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;
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 if (!IS_ENABLED(CONFIG_KEXEC_CORE))
83 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
84 &crash_size, &crash_base);
85 /* no crashkernel= or invalid value specified */
86 if (ret || !crash_size)
89 crash_size = PAGE_ALIGN(crash_size);
91 /* User specifies base address explicitly. */
93 crash_max = crash_base + crash_size;
95 /* Current arm64 boot protocol requires 2MB alignment */
96 crash_base = memblock_phys_alloc_range(crash_size, SZ_2M,
97 crash_base, crash_max);
99 pr_warn("cannot allocate crashkernel (size:0x%llx)\n",
104 pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n",
105 crash_base, crash_base + crash_size, crash_size >> 20);
108 * The crashkernel memory will be removed from the kernel linear
109 * map. Inform kmemleak so that it won't try to access it.
111 kmemleak_ignore_phys(crash_base);
112 crashk_res.start = crash_base;
113 crashk_res.end = crash_base + crash_size - 1;
117 * Return the maximum physical address for a zone accessible by the given bits
118 * limit. If DRAM starts above 32-bit, expand the zone to the maximum
119 * available memory, otherwise cap it at 32-bit.
121 static phys_addr_t __init max_zone_phys(unsigned int zone_bits)
123 phys_addr_t zone_mask = DMA_BIT_MASK(zone_bits);
124 phys_addr_t phys_start = memblock_start_of_DRAM();
126 if (phys_start > U32_MAX)
127 zone_mask = PHYS_ADDR_MAX;
128 else if (phys_start > zone_mask)
131 return min(zone_mask, memblock_end_of_DRAM() - 1) + 1;
134 static void __init zone_sizes_init(unsigned long min, unsigned long max)
136 unsigned long max_zone_pfns[MAX_NR_ZONES] = {0};
137 unsigned int __maybe_unused acpi_zone_dma_bits;
138 unsigned int __maybe_unused dt_zone_dma_bits;
139 phys_addr_t __maybe_unused dma32_phys_limit = max_zone_phys(32);
141 #ifdef CONFIG_ZONE_DMA
142 acpi_zone_dma_bits = fls64(acpi_iort_dma_get_max_cpu_address());
143 dt_zone_dma_bits = fls64(of_dma_get_max_cpu_address(NULL));
144 zone_dma_bits = min3(32U, dt_zone_dma_bits, acpi_zone_dma_bits);
145 arm64_dma_phys_limit = max_zone_phys(zone_dma_bits);
146 max_zone_pfns[ZONE_DMA] = PFN_DOWN(arm64_dma_phys_limit);
148 #ifdef CONFIG_ZONE_DMA32
149 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
150 if (!arm64_dma_phys_limit)
151 arm64_dma_phys_limit = dma32_phys_limit;
153 if (!arm64_dma_phys_limit)
154 arm64_dma_phys_limit = PHYS_MASK + 1;
155 max_zone_pfns[ZONE_NORMAL] = max;
157 free_area_init(max_zone_pfns);
160 int pfn_is_map_memory(unsigned long pfn)
162 phys_addr_t addr = PFN_PHYS(pfn);
164 /* avoid false positives for bogus PFNs, see comment in pfn_valid() */
165 if (PHYS_PFN(addr) != pfn)
168 return memblock_is_map_memory(addr);
170 EXPORT_SYMBOL(pfn_is_map_memory);
172 static phys_addr_t memory_limit __ro_after_init = PHYS_ADDR_MAX;
175 * Limit the memory size that was specified via FDT.
177 static int __init early_mem(char *p)
182 memory_limit = memparse(p, &p) & PAGE_MASK;
183 pr_notice("Memory limited to %lldMB\n", memory_limit >> 20);
187 early_param("mem", early_mem);
189 void __init arm64_memblock_init(void)
191 s64 linear_region_size = PAGE_END - _PAGE_OFFSET(vabits_actual);
194 * Corner case: 52-bit VA capable systems running KVM in nVHE mode may
195 * be limited in their ability to support a linear map that exceeds 51
196 * bits of VA space, depending on the placement of the ID map. Given
197 * that the placement of the ID map may be randomized, let's simply
198 * limit the kernel's linear map to 51 bits as well if we detect this
201 if (IS_ENABLED(CONFIG_KVM) && vabits_actual == 52 &&
202 is_hyp_mode_available() && !is_kernel_in_hyp_mode()) {
203 pr_info("Capping linear region to 51 bits for KVM in nVHE mode on LVA capable hardware.\n");
204 linear_region_size = min_t(u64, linear_region_size, BIT(51));
207 /* Remove memory above our supported physical address size */
208 memblock_remove(1ULL << PHYS_MASK_SHIFT, ULLONG_MAX);
211 * Select a suitable value for the base of physical memory.
213 memstart_addr = round_down(memblock_start_of_DRAM(),
214 ARM64_MEMSTART_ALIGN);
216 if ((memblock_end_of_DRAM() - memstart_addr) > linear_region_size)
217 pr_warn("Memory doesn't fit in the linear mapping, VA_BITS too small\n");
220 * Remove the memory that we will not be able to cover with the
221 * linear mapping. Take care not to clip the kernel which may be
224 memblock_remove(max_t(u64, memstart_addr + linear_region_size,
225 __pa_symbol(_end)), ULLONG_MAX);
226 if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {
227 /* ensure that memstart_addr remains sufficiently aligned */
228 memstart_addr = round_up(memblock_end_of_DRAM() - linear_region_size,
229 ARM64_MEMSTART_ALIGN);
230 memblock_remove(0, memstart_addr);
234 * If we are running with a 52-bit kernel VA config on a system that
235 * does not support it, we have to place the available physical
236 * memory in the 48-bit addressable part of the linear region, i.e.,
237 * we have to move it upward. Since memstart_addr represents the
238 * physical address of PAGE_OFFSET, we have to *subtract* from it.
240 if (IS_ENABLED(CONFIG_ARM64_VA_BITS_52) && (vabits_actual != 52))
241 memstart_addr -= _PAGE_OFFSET(48) - _PAGE_OFFSET(52);
244 * Apply the memory limit if it was set. Since the kernel may be loaded
245 * high up in memory, add back the kernel region that must be accessible
246 * via the linear mapping.
248 if (memory_limit != PHYS_ADDR_MAX) {
249 memblock_mem_limit_remove_map(memory_limit);
250 memblock_add(__pa_symbol(_text), (u64)(_end - _text));
253 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
255 * Add back the memory we just removed if it results in the
256 * initrd to become inaccessible via the linear mapping.
257 * Otherwise, this is a no-op
259 u64 base = phys_initrd_start & PAGE_MASK;
260 u64 size = PAGE_ALIGN(phys_initrd_start + phys_initrd_size) - base;
263 * We can only add back the initrd memory if we don't end up
264 * with more memory than we can address via the linear mapping.
265 * It is up to the bootloader to position the kernel and the
266 * initrd reasonably close to each other (i.e., within 32 GB of
267 * each other) so that all granule/#levels combinations can
268 * always access both.
270 if (WARN(base < memblock_start_of_DRAM() ||
271 base + size > memblock_start_of_DRAM() +
273 "initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
274 phys_initrd_size = 0;
276 memblock_remove(base, size); /* clear MEMBLOCK_ flags */
277 memblock_add(base, size);
278 memblock_reserve(base, size);
282 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
283 extern u16 memstart_offset_seed;
284 u64 mmfr0 = read_cpuid(ID_AA64MMFR0_EL1);
285 int parange = cpuid_feature_extract_unsigned_field(
286 mmfr0, ID_AA64MMFR0_PARANGE_SHIFT);
287 s64 range = linear_region_size -
288 BIT(id_aa64mmfr0_parange_to_phys_shift(parange));
291 * If the size of the linear region exceeds, by a sufficient
292 * margin, the size of the region that the physical memory can
293 * span, randomize the linear region as well.
295 if (memstart_offset_seed > 0 && range >= (s64)ARM64_MEMSTART_ALIGN) {
296 range /= ARM64_MEMSTART_ALIGN;
297 memstart_addr -= ARM64_MEMSTART_ALIGN *
298 ((range * memstart_offset_seed) >> 16);
303 * Register the kernel text, kernel data, initrd, and initial
304 * pagetables with memblock.
306 memblock_reserve(__pa_symbol(_stext), _end - _stext);
307 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
308 /* the generic initrd code expects virtual addresses */
309 initrd_start = __phys_to_virt(phys_initrd_start);
310 initrd_end = initrd_start + phys_initrd_size;
313 early_init_fdt_scan_reserved_mem();
315 high_memory = __va(memblock_end_of_DRAM() - 1) + 1;
318 void __init bootmem_init(void)
320 unsigned long min, max;
322 min = PFN_UP(memblock_start_of_DRAM());
323 max = PFN_DOWN(memblock_end_of_DRAM());
325 early_memtest(min << PAGE_SHIFT, max << PAGE_SHIFT);
327 max_pfn = max_low_pfn = max;
333 * must be done after arch_numa_init() which calls numa_init() to
334 * initialize node_online_map that gets used in hugetlb_cma_reserve()
335 * while allocating required CMA size across online nodes.
337 #if defined(CONFIG_HUGETLB_PAGE) && defined(CONFIG_CMA)
338 arm64_hugetlb_cma_reserve();
341 dma_pernuma_cma_reserve();
346 * sparse_init() tries to allocate memory from memblock, so must be
347 * done after the fixed reservations
350 zone_sizes_init(min, max);
353 * Reserve the CMA area after arm64_dma_phys_limit was initialised.
355 dma_contiguous_reserve(arm64_dma_phys_limit);
358 * request_standard_resources() depends on crashkernel's memory being
359 * reserved, so do it here.
361 reserve_crashkernel();
367 * mem_init() marks the free areas in the mem_map and tells us how much memory
368 * is free. This is done after various parts of the system have claimed their
369 * memory after the kernel image.
371 void __init mem_init(void)
373 if (swiotlb_force == SWIOTLB_FORCE ||
374 max_pfn > PFN_DOWN(arm64_dma_phys_limit))
376 else if (!xen_swiotlb_detect())
377 swiotlb_force = SWIOTLB_NO_FORCE;
379 /* this will put all unused low memory onto the freelists */
383 * Check boundaries twice: Some fundamental inconsistencies can be
384 * detected at build time already.
387 BUILD_BUG_ON(TASK_SIZE_32 > DEFAULT_MAP_WINDOW_64);
391 * Selected page table levels should match when derived from
392 * scratch using the virtual address range and page size.
394 BUILD_BUG_ON(ARM64_HW_PGTABLE_LEVELS(CONFIG_ARM64_VA_BITS) !=
395 CONFIG_PGTABLE_LEVELS);
397 if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) {
398 extern int sysctl_overcommit_memory;
400 * On a machine this small we won't get anywhere without
401 * overcommit, so turn it on by default.
403 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
407 void free_initmem(void)
409 free_reserved_area(lm_alias(__init_begin),
410 lm_alias(__init_end),
411 POISON_FREE_INITMEM, "unused kernel");
413 * Unmap the __init region but leave the VM area in place. This
414 * prevents the region from being reused for kernel modules, which
415 * is not supported by kallsyms.
417 vunmap_range((u64)__init_begin, (u64)__init_end);
420 void dump_mem_limit(void)
422 if (memory_limit != PHYS_ADDR_MAX) {
423 pr_emerg("Memory Limit: %llu MB\n", memory_limit >> 20);
425 pr_emerg("Memory Limit: none\n");