RO). */
611 .start
= (unsigned long)_stext
,
612 .end
= (unsigned long)__init_begin
,
613 #ifdef CONFIG_ARM_LPAE
614 .mask
= ~L_PMD_SECT_RDONLY
,
615 .prot
= L_PMD_SECT_RDONLY
,
617 .mask
= ~(PMD_SECT_APX
| PMD_SECT_AP_WRITE
),
618 .prot
= PMD_SECT_APX
| PMD_SECT_AP_WRITE
,
619 .clear
= PMD_SECT_AP_WRITE
,
626 * Updates section permissions only for the current mm (sections are
627 * copied into each mm). During startup, this is the init_mm. Is only
628 * safe to be called with preemption disabled, as under stop_machine().
630 static inline void section_update(unsigned long addr
, pmdval_t mask
,
631 pmdval_t prot
, struct mm_struct
*mm
)
635 pmd
= pmd_offset(pud_offset(pgd_offset(mm
, addr
), addr
), addr
);
637 #ifdef CONFIG_ARM_LPAE
638 pmd
[0] = __pmd((pmd_val(pmd
[0]) & mask
) | prot
);
640 if (addr
& SECTION_SIZE
)
641 pmd
[1] = __pmd((pmd_val(pmd
[1]) & mask
) | prot
);
643 pmd
[0] = __pmd((pmd_val(pmd
[0]) & mask
) | prot
);
645 flush_pmd_entry(pmd
);
646 local_flush_tlb_kernel_range(addr
, addr
+ SECTION_SIZE
);
649 /* Make sure extended page tables are in use. */
650 static inline bool arch_has_strict_perms(void)
652 if (cpu_architecture() < CPU_ARCH_ARMv6
)
655 return !!(get_cr() & CR_XP
);
658 void set_section_perms(struct section_perm
*perms
, int n
, bool set
,
659 struct mm_struct
*mm
)
664 if (!arch_has_strict_perms())
667 for (i
= 0; i
< n
; i
++) {
668 if (!IS_ALIGNED(perms
[i
].start
, SECTION_SIZE
) ||
669 !IS_ALIGNED(perms
[i
].end
, SECTION_SIZE
)) {
670 pr_err("BUG: section %lx-%lx not aligned to %lx\n",
671 perms
[i
].start
, perms
[i
].end
,
676 for (addr
= perms
[i
].start
;
678 addr
+= SECTION_SIZE
)
679 section_update(addr
, perms
[i
].mask
,
680 set
? perms
[i
].prot
: perms
[i
].clear
, mm
);
685 static void update_sections_early(struct section_perm perms
[], int n
)
687 struct task_struct
*t
, *s
;
689 read_lock(&tasklist_lock
);
690 for_each_process(t
) {
691 if (t
->flags
& PF_KTHREAD
)
693 for_each_thread(t
, s
)
694 set_section_perms(perms
, n
, true, s
->mm
);
696 read_unlock(&tasklist_lock
);
697 set_section_perms(perms
, n
, true, current
->active_mm
);
698 set_section_perms(perms
, n
, true, &init_mm
);
701 int __fix_kernmem_perms(void *unused
)
703 update_sections_early(nx_perms
, ARRAY_SIZE(nx_perms
));
707 void fix_kernmem_perms(void)
709 stop_machine(__fix_kernmem_perms
, NULL
, NULL
);
712 #ifdef CONFIG_DEBUG_RODATA
713 int __mark_rodata_ro(void *unused
)
715 update_sections_early(ro_perms
, ARRAY_SIZE(ro_perms
));
719 void mark_rodata_ro(void)
721 stop_machine(__mark_rodata_ro
, NULL
, NULL
);
724 void set_kernel_text_rw(void)
726 set_section_perms(ro_perms
, ARRAY_SIZE(ro_perms
), false,
730 void set_kernel_text_ro(void)
732 set_section_perms(ro_perms
, ARRAY_SIZE(ro_perms
), true,
735 #endif /* CONFIG_DEBUG_RODATA */
738 static inline void fix_kernmem_perms(void) { }
739 #endif /* CONFIG_ARM_KERNMEM_PERMS */
741 void free_tcmmem(void)
743 #ifdef CONFIG_HAVE_TCM
744 extern char __tcm_start
, __tcm_end
;
746 poison_init_mem(&__tcm_start
, &__tcm_end
- &__tcm_start
);
747 free_reserved_area(&__tcm_start
, &__tcm_end
, -1, "TCM link");
751 void free_initmem(void)
756 poison_init_mem(__init_begin
, __init_end
- __init_begin
);
757 if (!machine_is_integrator() && !machine_is_cintegrator())
758 free_initmem_default(-1);
761 #ifdef CONFIG_BLK_DEV_INITRD
763 static int keep_initrd
;
765 void free_initrd_mem(unsigned long start
, unsigned long end
)
768 if (start
== initrd_start
)
769 start
= round_down(start
, PAGE_SIZE
);
770 if (end
== initrd_end
)
771 end
= round_up(end
, PAGE_SIZE
);
773 poison_init_mem((void *)start
, PAGE_ALIGN(end
) - start
);
774 free_reserved_area((void *)start
, (void *)end
, -1, "initrd");
778 static int __init
keepinitrd_setup(char *__unused
)
784 __setup("keepinitrd", keepinitrd_setup
);