treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 500
[sfrench/cifs-2.6.git] / arch / arm / mm / init.c
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-contiguous.h>
22 #include <linux/sizes.h>
23 #include <linux/stop_machine.h>
24
25 #include <asm/cp15.h>
26 #include <asm/mach-types.h>
27 #include <asm/memblock.h>
28 #include <asm/memory.h>
29 #include <asm/prom.h>
30 #include <asm/sections.h>
31 #include <asm/setup.h>
32 #include <asm/system_info.h>
33 #include <asm/tlb.h>
34 #include <asm/fixmap.h>
35 #include <asm/ptdump.h>
36
37 #include <asm/mach/arch.h>
38 #include <asm/mach/map.h>
39
40 #include "mm.h"
41
42 #ifdef CONFIG_CPU_CP15_MMU
43 unsigned long __init __clear_cr(unsigned long mask)
44 {
45         cr_alignment = cr_alignment & ~mask;
46         return cr_alignment;
47 }
48 #endif
49
50 #ifdef CONFIG_BLK_DEV_INITRD
51 static int __init parse_tag_initrd(const struct tag *tag)
52 {
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;
57         return 0;
58 }
59
60 __tagtable(ATAG_INITRD, parse_tag_initrd);
61
62 static int __init parse_tag_initrd2(const struct tag *tag)
63 {
64         phys_initrd_start = tag->u.initrd.start;
65         phys_initrd_size = tag->u.initrd.size;
66         return 0;
67 }
68
69 __tagtable(ATAG_INITRD2, parse_tag_initrd2);
70 #endif
71
72 static void __init find_limits(unsigned long *min, unsigned long *max_low,
73                                unsigned long *max_high)
74 {
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());
78 }
79
80 #ifdef CONFIG_ZONE_DMA
81
82 phys_addr_t arm_dma_zone_size __read_mostly;
83 EXPORT_SYMBOL(arm_dma_zone_size);
84
85 /*
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.
90  */
91 phys_addr_t arm_dma_limit;
92 unsigned long arm_dma_pfn_limit;
93
94 static void __init arm_adjust_dma_zone(unsigned long *size, unsigned long *hole,
95         unsigned long dma_size)
96 {
97         if (size[0] <= dma_size)
98                 return;
99
100         size[ZONE_NORMAL] = size[0] - dma_size;
101         size[ZONE_DMA] = dma_size;
102         hole[ZONE_NORMAL] = hole[0];
103         hole[ZONE_DMA] = 0;
104 }
105 #endif
106
107 void __init setup_dma_zone(const struct machine_desc *mdesc)
108 {
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;
113         } else
114                 arm_dma_limit = 0xffffffff;
115         arm_dma_pfn_limit = arm_dma_limit >> PAGE_SHIFT;
116 #endif
117 }
118
119 static void __init zone_sizes_init(unsigned long min, unsigned long max_low,
120         unsigned long max_high)
121 {
122         unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES];
123         struct memblock_region *reg;
124
125         /*
126          * initialise the zones.
127          */
128         memset(zone_size, 0, sizeof(zone_size));
129
130         /*
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.
134          */
135         zone_size[0] = max_low - min;
136 #ifdef CONFIG_HIGHMEM
137         zone_size[ZONE_HIGHMEM] = max_high - max_low;
138 #endif
139
140         /*
141          * Calculate the size of the holes.
142          *  holes = node_size - sum(bank_sizes)
143          */
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);
148
149                 if (start < max_low) {
150                         unsigned long low_end = min(end, max_low);
151                         zhole_size[0] -= low_end - start;
152                 }
153 #ifdef CONFIG_HIGHMEM
154                 if (end > max_low) {
155                         unsigned long high_start = max(start, max_low);
156                         zhole_size[ZONE_HIGHMEM] -= end - high_start;
157                 }
158 #endif
159         }
160
161 #ifdef CONFIG_ZONE_DMA
162         /*
163          * Adjust the sizes according to any special requirements for
164          * this machine type.
165          */
166         if (arm_dma_zone_size)
167                 arm_adjust_dma_zone(zone_size, zhole_size,
168                         arm_dma_zone_size >> PAGE_SHIFT);
169 #endif
170
171         free_area_init_node(0, zone_size, min, zhole_size);
172 }
173
174 #ifdef CONFIG_HAVE_ARCH_PFN_VALID
175 int pfn_valid(unsigned long pfn)
176 {
177         return memblock_is_map_memory(__pfn_to_phys(pfn));
178 }
179 EXPORT_SYMBOL(pfn_valid);
180 #endif
181
182 static bool arm_memblock_steal_permitted = true;
183
184 phys_addr_t __init arm_memblock_steal(phys_addr_t size, phys_addr_t align)
185 {
186         phys_addr_t phys;
187
188         BUG_ON(!arm_memblock_steal_permitted);
189
190         phys = memblock_phys_alloc(size, align);
191         if (!phys)
192                 panic("Failed to steal %pa bytes at %pS\n",
193                       &size, (void *)_RET_IP_);
194
195         memblock_free(phys, size);
196         memblock_remove(phys, size);
197
198         return phys;
199 }
200
201 static void __init arm_initrd_init(void)
202 {
203 #ifdef CONFIG_BLK_DEV_INITRD
204         phys_addr_t start;
205         unsigned long size;
206
207         initrd_start = initrd_end = 0;
208
209         if (!phys_initrd_size)
210                 return;
211
212         /*
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.
217          */
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);
221
222         if (!memblock_is_region_memory(start, size)) {
223                 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region - disabling initrd\n",
224                        (u64)start, size);
225                 return;
226         }
227
228         if (memblock_is_region_reserved(start, size)) {
229                 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region - disabling initrd\n",
230                        (u64)start, size);
231                 return;
232         }
233
234         memblock_reserve(start, size);
235
236         /* Now convert initrd to virtual addresses */
237         initrd_start = __phys_to_virt(phys_initrd_start);
238         initrd_end = initrd_start + phys_initrd_size;
239 #endif
240 }
241
242 void __init arm_memblock_init(const struct machine_desc *mdesc)
243 {
244         /* Register the kernel text, kernel data and initrd with memblock. */
245         memblock_reserve(__pa(KERNEL_START), KERNEL_END - KERNEL_START);
246
247         arm_initrd_init();
248
249         arm_mm_memblock_reserve();
250
251         /* reserve any platform specific memblock areas */
252         if (mdesc->reserve)
253                 mdesc->reserve();
254
255         early_init_fdt_reserve_self();
256         early_init_fdt_scan_reserved_mem();
257
258         /* reserve memory for DMA contiguous allocations */
259         dma_contiguous_reserve(arm_dma_limit);
260
261         arm_memblock_steal_permitted = false;
262         memblock_dump_all();
263 }
264
265 void __init bootmem_init(void)
266 {
267         memblock_allow_resize();
268
269         find_limits(&min_low_pfn, &max_low_pfn, &max_pfn);
270
271         early_memtest((phys_addr_t)min_low_pfn << PAGE_SHIFT,
272                       (phys_addr_t)max_low_pfn << PAGE_SHIFT);
273
274         /*
275          * Sparsemem tries to allocate bootmem in memory_present(),
276          * so must be done after the fixed reservations
277          */
278         memblocks_present();
279
280         /*
281          * sparse_init() needs the bootmem allocator up and running.
282          */
283         sparse_init();
284
285         /*
286          * Now free the memory - free_area_init_node needs
287          * the sparse mem_map arrays initialized by sparse_init()
288          * for memmap_init_zone(), otherwise all PFNs are invalid.
289          */
290         zone_sizes_init(min_low_pfn, max_low_pfn, max_pfn);
291 }
292
293 /*
294  * Poison init memory with an undefined instruction (ARM) or a branch to an
295  * undefined instruction (Thumb).
296  */
297 static inline void poison_init_mem(void *s, size_t count)
298 {
299         u32 *p = (u32 *)s;
300         for (; count != 0; count -= 4)
301                 *p++ = 0xe7fddef0;
302 }
303
304 static inline void
305 free_memmap(unsigned long start_pfn, unsigned long end_pfn)
306 {
307         struct page *start_pg, *end_pg;
308         phys_addr_t pg, pgend;
309
310         /*
311          * Convert start_pfn/end_pfn to a struct page pointer.
312          */
313         start_pg = pfn_to_page(start_pfn - 1) + 1;
314         end_pg = pfn_to_page(end_pfn - 1) + 1;
315
316         /*
317          * Convert to physical addresses, and
318          * round start upwards and end downwards.
319          */
320         pg = PAGE_ALIGN(__pa(start_pg));
321         pgend = __pa(end_pg) & PAGE_MASK;
322
323         /*
324          * If there are free pages between these,
325          * free the section of the memmap array.
326          */
327         if (pg < pgend)
328                 memblock_free_early(pg, pgend - pg);
329 }
330
331 /*
332  * The mem_map array can get very big.  Free the unused area of the memory map.
333  */
334 static void __init free_unused_memmap(void)
335 {
336         unsigned long start, prev_end = 0;
337         struct memblock_region *reg;
338
339         /*
340          * This relies on each bank being in address order.
341          * The banks are sorted previously in bootmem_init().
342          */
343         for_each_memblock(memory, reg) {
344                 start = memblock_region_memory_base_pfn(reg);
345
346 #ifdef CONFIG_SPARSEMEM
347                 /*
348                  * Take care not to free memmap entries that don't exist
349                  * due to SPARSEMEM sections which aren't present.
350                  */
351                 start = min(start,
352                                  ALIGN(prev_end, PAGES_PER_SECTION));
353 #else
354                 /*
355                  * Align down here since the VM subsystem insists that the
356                  * memmap entries are valid from the bank start aligned to
357                  * MAX_ORDER_NR_PAGES.
358                  */
359                 start = round_down(start, MAX_ORDER_NR_PAGES);
360 #endif
361                 /*
362                  * If we had a previous bank, and there is a space
363                  * between the current bank and the previous, free it.
364                  */
365                 if (prev_end && prev_end < start)
366                         free_memmap(prev_end, start);
367
368                 /*
369                  * Align up here since the VM subsystem insists that the
370                  * memmap entries are valid from the bank end aligned to
371                  * MAX_ORDER_NR_PAGES.
372                  */
373                 prev_end = ALIGN(memblock_region_memory_end_pfn(reg),
374                                  MAX_ORDER_NR_PAGES);
375         }
376
377 #ifdef CONFIG_SPARSEMEM
378         if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
379                 free_memmap(prev_end,
380                             ALIGN(prev_end, PAGES_PER_SECTION));
381 #endif
382 }
383
384 #ifdef CONFIG_HIGHMEM
385 static inline void free_area_high(unsigned long pfn, unsigned long end)
386 {
387         for (; pfn < end; pfn++)
388                 free_highmem_page(pfn_to_page(pfn));
389 }
390 #endif
391
392 static void __init free_highpages(void)
393 {
394 #ifdef CONFIG_HIGHMEM
395         unsigned long max_low = max_low_pfn;
396         struct memblock_region *mem, *res;
397
398         /* set highmem page free */
399         for_each_memblock(memory, mem) {
400                 unsigned long start = memblock_region_memory_base_pfn(mem);
401                 unsigned long end = memblock_region_memory_end_pfn(mem);
402
403                 /* Ignore complete lowmem entries */
404                 if (end <= max_low)
405                         continue;
406
407                 if (memblock_is_nomap(mem))
408                         continue;
409
410                 /* Truncate partial highmem entries */
411                 if (start < max_low)
412                         start = max_low;
413
414                 /* Find and exclude any reserved regions */
415                 for_each_memblock(reserved, res) {
416                         unsigned long res_start, res_end;
417
418                         res_start = memblock_region_reserved_base_pfn(res);
419                         res_end = memblock_region_reserved_end_pfn(res);
420
421                         if (res_end < start)
422                                 continue;
423                         if (res_start < start)
424                                 res_start = start;
425                         if (res_start > end)
426                                 res_start = end;
427                         if (res_end > end)
428                                 res_end = end;
429                         if (res_start != start)
430                                 free_area_high(start, res_start);
431                         start = res_end;
432                         if (start == end)
433                                 break;
434                 }
435
436                 /* And now free anything which remains */
437                 if (start < end)
438                         free_area_high(start, end);
439         }
440 #endif
441 }
442
443 /*
444  * mem_init() marks the free areas in the mem_map and tells us how much
445  * memory is free.  This is done after various parts of the system have
446  * claimed their memory after the kernel image.
447  */
448 void __init mem_init(void)
449 {
450 #ifdef CONFIG_HAVE_TCM
451         /* These pointers are filled in on TCM detection */
452         extern u32 dtcm_end;
453         extern u32 itcm_end;
454 #endif
455
456         set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
457
458         /* this will put all unused low memory onto the freelists */
459         free_unused_memmap();
460         memblock_free_all();
461
462 #ifdef CONFIG_SA1111
463         /* now that our DMA memory is actually so designated, we can free it */
464         free_reserved_area(__va(PHYS_OFFSET), swapper_pg_dir, -1, NULL);
465 #endif
466
467         free_highpages();
468
469         mem_init_print_info(NULL);
470
471         /*
472          * Check boundaries twice: Some fundamental inconsistencies can
473          * be detected at build time already.
474          */
475 #ifdef CONFIG_MMU
476         BUILD_BUG_ON(TASK_SIZE                          > MODULES_VADDR);
477         BUG_ON(TASK_SIZE                                > MODULES_VADDR);
478 #endif
479
480 #ifdef CONFIG_HIGHMEM
481         BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
482         BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE      > PAGE_OFFSET);
483 #endif
484 }
485
486 #ifdef CONFIG_STRICT_KERNEL_RWX
487 struct section_perm {
488         const char *name;
489         unsigned long start;
490         unsigned long end;
491         pmdval_t mask;
492         pmdval_t prot;
493         pmdval_t clear;
494 };
495
496 /* First section-aligned location at or after __start_rodata. */
497 extern char __start_rodata_section_aligned[];
498
499 static struct section_perm nx_perms[] = {
500         /* Make pages tables, etc before _stext RW (set NX). */
501         {
502                 .name   = "pre-text NX",
503                 .start  = PAGE_OFFSET,
504                 .end    = (unsigned long)_stext,
505                 .mask   = ~PMD_SECT_XN,
506                 .prot   = PMD_SECT_XN,
507         },
508         /* Make init RW (set NX). */
509         {
510                 .name   = "init NX",
511                 .start  = (unsigned long)__init_begin,
512                 .end    = (unsigned long)_sdata,
513                 .mask   = ~PMD_SECT_XN,
514                 .prot   = PMD_SECT_XN,
515         },
516         /* Make rodata NX (set RO in ro_perms below). */
517         {
518                 .name   = "rodata NX",
519                 .start  = (unsigned long)__start_rodata_section_aligned,
520                 .end    = (unsigned long)__init_begin,
521                 .mask   = ~PMD_SECT_XN,
522                 .prot   = PMD_SECT_XN,
523         },
524 };
525
526 static struct section_perm ro_perms[] = {
527         /* Make kernel code and rodata RX (set RO). */
528         {
529                 .name   = "text/rodata RO",
530                 .start  = (unsigned long)_stext,
531                 .end    = (unsigned long)__init_begin,
532 #ifdef CONFIG_ARM_LPAE
533                 .mask   = ~(L_PMD_SECT_RDONLY | PMD_SECT_AP2),
534                 .prot   = L_PMD_SECT_RDONLY | PMD_SECT_AP2,
535 #else
536                 .mask   = ~(PMD_SECT_APX | PMD_SECT_AP_WRITE),
537                 .prot   = PMD_SECT_APX | PMD_SECT_AP_WRITE,
538                 .clear  = PMD_SECT_AP_WRITE,
539 #endif
540         },
541 };
542
543 /*
544  * Updates section permissions only for the current mm (sections are
545  * copied into each mm). During startup, this is the init_mm. Is only
546  * safe to be called with preemption disabled, as under stop_machine().
547  */
548 static inline void section_update(unsigned long addr, pmdval_t mask,
549                                   pmdval_t prot, struct mm_struct *mm)
550 {
551         pmd_t *pmd;
552
553         pmd = pmd_offset(pud_offset(pgd_offset(mm, addr), addr), addr);
554
555 #ifdef CONFIG_ARM_LPAE
556         pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
557 #else
558         if (addr & SECTION_SIZE)
559                 pmd[1] = __pmd((pmd_val(pmd[1]) & mask) | prot);
560         else
561                 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
562 #endif
563         flush_pmd_entry(pmd);
564         local_flush_tlb_kernel_range(addr, addr + SECTION_SIZE);
565 }
566
567 /* Make sure extended page tables are in use. */
568 static inline bool arch_has_strict_perms(void)
569 {
570         if (cpu_architecture() < CPU_ARCH_ARMv6)
571                 return false;
572
573         return !!(get_cr() & CR_XP);
574 }
575
576 void set_section_perms(struct section_perm *perms, int n, bool set,
577                         struct mm_struct *mm)
578 {
579         size_t i;
580         unsigned long addr;
581
582         if (!arch_has_strict_perms())
583                 return;
584
585         for (i = 0; i < n; i++) {
586                 if (!IS_ALIGNED(perms[i].start, SECTION_SIZE) ||
587                     !IS_ALIGNED(perms[i].end, SECTION_SIZE)) {
588                         pr_err("BUG: %s section %lx-%lx not aligned to %lx\n",
589                                 perms[i].name, perms[i].start, perms[i].end,
590                                 SECTION_SIZE);
591                         continue;
592                 }
593
594                 for (addr = perms[i].start;
595                      addr < perms[i].end;
596                      addr += SECTION_SIZE)
597                         section_update(addr, perms[i].mask,
598                                 set ? perms[i].prot : perms[i].clear, mm);
599         }
600
601 }
602
603 /**
604  * update_sections_early intended to be called only through stop_machine
605  * framework and executed by only one CPU while all other CPUs will spin and
606  * wait, so no locking is required in this function.
607  */
608 static void update_sections_early(struct section_perm perms[], int n)
609 {
610         struct task_struct *t, *s;
611
612         for_each_process(t) {
613                 if (t->flags & PF_KTHREAD)
614                         continue;
615                 for_each_thread(t, s)
616                         set_section_perms(perms, n, true, s->mm);
617         }
618         set_section_perms(perms, n, true, current->active_mm);
619         set_section_perms(perms, n, true, &init_mm);
620 }
621
622 static int __fix_kernmem_perms(void *unused)
623 {
624         update_sections_early(nx_perms, ARRAY_SIZE(nx_perms));
625         return 0;
626 }
627
628 static void fix_kernmem_perms(void)
629 {
630         stop_machine(__fix_kernmem_perms, NULL, NULL);
631 }
632
633 static int __mark_rodata_ro(void *unused)
634 {
635         update_sections_early(ro_perms, ARRAY_SIZE(ro_perms));
636         return 0;
637 }
638
639 static int kernel_set_to_readonly __read_mostly;
640
641 void mark_rodata_ro(void)
642 {
643         kernel_set_to_readonly = 1;
644         stop_machine(__mark_rodata_ro, NULL, NULL);
645         debug_checkwx();
646 }
647
648 void set_kernel_text_rw(void)
649 {
650         if (!kernel_set_to_readonly)
651                 return;
652
653         set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), false,
654                                 current->active_mm);
655 }
656
657 void set_kernel_text_ro(void)
658 {
659         if (!kernel_set_to_readonly)
660                 return;
661
662         set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), true,
663                                 current->active_mm);
664 }
665
666 #else
667 static inline void fix_kernmem_perms(void) { }
668 #endif /* CONFIG_STRICT_KERNEL_RWX */
669
670 void free_initmem(void)
671 {
672         fix_kernmem_perms();
673
674         poison_init_mem(__init_begin, __init_end - __init_begin);
675         if (!machine_is_integrator() && !machine_is_cintegrator())
676                 free_initmem_default(-1);
677 }
678
679 #ifdef CONFIG_BLK_DEV_INITRD
680 void free_initrd_mem(unsigned long start, unsigned long end)
681 {
682         if (start == initrd_start)
683                 start = round_down(start, PAGE_SIZE);
684         if (end == initrd_end)
685                 end = round_up(end, PAGE_SIZE);
686
687         poison_init_mem((void *)start, PAGE_ALIGN(end) - start);
688         free_reserved_area((void *)start, (void *)end, -1, "initrd");
689 }
690 #endif