arm64: cpuidle: make arm_cpuidle_suspend() a bit more efficient
[linux-2.6-block.git] / arch / arm64 / mm / init.c
CommitLineData
c1cc1552
CM
1/*
2 * Based on arch/arm/mm/init.c
3 *
4 * Copyright (C) 1995-2005 Russell King
5 * Copyright (C) 2012 ARM Ltd.
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program. If not, see <http://www.gnu.org/licenses/>.
18 */
19
20#include <linux/kernel.h>
21#include <linux/export.h>
22#include <linux/errno.h>
23#include <linux/swap.h>
24#include <linux/init.h>
25#include <linux/bootmem.h>
26#include <linux/mman.h>
27#include <linux/nodemask.h>
28#include <linux/initrd.h>
29#include <linux/gfp.h>
30#include <linux/memblock.h>
31#include <linux/sort.h>
32#include <linux/of_fdt.h>
19e7640d 33#include <linux/dma-mapping.h>
6ac2104d 34#include <linux/dma-contiguous.h>
86c8b27a 35#include <linux/efi.h>
a1e50a82 36#include <linux/swiotlb.h>
c1cc1552 37
a7f8de16 38#include <asm/boot.h>
08375198 39#include <asm/fixmap.h>
f9040773 40#include <asm/kasan.h>
a7f8de16 41#include <asm/kernel-pgtable.h>
aa03c428 42#include <asm/memory.h>
c1cc1552
CM
43#include <asm/sections.h>
44#include <asm/setup.h>
45#include <asm/sizes.h>
46#include <asm/tlb.h>
e039ee4e 47#include <asm/alternative.h>
c1cc1552
CM
48
49#include "mm.h"
50
a7f8de16
AB
51/*
52 * We need to be able to catch inadvertent references to memstart_addr
53 * that occur (potentially in generic code) before arm64_memblock_init()
54 * executes, which assigns it its actual value. So use a default value
55 * that cannot be mistaken for a real physical address.
56 */
020d044f 57s64 memstart_addr __read_mostly = -1;
a1e50a82 58phys_addr_t arm64_dma_phys_limit __read_mostly;
c1cc1552 59
ec2eaa73 60#ifdef CONFIG_BLK_DEV_INITRD
c1cc1552
CM
61static int __init early_initrd(char *p)
62{
63 unsigned long start, size;
64 char *endp;
65
66 start = memparse(p, &endp);
67 if (*endp == ',') {
68 size = memparse(endp + 1, NULL);
69
a89dea58
AB
70 initrd_start = start;
71 initrd_end = start + size;
c1cc1552
CM
72 }
73 return 0;
74}
75early_param("initrd", early_initrd);
ec2eaa73 76#endif
c1cc1552 77
d50314a6
CM
78/*
79 * Return the maximum physical address for ZONE_DMA (DMA_BIT_MASK(32)). It
80 * currently assumes that for memory starting above 4G, 32-bit devices will
81 * use a DMA offset.
82 */
a7c61a34 83static phys_addr_t __init max_zone_dma_phys(void)
d50314a6
CM
84{
85 phys_addr_t offset = memblock_start_of_DRAM() & GENMASK_ULL(63, 32);
86 return min(offset + (1ULL << 32), memblock_end_of_DRAM());
87}
88
c1cc1552
CM
89static void __init zone_sizes_init(unsigned long min, unsigned long max)
90{
91 struct memblock_region *reg;
92 unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES];
19e7640d 93 unsigned long max_dma = min;
c1cc1552
CM
94
95 memset(zone_size, 0, sizeof(zone_size));
96
c1cc1552 97 /* 4GB maximum for 32-bit only capable devices */
86a5906e
RM
98#ifdef CONFIG_ZONE_DMA
99 max_dma = PFN_DOWN(arm64_dma_phys_limit);
100 zone_size[ZONE_DMA] = max_dma - min;
101#endif
19e7640d 102 zone_size[ZONE_NORMAL] = max - max_dma;
c1cc1552
CM
103
104 memcpy(zhole_size, zone_size, sizeof(zhole_size));
105
106 for_each_memblock(memory, reg) {
107 unsigned long start = memblock_region_memory_base_pfn(reg);
108 unsigned long end = memblock_region_memory_end_pfn(reg);
109
110 if (start >= max)
111 continue;
19e7640d 112
86a5906e
RM
113#ifdef CONFIG_ZONE_DMA
114 if (start < max_dma) {
19e7640d
CM
115 unsigned long dma_end = min(end, max_dma);
116 zhole_size[ZONE_DMA] -= dma_end - start;
c1cc1552 117 }
86a5906e 118#endif
19e7640d 119 if (end > max_dma) {
c1cc1552 120 unsigned long normal_end = min(end, max);
19e7640d 121 unsigned long normal_start = max(start, max_dma);
c1cc1552
CM
122 zhole_size[ZONE_NORMAL] -= normal_end - normal_start;
123 }
124 }
125
126 free_area_init_node(0, zone_size, min, zhole_size);
127}
128
129#ifdef CONFIG_HAVE_ARCH_PFN_VALID
130int pfn_valid(unsigned long pfn)
131{
68709f45 132 return memblock_is_map_memory(pfn << PAGE_SHIFT);
c1cc1552
CM
133}
134EXPORT_SYMBOL(pfn_valid);
135#endif
136
137#ifndef CONFIG_SPARSEMEM
a7c61a34 138static void __init arm64_memory_present(void)
c1cc1552
CM
139{
140}
141#else
a7c61a34 142static void __init arm64_memory_present(void)
c1cc1552
CM
143{
144 struct memblock_region *reg;
145
146 for_each_memblock(memory, reg)
147 memory_present(0, memblock_region_memory_base_pfn(reg),
148 memblock_region_memory_end_pfn(reg));
149}
150#endif
151
6083fe74
MR
152static phys_addr_t memory_limit = (phys_addr_t)ULLONG_MAX;
153
154/*
155 * Limit the memory size that was specified via FDT.
156 */
157static int __init early_mem(char *p)
158{
159 if (!p)
160 return 1;
161
162 memory_limit = memparse(p, &p) & PAGE_MASK;
163 pr_notice("Memory limited to %lldMB\n", memory_limit >> 20);
164
165 return 0;
166}
167early_param("mem", early_mem);
168
c1cc1552
CM
169void __init arm64_memblock_init(void)
170{
a7f8de16
AB
171 const s64 linear_region_size = -(s64)PAGE_OFFSET;
172
6d2aa549
AB
173 /*
174 * Ensure that the linear region takes up exactly half of the kernel
175 * virtual address space. This way, we can distinguish a linear address
176 * from a kernel/module/vmalloc address by testing a single bit.
177 */
178 BUILD_BUG_ON(linear_region_size != BIT(VA_BITS - 1));
179
a7f8de16
AB
180 /*
181 * Select a suitable value for the base of physical memory.
182 */
183 memstart_addr = round_down(memblock_start_of_DRAM(),
184 ARM64_MEMSTART_ALIGN);
185
186 /*
187 * Remove the memory that we will not be able to cover with the
188 * linear mapping. Take care not to clip the kernel which may be
189 * high in memory.
190 */
020d044f 191 memblock_remove(max_t(u64, memstart_addr + linear_region_size, __pa(_end)),
a7f8de16
AB
192 ULLONG_MAX);
193 if (memblock_end_of_DRAM() > linear_region_size)
194 memblock_remove(0, memblock_end_of_DRAM() - linear_region_size);
195
196 /*
197 * Apply the memory limit if it was set. Since the kernel may be loaded
198 * high up in memory, add back the kernel region that must be accessible
199 * via the linear mapping.
200 */
201 if (memory_limit != (phys_addr_t)ULLONG_MAX) {
202 memblock_enforce_memory_limit(memory_limit);
203 memblock_add(__pa(_text), (u64)(_end - _text));
204 }
6083fe74 205
c031a421
AB
206 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
207 extern u16 memstart_offset_seed;
208 u64 range = linear_region_size -
209 (memblock_end_of_DRAM() - memblock_start_of_DRAM());
210
211 /*
212 * If the size of the linear region exceeds, by a sufficient
213 * margin, the size of the region that the available physical
214 * memory spans, randomize the linear region as well.
215 */
216 if (memstart_offset_seed > 0 && range >= ARM64_MEMSTART_ALIGN) {
217 range = range / ARM64_MEMSTART_ALIGN + 1;
218 memstart_addr -= ARM64_MEMSTART_ALIGN *
219 ((range * memstart_offset_seed) >> 16);
220 }
221 }
6083fe74 222
bd00cd5f
MR
223 /*
224 * Register the kernel text, kernel data, initrd, and initial
225 * pagetables with memblock.
226 */
c1cc1552
CM
227 memblock_reserve(__pa(_text), _end - _text);
228#ifdef CONFIG_BLK_DEV_INITRD
a89dea58
AB
229 if (initrd_start) {
230 memblock_reserve(initrd_start, initrd_end - initrd_start);
231
232 /* the generic initrd code expects virtual addresses */
233 initrd_start = __phys_to_virt(initrd_start);
234 initrd_end = __phys_to_virt(initrd_end);
235 }
c1cc1552
CM
236#endif
237
0ceac9e0 238 early_init_fdt_scan_reserved_mem();
2d5a5612
CM
239
240 /* 4GB maximum for 32-bit only capable devices */
241 if (IS_ENABLED(CONFIG_ZONE_DMA))
a1e50a82
CM
242 arm64_dma_phys_limit = max_zone_dma_phys();
243 else
244 arm64_dma_phys_limit = PHYS_MASK + 1;
245 dma_contiguous_reserve(arm64_dma_phys_limit);
6ac2104d 246
c1cc1552
CM
247 memblock_allow_resize();
248 memblock_dump_all();
249}
250
251void __init bootmem_init(void)
252{
253 unsigned long min, max;
254
255 min = PFN_UP(memblock_start_of_DRAM());
256 max = PFN_DOWN(memblock_end_of_DRAM());
257
36dd9086
VM
258 early_memtest(min << PAGE_SHIFT, max << PAGE_SHIFT);
259
c1cc1552
CM
260 /*
261 * Sparsemem tries to allocate bootmem in memory_present(), so must be
262 * done after the fixed reservations.
263 */
264 arm64_memory_present();
265
266 sparse_init();
267 zone_sizes_init(min, max);
268
269 high_memory = __va((max << PAGE_SHIFT) - 1) + 1;
270 max_pfn = max_low_pfn = max;
271}
272
c1cc1552
CM
273#ifndef CONFIG_SPARSEMEM_VMEMMAP
274static inline void free_memmap(unsigned long start_pfn, unsigned long end_pfn)
275{
276 struct page *start_pg, *end_pg;
277 unsigned long pg, pgend;
278
279 /*
280 * Convert start_pfn/end_pfn to a struct page pointer.
281 */
282 start_pg = pfn_to_page(start_pfn - 1) + 1;
283 end_pg = pfn_to_page(end_pfn - 1) + 1;
284
285 /*
286 * Convert to physical addresses, and round start upwards and end
287 * downwards.
288 */
289 pg = (unsigned long)PAGE_ALIGN(__pa(start_pg));
290 pgend = (unsigned long)__pa(end_pg) & PAGE_MASK;
291
292 /*
293 * If there are free pages between these, free the section of the
294 * memmap array.
295 */
296 if (pg < pgend)
297 free_bootmem(pg, pgend - pg);
298}
299
300/*
301 * The mem_map array can get very big. Free the unused area of the memory map.
302 */
303static void __init free_unused_memmap(void)
304{
305 unsigned long start, prev_end = 0;
306 struct memblock_region *reg;
307
308 for_each_memblock(memory, reg) {
309 start = __phys_to_pfn(reg->base);
310
311#ifdef CONFIG_SPARSEMEM
312 /*
313 * Take care not to free memmap entries that don't exist due
314 * to SPARSEMEM sections which aren't present.
315 */
316 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
317#endif
318 /*
319 * If we had a previous bank, and there is a space between the
320 * current bank and the previous, free it.
321 */
322 if (prev_end && prev_end < start)
323 free_memmap(prev_end, start);
324
325 /*
326 * Align up here since the VM subsystem insists that the
327 * memmap entries are valid from the bank end aligned to
328 * MAX_ORDER_NR_PAGES.
329 */
b9bcc919 330 prev_end = ALIGN(__phys_to_pfn(reg->base + reg->size),
c1cc1552
CM
331 MAX_ORDER_NR_PAGES);
332 }
333
334#ifdef CONFIG_SPARSEMEM
335 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
336 free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
337#endif
338}
339#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
340
341/*
342 * mem_init() marks the free areas in the mem_map and tells us how much memory
343 * is free. This is done after various parts of the system have claimed their
344 * memory after the kernel image.
345 */
346void __init mem_init(void)
347{
a1e50a82
CM
348 swiotlb_init(1);
349
a6583c7c 350 set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
c1cc1552
CM
351
352#ifndef CONFIG_SPARSEMEM_VMEMMAP
c1cc1552
CM
353 free_unused_memmap();
354#endif
bee4ebd1 355 /* this will put all unused low memory onto the freelists */
0c988534 356 free_all_bootmem();
c1cc1552 357
6879ea83 358 mem_init_print_info(NULL);
c1cc1552
CM
359
360#define MLK(b, t) b, t, ((t) - (b)) >> 10
361#define MLM(b, t) b, t, ((t) - (b)) >> 20
08375198 362#define MLG(b, t) b, t, ((t) - (b)) >> 30
c1cc1552
CM
363#define MLK_ROUNDUP(b, t) b, t, DIV_ROUND_UP(((t) - (b)), SZ_1K)
364
f09f1bac 365 pr_notice("Virtual kernel memory layout:\n");
ee7f881b 366#ifdef CONFIG_KASAN
f09f1bac
CM
367 pr_cont(" kasan : 0x%16lx - 0x%16lx (%6ld GB)\n",
368 MLG(KASAN_SHADOW_START, KASAN_SHADOW_END));
ee7f881b 369#endif
f09f1bac
CM
370 pr_cont(" modules : 0x%16lx - 0x%16lx (%6ld MB)\n",
371 MLM(MODULES_VADDR, MODULES_END));
372 pr_cont(" vmalloc : 0x%16lx - 0x%16lx (%6ld GB)\n",
373 MLG(VMALLOC_START, VMALLOC_END));
374 pr_cont(" .text : 0x%p" " - 0x%p" " (%6ld KB)\n"
375 " .rodata : 0x%p" " - 0x%p" " (%6ld KB)\n"
376 " .init : 0x%p" " - 0x%p" " (%6ld KB)\n"
377 " .data : 0x%p" " - 0x%p" " (%6ld KB)\n",
378 MLK_ROUNDUP(_text, __start_rodata),
379 MLK_ROUNDUP(__start_rodata, _etext),
380 MLK_ROUNDUP(__init_begin, __init_end),
381 MLK_ROUNDUP(_sdata, _edata));
c1cc1552 382#ifdef CONFIG_SPARSEMEM_VMEMMAP
f09f1bac
CM
383 pr_cont(" vmemmap : 0x%16lx - 0x%16lx (%6ld GB maximum)\n"
384 " 0x%16lx - 0x%16lx (%6ld MB actual)\n",
9d854607
LT
385 MLG(VMEMMAP_START,
386 VMEMMAP_START + VMEMMAP_SIZE),
f09f1bac
CM
387 MLM((unsigned long)phys_to_page(memblock_start_of_DRAM()),
388 (unsigned long)virt_to_page(high_memory)));
c1cc1552 389#endif
f09f1bac
CM
390 pr_cont(" fixed : 0x%16lx - 0x%16lx (%6ld KB)\n",
391 MLK(FIXADDR_START, FIXADDR_TOP));
392 pr_cont(" PCI I/O : 0x%16lx - 0x%16lx (%6ld MB)\n",
393 MLM(PCI_IO_START, PCI_IO_END));
394 pr_cont(" memory : 0x%16lx - 0x%16lx (%6ld MB)\n",
395 MLM(__phys_to_virt(memblock_start_of_DRAM()),
396 (unsigned long)high_memory));
c1cc1552
CM
397
398#undef MLK
399#undef MLM
400#undef MLK_ROUNDUP
401
402 /*
403 * Check boundaries twice: Some fundamental inconsistencies can be
404 * detected at build time already.
405 */
406#ifdef CONFIG_COMPAT
407 BUILD_BUG_ON(TASK_SIZE_32 > TASK_SIZE_64);
408#endif
c1cc1552 409
bee4ebd1 410 if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) {
c1cc1552
CM
411 extern int sysctl_overcommit_memory;
412 /*
413 * On a machine this small we won't get anywhere without
414 * overcommit, so turn it on by default.
415 */
416 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
417 }
418}
419
420void free_initmem(void)
421{
9af5b807 422 free_initmem_default(0);
f9040773 423 fixup_init();
c1cc1552
CM
424}
425
426#ifdef CONFIG_BLK_DEV_INITRD
427
662ba3db 428static int keep_initrd __initdata;
c1cc1552 429
662ba3db 430void __init free_initrd_mem(unsigned long start, unsigned long end)
c1cc1552 431{
0145058c 432 if (!keep_initrd)
9af5b807 433 free_reserved_area((void *)start, (void *)end, 0, "initrd");
c1cc1552
CM
434}
435
436static int __init keepinitrd_setup(char *__unused)
437{
438 keep_initrd = 1;
439 return 1;
440}
441
442__setup("keepinitrd", keepinitrd_setup);
443#endif
a7f8de16
AB
444
445/*
446 * Dump out memory limit information on panic.
447 */
448static int dump_mem_limit(struct notifier_block *self, unsigned long v, void *p)
449{
450 if (memory_limit != (phys_addr_t)ULLONG_MAX) {
451 pr_emerg("Memory Limit: %llu MB\n", memory_limit >> 20);
452 } else {
453 pr_emerg("Memory Limit: none\n");
454 }
455 return 0;
456}
457
458static struct notifier_block mem_limit_notifier = {
459 .notifier_call = dump_mem_limit,
460};
461
462static int __init register_mem_limit_dumper(void)
463{
464 atomic_notifier_chain_register(&panic_notifier_list,
465 &mem_limit_notifier);
466 return 0;
467}
468__initcall(register_mem_limit_dumper);