Commit | Line | Data |
---|---|---|
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> | |
5a9e3e15 | 26 | #include <linux/cache.h> |
c1cc1552 CM |
27 | #include <linux/mman.h> |
28 | #include <linux/nodemask.h> | |
29 | #include <linux/initrd.h> | |
30 | #include <linux/gfp.h> | |
31 | #include <linux/memblock.h> | |
32 | #include <linux/sort.h> | |
764b51ea | 33 | #include <linux/of.h> |
c1cc1552 | 34 | #include <linux/of_fdt.h> |
19e7640d | 35 | #include <linux/dma-mapping.h> |
6ac2104d | 36 | #include <linux/dma-contiguous.h> |
86c8b27a | 37 | #include <linux/efi.h> |
a1e50a82 | 38 | #include <linux/swiotlb.h> |
dae8c235 | 39 | #include <linux/vmalloc.h> |
2077be67 | 40 | #include <linux/mm.h> |
764b51ea | 41 | #include <linux/kexec.h> |
e62aaeac | 42 | #include <linux/crash_dump.h> |
c1cc1552 | 43 | |
a7f8de16 | 44 | #include <asm/boot.h> |
08375198 | 45 | #include <asm/fixmap.h> |
f9040773 | 46 | #include <asm/kasan.h> |
a7f8de16 | 47 | #include <asm/kernel-pgtable.h> |
aa03c428 | 48 | #include <asm/memory.h> |
1a2db300 | 49 | #include <asm/numa.h> |
c1cc1552 CM |
50 | #include <asm/sections.h> |
51 | #include <asm/setup.h> | |
52 | #include <asm/sizes.h> | |
53 | #include <asm/tlb.h> | |
e039ee4e | 54 | #include <asm/alternative.h> |
c1cc1552 | 55 | |
a7f8de16 AB |
56 | /* |
57 | * We need to be able to catch inadvertent references to memstart_addr | |
58 | * that occur (potentially in generic code) before arm64_memblock_init() | |
59 | * executes, which assigns it its actual value. So use a default value | |
60 | * that cannot be mistaken for a real physical address. | |
61 | */ | |
5a9e3e15 JZ |
62 | s64 memstart_addr __ro_after_init = -1; |
63 | phys_addr_t arm64_dma_phys_limit __ro_after_init; | |
c1cc1552 | 64 | |
ec2eaa73 | 65 | #ifdef CONFIG_BLK_DEV_INITRD |
c1cc1552 CM |
66 | static int __init early_initrd(char *p) |
67 | { | |
68 | unsigned long start, size; | |
69 | char *endp; | |
70 | ||
71 | start = memparse(p, &endp); | |
72 | if (*endp == ',') { | |
73 | size = memparse(endp + 1, NULL); | |
74 | ||
a89dea58 AB |
75 | initrd_start = start; |
76 | initrd_end = start + size; | |
c1cc1552 CM |
77 | } |
78 | return 0; | |
79 | } | |
80 | early_param("initrd", early_initrd); | |
ec2eaa73 | 81 | #endif |
c1cc1552 | 82 | |
764b51ea AT |
83 | #ifdef CONFIG_KEXEC_CORE |
84 | /* | |
85 | * reserve_crashkernel() - reserves memory for crash kernel | |
86 | * | |
87 | * This function reserves memory area given in "crashkernel=" kernel command | |
88 | * line parameter. The memory reserved is used by dump capture kernel when | |
89 | * primary kernel is crashing. | |
90 | */ | |
91 | static void __init reserve_crashkernel(void) | |
92 | { | |
93 | unsigned long long crash_base, crash_size; | |
94 | int ret; | |
95 | ||
96 | ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(), | |
97 | &crash_size, &crash_base); | |
98 | /* no crashkernel= or invalid value specified */ | |
99 | if (ret || !crash_size) | |
100 | return; | |
101 | ||
102 | crash_size = PAGE_ALIGN(crash_size); | |
103 | ||
104 | if (crash_base == 0) { | |
105 | /* Current arm64 boot protocol requires 2MB alignment */ | |
106 | crash_base = memblock_find_in_range(0, ARCH_LOW_ADDRESS_LIMIT, | |
107 | crash_size, SZ_2M); | |
108 | if (crash_base == 0) { | |
109 | pr_warn("cannot allocate crashkernel (size:0x%llx)\n", | |
110 | crash_size); | |
111 | return; | |
112 | } | |
113 | } else { | |
114 | /* User specifies base address explicitly. */ | |
115 | if (!memblock_is_region_memory(crash_base, crash_size)) { | |
116 | pr_warn("cannot reserve crashkernel: region is not memory\n"); | |
117 | return; | |
118 | } | |
119 | ||
120 | if (memblock_is_region_reserved(crash_base, crash_size)) { | |
121 | pr_warn("cannot reserve crashkernel: region overlaps reserved memory\n"); | |
122 | return; | |
123 | } | |
124 | ||
125 | if (!IS_ALIGNED(crash_base, SZ_2M)) { | |
126 | pr_warn("cannot reserve crashkernel: base address is not 2MB aligned\n"); | |
127 | return; | |
128 | } | |
129 | } | |
130 | memblock_reserve(crash_base, crash_size); | |
131 | ||
132 | pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n", | |
133 | crash_base, crash_base + crash_size, crash_size >> 20); | |
134 | ||
135 | crashk_res.start = crash_base; | |
136 | crashk_res.end = crash_base + crash_size - 1; | |
137 | } | |
254a41c0 AT |
138 | |
139 | static void __init kexec_reserve_crashkres_pages(void) | |
140 | { | |
141 | #ifdef CONFIG_HIBERNATION | |
142 | phys_addr_t addr; | |
143 | struct page *page; | |
144 | ||
145 | if (!crashk_res.end) | |
146 | return; | |
147 | ||
148 | /* | |
149 | * To reduce the size of hibernation image, all the pages are | |
150 | * marked as Reserved initially. | |
151 | */ | |
152 | for (addr = crashk_res.start; addr < (crashk_res.end + 1); | |
153 | addr += PAGE_SIZE) { | |
154 | page = phys_to_page(addr); | |
155 | SetPageReserved(page); | |
156 | } | |
157 | #endif | |
158 | } | |
764b51ea AT |
159 | #else |
160 | static void __init reserve_crashkernel(void) | |
161 | { | |
162 | } | |
254a41c0 AT |
163 | |
164 | static void __init kexec_reserve_crashkres_pages(void) | |
165 | { | |
166 | } | |
764b51ea AT |
167 | #endif /* CONFIG_KEXEC_CORE */ |
168 | ||
e62aaeac AT |
169 | #ifdef CONFIG_CRASH_DUMP |
170 | static int __init early_init_dt_scan_elfcorehdr(unsigned long node, | |
171 | const char *uname, int depth, void *data) | |
172 | { | |
173 | const __be32 *reg; | |
174 | int len; | |
175 | ||
176 | if (depth != 1 || strcmp(uname, "chosen") != 0) | |
177 | return 0; | |
178 | ||
179 | reg = of_get_flat_dt_prop(node, "linux,elfcorehdr", &len); | |
180 | if (!reg || (len < (dt_root_addr_cells + dt_root_size_cells))) | |
181 | return 1; | |
182 | ||
183 | elfcorehdr_addr = dt_mem_next_cell(dt_root_addr_cells, ®); | |
184 | elfcorehdr_size = dt_mem_next_cell(dt_root_size_cells, ®); | |
185 | ||
186 | return 1; | |
187 | } | |
188 | ||
189 | /* | |
190 | * reserve_elfcorehdr() - reserves memory for elf core header | |
191 | * | |
192 | * This function reserves the memory occupied by an elf core header | |
193 | * described in the device tree. This region contains all the | |
194 | * information about primary kernel's core image and is used by a dump | |
195 | * capture kernel to access the system memory on primary kernel. | |
196 | */ | |
197 | static void __init reserve_elfcorehdr(void) | |
198 | { | |
199 | of_scan_flat_dt(early_init_dt_scan_elfcorehdr, NULL); | |
200 | ||
201 | if (!elfcorehdr_size) | |
202 | return; | |
203 | ||
204 | if (memblock_is_region_reserved(elfcorehdr_addr, elfcorehdr_size)) { | |
205 | pr_warn("elfcorehdr is overlapped\n"); | |
206 | return; | |
207 | } | |
208 | ||
209 | memblock_reserve(elfcorehdr_addr, elfcorehdr_size); | |
210 | ||
211 | pr_info("Reserving %lldKB of memory at 0x%llx for elfcorehdr\n", | |
212 | elfcorehdr_size >> 10, elfcorehdr_addr); | |
213 | } | |
214 | #else | |
215 | static void __init reserve_elfcorehdr(void) | |
216 | { | |
217 | } | |
218 | #endif /* CONFIG_CRASH_DUMP */ | |
d50314a6 | 219 | /* |
ad67f5a6 | 220 | * Return the maximum physical address for ZONE_DMA32 (DMA_BIT_MASK(32)). It |
d50314a6 CM |
221 | * currently assumes that for memory starting above 4G, 32-bit devices will |
222 | * use a DMA offset. | |
223 | */ | |
a7c61a34 | 224 | static phys_addr_t __init max_zone_dma_phys(void) |
d50314a6 CM |
225 | { |
226 | phys_addr_t offset = memblock_start_of_DRAM() & GENMASK_ULL(63, 32); | |
227 | return min(offset + (1ULL << 32), memblock_end_of_DRAM()); | |
228 | } | |
229 | ||
1a2db300 GK |
230 | #ifdef CONFIG_NUMA |
231 | ||
232 | static void __init zone_sizes_init(unsigned long min, unsigned long max) | |
233 | { | |
234 | unsigned long max_zone_pfns[MAX_NR_ZONES] = {0}; | |
235 | ||
ad67f5a6 CH |
236 | if (IS_ENABLED(CONFIG_ZONE_DMA32)) |
237 | max_zone_pfns[ZONE_DMA32] = PFN_DOWN(max_zone_dma_phys()); | |
1a2db300 GK |
238 | max_zone_pfns[ZONE_NORMAL] = max; |
239 | ||
240 | free_area_init_nodes(max_zone_pfns); | |
241 | } | |
242 | ||
243 | #else | |
244 | ||
c1cc1552 CM |
245 | static void __init zone_sizes_init(unsigned long min, unsigned long max) |
246 | { | |
247 | struct memblock_region *reg; | |
248 | unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES]; | |
19e7640d | 249 | unsigned long max_dma = min; |
c1cc1552 CM |
250 | |
251 | memset(zone_size, 0, sizeof(zone_size)); | |
252 | ||
c1cc1552 | 253 | /* 4GB maximum for 32-bit only capable devices */ |
ad67f5a6 | 254 | #ifdef CONFIG_ZONE_DMA32 |
86a5906e | 255 | max_dma = PFN_DOWN(arm64_dma_phys_limit); |
ad67f5a6 | 256 | zone_size[ZONE_DMA32] = max_dma - min; |
86a5906e | 257 | #endif |
19e7640d | 258 | zone_size[ZONE_NORMAL] = max - max_dma; |
c1cc1552 CM |
259 | |
260 | memcpy(zhole_size, zone_size, sizeof(zhole_size)); | |
261 | ||
262 | for_each_memblock(memory, reg) { | |
263 | unsigned long start = memblock_region_memory_base_pfn(reg); | |
264 | unsigned long end = memblock_region_memory_end_pfn(reg); | |
265 | ||
266 | if (start >= max) | |
267 | continue; | |
19e7640d | 268 | |
ad67f5a6 | 269 | #ifdef CONFIG_ZONE_DMA32 |
86a5906e | 270 | if (start < max_dma) { |
19e7640d | 271 | unsigned long dma_end = min(end, max_dma); |
ad67f5a6 | 272 | zhole_size[ZONE_DMA32] -= dma_end - start; |
c1cc1552 | 273 | } |
86a5906e | 274 | #endif |
19e7640d | 275 | if (end > max_dma) { |
c1cc1552 | 276 | unsigned long normal_end = min(end, max); |
19e7640d | 277 | unsigned long normal_start = max(start, max_dma); |
c1cc1552 CM |
278 | zhole_size[ZONE_NORMAL] -= normal_end - normal_start; |
279 | } | |
280 | } | |
281 | ||
282 | free_area_init_node(0, zone_size, min, zhole_size); | |
283 | } | |
284 | ||
1a2db300 GK |
285 | #endif /* CONFIG_NUMA */ |
286 | ||
c1cc1552 CM |
287 | int pfn_valid(unsigned long pfn) |
288 | { | |
5ad356ea GH |
289 | phys_addr_t addr = pfn << PAGE_SHIFT; |
290 | ||
291 | if ((addr >> PAGE_SHIFT) != pfn) | |
292 | return 0; | |
293 | return memblock_is_map_memory(addr); | |
c1cc1552 CM |
294 | } |
295 | EXPORT_SYMBOL(pfn_valid); | |
c1cc1552 CM |
296 | |
297 | #ifndef CONFIG_SPARSEMEM | |
a7c61a34 | 298 | static void __init arm64_memory_present(void) |
c1cc1552 CM |
299 | { |
300 | } | |
301 | #else | |
a7c61a34 | 302 | static void __init arm64_memory_present(void) |
c1cc1552 CM |
303 | { |
304 | struct memblock_region *reg; | |
305 | ||
1a2db300 | 306 | for_each_memblock(memory, reg) { |
ea2cbee3 MR |
307 | int nid = memblock_get_region_node(reg); |
308 | ||
1a2db300 GK |
309 | memory_present(nid, memblock_region_memory_base_pfn(reg), |
310 | memblock_region_memory_end_pfn(reg)); | |
311 | } | |
c1cc1552 CM |
312 | } |
313 | #endif | |
314 | ||
d7dc899a | 315 | static phys_addr_t memory_limit = PHYS_ADDR_MAX; |
6083fe74 MR |
316 | |
317 | /* | |
318 | * Limit the memory size that was specified via FDT. | |
319 | */ | |
320 | static int __init early_mem(char *p) | |
321 | { | |
322 | if (!p) | |
323 | return 1; | |
324 | ||
325 | memory_limit = memparse(p, &p) & PAGE_MASK; | |
326 | pr_notice("Memory limited to %lldMB\n", memory_limit >> 20); | |
327 | ||
328 | return 0; | |
329 | } | |
330 | early_param("mem", early_mem); | |
331 | ||
8f579b1c AT |
332 | static int __init early_init_dt_scan_usablemem(unsigned long node, |
333 | const char *uname, int depth, void *data) | |
334 | { | |
335 | struct memblock_region *usablemem = data; | |
336 | const __be32 *reg; | |
337 | int len; | |
338 | ||
339 | if (depth != 1 || strcmp(uname, "chosen") != 0) | |
340 | return 0; | |
341 | ||
342 | reg = of_get_flat_dt_prop(node, "linux,usable-memory-range", &len); | |
343 | if (!reg || (len < (dt_root_addr_cells + dt_root_size_cells))) | |
344 | return 1; | |
345 | ||
346 | usablemem->base = dt_mem_next_cell(dt_root_addr_cells, ®); | |
347 | usablemem->size = dt_mem_next_cell(dt_root_size_cells, ®); | |
348 | ||
349 | return 1; | |
350 | } | |
351 | ||
352 | static void __init fdt_enforce_memory_region(void) | |
353 | { | |
354 | struct memblock_region reg = { | |
355 | .size = 0, | |
356 | }; | |
357 | ||
358 | of_scan_flat_dt(early_init_dt_scan_usablemem, ®); | |
359 | ||
360 | if (reg.size) | |
361 | memblock_cap_memory_range(reg.base, reg.size); | |
362 | } | |
363 | ||
c1cc1552 CM |
364 | void __init arm64_memblock_init(void) |
365 | { | |
a7f8de16 AB |
366 | const s64 linear_region_size = -(s64)PAGE_OFFSET; |
367 | ||
8f579b1c AT |
368 | /* Handle linux,usable-memory-range property */ |
369 | fdt_enforce_memory_region(); | |
370 | ||
e9eaa805 KM |
371 | /* Remove memory above our supported physical address size */ |
372 | memblock_remove(1ULL << PHYS_MASK_SHIFT, ULLONG_MAX); | |
373 | ||
6d2aa549 AB |
374 | /* |
375 | * Ensure that the linear region takes up exactly half of the kernel | |
376 | * virtual address space. This way, we can distinguish a linear address | |
377 | * from a kernel/module/vmalloc address by testing a single bit. | |
378 | */ | |
379 | BUILD_BUG_ON(linear_region_size != BIT(VA_BITS - 1)); | |
380 | ||
a7f8de16 AB |
381 | /* |
382 | * Select a suitable value for the base of physical memory. | |
383 | */ | |
384 | memstart_addr = round_down(memblock_start_of_DRAM(), | |
385 | ARM64_MEMSTART_ALIGN); | |
386 | ||
387 | /* | |
388 | * Remove the memory that we will not be able to cover with the | |
389 | * linear mapping. Take care not to clip the kernel which may be | |
390 | * high in memory. | |
391 | */ | |
2077be67 LA |
392 | memblock_remove(max_t(u64, memstart_addr + linear_region_size, |
393 | __pa_symbol(_end)), ULLONG_MAX); | |
2958987f AB |
394 | if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) { |
395 | /* ensure that memstart_addr remains sufficiently aligned */ | |
396 | memstart_addr = round_up(memblock_end_of_DRAM() - linear_region_size, | |
397 | ARM64_MEMSTART_ALIGN); | |
398 | memblock_remove(0, memstart_addr); | |
399 | } | |
a7f8de16 AB |
400 | |
401 | /* | |
402 | * Apply the memory limit if it was set. Since the kernel may be loaded | |
403 | * high up in memory, add back the kernel region that must be accessible | |
404 | * via the linear mapping. | |
405 | */ | |
d7dc899a | 406 | if (memory_limit != PHYS_ADDR_MAX) { |
cb0a6502 | 407 | memblock_mem_limit_remove_map(memory_limit); |
2077be67 | 408 | memblock_add(__pa_symbol(_text), (u64)(_end - _text)); |
a7f8de16 | 409 | } |
6083fe74 | 410 | |
177e15f0 AB |
411 | if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && initrd_start) { |
412 | /* | |
413 | * Add back the memory we just removed if it results in the | |
414 | * initrd to become inaccessible via the linear mapping. | |
415 | * Otherwise, this is a no-op | |
416 | */ | |
417 | u64 base = initrd_start & PAGE_MASK; | |
418 | u64 size = PAGE_ALIGN(initrd_end) - base; | |
419 | ||
420 | /* | |
421 | * We can only add back the initrd memory if we don't end up | |
422 | * with more memory than we can address via the linear mapping. | |
423 | * It is up to the bootloader to position the kernel and the | |
424 | * initrd reasonably close to each other (i.e., within 32 GB of | |
425 | * each other) so that all granule/#levels combinations can | |
426 | * always access both. | |
427 | */ | |
428 | if (WARN(base < memblock_start_of_DRAM() || | |
429 | base + size > memblock_start_of_DRAM() + | |
430 | linear_region_size, | |
431 | "initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) { | |
432 | initrd_start = 0; | |
433 | } else { | |
434 | memblock_remove(base, size); /* clear MEMBLOCK_ flags */ | |
435 | memblock_add(base, size); | |
436 | memblock_reserve(base, size); | |
437 | } | |
438 | } | |
439 | ||
c031a421 AB |
440 | if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) { |
441 | extern u16 memstart_offset_seed; | |
442 | u64 range = linear_region_size - | |
443 | (memblock_end_of_DRAM() - memblock_start_of_DRAM()); | |
444 | ||
445 | /* | |
446 | * If the size of the linear region exceeds, by a sufficient | |
447 | * margin, the size of the region that the available physical | |
448 | * memory spans, randomize the linear region as well. | |
449 | */ | |
450 | if (memstart_offset_seed > 0 && range >= ARM64_MEMSTART_ALIGN) { | |
451 | range = range / ARM64_MEMSTART_ALIGN + 1; | |
452 | memstart_addr -= ARM64_MEMSTART_ALIGN * | |
453 | ((range * memstart_offset_seed) >> 16); | |
454 | } | |
455 | } | |
6083fe74 | 456 | |
bd00cd5f MR |
457 | /* |
458 | * Register the kernel text, kernel data, initrd, and initial | |
459 | * pagetables with memblock. | |
460 | */ | |
2077be67 | 461 | memblock_reserve(__pa_symbol(_text), _end - _text); |
c1cc1552 | 462 | #ifdef CONFIG_BLK_DEV_INITRD |
a89dea58 AB |
463 | if (initrd_start) { |
464 | memblock_reserve(initrd_start, initrd_end - initrd_start); | |
465 | ||
466 | /* the generic initrd code expects virtual addresses */ | |
467 | initrd_start = __phys_to_virt(initrd_start); | |
468 | initrd_end = __phys_to_virt(initrd_end); | |
469 | } | |
c1cc1552 CM |
470 | #endif |
471 | ||
0ceac9e0 | 472 | early_init_fdt_scan_reserved_mem(); |
2d5a5612 CM |
473 | |
474 | /* 4GB maximum for 32-bit only capable devices */ | |
ad67f5a6 | 475 | if (IS_ENABLED(CONFIG_ZONE_DMA32)) |
a1e50a82 CM |
476 | arm64_dma_phys_limit = max_zone_dma_phys(); |
477 | else | |
478 | arm64_dma_phys_limit = PHYS_MASK + 1; | |
764b51ea AT |
479 | |
480 | reserve_crashkernel(); | |
481 | ||
e62aaeac AT |
482 | reserve_elfcorehdr(); |
483 | ||
f24e5834 SC |
484 | high_memory = __va(memblock_end_of_DRAM() - 1) + 1; |
485 | ||
a1e50a82 | 486 | dma_contiguous_reserve(arm64_dma_phys_limit); |
6ac2104d | 487 | |
c1cc1552 | 488 | memblock_allow_resize(); |
c1cc1552 CM |
489 | } |
490 | ||
491 | void __init bootmem_init(void) | |
492 | { | |
493 | unsigned long min, max; | |
494 | ||
495 | min = PFN_UP(memblock_start_of_DRAM()); | |
496 | max = PFN_DOWN(memblock_end_of_DRAM()); | |
497 | ||
36dd9086 VM |
498 | early_memtest(min << PAGE_SHIFT, max << PAGE_SHIFT); |
499 | ||
1a2db300 GK |
500 | max_pfn = max_low_pfn = max; |
501 | ||
502 | arm64_numa_init(); | |
c1cc1552 CM |
503 | /* |
504 | * Sparsemem tries to allocate bootmem in memory_present(), so must be | |
505 | * done after the fixed reservations. | |
506 | */ | |
507 | arm64_memory_present(); | |
508 | ||
509 | sparse_init(); | |
510 | zone_sizes_init(min, max); | |
511 | ||
1a2db300 | 512 | memblock_dump_all(); |
c1cc1552 CM |
513 | } |
514 | ||
c1cc1552 CM |
515 | #ifndef CONFIG_SPARSEMEM_VMEMMAP |
516 | static inline void free_memmap(unsigned long start_pfn, unsigned long end_pfn) | |
517 | { | |
518 | struct page *start_pg, *end_pg; | |
519 | unsigned long pg, pgend; | |
520 | ||
521 | /* | |
522 | * Convert start_pfn/end_pfn to a struct page pointer. | |
523 | */ | |
524 | start_pg = pfn_to_page(start_pfn - 1) + 1; | |
525 | end_pg = pfn_to_page(end_pfn - 1) + 1; | |
526 | ||
527 | /* | |
528 | * Convert to physical addresses, and round start upwards and end | |
529 | * downwards. | |
530 | */ | |
531 | pg = (unsigned long)PAGE_ALIGN(__pa(start_pg)); | |
532 | pgend = (unsigned long)__pa(end_pg) & PAGE_MASK; | |
533 | ||
534 | /* | |
535 | * If there are free pages between these, free the section of the | |
536 | * memmap array. | |
537 | */ | |
538 | if (pg < pgend) | |
2013288f | 539 | memblock_free(pg, pgend - pg); |
c1cc1552 CM |
540 | } |
541 | ||
542 | /* | |
543 | * The mem_map array can get very big. Free the unused area of the memory map. | |
544 | */ | |
545 | static void __init free_unused_memmap(void) | |
546 | { | |
547 | unsigned long start, prev_end = 0; | |
548 | struct memblock_region *reg; | |
549 | ||
550 | for_each_memblock(memory, reg) { | |
551 | start = __phys_to_pfn(reg->base); | |
552 | ||
553 | #ifdef CONFIG_SPARSEMEM | |
554 | /* | |
555 | * Take care not to free memmap entries that don't exist due | |
556 | * to SPARSEMEM sections which aren't present. | |
557 | */ | |
558 | start = min(start, ALIGN(prev_end, PAGES_PER_SECTION)); | |
559 | #endif | |
560 | /* | |
561 | * If we had a previous bank, and there is a space between the | |
562 | * current bank and the previous, free it. | |
563 | */ | |
564 | if (prev_end && prev_end < start) | |
565 | free_memmap(prev_end, start); | |
566 | ||
567 | /* | |
568 | * Align up here since the VM subsystem insists that the | |
569 | * memmap entries are valid from the bank end aligned to | |
570 | * MAX_ORDER_NR_PAGES. | |
571 | */ | |
b9bcc919 | 572 | prev_end = ALIGN(__phys_to_pfn(reg->base + reg->size), |
c1cc1552 CM |
573 | MAX_ORDER_NR_PAGES); |
574 | } | |
575 | ||
576 | #ifdef CONFIG_SPARSEMEM | |
577 | if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) | |
578 | free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION)); | |
579 | #endif | |
580 | } | |
581 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ | |
582 | ||
583 | /* | |
584 | * mem_init() marks the free areas in the mem_map and tells us how much memory | |
585 | * is free. This is done after various parts of the system have claimed their | |
586 | * memory after the kernel image. | |
587 | */ | |
588 | void __init mem_init(void) | |
589 | { | |
ae7871be GU |
590 | if (swiotlb_force == SWIOTLB_FORCE || |
591 | max_pfn > (arm64_dma_phys_limit >> PAGE_SHIFT)) | |
b67a8b29 | 592 | swiotlb_init(1); |
524dabe1 AG |
593 | else |
594 | swiotlb_force = SWIOTLB_NO_FORCE; | |
a1e50a82 | 595 | |
a6583c7c | 596 | set_max_mapnr(pfn_to_page(max_pfn) - mem_map); |
c1cc1552 CM |
597 | |
598 | #ifndef CONFIG_SPARSEMEM_VMEMMAP | |
c1cc1552 CM |
599 | free_unused_memmap(); |
600 | #endif | |
bee4ebd1 | 601 | /* this will put all unused low memory onto the freelists */ |
c6ffc5ca | 602 | memblock_free_all(); |
c1cc1552 | 603 | |
254a41c0 AT |
604 | kexec_reserve_crashkres_pages(); |
605 | ||
6879ea83 | 606 | mem_init_print_info(NULL); |
c1cc1552 | 607 | |
c1cc1552 CM |
608 | /* |
609 | * Check boundaries twice: Some fundamental inconsistencies can be | |
610 | * detected at build time already. | |
611 | */ | |
612 | #ifdef CONFIG_COMPAT | |
613 | BUILD_BUG_ON(TASK_SIZE_32 > TASK_SIZE_64); | |
614 | #endif | |
c1cc1552 | 615 | |
7b0eb6b4 | 616 | #ifdef CONFIG_SPARSEMEM_VMEMMAP |
3e1907d5 AB |
617 | /* |
618 | * Make sure we chose the upper bound of sizeof(struct page) | |
7b0eb6b4 | 619 | * correctly when sizing the VMEMMAP array. |
3e1907d5 AB |
620 | */ |
621 | BUILD_BUG_ON(sizeof(struct page) > (1 << STRUCT_PAGE_MAX_SHIFT)); | |
7b0eb6b4 | 622 | #endif |
3e1907d5 | 623 | |
bee4ebd1 | 624 | if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) { |
c1cc1552 CM |
625 | extern int sysctl_overcommit_memory; |
626 | /* | |
627 | * On a machine this small we won't get anywhere without | |
628 | * overcommit, so turn it on by default. | |
629 | */ | |
630 | sysctl_overcommit_memory = OVERCOMMIT_ALWAYS; | |
631 | } | |
632 | } | |
633 | ||
634 | void free_initmem(void) | |
635 | { | |
2077be67 LA |
636 | free_reserved_area(lm_alias(__init_begin), |
637 | lm_alias(__init_end), | |
d386825c | 638 | 0, "unused kernel"); |
dae8c235 KW |
639 | /* |
640 | * Unmap the __init region but leave the VM area in place. This | |
641 | * prevents the region from being reused for kernel modules, which | |
642 | * is not supported by kallsyms. | |
643 | */ | |
644 | unmap_kernel_range((u64)__init_begin, (u64)(__init_end - __init_begin)); | |
c1cc1552 CM |
645 | } |
646 | ||
647 | #ifdef CONFIG_BLK_DEV_INITRD | |
648 | ||
662ba3db | 649 | static int keep_initrd __initdata; |
c1cc1552 | 650 | |
662ba3db | 651 | void __init free_initrd_mem(unsigned long start, unsigned long end) |
c1cc1552 | 652 | { |
05c58752 | 653 | if (!keep_initrd) { |
9af5b807 | 654 | free_reserved_area((void *)start, (void *)end, 0, "initrd"); |
05c58752 CV |
655 | memblock_free(__virt_to_phys(start), end - start); |
656 | } | |
c1cc1552 CM |
657 | } |
658 | ||
659 | static int __init keepinitrd_setup(char *__unused) | |
660 | { | |
661 | keep_initrd = 1; | |
662 | return 1; | |
663 | } | |
664 | ||
665 | __setup("keepinitrd", keepinitrd_setup); | |
666 | #endif | |
a7f8de16 AB |
667 | |
668 | /* | |
669 | * Dump out memory limit information on panic. | |
670 | */ | |
671 | static int dump_mem_limit(struct notifier_block *self, unsigned long v, void *p) | |
672 | { | |
d7dc899a | 673 | if (memory_limit != PHYS_ADDR_MAX) { |
a7f8de16 AB |
674 | pr_emerg("Memory Limit: %llu MB\n", memory_limit >> 20); |
675 | } else { | |
676 | pr_emerg("Memory Limit: none\n"); | |
677 | } | |
678 | return 0; | |
679 | } | |
680 | ||
681 | static struct notifier_block mem_limit_notifier = { | |
682 | .notifier_call = dump_mem_limit, | |
683 | }; | |
684 | ||
685 | static int __init register_mem_limit_dumper(void) | |
686 | { | |
687 | atomic_notifier_chain_register(&panic_notifier_list, | |
688 | &mem_limit_notifier); | |
689 | return 0; | |
690 | } | |
691 | __initcall(register_mem_limit_dumper); |