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-map-ops.h>
22 #include <linux/sizes.h>
23 #include <linux/stop_machine.h>
24 #include <linux/swiotlb.h>
27 #include <asm/mach-types.h>
28 #include <asm/memblock.h>
29 #include <asm/memory.h>
31 #include <asm/sections.h>
32 #include <asm/setup.h>
33 #include <asm/set_memory.h>
34 #include <asm/system_info.h>
36 #include <asm/fixmap.h>
37 #include <asm/ptdump.h>
39 #include <asm/mach/arch.h>
40 #include <asm/mach/map.h>
44 #ifdef CONFIG_CPU_CP15_MMU
45 unsigned long __init __clear_cr(unsigned long mask)
47 cr_alignment = cr_alignment & ~mask;
52 #ifdef CONFIG_BLK_DEV_INITRD
53 static int __init parse_tag_initrd(const struct tag *tag)
55 pr_warn("ATAG_INITRD is deprecated; "
56 "please update your bootloader.\n");
57 phys_initrd_start = __virt_to_phys(tag->u.initrd.start);
58 phys_initrd_size = tag->u.initrd.size;
62 __tagtable(ATAG_INITRD, parse_tag_initrd);
64 static int __init parse_tag_initrd2(const struct tag *tag)
66 phys_initrd_start = tag->u.initrd.start;
67 phys_initrd_size = tag->u.initrd.size;
71 __tagtable(ATAG_INITRD2, parse_tag_initrd2);
74 static void __init find_limits(unsigned long *min, unsigned long *max_low,
75 unsigned long *max_high)
77 *max_low = PFN_DOWN(memblock_get_current_limit());
78 *min = PFN_UP(memblock_start_of_DRAM());
79 *max_high = PFN_DOWN(memblock_end_of_DRAM());
82 #ifdef CONFIG_ZONE_DMA
84 phys_addr_t arm_dma_zone_size __read_mostly;
85 EXPORT_SYMBOL(arm_dma_zone_size);
88 * The DMA mask corresponding to the maximum bus address allocatable
89 * using GFP_DMA. The default here places no restriction on DMA
90 * allocations. This must be the smallest DMA mask in the system,
91 * so a successful GFP_DMA allocation will always satisfy this.
93 phys_addr_t arm_dma_limit;
94 unsigned long arm_dma_pfn_limit;
97 void __init setup_dma_zone(const struct machine_desc *mdesc)
99 #ifdef CONFIG_ZONE_DMA
100 if (mdesc->dma_zone_size) {
101 arm_dma_zone_size = mdesc->dma_zone_size;
102 arm_dma_limit = PHYS_OFFSET + arm_dma_zone_size - 1;
104 arm_dma_limit = 0xffffffff;
105 arm_dma_pfn_limit = arm_dma_limit >> PAGE_SHIFT;
109 static void __init zone_sizes_init(unsigned long min, unsigned long max_low,
110 unsigned long max_high)
112 unsigned long max_zone_pfn[MAX_NR_ZONES] = { 0 };
114 #ifdef CONFIG_ZONE_DMA
115 max_zone_pfn[ZONE_DMA] = min(arm_dma_pfn_limit, max_low);
117 max_zone_pfn[ZONE_NORMAL] = max_low;
118 #ifdef CONFIG_HIGHMEM
119 max_zone_pfn[ZONE_HIGHMEM] = max_high;
121 free_area_init(max_zone_pfn);
124 #ifdef CONFIG_HAVE_ARCH_PFN_VALID
125 int pfn_valid(unsigned long pfn)
127 phys_addr_t addr = __pfn_to_phys(pfn);
129 if (__phys_to_pfn(addr) != pfn)
132 return memblock_is_map_memory(addr);
134 EXPORT_SYMBOL(pfn_valid);
137 static bool arm_memblock_steal_permitted = true;
139 phys_addr_t __init arm_memblock_steal(phys_addr_t size, phys_addr_t align)
143 BUG_ON(!arm_memblock_steal_permitted);
145 phys = memblock_phys_alloc(size, align);
147 panic("Failed to steal %pa bytes at %pS\n",
148 &size, (void *)_RET_IP_);
150 memblock_free(phys, size);
151 memblock_remove(phys, size);
156 static void __init arm_initrd_init(void)
158 #ifdef CONFIG_BLK_DEV_INITRD
162 initrd_start = initrd_end = 0;
164 if (!phys_initrd_size)
168 * Round the memory region to page boundaries as per free_initrd_mem()
169 * This allows us to detect whether the pages overlapping the initrd
170 * are in use, but more importantly, reserves the entire set of pages
171 * as we don't want these pages allocated for other purposes.
173 start = round_down(phys_initrd_start, PAGE_SIZE);
174 size = phys_initrd_size + (phys_initrd_start - start);
175 size = round_up(size, PAGE_SIZE);
177 if (!memblock_is_region_memory(start, size)) {
178 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region - disabling initrd\n",
183 if (memblock_is_region_reserved(start, size)) {
184 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region - disabling initrd\n",
189 memblock_reserve(start, size);
191 /* Now convert initrd to virtual addresses */
192 initrd_start = __phys_to_virt(phys_initrd_start);
193 initrd_end = initrd_start + phys_initrd_size;
197 #ifdef CONFIG_CPU_ICACHE_MISMATCH_WORKAROUND
198 void check_cpu_icache_size(int cpuid)
202 asm("mrc p15, 0, %0, c0, c0, 1" : "=r" (ctr));
204 size = 1 << ((ctr & 0xf) + 2);
205 if (cpuid != 0 && icache_size != size)
206 pr_info("CPU%u: detected I-Cache line size mismatch, workaround enabled\n",
208 if (icache_size > size)
213 void __init arm_memblock_init(const struct machine_desc *mdesc)
215 /* Register the kernel text, kernel data and initrd with memblock. */
216 memblock_reserve(__pa(KERNEL_START), KERNEL_END - KERNEL_START);
220 arm_mm_memblock_reserve();
222 /* reserve any platform specific memblock areas */
226 early_init_fdt_reserve_self();
227 early_init_fdt_scan_reserved_mem();
229 /* reserve memory for DMA contiguous allocations */
230 dma_contiguous_reserve(arm_dma_limit);
232 arm_memblock_steal_permitted = false;
236 void __init bootmem_init(void)
238 memblock_allow_resize();
240 find_limits(&min_low_pfn, &max_low_pfn, &max_pfn);
242 early_memtest((phys_addr_t)min_low_pfn << PAGE_SHIFT,
243 (phys_addr_t)max_low_pfn << PAGE_SHIFT);
246 * sparse_init() tries to allocate memory from memblock, so must be
247 * done after the fixed reservations
252 * Now free the memory - free_area_init needs
253 * the sparse mem_map arrays initialized by sparse_init()
254 * for memmap_init_zone(), otherwise all PFNs are invalid.
256 zone_sizes_init(min_low_pfn, max_low_pfn, max_pfn);
260 * Poison init memory with an undefined instruction (ARM) or a branch to an
261 * undefined instruction (Thumb).
263 static inline void poison_init_mem(void *s, size_t count)
266 for (; count != 0; count -= 4)
270 static inline void __init
271 free_memmap(unsigned long start_pfn, unsigned long end_pfn)
273 struct page *start_pg, *end_pg;
274 phys_addr_t pg, pgend;
277 * Convert start_pfn/end_pfn to a struct page pointer.
279 start_pg = pfn_to_page(start_pfn - 1) + 1;
280 end_pg = pfn_to_page(end_pfn - 1) + 1;
283 * Convert to physical addresses, and
284 * round start upwards and end downwards.
286 pg = PAGE_ALIGN(__pa(start_pg));
287 pgend = __pa(end_pg) & PAGE_MASK;
290 * If there are free pages between these,
291 * free the section of the memmap array.
294 memblock_free_early(pg, pgend - pg);
298 * The mem_map array can get very big. Free the unused area of the memory map.
300 static void __init free_unused_memmap(void)
302 unsigned long start, end, prev_end = 0;
306 * This relies on each bank being in address order.
307 * The banks are sorted previously in bootmem_init().
309 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) {
310 #ifdef CONFIG_SPARSEMEM
312 * Take care not to free memmap entries that don't exist
313 * due to SPARSEMEM sections which aren't present.
316 ALIGN(prev_end, PAGES_PER_SECTION));
319 * Align down here since the VM subsystem insists that the
320 * memmap entries are valid from the bank start aligned to
321 * MAX_ORDER_NR_PAGES.
323 start = round_down(start, MAX_ORDER_NR_PAGES);
326 * If we had a previous bank, and there is a space
327 * between the current bank and the previous, free it.
329 if (prev_end && prev_end < start)
330 free_memmap(prev_end, start);
333 * Align up here since the VM subsystem insists that the
334 * memmap entries are valid from the bank end aligned to
335 * MAX_ORDER_NR_PAGES.
337 prev_end = ALIGN(end, MAX_ORDER_NR_PAGES);
340 #ifdef CONFIG_SPARSEMEM
341 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
342 free_memmap(prev_end,
343 ALIGN(prev_end, PAGES_PER_SECTION));
347 static void __init free_highpages(void)
349 #ifdef CONFIG_HIGHMEM
350 unsigned long max_low = max_low_pfn;
351 phys_addr_t range_start, range_end;
354 /* set highmem page free */
355 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
356 &range_start, &range_end, NULL) {
357 unsigned long start = PHYS_PFN(range_start);
358 unsigned long end = PHYS_PFN(range_end);
360 /* Ignore complete lowmem entries */
364 /* Truncate partial highmem entries */
368 for (; start < end; start++)
369 free_highmem_page(pfn_to_page(start));
375 * mem_init() marks the free areas in the mem_map and tells us how much
376 * memory is free. This is done after various parts of the system have
377 * claimed their memory after the kernel image.
379 void __init mem_init(void)
381 #ifdef CONFIG_ARM_LPAE
385 set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
387 /* this will put all unused low memory onto the freelists */
388 free_unused_memmap();
392 /* now that our DMA memory is actually so designated, we can free it */
393 free_reserved_area(__va(PHYS_OFFSET), swapper_pg_dir, -1, NULL);
398 mem_init_print_info(NULL);
401 * Check boundaries twice: Some fundamental inconsistencies can
402 * be detected at build time already.
405 BUILD_BUG_ON(TASK_SIZE > MODULES_VADDR);
406 BUG_ON(TASK_SIZE > MODULES_VADDR);
409 #ifdef CONFIG_HIGHMEM
410 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
411 BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
415 #ifdef CONFIG_STRICT_KERNEL_RWX
416 struct section_perm {
425 /* First section-aligned location at or after __start_rodata. */
426 extern char __start_rodata_section_aligned[];
428 static struct section_perm nx_perms[] = {
429 /* Make pages tables, etc before _stext RW (set NX). */
431 .name = "pre-text NX",
432 .start = PAGE_OFFSET,
433 .end = (unsigned long)_stext,
434 .mask = ~PMD_SECT_XN,
437 /* Make init RW (set NX). */
440 .start = (unsigned long)__init_begin,
441 .end = (unsigned long)_sdata,
442 .mask = ~PMD_SECT_XN,
445 /* Make rodata NX (set RO in ro_perms below). */
448 .start = (unsigned long)__start_rodata_section_aligned,
449 .end = (unsigned long)__init_begin,
450 .mask = ~PMD_SECT_XN,
455 static struct section_perm ro_perms[] = {
456 /* Make kernel code and rodata RX (set RO). */
458 .name = "text/rodata RO",
459 .start = (unsigned long)_stext,
460 .end = (unsigned long)__init_begin,
461 #ifdef CONFIG_ARM_LPAE
462 .mask = ~(L_PMD_SECT_RDONLY | PMD_SECT_AP2),
463 .prot = L_PMD_SECT_RDONLY | PMD_SECT_AP2,
465 .mask = ~(PMD_SECT_APX | PMD_SECT_AP_WRITE),
466 .prot = PMD_SECT_APX | PMD_SECT_AP_WRITE,
467 .clear = PMD_SECT_AP_WRITE,
473 * Updates section permissions only for the current mm (sections are
474 * copied into each mm). During startup, this is the init_mm. Is only
475 * safe to be called with preemption disabled, as under stop_machine().
477 static inline void section_update(unsigned long addr, pmdval_t mask,
478 pmdval_t prot, struct mm_struct *mm)
482 pmd = pmd_offset(pud_offset(p4d_offset(pgd_offset(mm, addr), addr), addr), addr);
484 #ifdef CONFIG_ARM_LPAE
485 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
487 if (addr & SECTION_SIZE)
488 pmd[1] = __pmd((pmd_val(pmd[1]) & mask) | prot);
490 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
492 flush_pmd_entry(pmd);
493 local_flush_tlb_kernel_range(addr, addr + SECTION_SIZE);
496 /* Make sure extended page tables are in use. */
497 static inline bool arch_has_strict_perms(void)
499 if (cpu_architecture() < CPU_ARCH_ARMv6)
502 return !!(get_cr() & CR_XP);
505 static void set_section_perms(struct section_perm *perms, int n, bool set,
506 struct mm_struct *mm)
511 if (!arch_has_strict_perms())
514 for (i = 0; i < n; i++) {
515 if (!IS_ALIGNED(perms[i].start, SECTION_SIZE) ||
516 !IS_ALIGNED(perms[i].end, SECTION_SIZE)) {
517 pr_err("BUG: %s section %lx-%lx not aligned to %lx\n",
518 perms[i].name, perms[i].start, perms[i].end,
523 for (addr = perms[i].start;
525 addr += SECTION_SIZE)
526 section_update(addr, perms[i].mask,
527 set ? perms[i].prot : perms[i].clear, mm);
533 * update_sections_early intended to be called only through stop_machine
534 * framework and executed by only one CPU while all other CPUs will spin and
535 * wait, so no locking is required in this function.
537 static void update_sections_early(struct section_perm perms[], int n)
539 struct task_struct *t, *s;
541 for_each_process(t) {
542 if (t->flags & PF_KTHREAD)
544 for_each_thread(t, s)
546 set_section_perms(perms, n, true, s->mm);
548 set_section_perms(perms, n, true, current->active_mm);
549 set_section_perms(perms, n, true, &init_mm);
552 static int __fix_kernmem_perms(void *unused)
554 update_sections_early(nx_perms, ARRAY_SIZE(nx_perms));
558 static void fix_kernmem_perms(void)
560 stop_machine(__fix_kernmem_perms, NULL, NULL);
563 static int __mark_rodata_ro(void *unused)
565 update_sections_early(ro_perms, ARRAY_SIZE(ro_perms));
569 static int kernel_set_to_readonly __read_mostly;
571 void mark_rodata_ro(void)
573 kernel_set_to_readonly = 1;
574 stop_machine(__mark_rodata_ro, NULL, NULL);
578 void set_kernel_text_rw(void)
580 if (!kernel_set_to_readonly)
583 set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), false,
587 void set_kernel_text_ro(void)
589 if (!kernel_set_to_readonly)
592 set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), true,
597 static inline void fix_kernmem_perms(void) { }
598 #endif /* CONFIG_STRICT_KERNEL_RWX */
600 void free_initmem(void)
604 poison_init_mem(__init_begin, __init_end - __init_begin);
605 if (!machine_is_integrator() && !machine_is_cintegrator())
606 free_initmem_default(-1);
609 #ifdef CONFIG_BLK_DEV_INITRD
610 void free_initrd_mem(unsigned long start, unsigned long end)
612 if (start == initrd_start)
613 start = round_down(start, PAGE_SIZE);
614 if (end == initrd_end)
615 end = round_up(end, PAGE_SIZE);
617 poison_init_mem((void *)start, PAGE_ALIGN(end) - start);
618 free_reserved_area((void *)start, (void *)end, -1, "initrd");