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/mman.h>
15 #include <linux/sched/signal.h>
16 #include <linux/sched/task.h>
17 #include <linux/export.h>
18 #include <linux/nodemask.h>
19 #include <linux/initrd.h>
20 #include <linux/of_fdt.h>
21 #include <linux/highmem.h>
22 #include <linux/gfp.h>
23 #include <linux/memblock.h>
24 #include <linux/dma-contiguous.h>
25 #include <linux/sizes.h>
26 #include <linux/stop_machine.h>
29 #include <asm/mach-types.h>
30 #include <asm/memblock.h>
31 #include <asm/memory.h>
33 #include <asm/sections.h>
34 #include <asm/setup.h>
35 #include <asm/system_info.h>
37 #include <asm/fixmap.h>
38 #include <asm/ptdump.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 #ifdef CONFIG_BLK_DEV_INITRD
54 static int __init parse_tag_initrd(const struct tag *tag)
56 pr_warn("ATAG_INITRD is deprecated; "
57 "please update your bootloader.\n");
58 phys_initrd_start = __virt_to_phys(tag->u.initrd.start);
59 phys_initrd_size = tag->u.initrd.size;
63 __tagtable(ATAG_INITRD, parse_tag_initrd);
65 static int __init parse_tag_initrd2(const struct tag *tag)
67 phys_initrd_start = tag->u.initrd.start;
68 phys_initrd_size = tag->u.initrd.size;
72 __tagtable(ATAG_INITRD2, parse_tag_initrd2);
75 static void __init find_limits(unsigned long *min, unsigned long *max_low,
76 unsigned long *max_high)
78 *max_low = PFN_DOWN(memblock_get_current_limit());
79 *min = PFN_UP(memblock_start_of_DRAM());
80 *max_high = PFN_DOWN(memblock_end_of_DRAM());
83 #ifdef CONFIG_ZONE_DMA
85 phys_addr_t arm_dma_zone_size __read_mostly;
86 EXPORT_SYMBOL(arm_dma_zone_size);
89 * The DMA mask corresponding to the maximum bus address allocatable
90 * using GFP_DMA. The default here places no restriction on DMA
91 * allocations. This must be the smallest DMA mask in the system,
92 * so a successful GFP_DMA allocation will always satisfy this.
94 phys_addr_t arm_dma_limit;
95 unsigned long arm_dma_pfn_limit;
97 static void __init arm_adjust_dma_zone(unsigned long *size, unsigned long *hole,
98 unsigned long dma_size)
100 if (size[0] <= dma_size)
103 size[ZONE_NORMAL] = size[0] - dma_size;
104 size[ZONE_DMA] = dma_size;
105 hole[ZONE_NORMAL] = hole[0];
110 void __init setup_dma_zone(const struct machine_desc *mdesc)
112 #ifdef CONFIG_ZONE_DMA
113 if (mdesc->dma_zone_size) {
114 arm_dma_zone_size = mdesc->dma_zone_size;
115 arm_dma_limit = PHYS_OFFSET + arm_dma_zone_size - 1;
117 arm_dma_limit = 0xffffffff;
118 arm_dma_pfn_limit = arm_dma_limit >> PAGE_SHIFT;
122 static void __init zone_sizes_init(unsigned long min, unsigned long max_low,
123 unsigned long max_high)
125 unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES];
126 struct memblock_region *reg;
129 * initialise the zones.
131 memset(zone_size, 0, sizeof(zone_size));
134 * The memory size has already been determined. If we need
135 * to do anything fancy with the allocation of this memory
136 * to the zones, now is the time to do it.
138 zone_size[0] = max_low - min;
139 #ifdef CONFIG_HIGHMEM
140 zone_size[ZONE_HIGHMEM] = max_high - max_low;
144 * Calculate the size of the holes.
145 * holes = node_size - sum(bank_sizes)
147 memcpy(zhole_size, zone_size, sizeof(zhole_size));
148 for_each_memblock(memory, reg) {
149 unsigned long start = memblock_region_memory_base_pfn(reg);
150 unsigned long end = memblock_region_memory_end_pfn(reg);
152 if (start < max_low) {
153 unsigned long low_end = min(end, max_low);
154 zhole_size[0] -= low_end - start;
156 #ifdef CONFIG_HIGHMEM
158 unsigned long high_start = max(start, max_low);
159 zhole_size[ZONE_HIGHMEM] -= end - high_start;
164 #ifdef CONFIG_ZONE_DMA
166 * Adjust the sizes according to any special requirements for
169 if (arm_dma_zone_size)
170 arm_adjust_dma_zone(zone_size, zhole_size,
171 arm_dma_zone_size >> PAGE_SHIFT);
174 free_area_init_node(0, zone_size, min, zhole_size);
177 #ifdef CONFIG_HAVE_ARCH_PFN_VALID
178 int pfn_valid(unsigned long pfn)
180 return memblock_is_map_memory(__pfn_to_phys(pfn));
182 EXPORT_SYMBOL(pfn_valid);
185 #ifndef CONFIG_SPARSEMEM
186 static void __init arm_memory_present(void)
190 static void __init arm_memory_present(void)
192 struct memblock_region *reg;
194 for_each_memblock(memory, reg)
195 memory_present(0, memblock_region_memory_base_pfn(reg),
196 memblock_region_memory_end_pfn(reg));
200 static bool arm_memblock_steal_permitted = true;
202 phys_addr_t __init arm_memblock_steal(phys_addr_t size, phys_addr_t align)
206 BUG_ON(!arm_memblock_steal_permitted);
208 phys = memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ANYWHERE);
209 memblock_free(phys, size);
210 memblock_remove(phys, size);
215 static void __init arm_initrd_init(void)
217 #ifdef CONFIG_BLK_DEV_INITRD
221 initrd_start = initrd_end = 0;
223 if (!phys_initrd_size)
227 * Round the memory region to page boundaries as per free_initrd_mem()
228 * This allows us to detect whether the pages overlapping the initrd
229 * are in use, but more importantly, reserves the entire set of pages
230 * as we don't want these pages allocated for other purposes.
232 start = round_down(phys_initrd_start, PAGE_SIZE);
233 size = phys_initrd_size + (phys_initrd_start - start);
234 size = round_up(size, PAGE_SIZE);
236 if (!memblock_is_region_memory(start, size)) {
237 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region - disabling initrd\n",
242 if (memblock_is_region_reserved(start, size)) {
243 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region - disabling initrd\n",
248 memblock_reserve(start, size);
250 /* Now convert initrd to virtual addresses */
251 initrd_start = __phys_to_virt(phys_initrd_start);
252 initrd_end = initrd_start + phys_initrd_size;
256 void __init arm_memblock_init(const struct machine_desc *mdesc)
258 /* Register the kernel text, kernel data and initrd with memblock. */
259 memblock_reserve(__pa(KERNEL_START), KERNEL_END - KERNEL_START);
263 arm_mm_memblock_reserve();
265 /* reserve any platform specific memblock areas */
269 early_init_fdt_reserve_self();
270 early_init_fdt_scan_reserved_mem();
272 /* reserve memory for DMA contiguous allocations */
273 dma_contiguous_reserve(arm_dma_limit);
275 arm_memblock_steal_permitted = false;
279 void __init bootmem_init(void)
281 unsigned long min, max_low, max_high;
283 memblock_allow_resize();
284 max_low = max_high = 0;
286 find_limits(&min, &max_low, &max_high);
288 early_memtest((phys_addr_t)min << PAGE_SHIFT,
289 (phys_addr_t)max_low << PAGE_SHIFT);
292 * Sparsemem tries to allocate bootmem in memory_present(),
293 * so must be done after the fixed reservations
295 arm_memory_present();
298 * sparse_init() needs the bootmem allocator up and running.
303 * Now free the memory - free_area_init_node needs
304 * the sparse mem_map arrays initialized by sparse_init()
305 * for memmap_init_zone(), otherwise all PFNs are invalid.
307 zone_sizes_init(min, max_low, max_high);
310 * This doesn't seem to be used by the Linux memory manager any
311 * more, but is used by ll_rw_block. If we can get rid of it, we
312 * also get rid of some of the stuff above as well.
315 max_low_pfn = max_low;
320 * Poison init memory with an undefined instruction (ARM) or a branch to an
321 * undefined instruction (Thumb).
323 static inline void poison_init_mem(void *s, size_t count)
326 for (; count != 0; count -= 4)
331 free_memmap(unsigned long start_pfn, unsigned long end_pfn)
333 struct page *start_pg, *end_pg;
334 phys_addr_t pg, pgend;
337 * Convert start_pfn/end_pfn to a struct page pointer.
339 start_pg = pfn_to_page(start_pfn - 1) + 1;
340 end_pg = pfn_to_page(end_pfn - 1) + 1;
343 * Convert to physical addresses, and
344 * round start upwards and end downwards.
346 pg = PAGE_ALIGN(__pa(start_pg));
347 pgend = __pa(end_pg) & PAGE_MASK;
350 * If there are free pages between these,
351 * free the section of the memmap array.
354 memblock_free_early(pg, pgend - pg);
358 * The mem_map array can get very big. Free the unused area of the memory map.
360 static void __init free_unused_memmap(void)
362 unsigned long start, prev_end = 0;
363 struct memblock_region *reg;
366 * This relies on each bank being in address order.
367 * The banks are sorted previously in bootmem_init().
369 for_each_memblock(memory, reg) {
370 start = memblock_region_memory_base_pfn(reg);
372 #ifdef CONFIG_SPARSEMEM
374 * Take care not to free memmap entries that don't exist
375 * due to SPARSEMEM sections which aren't present.
378 ALIGN(prev_end, PAGES_PER_SECTION));
381 * Align down here since the VM subsystem insists that the
382 * memmap entries are valid from the bank start aligned to
383 * MAX_ORDER_NR_PAGES.
385 start = round_down(start, MAX_ORDER_NR_PAGES);
388 * If we had a previous bank, and there is a space
389 * between the current bank and the previous, free it.
391 if (prev_end && prev_end < start)
392 free_memmap(prev_end, start);
395 * Align up here since the VM subsystem insists that the
396 * memmap entries are valid from the bank end aligned to
397 * MAX_ORDER_NR_PAGES.
399 prev_end = ALIGN(memblock_region_memory_end_pfn(reg),
403 #ifdef CONFIG_SPARSEMEM
404 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
405 free_memmap(prev_end,
406 ALIGN(prev_end, PAGES_PER_SECTION));
410 #ifdef CONFIG_HIGHMEM
411 static inline void free_area_high(unsigned long pfn, unsigned long end)
413 for (; pfn < end; pfn++)
414 free_highmem_page(pfn_to_page(pfn));
418 static void __init free_highpages(void)
420 #ifdef CONFIG_HIGHMEM
421 unsigned long max_low = max_low_pfn;
422 struct memblock_region *mem, *res;
424 /* set highmem page free */
425 for_each_memblock(memory, mem) {
426 unsigned long start = memblock_region_memory_base_pfn(mem);
427 unsigned long end = memblock_region_memory_end_pfn(mem);
429 /* Ignore complete lowmem entries */
433 if (memblock_is_nomap(mem))
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(VECTORS_BASE, VECTORS_BASE + PAGE_SIZE),
522 #ifdef CONFIG_HAVE_TCM
523 MLK(DTCM_OFFSET, (unsigned long) dtcm_end),
524 MLK(ITCM_OFFSET, (unsigned long) itcm_end),
526 MLK(FIXADDR_START, FIXADDR_END),
527 MLM(VMALLOC_START, VMALLOC_END),
528 MLM(PAGE_OFFSET, (unsigned long)high_memory),
529 #ifdef CONFIG_HIGHMEM
530 MLM(PKMAP_BASE, (PKMAP_BASE) + (LAST_PKMAP) *
533 #ifdef CONFIG_MODULES
534 MLM(MODULES_VADDR, MODULES_END),
537 MLK_ROUNDUP(_text, _etext),
538 MLK_ROUNDUP(__init_begin, __init_end),
539 MLK_ROUNDUP(_sdata, _edata),
540 MLK_ROUNDUP(__bss_start, __bss_stop));
547 * Check boundaries twice: Some fundamental inconsistencies can
548 * be detected at build time already.
551 BUILD_BUG_ON(TASK_SIZE > MODULES_VADDR);
552 BUG_ON(TASK_SIZE > MODULES_VADDR);
555 #ifdef CONFIG_HIGHMEM
556 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
557 BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
561 #ifdef CONFIG_STRICT_KERNEL_RWX
562 struct section_perm {
571 /* First section-aligned location at or after __start_rodata. */
572 extern char __start_rodata_section_aligned[];
574 static struct section_perm nx_perms[] = {
575 /* Make pages tables, etc before _stext RW (set NX). */
577 .name = "pre-text NX",
578 .start = PAGE_OFFSET,
579 .end = (unsigned long)_stext,
580 .mask = ~PMD_SECT_XN,
583 /* Make init RW (set NX). */
586 .start = (unsigned long)__init_begin,
587 .end = (unsigned long)_sdata,
588 .mask = ~PMD_SECT_XN,
591 /* Make rodata NX (set RO in ro_perms below). */
594 .start = (unsigned long)__start_rodata_section_aligned,
595 .end = (unsigned long)__init_begin,
596 .mask = ~PMD_SECT_XN,
601 static struct section_perm ro_perms[] = {
602 /* Make kernel code and rodata RX (set RO). */
604 .name = "text/rodata RO",
605 .start = (unsigned long)_stext,
606 .end = (unsigned long)__init_begin,
607 #ifdef CONFIG_ARM_LPAE
608 .mask = ~(L_PMD_SECT_RDONLY | PMD_SECT_AP2),
609 .prot = L_PMD_SECT_RDONLY | PMD_SECT_AP2,
611 .mask = ~(PMD_SECT_APX | PMD_SECT_AP_WRITE),
612 .prot = PMD_SECT_APX | PMD_SECT_AP_WRITE,
613 .clear = PMD_SECT_AP_WRITE,
619 * Updates section permissions only for the current mm (sections are
620 * copied into each mm). During startup, this is the init_mm. Is only
621 * safe to be called with preemption disabled, as under stop_machine().
623 static inline void section_update(unsigned long addr, pmdval_t mask,
624 pmdval_t prot, struct mm_struct *mm)
628 pmd = pmd_offset(pud_offset(pgd_offset(mm, addr), addr), addr);
630 #ifdef CONFIG_ARM_LPAE
631 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
633 if (addr & SECTION_SIZE)
634 pmd[1] = __pmd((pmd_val(pmd[1]) & mask) | prot);
636 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
638 flush_pmd_entry(pmd);
639 local_flush_tlb_kernel_range(addr, addr + SECTION_SIZE);
642 /* Make sure extended page tables are in use. */
643 static inline bool arch_has_strict_perms(void)
645 if (cpu_architecture() < CPU_ARCH_ARMv6)
648 return !!(get_cr() & CR_XP);
651 void set_section_perms(struct section_perm *perms, int n, bool set,
652 struct mm_struct *mm)
657 if (!arch_has_strict_perms())
660 for (i = 0; i < n; i++) {
661 if (!IS_ALIGNED(perms[i].start, SECTION_SIZE) ||
662 !IS_ALIGNED(perms[i].end, SECTION_SIZE)) {
663 pr_err("BUG: %s section %lx-%lx not aligned to %lx\n",
664 perms[i].name, perms[i].start, perms[i].end,
669 for (addr = perms[i].start;
671 addr += SECTION_SIZE)
672 section_update(addr, perms[i].mask,
673 set ? perms[i].prot : perms[i].clear, mm);
679 * update_sections_early intended to be called only through stop_machine
680 * framework and executed by only one CPU while all other CPUs will spin and
681 * wait, so no locking is required in this function.
683 static void update_sections_early(struct section_perm perms[], int n)
685 struct task_struct *t, *s;
687 for_each_process(t) {
688 if (t->flags & PF_KTHREAD)
690 for_each_thread(t, s)
691 set_section_perms(perms, n, true, s->mm);
693 set_section_perms(perms, n, true, current->active_mm);
694 set_section_perms(perms, n, true, &init_mm);
697 static int __fix_kernmem_perms(void *unused)
699 update_sections_early(nx_perms, ARRAY_SIZE(nx_perms));
703 static void fix_kernmem_perms(void)
705 stop_machine(__fix_kernmem_perms, NULL, NULL);
708 static int __mark_rodata_ro(void *unused)
710 update_sections_early(ro_perms, ARRAY_SIZE(ro_perms));
714 static int kernel_set_to_readonly __read_mostly;
716 void mark_rodata_ro(void)
718 kernel_set_to_readonly = 1;
719 stop_machine(__mark_rodata_ro, NULL, NULL);
723 void set_kernel_text_rw(void)
725 if (!kernel_set_to_readonly)
728 set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), false,
732 void set_kernel_text_ro(void)
734 if (!kernel_set_to_readonly)
737 set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), true,
742 static inline void fix_kernmem_perms(void) { }
743 #endif /* CONFIG_STRICT_KERNEL_RWX */
745 void free_initmem(void)
749 poison_init_mem(__init_begin, __init_end - __init_begin);
750 if (!machine_is_integrator() && !machine_is_cintegrator())
751 free_initmem_default(-1);
754 #ifdef CONFIG_BLK_DEV_INITRD
756 static int keep_initrd;
758 void free_initrd_mem(unsigned long start, unsigned long end)
761 if (start == initrd_start)
762 start = round_down(start, PAGE_SIZE);
763 if (end == initrd_end)
764 end = round_up(end, PAGE_SIZE);
766 poison_init_mem((void *)start, PAGE_ALIGN(end) - start);
767 free_reserved_area((void *)start, (void *)end, -1, "initrd");
771 static int __init keepinitrd_setup(char *__unused)
777 __setup("keepinitrd", keepinitrd_setup);