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/memory.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
__clear_cr(unsigned long mask)45 unsigned 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
parse_tag_initrd(const struct tag * tag)53 static 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
parse_tag_initrd2(const struct tag * tag)64 static 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
find_limits(unsigned long * min,unsigned long * max_low,unsigned long * max_high)74 static 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
84 phys_addr_t arm_dma_zone_size __read_mostly;
85 EXPORT_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 */
93 phys_addr_t arm_dma_limit;
94 unsigned long arm_dma_pfn_limit;
95 #endif
96
setup_dma_zone(const struct machine_desc * mdesc)97 void __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
zone_sizes_init(unsigned long min,unsigned long max_low,unsigned long max_high)109 static 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
pfn_valid(unsigned long pfn)125 int pfn_valid(unsigned long pfn)
126 {
127 phys_addr_t addr = __pfn_to_phys(pfn);
128
129 if (__phys_to_pfn(addr) != pfn)
130 return 0;
131
132 return memblock_is_map_memory(addr);
133 }
134 EXPORT_SYMBOL(pfn_valid);
135 #endif
136
137 static bool arm_memblock_steal_permitted = true;
138
arm_memblock_steal(phys_addr_t size,phys_addr_t align)139 phys_addr_t __init arm_memblock_steal(phys_addr_t size, phys_addr_t align)
140 {
141 phys_addr_t phys;
142
143 BUG_ON(!arm_memblock_steal_permitted);
144
145 phys = memblock_phys_alloc(size, align);
146 if (!phys)
147 panic("Failed to steal %pa bytes at %pS\n",
148 &size, (void *)_RET_IP_);
149
150 memblock_free(phys, size);
151 memblock_remove(phys, size);
152
153 return phys;
154 }
155
arm_initrd_init(void)156 static void __init arm_initrd_init(void)
157 {
158 #ifdef CONFIG_BLK_DEV_INITRD
159 phys_addr_t start;
160 unsigned long size;
161
162 initrd_start = initrd_end = 0;
163
164 if (!phys_initrd_size)
165 return;
166
167 /*
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.
172 */
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);
176
177 if (!memblock_is_region_memory(start, size)) {
178 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region - disabling initrd\n",
179 (u64)start, size);
180 return;
181 }
182
183 if (memblock_is_region_reserved(start, size)) {
184 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region - disabling initrd\n",
185 (u64)start, size);
186 return;
187 }
188
189 memblock_reserve(start, size);
190
191 /* Now convert initrd to virtual addresses */
192 initrd_start = __phys_to_virt(phys_initrd_start);
193 initrd_end = initrd_start + phys_initrd_size;
194 #endif
195 }
196
197 #ifdef CONFIG_CPU_ICACHE_MISMATCH_WORKAROUND
check_cpu_icache_size(int cpuid)198 void check_cpu_icache_size(int cpuid)
199 {
200 u32 size, ctr;
201
202 asm("mrc p15, 0, %0, c0, c0, 1" : "=r" (ctr));
203
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",
207 cpuid);
208 if (icache_size > size)
209 icache_size = size;
210 }
211 #endif
212
arm_memblock_init(const struct machine_desc * mdesc)213 void __init arm_memblock_init(const struct machine_desc *mdesc)
214 {
215 /* Register the kernel text, kernel data and initrd with memblock. */
216 memblock_reserve(__pa(KERNEL_START), KERNEL_END - KERNEL_START);
217
218 arm_initrd_init();
219
220 arm_mm_memblock_reserve();
221
222 /* reserve any platform specific memblock areas */
223 if (mdesc->reserve)
224 mdesc->reserve();
225
226 early_init_fdt_scan_reserved_mem();
227
228 /* reserve memory for DMA contiguous allocations */
229 dma_contiguous_reserve(arm_dma_limit);
230
231 arm_memblock_steal_permitted = false;
232 memblock_dump_all();
233 }
234
bootmem_init(void)235 void __init bootmem_init(void)
236 {
237 memblock_allow_resize();
238
239 find_limits(&min_low_pfn, &max_low_pfn, &max_pfn);
240
241 early_memtest((phys_addr_t)min_low_pfn << PAGE_SHIFT,
242 (phys_addr_t)max_low_pfn << PAGE_SHIFT);
243
244 /*
245 * sparse_init() tries to allocate memory from memblock, so must be
246 * done after the fixed reservations
247 */
248 sparse_init();
249
250 /*
251 * Now free the memory - free_area_init needs
252 * the sparse mem_map arrays initialized by sparse_init()
253 * for memmap_init_zone(), otherwise all PFNs are invalid.
254 */
255 zone_sizes_init(min_low_pfn, max_low_pfn, max_pfn);
256 }
257
258 /*
259 * Poison init memory with an undefined instruction (ARM) or a branch to an
260 * undefined instruction (Thumb).
261 */
poison_init_mem(void * s,size_t count)262 static inline void poison_init_mem(void *s, size_t count)
263 {
264 u32 *p = (u32 *)s;
265 for (; count != 0; count -= 4)
266 *p++ = 0xe7fddef0;
267 }
268
free_highpages(void)269 static void __init free_highpages(void)
270 {
271 #ifdef CONFIG_HIGHMEM
272 unsigned long max_low = max_low_pfn;
273 phys_addr_t range_start, range_end;
274 u64 i;
275
276 /* set highmem page free */
277 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
278 &range_start, &range_end, NULL) {
279 unsigned long start = PFN_UP(range_start);
280 unsigned long end = PFN_DOWN(range_end);
281
282 /* Ignore complete lowmem entries */
283 if (end <= max_low)
284 continue;
285
286 /* Truncate partial highmem entries */
287 if (start < max_low)
288 start = max_low;
289
290 for (; start < end; start++)
291 free_highmem_page(pfn_to_page(start));
292 }
293 #endif
294 }
295
296 /*
297 * mem_init() marks the free areas in the mem_map and tells us how much
298 * memory is free. This is done after various parts of the system have
299 * claimed their memory after the kernel image.
300 */
mem_init(void)301 void __init mem_init(void)
302 {
303 #ifdef CONFIG_ARM_LPAE
304 if (swiotlb_force == SWIOTLB_FORCE ||
305 max_pfn > arm_dma_pfn_limit)
306 swiotlb_init(1);
307 else
308 swiotlb_force = SWIOTLB_NO_FORCE;
309 #endif
310
311 set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
312
313 /* this will put all unused low memory onto the freelists */
314 memblock_free_all();
315
316 #ifdef CONFIG_SA1111
317 /* now that our DMA memory is actually so designated, we can free it */
318 free_reserved_area(__va(PHYS_OFFSET), swapper_pg_dir, -1, NULL);
319 #endif
320
321 free_highpages();
322
323 /*
324 * Check boundaries twice: Some fundamental inconsistencies can
325 * be detected at build time already.
326 */
327 #ifdef CONFIG_MMU
328 BUILD_BUG_ON(TASK_SIZE > MODULES_VADDR);
329 BUG_ON(TASK_SIZE > MODULES_VADDR);
330 #endif
331
332 #ifdef CONFIG_HIGHMEM
333 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
334 BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
335 #endif
336 }
337
338 #ifdef CONFIG_STRICT_KERNEL_RWX
339 struct section_perm {
340 const char *name;
341 unsigned long start;
342 unsigned long end;
343 pmdval_t mask;
344 pmdval_t prot;
345 pmdval_t clear;
346 };
347
348 /* First section-aligned location at or after __start_rodata. */
349 extern char __start_rodata_section_aligned[];
350
351 static struct section_perm nx_perms[] = {
352 /* Make pages tables, etc before _stext RW (set NX). */
353 {
354 .name = "pre-text NX",
355 .start = PAGE_OFFSET,
356 .end = (unsigned long)_stext,
357 .mask = ~PMD_SECT_XN,
358 .prot = PMD_SECT_XN,
359 },
360 /* Make init RW (set NX). */
361 {
362 .name = "init NX",
363 .start = (unsigned long)__init_begin,
364 .end = (unsigned long)_sdata,
365 .mask = ~PMD_SECT_XN,
366 .prot = PMD_SECT_XN,
367 },
368 /* Make rodata NX (set RO in ro_perms below). */
369 {
370 .name = "rodata NX",
371 .start = (unsigned long)__start_rodata_section_aligned,
372 .end = (unsigned long)__init_begin,
373 .mask = ~PMD_SECT_XN,
374 .prot = PMD_SECT_XN,
375 },
376 };
377
378 static struct section_perm ro_perms[] = {
379 /* Make kernel code and rodata RX (set RO). */
380 {
381 .name = "text/rodata RO",
382 .start = (unsigned long)_stext,
383 .end = (unsigned long)__init_begin,
384 #ifdef CONFIG_ARM_LPAE
385 .mask = ~(L_PMD_SECT_RDONLY | PMD_SECT_AP2),
386 .prot = L_PMD_SECT_RDONLY | PMD_SECT_AP2,
387 #else
388 .mask = ~(PMD_SECT_APX | PMD_SECT_AP_WRITE),
389 .prot = PMD_SECT_APX | PMD_SECT_AP_WRITE,
390 .clear = PMD_SECT_AP_WRITE,
391 #endif
392 },
393 };
394
395 /*
396 * Updates section permissions only for the current mm (sections are
397 * copied into each mm). During startup, this is the init_mm. Is only
398 * safe to be called with preemption disabled, as under stop_machine().
399 */
section_update(unsigned long addr,pmdval_t mask,pmdval_t prot,struct mm_struct * mm)400 static inline void section_update(unsigned long addr, pmdval_t mask,
401 pmdval_t prot, struct mm_struct *mm)
402 {
403 pmd_t *pmd;
404
405 pmd = pmd_offset(pud_offset(p4d_offset(pgd_offset(mm, addr), addr), addr), addr);
406
407 #ifdef CONFIG_ARM_LPAE
408 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
409 #else
410 if (addr & SECTION_SIZE)
411 pmd[1] = __pmd((pmd_val(pmd[1]) & mask) | prot);
412 else
413 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
414 #endif
415 flush_pmd_entry(pmd);
416 local_flush_tlb_kernel_range(addr, addr + SECTION_SIZE);
417 }
418
419 /* Make sure extended page tables are in use. */
arch_has_strict_perms(void)420 static inline bool arch_has_strict_perms(void)
421 {
422 if (cpu_architecture() < CPU_ARCH_ARMv6)
423 return false;
424
425 return !!(get_cr() & CR_XP);
426 }
427
set_section_perms(struct section_perm * perms,int n,bool set,struct mm_struct * mm)428 static void set_section_perms(struct section_perm *perms, int n, bool set,
429 struct mm_struct *mm)
430 {
431 size_t i;
432 unsigned long addr;
433
434 if (!arch_has_strict_perms())
435 return;
436
437 for (i = 0; i < n; i++) {
438 if (!IS_ALIGNED(perms[i].start, SECTION_SIZE) ||
439 !IS_ALIGNED(perms[i].end, SECTION_SIZE)) {
440 pr_err("BUG: %s section %lx-%lx not aligned to %lx\n",
441 perms[i].name, perms[i].start, perms[i].end,
442 SECTION_SIZE);
443 continue;
444 }
445
446 for (addr = perms[i].start;
447 addr < perms[i].end;
448 addr += SECTION_SIZE)
449 section_update(addr, perms[i].mask,
450 set ? perms[i].prot : perms[i].clear, mm);
451 }
452
453 }
454
455 /**
456 * update_sections_early intended to be called only through stop_machine
457 * framework and executed by only one CPU while all other CPUs will spin and
458 * wait, so no locking is required in this function.
459 */
update_sections_early(struct section_perm perms[],int n)460 static void update_sections_early(struct section_perm perms[], int n)
461 {
462 struct task_struct *t, *s;
463
464 for_each_process(t) {
465 if (t->flags & PF_KTHREAD)
466 continue;
467 for_each_thread(t, s)
468 if (s->mm)
469 set_section_perms(perms, n, true, s->mm);
470 }
471 set_section_perms(perms, n, true, current->active_mm);
472 set_section_perms(perms, n, true, &init_mm);
473 }
474
__fix_kernmem_perms(void * unused)475 static int __fix_kernmem_perms(void *unused)
476 {
477 update_sections_early(nx_perms, ARRAY_SIZE(nx_perms));
478 return 0;
479 }
480
fix_kernmem_perms(void)481 static void fix_kernmem_perms(void)
482 {
483 stop_machine(__fix_kernmem_perms, NULL, NULL);
484 }
485
__mark_rodata_ro(void * unused)486 static int __mark_rodata_ro(void *unused)
487 {
488 update_sections_early(ro_perms, ARRAY_SIZE(ro_perms));
489 return 0;
490 }
491
mark_rodata_ro(void)492 void mark_rodata_ro(void)
493 {
494 stop_machine(__mark_rodata_ro, NULL, NULL);
495 debug_checkwx();
496 }
497
498 #else
fix_kernmem_perms(void)499 static inline void fix_kernmem_perms(void) { }
500 #endif /* CONFIG_STRICT_KERNEL_RWX */
501
free_initmem(void)502 void free_initmem(void)
503 {
504 fix_kernmem_perms();
505
506 poison_init_mem(__init_begin, __init_end - __init_begin);
507 if (!machine_is_integrator() && !machine_is_cintegrator())
508 free_initmem_default(-1);
509 }
510
511 #ifdef CONFIG_BLK_DEV_INITRD
free_initrd_mem(unsigned long start,unsigned long end)512 void free_initrd_mem(unsigned long start, unsigned long end)
513 {
514 if (start == initrd_start)
515 start = round_down(start, PAGE_SIZE);
516 if (end == initrd_end)
517 end = round_up(end, PAGE_SIZE);
518
519 poison_init_mem((void *)start, PAGE_ALIGN(end) - start);
520 free_reserved_area((void *)start, (void *)end, -1, "initrd");
521 }
522 #endif
523