Loading...
Note: File does not exist in v3.1.
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Based on arch/arm/mm/init.c
4 *
5 * Copyright (C) 1995-2005 Russell King
6 * Copyright (C) 2012 ARM Ltd.
7 */
8
9#include <linux/kernel.h>
10#include <linux/export.h>
11#include <linux/errno.h>
12#include <linux/swap.h>
13#include <linux/init.h>
14#include <linux/cache.h>
15#include <linux/mman.h>
16#include <linux/nodemask.h>
17#include <linux/initrd.h>
18#include <linux/gfp.h>
19#include <linux/memblock.h>
20#include <linux/sort.h>
21#include <linux/of.h>
22#include <linux/of_fdt.h>
23#include <linux/dma-direct.h>
24#include <linux/dma-mapping.h>
25#include <linux/dma-contiguous.h>
26#include <linux/efi.h>
27#include <linux/swiotlb.h>
28#include <linux/vmalloc.h>
29#include <linux/mm.h>
30#include <linux/kexec.h>
31#include <linux/crash_dump.h>
32#include <linux/hugetlb.h>
33
34#include <asm/boot.h>
35#include <asm/fixmap.h>
36#include <asm/kasan.h>
37#include <asm/kernel-pgtable.h>
38#include <asm/memory.h>
39#include <asm/numa.h>
40#include <asm/sections.h>
41#include <asm/setup.h>
42#include <linux/sizes.h>
43#include <asm/tlb.h>
44#include <asm/alternative.h>
45
46#define ARM64_ZONE_DMA_BITS 30
47
48/*
49 * We need to be able to catch inadvertent references to memstart_addr
50 * that occur (potentially in generic code) before arm64_memblock_init()
51 * executes, which assigns it its actual value. So use a default value
52 * that cannot be mistaken for a real physical address.
53 */
54s64 memstart_addr __ro_after_init = -1;
55EXPORT_SYMBOL(memstart_addr);
56
57s64 physvirt_offset __ro_after_init;
58EXPORT_SYMBOL(physvirt_offset);
59
60struct page *vmemmap __ro_after_init;
61EXPORT_SYMBOL(vmemmap);
62
63/*
64 * We create both ZONE_DMA and ZONE_DMA32. ZONE_DMA covers the first 1G of
65 * memory as some devices, namely the Raspberry Pi 4, have peripherals with
66 * this limited view of the memory. ZONE_DMA32 will cover the rest of the 32
67 * bit addressable memory area.
68 */
69phys_addr_t arm64_dma_phys_limit __ro_after_init;
70static phys_addr_t arm64_dma32_phys_limit __ro_after_init;
71
72#ifdef CONFIG_KEXEC_CORE
73/*
74 * reserve_crashkernel() - reserves memory for crash kernel
75 *
76 * This function reserves memory area given in "crashkernel=" kernel command
77 * line parameter. The memory reserved is used by dump capture kernel when
78 * primary kernel is crashing.
79 */
80static void __init reserve_crashkernel(void)
81{
82 unsigned long long crash_base, crash_size;
83 int ret;
84
85 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
86 &crash_size, &crash_base);
87 /* no crashkernel= or invalid value specified */
88 if (ret || !crash_size)
89 return;
90
91 crash_size = PAGE_ALIGN(crash_size);
92
93 if (crash_base == 0) {
94 /* Current arm64 boot protocol requires 2MB alignment */
95 crash_base = memblock_find_in_range(0, arm64_dma32_phys_limit,
96 crash_size, SZ_2M);
97 if (crash_base == 0) {
98 pr_warn("cannot allocate crashkernel (size:0x%llx)\n",
99 crash_size);
100 return;
101 }
102 } else {
103 /* User specifies base address explicitly. */
104 if (!memblock_is_region_memory(crash_base, crash_size)) {
105 pr_warn("cannot reserve crashkernel: region is not memory\n");
106 return;
107 }
108
109 if (memblock_is_region_reserved(crash_base, crash_size)) {
110 pr_warn("cannot reserve crashkernel: region overlaps reserved memory\n");
111 return;
112 }
113
114 if (!IS_ALIGNED(crash_base, SZ_2M)) {
115 pr_warn("cannot reserve crashkernel: base address is not 2MB aligned\n");
116 return;
117 }
118 }
119 memblock_reserve(crash_base, crash_size);
120
121 pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n",
122 crash_base, crash_base + crash_size, crash_size >> 20);
123
124 crashk_res.start = crash_base;
125 crashk_res.end = crash_base + crash_size - 1;
126}
127#else
128static void __init reserve_crashkernel(void)
129{
130}
131#endif /* CONFIG_KEXEC_CORE */
132
133#ifdef CONFIG_CRASH_DUMP
134static int __init early_init_dt_scan_elfcorehdr(unsigned long node,
135 const char *uname, int depth, void *data)
136{
137 const __be32 *reg;
138 int len;
139
140 if (depth != 1 || strcmp(uname, "chosen") != 0)
141 return 0;
142
143 reg = of_get_flat_dt_prop(node, "linux,elfcorehdr", &len);
144 if (!reg || (len < (dt_root_addr_cells + dt_root_size_cells)))
145 return 1;
146
147 elfcorehdr_addr = dt_mem_next_cell(dt_root_addr_cells, ®);
148 elfcorehdr_size = dt_mem_next_cell(dt_root_size_cells, ®);
149
150 return 1;
151}
152
153/*
154 * reserve_elfcorehdr() - reserves memory for elf core header
155 *
156 * This function reserves the memory occupied by an elf core header
157 * described in the device tree. This region contains all the
158 * information about primary kernel's core image and is used by a dump
159 * capture kernel to access the system memory on primary kernel.
160 */
161static void __init reserve_elfcorehdr(void)
162{
163 of_scan_flat_dt(early_init_dt_scan_elfcorehdr, NULL);
164
165 if (!elfcorehdr_size)
166 return;
167
168 if (memblock_is_region_reserved(elfcorehdr_addr, elfcorehdr_size)) {
169 pr_warn("elfcorehdr is overlapped\n");
170 return;
171 }
172
173 memblock_reserve(elfcorehdr_addr, elfcorehdr_size);
174
175 pr_info("Reserving %lldKB of memory at 0x%llx for elfcorehdr\n",
176 elfcorehdr_size >> 10, elfcorehdr_addr);
177}
178#else
179static void __init reserve_elfcorehdr(void)
180{
181}
182#endif /* CONFIG_CRASH_DUMP */
183
184/*
185 * Return the maximum physical address for a zone with a given address size
186 * limit. It currently assumes that for memory starting above 4G, 32-bit
187 * devices will use a DMA offset.
188 */
189static phys_addr_t __init max_zone_phys(unsigned int zone_bits)
190{
191 phys_addr_t offset = memblock_start_of_DRAM() & GENMASK_ULL(63, zone_bits);
192 return min(offset + (1ULL << zone_bits), memblock_end_of_DRAM());
193}
194
195static void __init zone_sizes_init(unsigned long min, unsigned long max)
196{
197 unsigned long max_zone_pfns[MAX_NR_ZONES] = {0};
198
199#ifdef CONFIG_ZONE_DMA
200 max_zone_pfns[ZONE_DMA] = PFN_DOWN(arm64_dma_phys_limit);
201#endif
202#ifdef CONFIG_ZONE_DMA32
203 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(arm64_dma32_phys_limit);
204#endif
205 max_zone_pfns[ZONE_NORMAL] = max;
206
207 free_area_init(max_zone_pfns);
208}
209
210int pfn_valid(unsigned long pfn)
211{
212 phys_addr_t addr = pfn << PAGE_SHIFT;
213
214 if ((addr >> PAGE_SHIFT) != pfn)
215 return 0;
216
217#ifdef CONFIG_SPARSEMEM
218 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
219 return 0;
220
221 if (!valid_section(__pfn_to_section(pfn)))
222 return 0;
223#endif
224 return memblock_is_map_memory(addr);
225}
226EXPORT_SYMBOL(pfn_valid);
227
228static phys_addr_t memory_limit = PHYS_ADDR_MAX;
229
230/*
231 * Limit the memory size that was specified via FDT.
232 */
233static int __init early_mem(char *p)
234{
235 if (!p)
236 return 1;
237
238 memory_limit = memparse(p, &p) & PAGE_MASK;
239 pr_notice("Memory limited to %lldMB\n", memory_limit >> 20);
240
241 return 0;
242}
243early_param("mem", early_mem);
244
245static int __init early_init_dt_scan_usablemem(unsigned long node,
246 const char *uname, int depth, void *data)
247{
248 struct memblock_region *usablemem = data;
249 const __be32 *reg;
250 int len;
251
252 if (depth != 1 || strcmp(uname, "chosen") != 0)
253 return 0;
254
255 reg = of_get_flat_dt_prop(node, "linux,usable-memory-range", &len);
256 if (!reg || (len < (dt_root_addr_cells + dt_root_size_cells)))
257 return 1;
258
259 usablemem->base = dt_mem_next_cell(dt_root_addr_cells, ®);
260 usablemem->size = dt_mem_next_cell(dt_root_size_cells, ®);
261
262 return 1;
263}
264
265static void __init fdt_enforce_memory_region(void)
266{
267 struct memblock_region reg = {
268 .size = 0,
269 };
270
271 of_scan_flat_dt(early_init_dt_scan_usablemem, ®);
272
273 if (reg.size)
274 memblock_cap_memory_range(reg.base, reg.size);
275}
276
277void __init arm64_memblock_init(void)
278{
279 const s64 linear_region_size = BIT(vabits_actual - 1);
280
281 /* Handle linux,usable-memory-range property */
282 fdt_enforce_memory_region();
283
284 /* Remove memory above our supported physical address size */
285 memblock_remove(1ULL << PHYS_MASK_SHIFT, ULLONG_MAX);
286
287 /*
288 * Select a suitable value for the base of physical memory.
289 */
290 memstart_addr = round_down(memblock_start_of_DRAM(),
291 ARM64_MEMSTART_ALIGN);
292
293 physvirt_offset = PHYS_OFFSET - PAGE_OFFSET;
294
295 vmemmap = ((struct page *)VMEMMAP_START - (memstart_addr >> PAGE_SHIFT));
296
297 /*
298 * If we are running with a 52-bit kernel VA config on a system that
299 * does not support it, we have to offset our vmemmap and physvirt_offset
300 * s.t. we avoid the 52-bit portion of the direct linear map
301 */
302 if (IS_ENABLED(CONFIG_ARM64_VA_BITS_52) && (vabits_actual != 52)) {
303 vmemmap += (_PAGE_OFFSET(48) - _PAGE_OFFSET(52)) >> PAGE_SHIFT;
304 physvirt_offset = PHYS_OFFSET - _PAGE_OFFSET(48);
305 }
306
307 /*
308 * Remove the memory that we will not be able to cover with the
309 * linear mapping. Take care not to clip the kernel which may be
310 * high in memory.
311 */
312 memblock_remove(max_t(u64, memstart_addr + linear_region_size,
313 __pa_symbol(_end)), ULLONG_MAX);
314 if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {
315 /* ensure that memstart_addr remains sufficiently aligned */
316 memstart_addr = round_up(memblock_end_of_DRAM() - linear_region_size,
317 ARM64_MEMSTART_ALIGN);
318 memblock_remove(0, memstart_addr);
319 }
320
321 /*
322 * Apply the memory limit if it was set. Since the kernel may be loaded
323 * high up in memory, add back the kernel region that must be accessible
324 * via the linear mapping.
325 */
326 if (memory_limit != PHYS_ADDR_MAX) {
327 memblock_mem_limit_remove_map(memory_limit);
328 memblock_add(__pa_symbol(_text), (u64)(_end - _text));
329 }
330
331 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
332 /*
333 * Add back the memory we just removed if it results in the
334 * initrd to become inaccessible via the linear mapping.
335 * Otherwise, this is a no-op
336 */
337 u64 base = phys_initrd_start & PAGE_MASK;
338 u64 size = PAGE_ALIGN(phys_initrd_start + phys_initrd_size) - base;
339
340 /*
341 * We can only add back the initrd memory if we don't end up
342 * with more memory than we can address via the linear mapping.
343 * It is up to the bootloader to position the kernel and the
344 * initrd reasonably close to each other (i.e., within 32 GB of
345 * each other) so that all granule/#levels combinations can
346 * always access both.
347 */
348 if (WARN(base < memblock_start_of_DRAM() ||
349 base + size > memblock_start_of_DRAM() +
350 linear_region_size,
351 "initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
352 phys_initrd_size = 0;
353 } else {
354 memblock_remove(base, size); /* clear MEMBLOCK_ flags */
355 memblock_add(base, size);
356 memblock_reserve(base, size);
357 }
358 }
359
360 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
361 extern u16 memstart_offset_seed;
362 u64 range = linear_region_size -
363 (memblock_end_of_DRAM() - memblock_start_of_DRAM());
364
365 /*
366 * If the size of the linear region exceeds, by a sufficient
367 * margin, the size of the region that the available physical
368 * memory spans, randomize the linear region as well.
369 */
370 if (memstart_offset_seed > 0 && range >= ARM64_MEMSTART_ALIGN) {
371 range /= ARM64_MEMSTART_ALIGN;
372 memstart_addr -= ARM64_MEMSTART_ALIGN *
373 ((range * memstart_offset_seed) >> 16);
374 }
375 }
376
377 /*
378 * Register the kernel text, kernel data, initrd, and initial
379 * pagetables with memblock.
380 */
381 memblock_reserve(__pa_symbol(_text), _end - _text);
382 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
383 /* the generic initrd code expects virtual addresses */
384 initrd_start = __phys_to_virt(phys_initrd_start);
385 initrd_end = initrd_start + phys_initrd_size;
386 }
387
388 early_init_fdt_scan_reserved_mem();
389
390 if (IS_ENABLED(CONFIG_ZONE_DMA)) {
391 zone_dma_bits = ARM64_ZONE_DMA_BITS;
392 arm64_dma_phys_limit = max_zone_phys(ARM64_ZONE_DMA_BITS);
393 }
394
395 if (IS_ENABLED(CONFIG_ZONE_DMA32))
396 arm64_dma32_phys_limit = max_zone_phys(32);
397 else
398 arm64_dma32_phys_limit = PHYS_MASK + 1;
399
400 reserve_crashkernel();
401
402 reserve_elfcorehdr();
403
404 high_memory = __va(memblock_end_of_DRAM() - 1) + 1;
405
406 dma_contiguous_reserve(arm64_dma32_phys_limit);
407}
408
409void __init bootmem_init(void)
410{
411 unsigned long min, max;
412
413 min = PFN_UP(memblock_start_of_DRAM());
414 max = PFN_DOWN(memblock_end_of_DRAM());
415
416 early_memtest(min << PAGE_SHIFT, max << PAGE_SHIFT);
417
418 max_pfn = max_low_pfn = max;
419 min_low_pfn = min;
420
421 arm64_numa_init();
422
423 /*
424 * must be done after arm64_numa_init() which calls numa_init() to
425 * initialize node_online_map that gets used in hugetlb_cma_reserve()
426 * while allocating required CMA size across online nodes.
427 */
428#if defined(CONFIG_HUGETLB_PAGE) && defined(CONFIG_CMA)
429 arm64_hugetlb_cma_reserve();
430#endif
431
432 /*
433 * sparse_init() tries to allocate memory from memblock, so must be
434 * done after the fixed reservations
435 */
436 sparse_init();
437 zone_sizes_init(min, max);
438
439 memblock_dump_all();
440}
441
442#ifndef CONFIG_SPARSEMEM_VMEMMAP
443static inline void free_memmap(unsigned long start_pfn, unsigned long end_pfn)
444{
445 struct page *start_pg, *end_pg;
446 unsigned long pg, pgend;
447
448 /*
449 * Convert start_pfn/end_pfn to a struct page pointer.
450 */
451 start_pg = pfn_to_page(start_pfn - 1) + 1;
452 end_pg = pfn_to_page(end_pfn - 1) + 1;
453
454 /*
455 * Convert to physical addresses, and round start upwards and end
456 * downwards.
457 */
458 pg = (unsigned long)PAGE_ALIGN(__pa(start_pg));
459 pgend = (unsigned long)__pa(end_pg) & PAGE_MASK;
460
461 /*
462 * If there are free pages between these, free the section of the
463 * memmap array.
464 */
465 if (pg < pgend)
466 memblock_free(pg, pgend - pg);
467}
468
469/*
470 * The mem_map array can get very big. Free the unused area of the memory map.
471 */
472static void __init free_unused_memmap(void)
473{
474 unsigned long start, prev_end = 0;
475 struct memblock_region *reg;
476
477 for_each_memblock(memory, reg) {
478 start = __phys_to_pfn(reg->base);
479
480#ifdef CONFIG_SPARSEMEM
481 /*
482 * Take care not to free memmap entries that don't exist due
483 * to SPARSEMEM sections which aren't present.
484 */
485 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
486#endif
487 /*
488 * If we had a previous bank, and there is a space between the
489 * current bank and the previous, free it.
490 */
491 if (prev_end && prev_end < start)
492 free_memmap(prev_end, start);
493
494 /*
495 * Align up here since the VM subsystem insists that the
496 * memmap entries are valid from the bank end aligned to
497 * MAX_ORDER_NR_PAGES.
498 */
499 prev_end = ALIGN(__phys_to_pfn(reg->base + reg->size),
500 MAX_ORDER_NR_PAGES);
501 }
502
503#ifdef CONFIG_SPARSEMEM
504 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
505 free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
506#endif
507}
508#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
509
510/*
511 * mem_init() marks the free areas in the mem_map and tells us how much memory
512 * is free. This is done after various parts of the system have claimed their
513 * memory after the kernel image.
514 */
515void __init mem_init(void)
516{
517 if (swiotlb_force == SWIOTLB_FORCE ||
518 max_pfn > PFN_DOWN(arm64_dma_phys_limit ? : arm64_dma32_phys_limit))
519 swiotlb_init(1);
520 else
521 swiotlb_force = SWIOTLB_NO_FORCE;
522
523 set_max_mapnr(max_pfn - PHYS_PFN_OFFSET);
524
525#ifndef CONFIG_SPARSEMEM_VMEMMAP
526 free_unused_memmap();
527#endif
528 /* this will put all unused low memory onto the freelists */
529 memblock_free_all();
530
531 mem_init_print_info(NULL);
532
533 /*
534 * Check boundaries twice: Some fundamental inconsistencies can be
535 * detected at build time already.
536 */
537#ifdef CONFIG_COMPAT
538 BUILD_BUG_ON(TASK_SIZE_32 > DEFAULT_MAP_WINDOW_64);
539#endif
540
541 if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) {
542 extern int sysctl_overcommit_memory;
543 /*
544 * On a machine this small we won't get anywhere without
545 * overcommit, so turn it on by default.
546 */
547 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
548 }
549}
550
551void free_initmem(void)
552{
553 free_reserved_area(lm_alias(__init_begin),
554 lm_alias(__init_end),
555 POISON_FREE_INITMEM, "unused kernel");
556 /*
557 * Unmap the __init region but leave the VM area in place. This
558 * prevents the region from being reused for kernel modules, which
559 * is not supported by kallsyms.
560 */
561 unmap_kernel_range((u64)__init_begin, (u64)(__init_end - __init_begin));
562}
563
564void dump_mem_limit(void)
565{
566 if (memory_limit != PHYS_ADDR_MAX) {
567 pr_emerg("Memory Limit: %llu MB\n", memory_limit >> 20);
568 } else {
569 pr_emerg("Memory Limit: none\n");
570 }
571}