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>
35 #include <asm/fixmap.h>
36 #include <asm/kasan.h>
37 #include <asm/kernel-pgtable.h>
38 #include <asm/kvm_host.h>
39 #include <asm/memory.h>
41 #include <asm/sections.h>
42 #include <asm/setup.h>
43 #include <linux/sizes.h>
45 #include <asm/alternative.h>
46 #include <asm/xen/swiotlb-xen.h>
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.
54 s64 memstart_addr __ro_after_init = -1;
55 EXPORT_SYMBOL(memstart_addr);
58 * If the corresponding config options are enabled, we create both ZONE_DMA
59 * and ZONE_DMA32. By default ZONE_DMA covers the 32-bit addressable memory
60 * unless restricted on specific platforms (e.g. 30-bit on Raspberry Pi 4).
61 * In such case, ZONE_DMA32 covers the rest of the 32-bit addressable memory,
62 * otherwise it is empty.
64 phys_addr_t arm64_dma_phys_limit __ro_after_init;
66 #ifdef CONFIG_KEXEC_CORE
68 * reserve_crashkernel() - reserves memory for crash kernel
70 * This function reserves memory area given in "crashkernel=" kernel command
71 * line parameter. The memory reserved is used by dump capture kernel when
72 * primary kernel is crashing.
74 static void __init reserve_crashkernel(void)
76 unsigned long long crash_base, crash_size;
77 unsigned long long crash_max = arm64_dma_phys_limit;
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 /* User specifies base address explicitly. */
90 crash_max = crash_base + crash_size;
92 /* Current arm64 boot protocol requires 2MB alignment */
93 crash_base = memblock_phys_alloc_range(crash_size, SZ_2M,
94 crash_base, crash_max);
96 pr_warn("cannot allocate crashkernel (size:0x%llx)\n",
101 pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n",
102 crash_base, crash_base + crash_size, crash_size >> 20);
104 crashk_res.start = crash_base;
105 crashk_res.end = crash_base + crash_size - 1;
108 static void __init reserve_crashkernel(void)
111 #endif /* CONFIG_KEXEC_CORE */
114 * Return the maximum physical address for a zone accessible by the given bits
115 * limit. If DRAM starts above 32-bit, expand the zone to the maximum
116 * available memory, otherwise cap it at 32-bit.
118 static phys_addr_t __init max_zone_phys(unsigned int zone_bits)
120 phys_addr_t zone_mask = DMA_BIT_MASK(zone_bits);
121 phys_addr_t phys_start = memblock_start_of_DRAM();
123 if (phys_start > U32_MAX)
124 zone_mask = PHYS_ADDR_MAX;
125 else if (phys_start > zone_mask)
128 return min(zone_mask, memblock_end_of_DRAM() - 1) + 1;
131 static void __init zone_sizes_init(unsigned long min, unsigned long max)
133 unsigned long max_zone_pfns[MAX_NR_ZONES] = {0};
134 unsigned int __maybe_unused acpi_zone_dma_bits;
135 unsigned int __maybe_unused dt_zone_dma_bits;
136 phys_addr_t __maybe_unused dma32_phys_limit = max_zone_phys(32);
138 #ifdef CONFIG_ZONE_DMA
139 acpi_zone_dma_bits = fls64(acpi_iort_dma_get_max_cpu_address());
140 dt_zone_dma_bits = fls64(of_dma_get_max_cpu_address(NULL));
141 zone_dma_bits = min3(32U, dt_zone_dma_bits, acpi_zone_dma_bits);
142 arm64_dma_phys_limit = max_zone_phys(zone_dma_bits);
143 max_zone_pfns[ZONE_DMA] = PFN_DOWN(arm64_dma_phys_limit);
145 #ifdef CONFIG_ZONE_DMA32
146 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
147 if (!arm64_dma_phys_limit)
148 arm64_dma_phys_limit = dma32_phys_limit;
150 if (!arm64_dma_phys_limit)
151 arm64_dma_phys_limit = PHYS_MASK + 1;
152 max_zone_pfns[ZONE_NORMAL] = max;
154 free_area_init(max_zone_pfns);
157 int pfn_valid(unsigned long pfn)
159 phys_addr_t addr = PFN_PHYS(pfn);
160 struct mem_section *ms;
163 * Ensure the upper PAGE_SHIFT bits are clear in the
164 * pfn. Else it might lead to false positives when
165 * some of the upper bits are set, but the lower bits
168 if (PHYS_PFN(addr) != pfn)
171 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
174 ms = __pfn_to_section(pfn);
175 if (!valid_section(ms))
179 * ZONE_DEVICE memory does not have the memblock entries.
180 * memblock_is_map_memory() check for ZONE_DEVICE based
181 * addresses will always fail. Even the normal hotplugged
182 * memory will never have MEMBLOCK_NOMAP flag set in their
183 * memblock entries. Skip memblock search for all non early
184 * memory sections covering all of hotplug memory including
185 * both normal and ZONE_DEVICE based.
187 if (!early_section(ms))
188 return pfn_section_valid(ms, pfn);
190 return memblock_is_memory(addr);
192 EXPORT_SYMBOL(pfn_valid);
194 int pfn_is_map_memory(unsigned long pfn)
196 phys_addr_t addr = PFN_PHYS(pfn);
198 /* avoid false positives for bogus PFNs, see comment in pfn_valid() */
199 if (PHYS_PFN(addr) != pfn)
202 return memblock_is_map_memory(addr);
204 EXPORT_SYMBOL(pfn_is_map_memory);
206 static phys_addr_t memory_limit = PHYS_ADDR_MAX;
209 * Limit the memory size that was specified via FDT.
211 static int __init early_mem(char *p)
216 memory_limit = memparse(p, &p) & PAGE_MASK;
217 pr_notice("Memory limited to %lldMB\n", memory_limit >> 20);
221 early_param("mem", early_mem);
223 void __init arm64_memblock_init(void)
225 const s64 linear_region_size = PAGE_END - _PAGE_OFFSET(vabits_actual);
227 /* Remove memory above our supported physical address size */
228 memblock_remove(1ULL << PHYS_MASK_SHIFT, ULLONG_MAX);
231 * Select a suitable value for the base of physical memory.
233 memstart_addr = round_down(memblock_start_of_DRAM(),
234 ARM64_MEMSTART_ALIGN);
236 if ((memblock_end_of_DRAM() - memstart_addr) > linear_region_size)
237 pr_warn("Memory doesn't fit in the linear mapping, VA_BITS too small\n");
240 * Remove the memory that we will not be able to cover with the
241 * linear mapping. Take care not to clip the kernel which may be
244 memblock_remove(max_t(u64, memstart_addr + linear_region_size,
245 __pa_symbol(_end)), ULLONG_MAX);
246 if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {
247 /* ensure that memstart_addr remains sufficiently aligned */
248 memstart_addr = round_up(memblock_end_of_DRAM() - linear_region_size,
249 ARM64_MEMSTART_ALIGN);
250 memblock_remove(0, memstart_addr);
254 * If we are running with a 52-bit kernel VA config on a system that
255 * does not support it, we have to place the available physical
256 * memory in the 48-bit addressable part of the linear region, i.e.,
257 * we have to move it upward. Since memstart_addr represents the
258 * physical address of PAGE_OFFSET, we have to *subtract* from it.
260 if (IS_ENABLED(CONFIG_ARM64_VA_BITS_52) && (vabits_actual != 52))
261 memstart_addr -= _PAGE_OFFSET(48) - _PAGE_OFFSET(52);
264 * Apply the memory limit if it was set. Since the kernel may be loaded
265 * high up in memory, add back the kernel region that must be accessible
266 * via the linear mapping.
268 if (memory_limit != PHYS_ADDR_MAX) {
269 memblock_mem_limit_remove_map(memory_limit);
270 memblock_add(__pa_symbol(_text), (u64)(_end - _text));
273 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
275 * Add back the memory we just removed if it results in the
276 * initrd to become inaccessible via the linear mapping.
277 * Otherwise, this is a no-op
279 u64 base = phys_initrd_start & PAGE_MASK;
280 u64 size = PAGE_ALIGN(phys_initrd_start + phys_initrd_size) - base;
283 * We can only add back the initrd memory if we don't end up
284 * with more memory than we can address via the linear mapping.
285 * It is up to the bootloader to position the kernel and the
286 * initrd reasonably close to each other (i.e., within 32 GB of
287 * each other) so that all granule/#levels combinations can
288 * always access both.
290 if (WARN(base < memblock_start_of_DRAM() ||
291 base + size > memblock_start_of_DRAM() +
293 "initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
294 phys_initrd_size = 0;
296 memblock_remove(base, size); /* clear MEMBLOCK_ flags */
297 memblock_add(base, size);
298 memblock_reserve(base, size);
302 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
303 extern u16 memstart_offset_seed;
304 u64 mmfr0 = read_cpuid(ID_AA64MMFR0_EL1);
305 int parange = cpuid_feature_extract_unsigned_field(
306 mmfr0, ID_AA64MMFR0_PARANGE_SHIFT);
307 s64 range = linear_region_size -
308 BIT(id_aa64mmfr0_parange_to_phys_shift(parange));
311 * If the size of the linear region exceeds, by a sufficient
312 * margin, the size of the region that the physical memory can
313 * span, randomize the linear region as well.
315 if (memstart_offset_seed > 0 && range >= (s64)ARM64_MEMSTART_ALIGN) {
316 range /= ARM64_MEMSTART_ALIGN;
317 memstart_addr -= ARM64_MEMSTART_ALIGN *
318 ((range * memstart_offset_seed) >> 16);
323 * Register the kernel text, kernel data, initrd, and initial
324 * pagetables with memblock.
326 memblock_reserve(__pa_symbol(_stext), _end - _stext);
327 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
328 /* the generic initrd code expects virtual addresses */
329 initrd_start = __phys_to_virt(phys_initrd_start);
330 initrd_end = initrd_start + phys_initrd_size;
333 early_init_fdt_scan_reserved_mem();
335 high_memory = __va(memblock_end_of_DRAM() - 1) + 1;
338 void __init bootmem_init(void)
340 unsigned long min, max;
342 min = PFN_UP(memblock_start_of_DRAM());
343 max = PFN_DOWN(memblock_end_of_DRAM());
345 early_memtest(min << PAGE_SHIFT, max << PAGE_SHIFT);
347 max_pfn = max_low_pfn = max;
353 * must be done after arch_numa_init() which calls numa_init() to
354 * initialize node_online_map that gets used in hugetlb_cma_reserve()
355 * while allocating required CMA size across online nodes.
357 #if defined(CONFIG_HUGETLB_PAGE) && defined(CONFIG_CMA)
358 arm64_hugetlb_cma_reserve();
361 dma_pernuma_cma_reserve();
366 * sparse_init() tries to allocate memory from memblock, so must be
367 * done after the fixed reservations
370 zone_sizes_init(min, max);
373 * Reserve the CMA area after arm64_dma_phys_limit was initialised.
375 dma_contiguous_reserve(arm64_dma_phys_limit);
378 * request_standard_resources() depends on crashkernel's memory being
379 * reserved, so do it here.
381 reserve_crashkernel();
387 * mem_init() marks the free areas in the mem_map and tells us how much memory
388 * is free. This is done after various parts of the system have claimed their
389 * memory after the kernel image.
391 void __init mem_init(void)
393 if (swiotlb_force == SWIOTLB_FORCE ||
394 max_pfn > PFN_DOWN(arm64_dma_phys_limit))
396 else if (!xen_swiotlb_detect())
397 swiotlb_force = SWIOTLB_NO_FORCE;
399 set_max_mapnr(max_pfn - PHYS_PFN_OFFSET);
401 /* this will put all unused low memory onto the freelists */
405 * Check boundaries twice: Some fundamental inconsistencies can be
406 * detected at build time already.
409 BUILD_BUG_ON(TASK_SIZE_32 > DEFAULT_MAP_WINDOW_64);
413 * Selected page table levels should match when derived from
414 * scratch using the virtual address range and page size.
416 BUILD_BUG_ON(ARM64_HW_PGTABLE_LEVELS(CONFIG_ARM64_VA_BITS) !=
417 CONFIG_PGTABLE_LEVELS);
419 if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) {
420 extern int sysctl_overcommit_memory;
422 * On a machine this small we won't get anywhere without
423 * overcommit, so turn it on by default.
425 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
429 void free_initmem(void)
431 free_reserved_area(lm_alias(__init_begin),
432 lm_alias(__init_end),
433 POISON_FREE_INITMEM, "unused kernel");
435 * Unmap the __init region but leave the VM area in place. This
436 * prevents the region from being reused for kernel modules, which
437 * is not supported by kallsyms.
439 vunmap_range((u64)__init_begin, (u64)__init_end);
442 void dump_mem_limit(void)
444 if (memory_limit != PHYS_ADDR_MAX) {
445 pr_emerg("Memory Limit: %llu MB\n", memory_limit >> 20);
447 pr_emerg("Memory Limit: none\n");