Merge tag 'csky-for-linus-4.20-fixup-dtb' of https://github.com/c-sky/csky-linux
[linux-block.git] / arch / arm / mm / init.c
CommitLineData
1da177e4
LT
1/*
2 * linux/arch/arm/mm/init.c
3 *
90072059 4 * Copyright (C) 1995-2005 Russell King
1da177e4
LT
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 */
1da177e4
LT
10#include <linux/kernel.h>
11#include <linux/errno.h>
1da177e4
LT
12#include <linux/swap.h>
13#include <linux/init.h>
1da177e4 14#include <linux/mman.h>
3f07c014 15#include <linux/sched/signal.h>
29930025 16#include <linux/sched/task.h>
dc28094b 17#include <linux/export.h>
1da177e4
LT
18#include <linux/nodemask.h>
19#include <linux/initrd.h>
9eb8f674 20#include <linux/of_fdt.h>
3835f6cb 21#include <linux/highmem.h>
5a0e3ad6 22#include <linux/gfp.h>
2778f620 23#include <linux/memblock.h>
c7909509 24#include <linux/dma-contiguous.h>
158e8bfe 25#include <linux/sizes.h>
08925c2f 26#include <linux/stop_machine.h>
1da177e4 27
b4b20ad8 28#include <asm/cp15.h>
1da177e4 29#include <asm/mach-types.h>
716a3dc2 30#include <asm/memblock.h>
d2ca5f24 31#include <asm/memory.h>
93c02ab4 32#include <asm/prom.h>
37efe642 33#include <asm/sections.h>
1da177e4 34#include <asm/setup.h>
1e6b4811 35#include <asm/system_info.h>
1da177e4 36#include <asm/tlb.h>
db9ef1af 37#include <asm/fixmap.h>
a8e53c15 38#include <asm/ptdump.h>
1da177e4
LT
39
40#include <asm/mach/arch.h>
41#include <asm/mach/map.h>
42
1b2e2b73
RK
43#include "mm.h"
44
b4b20ad8
RK
45#ifdef CONFIG_CPU_CP15_MMU
46unsigned long __init __clear_cr(unsigned long mask)
47{
b4b20ad8
RK
48 cr_alignment = cr_alignment & ~mask;
49 return cr_alignment;
50}
51#endif
52
de22cc6e 53static phys_addr_t phys_initrd_start __initdata = 0;
012d1f4a
RK
54static unsigned long phys_initrd_size __initdata = 0;
55
2b0d8c25 56static int __init early_initrd(char *p)
012d1f4a 57{
de22cc6e
VA
58 phys_addr_t start;
59 unsigned long size;
2b0d8c25 60 char *endp;
012d1f4a 61
2b0d8c25
JK
62 start = memparse(p, &endp);
63 if (*endp == ',') {
64 size = memparse(endp + 1, NULL);
012d1f4a
RK
65
66 phys_initrd_start = start;
67 phys_initrd_size = size;
68 }
2b0d8c25 69 return 0;
012d1f4a 70}
2b0d8c25 71early_param("initrd", early_initrd);
012d1f4a
RK
72
73static int __init parse_tag_initrd(const struct tag *tag)
74{
4ed89f22 75 pr_warn("ATAG_INITRD is deprecated; "
012d1f4a
RK
76 "please update your bootloader.\n");
77 phys_initrd_start = __virt_to_phys(tag->u.initrd.start);
78 phys_initrd_size = tag->u.initrd.size;
79 return 0;
80}
81
82__tagtable(ATAG_INITRD, parse_tag_initrd);
83
84static int __init parse_tag_initrd2(const struct tag *tag)
85{
86 phys_initrd_start = tag->u.initrd.start;
87 phys_initrd_size = tag->u.initrd.size;
88 return 0;
89}
90
91__tagtable(ATAG_INITRD2, parse_tag_initrd2);
1da177e4 92
f25b4b4c 93static void __init find_limits(unsigned long *min, unsigned long *max_low,
27a3f0e9 94 unsigned long *max_high)
dde5828f 95{
1c2f87c2
LA
96 *max_low = PFN_DOWN(memblock_get_current_limit());
97 *min = PFN_UP(memblock_start_of_DRAM());
98 *max_high = PFN_DOWN(memblock_end_of_DRAM());
dde5828f
RK
99}
100
be20902b 101#ifdef CONFIG_ZONE_DMA
65032018 102
364230b9 103phys_addr_t arm_dma_zone_size __read_mostly;
65032018
NP
104EXPORT_SYMBOL(arm_dma_zone_size);
105
022ae537
RK
106/*
107 * The DMA mask corresponding to the maximum bus address allocatable
108 * using GFP_DMA. The default here places no restriction on DMA
109 * allocations. This must be the smallest DMA mask in the system,
110 * so a successful GFP_DMA allocation will always satisfy this.
111 */
4986e5c7 112phys_addr_t arm_dma_limit;
4dcfa600 113unsigned long arm_dma_pfn_limit;
022ae537 114
be20902b
RK
115static void __init arm_adjust_dma_zone(unsigned long *size, unsigned long *hole,
116 unsigned long dma_size)
117{
118 if (size[0] <= dma_size)
119 return;
120
121 size[ZONE_NORMAL] = size[0] - dma_size;
122 size[ZONE_DMA] = dma_size;
123 hole[ZONE_NORMAL] = hole[0];
124 hole[ZONE_DMA] = 0;
125}
126#endif
127
ff69a4c8 128void __init setup_dma_zone(const struct machine_desc *mdesc)
c7909509
MS
129{
130#ifdef CONFIG_ZONE_DMA
131 if (mdesc->dma_zone_size) {
132 arm_dma_zone_size = mdesc->dma_zone_size;
6bcac805 133 arm_dma_limit = PHYS_OFFSET + arm_dma_zone_size - 1;
c7909509
MS
134 } else
135 arm_dma_limit = 0xffffffff;
4dcfa600 136 arm_dma_pfn_limit = arm_dma_limit >> PAGE_SHIFT;
c7909509
MS
137#endif
138}
139
84f452b1 140static void __init zone_sizes_init(unsigned long min, unsigned long max_low,
a2c54d2a 141 unsigned long max_high)
b7a69ac3
RK
142{
143 unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES];
a2c54d2a 144 struct memblock_region *reg;
b7a69ac3 145
90072059 146 /*
be370302 147 * initialise the zones.
90072059
RK
148 */
149 memset(zone_size, 0, sizeof(zone_size));
90072059
RK
150
151 /*
be370302
RK
152 * The memory size has already been determined. If we need
153 * to do anything fancy with the allocation of this memory
154 * to the zones, now is the time to do it.
90072059 155 */
dde5828f
RK
156 zone_size[0] = max_low - min;
157#ifdef CONFIG_HIGHMEM
158 zone_size[ZONE_HIGHMEM] = max_high - max_low;
159#endif
90072059
RK
160
161 /*
be370302
RK
162 * Calculate the size of the holes.
163 * holes = node_size - sum(bank_sizes)
90072059 164 */
dde5828f 165 memcpy(zhole_size, zone_size, sizeof(zhole_size));
a2c54d2a
RK
166 for_each_memblock(memory, reg) {
167 unsigned long start = memblock_region_memory_base_pfn(reg);
168 unsigned long end = memblock_region_memory_end_pfn(reg);
169
170 if (start < max_low) {
171 unsigned long low_end = min(end, max_low);
172 zhole_size[0] -= low_end - start;
173 }
dde5828f 174#ifdef CONFIG_HIGHMEM
a2c54d2a
RK
175 if (end > max_low) {
176 unsigned long high_start = max(start, max_low);
177 zhole_size[ZONE_HIGHMEM] -= end - high_start;
178 }
dde5828f 179#endif
dde5828f 180 }
90072059 181
65032018 182#ifdef CONFIG_ZONE_DMA
90072059
RK
183 /*
184 * Adjust the sizes according to any special requirements for
185 * this machine type.
186 */
c7909509 187 if (arm_dma_zone_size)
65032018
NP
188 arm_adjust_dma_zone(zone_size, zhole_size,
189 arm_dma_zone_size >> PAGE_SHIFT);
be20902b 190#endif
90072059 191
be370302 192 free_area_init_node(0, zone_size, min, zhole_size);
1da177e4
LT
193}
194
7b7bf499 195#ifdef CONFIG_HAVE_ARCH_PFN_VALID
b7cfda9f
RK
196int pfn_valid(unsigned long pfn)
197{
09414d00 198 return memblock_is_map_memory(__pfn_to_phys(pfn));
b7cfda9f
RK
199}
200EXPORT_SYMBOL(pfn_valid);
7b7bf499 201#endif
657e12fd 202
7b7bf499 203#ifndef CONFIG_SPARSEMEM
14904927 204static void __init arm_memory_present(void)
657e12fd
RK
205{
206}
207#else
14904927 208static void __init arm_memory_present(void)
657e12fd 209{
719c1514 210 struct memblock_region *reg;
719c1514 211
7c996361 212 for_each_memblock(memory, reg)
c7fc2de0
YL
213 memory_present(0, memblock_region_memory_base_pfn(reg),
214 memblock_region_memory_end_pfn(reg));
657e12fd 215}
b7cfda9f
RK
216#endif
217
716a3dc2
RK
218static bool arm_memblock_steal_permitted = true;
219
bc2827d0 220phys_addr_t __init arm_memblock_steal(phys_addr_t size, phys_addr_t align)
716a3dc2
RK
221{
222 phys_addr_t phys;
223
224 BUG_ON(!arm_memblock_steal_permitted);
225
7ac68a4c 226 phys = memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ANYWHERE);
716a3dc2
RK
227 memblock_free(phys, size);
228 memblock_remove(phys, size);
229
230 return phys;
231}
232
39286248 233static void __init arm_initrd_init(void)
2778f620 234{
2778f620 235#ifdef CONFIG_BLK_DEV_INITRD
cdcc5fa0
RK
236 phys_addr_t start;
237 unsigned long size;
238
65939301 239 /* FDT scan will populate initrd_start */
4c235cb9 240 if (initrd_start && !phys_initrd_size) {
65939301
RH
241 phys_initrd_start = __virt_to_phys(initrd_start);
242 phys_initrd_size = initrd_end - initrd_start;
243 }
68b32f36 244
4c235cb9 245 initrd_start = initrd_end = 0;
68b32f36
RK
246
247 if (!phys_initrd_size)
248 return;
249
cdcc5fa0
RK
250 /*
251 * Round the memory region to page boundaries as per free_initrd_mem()
252 * This allows us to detect whether the pages overlapping the initrd
253 * are in use, but more importantly, reserves the entire set of pages
254 * as we don't want these pages allocated for other purposes.
255 */
256 start = round_down(phys_initrd_start, PAGE_SIZE);
257 size = phys_initrd_size + (phys_initrd_start - start);
258 size = round_up(size, PAGE_SIZE);
259
260 if (!memblock_is_region_memory(start, size)) {
de22cc6e 261 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region - disabling initrd\n",
cdcc5fa0 262 (u64)start, size);
68b32f36 263 return;
8f4b8c76 264 }
68b32f36 265
cdcc5fa0 266 if (memblock_is_region_reserved(start, size)) {
de22cc6e 267 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region - disabling initrd\n",
cdcc5fa0 268 (u64)start, size);
68b32f36 269 return;
b0a2679d 270 }
2778f620 271
cdcc5fa0 272 memblock_reserve(start, size);
68b32f36
RK
273
274 /* Now convert initrd to virtual addresses */
275 initrd_start = __phys_to_virt(phys_initrd_start);
276 initrd_end = initrd_start + phys_initrd_size;
2778f620 277#endif
39286248
RK
278}
279
280void __init arm_memblock_init(const struct machine_desc *mdesc)
281{
282 /* Register the kernel text, kernel data and initrd with memblock. */
283 memblock_reserve(__pa(KERNEL_START), KERNEL_END - KERNEL_START);
284
285 arm_initrd_init();
2778f620
RK
286
287 arm_mm_memblock_reserve();
288
8d717a52
RK
289 /* reserve any platform specific memblock areas */
290 if (mdesc->reserve)
291 mdesc->reserve();
292
24bbd929 293 early_init_fdt_reserve_self();
bcedb5f9
MS
294 early_init_fdt_scan_reserved_mem();
295
99a468d7 296 /* reserve memory for DMA contiguous allocations */
95b0e655 297 dma_contiguous_reserve(arm_dma_limit);
c7909509 298
716a3dc2 299 arm_memblock_steal_permitted = false;
2778f620
RK
300 memblock_dump_all();
301}
302
8d717a52 303void __init bootmem_init(void)
1da177e4 304{
dde5828f 305 unsigned long min, max_low, max_high;
1da177e4 306
8e58caef 307 memblock_allow_resize();
dde5828f
RK
308 max_low = max_high = 0;
309
f25b4b4c 310 find_limits(&min, &max_low, &max_high);
dde5828f 311
d30eae47
VM
312 early_memtest((phys_addr_t)min << PAGE_SHIFT,
313 (phys_addr_t)max_low << PAGE_SHIFT);
314
be370302
RK
315 /*
316 * Sparsemem tries to allocate bootmem in memory_present(),
317 * so must be done after the fixed reservations
318 */
eda2e5dc 319 arm_memory_present();
1da177e4 320
b7a69ac3
RK
321 /*
322 * sparse_init() needs the bootmem allocator up and running.
323 */
324 sparse_init();
325
326 /*
be370302 327 * Now free the memory - free_area_init_node needs
b7a69ac3
RK
328 * the sparse mem_map arrays initialized by sparse_init()
329 * for memmap_init_zone(), otherwise all PFNs are invalid.
330 */
84f452b1 331 zone_sizes_init(min, max_low, max_high);
b7a69ac3 332
90072059
RK
333 /*
334 * This doesn't seem to be used by the Linux memory manager any
335 * more, but is used by ll_rw_block. If we can get rid of it, we
336 * also get rid of some of the stuff above as well.
90072059 337 */
26ba47b1
SS
338 min_low_pfn = min;
339 max_low_pfn = max_low;
340 max_pfn = max_high;
90072059 341}
1da177e4 342
54d52573
SB
343/*
344 * Poison init memory with an undefined instruction (ARM) or a branch to an
345 * undefined instruction (Thumb).
346 */
347static inline void poison_init_mem(void *s, size_t count)
348{
349 u32 *p = (u32 *)s;
bf912d99 350 for (; count != 0; count -= 4)
54d52573
SB
351 *p++ = 0xe7fddef0;
352}
353
a013053d 354static inline void
be370302 355free_memmap(unsigned long start_pfn, unsigned long end_pfn)
a013053d
RK
356{
357 struct page *start_pg, *end_pg;
56bc6286 358 phys_addr_t pg, pgend;
a013053d
RK
359
360 /*
361 * Convert start_pfn/end_pfn to a struct page pointer.
362 */
3257f43d 363 start_pg = pfn_to_page(start_pfn - 1) + 1;
9af386c8 364 end_pg = pfn_to_page(end_pfn - 1) + 1;
a013053d
RK
365
366 /*
367 * Convert to physical addresses, and
368 * round start upwards and end downwards.
369 */
56bc6286
VA
370 pg = PAGE_ALIGN(__pa(start_pg));
371 pgend = __pa(end_pg) & PAGE_MASK;
a013053d
RK
372
373 /*
374 * If there are free pages between these,
375 * free the section of the memmap array.
376 */
377 if (pg < pgend)
cfb66586 378 memblock_free_early(pg, pgend - pg);
a013053d
RK
379}
380
381/*
382 * The mem_map array can get very big. Free the unused area of the memory map.
383 */
1c2f87c2 384static void __init free_unused_memmap(void)
a013053d 385{
1c2f87c2
LA
386 unsigned long start, prev_end = 0;
387 struct memblock_region *reg;
a013053d
RK
388
389 /*
3260e529
MB
390 * This relies on each bank being in address order.
391 * The banks are sorted previously in bootmem_init().
a013053d 392 */
1c2f87c2
LA
393 for_each_memblock(memory, reg) {
394 start = memblock_region_memory_base_pfn(reg);
a013053d 395
9af386c8
WD
396#ifdef CONFIG_SPARSEMEM
397 /*
398 * Take care not to free memmap entries that don't exist
399 * due to SPARSEMEM sections which aren't present.
400 */
1c2f87c2
LA
401 start = min(start,
402 ALIGN(prev_end, PAGES_PER_SECTION));
002ea9ee
LW
403#else
404 /*
405 * Align down here since the VM subsystem insists that the
406 * memmap entries are valid from the bank start aligned to
407 * MAX_ORDER_NR_PAGES.
408 */
1c2f87c2 409 start = round_down(start, MAX_ORDER_NR_PAGES);
9af386c8 410#endif
a013053d
RK
411 /*
412 * If we had a previous bank, and there is a space
413 * between the current bank and the previous, free it.
414 */
1c2f87c2
LA
415 if (prev_end && prev_end < start)
416 free_memmap(prev_end, start);
a013053d 417
3260e529
MB
418 /*
419 * Align up here since the VM subsystem insists that the
420 * memmap entries are valid from the bank end aligned to
421 * MAX_ORDER_NR_PAGES.
422 */
1c2f87c2
LA
423 prev_end = ALIGN(memblock_region_memory_end_pfn(reg),
424 MAX_ORDER_NR_PAGES);
a013053d 425 }
9af386c8
WD
426
427#ifdef CONFIG_SPARSEMEM
1c2f87c2
LA
428 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
429 free_memmap(prev_end,
430 ALIGN(prev_end, PAGES_PER_SECTION));
9af386c8 431#endif
a013053d
RK
432}
433
83db0384
JL
434#ifdef CONFIG_HIGHMEM
435static inline void free_area_high(unsigned long pfn, unsigned long end)
436{
dd6911ef
JL
437 for (; pfn < end; pfn++)
438 free_highmem_page(pfn_to_page(pfn));
83db0384
JL
439}
440#endif
441
d0e775af
RK
442static void __init free_highpages(void)
443{
444#ifdef CONFIG_HIGHMEM
26ba47b1 445 unsigned long max_low = max_low_pfn;
df4f14c7 446 struct memblock_region *mem, *res;
d0e775af
RK
447
448 /* set highmem page free */
df4f14c7
RK
449 for_each_memblock(memory, mem) {
450 unsigned long start = memblock_region_memory_base_pfn(mem);
451 unsigned long end = memblock_region_memory_end_pfn(mem);
452
453 /* Ignore complete lowmem entries */
454 if (end <= max_low)
455 continue;
456
09414d00
AB
457 if (memblock_is_nomap(mem))
458 continue;
459
df4f14c7
RK
460 /* Truncate partial highmem entries */
461 if (start < max_low)
462 start = max_low;
463
464 /* Find and exclude any reserved regions */
465 for_each_memblock(reserved, res) {
466 unsigned long res_start, res_end;
467
468 res_start = memblock_region_reserved_base_pfn(res);
469 res_end = memblock_region_reserved_end_pfn(res);
470
471 if (res_end < start)
472 continue;
473 if (res_start < start)
474 res_start = start;
475 if (res_start > end)
476 res_start = end;
477 if (res_end > end)
478 res_end = end;
479 if (res_start != start)
83db0384 480 free_area_high(start, res_start);
df4f14c7
RK
481 start = res_end;
482 if (start == end)
483 break;
484 }
485
486 /* And now free anything which remains */
487 if (start < end)
83db0384 488 free_area_high(start, end);
d0e775af 489 }
d0e775af
RK
490#endif
491}
492
1da177e4
LT
493/*
494 * mem_init() marks the free areas in the mem_map and tells us how much
495 * memory is free. This is done after various parts of the system have
496 * claimed their memory after the kernel image.
497 */
498void __init mem_init(void)
499{
1dbd30e9
LW
500#ifdef CONFIG_HAVE_TCM
501 /* These pointers are filled in on TCM detection */
502 extern u32 dtcm_end;
503 extern u32 itcm_end;
504#endif
1da177e4 505
b3ba41f2 506 set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
1da177e4 507
1da177e4 508 /* this will put all unused low memory onto the freelists */
1c2f87c2 509 free_unused_memmap();
c6ffc5ca 510 memblock_free_all();
1da177e4
LT
511
512#ifdef CONFIG_SA1111
513 /* now that our DMA memory is actually so designated, we can free it */
bfd65dd9 514 free_reserved_area(__va(PHYS_OFFSET), swapper_pg_dir, -1, NULL);
1da177e4
LT
515#endif
516
d0e775af 517 free_highpages();
3835f6cb 518
2450c973 519 mem_init_print_info(NULL);
1da177e4 520
db9ef1af
FB
521#define MLK(b, t) b, t, ((t) - (b)) >> 10
522#define MLM(b, t) b, t, ((t) - (b)) >> 20
523#define MLK_ROUNDUP(b, t) b, t, DIV_ROUND_UP(((t) - (b)), SZ_1K)
524
4ed89f22 525 pr_notice("Virtual kernel memory layout:\n"
db9ef1af 526 " vector : 0x%08lx - 0x%08lx (%4ld kB)\n"
07d2a5c7 527#ifdef CONFIG_HAVE_TCM
07d2a5c7 528 " DTCM : 0x%08lx - 0x%08lx (%4ld kB)\n"
07d2a5c7 529 " ITCM : 0x%08lx - 0x%08lx (%4ld kB)\n"
07d2a5c7 530#endif
db9ef1af
FB
531 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n"
532 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n"
533 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n"
534#ifdef CONFIG_HIGHMEM
535 " pkmap : 0x%08lx - 0x%08lx (%4ld MB)\n"
536#endif
d9277d51 537#ifdef CONFIG_MODULES
db9ef1af 538 " modules : 0x%08lx - 0x%08lx (%4ld MB)\n"
d9277d51 539#endif
178c3dfe
RK
540 " .text : 0x%p" " - 0x%p" " (%4td kB)\n"
541 " .init : 0x%p" " - 0x%p" " (%4td kB)\n"
542 " .data : 0x%p" " - 0x%p" " (%4td kB)\n"
543 " .bss : 0x%p" " - 0x%p" " (%4td kB)\n",
db9ef1af 544
d2ca5f24 545 MLK(VECTORS_BASE, VECTORS_BASE + PAGE_SIZE),
07d2a5c7 546#ifdef CONFIG_HAVE_TCM
1dbd30e9
LW
547 MLK(DTCM_OFFSET, (unsigned long) dtcm_end),
548 MLK(ITCM_OFFSET, (unsigned long) itcm_end),
07d2a5c7 549#endif
b615bbbf 550 MLK(FIXADDR_START, FIXADDR_END),
c931b4f6 551 MLM(VMALLOC_START, VMALLOC_END),
db9ef1af
FB
552 MLM(PAGE_OFFSET, (unsigned long)high_memory),
553#ifdef CONFIG_HIGHMEM
554 MLM(PKMAP_BASE, (PKMAP_BASE) + (LAST_PKMAP) *
555 (PAGE_SIZE)),
556#endif
d9277d51 557#ifdef CONFIG_MODULES
db9ef1af 558 MLM(MODULES_VADDR, MODULES_END),
d9277d51 559#endif
db9ef1af 560
db9ef1af 561 MLK_ROUNDUP(_text, _etext),
3835d69a 562 MLK_ROUNDUP(__init_begin, __init_end),
45f6d7e0
RV
563 MLK_ROUNDUP(_sdata, _edata),
564 MLK_ROUNDUP(__bss_start, __bss_stop));
db9ef1af
FB
565
566#undef MLK
567#undef MLM
568#undef MLK_ROUNDUP
569
a1839272
FB
570 /*
571 * Check boundaries twice: Some fundamental inconsistencies can
572 * be detected at build time already.
573 */
574#ifdef CONFIG_MMU
a1839272
FB
575 BUILD_BUG_ON(TASK_SIZE > MODULES_VADDR);
576 BUG_ON(TASK_SIZE > MODULES_VADDR);
577#endif
578
579#ifdef CONFIG_HIGHMEM
580 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
581 BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
582#endif
1da177e4
LT
583}
584
0f5bf6d0 585#ifdef CONFIG_STRICT_KERNEL_RWX
1e6b4811 586struct section_perm {
25362dc4 587 const char *name;
1e6b4811
KC
588 unsigned long start;
589 unsigned long end;
590 pmdval_t mask;
591 pmdval_t prot;
80d6b0c2 592 pmdval_t clear;
1e6b4811
KC
593};
594
64ac2e74
KC
595/* First section-aligned location at or after __start_rodata. */
596extern char __start_rodata_section_aligned[];
597
80d6b0c2 598static struct section_perm nx_perms[] = {
1e6b4811
KC
599 /* Make pages tables, etc before _stext RW (set NX). */
600 {
25362dc4 601 .name = "pre-text NX",
1e6b4811
KC
602 .start = PAGE_OFFSET,
603 .end = (unsigned long)_stext,
604 .mask = ~PMD_SECT_XN,
605 .prot = PMD_SECT_XN,
606 },
607 /* Make init RW (set NX). */
608 {
25362dc4 609 .name = "init NX",
1e6b4811
KC
610 .start = (unsigned long)__init_begin,
611 .end = (unsigned long)_sdata,
612 .mask = ~PMD_SECT_XN,
613 .prot = PMD_SECT_XN,
614 },
80d6b0c2
KC
615 /* Make rodata NX (set RO in ro_perms below). */
616 {
25362dc4 617 .name = "rodata NX",
64ac2e74 618 .start = (unsigned long)__start_rodata_section_aligned,
80d6b0c2
KC
619 .end = (unsigned long)__init_begin,
620 .mask = ~PMD_SECT_XN,
621 .prot = PMD_SECT_XN,
622 },
1e6b4811
KC
623};
624
80d6b0c2
KC
625static struct section_perm ro_perms[] = {
626 /* Make kernel code and rodata RX (set RO). */
627 {
25362dc4 628 .name = "text/rodata RO",
80d6b0c2
KC
629 .start = (unsigned long)_stext,
630 .end = (unsigned long)__init_begin,
631#ifdef CONFIG_ARM_LPAE
400eeffa
PD
632 .mask = ~(L_PMD_SECT_RDONLY | PMD_SECT_AP2),
633 .prot = L_PMD_SECT_RDONLY | PMD_SECT_AP2,
80d6b0c2
KC
634#else
635 .mask = ~(PMD_SECT_APX | PMD_SECT_AP_WRITE),
636 .prot = PMD_SECT_APX | PMD_SECT_AP_WRITE,
637 .clear = PMD_SECT_AP_WRITE,
638#endif
639 },
640};
80d6b0c2 641
1e6b4811
KC
642/*
643 * Updates section permissions only for the current mm (sections are
644 * copied into each mm). During startup, this is the init_mm. Is only
645 * safe to be called with preemption disabled, as under stop_machine().
646 */
647static inline void section_update(unsigned long addr, pmdval_t mask,
08925c2f 648 pmdval_t prot, struct mm_struct *mm)
1e6b4811 649{
1e6b4811
KC
650 pmd_t *pmd;
651
1e6b4811
KC
652 pmd = pmd_offset(pud_offset(pgd_offset(mm, addr), addr), addr);
653
654#ifdef CONFIG_ARM_LPAE
655 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
656#else
657 if (addr & SECTION_SIZE)
658 pmd[1] = __pmd((pmd_val(pmd[1]) & mask) | prot);
659 else
660 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
661#endif
662 flush_pmd_entry(pmd);
663 local_flush_tlb_kernel_range(addr, addr + SECTION_SIZE);
664}
665
666/* Make sure extended page tables are in use. */
667static inline bool arch_has_strict_perms(void)
668{
669 if (cpu_architecture() < CPU_ARCH_ARMv6)
670 return false;
671
672 return !!(get_cr() & CR_XP);
673}
674
08925c2f
LA
675void set_section_perms(struct section_perm *perms, int n, bool set,
676 struct mm_struct *mm)
677{
678 size_t i;
679 unsigned long addr;
680
681 if (!arch_has_strict_perms())
682 return;
683
684 for (i = 0; i < n; i++) {
685 if (!IS_ALIGNED(perms[i].start, SECTION_SIZE) ||
686 !IS_ALIGNED(perms[i].end, SECTION_SIZE)) {
25362dc4
KC
687 pr_err("BUG: %s section %lx-%lx not aligned to %lx\n",
688 perms[i].name, perms[i].start, perms[i].end,
08925c2f
LA
689 SECTION_SIZE);
690 continue;
691 }
692
693 for (addr = perms[i].start;
694 addr < perms[i].end;
695 addr += SECTION_SIZE)
696 section_update(addr, perms[i].mask,
697 set ? perms[i].prot : perms[i].clear, mm);
698 }
699
1e6b4811
KC
700}
701
11ce4b33
GS
702/**
703 * update_sections_early intended to be called only through stop_machine
704 * framework and executed by only one CPU while all other CPUs will spin and
705 * wait, so no locking is required in this function.
706 */
08925c2f 707static void update_sections_early(struct section_perm perms[], int n)
1e6b4811 708{
08925c2f
LA
709 struct task_struct *t, *s;
710
08925c2f
LA
711 for_each_process(t) {
712 if (t->flags & PF_KTHREAD)
713 continue;
714 for_each_thread(t, s)
715 set_section_perms(perms, n, true, s->mm);
716 }
08925c2f
LA
717 set_section_perms(perms, n, true, current->active_mm);
718 set_section_perms(perms, n, true, &init_mm);
719}
720
11ce4b33 721static int __fix_kernmem_perms(void *unused)
08925c2f
LA
722{
723 update_sections_early(nx_perms, ARRAY_SIZE(nx_perms));
724 return 0;
725}
726
11ce4b33 727static void fix_kernmem_perms(void)
08925c2f
LA
728{
729 stop_machine(__fix_kernmem_perms, NULL, NULL);
1e6b4811 730}
80d6b0c2 731
11ce4b33 732static int __mark_rodata_ro(void *unused)
08925c2f
LA
733{
734 update_sections_early(ro_perms, ARRAY_SIZE(ro_perms));
735 return 0;
736}
737
b4c7e2bd
SRV
738static int kernel_set_to_readonly __read_mostly;
739
80d6b0c2
KC
740void mark_rodata_ro(void)
741{
b4c7e2bd 742 kernel_set_to_readonly = 1;
08925c2f 743 stop_machine(__mark_rodata_ro, NULL, NULL);
a8e53c15 744 debug_checkwx();
80d6b0c2
KC
745}
746
747void set_kernel_text_rw(void)
748{
b4c7e2bd
SRV
749 if (!kernel_set_to_readonly)
750 return;
751
08925c2f
LA
752 set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), false,
753 current->active_mm);
80d6b0c2
KC
754}
755
756void set_kernel_text_ro(void)
757{
b4c7e2bd
SRV
758 if (!kernel_set_to_readonly)
759 return;
760
08925c2f
LA
761 set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), true,
762 current->active_mm);
80d6b0c2 763}
80d6b0c2 764
1e6b4811
KC
765#else
766static inline void fix_kernmem_perms(void) { }
0f5bf6d0 767#endif /* CONFIG_STRICT_KERNEL_RWX */
1e6b4811 768
1e6b4811
KC
769void free_initmem(void)
770{
771 fix_kernmem_perms();
bc581770 772
54d52573 773 poison_init_mem(__init_begin, __init_end - __init_begin);
6db015e4 774 if (!machine_is_integrator() && !machine_is_cintegrator())
dbe67df4 775 free_initmem_default(-1);
1da177e4
LT
776}
777
778#ifdef CONFIG_BLK_DEV_INITRD
779
780static int keep_initrd;
781
782void free_initrd_mem(unsigned long start, unsigned long end)
783{
54d52573 784 if (!keep_initrd) {
421520ba
YW
785 if (start == initrd_start)
786 start = round_down(start, PAGE_SIZE);
787 if (end == initrd_end)
788 end = round_up(end, PAGE_SIZE);
789
54d52573 790 poison_init_mem((void *)start, PAGE_ALIGN(end) - start);
dbe67df4 791 free_reserved_area((void *)start, (void *)end, -1, "initrd");
54d52573 792 }
1da177e4
LT
793}
794
795static int __init keepinitrd_setup(char *__unused)
796{
797 keep_initrd = 1;
798 return 1;
799}
800
801__setup("keepinitrd", keepinitrd_setup);
802#endif