• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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
__clear_cr(unsigned long mask)47 unsigned long __init __clear_cr(unsigned long mask)
48 {
49 	cr_alignment = cr_alignment & ~mask;
50 	return cr_alignment;
51 }
52 #endif
53 
54 static phys_addr_t phys_initrd_start __initdata = 0;
55 static unsigned long phys_initrd_size __initdata = 0;
56 
early_initrd(char * p)57 static 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 }
72 early_param("initrd", early_initrd);
73 
parse_tag_initrd(const struct tag * tag)74 static 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 
parse_tag_initrd2(const struct tag * tag)85 static 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 
find_limits(unsigned long * min,unsigned long * max_low,unsigned long * max_high)94 static 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 
104 phys_addr_t arm_dma_zone_size __read_mostly;
105 EXPORT_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  */
113 phys_addr_t arm_dma_limit;
114 unsigned long arm_dma_pfn_limit;
115 
arm_adjust_dma_zone(unsigned long * size,unsigned long * hole,unsigned long dma_size)116 static 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 
setup_dma_zone(const struct machine_desc * mdesc)129 void __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 
zone_sizes_init(unsigned long min,unsigned long max_low,unsigned long max_high)141 static 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
pfn_valid(unsigned long pfn)197 int pfn_valid(unsigned long pfn)
198 {
199 	phys_addr_t addr = __pfn_to_phys(pfn);
200 
201 	if (__phys_to_pfn(addr) != pfn)
202 		return 0;
203 
204 	return memblock_is_map_memory(__pfn_to_phys(pfn));
205 }
206 EXPORT_SYMBOL(pfn_valid);
207 #endif
208 
209 #ifndef CONFIG_SPARSEMEM
arm_memory_present(void)210 static void __init arm_memory_present(void)
211 {
212 }
213 #else
arm_memory_present(void)214 static void __init arm_memory_present(void)
215 {
216 	struct memblock_region *reg;
217 
218 	for_each_memblock(memory, reg)
219 		memory_present(0, memblock_region_memory_base_pfn(reg),
220 			       memblock_region_memory_end_pfn(reg));
221 }
222 #endif
223 
224 static bool arm_memblock_steal_permitted = true;
225 
arm_memblock_steal(phys_addr_t size,phys_addr_t align)226 phys_addr_t __init arm_memblock_steal(phys_addr_t size, phys_addr_t align)
227 {
228 	phys_addr_t phys;
229 
230 	BUG_ON(!arm_memblock_steal_permitted);
231 
232 	phys = memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ANYWHERE);
233 	memblock_free(phys, size);
234 	memblock_remove(phys, size);
235 
236 	return phys;
237 }
238 
arm_initrd_init(void)239 static void __init arm_initrd_init(void)
240 {
241 #ifdef CONFIG_BLK_DEV_INITRD
242 	phys_addr_t start;
243 	unsigned long size;
244 
245 	/* FDT scan will populate initrd_start */
246 	if (initrd_start && !phys_initrd_size) {
247 		phys_initrd_start = __virt_to_phys(initrd_start);
248 		phys_initrd_size = initrd_end - initrd_start;
249 	}
250 
251 	initrd_start = initrd_end = 0;
252 
253 	if (!phys_initrd_size)
254 		return;
255 
256 	/*
257 	 * Round the memory region to page boundaries as per free_initrd_mem()
258 	 * This allows us to detect whether the pages overlapping the initrd
259 	 * are in use, but more importantly, reserves the entire set of pages
260 	 * as we don't want these pages allocated for other purposes.
261 	 */
262 	start = round_down(phys_initrd_start, PAGE_SIZE);
263 	size = phys_initrd_size + (phys_initrd_start - start);
264 	size = round_up(size, PAGE_SIZE);
265 
266 	if (!memblock_is_region_memory(start, size)) {
267 		pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region - disabling initrd\n",
268 		       (u64)start, size);
269 		return;
270 	}
271 
272 	if (memblock_is_region_reserved(start, size)) {
273 		pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region - disabling initrd\n",
274 		       (u64)start, size);
275 		return;
276 	}
277 
278 	memblock_reserve(start, size);
279 
280 	/* Now convert initrd to virtual addresses */
281 	initrd_start = __phys_to_virt(phys_initrd_start);
282 	initrd_end = initrd_start + phys_initrd_size;
283 #endif
284 }
285 
arm_memblock_init(const struct machine_desc * mdesc)286 void __init arm_memblock_init(const struct machine_desc *mdesc)
287 {
288 	/* Register the kernel text, kernel data and initrd with memblock. */
289 	memblock_reserve(__pa(KERNEL_START), KERNEL_END - KERNEL_START);
290 
291 	arm_initrd_init();
292 
293 	arm_mm_memblock_reserve();
294 
295 	/* reserve any platform specific memblock areas */
296 	if (mdesc->reserve)
297 		mdesc->reserve();
298 
299 	early_init_fdt_reserve_self();
300 	early_init_fdt_scan_reserved_mem();
301 
302 	/* reserve memory for DMA contiguous allocations */
303 	dma_contiguous_reserve(arm_dma_limit);
304 
305 	arm_memblock_steal_permitted = false;
306 	memblock_dump_all();
307 }
308 
bootmem_init(void)309 void __init bootmem_init(void)
310 {
311 	unsigned long min, max_low, max_high;
312 
313 	memblock_allow_resize();
314 	max_low = max_high = 0;
315 
316 	find_limits(&min, &max_low, &max_high);
317 
318 	early_memtest((phys_addr_t)min << PAGE_SHIFT,
319 		      (phys_addr_t)max_low << PAGE_SHIFT);
320 
321 	/*
322 	 * Sparsemem tries to allocate bootmem in memory_present(),
323 	 * so must be done after the fixed reservations
324 	 */
325 	arm_memory_present();
326 
327 	/*
328 	 * sparse_init() needs the bootmem allocator up and running.
329 	 */
330 	sparse_init();
331 
332 	/*
333 	 * Now free the memory - free_area_init_node needs
334 	 * the sparse mem_map arrays initialized by sparse_init()
335 	 * for memmap_init_zone(), otherwise all PFNs are invalid.
336 	 */
337 	zone_sizes_init(min, max_low, max_high);
338 
339 	/*
340 	 * This doesn't seem to be used by the Linux memory manager any
341 	 * more, but is used by ll_rw_block.  If we can get rid of it, we
342 	 * also get rid of some of the stuff above as well.
343 	 */
344 	min_low_pfn = min;
345 	max_low_pfn = max_low;
346 	max_pfn = max_high;
347 }
348 
349 /*
350  * Poison init memory with an undefined instruction (ARM) or a branch to an
351  * undefined instruction (Thumb).
352  */
poison_init_mem(void * s,size_t count)353 static inline void poison_init_mem(void *s, size_t count)
354 {
355 	u32 *p = (u32 *)s;
356 	for (; count != 0; count -= 4)
357 		*p++ = 0xe7fddef0;
358 }
359 
360 static inline void __init
free_memmap(unsigned long start_pfn,unsigned long end_pfn)361 free_memmap(unsigned long start_pfn, unsigned long end_pfn)
362 {
363 	struct page *start_pg, *end_pg;
364 	phys_addr_t pg, pgend;
365 
366 	/*
367 	 * Convert start_pfn/end_pfn to a struct page pointer.
368 	 */
369 	start_pg = pfn_to_page(start_pfn - 1) + 1;
370 	end_pg = pfn_to_page(end_pfn - 1) + 1;
371 
372 	/*
373 	 * Convert to physical addresses, and
374 	 * round start upwards and end downwards.
375 	 */
376 	pg = PAGE_ALIGN(__pa(start_pg));
377 	pgend = __pa(end_pg) & PAGE_MASK;
378 
379 	/*
380 	 * If there are free pages between these,
381 	 * free the section of the memmap array.
382 	 */
383 	if (pg < pgend)
384 		memblock_free_early(pg, pgend - pg);
385 }
386 
387 /*
388  * The mem_map array can get very big.  Free the unused area of the memory map.
389  */
free_unused_memmap(void)390 static void __init free_unused_memmap(void)
391 {
392 	unsigned long start, prev_end = 0;
393 	struct memblock_region *reg;
394 
395 	/*
396 	 * This relies on each bank being in address order.
397 	 * The banks are sorted previously in bootmem_init().
398 	 */
399 	for_each_memblock(memory, reg) {
400 		start = memblock_region_memory_base_pfn(reg);
401 
402 #ifdef CONFIG_SPARSEMEM
403 		/*
404 		 * Take care not to free memmap entries that don't exist
405 		 * due to SPARSEMEM sections which aren't present.
406 		 */
407 		start = min(start,
408 				 ALIGN(prev_end, PAGES_PER_SECTION));
409 #else
410 		/*
411 		 * Align down here since the VM subsystem insists that the
412 		 * memmap entries are valid from the bank start aligned to
413 		 * MAX_ORDER_NR_PAGES.
414 		 */
415 		start = round_down(start, MAX_ORDER_NR_PAGES);
416 #endif
417 		/*
418 		 * If we had a previous bank, and there is a space
419 		 * between the current bank and the previous, free it.
420 		 */
421 		if (prev_end && prev_end < start)
422 			free_memmap(prev_end, start);
423 
424 		/*
425 		 * Align up here since the VM subsystem insists that the
426 		 * memmap entries are valid from the bank end aligned to
427 		 * MAX_ORDER_NR_PAGES.
428 		 */
429 		prev_end = ALIGN(memblock_region_memory_end_pfn(reg),
430 				 MAX_ORDER_NR_PAGES);
431 	}
432 
433 #ifdef CONFIG_SPARSEMEM
434 	if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
435 		free_memmap(prev_end,
436 			    ALIGN(prev_end, PAGES_PER_SECTION));
437 #endif
438 }
439 
440 #ifdef CONFIG_HIGHMEM
free_area_high(unsigned long pfn,unsigned long end)441 static inline void free_area_high(unsigned long pfn, unsigned long end)
442 {
443 	for (; pfn < end; pfn++)
444 		free_highmem_page(pfn_to_page(pfn));
445 }
446 #endif
447 
free_highpages(void)448 static void __init free_highpages(void)
449 {
450 #ifdef CONFIG_HIGHMEM
451 	unsigned long max_low = max_low_pfn;
452 	struct memblock_region *mem, *res;
453 
454 	/* set highmem page free */
455 	for_each_memblock(memory, mem) {
456 		unsigned long start = memblock_region_memory_base_pfn(mem);
457 		unsigned long end = memblock_region_memory_end_pfn(mem);
458 
459 		/* Ignore complete lowmem entries */
460 		if (end <= max_low)
461 			continue;
462 
463 		if (memblock_is_nomap(mem))
464 			continue;
465 
466 		/* Truncate partial highmem entries */
467 		if (start < max_low)
468 			start = max_low;
469 
470 		/* Find and exclude any reserved regions */
471 		for_each_memblock(reserved, res) {
472 			unsigned long res_start, res_end;
473 
474 			res_start = memblock_region_reserved_base_pfn(res);
475 			res_end = memblock_region_reserved_end_pfn(res);
476 
477 			if (res_end < start)
478 				continue;
479 			if (res_start < start)
480 				res_start = start;
481 			if (res_start > end)
482 				res_start = end;
483 			if (res_end > end)
484 				res_end = end;
485 			if (res_start != start)
486 				free_area_high(start, res_start);
487 			start = res_end;
488 			if (start == end)
489 				break;
490 		}
491 
492 		/* And now free anything which remains */
493 		if (start < end)
494 			free_area_high(start, end);
495 	}
496 #endif
497 }
498 
499 /*
500  * mem_init() marks the free areas in the mem_map and tells us how much
501  * memory is free.  This is done after various parts of the system have
502  * claimed their memory after the kernel image.
503  */
mem_init(void)504 void __init mem_init(void)
505 {
506 #ifdef CONFIG_HAVE_TCM
507 	/* These pointers are filled in on TCM detection */
508 	extern u32 dtcm_end;
509 	extern u32 itcm_end;
510 #endif
511 
512 	set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
513 
514 	/* this will put all unused low memory onto the freelists */
515 	free_unused_memmap();
516 	free_all_bootmem();
517 
518 #ifdef CONFIG_SA1111
519 	/* now that our DMA memory is actually so designated, we can free it */
520 	free_reserved_area(__va(PHYS_OFFSET), swapper_pg_dir, -1, NULL);
521 #endif
522 
523 	free_highpages();
524 
525 	mem_init_print_info(NULL);
526 
527 #define MLK(b, t) b, t, ((t) - (b)) >> 10
528 #define MLM(b, t) b, t, ((t) - (b)) >> 20
529 #define MLK_ROUNDUP(b, t) b, t, DIV_ROUND_UP(((t) - (b)), SZ_1K)
530 
531 	pr_notice("Virtual kernel memory layout:\n"
532 			"    vector  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
533 #ifdef CONFIG_HAVE_TCM
534 			"    DTCM    : 0x%08lx - 0x%08lx   (%4ld kB)\n"
535 			"    ITCM    : 0x%08lx - 0x%08lx   (%4ld kB)\n"
536 #endif
537 			"    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
538 			"    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
539 			"    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
540 #ifdef CONFIG_HIGHMEM
541 			"    pkmap   : 0x%08lx - 0x%08lx   (%4ld MB)\n"
542 #endif
543 #ifdef CONFIG_MODULES
544 			"    modules : 0x%08lx - 0x%08lx   (%4ld MB)\n"
545 #endif
546 			"      .text : 0x%p" " - 0x%p" "   (%4td kB)\n"
547 			"      .init : 0x%p" " - 0x%p" "   (%4td kB)\n"
548 			"      .data : 0x%p" " - 0x%p" "   (%4td kB)\n"
549 			"       .bss : 0x%p" " - 0x%p" "   (%4td kB)\n",
550 
551 			MLK(VECTORS_BASE, VECTORS_BASE + PAGE_SIZE),
552 #ifdef CONFIG_HAVE_TCM
553 			MLK(DTCM_OFFSET, (unsigned long) dtcm_end),
554 			MLK(ITCM_OFFSET, (unsigned long) itcm_end),
555 #endif
556 			MLK(FIXADDR_START, FIXADDR_END),
557 			MLM(VMALLOC_START, VMALLOC_END),
558 			MLM(PAGE_OFFSET, (unsigned long)high_memory),
559 #ifdef CONFIG_HIGHMEM
560 			MLM(PKMAP_BASE, (PKMAP_BASE) + (LAST_PKMAP) *
561 				(PAGE_SIZE)),
562 #endif
563 #ifdef CONFIG_MODULES
564 			MLM(MODULES_VADDR, MODULES_END),
565 #endif
566 
567 			MLK_ROUNDUP(_text, _etext),
568 			MLK_ROUNDUP(__init_begin, __init_end),
569 			MLK_ROUNDUP(_sdata, _edata),
570 			MLK_ROUNDUP(__bss_start, __bss_stop));
571 
572 #undef MLK
573 #undef MLM
574 #undef MLK_ROUNDUP
575 
576 	/*
577 	 * Check boundaries twice: Some fundamental inconsistencies can
578 	 * be detected at build time already.
579 	 */
580 #ifdef CONFIG_MMU
581 	BUILD_BUG_ON(TASK_SIZE				> MODULES_VADDR);
582 	BUG_ON(TASK_SIZE 				> MODULES_VADDR);
583 #endif
584 
585 #ifdef CONFIG_HIGHMEM
586 	BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
587 	BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE	> PAGE_OFFSET);
588 #endif
589 }
590 
591 #ifdef CONFIG_STRICT_KERNEL_RWX
592 struct section_perm {
593 	const char *name;
594 	unsigned long start;
595 	unsigned long end;
596 	pmdval_t mask;
597 	pmdval_t prot;
598 	pmdval_t clear;
599 };
600 
601 /* First section-aligned location at or after __start_rodata. */
602 extern char __start_rodata_section_aligned[];
603 
604 static struct section_perm nx_perms[] = {
605 	/* Make pages tables, etc before _stext RW (set NX). */
606 	{
607 		.name	= "pre-text NX",
608 		.start	= PAGE_OFFSET,
609 		.end	= (unsigned long)_stext,
610 		.mask	= ~PMD_SECT_XN,
611 		.prot	= PMD_SECT_XN,
612 	},
613 	/* Make init RW (set NX). */
614 	{
615 		.name	= "init NX",
616 		.start	= (unsigned long)__init_begin,
617 		.end	= (unsigned long)_sdata,
618 		.mask	= ~PMD_SECT_XN,
619 		.prot	= PMD_SECT_XN,
620 	},
621 	/* Make rodata NX (set RO in ro_perms below). */
622 	{
623 		.name	= "rodata NX",
624 		.start  = (unsigned long)__start_rodata_section_aligned,
625 		.end    = (unsigned long)__init_begin,
626 		.mask   = ~PMD_SECT_XN,
627 		.prot   = PMD_SECT_XN,
628 	},
629 };
630 
631 static struct section_perm ro_perms[] = {
632 	/* Make kernel code and rodata RX (set RO). */
633 	{
634 		.name	= "text/rodata RO",
635 		.start  = (unsigned long)_stext,
636 		.end    = (unsigned long)__init_begin,
637 #ifdef CONFIG_ARM_LPAE
638 		.mask   = ~(L_PMD_SECT_RDONLY | PMD_SECT_AP2),
639 		.prot   = L_PMD_SECT_RDONLY | PMD_SECT_AP2,
640 #else
641 		.mask   = ~(PMD_SECT_APX | PMD_SECT_AP_WRITE),
642 		.prot   = PMD_SECT_APX | PMD_SECT_AP_WRITE,
643 		.clear  = PMD_SECT_AP_WRITE,
644 #endif
645 	},
646 };
647 
648 /*
649  * Updates section permissions only for the current mm (sections are
650  * copied into each mm). During startup, this is the init_mm. Is only
651  * safe to be called with preemption disabled, as under stop_machine().
652  */
section_update(unsigned long addr,pmdval_t mask,pmdval_t prot,struct mm_struct * mm)653 static inline void section_update(unsigned long addr, pmdval_t mask,
654 				  pmdval_t prot, struct mm_struct *mm)
655 {
656 	pmd_t *pmd;
657 
658 	pmd = pmd_offset(pud_offset(pgd_offset(mm, addr), addr), addr);
659 
660 #ifdef CONFIG_ARM_LPAE
661 	pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
662 #else
663 	if (addr & SECTION_SIZE)
664 		pmd[1] = __pmd((pmd_val(pmd[1]) & mask) | prot);
665 	else
666 		pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
667 #endif
668 	flush_pmd_entry(pmd);
669 	local_flush_tlb_kernel_range(addr, addr + SECTION_SIZE);
670 }
671 
672 /* Make sure extended page tables are in use. */
arch_has_strict_perms(void)673 static inline bool arch_has_strict_perms(void)
674 {
675 	if (cpu_architecture() < CPU_ARCH_ARMv6)
676 		return false;
677 
678 	return !!(get_cr() & CR_XP);
679 }
680 
set_section_perms(struct section_perm * perms,int n,bool set,struct mm_struct * mm)681 void set_section_perms(struct section_perm *perms, int n, bool set,
682 			struct mm_struct *mm)
683 {
684 	size_t i;
685 	unsigned long addr;
686 
687 	if (!arch_has_strict_perms())
688 		return;
689 
690 	for (i = 0; i < n; i++) {
691 		if (!IS_ALIGNED(perms[i].start, SECTION_SIZE) ||
692 		    !IS_ALIGNED(perms[i].end, SECTION_SIZE)) {
693 			pr_err("BUG: %s section %lx-%lx not aligned to %lx\n",
694 				perms[i].name, perms[i].start, perms[i].end,
695 				SECTION_SIZE);
696 			continue;
697 		}
698 
699 		for (addr = perms[i].start;
700 		     addr < perms[i].end;
701 		     addr += SECTION_SIZE)
702 			section_update(addr, perms[i].mask,
703 				set ? perms[i].prot : perms[i].clear, mm);
704 	}
705 
706 }
707 
708 /**
709  * update_sections_early intended to be called only through stop_machine
710  * framework and executed by only one CPU while all other CPUs will spin and
711  * wait, so no locking is required in this function.
712  */
update_sections_early(struct section_perm perms[],int n)713 static void update_sections_early(struct section_perm perms[], int n)
714 {
715 	struct task_struct *t, *s;
716 
717 	for_each_process(t) {
718 		if (t->flags & PF_KTHREAD)
719 			continue;
720 		for_each_thread(t, s)
721 			if (s->mm)
722 				set_section_perms(perms, n, true, s->mm);
723 	}
724 	set_section_perms(perms, n, true, current->active_mm);
725 	set_section_perms(perms, n, true, &init_mm);
726 }
727 
__fix_kernmem_perms(void * unused)728 static int __fix_kernmem_perms(void *unused)
729 {
730 	update_sections_early(nx_perms, ARRAY_SIZE(nx_perms));
731 	return 0;
732 }
733 
fix_kernmem_perms(void)734 static void fix_kernmem_perms(void)
735 {
736 	stop_machine(__fix_kernmem_perms, NULL, NULL);
737 }
738 
__mark_rodata_ro(void * unused)739 static int __mark_rodata_ro(void *unused)
740 {
741 	update_sections_early(ro_perms, ARRAY_SIZE(ro_perms));
742 	return 0;
743 }
744 
745 static int kernel_set_to_readonly __read_mostly;
746 
mark_rodata_ro(void)747 void mark_rodata_ro(void)
748 {
749 	kernel_set_to_readonly = 1;
750 	stop_machine(__mark_rodata_ro, NULL, NULL);
751 	debug_checkwx();
752 }
753 
set_kernel_text_rw(void)754 void set_kernel_text_rw(void)
755 {
756 	if (!kernel_set_to_readonly)
757 		return;
758 
759 	set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), false,
760 				current->active_mm);
761 }
762 
set_kernel_text_ro(void)763 void set_kernel_text_ro(void)
764 {
765 	if (!kernel_set_to_readonly)
766 		return;
767 
768 	set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), true,
769 				current->active_mm);
770 }
771 
772 #else
fix_kernmem_perms(void)773 static inline void fix_kernmem_perms(void) { }
774 #endif /* CONFIG_STRICT_KERNEL_RWX */
775 
free_initmem(void)776 void free_initmem(void)
777 {
778 	fix_kernmem_perms();
779 
780 	poison_init_mem(__init_begin, __init_end - __init_begin);
781 	if (!machine_is_integrator() && !machine_is_cintegrator())
782 		free_initmem_default(-1);
783 }
784 
785 #ifdef CONFIG_BLK_DEV_INITRD
786 
787 static int keep_initrd;
788 
free_initrd_mem(unsigned long start,unsigned long end)789 void free_initrd_mem(unsigned long start, unsigned long end)
790 {
791 	if (!keep_initrd) {
792 		if (start == initrd_start)
793 			start = round_down(start, PAGE_SIZE);
794 		if (end == initrd_end)
795 			end = round_up(end, PAGE_SIZE);
796 
797 		poison_init_mem((void *)start, PAGE_ALIGN(end) - start);
798 		free_reserved_area((void *)start, (void *)end, -1, "initrd");
799 	}
800 }
801 
keepinitrd_setup(char * __unused)802 static int __init keepinitrd_setup(char *__unused)
803 {
804 	keep_initrd = 1;
805 	return 1;
806 }
807 
808 __setup("keepinitrd", keepinitrd_setup);
809 #endif
810