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/sched/signal.h>
17 #include <linux/sched/task.h>
18 #include <linux/export.h>
19 #include <linux/nodemask.h>
20 #include <linux/initrd.h>
21 #include <linux/of_fdt.h>
22 #include <linux/highmem.h>
23 #include <linux/gfp.h>
24 #include <linux/memblock.h>
25 #include <linux/dma-contiguous.h>
26 #include <linux/sizes.h>
27 #include <linux/stop_machine.h>
30 #include <asm/mach-types.h>
31 #include <asm/memblock.h>
32 #include <asm/memory.h>
34 #include <asm/sections.h>
35 #include <asm/setup.h>
36 #include <asm/system_info.h>
38 #include <asm/fixmap.h>
40 #include <asm/mach/arch.h>
41 #include <asm/mach/map.h>
45 #ifdef CONFIG_CPU_CP15_MMU
46 unsigned long __init __clear_cr(unsigned long mask)
48 cr_alignment = cr_alignment & ~mask;
53 static phys_addr_t phys_initrd_start __initdata = 0;
54 static unsigned long phys_initrd_size __initdata = 0;
56 static int __init early_initrd(char *p)
62 start = memparse(p, &endp);
64 size = memparse(endp + 1, NULL);
66 phys_initrd_start = start;
67 phys_initrd_size = size;
71 early_param("initrd", early_initrd);
73 static int __init parse_tag_initrd(const struct tag *tag)
75 pr_warn("ATAG_INITRD is deprecated; "
76 "please update your bootloader.\n");
77 phys_initrd_start = __virt_to_phys(tag->u.initrd.start);
78 phys_initrd_size = tag->u.initrd.size;
82 __tagtable(ATAG_INITRD, parse_tag_initrd);
84 static int __init parse_tag_initrd2(const struct tag *tag)
86 phys_initrd_start = tag->u.initrd.start;
87 phys_initrd_size = tag->u.initrd.size;
91 __tagtable(ATAG_INITRD2, parse_tag_initrd2);
93 static void __init find_limits(unsigned long *min, unsigned long *max_low,
94 unsigned long *max_high)
96 *max_low = PFN_DOWN(memblock_get_current_limit());
97 *min = PFN_UP(memblock_start_of_DRAM());
98 *max_high = PFN_DOWN(memblock_end_of_DRAM());
101 #ifdef CONFIG_ZONE_DMA
103 phys_addr_t arm_dma_zone_size __read_mostly;
104 EXPORT_SYMBOL(arm_dma_zone_size);
107 * The DMA mask corresponding to the maximum bus address allocatable
108 * using GFP_DMA. The default here places no restriction on DMA
109 * allocations. This must be the smallest DMA mask in the system,
110 * so a successful GFP_DMA allocation will always satisfy this.
112 phys_addr_t arm_dma_limit;
113 unsigned long arm_dma_pfn_limit;
115 static void __init arm_adjust_dma_zone(unsigned long *size, unsigned long *hole,
116 unsigned long dma_size)
118 if (size[0] <= dma_size)
121 size[ZONE_NORMAL] = size[0] - dma_size;
122 size[ZONE_DMA] = dma_size;
123 hole[ZONE_NORMAL] = hole[0];
128 void __init setup_dma_zone(const struct machine_desc *mdesc)
130 #ifdef CONFIG_ZONE_DMA
131 if (mdesc->dma_zone_size) {
132 arm_dma_zone_size = mdesc->dma_zone_size;
133 arm_dma_limit = PHYS_OFFSET + arm_dma_zone_size - 1;
135 arm_dma_limit = 0xffffffff;
136 arm_dma_pfn_limit = arm_dma_limit >> PAGE_SHIFT;
140 static void __init zone_sizes_init(unsigned long min, unsigned long max_low,
141 unsigned long max_high)
143 unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES];
144 struct memblock_region *reg;
147 * initialise the zones.
149 memset(zone_size, 0, sizeof(zone_size));
152 * The memory size has already been determined. If we need
153 * to do anything fancy with the allocation of this memory
154 * to the zones, now is the time to do it.
156 zone_size[0] = max_low - min;
157 #ifdef CONFIG_HIGHMEM
158 zone_size[ZONE_HIGHMEM] = max_high - max_low;
162 * Calculate the size of the holes.
163 * holes = node_size - sum(bank_sizes)
165 memcpy(zhole_size, zone_size, sizeof(zhole_size));
166 for_each_memblock(memory, reg) {
167 unsigned long start = memblock_region_memory_base_pfn(reg);
168 unsigned long end = memblock_region_memory_end_pfn(reg);
170 if (start < max_low) {
171 unsigned long low_end = min(end, max_low);
172 zhole_size[0] -= low_end - start;
174 #ifdef CONFIG_HIGHMEM
176 unsigned long high_start = max(start, max_low);
177 zhole_size[ZONE_HIGHMEM] -= end - high_start;
182 #ifdef CONFIG_ZONE_DMA
184 * Adjust the sizes according to any special requirements for
187 if (arm_dma_zone_size)
188 arm_adjust_dma_zone(zone_size, zhole_size,
189 arm_dma_zone_size >> PAGE_SHIFT);
192 free_area_init_node(0, zone_size, min, zhole_size);
195 #ifdef CONFIG_HAVE_ARCH_PFN_VALID
196 int pfn_valid(unsigned long pfn)
198 phys_addr_t addr = __pfn_to_phys(pfn);
200 if (__phys_to_pfn(addr) != pfn)
203 return memblock_is_map_memory(__pfn_to_phys(pfn));
205 EXPORT_SYMBOL(pfn_valid);
208 #ifndef CONFIG_SPARSEMEM
209 static void __init arm_memory_present(void)
213 static void __init arm_memory_present(void)
215 struct memblock_region *reg;
217 for_each_memblock(memory, reg)
218 memory_present(0, memblock_region_memory_base_pfn(reg),
219 memblock_region_memory_end_pfn(reg));
223 static bool arm_memblock_steal_permitted = true;
225 phys_addr_t __init arm_memblock_steal(phys_addr_t size, phys_addr_t align)
229 BUG_ON(!arm_memblock_steal_permitted);
231 phys = memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ANYWHERE);
232 memblock_free(phys, size);
233 memblock_remove(phys, size);
238 static void __init arm_initrd_init(void)
240 #ifdef CONFIG_BLK_DEV_INITRD
244 /* FDT scan will populate initrd_start */
245 if (initrd_start && !phys_initrd_size) {
246 phys_initrd_start = __virt_to_phys(initrd_start);
247 phys_initrd_size = initrd_end - initrd_start;
250 initrd_start = initrd_end = 0;
252 if (!phys_initrd_size)
256 * Round the memory region to page boundaries as per free_initrd_mem()
257 * This allows us to detect whether the pages overlapping the initrd
258 * are in use, but more importantly, reserves the entire set of pages
259 * as we don't want these pages allocated for other purposes.
261 start = round_down(phys_initrd_start, PAGE_SIZE);
262 size = phys_initrd_size + (phys_initrd_start - start);
263 size = round_up(size, PAGE_SIZE);
265 if (!memblock_is_region_memory(start, size)) {
266 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region - disabling initrd\n",
271 if (memblock_is_region_reserved(start, size)) {
272 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region - disabling initrd\n",
277 memblock_reserve(start, size);
279 /* Now convert initrd to virtual addresses */
280 initrd_start = __phys_to_virt(phys_initrd_start);
281 initrd_end = initrd_start + phys_initrd_size;
285 void __init arm_memblock_init(const struct machine_desc *mdesc)
287 /* Register the kernel text, kernel data and initrd with memblock. */
288 memblock_reserve(__pa(KERNEL_START), KERNEL_END - KERNEL_START);
292 arm_mm_memblock_reserve();
294 /* reserve any platform specific memblock areas */
298 early_init_fdt_reserve_self();
299 early_init_fdt_scan_reserved_mem();
301 /* reserve memory for DMA contiguous allocations */
302 dma_contiguous_reserve(arm_dma_limit);
304 arm_memblock_steal_permitted = false;
308 void __init bootmem_init(void)
310 unsigned long min, max_low, max_high;
312 memblock_allow_resize();
313 max_low = max_high = 0;
315 find_limits(&min, &max_low, &max_high);
317 early_memtest((phys_addr_t)min << PAGE_SHIFT,
318 (phys_addr_t)max_low << PAGE_SHIFT);
321 * Sparsemem tries to allocate bootmem in memory_present(),
322 * so must be done after the fixed reservations
324 arm_memory_present();
327 * sparse_init() needs the bootmem allocator up and running.
332 * Now free the memory - free_area_init_node needs
333 * the sparse mem_map arrays initialized by sparse_init()
334 * for memmap_init_zone(), otherwise all PFNs are invalid.
336 zone_sizes_init(min, max_low, max_high);
339 * This doesn't seem to be used by the Linux memory manager any
340 * more, but is used by ll_rw_block. If we can get rid of it, we
341 * also get rid of some of the stuff above as well.
344 max_low_pfn = max_low;
349 * Poison init memory with an undefined instruction (ARM) or a branch to an
350 * undefined instruction (Thumb).
352 static inline void poison_init_mem(void *s, size_t count)
355 for (; count != 0; count -= 4)
359 static inline void __init
360 free_memmap(unsigned long start_pfn, unsigned long end_pfn)
362 struct page *start_pg, *end_pg;
363 phys_addr_t pg, pgend;
366 * Convert start_pfn/end_pfn to a struct page pointer.
368 start_pg = pfn_to_page(start_pfn - 1) + 1;
369 end_pg = pfn_to_page(end_pfn - 1) + 1;
372 * Convert to physical addresses, and
373 * round start upwards and end downwards.
375 pg = PAGE_ALIGN(__pa(start_pg));
376 pgend = __pa(end_pg) & PAGE_MASK;
379 * If there are free pages between these,
380 * free the section of the memmap array.
383 memblock_free_early(pg, pgend - pg);
387 * The mem_map array can get very big. Free the unused area of the memory map.
389 static void __init free_unused_memmap(void)
391 unsigned long start, prev_end = 0;
392 struct memblock_region *reg;
395 * This relies on each bank being in address order.
396 * The banks are sorted previously in bootmem_init().
398 for_each_memblock(memory, reg) {
399 start = memblock_region_memory_base_pfn(reg);
401 #ifdef CONFIG_SPARSEMEM
403 * Take care not to free memmap entries that don't exist
404 * due to SPARSEMEM sections which aren't present.
407 ALIGN(prev_end, PAGES_PER_SECTION));
410 * Align down here since the VM subsystem insists that the
411 * memmap entries are valid from the bank start aligned to
412 * MAX_ORDER_NR_PAGES.
414 start = round_down(start, MAX_ORDER_NR_PAGES);
417 * If we had a previous bank, and there is a space
418 * between the current bank and the previous, free it.
420 if (prev_end && prev_end < start)
421 free_memmap(prev_end, start);
424 * Align up here since the VM subsystem insists that the
425 * memmap entries are valid from the bank end aligned to
426 * MAX_ORDER_NR_PAGES.
428 prev_end = ALIGN(memblock_region_memory_end_pfn(reg),
432 #ifdef CONFIG_SPARSEMEM
433 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
434 free_memmap(prev_end,
435 ALIGN(prev_end, PAGES_PER_SECTION));
439 #ifdef CONFIG_HIGHMEM
440 static inline void free_area_high(unsigned long pfn, unsigned long end)
442 for (; pfn < end; pfn++)
443 free_highmem_page(pfn_to_page(pfn));
447 static void __init free_highpages(void)
449 #ifdef CONFIG_HIGHMEM
450 unsigned long max_low = max_low_pfn;
451 struct memblock_region *mem, *res;
453 /* set highmem page free */
454 for_each_memblock(memory, mem) {
455 unsigned long start = memblock_region_memory_base_pfn(mem);
456 unsigned long end = memblock_region_memory_end_pfn(mem);
458 /* Ignore complete lowmem entries */
462 if (memblock_is_nomap(mem))
465 /* Truncate partial highmem entries */
469 /* Find and exclude any reserved regions */
470 for_each_memblock(reserved, res) {
471 unsigned long res_start, res_end;
473 res_start = memblock_region_reserved_base_pfn(res);
474 res_end = memblock_region_reserved_end_pfn(res);
478 if (res_start < start)
484 if (res_start != start)
485 free_area_high(start, res_start);
491 /* And now free anything which remains */
493 free_area_high(start, end);
499 * mem_init() marks the free areas in the mem_map and tells us how much
500 * memory is free. This is done after various parts of the system have
501 * claimed their memory after the kernel image.
503 void __init mem_init(void)
505 #ifdef CONFIG_HAVE_TCM
506 /* These pointers are filled in on TCM detection */
511 set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
513 /* this will put all unused low memory onto the freelists */
514 free_unused_memmap();
518 /* now that our DMA memory is actually so designated, we can free it */
519 free_reserved_area(__va(PHYS_OFFSET), swapper_pg_dir, -1, NULL);
524 mem_init_print_info(NULL);
526 #define MLK(b, t) b, t, ((t) - (b)) >> 10
527 #define MLM(b, t) b, t, ((t) - (b)) >> 20
528 #define MLK_ROUNDUP(b, t) b, t, DIV_ROUND_UP(((t) - (b)), SZ_1K)
530 pr_notice("Virtual kernel memory layout:\n"
531 " vector : 0x%08lx - 0x%08lx (%4ld kB)\n"
532 #ifdef CONFIG_HAVE_TCM
533 " DTCM : 0x%08lx - 0x%08lx (%4ld kB)\n"
534 " ITCM : 0x%08lx - 0x%08lx (%4ld kB)\n"
536 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
537 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
538 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
539 #ifdef CONFIG_HIGHMEM
540 " pkmap : 0x%08lx - 0x%08lx (%4ld MB)\n"
542 #ifdef CONFIG_MODULES
543 " modules : 0x%08lx - 0x%08lx (%4ld MB)\n"
545 " .text : 0x%p" " - 0x%p" " (%4td kB)\n"
546 " .init : 0x%p" " - 0x%p" " (%4td kB)\n"
547 " .data : 0x%p" " - 0x%p" " (%4td kB)\n"
548 " .bss : 0x%p" " - 0x%p" " (%4td kB)\n",
550 MLK(VECTORS_BASE, VECTORS_BASE + PAGE_SIZE),
551 #ifdef CONFIG_HAVE_TCM
552 MLK(DTCM_OFFSET, (unsigned long) dtcm_end),
553 MLK(ITCM_OFFSET, (unsigned long) itcm_end),
555 MLK(FIXADDR_START, FIXADDR_END),
556 MLM(VMALLOC_START, VMALLOC_END),
557 MLM(PAGE_OFFSET, (unsigned long)high_memory),
558 #ifdef CONFIG_HIGHMEM
559 MLM(PKMAP_BASE, (PKMAP_BASE) + (LAST_PKMAP) *
562 #ifdef CONFIG_MODULES
563 MLM(MODULES_VADDR, MODULES_END),
566 MLK_ROUNDUP(_text, _etext),
567 MLK_ROUNDUP(__init_begin, __init_end),
568 MLK_ROUNDUP(_sdata, _edata),
569 MLK_ROUNDUP(__bss_start, __bss_stop));
576 * Check boundaries twice: Some fundamental inconsistencies can
577 * be detected at build time already.
580 BUILD_BUG_ON(TASK_SIZE > MODULES_VADDR);
581 BUG_ON(TASK_SIZE > MODULES_VADDR);
584 #ifdef CONFIG_HIGHMEM
585 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
586 BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
589 if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) {
590 extern int sysctl_overcommit_memory;
592 * On a machine this small we won't get
593 * anywhere without overcommit, so turn
596 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
600 #ifdef CONFIG_STRICT_KERNEL_RWX
601 struct section_perm {
610 /* First section-aligned location at or after __start_rodata. */
611 extern char __start_rodata_section_aligned[];
613 static struct section_perm nx_perms[] = {
614 /* Make pages tables, etc before _stext RW (set NX). */
616 .name = "pre-text NX",
617 .start = PAGE_OFFSET,
618 .end = (unsigned long)_stext,
619 .mask = ~PMD_SECT_XN,
622 /* Make init RW (set NX). */
625 .start = (unsigned long)__init_begin,
626 .end = (unsigned long)_sdata,
627 .mask = ~PMD_SECT_XN,
630 /* Make rodata NX (set RO in ro_perms below). */
633 .start = (unsigned long)__start_rodata_section_aligned,
634 .end = (unsigned long)__init_begin,
635 .mask = ~PMD_SECT_XN,
640 static struct section_perm ro_perms[] = {
641 /* Make kernel code and rodata RX (set RO). */
643 .name = "text/rodata RO",
644 .start = (unsigned long)_stext,
645 .end = (unsigned long)__init_begin,
646 #ifdef CONFIG_ARM_LPAE
647 .mask = ~(L_PMD_SECT_RDONLY | PMD_SECT_AP2),
648 .prot = L_PMD_SECT_RDONLY | PMD_SECT_AP2,
650 .mask = ~(PMD_SECT_APX | PMD_SECT_AP_WRITE),
651 .prot = PMD_SECT_APX | PMD_SECT_AP_WRITE,
652 .clear = PMD_SECT_AP_WRITE,
658 * Updates section permissions only for the current mm (sections are
659 * copied into each mm). During startup, this is the init_mm. Is only
660 * safe to be called with preemption disabled, as under stop_machine().
662 static inline void section_update(unsigned long addr, pmdval_t mask,
663 pmdval_t prot, struct mm_struct *mm)
667 pmd = pmd_offset(pud_offset(pgd_offset(mm, addr), addr), addr);
669 #ifdef CONFIG_ARM_LPAE
670 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
672 if (addr & SECTION_SIZE)
673 pmd[1] = __pmd((pmd_val(pmd[1]) & mask) | prot);
675 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
677 flush_pmd_entry(pmd);
678 local_flush_tlb_kernel_range(addr, addr + SECTION_SIZE);
681 /* Make sure extended page tables are in use. */
682 static inline bool arch_has_strict_perms(void)
684 if (cpu_architecture() < CPU_ARCH_ARMv6)
687 return !!(get_cr() & CR_XP);
690 void set_section_perms(struct section_perm *perms, int n, bool set,
691 struct mm_struct *mm)
696 if (!arch_has_strict_perms())
699 for (i = 0; i < n; i++) {
700 if (!IS_ALIGNED(perms[i].start, SECTION_SIZE) ||
701 !IS_ALIGNED(perms[i].end, SECTION_SIZE)) {
702 pr_err("BUG: %s section %lx-%lx not aligned to %lx\n",
703 perms[i].name, perms[i].start, perms[i].end,
708 for (addr = perms[i].start;
710 addr += SECTION_SIZE)
711 section_update(addr, perms[i].mask,
712 set ? perms[i].prot : perms[i].clear, mm);
718 * update_sections_early intended to be called only through stop_machine
719 * framework and executed by only one CPU while all other CPUs will spin and
720 * wait, so no locking is required in this function.
722 static void update_sections_early(struct section_perm perms[], int n)
724 struct task_struct *t, *s;
726 for_each_process(t) {
727 if (t->flags & PF_KTHREAD)
729 for_each_thread(t, s)
731 set_section_perms(perms, n, true, s->mm);
733 set_section_perms(perms, n, true, current->active_mm);
734 set_section_perms(perms, n, true, &init_mm);
737 static int __fix_kernmem_perms(void *unused)
739 update_sections_early(nx_perms, ARRAY_SIZE(nx_perms));
743 static void fix_kernmem_perms(void)
745 stop_machine(__fix_kernmem_perms, NULL, NULL);
748 static int __mark_rodata_ro(void *unused)
750 update_sections_early(ro_perms, ARRAY_SIZE(ro_perms));
754 static int kernel_set_to_readonly __read_mostly;
756 void mark_rodata_ro(void)
758 kernel_set_to_readonly = 1;
759 stop_machine(__mark_rodata_ro, NULL, NULL);
762 void set_kernel_text_rw(void)
764 if (!kernel_set_to_readonly)
767 set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), false,
771 void set_kernel_text_ro(void)
773 if (!kernel_set_to_readonly)
776 set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), true,
781 static inline void fix_kernmem_perms(void) { }
782 #endif /* CONFIG_STRICT_KERNEL_RWX */
784 void free_tcmmem(void)
786 #ifdef CONFIG_HAVE_TCM
787 extern char __tcm_start, __tcm_end;
789 poison_init_mem(&__tcm_start, &__tcm_end - &__tcm_start);
790 free_reserved_area(&__tcm_start, &__tcm_end, -1, "TCM link");
794 void free_initmem(void)
799 poison_init_mem(__init_begin, __init_end - __init_begin);
800 if (!machine_is_integrator() && !machine_is_cintegrator())
801 free_initmem_default(-1);
804 #ifdef CONFIG_BLK_DEV_INITRD
806 static int keep_initrd;
808 void free_initrd_mem(unsigned long start, unsigned long end)
811 if (start == initrd_start)
812 start = round_down(start, PAGE_SIZE);
813 if (end == initrd_end)
814 end = round_up(end, PAGE_SIZE);
816 poison_init_mem((void *)start, PAGE_ALIGN(end) - start);
817 free_reserved_area((void *)start, (void *)end, -1, "initrd");
821 static int __init keepinitrd_setup(char *__unused)
827 __setup("keepinitrd", keepinitrd_setup);