1 // SPDX-License-Identifier: GPL-2.0-only
3 * Based on arch/arm/mm/init.c
5 * Copyright (C) 1995-2005 Russell King
6 * Copyright (C) 2012 ARM Ltd.
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>
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>
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>
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>
42 #include <asm/sections.h>
43 #include <asm/setup.h>
44 #include <linux/sizes.h>
46 #include <asm/alternative.h>
47 #include <asm/xen/swiotlb-xen.h>
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.
55 s64 memstart_addr __ro_after_init = -1;
56 EXPORT_SYMBOL(memstart_addr);
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.
65 phys_addr_t arm64_dma_phys_limit __ro_after_init;
67 #ifdef CONFIG_KEXEC_CORE
69 * reserve_crashkernel() - reserves memory for crash kernel
71 * This function reserves memory area given in "crashkernel=" kernel command
72 * line parameter. The memory reserved is used by dump capture kernel when
73 * primary kernel is crashing.
75 static void __init reserve_crashkernel(void)
77 unsigned long long crash_base, crash_size;
80 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
81 &crash_size, &crash_base);
82 /* no crashkernel= or invalid value specified */
83 if (ret || !crash_size)
86 crash_size = PAGE_ALIGN(crash_size);
88 if (crash_base == 0) {
89 /* Current arm64 boot protocol requires 2MB alignment */
90 crash_base = memblock_find_in_range(0, arm64_dma_phys_limit,
92 if (crash_base == 0) {
93 pr_warn("cannot allocate crashkernel (size:0x%llx)\n",
98 /* User specifies base address explicitly. */
99 if (!memblock_is_region_memory(crash_base, crash_size)) {
100 pr_warn("cannot reserve crashkernel: region is not memory\n");
104 if (memblock_is_region_reserved(crash_base, crash_size)) {
105 pr_warn("cannot reserve crashkernel: region overlaps reserved memory\n");
109 if (!IS_ALIGNED(crash_base, SZ_2M)) {
110 pr_warn("cannot reserve crashkernel: base address is not 2MB aligned\n");
114 memblock_reserve(crash_base, crash_size);
116 pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n",
117 crash_base, crash_base + crash_size, crash_size >> 20);
120 * The crashkernel memory will be removed from the kernel linear
121 * map. Inform kmemleak so that it won't try to access it.
123 kmemleak_ignore_phys(crash_base);
124 crashk_res.start = crash_base;
125 crashk_res.end = crash_base + crash_size - 1;
128 static void __init reserve_crashkernel(void)
131 #endif /* CONFIG_KEXEC_CORE */
133 #ifdef CONFIG_CRASH_DUMP
134 static int __init early_init_dt_scan_elfcorehdr(unsigned long node,
135 const char *uname, int depth, void *data)
140 if (depth != 1 || strcmp(uname, "chosen") != 0)
143 reg = of_get_flat_dt_prop(node, "linux,elfcorehdr", &len);
144 if (!reg || (len < (dt_root_addr_cells + dt_root_size_cells)))
147 elfcorehdr_addr = dt_mem_next_cell(dt_root_addr_cells, ®);
148 elfcorehdr_size = dt_mem_next_cell(dt_root_size_cells, ®);
154 * reserve_elfcorehdr() - reserves memory for elf core header
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.
161 static void __init reserve_elfcorehdr(void)
163 of_scan_flat_dt(early_init_dt_scan_elfcorehdr, NULL);
165 if (!elfcorehdr_size)
168 if (memblock_is_region_reserved(elfcorehdr_addr, elfcorehdr_size)) {
169 pr_warn("elfcorehdr is overlapped\n");
173 memblock_reserve(elfcorehdr_addr, elfcorehdr_size);
175 pr_info("Reserving %lldKB of memory at 0x%llx for elfcorehdr\n",
176 elfcorehdr_size >> 10, elfcorehdr_addr);
179 static void __init reserve_elfcorehdr(void)
182 #endif /* CONFIG_CRASH_DUMP */
185 * Return the maximum physical address for a zone accessible by the given bits
186 * limit. If DRAM starts above 32-bit, expand the zone to the maximum
187 * available memory, otherwise cap it at 32-bit.
189 static phys_addr_t __init max_zone_phys(unsigned int zone_bits)
191 phys_addr_t zone_mask = DMA_BIT_MASK(zone_bits);
192 phys_addr_t phys_start = memblock_start_of_DRAM();
194 if (phys_start > U32_MAX)
195 zone_mask = PHYS_ADDR_MAX;
196 else if (phys_start > zone_mask)
199 return min(zone_mask, memblock_end_of_DRAM() - 1) + 1;
202 static void __init zone_sizes_init(unsigned long min, unsigned long max)
204 unsigned long max_zone_pfns[MAX_NR_ZONES] = {0};
205 unsigned int __maybe_unused acpi_zone_dma_bits;
206 unsigned int __maybe_unused dt_zone_dma_bits;
207 phys_addr_t __maybe_unused dma32_phys_limit = max_zone_phys(32);
209 #ifdef CONFIG_ZONE_DMA
210 acpi_zone_dma_bits = fls64(acpi_iort_dma_get_max_cpu_address());
211 dt_zone_dma_bits = fls64(of_dma_get_max_cpu_address(NULL));
212 zone_dma_bits = min3(32U, dt_zone_dma_bits, acpi_zone_dma_bits);
213 arm64_dma_phys_limit = max_zone_phys(zone_dma_bits);
214 max_zone_pfns[ZONE_DMA] = PFN_DOWN(arm64_dma_phys_limit);
216 #ifdef CONFIG_ZONE_DMA32
217 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
218 if (!arm64_dma_phys_limit)
219 arm64_dma_phys_limit = dma32_phys_limit;
221 if (!arm64_dma_phys_limit)
222 arm64_dma_phys_limit = PHYS_MASK + 1;
223 max_zone_pfns[ZONE_NORMAL] = max;
225 free_area_init(max_zone_pfns);
228 int pfn_is_map_memory(unsigned long pfn)
230 phys_addr_t addr = PFN_PHYS(pfn);
232 /* avoid false positives for bogus PFNs, see comment in pfn_valid() */
233 if (PHYS_PFN(addr) != pfn)
236 return memblock_is_map_memory(addr);
238 EXPORT_SYMBOL(pfn_is_map_memory);
240 static phys_addr_t memory_limit = PHYS_ADDR_MAX;
243 * Limit the memory size that was specified via FDT.
245 static int __init early_mem(char *p)
250 memory_limit = memparse(p, &p) & PAGE_MASK;
251 pr_notice("Memory limited to %lldMB\n", memory_limit >> 20);
255 early_param("mem", early_mem);
257 static int __init early_init_dt_scan_usablemem(unsigned long node,
258 const char *uname, int depth, void *data)
260 struct memblock_region *usablemem = data;
264 if (depth != 1 || strcmp(uname, "chosen") != 0)
267 reg = of_get_flat_dt_prop(node, "linux,usable-memory-range", &len);
268 if (!reg || (len < (dt_root_addr_cells + dt_root_size_cells)))
271 usablemem->base = dt_mem_next_cell(dt_root_addr_cells, ®);
272 usablemem->size = dt_mem_next_cell(dt_root_size_cells, ®);
277 static void __init fdt_enforce_memory_region(void)
279 struct memblock_region reg = {
283 of_scan_flat_dt(early_init_dt_scan_usablemem, ®);
286 memblock_cap_memory_range(reg.base, reg.size);
289 void __init arm64_memblock_init(void)
291 s64 linear_region_size = PAGE_END - _PAGE_OFFSET(vabits_actual);
294 * Corner case: 52-bit VA capable systems running KVM in nVHE mode may
295 * be limited in their ability to support a linear map that exceeds 51
296 * bits of VA space, depending on the placement of the ID map. Given
297 * that the placement of the ID map may be randomized, let's simply
298 * limit the kernel's linear map to 51 bits as well if we detect this
301 if (IS_ENABLED(CONFIG_KVM) && vabits_actual == 52 &&
302 is_hyp_mode_available() && !is_kernel_in_hyp_mode()) {
303 pr_info("Capping linear region to 51 bits for KVM in nVHE mode on LVA capable hardware.\n");
304 linear_region_size = min_t(u64, linear_region_size, BIT(51));
307 /* Handle linux,usable-memory-range property */
308 fdt_enforce_memory_region();
310 /* Remove memory above our supported physical address size */
311 memblock_remove(1ULL << PHYS_MASK_SHIFT, ULLONG_MAX);
314 * Select a suitable value for the base of physical memory.
316 memstart_addr = round_down(memblock_start_of_DRAM(),
317 ARM64_MEMSTART_ALIGN);
319 if ((memblock_end_of_DRAM() - memstart_addr) > linear_region_size)
320 pr_warn("Memory doesn't fit in the linear mapping, VA_BITS too small\n");
323 * Remove the memory that we will not be able to cover with the
324 * linear mapping. Take care not to clip the kernel which may be
327 memblock_remove(max_t(u64, memstart_addr + linear_region_size,
328 __pa_symbol(_end)), ULLONG_MAX);
329 if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {
330 /* ensure that memstart_addr remains sufficiently aligned */
331 memstart_addr = round_up(memblock_end_of_DRAM() - linear_region_size,
332 ARM64_MEMSTART_ALIGN);
333 memblock_remove(0, memstart_addr);
337 * If we are running with a 52-bit kernel VA config on a system that
338 * does not support it, we have to place the available physical
339 * memory in the 48-bit addressable part of the linear region, i.e.,
340 * we have to move it upward. Since memstart_addr represents the
341 * physical address of PAGE_OFFSET, we have to *subtract* from it.
343 if (IS_ENABLED(CONFIG_ARM64_VA_BITS_52) && (vabits_actual != 52))
344 memstart_addr -= _PAGE_OFFSET(48) - _PAGE_OFFSET(52);
347 * Apply the memory limit if it was set. Since the kernel may be loaded
348 * high up in memory, add back the kernel region that must be accessible
349 * via the linear mapping.
351 if (memory_limit != PHYS_ADDR_MAX) {
352 memblock_mem_limit_remove_map(memory_limit);
353 memblock_add(__pa_symbol(_text), (u64)(_end - _text));
356 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
358 * Add back the memory we just removed if it results in the
359 * initrd to become inaccessible via the linear mapping.
360 * Otherwise, this is a no-op
362 u64 base = phys_initrd_start & PAGE_MASK;
363 u64 size = PAGE_ALIGN(phys_initrd_start + phys_initrd_size) - base;
366 * We can only add back the initrd memory if we don't end up
367 * with more memory than we can address via the linear mapping.
368 * It is up to the bootloader to position the kernel and the
369 * initrd reasonably close to each other (i.e., within 32 GB of
370 * each other) so that all granule/#levels combinations can
371 * always access both.
373 if (WARN(base < memblock_start_of_DRAM() ||
374 base + size > memblock_start_of_DRAM() +
376 "initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
377 phys_initrd_size = 0;
379 memblock_remove(base, size); /* clear MEMBLOCK_ flags */
380 memblock_add(base, size);
381 memblock_reserve(base, size);
385 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
386 extern u16 memstart_offset_seed;
387 u64 mmfr0 = read_cpuid(ID_AA64MMFR0_EL1);
388 int parange = cpuid_feature_extract_unsigned_field(
389 mmfr0, ID_AA64MMFR0_PARANGE_SHIFT);
390 s64 range = linear_region_size -
391 BIT(id_aa64mmfr0_parange_to_phys_shift(parange));
394 * If the size of the linear region exceeds, by a sufficient
395 * margin, the size of the region that the physical memory can
396 * span, randomize the linear region as well.
398 if (memstart_offset_seed > 0 && range >= (s64)ARM64_MEMSTART_ALIGN) {
399 range /= ARM64_MEMSTART_ALIGN;
400 memstart_addr -= ARM64_MEMSTART_ALIGN *
401 ((range * memstart_offset_seed) >> 16);
406 * Register the kernel text, kernel data, initrd, and initial
407 * pagetables with memblock.
409 memblock_reserve(__pa_symbol(_stext), _end - _stext);
410 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
411 /* the generic initrd code expects virtual addresses */
412 initrd_start = __phys_to_virt(phys_initrd_start);
413 initrd_end = initrd_start + phys_initrd_size;
416 early_init_fdt_scan_reserved_mem();
418 reserve_elfcorehdr();
420 high_memory = __va(memblock_end_of_DRAM() - 1) + 1;
423 void __init bootmem_init(void)
425 unsigned long min, max;
427 min = PFN_UP(memblock_start_of_DRAM());
428 max = PFN_DOWN(memblock_end_of_DRAM());
430 early_memtest(min << PAGE_SHIFT, max << PAGE_SHIFT);
432 max_pfn = max_low_pfn = max;
438 * must be done after arch_numa_init() which calls numa_init() to
439 * initialize node_online_map that gets used in hugetlb_cma_reserve()
440 * while allocating required CMA size across online nodes.
442 #if defined(CONFIG_HUGETLB_PAGE) && defined(CONFIG_CMA)
443 arm64_hugetlb_cma_reserve();
446 dma_pernuma_cma_reserve();
451 * sparse_init() tries to allocate memory from memblock, so must be
452 * done after the fixed reservations
455 zone_sizes_init(min, max);
458 * Reserve the CMA area after arm64_dma_phys_limit was initialised.
460 dma_contiguous_reserve(arm64_dma_phys_limit);
463 * request_standard_resources() depends on crashkernel's memory being
464 * reserved, so do it here.
466 reserve_crashkernel();
472 * mem_init() marks the free areas in the mem_map and tells us how much memory
473 * is free. This is done after various parts of the system have claimed their
474 * memory after the kernel image.
476 void __init mem_init(void)
478 if (swiotlb_force == SWIOTLB_FORCE ||
479 max_pfn > PFN_DOWN(arm64_dma_phys_limit))
481 else if (!xen_swiotlb_detect())
482 swiotlb_force = SWIOTLB_NO_FORCE;
484 set_max_mapnr(max_pfn - PHYS_PFN_OFFSET);
486 /* this will put all unused low memory onto the freelists */
490 * Check boundaries twice: Some fundamental inconsistencies can be
491 * detected at build time already.
494 BUILD_BUG_ON(TASK_SIZE_32 > DEFAULT_MAP_WINDOW_64);
498 * Selected page table levels should match when derived from
499 * scratch using the virtual address range and page size.
501 BUILD_BUG_ON(ARM64_HW_PGTABLE_LEVELS(CONFIG_ARM64_VA_BITS) !=
502 CONFIG_PGTABLE_LEVELS);
504 if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) {
505 extern int sysctl_overcommit_memory;
507 * On a machine this small we won't get anywhere without
508 * overcommit, so turn it on by default.
510 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
514 void free_initmem(void)
516 free_reserved_area(lm_alias(__init_begin),
517 lm_alias(__init_end),
518 POISON_FREE_INITMEM, "unused kernel");
520 * Unmap the __init region but leave the VM area in place. This
521 * prevents the region from being reused for kernel modules, which
522 * is not supported by kallsyms.
524 vunmap_range((u64)__init_begin, (u64)__init_end);
527 void dump_mem_limit(void)
529 if (memory_limit != PHYS_ADDR_MAX) {
530 pr_emerg("Memory Limit: %llu MB\n", memory_limit >> 20);
532 pr_emerg("Memory Limit: none\n");