Loading...
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
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 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
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#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
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
129 if (__phys_to_pfn(addr) != pfn)
130 return 0;
131
132 return memblock_is_map_memory(addr);
133}
134EXPORT_SYMBOL(pfn_valid);
135#endif
136
137static bool arm_memblock_steal_permitted = true;
138
139phys_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
156static 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
198void 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
213void __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_reserve_self();
227 early_init_fdt_scan_reserved_mem();
228
229 /* reserve memory for DMA contiguous allocations */
230 dma_contiguous_reserve(arm_dma_limit);
231
232 arm_memblock_steal_permitted = false;
233 memblock_dump_all();
234}
235
236void __init bootmem_init(void)
237{
238 memblock_allow_resize();
239
240 find_limits(&min_low_pfn, &max_low_pfn, &max_pfn);
241
242 early_memtest((phys_addr_t)min_low_pfn << PAGE_SHIFT,
243 (phys_addr_t)max_low_pfn << PAGE_SHIFT);
244
245 /*
246 * sparse_init() tries to allocate memory from memblock, so must be
247 * done after the fixed reservations
248 */
249 sparse_init();
250
251 /*
252 * Now free the memory - free_area_init needs
253 * the sparse mem_map arrays initialized by sparse_init()
254 * for memmap_init_zone(), otherwise all PFNs are invalid.
255 */
256 zone_sizes_init(min_low_pfn, max_low_pfn, max_pfn);
257}
258
259/*
260 * Poison init memory with an undefined instruction (ARM) or a branch to an
261 * undefined instruction (Thumb).
262 */
263static inline void poison_init_mem(void *s, size_t count)
264{
265 u32 *p = (u32 *)s;
266 for (; count != 0; count -= 4)
267 *p++ = 0xe7fddef0;
268}
269
270static inline void __init
271free_memmap(unsigned long start_pfn, unsigned long end_pfn)
272{
273 struct page *start_pg, *end_pg;
274 phys_addr_t pg, pgend;
275
276 /*
277 * Convert start_pfn/end_pfn to a struct page pointer.
278 */
279 start_pg = pfn_to_page(start_pfn - 1) + 1;
280 end_pg = pfn_to_page(end_pfn - 1) + 1;
281
282 /*
283 * Convert to physical addresses, and
284 * round start upwards and end downwards.
285 */
286 pg = PAGE_ALIGN(__pa(start_pg));
287 pgend = __pa(end_pg) & PAGE_MASK;
288
289 /*
290 * If there are free pages between these,
291 * free the section of the memmap array.
292 */
293 if (pg < pgend)
294 memblock_free_early(pg, pgend - pg);
295}
296
297/*
298 * The mem_map array can get very big. Free the unused area of the memory map.
299 */
300static void __init free_unused_memmap(void)
301{
302 unsigned long start, prev_end = 0;
303 struct memblock_region *reg;
304
305 /*
306 * This relies on each bank being in address order.
307 * The banks are sorted previously in bootmem_init().
308 */
309 for_each_memblock(memory, reg) {
310 start = memblock_region_memory_base_pfn(reg);
311
312#ifdef CONFIG_SPARSEMEM
313 /*
314 * Take care not to free memmap entries that don't exist
315 * due to SPARSEMEM sections which aren't present.
316 */
317 start = min(start,
318 ALIGN(prev_end, PAGES_PER_SECTION));
319#else
320 /*
321 * Align down here since the VM subsystem insists that the
322 * memmap entries are valid from the bank start aligned to
323 * MAX_ORDER_NR_PAGES.
324 */
325 start = round_down(start, MAX_ORDER_NR_PAGES);
326#endif
327 /*
328 * If we had a previous bank, and there is a space
329 * between the current bank and the previous, free it.
330 */
331 if (prev_end && prev_end < start)
332 free_memmap(prev_end, start);
333
334 /*
335 * Align up here since the VM subsystem insists that the
336 * memmap entries are valid from the bank end aligned to
337 * MAX_ORDER_NR_PAGES.
338 */
339 prev_end = ALIGN(memblock_region_memory_end_pfn(reg),
340 MAX_ORDER_NR_PAGES);
341 }
342
343#ifdef CONFIG_SPARSEMEM
344 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
345 free_memmap(prev_end,
346 ALIGN(prev_end, PAGES_PER_SECTION));
347#endif
348}
349
350#ifdef CONFIG_HIGHMEM
351static inline void free_area_high(unsigned long pfn, unsigned long end)
352{
353 for (; pfn < end; pfn++)
354 free_highmem_page(pfn_to_page(pfn));
355}
356#endif
357
358static void __init free_highpages(void)
359{
360#ifdef CONFIG_HIGHMEM
361 unsigned long max_low = max_low_pfn;
362 struct memblock_region *mem, *res;
363
364 /* set highmem page free */
365 for_each_memblock(memory, mem) {
366 unsigned long start = memblock_region_memory_base_pfn(mem);
367 unsigned long end = memblock_region_memory_end_pfn(mem);
368
369 /* Ignore complete lowmem entries */
370 if (end <= max_low)
371 continue;
372
373 if (memblock_is_nomap(mem))
374 continue;
375
376 /* Truncate partial highmem entries */
377 if (start < max_low)
378 start = max_low;
379
380 /* Find and exclude any reserved regions */
381 for_each_memblock(reserved, res) {
382 unsigned long res_start, res_end;
383
384 res_start = memblock_region_reserved_base_pfn(res);
385 res_end = memblock_region_reserved_end_pfn(res);
386
387 if (res_end < start)
388 continue;
389 if (res_start < start)
390 res_start = start;
391 if (res_start > end)
392 res_start = end;
393 if (res_end > end)
394 res_end = end;
395 if (res_start != start)
396 free_area_high(start, res_start);
397 start = res_end;
398 if (start == end)
399 break;
400 }
401
402 /* And now free anything which remains */
403 if (start < end)
404 free_area_high(start, end);
405 }
406#endif
407}
408
409/*
410 * mem_init() marks the free areas in the mem_map and tells us how much
411 * memory is free. This is done after various parts of the system have
412 * claimed their memory after the kernel image.
413 */
414void __init mem_init(void)
415{
416#ifdef CONFIG_ARM_LPAE
417 swiotlb_init(1);
418#endif
419
420 set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
421
422 /* this will put all unused low memory onto the freelists */
423 free_unused_memmap();
424 memblock_free_all();
425
426#ifdef CONFIG_SA1111
427 /* now that our DMA memory is actually so designated, we can free it */
428 free_reserved_area(__va(PHYS_OFFSET), swapper_pg_dir, -1, NULL);
429#endif
430
431 free_highpages();
432
433 mem_init_print_info(NULL);
434
435 /*
436 * Check boundaries twice: Some fundamental inconsistencies can
437 * be detected at build time already.
438 */
439#ifdef CONFIG_MMU
440 BUILD_BUG_ON(TASK_SIZE > MODULES_VADDR);
441 BUG_ON(TASK_SIZE > MODULES_VADDR);
442#endif
443
444#ifdef CONFIG_HIGHMEM
445 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
446 BUG_ON(PKMAP_BASE + LAST_PKMAP * PAGE_SIZE > PAGE_OFFSET);
447#endif
448}
449
450#ifdef CONFIG_STRICT_KERNEL_RWX
451struct section_perm {
452 const char *name;
453 unsigned long start;
454 unsigned long end;
455 pmdval_t mask;
456 pmdval_t prot;
457 pmdval_t clear;
458};
459
460/* First section-aligned location at or after __start_rodata. */
461extern char __start_rodata_section_aligned[];
462
463static struct section_perm nx_perms[] = {
464 /* Make pages tables, etc before _stext RW (set NX). */
465 {
466 .name = "pre-text NX",
467 .start = PAGE_OFFSET,
468 .end = (unsigned long)_stext,
469 .mask = ~PMD_SECT_XN,
470 .prot = PMD_SECT_XN,
471 },
472 /* Make init RW (set NX). */
473 {
474 .name = "init NX",
475 .start = (unsigned long)__init_begin,
476 .end = (unsigned long)_sdata,
477 .mask = ~PMD_SECT_XN,
478 .prot = PMD_SECT_XN,
479 },
480 /* Make rodata NX (set RO in ro_perms below). */
481 {
482 .name = "rodata NX",
483 .start = (unsigned long)__start_rodata_section_aligned,
484 .end = (unsigned long)__init_begin,
485 .mask = ~PMD_SECT_XN,
486 .prot = PMD_SECT_XN,
487 },
488};
489
490static struct section_perm ro_perms[] = {
491 /* Make kernel code and rodata RX (set RO). */
492 {
493 .name = "text/rodata RO",
494 .start = (unsigned long)_stext,
495 .end = (unsigned long)__init_begin,
496#ifdef CONFIG_ARM_LPAE
497 .mask = ~(L_PMD_SECT_RDONLY | PMD_SECT_AP2),
498 .prot = L_PMD_SECT_RDONLY | PMD_SECT_AP2,
499#else
500 .mask = ~(PMD_SECT_APX | PMD_SECT_AP_WRITE),
501 .prot = PMD_SECT_APX | PMD_SECT_AP_WRITE,
502 .clear = PMD_SECT_AP_WRITE,
503#endif
504 },
505};
506
507/*
508 * Updates section permissions only for the current mm (sections are
509 * copied into each mm). During startup, this is the init_mm. Is only
510 * safe to be called with preemption disabled, as under stop_machine().
511 */
512static inline void section_update(unsigned long addr, pmdval_t mask,
513 pmdval_t prot, struct mm_struct *mm)
514{
515 pmd_t *pmd;
516
517 pmd = pmd_offset(pud_offset(p4d_offset(pgd_offset(mm, addr), addr), addr), addr);
518
519#ifdef CONFIG_ARM_LPAE
520 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
521#else
522 if (addr & SECTION_SIZE)
523 pmd[1] = __pmd((pmd_val(pmd[1]) & mask) | prot);
524 else
525 pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
526#endif
527 flush_pmd_entry(pmd);
528 local_flush_tlb_kernel_range(addr, addr + SECTION_SIZE);
529}
530
531/* Make sure extended page tables are in use. */
532static inline bool arch_has_strict_perms(void)
533{
534 if (cpu_architecture() < CPU_ARCH_ARMv6)
535 return false;
536
537 return !!(get_cr() & CR_XP);
538}
539
540static void set_section_perms(struct section_perm *perms, int n, bool set,
541 struct mm_struct *mm)
542{
543 size_t i;
544 unsigned long addr;
545
546 if (!arch_has_strict_perms())
547 return;
548
549 for (i = 0; i < n; i++) {
550 if (!IS_ALIGNED(perms[i].start, SECTION_SIZE) ||
551 !IS_ALIGNED(perms[i].end, SECTION_SIZE)) {
552 pr_err("BUG: %s section %lx-%lx not aligned to %lx\n",
553 perms[i].name, perms[i].start, perms[i].end,
554 SECTION_SIZE);
555 continue;
556 }
557
558 for (addr = perms[i].start;
559 addr < perms[i].end;
560 addr += SECTION_SIZE)
561 section_update(addr, perms[i].mask,
562 set ? perms[i].prot : perms[i].clear, mm);
563 }
564
565}
566
567/**
568 * update_sections_early intended to be called only through stop_machine
569 * framework and executed by only one CPU while all other CPUs will spin and
570 * wait, so no locking is required in this function.
571 */
572static void update_sections_early(struct section_perm perms[], int n)
573{
574 struct task_struct *t, *s;
575
576 for_each_process(t) {
577 if (t->flags & PF_KTHREAD)
578 continue;
579 for_each_thread(t, s)
580 if (s->mm)
581 set_section_perms(perms, n, true, s->mm);
582 }
583 set_section_perms(perms, n, true, current->active_mm);
584 set_section_perms(perms, n, true, &init_mm);
585}
586
587static int __fix_kernmem_perms(void *unused)
588{
589 update_sections_early(nx_perms, ARRAY_SIZE(nx_perms));
590 return 0;
591}
592
593static void fix_kernmem_perms(void)
594{
595 stop_machine(__fix_kernmem_perms, NULL, NULL);
596}
597
598static int __mark_rodata_ro(void *unused)
599{
600 update_sections_early(ro_perms, ARRAY_SIZE(ro_perms));
601 return 0;
602}
603
604static int kernel_set_to_readonly __read_mostly;
605
606void mark_rodata_ro(void)
607{
608 kernel_set_to_readonly = 1;
609 stop_machine(__mark_rodata_ro, NULL, NULL);
610 debug_checkwx();
611}
612
613void set_kernel_text_rw(void)
614{
615 if (!kernel_set_to_readonly)
616 return;
617
618 set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), false,
619 current->active_mm);
620}
621
622void set_kernel_text_ro(void)
623{
624 if (!kernel_set_to_readonly)
625 return;
626
627 set_section_perms(ro_perms, ARRAY_SIZE(ro_perms), true,
628 current->active_mm);
629}
630
631#else
632static inline void fix_kernmem_perms(void) { }
633#endif /* CONFIG_STRICT_KERNEL_RWX */
634
635void free_initmem(void)
636{
637 fix_kernmem_perms();
638
639 poison_init_mem(__init_begin, __init_end - __init_begin);
640 if (!machine_is_integrator() && !machine_is_cintegrator())
641 free_initmem_default(-1);
642}
643
644#ifdef CONFIG_BLK_DEV_INITRD
645void free_initrd_mem(unsigned long start, unsigned long end)
646{
647 if (start == initrd_start)
648 start = round_down(start, PAGE_SIZE);
649 if (end == initrd_end)
650 end = round_up(end, PAGE_SIZE);
651
652 poison_init_mem((void *)start, PAGE_ALIGN(end) - start);
653 free_reserved_area((void *)start, (void *)end, -1, "initrd");
654}
655#endif