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-map-ops.h>
25 #include <linux/efi.h>
26 #include <linux/swiotlb.h>
27 #include <linux/vmalloc.h>
28 #include <linux/mm.h>
29 #include <linux/kexec.h>
30 #include <linux/crash_dump.h>
31 #include <linux/hugetlb.h>
32 #include <linux/acpi_iort.h>
33 #include <linux/kmemleak.h>
34
35 #include <asm/boot.h>
36 #include <asm/fixmap.h>
37 #include <asm/kasan.h>
38 #include <asm/kernel-pgtable.h>
39 #include <asm/kvm_host.h>
40 #include <asm/memory.h>
41 #include <asm/numa.h>
42 #include <asm/sections.h>
43 #include <asm/setup.h>
44 #include <linux/sizes.h>
45 #include <asm/tlb.h>
46 #include <asm/alternative.h>
47 #include <asm/xen/swiotlb-xen.h>
48
49 /*
50 * We need to be able to catch inadvertent references to memstart_addr
51 * that occur (potentially in generic code) before arm64_memblock_init()
52 * executes, which assigns it its actual value. So use a default value
53 * that cannot be mistaken for a real physical address.
54 */
55 s64 memstart_addr __ro_after_init = -1;
56 EXPORT_SYMBOL(memstart_addr);
57
58 /*
59 * If the corresponding config options are enabled, we create both ZONE_DMA
60 * and ZONE_DMA32. By default ZONE_DMA covers the 32-bit addressable memory
61 * unless restricted on specific platforms (e.g. 30-bit on Raspberry Pi 4).
62 * In such case, ZONE_DMA32 covers the rest of the 32-bit addressable memory,
63 * otherwise it is empty.
64 *
65 * Memory reservation for crash kernel either done early or deferred
66 * depending on DMA memory zones configs (ZONE_DMA) --
67 *
68 * In absence of ZONE_DMA configs arm64_dma_phys_limit initialized
69 * here instead of max_zone_phys(). This lets early reservation of
70 * crash kernel memory which has a dependency on arm64_dma_phys_limit.
71 * Reserving memory early for crash kernel allows linear creation of block
72 * mappings (greater than page-granularity) for all the memory bank rangs.
73 * In this scheme a comparatively quicker boot is observed.
74 *
75 * If ZONE_DMA configs are defined, crash kernel memory reservation
76 * is delayed until DMA zone memory range size initilazation performed in
77 * zone_sizes_init(). The defer is necessary to steer clear of DMA zone
78 * memory range to avoid overlap allocation. So crash kernel memory boundaries
79 * are not known when mapping all bank memory ranges, which otherwise means
80 * not possible to exclude crash kernel range from creating block mappings
81 * so page-granularity mappings are created for the entire memory range.
82 * Hence a slightly slower boot is observed.
83 *
84 * Note: Page-granularity mapppings are necessary for crash kernel memory
85 * range for shrinking its size via /sys/kernel/kexec_crash_size interface.
86 */
87 #if IS_ENABLED(CONFIG_ZONE_DMA) || IS_ENABLED(CONFIG_ZONE_DMA32)
88 phys_addr_t __ro_after_init arm64_dma_phys_limit;
89 #else
90 phys_addr_t __ro_after_init arm64_dma_phys_limit = PHYS_MASK + 1;
91 #endif
92
93 /*
94 * Provide a run-time mean of disabling ZONE_DMA32 if it is enabled via
95 * CONFIG_ZONE_DMA32.
96 */
97 static bool disable_dma32 __ro_after_init;
98
99 #ifdef CONFIG_KEXEC_CORE
100 /*
101 * reserve_crashkernel() - reserves memory for crash kernel
102 *
103 * This function reserves memory area given in "crashkernel=" kernel command
104 * line parameter. The memory reserved is used by dump capture kernel when
105 * primary kernel is crashing.
106 */
reserve_crashkernel(void)107 static void __init reserve_crashkernel(void)
108 {
109 unsigned long long crash_base, crash_size;
110 unsigned long long crash_max = arm64_dma_phys_limit;
111 int ret;
112
113 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
114 &crash_size, &crash_base);
115 /* no crashkernel= or invalid value specified */
116 if (ret || !crash_size)
117 return;
118
119 crash_size = PAGE_ALIGN(crash_size);
120
121 /* User specifies base address explicitly. */
122 if (crash_base)
123 crash_max = crash_base + crash_size;
124
125 /* Current arm64 boot protocol requires 2MB alignment */
126 crash_base = memblock_phys_alloc_range(crash_size, SZ_2M,
127 crash_base, crash_max);
128 if (!crash_base) {
129 pr_warn("cannot allocate crashkernel (size:0x%llx)\n",
130 crash_size);
131 return;
132 }
133
134 pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n",
135 crash_base, crash_base + crash_size, crash_size >> 20);
136
137 /*
138 * The crashkernel memory will be removed from the kernel linear
139 * map. Inform kmemleak so that it won't try to access it.
140 */
141 kmemleak_ignore_phys(crash_base);
142 crashk_res.start = crash_base;
143 crashk_res.end = crash_base + crash_size - 1;
144 }
145 #else
reserve_crashkernel(void)146 static void __init reserve_crashkernel(void)
147 {
148 }
149 #endif /* CONFIG_KEXEC_CORE */
150
151 /*
152 * Return the maximum physical address for a zone accessible by the given bits
153 * limit. If DRAM starts above 32-bit, expand the zone to the maximum
154 * available memory, otherwise cap it at 32-bit.
155 */
max_zone_phys(unsigned int zone_bits)156 static phys_addr_t __init max_zone_phys(unsigned int zone_bits)
157 {
158 phys_addr_t zone_mask = DMA_BIT_MASK(zone_bits);
159 phys_addr_t phys_start = memblock_start_of_DRAM();
160
161 if (phys_start > U32_MAX)
162 zone_mask = PHYS_ADDR_MAX;
163 else if (phys_start > zone_mask)
164 zone_mask = U32_MAX;
165
166 return min(zone_mask, memblock_end_of_DRAM() - 1) + 1;
167 }
168
zone_sizes_init(unsigned long min,unsigned long max)169 static void __init zone_sizes_init(unsigned long min, unsigned long max)
170 {
171 unsigned long max_zone_pfns[MAX_NR_ZONES] = {0};
172 unsigned int __maybe_unused acpi_zone_dma_bits;
173 unsigned int __maybe_unused dt_zone_dma_bits;
174 phys_addr_t __maybe_unused dma32_phys_limit = max_zone_phys(32);
175
176 #ifdef CONFIG_ZONE_DMA
177 acpi_zone_dma_bits = fls64(acpi_iort_dma_get_max_cpu_address());
178 dt_zone_dma_bits = fls64(of_dma_get_max_cpu_address(NULL));
179 zone_dma_bits = min3(32U, dt_zone_dma_bits, acpi_zone_dma_bits);
180 arm64_dma_phys_limit = max_zone_phys(zone_dma_bits);
181 max_zone_pfns[ZONE_DMA] = PFN_DOWN(arm64_dma_phys_limit);
182 #endif
183 #ifdef CONFIG_ZONE_DMA32
184 max_zone_pfns[ZONE_DMA32] = disable_dma32 ? 0 : PFN_DOWN(dma32_phys_limit);
185 if (!arm64_dma_phys_limit)
186 arm64_dma_phys_limit = dma32_phys_limit;
187 #endif
188 max_zone_pfns[ZONE_NORMAL] = max;
189
190 free_area_init(max_zone_pfns);
191 }
192
early_disable_dma32(char * buf)193 static int __init early_disable_dma32(char *buf)
194 {
195 if (!buf)
196 return -EINVAL;
197
198 if (!strcmp(buf, "on"))
199 disable_dma32 = true;
200
201 return 0;
202 }
203 early_param("disable_dma32", early_disable_dma32);
204
pfn_is_map_memory(unsigned long pfn)205 int pfn_is_map_memory(unsigned long pfn)
206 {
207 phys_addr_t addr = PFN_PHYS(pfn);
208
209 /* avoid false positives for bogus PFNs, see comment in pfn_valid() */
210 if (PHYS_PFN(addr) != pfn)
211 return 0;
212
213 return memblock_is_map_memory(addr);
214 }
215 EXPORT_SYMBOL(pfn_is_map_memory);
216
217 static phys_addr_t memory_limit = PHYS_ADDR_MAX;
218
219 /*
220 * Limit the memory size that was specified via FDT.
221 */
early_mem(char * p)222 static int __init early_mem(char *p)
223 {
224 if (!p)
225 return 1;
226
227 memory_limit = memparse(p, &p) & PAGE_MASK;
228 pr_notice("Memory limited to %lldMB\n", memory_limit >> 20);
229
230 return 0;
231 }
232 early_param("mem", early_mem);
233
arm64_memblock_init(void)234 void __init arm64_memblock_init(void)
235 {
236 s64 linear_region_size = PAGE_END - _PAGE_OFFSET(vabits_actual);
237
238 /*
239 * Corner case: 52-bit VA capable systems running KVM in nVHE mode may
240 * be limited in their ability to support a linear map that exceeds 51
241 * bits of VA space, depending on the placement of the ID map. Given
242 * that the placement of the ID map may be randomized, let's simply
243 * limit the kernel's linear map to 51 bits as well if we detect this
244 * configuration.
245 */
246 if (IS_ENABLED(CONFIG_KVM) && vabits_actual == 52 &&
247 is_hyp_mode_available() && !is_kernel_in_hyp_mode()) {
248 pr_info("Capping linear region to 51 bits for KVM in nVHE mode on LVA capable hardware.\n");
249 linear_region_size = min_t(u64, linear_region_size, BIT(51));
250 }
251
252 /* Remove memory above our supported physical address size */
253 memblock_remove(1ULL << PHYS_MASK_SHIFT, ULLONG_MAX);
254
255 /*
256 * Select a suitable value for the base of physical memory.
257 */
258 memstart_addr = round_down(memblock_start_of_DRAM(),
259 ARM64_MEMSTART_ALIGN);
260
261 if ((memblock_end_of_DRAM() - memstart_addr) > linear_region_size)
262 pr_warn("Memory doesn't fit in the linear mapping, VA_BITS too small\n");
263
264 /*
265 * Remove the memory that we will not be able to cover with the
266 * linear mapping. Take care not to clip the kernel which may be
267 * high in memory.
268 */
269 memblock_remove(max_t(u64, memstart_addr + linear_region_size,
270 __pa_symbol(_end)), ULLONG_MAX);
271 if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {
272 /* ensure that memstart_addr remains sufficiently aligned */
273 memstart_addr = round_up(memblock_end_of_DRAM() - linear_region_size,
274 ARM64_MEMSTART_ALIGN);
275 memblock_remove(0, memstart_addr);
276 }
277
278 /*
279 * If we are running with a 52-bit kernel VA config on a system that
280 * does not support it, we have to place the available physical
281 * memory in the 48-bit addressable part of the linear region, i.e.,
282 * we have to move it upward. Since memstart_addr represents the
283 * physical address of PAGE_OFFSET, we have to *subtract* from it.
284 */
285 if (IS_ENABLED(CONFIG_ARM64_VA_BITS_52) && (vabits_actual != 52))
286 memstart_addr -= _PAGE_OFFSET(48) - _PAGE_OFFSET(52);
287
288 /*
289 * Apply the memory limit if it was set. Since the kernel may be loaded
290 * high up in memory, add back the kernel region that must be accessible
291 * via the linear mapping.
292 */
293 if (memory_limit != PHYS_ADDR_MAX) {
294 memblock_mem_limit_remove_map(memory_limit);
295 memblock_add(__pa_symbol(_text), (u64)(_end - _text));
296 }
297
298 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
299 /*
300 * Add back the memory we just removed if it results in the
301 * initrd to become inaccessible via the linear mapping.
302 * Otherwise, this is a no-op
303 */
304 u64 base = phys_initrd_start & PAGE_MASK;
305 u64 size = PAGE_ALIGN(phys_initrd_start + phys_initrd_size) - base;
306
307 /*
308 * We can only add back the initrd memory if we don't end up
309 * with more memory than we can address via the linear mapping.
310 * It is up to the bootloader to position the kernel and the
311 * initrd reasonably close to each other (i.e., within 32 GB of
312 * each other) so that all granule/#levels combinations can
313 * always access both.
314 */
315 if (WARN(base < memblock_start_of_DRAM() ||
316 base + size > memblock_start_of_DRAM() +
317 linear_region_size,
318 "initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
319 phys_initrd_size = 0;
320 } else {
321 memblock_remove(base, size); /* clear MEMBLOCK_ flags */
322 memblock_add(base, size);
323 memblock_reserve(base, size);
324 }
325 }
326
327 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
328 extern u16 memstart_offset_seed;
329 u64 mmfr0 = read_cpuid(ID_AA64MMFR0_EL1);
330 int parange = cpuid_feature_extract_unsigned_field(
331 mmfr0, ID_AA64MMFR0_EL1_PARANGE_SHIFT);
332 s64 range = linear_region_size -
333 BIT(id_aa64mmfr0_parange_to_phys_shift(parange));
334
335 /*
336 * If the size of the linear region exceeds, by a sufficient
337 * margin, the size of the region that the physical memory can
338 * span, randomize the linear region as well.
339 */
340 if (memstart_offset_seed > 0 && range >= (s64)ARM64_MEMSTART_ALIGN) {
341 range /= ARM64_MEMSTART_ALIGN;
342 memstart_addr -= ARM64_MEMSTART_ALIGN *
343 ((range * memstart_offset_seed) >> 16);
344 }
345 }
346
347 /*
348 * Register the kernel text, kernel data, initrd, and initial
349 * pagetables with memblock.
350 */
351 memblock_reserve(__pa_symbol(_stext), _end - _stext);
352 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
353 /* the generic initrd code expects virtual addresses */
354 initrd_start = __phys_to_virt(phys_initrd_start);
355 initrd_end = initrd_start + phys_initrd_size;
356 }
357
358 early_init_fdt_scan_reserved_mem();
359
360 if (!IS_ENABLED(CONFIG_ZONE_DMA) && !IS_ENABLED(CONFIG_ZONE_DMA32))
361 reserve_crashkernel();
362
363 high_memory = __va(memblock_end_of_DRAM() - 1) + 1;
364 }
365
bootmem_init(void)366 void __init bootmem_init(void)
367 {
368 unsigned long min, max;
369
370 min = PFN_UP(memblock_start_of_DRAM());
371 max = PFN_DOWN(memblock_end_of_DRAM());
372
373 early_memtest(min << PAGE_SHIFT, max << PAGE_SHIFT);
374
375 max_pfn = max_low_pfn = max;
376 min_low_pfn = min;
377
378 arch_numa_init();
379
380 /*
381 * must be done after arch_numa_init() which calls numa_init() to
382 * initialize node_online_map that gets used in hugetlb_cma_reserve()
383 * while allocating required CMA size across online nodes.
384 */
385 #if defined(CONFIG_HUGETLB_PAGE) && defined(CONFIG_CMA)
386 arm64_hugetlb_cma_reserve();
387 #endif
388
389 dma_pernuma_cma_reserve();
390
391 kvm_hyp_reserve();
392
393 /*
394 * sparse_init() tries to allocate memory from memblock, so must be
395 * done after the fixed reservations
396 */
397 sparse_init();
398 zone_sizes_init(min, max);
399
400 /*
401 * Reserve the CMA area after arm64_dma_phys_limit was initialised.
402 */
403 dma_contiguous_reserve(arm64_dma_phys_limit);
404
405 /*
406 * request_standard_resources() depends on crashkernel's memory being
407 * reserved, so do it here.
408 */
409 if (IS_ENABLED(CONFIG_ZONE_DMA) || IS_ENABLED(CONFIG_ZONE_DMA32))
410 reserve_crashkernel();
411
412 memblock_dump_all();
413 }
414
415 /*
416 * mem_init() marks the free areas in the mem_map and tells us how much memory
417 * is free. This is done after various parts of the system have claimed their
418 * memory after the kernel image.
419 */
mem_init(void)420 void __init mem_init(void)
421 {
422 if (swiotlb_force == SWIOTLB_FORCE ||
423 max_pfn > PFN_DOWN(arm64_dma_phys_limit))
424 swiotlb_init(1);
425 else if (!xen_swiotlb_detect())
426 swiotlb_force = SWIOTLB_NO_FORCE;
427
428 set_max_mapnr(max_pfn - PHYS_PFN_OFFSET);
429
430 /* this will put all unused low memory onto the freelists */
431 memblock_free_all();
432
433 /*
434 * Check boundaries twice: Some fundamental inconsistencies can be
435 * detected at build time already.
436 */
437 #ifdef CONFIG_COMPAT
438 BUILD_BUG_ON(TASK_SIZE_32 > DEFAULT_MAP_WINDOW_64);
439 #endif
440
441 /*
442 * Selected page table levels should match when derived from
443 * scratch using the virtual address range and page size.
444 */
445 BUILD_BUG_ON(ARM64_HW_PGTABLE_LEVELS(CONFIG_ARM64_VA_BITS) !=
446 CONFIG_PGTABLE_LEVELS);
447
448 if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) {
449 extern int sysctl_overcommit_memory;
450 /*
451 * On a machine this small we won't get anywhere without
452 * overcommit, so turn it on by default.
453 */
454 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
455 }
456 }
457
free_initmem(void)458 void free_initmem(void)
459 {
460 free_reserved_area(lm_alias(__init_begin),
461 lm_alias(__init_end),
462 POISON_FREE_INITMEM, "unused kernel");
463 /*
464 * Unmap the __init region but leave the VM area in place. This
465 * prevents the region from being reused for kernel modules, which
466 * is not supported by kallsyms.
467 */
468 vunmap_range((u64)__init_begin, (u64)__init_end);
469 }
470
dump_mem_limit(void)471 void dump_mem_limit(void)
472 {
473 if (memory_limit != PHYS_ADDR_MAX) {
474 pr_emerg("Memory Limit: %llu MB\n", memory_limit >> 20);
475 } else {
476 pr_emerg("Memory Limit: none\n");
477 }
478 }
479