Loading...
1/*
2 * linux/arch/arm/mm/init.c
3 *
4 * Copyright (C) 1995-2005 Russell King
5 *
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.
9 */
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>
28
29#include <asm/cp15.h>
30#include <asm/mach-types.h>
31#include <asm/memblock.h>
32#include <asm/memory.h>
33#include <asm/prom.h>
34#include <asm/sections.h>
35#include <asm/setup.h>
36#include <asm/system_info.h>
37#include <asm/tlb.h>
38#include <asm/fixmap.h>
39#include <asm/ptdump.h>
40
41#include <asm/mach/arch.h>
42#include <asm/mach/map.h>
43
44#include "mm.h"
45
46#ifdef CONFIG_CPU_CP15_MMU
47unsigned long __init __clear_cr(unsigned long mask)
48{
49 cr_alignment = cr_alignment & ~mask;
50 return cr_alignment;
51}
52#endif
53
54static phys_addr_t phys_initrd_start __initdata = 0;
55static unsigned long phys_initrd_size __initdata = 0;
56
57static int __init early_initrd(char *p)
58{
59 phys_addr_t start;
60 unsigned long size;
61 char *endp;
62
63 start = memparse(p, &endp);
64 if (*endp == ',') {
65 size = memparse(endp + 1, NULL);
66
67 phys_initrd_start = start;
68 phys_initrd_size = size;
69 }
70 return 0;
71}
72early_param("initrd", early_initrd);
73
74static int __init parse_tag_initrd(const struct tag *tag)
75{
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;
80 return 0;
81}
82
83__tagtable(ATAG_INITRD, parse_tag_initrd);
84
85static int __init parse_tag_initrd2(const struct tag *tag)
86{
87 phys_initrd_start = tag->u.initrd.start;
88 phys_initrd_size = tag->u.initrd.size;
89 return 0;
90}
91
92__tagtable(ATAG_INITRD2, parse_tag_initrd2);
93
94static void __init find_limits(unsigned long *min, unsigned long *max_low,
95 unsigned long *max_high)
96{
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());
100}
101
102#ifdef CONFIG_ZONE_DMA
103
104phys_addr_t arm_dma_zone_size __read_mostly;
105EXPORT_SYMBOL(arm_dma_zone_size);
106
107/*
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.
112 */
113phys_addr_t arm_dma_limit;
114unsigned long arm_dma_pfn_limit;
115
116static void __init arm_adjust_dma_zone(unsigned long *size, unsigned long *hole,
117 unsigned long dma_size)
118{
119 if (size[0] <= dma_size)
120 return;
121
122 size[ZONE_NORMAL] = size[0] - dma_size;
123 size[ZONE_DMA] = dma_size;
124 hole[ZONE_NORMAL] = hole[0];
125 hole[ZONE_DMA] = 0;
126}
127#endif
128
129void __init setup_dma_zone(const struct machine_desc *mdesc)
130{
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;
135 } else
136 arm_dma_limit = 0xffffffff;
137 arm_dma_pfn_limit = arm_dma_limit >> PAGE_SHIFT;
138#endif
139}
140
141static void __init zone_sizes_init(unsigned long min, unsigned long max_low,
142 unsigned long max_high)
143{
144 unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES];
145 struct memblock_region *reg;
146
147 /*
148 * initialise the zones.
149 */
150 memset(zone_size, 0, sizeof(zone_size));
151
152 /*
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.
156 */
157 zone_size[0] = max_low - min;
158#ifdef CONFIG_HIGHMEM
159 zone_size[ZONE_HIGHMEM] = max_high - max_low;
160#endif
161
162 /*
163 * Calculate the size of the holes.
164 * holes = node_size - sum(bank_sizes)
165 */
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);
170
171 if (start < max_low) {
172 unsigned long low_end = min(end, max_low);
173 zhole_size[0] -= low_end - start;
174 }
175#ifdef CONFIG_HIGHMEM
176 if (end > max_low) {
177 unsigned long high_start = max(start, max_low);
178 zhole_size[ZONE_HIGHMEM] -= end - high_start;
179 }
180#endif
181 }
182
183#ifdef CONFIG_ZONE_DMA
184 /*
185 * Adjust the sizes according to any special requirements for
186 * this machine type.
187 */
188 if (arm_dma_zone_size)
189 arm_adjust_dma_zone(zone_size, zhole_size,
190 arm_dma_zone_size >> PAGE_SHIFT);
191#endif
192
193 free_area_init_node(0, zone_size, min, zhole_size);
194}
195
196#ifdef CONFIG_HAVE_ARCH_PFN_VALID
197int pfn_valid(unsigned long pfn)
198{
199 return memblock_is_map_memory(__pfn_to_phys(pfn));
200}
201EXPORT_SYMBOL(pfn_valid);
202#endif
203
204#ifndef CONFIG_SPARSEMEM
205static void __init arm_memory_present(void)
206{
207}
208#else
209static void __init arm_memory_present(void)
210{
211 struct memblock_region *reg;
212
213 for_each_memblock(memory, reg)
214 memory_present(0, memblock_region_memory_base_pfn(reg),
215 memblock_region_memory_end_pfn(reg));
216}
217#endif
218
219static bool arm_memblock_steal_permitted = true;
220
221phys_addr_t __init arm_memblock_steal(phys_addr_t size, phys_addr_t align)
222{
223 phys_addr_t phys;
224
225 BUG_ON(!arm_memblock_steal_permitted);
226
227 phys = memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ANYWHERE);
228 memblock_free(phys, size);
229 memblock_remove(phys, size);
230
231 return phys;
232}
233
234static void __init arm_initrd_init(void)
235{
236#ifdef CONFIG_BLK_DEV_INITRD
237 phys_addr_t start;
238 unsigned long size;
239
240 /* FDT scan will populate initrd_start */
241 if (initrd_start && !phys_initrd_size) {
242 phys_initrd_start = __virt_to_phys(initrd_start);
243 phys_initrd_size = initrd_end - initrd_start;
244 }
245
246 initrd_start = initrd_end = 0;
247
248 if (!phys_initrd_size)
249 return;
250
251 /*
252 * Round the memory region to page boundaries as per free_initrd_mem()
253 * This allows us to detect whether the pages overlapping the initrd
254 * are in use, but more importantly, reserves the entire set of pages
255 * as we don't want these pages allocated for other purposes.
256 */
257 start = round_down(phys_initrd_start, PAGE_SIZE);
258 size = phys_initrd_size + (phys_initrd_start - start);
259 size = round_up(size, PAGE_SIZE);
260
261 if (!memblock_is_region_memory(start, size)) {
262 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region - disabling initrd\n",
263 (u64)start, size);
264 return;
265 }
266
267 if (memblock_is_region_reserved(start, size)) {
268 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region - disabling initrd\n",
269 (u64)start, size);
270 return;
271 }
272
273 memblock_reserve(start, size);
274
275 /* Now convert initrd to virtual addresses */
276 initrd_start = __phys_to_virt(phys_initrd_start);
277 initrd_end = initrd_start + phys_initrd_size;
278#endif
279}
280
281void __init arm_memblock_init(const struct machine_desc *mdesc)
282{
283 /* Register the kernel text, kernel data and initrd with memblock. */
284 memblock_reserve(__pa(KERNEL_START), KERNEL_END - KERNEL_START);
285
286 arm_initrd_init();
287
288 arm_mm_memblock_reserve();
289
290 /* reserve any platform specific memblock areas */
291 if (mdesc->reserve)
292 mdesc->reserve();
293
294 early_init_fdt_reserve_self();
295 early_init_fdt_scan_reserved_mem();
296
297 /* reserve memory for DMA contiguous allocations */
298 dma_contiguous_reserve(arm_dma_limit);
299
300 arm_memblock_steal_permitted = false;
301 memblock_dump_all();
302}
303
304void __init bootmem_init(void)
305{
306 unsigned long min, max_low, max_high;
307
308 memblock_allow_resize();
309 max_low = max_high = 0;
310
311 find_limits(&min, &max_low, &max_high);
312
313 early_memtest((phys_addr_t)min << PAGE_SHIFT,
314 (phys_addr_t)max_low << PAGE_SHIFT);
315
316 /*
317 * Sparsemem tries to allocate bootmem in memory_present(),
318 * so must be done after the fixed reservations
319 */
320 arm_memory_present();
321
322 /*
323 * sparse_init() needs the bootmem allocator up and running.
324 */
325 sparse_init();
326
327 /*
328 * Now free the memory - free_area_init_node needs
329 * the sparse mem_map arrays initialized by sparse_init()
330 * for memmap_init_zone(), otherwise all PFNs are invalid.
331 */
332 zone_sizes_init(min, max_low, max_high);
333
334 /*
335 * This doesn't seem to be used by the Linux memory manager any
336 * more, but is used by ll_rw_block. If we can get rid of it, we
337 * also get rid of some of the stuff above as well.
338 */
339 min_low_pfn = min;
340 max_low_pfn = max_low;
341 max_pfn = max_high;
342}
343
344/*
345 * Poison init memory with an undefined instruction (ARM) or a branch to an
346 * undefined instruction (Thumb).
347 */
348static inline void poison_init_mem(void *s, size_t count)
349{
350 u32 *p = (u32 *)s;
351 for (; count != 0; count -= 4)
352 *p++ = 0xe7fddef0;
353}
354
355static inline void
356free_memmap(unsigned long start_pfn, unsigned long end_pfn)
357{
358 struct page *start_pg, *end_pg;
359 phys_addr_t pg, pgend;
360
361 /*
362 * Convert start_pfn/end_pfn to a struct page pointer.
363 */
364 start_pg = pfn_to_page(start_pfn - 1) + 1;
365 end_pg = pfn_to_page(end_pfn - 1) + 1;
366
367 /*
368 * Convert to physical addresses, and
369 * round start upwards and end downwards.
370 */
371 pg = PAGE_ALIGN(__pa(start_pg));
372 pgend = __pa(end_pg) & PAGE_MASK;
373
374 /*
375 * If there are free pages between these,
376 * free the section of the memmap array.
377 */
378 if (pg < pgend)
379 memblock_free_early(pg, pgend - pg);
380}
381
382/*
383 * The mem_map array can get very big. Free the unused area of the memory map.
384 */
385static void __init free_unused_memmap(void)
386{
387 unsigned long start, prev_end = 0;
388 struct memblock_region *reg;
389
390 /*
391 * This relies on each bank being in address order.
392 * The banks are sorted previously in bootmem_init().
393 */
394 for_each_memblock(memory, reg) {
395 start = memblock_region_memory_base_pfn(reg);
396
397#ifdef CONFIG_SPARSEMEM
398 /*
399 * Take care not to free memmap entries that don't exist
400 * due to SPARSEMEM sections which aren't present.
401 */
402 start = min(start,
403 ALIGN(prev_end, PAGES_PER_SECTION));
404#else
405 /*
406 * Align down here since the VM subsystem insists that the
407 * memmap entries are valid from the bank start aligned to
408 * MAX_ORDER_NR_PAGES.
409 */
410 start = round_down(start, MAX_ORDER_NR_PAGES);
411#endif
412 /*
413 * If we had a previous bank, and there is a space
414 * between the current bank and the previous, free it.
415 */
416 if (prev_end && prev_end < start)
417 free_memmap(prev_end, start);
418
419 /*
420 * Align up here since the VM subsystem insists that the
421 * memmap entries are valid from the bank end aligned to
422 * MAX_ORDER_NR_PAGES.
423 */
424 prev_end = ALIGN(memblock_region_memory_end_pfn(reg),
425 MAX_ORDER_NR_PAGES);
426 }
427
428#ifdef CONFIG_SPARSEMEM
429 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
430 free_memmap(prev_end,
431 ALIGN(prev_end, PAGES_PER_SECTION));
432#endif
433}
434
435#ifdef CONFIG_HIGHMEM
436static inline void free_area_high(unsigned long pfn, unsigned long end)
437{
438 for (; pfn < end; pfn++)
439 free_highmem_page(pfn_to_page(pfn));
440}
441#endif
442
443static void __init free_highpages(void)
444{
445#ifdef CONFIG_HIGHMEM
446 unsigned long max_low = max_low_pfn;
447 struct memblock_region *mem, *res;
448
449 /* set highmem page free */
450 for_each_memblock(memory, mem) {
451 unsigned long start = memblock_region_memory_base_pfn(mem);
452 unsigned long end = memblock_region_memory_end_pfn(mem);
453
454 /* Ignore complete lowmem entries */
455 if (end <= max_low)
456 continue;
457
458 if (memblock_is_nomap(mem))
459 continue;
460
461 /* Truncate partial highmem entries */
462 if (start < max_low)
463 start = max_low;
464
465 /* Find and exclude any reserved regions */
466 for_each_memblock(reserved, res) {
467 unsigned long res_start, res_end;
468
469 res_start = memblock_region_reserved_base_pfn(res);
470 res_end = memblock_region_reserved_end_pfn(res);
471
472 if (res_end < start)
473 continue;
474 if (res_start < start)
475 res_start = start;
476 if (res_start > end)
477 res_start = end;
478 if (res_end > end)
479 res_end = end;
480 if (res_start != start)
481 free_area_high(start, res_start);
482 start = res_end;
483 if (start == end)
484 break;
485 }
486
487 /* And now free anything which remains */
488 if (start < end)
489 free_area_high(start, end);
490 }
491#endif
492}
493
494/*
495 * mem_init() marks the free areas in the mem_map and tells us how much
496 * memory is free. This is done after various parts of the system have
497 * claimed their memory after the kernel image.
498 */
499void __init mem_init(void)
500{
501#ifdef CONFIG_HAVE_TCM
502 /* These pointers are filled in on TCM detection */
503 extern u32 dtcm_end;
504 extern u32 itcm_end;
505#endif
506
507 set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
508
509 /* this will put all unused low memory onto the freelists */
510 free_unused_memmap();
511 free_all_bootmem();
512
513#ifdef CONFIG_SA1111
514 /* now that our DMA memory is actually so designated, we can free it */
515 free_reserved_area(__va(PHYS_OFFSET), swapper_pg_dir, -1, NULL);
516#endif
517
518 free_highpages();
519
520 mem_init_print_info(NULL);
521
522#define MLK(b, t) b, t, ((t) - (b)) >> 10
523#define MLM(b, t) b, t, ((t) - (b)) >> 20
524#define MLK_ROUNDUP(b, t) b, t, DIV_ROUND_UP(((t) - (b)), SZ_1K)
525
526 pr_notice("Virtual kernel memory layout:\n"
527 " vector : 0x%08lx - 0x%08lx (%4ld kB)\n"
528#ifdef CONFIG_HAVE_TCM
529 " DTCM : 0x%08lx - 0x%08lx (%4ld kB)\n"
530 " ITCM : 0x%08lx - 0x%08lx (%4ld kB)\n"
531#endif
532 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
533 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
534 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
535#ifdef CONFIG_HIGHMEM
536 " pkmap : 0x%08lx - 0x%08lx (%4ld MB)\n"
537#endif
538#ifdef CONFIG_MODULES
539 " modules : 0x%08lx - 0x%08lx (%4ld MB)\n"
540#endif
541 " .text : 0x%p" " - 0x%p" " (%4td kB)\n"
542 " .init : 0x%p" " - 0x%p" " (%4td kB)\n"
543 " .data : 0x%p" " - 0x%p" " (%4td kB)\n"
544 " .bss : 0x%p" " - 0x%p" " (%4td kB)\n",
545
546 MLK(VECTORS_BASE, VECTORS_BASE + PAGE_SIZE),
547#ifdef CONFIG_HAVE_TCM
548 MLK(DTCM_OFFSET, (unsigned long) dtcm_end),
549 MLK(ITCM_OFFSET, (unsigned long) itcm_end),
550#endif
551 MLK(FIXADDR_START, FIXADDR_END),
552 MLM(VMALLOC_START, VMALLOC_END),
553 MLM(PAGE_OFFSET, (unsigned long)high_memory),
554#ifdef CONFIG_HIGHMEM
555 MLM(PKMAP_BASE, (PKMAP_BASE) + (LAST_PKMAP) *
556 (PAGE_SIZE)),
557#endif
558#ifdef CONFIG_MODULES
559 MLM(MODULES_VADDR, MODULES_END),
560#endif
561
562 MLK_ROUNDUP(_text, _etext),
563 MLK_ROUNDUP(__init_begin, __init_end),
564 MLK_ROUNDUP(_sdata, _edata),
565 MLK_ROUNDUP(__bss_start, __bss_stop));
566
567#undef MLK
568#undef MLM
569#undef MLK_ROUNDUP
570
571 /*
572 * Check boundaries twice: Some fundamental inconsistencies can
573 * be detected at build time already.
574 */
575#ifdef CONFIG_MMU
576 BUILD_BUG_ON(TASK_SIZE > MODULES_VADDR);
577 BUG_ON(TASK_SIZE > MODULES_VADDR);
578#endif
579
580#ifdef CONFIG_HIGHMEM
581 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
582 BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
583#endif
584}
585
586#ifdef CONFIG_STRICT_KERNEL_RWX
587struct section_perm {
588 const char *name;
589 unsigned long start;
590 unsigned long end;
591 pmdval_t mask;
592 pmdval_t prot;
593 pmdval_t clear;
594};
595
596/* First section-aligned location at or after __start_rodata. */
597extern char __start_rodata_section_aligned[];
598
599static struct section_perm nx_perms[] = {
600 /* Make pages tables, etc before _stext RW (set NX). */
601 {
602 .name = "pre-text NX",
603 .start = PAGE_OFFSET,
604 .end = (unsigned long)_stext,
605 .mask = ~PMD_SECT_XN,
606 .prot = PMD_SECT_XN,
607 },
608 /* Make init RW (set NX). */
609 {
610 .name = "init NX",
611 .start = (unsigned long)__init_begin,
612 .end = (unsigned long)_sdata,
613 .mask = ~PMD_SECT_XN,
614 .prot = PMD_SECT_XN,
615 },
616 /* Make rodata NX (set RO in ro_perms below). */
617 {
618 .name = "rodata NX",
619 .start = (unsigned long)__start_rodata_section_aligned,
620 .end = (unsigned long)__init_begin,
621 .mask = ~PMD_SECT_XN,
622 .prot = PMD_SECT_XN,
623 },
624};
625
626static struct section_perm ro_perms[] = {
627 /* Make kernel code and rodata RX (set RO). */
628 {
629 .name = "text/rodata RO",
630 .start = (unsigned long)_stext,
631 .end = (unsigned long)__init_begin,
632#ifdef CONFIG_ARM_LPAE
633 .mask = ~(L_PMD_SECT_RDONLY | PMD_SECT_AP2),
634 .prot = L_PMD_SECT_RDONLY | PMD_SECT_AP2,
635#else
636 .mask = ~(PMD_SECT_APX | PMD_SECT_AP_WRITE),
637 .prot = PMD_SECT_APX | PMD_SECT_AP_WRITE,
638 .clear = PMD_SECT_AP_WRITE,
639#endif
640 },
641};
642
643/*
644 * Updates section permissions only for the current mm (sections are
645 * copied into each mm). During startup, this is the init_mm. Is only
646 * safe to be called with preemption disabled, as under stop_machine().
647 */
648static inline void section_update(unsigned long addr, pmdval_t mask,
649 pmdval_t prot, struct mm_struct *mm)
650{
651 pmd_t *pmd;
652
653 pmd = pmd_offset(pud_offset(pgd_offset(mm, addr), addr), addr);
654
655#ifdef CONFIG_ARM_LPAE
656 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
657#else
658 if (addr & SECTION_SIZE)
659 pmd[1] = __pmd((pmd_val(pmd[1]) & mask) | prot);
660 else
661 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
662#endif
663 flush_pmd_entry(pmd);
664 local_flush_tlb_kernel_range(addr, addr + SECTION_SIZE);
665}
666
667/* Make sure extended page tables are in use. */
668static inline bool arch_has_strict_perms(void)
669{
670 if (cpu_architecture() < CPU_ARCH_ARMv6)
671 return false;
672
673 return !!(get_cr() & CR_XP);
674}
675
676void set_section_perms(struct section_perm *perms, int n, bool set,
677 struct mm_struct *mm)
678{
679 size_t i;
680 unsigned long addr;
681
682 if (!arch_has_strict_perms())
683 return;
684
685 for (i = 0; i < n; i++) {
686 if (!IS_ALIGNED(perms[i].start, SECTION_SIZE) ||
687 !IS_ALIGNED(perms[i].end, SECTION_SIZE)) {
688 pr_err("BUG: %s section %lx-%lx not aligned to %lx\n",
689 perms[i].name, perms[i].start, perms[i].end,
690 SECTION_SIZE);
691 continue;
692 }
693
694 for (addr = perms[i].start;
695 addr < perms[i].end;
696 addr += SECTION_SIZE)
697 section_update(addr, perms[i].mask,
698 set ? perms[i].prot : perms[i].clear, mm);
699 }
700
701}
702
703/**
704 * update_sections_early intended to be called only through stop_machine
705 * framework and executed by only one CPU while all other CPUs will spin and
706 * wait, so no locking is required in this function.
707 */
708static void update_sections_early(struct section_perm perms[], int n)
709{
710 struct task_struct *t, *s;
711
712 for_each_process(t) {
713 if (t->flags & PF_KTHREAD)
714 continue;
715 for_each_thread(t, s)
716 set_section_perms(perms, n, true, s->mm);
717 }
718 set_section_perms(perms, n, true, current->active_mm);
719 set_section_perms(perms, n, true, &init_mm);
720}
721
722static int __fix_kernmem_perms(void *unused)
723{
724 update_sections_early(nx_perms, ARRAY_SIZE(nx_perms));
725 return 0;
726}
727
728static void fix_kernmem_perms(void)
729{
730 stop_machine(__fix_kernmem_perms, NULL, NULL);
731}
732
733static int __mark_rodata_ro(void *unused)
734{
735 update_sections_early(ro_perms, ARRAY_SIZE(ro_perms));
736 return 0;
737}
738
739void mark_rodata_ro(void)
740{
741 stop_machine(__mark_rodata_ro, NULL, NULL);
742 debug_checkwx();
743}
744
745void set_kernel_text_rw(void)
746{
747 set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), false,
748 current->active_mm);
749}
750
751void set_kernel_text_ro(void)
752{
753 set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), true,
754 current->active_mm);
755}
756
757#else
758static inline void fix_kernmem_perms(void) { }
759#endif /* CONFIG_STRICT_KERNEL_RWX */
760
761void free_initmem(void)
762{
763 fix_kernmem_perms();
764
765 poison_init_mem(__init_begin, __init_end - __init_begin);
766 if (!machine_is_integrator() && !machine_is_cintegrator())
767 free_initmem_default(-1);
768}
769
770#ifdef CONFIG_BLK_DEV_INITRD
771
772static int keep_initrd;
773
774void free_initrd_mem(unsigned long start, unsigned long end)
775{
776 if (!keep_initrd) {
777 if (start == initrd_start)
778 start = round_down(start, PAGE_SIZE);
779 if (end == initrd_end)
780 end = round_up(end, PAGE_SIZE);
781
782 poison_init_mem((void *)start, PAGE_ALIGN(end) - start);
783 free_reserved_area((void *)start, (void *)end, -1, "initrd");
784 }
785}
786
787static int __init keepinitrd_setup(char *__unused)
788{
789 keep_initrd = 1;
790 return 1;
791}
792
793__setup("keepinitrd", keepinitrd_setup);
794#endif
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/arch/arm/mm/init.c
4 *
5 * Copyright (C) 1995-2005 Russell King
6 */
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>
25
26#include <asm/cp15.h>
27#include <asm/mach-types.h>
28#include <asm/memblock.h>
29#include <asm/page.h>
30#include <asm/prom.h>
31#include <asm/sections.h>
32#include <asm/setup.h>
33#include <asm/set_memory.h>
34#include <asm/system_info.h>
35#include <asm/tlb.h>
36#include <asm/fixmap.h>
37#include <asm/ptdump.h>
38
39#include <asm/mach/arch.h>
40#include <asm/mach/map.h>
41
42#include "mm.h"
43
44#ifdef CONFIG_CPU_CP15_MMU
45unsigned long __init __clear_cr(unsigned long mask)
46{
47 cr_alignment = cr_alignment & ~mask;
48 return cr_alignment;
49}
50#endif
51
52#ifdef CONFIG_BLK_DEV_INITRD
53static int __init parse_tag_initrd(const struct tag *tag)
54{
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;
59 return 0;
60}
61
62__tagtable(ATAG_INITRD, parse_tag_initrd);
63
64static int __init parse_tag_initrd2(const struct tag *tag)
65{
66 phys_initrd_start = tag->u.initrd.start;
67 phys_initrd_size = tag->u.initrd.size;
68 return 0;
69}
70
71__tagtable(ATAG_INITRD2, parse_tag_initrd2);
72#endif
73
74static void __init find_limits(unsigned long *min, unsigned long *max_low,
75 unsigned long *max_high)
76{
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());
80}
81
82#ifdef CONFIG_ZONE_DMA
83
84phys_addr_t arm_dma_zone_size __read_mostly;
85EXPORT_SYMBOL(arm_dma_zone_size);
86
87/*
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.
92 */
93phys_addr_t arm_dma_limit;
94unsigned long arm_dma_pfn_limit;
95#endif
96
97void __init setup_dma_zone(const struct machine_desc *mdesc)
98{
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;
103 } else
104 arm_dma_limit = 0xffffffff;
105 arm_dma_pfn_limit = arm_dma_limit >> PAGE_SHIFT;
106#endif
107}
108
109static void __init zone_sizes_init(unsigned long min, unsigned long max_low,
110 unsigned long max_high)
111{
112 unsigned long max_zone_pfn[MAX_NR_ZONES] = { 0 };
113
114#ifdef CONFIG_ZONE_DMA
115 max_zone_pfn[ZONE_DMA] = min(arm_dma_pfn_limit, max_low);
116#endif
117 max_zone_pfn[ZONE_NORMAL] = max_low;
118#ifdef CONFIG_HIGHMEM
119 max_zone_pfn[ZONE_HIGHMEM] = max_high;
120#endif
121 free_area_init(max_zone_pfn);
122}
123
124#ifdef CONFIG_HAVE_ARCH_PFN_VALID
125int pfn_valid(unsigned long pfn)
126{
127 phys_addr_t addr = __pfn_to_phys(pfn);
128 unsigned long pageblock_size = PAGE_SIZE * pageblock_nr_pages;
129
130 if (__phys_to_pfn(addr) != pfn)
131 return 0;
132
133 /*
134 * If address less than pageblock_size bytes away from a present
135 * memory chunk there still will be a memory map entry for it
136 * because we round freed memory map to the pageblock boundaries.
137 */
138 if (memblock_overlaps_region(&memblock.memory,
139 ALIGN_DOWN(addr, pageblock_size),
140 pageblock_size))
141 return 1;
142
143 return 0;
144}
145EXPORT_SYMBOL(pfn_valid);
146#endif
147
148static bool arm_memblock_steal_permitted = true;
149
150phys_addr_t __init arm_memblock_steal(phys_addr_t size, phys_addr_t align)
151{
152 phys_addr_t phys;
153
154 BUG_ON(!arm_memblock_steal_permitted);
155
156 phys = memblock_phys_alloc(size, align);
157 if (!phys)
158 panic("Failed to steal %pa bytes at %pS\n",
159 &size, (void *)_RET_IP_);
160
161 memblock_phys_free(phys, size);
162 memblock_remove(phys, size);
163
164 return phys;
165}
166
167#ifdef CONFIG_CPU_ICACHE_MISMATCH_WORKAROUND
168void check_cpu_icache_size(int cpuid)
169{
170 u32 size, ctr;
171
172 asm("mrc p15, 0, %0, c0, c0, 1" : "=r" (ctr));
173
174 size = 1 << ((ctr & 0xf) + 2);
175 if (cpuid != 0 && icache_size != size)
176 pr_info("CPU%u: detected I-Cache line size mismatch, workaround enabled\n",
177 cpuid);
178 if (icache_size > size)
179 icache_size = size;
180}
181#endif
182
183void __init arm_memblock_init(const struct machine_desc *mdesc)
184{
185 /* Register the kernel text, kernel data and initrd with memblock. */
186 memblock_reserve(__pa(KERNEL_START), KERNEL_END - KERNEL_START);
187
188 reserve_initrd_mem();
189
190 arm_mm_memblock_reserve();
191
192 /* reserve any platform specific memblock areas */
193 if (mdesc->reserve)
194 mdesc->reserve();
195
196 early_init_fdt_scan_reserved_mem();
197
198 /* reserve memory for DMA contiguous allocations */
199 dma_contiguous_reserve(arm_dma_limit);
200
201 arm_memblock_steal_permitted = false;
202 memblock_dump_all();
203}
204
205void __init bootmem_init(void)
206{
207 memblock_allow_resize();
208
209 find_limits(&min_low_pfn, &max_low_pfn, &max_pfn);
210
211 early_memtest((phys_addr_t)min_low_pfn << PAGE_SHIFT,
212 (phys_addr_t)max_low_pfn << PAGE_SHIFT);
213
214 /*
215 * sparse_init() tries to allocate memory from memblock, so must be
216 * done after the fixed reservations
217 */
218 sparse_init();
219
220 /*
221 * Now free the memory - free_area_init needs
222 * the sparse mem_map arrays initialized by sparse_init()
223 * for memmap_init_zone(), otherwise all PFNs are invalid.
224 */
225 zone_sizes_init(min_low_pfn, max_low_pfn, max_pfn);
226}
227
228/*
229 * Poison init memory with an undefined instruction (ARM) or a branch to an
230 * undefined instruction (Thumb).
231 */
232static inline void poison_init_mem(void *s, size_t count)
233{
234 u32 *p = (u32 *)s;
235 for (; count != 0; count -= 4)
236 *p++ = 0xe7fddef0;
237}
238
239static void __init free_highpages(void)
240{
241#ifdef CONFIG_HIGHMEM
242 unsigned long max_low = max_low_pfn;
243 phys_addr_t range_start, range_end;
244 u64 i;
245
246 /* set highmem page free */
247 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
248 &range_start, &range_end, NULL) {
249 unsigned long start = PFN_UP(range_start);
250 unsigned long end = PFN_DOWN(range_end);
251
252 /* Ignore complete lowmem entries */
253 if (end <= max_low)
254 continue;
255
256 /* Truncate partial highmem entries */
257 if (start < max_low)
258 start = max_low;
259
260 for (; start < end; start++)
261 free_highmem_page(pfn_to_page(start));
262 }
263#endif
264}
265
266/*
267 * mem_init() marks the free areas in the mem_map and tells us how much
268 * memory is free. This is done after various parts of the system have
269 * claimed their memory after the kernel image.
270 */
271void __init mem_init(void)
272{
273#ifdef CONFIG_ARM_LPAE
274 swiotlb_init(max_pfn > arm_dma_pfn_limit, SWIOTLB_VERBOSE);
275#endif
276
277 set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
278
279 /* this will put all unused low memory onto the freelists */
280 memblock_free_all();
281
282#ifdef CONFIG_SA1111
283 /* now that our DMA memory is actually so designated, we can free it */
284 free_reserved_area(__va(PHYS_OFFSET), swapper_pg_dir, -1, NULL);
285#endif
286
287 free_highpages();
288
289 /*
290 * Check boundaries twice: Some fundamental inconsistencies can
291 * be detected at build time already.
292 */
293#ifdef CONFIG_MMU
294 BUILD_BUG_ON(TASK_SIZE > MODULES_VADDR);
295 BUG_ON(TASK_SIZE > MODULES_VADDR);
296#endif
297
298#ifdef CONFIG_HIGHMEM
299 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
300 BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
301#endif
302}
303
304#ifdef CONFIG_STRICT_KERNEL_RWX
305struct section_perm {
306 const char *name;
307 unsigned long start;
308 unsigned long end;
309 pmdval_t mask;
310 pmdval_t prot;
311 pmdval_t clear;
312};
313
314/* First section-aligned location at or after __start_rodata. */
315extern char __start_rodata_section_aligned[];
316
317static struct section_perm nx_perms[] = {
318 /* Make pages tables, etc before _stext RW (set NX). */
319 {
320 .name = "pre-text NX",
321 .start = PAGE_OFFSET,
322 .end = (unsigned long)_stext,
323 .mask = ~PMD_SECT_XN,
324 .prot = PMD_SECT_XN,
325 },
326 /* Make init RW (set NX). */
327 {
328 .name = "init NX",
329 .start = (unsigned long)__init_begin,
330 .end = (unsigned long)_sdata,
331 .mask = ~PMD_SECT_XN,
332 .prot = PMD_SECT_XN,
333 },
334 /* Make rodata NX (set RO in ro_perms below). */
335 {
336 .name = "rodata NX",
337 .start = (unsigned long)__start_rodata_section_aligned,
338 .end = (unsigned long)__init_begin,
339 .mask = ~PMD_SECT_XN,
340 .prot = PMD_SECT_XN,
341 },
342};
343
344static struct section_perm ro_perms[] = {
345 /* Make kernel code and rodata RX (set RO). */
346 {
347 .name = "text/rodata RO",
348 .start = (unsigned long)_stext,
349 .end = (unsigned long)__init_begin,
350#ifdef CONFIG_ARM_LPAE
351 .mask = ~(L_PMD_SECT_RDONLY | PMD_SECT_AP2),
352 .prot = L_PMD_SECT_RDONLY | PMD_SECT_AP2,
353#else
354 .mask = ~(PMD_SECT_APX | PMD_SECT_AP_WRITE),
355 .prot = PMD_SECT_APX | PMD_SECT_AP_WRITE,
356 .clear = PMD_SECT_AP_WRITE,
357#endif
358 },
359};
360
361/*
362 * Updates section permissions only for the current mm (sections are
363 * copied into each mm). During startup, this is the init_mm. Is only
364 * safe to be called with preemption disabled, as under stop_machine().
365 */
366static inline void section_update(unsigned long addr, pmdval_t mask,
367 pmdval_t prot, struct mm_struct *mm)
368{
369 pmd_t *pmd;
370
371 pmd = pmd_offset(pud_offset(p4d_offset(pgd_offset(mm, addr), addr), addr), addr);
372
373#ifdef CONFIG_ARM_LPAE
374 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
375#else
376 if (addr & SECTION_SIZE)
377 pmd[1] = __pmd((pmd_val(pmd[1]) & mask) | prot);
378 else
379 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
380#endif
381 flush_pmd_entry(pmd);
382 local_flush_tlb_kernel_range(addr, addr + SECTION_SIZE);
383}
384
385/* Make sure extended page tables are in use. */
386static inline bool arch_has_strict_perms(void)
387{
388 if (cpu_architecture() < CPU_ARCH_ARMv6)
389 return false;
390
391 return !!(get_cr() & CR_XP);
392}
393
394static void set_section_perms(struct section_perm *perms, int n, bool set,
395 struct mm_struct *mm)
396{
397 size_t i;
398 unsigned long addr;
399
400 if (!arch_has_strict_perms())
401 return;
402
403 for (i = 0; i < n; i++) {
404 if (!IS_ALIGNED(perms[i].start, SECTION_SIZE) ||
405 !IS_ALIGNED(perms[i].end, SECTION_SIZE)) {
406 pr_err("BUG: %s section %lx-%lx not aligned to %lx\n",
407 perms[i].name, perms[i].start, perms[i].end,
408 SECTION_SIZE);
409 continue;
410 }
411
412 for (addr = perms[i].start;
413 addr < perms[i].end;
414 addr += SECTION_SIZE)
415 section_update(addr, perms[i].mask,
416 set ? perms[i].prot : perms[i].clear, mm);
417 }
418
419}
420
421/*
422 * update_sections_early intended to be called only through stop_machine
423 * framework and executed by only one CPU while all other CPUs will spin and
424 * wait, so no locking is required in this function.
425 */
426static void update_sections_early(struct section_perm perms[], int n)
427{
428 struct task_struct *t, *s;
429
430 for_each_process(t) {
431 if (t->flags & PF_KTHREAD)
432 continue;
433 for_each_thread(t, s)
434 if (s->mm)
435 set_section_perms(perms, n, true, s->mm);
436 }
437 set_section_perms(perms, n, true, current->active_mm);
438 set_section_perms(perms, n, true, &init_mm);
439}
440
441static int __fix_kernmem_perms(void *unused)
442{
443 update_sections_early(nx_perms, ARRAY_SIZE(nx_perms));
444 return 0;
445}
446
447static void fix_kernmem_perms(void)
448{
449 stop_machine(__fix_kernmem_perms, NULL, NULL);
450}
451
452static int __mark_rodata_ro(void *unused)
453{
454 update_sections_early(ro_perms, ARRAY_SIZE(ro_perms));
455 return 0;
456}
457
458void mark_rodata_ro(void)
459{
460 stop_machine(__mark_rodata_ro, NULL, NULL);
461 arm_debug_checkwx();
462}
463
464#else
465static inline void fix_kernmem_perms(void) { }
466#endif /* CONFIG_STRICT_KERNEL_RWX */
467
468void free_initmem(void)
469{
470 fix_kernmem_perms();
471
472 poison_init_mem(__init_begin, __init_end - __init_begin);
473 if (!machine_is_integrator() && !machine_is_cintegrator())
474 free_initmem_default(-1);
475}
476
477#ifdef CONFIG_BLK_DEV_INITRD
478void free_initrd_mem(unsigned long start, unsigned long end)
479{
480 if (start == initrd_start)
481 start = round_down(start, PAGE_SIZE);
482 if (end == initrd_end)
483 end = round_up(end, PAGE_SIZE);
484
485 poison_init_mem((void *)start, PAGE_ALIGN(end) - start);
486 free_reserved_area((void *)start, (void *)end, -1, "initrd");
487}
488#endif