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>
39 #include <asm/ptdump.h>
41 #include <asm/mach/arch.h>
42 #include <asm/mach/map.h>
46 #ifdef CONFIG_CPU_CP15_MMU
47 unsigned long __init __clear_cr(unsigned long mask)
49 cr_alignment = cr_alignment & ~mask;
54 static phys_addr_t phys_initrd_start __initdata = 0;
55 static unsigned long phys_initrd_size __initdata = 0;
57 static int __init early_initrd(char *p)
63 start = memparse(p, &endp);
65 size = memparse(endp + 1, NULL);
67 phys_initrd_start = start;
68 phys_initrd_size = size;
72 early_param("initrd", early_initrd);
74 static int __init parse_tag_initrd(const struct tag *tag)
76 pr_warn("ATAG_INITRD is deprecated; "
77 "please update your bootloader.\n");
78 phys_initrd_start = __virt_to_phys(tag->u.initrd.start);
79 phys_initrd_size = tag->u.initrd.size;
83 __tagtable(ATAG_INITRD, parse_tag_initrd);
85 static int __init parse_tag_initrd2(const struct tag *tag)
87 phys_initrd_start = tag->u.initrd.start;
88 phys_initrd_size = tag->u.initrd.size;
92 __tagtable(ATAG_INITRD2, parse_tag_initrd2);
94 static void __init find_limits(unsigned long *min, unsigned long *max_low,
95 unsigned long *max_high)
97 *max_low = PFN_DOWN(memblock_get_current_limit());
98 *min = PFN_UP(memblock_start_of_DRAM());
99 *max_high = PFN_DOWN(memblock_end_of_DRAM());
102 #ifdef CONFIG_ZONE_DMA
104 phys_addr_t arm_dma_zone_size __read_mostly;
105 EXPORT_SYMBOL(arm_dma_zone_size);
108 * The DMA mask corresponding to the maximum bus address allocatable
109 * using GFP_DMA. The default here places no restriction on DMA
110 * allocations. This must be the smallest DMA mask in the system,
111 * so a successful GFP_DMA allocation will always satisfy this.
113 phys_addr_t arm_dma_limit;
114 unsigned long arm_dma_pfn_limit;
116 static void __init arm_adjust_dma_zone(unsigned long *size, unsigned long *hole,
117 unsigned long dma_size)
119 if (size[0] <= dma_size)
122 size[ZONE_NORMAL] = size[0] - dma_size;
123 size[ZONE_DMA] = dma_size;
124 hole[ZONE_NORMAL] = hole[0];
129 void __init setup_dma_zone(const struct machine_desc *mdesc)
131 #ifdef CONFIG_ZONE_DMA
132 if (mdesc->dma_zone_size) {
133 arm_dma_zone_size = mdesc->dma_zone_size;
134 arm_dma_limit = PHYS_OFFSET + arm_dma_zone_size - 1;
136 arm_dma_limit = 0xffffffff;
137 arm_dma_pfn_limit = arm_dma_limit >> PAGE_SHIFT;
141 static void __init zone_sizes_init(unsigned long min, unsigned long max_low,
142 unsigned long max_high)
144 unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES];
145 struct memblock_region *reg;
148 * initialise the zones.
150 memset(zone_size, 0, sizeof(zone_size));
153 * The memory size has already been determined. If we need
154 * to do anything fancy with the allocation of this memory
155 * to the zones, now is the time to do it.
157 zone_size[0] = max_low - min;
158 #ifdef CONFIG_HIGHMEM
159 zone_size[ZONE_HIGHMEM] = max_high - max_low;
163 * Calculate the size of the holes.
164 * holes = node_size - sum(bank_sizes)
166 memcpy(zhole_size, zone_size, sizeof(zhole_size));
167 for_each_memblock(memory, reg) {
168 unsigned long start = memblock_region_memory_base_pfn(reg);
169 unsigned long end = memblock_region_memory_end_pfn(reg);
171 if (start < max_low) {
172 unsigned long low_end = min(end, max_low);
173 zhole_size[0] -= low_end - start;
175 #ifdef CONFIG_HIGHMEM
177 unsigned long high_start = max(start, max_low);
178 zhole_size[ZONE_HIGHMEM] -= end - high_start;
183 #ifdef CONFIG_ZONE_DMA
185 * Adjust the sizes according to any special requirements for
188 if (arm_dma_zone_size)
189 arm_adjust_dma_zone(zone_size, zhole_size,
190 arm_dma_zone_size >> PAGE_SHIFT);
193 free_area_init_node(0, zone_size, min, zhole_size);
196 #ifdef CONFIG_HAVE_ARCH_PFN_VALID
197 int pfn_valid(unsigned long pfn)
199 phys_addr_t addr = __pfn_to_phys(pfn);
201 if (__phys_to_pfn(addr) != pfn)
204 return memblock_is_map_memory(__pfn_to_phys(pfn));
206 EXPORT_SYMBOL(pfn_valid);
209 #ifndef CONFIG_SPARSEMEM
210 static void __init arm_memory_present(void)
214 static void __init arm_memory_present(void)
216 struct memblock_region *reg;
218 for_each_memblock(memory, reg)
219 memory_present(0, memblock_region_memory_base_pfn(reg),
220 memblock_region_memory_end_pfn(reg));
224 static bool arm_memblock_steal_permitted = true;
226 phys_addr_t __init arm_memblock_steal(phys_addr_t size, phys_addr_t align)
230 BUG_ON(!arm_memblock_steal_permitted);
232 phys = memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ANYWHERE);
233 memblock_free(phys, size);
234 memblock_remove(phys, size);
239 static void __init arm_initrd_init(void)
241 #ifdef CONFIG_BLK_DEV_INITRD
245 /* FDT scan will populate initrd_start */
246 if (initrd_start && !phys_initrd_size) {
247 phys_initrd_start = __virt_to_phys(initrd_start);
248 phys_initrd_size = initrd_end - initrd_start;
251 initrd_start = initrd_end = 0;
253 if (!phys_initrd_size)
257 * Round the memory region to page boundaries as per free_initrd_mem()
258 * This allows us to detect whether the pages overlapping the initrd
259 * are in use, but more importantly, reserves the entire set of pages
260 * as we don't want these pages allocated for other purposes.
262 start = round_down(phys_initrd_start, PAGE_SIZE);
263 size = phys_initrd_size + (phys_initrd_start - start);
264 size = round_up(size, PAGE_SIZE);
266 if (!memblock_is_region_memory(start, size)) {
267 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region - disabling initrd\n",
272 if (memblock_is_region_reserved(start, size)) {
273 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region - disabling initrd\n",
278 memblock_reserve(start, size);
280 /* Now convert initrd to virtual addresses */
281 initrd_start = __phys_to_virt(phys_initrd_start);
282 initrd_end = initrd_start + phys_initrd_size;
286 void __init arm_memblock_init(const struct machine_desc *mdesc)
288 /* Register the kernel text, kernel data and initrd with memblock. */
289 memblock_reserve(__pa(KERNEL_START), KERNEL_END - KERNEL_START);
293 arm_mm_memblock_reserve();
295 /* reserve any platform specific memblock areas */
299 early_init_fdt_reserve_self();
300 early_init_fdt_scan_reserved_mem();
302 /* reserve memory for DMA contiguous allocations */
303 dma_contiguous_reserve(arm_dma_limit);
305 arm_memblock_steal_permitted = false;
309 void __init bootmem_init(void)
311 unsigned long min, max_low, max_high;
313 memblock_allow_resize();
314 max_low = max_high = 0;
316 find_limits(&min, &max_low, &max_high);
318 early_memtest((phys_addr_t)min << PAGE_SHIFT,
319 (phys_addr_t)max_low << PAGE_SHIFT);
322 * Sparsemem tries to allocate bootmem in memory_present(),
323 * so must be done after the fixed reservations
325 arm_memory_present();
328 * sparse_init() needs the bootmem allocator up and running.
333 * Now free the memory - free_area_init_node needs
334 * the sparse mem_map arrays initialized by sparse_init()
335 * for memmap_init_zone(), otherwise all PFNs are invalid.
337 zone_sizes_init(min, max_low, max_high);
340 * This doesn't seem to be used by the Linux memory manager any
341 * more, but is used by ll_rw_block. If we can get rid of it, we
342 * also get rid of some of the stuff above as well.
345 max_low_pfn = max_low;
350 * Poison init memory with an undefined instruction (ARM) or a branch to an
351 * undefined instruction (Thumb).
353 static inline void poison_init_mem(void *s, size_t count)
356 for (; count != 0; count -= 4)
360 static inline void __init
361 free_memmap(unsigned long start_pfn, unsigned long end_pfn)
363 struct page *start_pg, *end_pg;
364 phys_addr_t pg, pgend;
367 * Convert start_pfn/end_pfn to a struct page pointer.
369 start_pg = pfn_to_page(start_pfn - 1) + 1;
370 end_pg = pfn_to_page(end_pfn - 1) + 1;
373 * Convert to physical addresses, and
374 * round start upwards and end downwards.
376 pg = PAGE_ALIGN(__pa(start_pg));
377 pgend = __pa(end_pg) & PAGE_MASK;
380 * If there are free pages between these,
381 * free the section of the memmap array.
384 memblock_free_early(pg, pgend - pg);
388 * The mem_map array can get very big. Free the unused area of the memory map.
390 static void __init free_unused_memmap(void)
392 unsigned long start, prev_end = 0;
393 struct memblock_region *reg;
396 * This relies on each bank being in address order.
397 * The banks are sorted previously in bootmem_init().
399 for_each_memblock(memory, reg) {
400 start = memblock_region_memory_base_pfn(reg);
402 #ifdef CONFIG_SPARSEMEM
404 * Take care not to free memmap entries that don't exist
405 * due to SPARSEMEM sections which aren't present.
408 ALIGN(prev_end, PAGES_PER_SECTION));
411 * Align down here since the VM subsystem insists that the
412 * memmap entries are valid from the bank start aligned to
413 * MAX_ORDER_NR_PAGES.
415 start = round_down(start, MAX_ORDER_NR_PAGES);
418 * If we had a previous bank, and there is a space
419 * between the current bank and the previous, free it.
421 if (prev_end && prev_end < start)
422 free_memmap(prev_end, start);
425 * Align up here since the VM subsystem insists that the
426 * memmap entries are valid from the bank end aligned to
427 * MAX_ORDER_NR_PAGES.
429 prev_end = ALIGN(memblock_region_memory_end_pfn(reg),
433 #ifdef CONFIG_SPARSEMEM
434 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
435 free_memmap(prev_end,
436 ALIGN(prev_end, PAGES_PER_SECTION));
440 #ifdef CONFIG_HIGHMEM
441 static inline void free_area_high(unsigned long pfn, unsigned long end)
443 for (; pfn < end; pfn++)
444 free_highmem_page(pfn_to_page(pfn));
448 static void __init free_highpages(void)
450 #ifdef CONFIG_HIGHMEM
451 unsigned long max_low = max_low_pfn;
452 struct memblock_region *mem, *res;
454 /* set highmem page free */
455 for_each_memblock(memory, mem) {
456 unsigned long start = memblock_region_memory_base_pfn(mem);
457 unsigned long end = memblock_region_memory_end_pfn(mem);
459 /* Ignore complete lowmem entries */
463 if (memblock_is_nomap(mem))
466 /* Truncate partial highmem entries */
470 /* Find and exclude any reserved regions */
471 for_each_memblock(reserved, res) {
472 unsigned long res_start, res_end;
474 res_start = memblock_region_reserved_base_pfn(res);
475 res_end = memblock_region_reserved_end_pfn(res);
479 if (res_start < start)
485 if (res_start != start)
486 free_area_high(start, res_start);
492 /* And now free anything which remains */
494 free_area_high(start, end);
500 * mem_init() marks the free areas in the mem_map and tells us how much
501 * memory is free. This is done after various parts of the system have
502 * claimed their memory after the kernel image.
504 void __init mem_init(void)
506 #ifdef CONFIG_HAVE_TCM
507 /* These pointers are filled in on TCM detection */
512 set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
514 /* this will put all unused low memory onto the freelists */
515 free_unused_memmap();
519 /* now that our DMA memory is actually so designated, we can free it */
520 free_reserved_area(__va(PHYS_OFFSET), swapper_pg_dir, -1, NULL);
525 mem_init_print_info(NULL);
527 #define MLK(b, t) b, t, ((t) - (b)) >> 10
528 #define MLM(b, t) b, t, ((t) - (b)) >> 20
529 #define MLK_ROUNDUP(b, t) b, t, DIV_ROUND_UP(((t) - (b)), SZ_1K)
531 pr_notice("Virtual kernel memory layout:\n"
532 " vector : 0x%08lx - 0x%08lx (%4ld kB)\n"
533 #ifdef CONFIG_HAVE_TCM
534 " DTCM : 0x%08lx - 0x%08lx (%4ld kB)\n"
535 " ITCM : 0x%08lx - 0x%08lx (%4ld kB)\n"
537 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
538 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
539 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
540 #ifdef CONFIG_HIGHMEM
541 " pkmap : 0x%08lx - 0x%08lx (%4ld MB)\n"
543 #ifdef CONFIG_MODULES
544 " modules : 0x%08lx - 0x%08lx (%4ld MB)\n"
546 " .text : 0x%p" " - 0x%p" " (%4td kB)\n"
547 " .init : 0x%p" " - 0x%p" " (%4td kB)\n"
548 " .data : 0x%p" " - 0x%p" " (%4td kB)\n"
549 " .bss : 0x%p" " - 0x%p" " (%4td kB)\n",
551 MLK(VECTORS_BASE, VECTORS_BASE + PAGE_SIZE),
552 #ifdef CONFIG_HAVE_TCM
553 MLK(DTCM_OFFSET, (unsigned long) dtcm_end),
554 MLK(ITCM_OFFSET, (unsigned long) itcm_end),
556 MLK(FIXADDR_START, FIXADDR_END),
557 MLM(VMALLOC_START, VMALLOC_END),
558 MLM(PAGE_OFFSET, (unsigned long)high_memory),
559 #ifdef CONFIG_HIGHMEM
560 MLM(PKMAP_BASE, (PKMAP_BASE) + (LAST_PKMAP) *
563 #ifdef CONFIG_MODULES
564 MLM(MODULES_VADDR, MODULES_END),
567 MLK_ROUNDUP(_text, _etext),
568 MLK_ROUNDUP(__init_begin, __init_end),
569 MLK_ROUNDUP(_sdata, _edata),
570 MLK_ROUNDUP(__bss_start, __bss_stop));
577 * Check boundaries twice: Some fundamental inconsistencies can
578 * be detected at build time already.
581 BUILD_BUG_ON(TASK_SIZE > MODULES_VADDR);
582 BUG_ON(TASK_SIZE > MODULES_VADDR);
585 #ifdef CONFIG_HIGHMEM
586 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
587 BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
591 #ifdef CONFIG_STRICT_KERNEL_RWX
592 struct section_perm {
601 /* First section-aligned location at or after __start_rodata. */
602 extern char __start_rodata_section_aligned[];
604 static struct section_perm nx_perms[] = {
605 /* Make pages tables, etc before _stext RW (set NX). */
607 .name = "pre-text NX",
608 .start = PAGE_OFFSET,
609 .end = (unsigned long)_stext,
610 .mask = ~PMD_SECT_XN,
613 /* Make init RW (set NX). */
616 .start = (unsigned long)__init_begin,
617 .end = (unsigned long)_sdata,
618 .mask = ~PMD_SECT_XN,
621 /* Make rodata NX (set RO in ro_perms below). */
624 .start = (unsigned long)__start_rodata_section_aligned,
625 .end = (unsigned long)__init_begin,
626 .mask = ~PMD_SECT_XN,
631 static struct section_perm ro_perms[] = {
632 /* Make kernel code and rodata RX (set RO). */
634 .name = "text/rodata RO",
635 .start = (unsigned long)_stext,
636 .end = (unsigned long)__init_begin,
637 #ifdef CONFIG_ARM_LPAE
638 .mask = ~(L_PMD_SECT_RDONLY | PMD_SECT_AP2),
639 .prot = L_PMD_SECT_RDONLY | PMD_SECT_AP2,
641 .mask = ~(PMD_SECT_APX | PMD_SECT_AP_WRITE),
642 .prot = PMD_SECT_APX | PMD_SECT_AP_WRITE,
643 .clear = PMD_SECT_AP_WRITE,
649 * Updates section permissions only for the current mm (sections are
650 * copied into each mm). During startup, this is the init_mm. Is only
651 * safe to be called with preemption disabled, as under stop_machine().
653 static inline void section_update(unsigned long addr, pmdval_t mask,
654 pmdval_t prot, struct mm_struct *mm)
658 pmd = pmd_offset(pud_offset(pgd_offset(mm, addr), addr), addr);
660 #ifdef CONFIG_ARM_LPAE
661 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
663 if (addr & SECTION_SIZE)
664 pmd[1] = __pmd((pmd_val(pmd[1]) & mask) | prot);
666 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
668 flush_pmd_entry(pmd);
669 local_flush_tlb_kernel_range(addr, addr + SECTION_SIZE);
672 /* Make sure extended page tables are in use. */
673 static inline bool arch_has_strict_perms(void)
675 if (cpu_architecture() < CPU_ARCH_ARMv6)
678 return !!(get_cr() & CR_XP);
681 void set_section_perms(struct section_perm *perms, int n, bool set,
682 struct mm_struct *mm)
687 if (!arch_has_strict_perms())
690 for (i = 0; i < n; i++) {
691 if (!IS_ALIGNED(perms[i].start, SECTION_SIZE) ||
692 !IS_ALIGNED(perms[i].end, SECTION_SIZE)) {
693 pr_err("BUG: %s section %lx-%lx not aligned to %lx\n",
694 perms[i].name, perms[i].start, perms[i].end,
699 for (addr = perms[i].start;
701 addr += SECTION_SIZE)
702 section_update(addr, perms[i].mask,
703 set ? perms[i].prot : perms[i].clear, mm);
709 * update_sections_early intended to be called only through stop_machine
710 * framework and executed by only one CPU while all other CPUs will spin and
711 * wait, so no locking is required in this function.
713 static void update_sections_early(struct section_perm perms[], int n)
715 struct task_struct *t, *s;
717 for_each_process(t) {
718 if (t->flags & PF_KTHREAD)
720 for_each_thread(t, s)
722 set_section_perms(perms, n, true, s->mm);
724 set_section_perms(perms, n, true, current->active_mm);
725 set_section_perms(perms, n, true, &init_mm);
728 static int __fix_kernmem_perms(void *unused)
730 update_sections_early(nx_perms, ARRAY_SIZE(nx_perms));
734 static void fix_kernmem_perms(void)
736 stop_machine(__fix_kernmem_perms, NULL, NULL);
739 static int __mark_rodata_ro(void *unused)
741 update_sections_early(ro_perms, ARRAY_SIZE(ro_perms));
745 static int kernel_set_to_readonly __read_mostly;
747 void mark_rodata_ro(void)
749 kernel_set_to_readonly = 1;
750 stop_machine(__mark_rodata_ro, NULL, NULL);
754 void set_kernel_text_rw(void)
756 if (!kernel_set_to_readonly)
759 set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), false,
763 void set_kernel_text_ro(void)
765 if (!kernel_set_to_readonly)
768 set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), true,
773 static inline void fix_kernmem_perms(void) { }
774 #endif /* CONFIG_STRICT_KERNEL_RWX */
776 void free_initmem(void)
780 poison_init_mem(__init_begin, __init_end - __init_begin);
781 if (!machine_is_integrator() && !machine_is_cintegrator())
782 free_initmem_default(-1);
785 #ifdef CONFIG_BLK_DEV_INITRD
787 static int keep_initrd;
789 void free_initrd_mem(unsigned long start, unsigned long end)
792 if (start == initrd_start)
793 start = round_down(start, PAGE_SIZE);
794 if (end == initrd_end)
795 end = round_up(end, PAGE_SIZE);
797 poison_init_mem((void *)start, PAGE_ALIGN(end) - start);
798 free_reserved_area((void *)start, (void *)end, -1, "initrd");
802 static int __init keepinitrd_setup(char *__unused)
808 __setup("keepinitrd", keepinitrd_setup);