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>
26 #include <asm/mach-types.h>
27 #include <asm/memblock.h>
28 #include <asm/memory.h>
30 #include <asm/sections.h>
31 #include <asm/setup.h>
32 #include <asm/system_info.h>
34 #include <asm/fixmap.h>
35 #include <asm/ptdump.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 #ifdef CONFIG_BLK_DEV_INITRD
51 static int __init parse_tag_initrd(const struct tag *tag)
53 pr_warn("ATAG_INITRD is deprecated; "
54 "please update your bootloader.\n");
55 phys_initrd_start = __virt_to_phys(tag->u.initrd.start);
56 phys_initrd_size = tag->u.initrd.size;
60 __tagtable(ATAG_INITRD, parse_tag_initrd);
62 static int __init parse_tag_initrd2(const struct tag *tag)
64 phys_initrd_start = tag->u.initrd.start;
65 phys_initrd_size = tag->u.initrd.size;
69 __tagtable(ATAG_INITRD2, parse_tag_initrd2);
72 static void __init find_limits(unsigned long *min, unsigned long *max_low,
73 unsigned long *max_high)
75 *max_low = PFN_DOWN(memblock_get_current_limit());
76 *min = PFN_UP(memblock_start_of_DRAM());
77 *max_high = PFN_DOWN(memblock_end_of_DRAM());
80 #ifdef CONFIG_ZONE_DMA
82 phys_addr_t arm_dma_zone_size __read_mostly;
83 EXPORT_SYMBOL(arm_dma_zone_size);
86 * The DMA mask corresponding to the maximum bus address allocatable
87 * using GFP_DMA. The default here places no restriction on DMA
88 * allocations. This must be the smallest DMA mask in the system,
89 * so a successful GFP_DMA allocation will always satisfy this.
91 phys_addr_t arm_dma_limit;
92 unsigned long arm_dma_pfn_limit;
94 static void __init arm_adjust_dma_zone(unsigned long *size, unsigned long *hole,
95 unsigned long dma_size)
97 if (size[0] <= dma_size)
100 size[ZONE_NORMAL] = size[0] - dma_size;
101 size[ZONE_DMA] = dma_size;
102 hole[ZONE_NORMAL] = hole[0];
107 void __init setup_dma_zone(const struct machine_desc *mdesc)
109 #ifdef CONFIG_ZONE_DMA
110 if (mdesc->dma_zone_size) {
111 arm_dma_zone_size = mdesc->dma_zone_size;
112 arm_dma_limit = PHYS_OFFSET + arm_dma_zone_size - 1;
114 arm_dma_limit = 0xffffffff;
115 arm_dma_pfn_limit = arm_dma_limit >> PAGE_SHIFT;
119 static void __init zone_sizes_init(unsigned long min, unsigned long max_low,
120 unsigned long max_high)
122 unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES];
123 struct memblock_region *reg;
126 * initialise the zones.
128 memset(zone_size, 0, sizeof(zone_size));
131 * The memory size has already been determined. If we need
132 * to do anything fancy with the allocation of this memory
133 * to the zones, now is the time to do it.
135 zone_size[0] = max_low - min;
136 #ifdef CONFIG_HIGHMEM
137 zone_size[ZONE_HIGHMEM] = max_high - max_low;
141 * Calculate the size of the holes.
142 * holes = node_size - sum(bank_sizes)
144 memcpy(zhole_size, zone_size, sizeof(zhole_size));
145 for_each_memblock(memory, reg) {
146 unsigned long start = memblock_region_memory_base_pfn(reg);
147 unsigned long end = memblock_region_memory_end_pfn(reg);
149 if (start < max_low) {
150 unsigned long low_end = min(end, max_low);
151 zhole_size[0] -= low_end - start;
153 #ifdef CONFIG_HIGHMEM
155 unsigned long high_start = max(start, max_low);
156 zhole_size[ZONE_HIGHMEM] -= end - high_start;
161 #ifdef CONFIG_ZONE_DMA
163 * Adjust the sizes according to any special requirements for
166 if (arm_dma_zone_size)
167 arm_adjust_dma_zone(zone_size, zhole_size,
168 arm_dma_zone_size >> PAGE_SHIFT);
171 free_area_init_node(0, zone_size, min, zhole_size);
174 #ifdef CONFIG_HAVE_ARCH_PFN_VALID
175 int pfn_valid(unsigned long pfn)
177 return memblock_is_map_memory(__pfn_to_phys(pfn));
179 EXPORT_SYMBOL(pfn_valid);
182 static bool arm_memblock_steal_permitted = true;
184 phys_addr_t __init arm_memblock_steal(phys_addr_t size, phys_addr_t align)
188 BUG_ON(!arm_memblock_steal_permitted);
190 phys = memblock_phys_alloc(size, align);
192 panic("Failed to steal %pa bytes at %pS\n",
193 &size, (void *)_RET_IP_);
195 memblock_free(phys, size);
196 memblock_remove(phys, size);
201 static void __init arm_initrd_init(void)
203 #ifdef CONFIG_BLK_DEV_INITRD
207 initrd_start = initrd_end = 0;
209 if (!phys_initrd_size)
213 * Round the memory region to page boundaries as per free_initrd_mem()
214 * This allows us to detect whether the pages overlapping the initrd
215 * are in use, but more importantly, reserves the entire set of pages
216 * as we don't want these pages allocated for other purposes.
218 start = round_down(phys_initrd_start, PAGE_SIZE);
219 size = phys_initrd_size + (phys_initrd_start - start);
220 size = round_up(size, PAGE_SIZE);
222 if (!memblock_is_region_memory(start, size)) {
223 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region - disabling initrd\n",
228 if (memblock_is_region_reserved(start, size)) {
229 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region - disabling initrd\n",
234 memblock_reserve(start, size);
236 /* Now convert initrd to virtual addresses */
237 initrd_start = __phys_to_virt(phys_initrd_start);
238 initrd_end = initrd_start + phys_initrd_size;
242 #ifdef CONFIG_CPU_ICACHE_MISMATCH_WORKAROUND
243 void check_cpu_icache_size(int cpuid)
247 asm("mrc p15, 0, %0, c0, c0, 1" : "=r" (ctr));
249 size = 1 << ((ctr & 0xf) + 2);
250 if (cpuid != 0 && icache_size != size)
251 pr_info("CPU%u: detected I-Cache line size mismatch, workaround enabled\n",
253 if (icache_size > size)
258 void __init arm_memblock_init(const struct machine_desc *mdesc)
260 /* Register the kernel text, kernel data and initrd with memblock. */
261 memblock_reserve(__pa(KERNEL_START), KERNEL_END - KERNEL_START);
265 arm_mm_memblock_reserve();
267 /* reserve any platform specific memblock areas */
271 early_init_fdt_reserve_self();
272 early_init_fdt_scan_reserved_mem();
274 /* reserve memory for DMA contiguous allocations */
275 dma_contiguous_reserve(arm_dma_limit);
277 arm_memblock_steal_permitted = false;
281 void __init bootmem_init(void)
283 memblock_allow_resize();
285 find_limits(&min_low_pfn, &max_low_pfn, &max_pfn);
287 early_memtest((phys_addr_t)min_low_pfn << PAGE_SHIFT,
288 (phys_addr_t)max_low_pfn << PAGE_SHIFT);
291 * Sparsemem tries to allocate bootmem in memory_present(),
292 * so must be done after the fixed reservations
297 * sparse_init() needs the bootmem allocator up and running.
302 * Now free the memory - free_area_init_node needs
303 * the sparse mem_map arrays initialized by sparse_init()
304 * for memmap_init_zone(), otherwise all PFNs are invalid.
306 zone_sizes_init(min_low_pfn, max_low_pfn, max_pfn);
310 * Poison init memory with an undefined instruction (ARM) or a branch to an
311 * undefined instruction (Thumb).
313 static inline void poison_init_mem(void *s, size_t count)
316 for (; count != 0; count -= 4)
321 free_memmap(unsigned long start_pfn, unsigned long end_pfn)
323 struct page *start_pg, *end_pg;
324 phys_addr_t pg, pgend;
327 * Convert start_pfn/end_pfn to a struct page pointer.
329 start_pg = pfn_to_page(start_pfn - 1) + 1;
330 end_pg = pfn_to_page(end_pfn - 1) + 1;
333 * Convert to physical addresses, and
334 * round start upwards and end downwards.
336 pg = PAGE_ALIGN(__pa(start_pg));
337 pgend = __pa(end_pg) & PAGE_MASK;
340 * If there are free pages between these,
341 * free the section of the memmap array.
344 memblock_free_early(pg, pgend - pg);
348 * The mem_map array can get very big. Free the unused area of the memory map.
350 static void __init free_unused_memmap(void)
352 unsigned long start, prev_end = 0;
353 struct memblock_region *reg;
356 * This relies on each bank being in address order.
357 * The banks are sorted previously in bootmem_init().
359 for_each_memblock(memory, reg) {
360 start = memblock_region_memory_base_pfn(reg);
362 #ifdef CONFIG_SPARSEMEM
364 * Take care not to free memmap entries that don't exist
365 * due to SPARSEMEM sections which aren't present.
368 ALIGN(prev_end, PAGES_PER_SECTION));
371 * Align down here since the VM subsystem insists that the
372 * memmap entries are valid from the bank start aligned to
373 * MAX_ORDER_NR_PAGES.
375 start = round_down(start, MAX_ORDER_NR_PAGES);
378 * If we had a previous bank, and there is a space
379 * between the current bank and the previous, free it.
381 if (prev_end && prev_end < start)
382 free_memmap(prev_end, start);
385 * Align up here since the VM subsystem insists that the
386 * memmap entries are valid from the bank end aligned to
387 * MAX_ORDER_NR_PAGES.
389 prev_end = ALIGN(memblock_region_memory_end_pfn(reg),
393 #ifdef CONFIG_SPARSEMEM
394 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
395 free_memmap(prev_end,
396 ALIGN(prev_end, PAGES_PER_SECTION));
400 #ifdef CONFIG_HIGHMEM
401 static inline void free_area_high(unsigned long pfn, unsigned long end)
403 for (; pfn < end; pfn++)
404 free_highmem_page(pfn_to_page(pfn));
408 static void __init free_highpages(void)
410 #ifdef CONFIG_HIGHMEM
411 unsigned long max_low = max_low_pfn;
412 struct memblock_region *mem, *res;
414 /* set highmem page free */
415 for_each_memblock(memory, mem) {
416 unsigned long start = memblock_region_memory_base_pfn(mem);
417 unsigned long end = memblock_region_memory_end_pfn(mem);
419 /* Ignore complete lowmem entries */
423 if (memblock_is_nomap(mem))
426 /* Truncate partial highmem entries */
430 /* Find and exclude any reserved regions */
431 for_each_memblock(reserved, res) {
432 unsigned long res_start, res_end;
434 res_start = memblock_region_reserved_base_pfn(res);
435 res_end = memblock_region_reserved_end_pfn(res);
439 if (res_start < start)
445 if (res_start != start)
446 free_area_high(start, res_start);
452 /* And now free anything which remains */
454 free_area_high(start, end);
460 * mem_init() marks the free areas in the mem_map and tells us how much
461 * memory is free. This is done after various parts of the system have
462 * claimed their memory after the kernel image.
464 void __init mem_init(void)
466 set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
468 /* this will put all unused low memory onto the freelists */
469 free_unused_memmap();
473 /* now that our DMA memory is actually so designated, we can free it */
474 free_reserved_area(__va(PHYS_OFFSET), swapper_pg_dir, -1, NULL);
479 mem_init_print_info(NULL);
482 * Check boundaries twice: Some fundamental inconsistencies can
483 * be detected at build time already.
486 BUILD_BUG_ON(TASK_SIZE > MODULES_VADDR);
487 BUG_ON(TASK_SIZE > MODULES_VADDR);
490 #ifdef CONFIG_HIGHMEM
491 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
492 BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
496 #ifdef CONFIG_STRICT_KERNEL_RWX
497 struct section_perm {
506 /* First section-aligned location at or after __start_rodata. */
507 extern char __start_rodata_section_aligned[];
509 static struct section_perm nx_perms[] = {
510 /* Make pages tables, etc before _stext RW (set NX). */
512 .name = "pre-text NX",
513 .start = PAGE_OFFSET,
514 .end = (unsigned long)_stext,
515 .mask = ~PMD_SECT_XN,
518 /* Make init RW (set NX). */
521 .start = (unsigned long)__init_begin,
522 .end = (unsigned long)_sdata,
523 .mask = ~PMD_SECT_XN,
526 /* Make rodata NX (set RO in ro_perms below). */
529 .start = (unsigned long)__start_rodata_section_aligned,
530 .end = (unsigned long)__init_begin,
531 .mask = ~PMD_SECT_XN,
536 static struct section_perm ro_perms[] = {
537 /* Make kernel code and rodata RX (set RO). */
539 .name = "text/rodata RO",
540 .start = (unsigned long)_stext,
541 .end = (unsigned long)__init_begin,
542 #ifdef CONFIG_ARM_LPAE
543 .mask = ~(L_PMD_SECT_RDONLY | PMD_SECT_AP2),
544 .prot = L_PMD_SECT_RDONLY | PMD_SECT_AP2,
546 .mask = ~(PMD_SECT_APX | PMD_SECT_AP_WRITE),
547 .prot = PMD_SECT_APX | PMD_SECT_AP_WRITE,
548 .clear = PMD_SECT_AP_WRITE,
554 * Updates section permissions only for the current mm (sections are
555 * copied into each mm). During startup, this is the init_mm. Is only
556 * safe to be called with preemption disabled, as under stop_machine().
558 static inline void section_update(unsigned long addr, pmdval_t mask,
559 pmdval_t prot, struct mm_struct *mm)
563 pmd = pmd_offset(pud_offset(pgd_offset(mm, addr), addr), addr);
565 #ifdef CONFIG_ARM_LPAE
566 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
568 if (addr & SECTION_SIZE)
569 pmd[1] = __pmd((pmd_val(pmd[1]) & mask) | prot);
571 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
573 flush_pmd_entry(pmd);
574 local_flush_tlb_kernel_range(addr, addr + SECTION_SIZE);
577 /* Make sure extended page tables are in use. */
578 static inline bool arch_has_strict_perms(void)
580 if (cpu_architecture() < CPU_ARCH_ARMv6)
583 return !!(get_cr() & CR_XP);
586 void set_section_perms(struct section_perm *perms, int n, bool set,
587 struct mm_struct *mm)
592 if (!arch_has_strict_perms())
595 for (i = 0; i < n; i++) {
596 if (!IS_ALIGNED(perms[i].start, SECTION_SIZE) ||
597 !IS_ALIGNED(perms[i].end, SECTION_SIZE)) {
598 pr_err("BUG: %s section %lx-%lx not aligned to %lx\n",
599 perms[i].name, perms[i].start, perms[i].end,
604 for (addr = perms[i].start;
606 addr += SECTION_SIZE)
607 section_update(addr, perms[i].mask,
608 set ? perms[i].prot : perms[i].clear, mm);
614 * update_sections_early intended to be called only through stop_machine
615 * framework and executed by only one CPU while all other CPUs will spin and
616 * wait, so no locking is required in this function.
618 static void update_sections_early(struct section_perm perms[], int n)
620 struct task_struct *t, *s;
622 for_each_process(t) {
623 if (t->flags & PF_KTHREAD)
625 for_each_thread(t, s)
626 set_section_perms(perms, n, true, s->mm);
628 set_section_perms(perms, n, true, current->active_mm);
629 set_section_perms(perms, n, true, &init_mm);
632 static int __fix_kernmem_perms(void *unused)
634 update_sections_early(nx_perms, ARRAY_SIZE(nx_perms));
638 static void fix_kernmem_perms(void)
640 stop_machine(__fix_kernmem_perms, NULL, NULL);
643 static int __mark_rodata_ro(void *unused)
645 update_sections_early(ro_perms, ARRAY_SIZE(ro_perms));
649 static int kernel_set_to_readonly __read_mostly;
651 void mark_rodata_ro(void)
653 kernel_set_to_readonly = 1;
654 stop_machine(__mark_rodata_ro, NULL, NULL);
658 void set_kernel_text_rw(void)
660 if (!kernel_set_to_readonly)
663 set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), false,
667 void set_kernel_text_ro(void)
669 if (!kernel_set_to_readonly)
672 set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), true,
677 static inline void fix_kernmem_perms(void) { }
678 #endif /* CONFIG_STRICT_KERNEL_RWX */
680 void free_initmem(void)
684 poison_init_mem(__init_begin, __init_end - __init_begin);
685 if (!machine_is_integrator() && !machine_is_cintegrator())
686 free_initmem_default(-1);
689 #ifdef CONFIG_BLK_DEV_INITRD
690 void free_initrd_mem(unsigned long start, unsigned long end)
692 if (start == initrd_start)
693 start = round_down(start, PAGE_SIZE);
694 if (end == initrd_end)
695 end = round_up(end, PAGE_SIZE);
697 poison_init_mem((void *)start, PAGE_ALIGN(end) - start);
698 free_reserved_area((void *)start, (void *)end, -1, "initrd");