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 * Memory reservation for crash kernel either done early or deferred
66 * depending on DMA memory zones configs (ZONE_DMA) --
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.
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.
84 * Note: Page-granularity mapppings are necessary for crash kernel memory
85 * range for shrinking its size via /sys/kernel/kexec_crash_size interface.
87 #if IS_ENABLED(CONFIG_ZONE_DMA) || IS_ENABLED(CONFIG_ZONE_DMA32)
88 phys_addr_t __ro_after_init arm64_dma_phys_limit;
90 phys_addr_t __ro_after_init arm64_dma_phys_limit = PHYS_MASK + 1;
93 #ifdef CONFIG_KEXEC_CORE
95 * reserve_crashkernel() - reserves memory for crash kernel
97 * This function reserves memory area given in "crashkernel=" kernel command
98 * line parameter. The memory reserved is used by dump capture kernel when
99 * primary kernel is crashing.
101 static void __init reserve_crashkernel(void)
103 unsigned long long crash_base, crash_size;
104 unsigned long long crash_max = arm64_dma_phys_limit;
107 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
108 &crash_size, &crash_base);
109 /* no crashkernel= or invalid value specified */
110 if (ret || !crash_size)
113 crash_size = PAGE_ALIGN(crash_size);
115 /* User specifies base address explicitly. */
117 crash_max = crash_base + crash_size;
119 /* Current arm64 boot protocol requires 2MB alignment */
120 crash_base = memblock_phys_alloc_range(crash_size, SZ_2M,
121 crash_base, crash_max);
123 pr_warn("cannot allocate crashkernel (size:0x%llx)\n",
128 pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n",
129 crash_base, crash_base + crash_size, crash_size >> 20);
132 * The crashkernel memory will be removed from the kernel linear
133 * map. Inform kmemleak so that it won't try to access it.
135 kmemleak_ignore_phys(crash_base);
136 crashk_res.start = crash_base;
137 crashk_res.end = crash_base + crash_size - 1;
140 static void __init reserve_crashkernel(void)
143 #endif /* CONFIG_KEXEC_CORE */
146 * Return the maximum physical address for a zone accessible by the given bits
147 * limit. If DRAM starts above 32-bit, expand the zone to the maximum
148 * available memory, otherwise cap it at 32-bit.
150 static phys_addr_t __init max_zone_phys(unsigned int zone_bits)
152 phys_addr_t zone_mask = DMA_BIT_MASK(zone_bits);
153 phys_addr_t phys_start = memblock_start_of_DRAM();
155 if (phys_start > U32_MAX)
156 zone_mask = PHYS_ADDR_MAX;
157 else if (phys_start > zone_mask)
160 return min(zone_mask, memblock_end_of_DRAM() - 1) + 1;
163 static void __init zone_sizes_init(unsigned long min, unsigned long max)
165 unsigned long max_zone_pfns[MAX_NR_ZONES] = {0};
166 unsigned int __maybe_unused acpi_zone_dma_bits;
167 unsigned int __maybe_unused dt_zone_dma_bits;
168 phys_addr_t __maybe_unused dma32_phys_limit = max_zone_phys(32);
170 #ifdef CONFIG_ZONE_DMA
171 acpi_zone_dma_bits = fls64(acpi_iort_dma_get_max_cpu_address());
172 dt_zone_dma_bits = fls64(of_dma_get_max_cpu_address(NULL));
173 zone_dma_bits = min3(32U, dt_zone_dma_bits, acpi_zone_dma_bits);
174 arm64_dma_phys_limit = max_zone_phys(zone_dma_bits);
175 max_zone_pfns[ZONE_DMA] = PFN_DOWN(arm64_dma_phys_limit);
177 #ifdef CONFIG_ZONE_DMA32
178 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
179 if (!arm64_dma_phys_limit)
180 arm64_dma_phys_limit = dma32_phys_limit;
182 max_zone_pfns[ZONE_NORMAL] = max;
184 free_area_init(max_zone_pfns);
187 int pfn_is_map_memory(unsigned long pfn)
189 phys_addr_t addr = PFN_PHYS(pfn);
191 /* avoid false positives for bogus PFNs, see comment in pfn_valid() */
192 if (PHYS_PFN(addr) != pfn)
195 return memblock_is_map_memory(addr);
197 EXPORT_SYMBOL(pfn_is_map_memory);
199 static phys_addr_t memory_limit = PHYS_ADDR_MAX;
202 * Limit the memory size that was specified via FDT.
204 static int __init early_mem(char *p)
209 memory_limit = memparse(p, &p) & PAGE_MASK;
210 pr_notice("Memory limited to %lldMB\n", memory_limit >> 20);
214 early_param("mem", early_mem);
216 void __init arm64_memblock_init(void)
218 s64 linear_region_size = PAGE_END - _PAGE_OFFSET(vabits_actual);
221 * Corner case: 52-bit VA capable systems running KVM in nVHE mode may
222 * be limited in their ability to support a linear map that exceeds 51
223 * bits of VA space, depending on the placement of the ID map. Given
224 * that the placement of the ID map may be randomized, let's simply
225 * limit the kernel's linear map to 51 bits as well if we detect this
228 if (IS_ENABLED(CONFIG_KVM) && vabits_actual == 52 &&
229 is_hyp_mode_available() && !is_kernel_in_hyp_mode()) {
230 pr_info("Capping linear region to 51 bits for KVM in nVHE mode on LVA capable hardware.\n");
231 linear_region_size = min_t(u64, linear_region_size, BIT(51));
234 /* Remove memory above our supported physical address size */
235 memblock_remove(1ULL << PHYS_MASK_SHIFT, ULLONG_MAX);
238 * Select a suitable value for the base of physical memory.
240 memstart_addr = round_down(memblock_start_of_DRAM(),
241 ARM64_MEMSTART_ALIGN);
243 if ((memblock_end_of_DRAM() - memstart_addr) > linear_region_size)
244 pr_warn("Memory doesn't fit in the linear mapping, VA_BITS too small\n");
247 * Remove the memory that we will not be able to cover with the
248 * linear mapping. Take care not to clip the kernel which may be
251 memblock_remove(max_t(u64, memstart_addr + linear_region_size,
252 __pa_symbol(_end)), ULLONG_MAX);
253 if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {
254 /* ensure that memstart_addr remains sufficiently aligned */
255 memstart_addr = round_up(memblock_end_of_DRAM() - linear_region_size,
256 ARM64_MEMSTART_ALIGN);
257 memblock_remove(0, memstart_addr);
261 * If we are running with a 52-bit kernel VA config on a system that
262 * does not support it, we have to place the available physical
263 * memory in the 48-bit addressable part of the linear region, i.e.,
264 * we have to move it upward. Since memstart_addr represents the
265 * physical address of PAGE_OFFSET, we have to *subtract* from it.
267 if (IS_ENABLED(CONFIG_ARM64_VA_BITS_52) && (vabits_actual != 52))
268 memstart_addr -= _PAGE_OFFSET(48) - _PAGE_OFFSET(52);
271 * Apply the memory limit if it was set. Since the kernel may be loaded
272 * high up in memory, add back the kernel region that must be accessible
273 * via the linear mapping.
275 if (memory_limit != PHYS_ADDR_MAX) {
276 memblock_mem_limit_remove_map(memory_limit);
277 memblock_add(__pa_symbol(_text), (u64)(_end - _text));
280 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
282 * Add back the memory we just removed if it results in the
283 * initrd to become inaccessible via the linear mapping.
284 * Otherwise, this is a no-op
286 u64 base = phys_initrd_start & PAGE_MASK;
287 u64 size = PAGE_ALIGN(phys_initrd_start + phys_initrd_size) - base;
290 * We can only add back the initrd memory if we don't end up
291 * with more memory than we can address via the linear mapping.
292 * It is up to the bootloader to position the kernel and the
293 * initrd reasonably close to each other (i.e., within 32 GB of
294 * each other) so that all granule/#levels combinations can
295 * always access both.
297 if (WARN(base < memblock_start_of_DRAM() ||
298 base + size > memblock_start_of_DRAM() +
300 "initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
301 phys_initrd_size = 0;
303 memblock_remove(base, size); /* clear MEMBLOCK_ flags */
304 memblock_add(base, size);
305 memblock_reserve(base, size);
309 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
310 extern u16 memstart_offset_seed;
311 u64 mmfr0 = read_cpuid(ID_AA64MMFR0_EL1);
312 int parange = cpuid_feature_extract_unsigned_field(
313 mmfr0, ID_AA64MMFR0_PARANGE_SHIFT);
314 s64 range = linear_region_size -
315 BIT(id_aa64mmfr0_parange_to_phys_shift(parange));
318 * If the size of the linear region exceeds, by a sufficient
319 * margin, the size of the region that the physical memory can
320 * span, randomize the linear region as well.
322 if (memstart_offset_seed > 0 && range >= (s64)ARM64_MEMSTART_ALIGN) {
323 range /= ARM64_MEMSTART_ALIGN;
324 memstart_addr -= ARM64_MEMSTART_ALIGN *
325 ((range * memstart_offset_seed) >> 16);
330 * Register the kernel text, kernel data, initrd, and initial
331 * pagetables with memblock.
333 memblock_reserve(__pa_symbol(_stext), _end - _stext);
334 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
335 /* the generic initrd code expects virtual addresses */
336 initrd_start = __phys_to_virt(phys_initrd_start);
337 initrd_end = initrd_start + phys_initrd_size;
340 early_init_fdt_scan_reserved_mem();
342 if (!IS_ENABLED(CONFIG_ZONE_DMA) && !IS_ENABLED(CONFIG_ZONE_DMA32))
343 reserve_crashkernel();
345 high_memory = __va(memblock_end_of_DRAM() - 1) + 1;
348 void __init bootmem_init(void)
350 unsigned long min, max;
352 min = PFN_UP(memblock_start_of_DRAM());
353 max = PFN_DOWN(memblock_end_of_DRAM());
355 early_memtest(min << PAGE_SHIFT, max << PAGE_SHIFT);
357 max_pfn = max_low_pfn = max;
363 * must be done after arch_numa_init() which calls numa_init() to
364 * initialize node_online_map that gets used in hugetlb_cma_reserve()
365 * while allocating required CMA size across online nodes.
367 #if defined(CONFIG_HUGETLB_PAGE) && defined(CONFIG_CMA)
368 arm64_hugetlb_cma_reserve();
371 dma_pernuma_cma_reserve();
376 * sparse_init() tries to allocate memory from memblock, so must be
377 * done after the fixed reservations
380 zone_sizes_init(min, max);
383 * Reserve the CMA area after arm64_dma_phys_limit was initialised.
385 dma_contiguous_reserve(arm64_dma_phys_limit);
388 * request_standard_resources() depends on crashkernel's memory being
389 * reserved, so do it here.
391 if (IS_ENABLED(CONFIG_ZONE_DMA) || IS_ENABLED(CONFIG_ZONE_DMA32))
392 reserve_crashkernel();
398 * mem_init() marks the free areas in the mem_map and tells us how much memory
399 * is free. This is done after various parts of the system have claimed their
400 * memory after the kernel image.
402 void __init mem_init(void)
404 if (swiotlb_force == SWIOTLB_FORCE ||
405 max_pfn > PFN_DOWN(arm64_dma_phys_limit))
407 else if (!xen_swiotlb_detect())
408 swiotlb_force = SWIOTLB_NO_FORCE;
410 set_max_mapnr(max_pfn - PHYS_PFN_OFFSET);
412 /* this will put all unused low memory onto the freelists */
416 * Check boundaries twice: Some fundamental inconsistencies can be
417 * detected at build time already.
420 BUILD_BUG_ON(TASK_SIZE_32 > DEFAULT_MAP_WINDOW_64);
424 * Selected page table levels should match when derived from
425 * scratch using the virtual address range and page size.
427 BUILD_BUG_ON(ARM64_HW_PGTABLE_LEVELS(CONFIG_ARM64_VA_BITS) !=
428 CONFIG_PGTABLE_LEVELS);
430 if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) {
431 extern int sysctl_overcommit_memory;
433 * On a machine this small we won't get anywhere without
434 * overcommit, so turn it on by default.
436 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
440 void free_initmem(void)
442 free_reserved_area(lm_alias(__init_begin),
443 lm_alias(__init_end),
444 POISON_FREE_INITMEM, "unused kernel");
446 * Unmap the __init region but leave the VM area in place. This
447 * prevents the region from being reused for kernel modules, which
448 * is not supported by kallsyms.
450 vunmap_range((u64)__init_begin, (u64)__init_end);
453 void dump_mem_limit(void)
455 if (memory_limit != PHYS_ADDR_MAX) {
456 pr_emerg("Memory Limit: %llu MB\n", memory_limit >> 20);
458 pr_emerg("Memory Limit: none\n");