xref: /linux/arch/arm64/mm/init.c (revision fbd126f5a658b92c7f6af986a6d89cf5e5693268)
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 phys_addr_t __ro_after_init arm64_dma_phys_limit;
66 
67 /*
68  * To make optimal use of block mappings when laying out the linear
69  * mapping, round down the base of physical memory to a size that can
70  * be mapped efficiently, i.e., either PUD_SIZE (4k granule) or PMD_SIZE
71  * (64k granule), or a multiple that can be mapped using contiguous bits
72  * in the page tables: 32 * PMD_SIZE (16k granule)
73  */
74 #if defined(CONFIG_ARM64_4K_PAGES)
75 #define ARM64_MEMSTART_SHIFT		PUD_SHIFT
76 #elif defined(CONFIG_ARM64_16K_PAGES)
77 #define ARM64_MEMSTART_SHIFT		CONT_PMD_SHIFT
78 #else
79 #define ARM64_MEMSTART_SHIFT		PMD_SHIFT
80 #endif
81 
82 /*
83  * sparsemem vmemmap imposes an additional requirement on the alignment of
84  * memstart_addr, due to the fact that the base of the vmemmap region
85  * has a direct correspondence, and needs to appear sufficiently aligned
86  * in the virtual address space.
87  */
88 #if ARM64_MEMSTART_SHIFT < SECTION_SIZE_BITS
89 #define ARM64_MEMSTART_ALIGN	(1UL << SECTION_SIZE_BITS)
90 #else
91 #define ARM64_MEMSTART_ALIGN	(1UL << ARM64_MEMSTART_SHIFT)
92 #endif
93 
94 static void __init arch_reserve_crashkernel(void)
95 {
96 	unsigned long long low_size = 0;
97 	unsigned long long crash_base, crash_size;
98 	char *cmdline = boot_command_line;
99 	bool high = false;
100 	int ret;
101 
102 	if (!IS_ENABLED(CONFIG_KEXEC_CORE))
103 		return;
104 
105 	ret = parse_crashkernel(cmdline, memblock_phys_mem_size(),
106 				&crash_size, &crash_base,
107 				&low_size, &high);
108 	if (ret)
109 		return;
110 
111 	reserve_crashkernel_generic(cmdline, crash_size, crash_base,
112 				    low_size, high);
113 }
114 
115 /*
116  * Return the maximum physical address for a zone accessible by the given bits
117  * limit. If DRAM starts above 32-bit, expand the zone to the maximum
118  * available memory, otherwise cap it at 32-bit.
119  */
120 static phys_addr_t __init max_zone_phys(unsigned int zone_bits)
121 {
122 	phys_addr_t zone_mask = DMA_BIT_MASK(zone_bits);
123 	phys_addr_t phys_start = memblock_start_of_DRAM();
124 
125 	if (phys_start > U32_MAX)
126 		zone_mask = PHYS_ADDR_MAX;
127 	else if (phys_start > zone_mask)
128 		zone_mask = U32_MAX;
129 
130 	return min(zone_mask, memblock_end_of_DRAM() - 1) + 1;
131 }
132 
133 static void __init zone_sizes_init(void)
134 {
135 	unsigned long max_zone_pfns[MAX_NR_ZONES]  = {0};
136 	unsigned int __maybe_unused acpi_zone_dma_bits;
137 	unsigned int __maybe_unused dt_zone_dma_bits;
138 	phys_addr_t __maybe_unused dma32_phys_limit = max_zone_phys(32);
139 
140 #ifdef CONFIG_ZONE_DMA
141 	acpi_zone_dma_bits = fls64(acpi_iort_dma_get_max_cpu_address());
142 	dt_zone_dma_bits = fls64(of_dma_get_max_cpu_address(NULL));
143 	zone_dma_bits = min3(32U, dt_zone_dma_bits, acpi_zone_dma_bits);
144 	arm64_dma_phys_limit = max_zone_phys(zone_dma_bits);
145 	max_zone_pfns[ZONE_DMA] = PFN_DOWN(arm64_dma_phys_limit);
146 #endif
147 #ifdef CONFIG_ZONE_DMA32
148 	max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
149 	if (!arm64_dma_phys_limit)
150 		arm64_dma_phys_limit = dma32_phys_limit;
151 #endif
152 	if (!arm64_dma_phys_limit)
153 		arm64_dma_phys_limit = PHYS_MASK + 1;
154 	max_zone_pfns[ZONE_NORMAL] = max_pfn;
155 
156 	free_area_init(max_zone_pfns);
157 }
158 
159 int pfn_is_map_memory(unsigned long pfn)
160 {
161 	phys_addr_t addr = PFN_PHYS(pfn);
162 
163 	/* avoid false positives for bogus PFNs, see comment in pfn_valid() */
164 	if (PHYS_PFN(addr) != pfn)
165 		return 0;
166 
167 	return memblock_is_map_memory(addr);
168 }
169 EXPORT_SYMBOL(pfn_is_map_memory);
170 
171 static phys_addr_t memory_limit __ro_after_init = PHYS_ADDR_MAX;
172 
173 /*
174  * Limit the memory size that was specified via FDT.
175  */
176 static int __init early_mem(char *p)
177 {
178 	if (!p)
179 		return 1;
180 
181 	memory_limit = memparse(p, &p) & PAGE_MASK;
182 	pr_notice("Memory limited to %lldMB\n", memory_limit >> 20);
183 
184 	return 0;
185 }
186 early_param("mem", early_mem);
187 
188 void __init arm64_memblock_init(void)
189 {
190 	s64 linear_region_size = PAGE_END - _PAGE_OFFSET(vabits_actual);
191 
192 	/*
193 	 * Corner case: 52-bit VA capable systems running KVM in nVHE mode may
194 	 * be limited in their ability to support a linear map that exceeds 51
195 	 * bits of VA space, depending on the placement of the ID map. Given
196 	 * that the placement of the ID map may be randomized, let's simply
197 	 * limit the kernel's linear map to 51 bits as well if we detect this
198 	 * configuration.
199 	 */
200 	if (IS_ENABLED(CONFIG_KVM) && vabits_actual == 52 &&
201 	    is_hyp_mode_available() && !is_kernel_in_hyp_mode()) {
202 		pr_info("Capping linear region to 51 bits for KVM in nVHE mode on LVA capable hardware.\n");
203 		linear_region_size = min_t(u64, linear_region_size, BIT(51));
204 	}
205 
206 	/* Remove memory above our supported physical address size */
207 	memblock_remove(1ULL << PHYS_MASK_SHIFT, ULLONG_MAX);
208 
209 	/*
210 	 * Select a suitable value for the base of physical memory.
211 	 */
212 	memstart_addr = round_down(memblock_start_of_DRAM(),
213 				   ARM64_MEMSTART_ALIGN);
214 
215 	if ((memblock_end_of_DRAM() - memstart_addr) > linear_region_size)
216 		pr_warn("Memory doesn't fit in the linear mapping, VA_BITS too small\n");
217 
218 	/*
219 	 * Remove the memory that we will not be able to cover with the
220 	 * linear mapping. Take care not to clip the kernel which may be
221 	 * high in memory.
222 	 */
223 	memblock_remove(max_t(u64, memstart_addr + linear_region_size,
224 			__pa_symbol(_end)), ULLONG_MAX);
225 	if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {
226 		/* ensure that memstart_addr remains sufficiently aligned */
227 		memstart_addr = round_up(memblock_end_of_DRAM() - linear_region_size,
228 					 ARM64_MEMSTART_ALIGN);
229 		memblock_remove(0, memstart_addr);
230 	}
231 
232 	/*
233 	 * If we are running with a 52-bit kernel VA config on a system that
234 	 * does not support it, we have to place the available physical
235 	 * memory in the 48-bit addressable part of the linear region, i.e.,
236 	 * we have to move it upward. Since memstart_addr represents the
237 	 * physical address of PAGE_OFFSET, we have to *subtract* from it.
238 	 */
239 	if (IS_ENABLED(CONFIG_ARM64_VA_BITS_52) && (vabits_actual != 52))
240 		memstart_addr -= _PAGE_OFFSET(48) - _PAGE_OFFSET(52);
241 
242 	/*
243 	 * Apply the memory limit if it was set. Since the kernel may be loaded
244 	 * high up in memory, add back the kernel region that must be accessible
245 	 * via the linear mapping.
246 	 */
247 	if (memory_limit != PHYS_ADDR_MAX) {
248 		memblock_mem_limit_remove_map(memory_limit);
249 		memblock_add(__pa_symbol(_text), (u64)(_end - _text));
250 	}
251 
252 	if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
253 		/*
254 		 * Add back the memory we just removed if it results in the
255 		 * initrd to become inaccessible via the linear mapping.
256 		 * Otherwise, this is a no-op
257 		 */
258 		u64 base = phys_initrd_start & PAGE_MASK;
259 		u64 size = PAGE_ALIGN(phys_initrd_start + phys_initrd_size) - base;
260 
261 		/*
262 		 * We can only add back the initrd memory if we don't end up
263 		 * with more memory than we can address via the linear mapping.
264 		 * It is up to the bootloader to position the kernel and the
265 		 * initrd reasonably close to each other (i.e., within 32 GB of
266 		 * each other) so that all granule/#levels combinations can
267 		 * always access both.
268 		 */
269 		if (WARN(base < memblock_start_of_DRAM() ||
270 			 base + size > memblock_start_of_DRAM() +
271 				       linear_region_size,
272 			"initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
273 			phys_initrd_size = 0;
274 		} else {
275 			memblock_add(base, size);
276 			memblock_clear_nomap(base, size);
277 			memblock_reserve(base, size);
278 		}
279 	}
280 
281 	if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
282 		extern u16 memstart_offset_seed;
283 		u64 mmfr0 = read_cpuid(ID_AA64MMFR0_EL1);
284 		int parange = cpuid_feature_extract_unsigned_field(
285 					mmfr0, ID_AA64MMFR0_EL1_PARANGE_SHIFT);
286 		s64 range = linear_region_size -
287 			    BIT(id_aa64mmfr0_parange_to_phys_shift(parange));
288 
289 		/*
290 		 * If the size of the linear region exceeds, by a sufficient
291 		 * margin, the size of the region that the physical memory can
292 		 * span, randomize the linear region as well.
293 		 */
294 		if (memstart_offset_seed > 0 && range >= (s64)ARM64_MEMSTART_ALIGN) {
295 			range /= ARM64_MEMSTART_ALIGN;
296 			memstart_addr -= ARM64_MEMSTART_ALIGN *
297 					 ((range * memstart_offset_seed) >> 16);
298 		}
299 	}
300 
301 	/*
302 	 * Register the kernel text, kernel data, initrd, and initial
303 	 * pagetables with memblock.
304 	 */
305 	memblock_reserve(__pa_symbol(_stext), _end - _stext);
306 	if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
307 		/* the generic initrd code expects virtual addresses */
308 		initrd_start = __phys_to_virt(phys_initrd_start);
309 		initrd_end = initrd_start + phys_initrd_size;
310 	}
311 
312 	early_init_fdt_scan_reserved_mem();
313 
314 	high_memory = __va(memblock_end_of_DRAM() - 1) + 1;
315 }
316 
317 void __init bootmem_init(void)
318 {
319 	unsigned long min, max;
320 
321 	min = PFN_UP(memblock_start_of_DRAM());
322 	max = PFN_DOWN(memblock_end_of_DRAM());
323 
324 	early_memtest(min << PAGE_SHIFT, max << PAGE_SHIFT);
325 
326 	max_pfn = max_low_pfn = max;
327 	min_low_pfn = min;
328 
329 	arch_numa_init();
330 
331 	/*
332 	 * must be done after arch_numa_init() which calls numa_init() to
333 	 * initialize node_online_map that gets used in hugetlb_cma_reserve()
334 	 * while allocating required CMA size across online nodes.
335 	 */
336 #if defined(CONFIG_HUGETLB_PAGE) && defined(CONFIG_CMA)
337 	arm64_hugetlb_cma_reserve();
338 #endif
339 
340 	kvm_hyp_reserve();
341 
342 	/*
343 	 * sparse_init() tries to allocate memory from memblock, so must be
344 	 * done after the fixed reservations
345 	 */
346 	sparse_init();
347 	zone_sizes_init();
348 
349 	/*
350 	 * Reserve the CMA area after arm64_dma_phys_limit was initialised.
351 	 */
352 	dma_contiguous_reserve(arm64_dma_phys_limit);
353 
354 	/*
355 	 * request_standard_resources() depends on crashkernel's memory being
356 	 * reserved, so do it here.
357 	 */
358 	arch_reserve_crashkernel();
359 
360 	memblock_dump_all();
361 }
362 
363 /*
364  * mem_init() marks the free areas in the mem_map and tells us how much memory
365  * is free.  This is done after various parts of the system have claimed their
366  * memory after the kernel image.
367  */
368 void __init mem_init(void)
369 {
370 	bool swiotlb = max_pfn > PFN_DOWN(arm64_dma_phys_limit);
371 
372 	if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC))
373 		swiotlb = true;
374 
375 	swiotlb_init(swiotlb, SWIOTLB_VERBOSE);
376 
377 	/* this will put all unused low memory onto the freelists */
378 	memblock_free_all();
379 
380 	/*
381 	 * Check boundaries twice: Some fundamental inconsistencies can be
382 	 * detected at build time already.
383 	 */
384 #ifdef CONFIG_COMPAT
385 	BUILD_BUG_ON(TASK_SIZE_32 > DEFAULT_MAP_WINDOW_64);
386 #endif
387 
388 	/*
389 	 * Selected page table levels should match when derived from
390 	 * scratch using the virtual address range and page size.
391 	 */
392 	BUILD_BUG_ON(ARM64_HW_PGTABLE_LEVELS(CONFIG_ARM64_VA_BITS) !=
393 		     CONFIG_PGTABLE_LEVELS);
394 
395 	if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) {
396 		extern int sysctl_overcommit_memory;
397 		/*
398 		 * On a machine this small we won't get anywhere without
399 		 * overcommit, so turn it on by default.
400 		 */
401 		sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
402 	}
403 }
404 
405 void free_initmem(void)
406 {
407 	free_reserved_area(lm_alias(__init_begin),
408 			   lm_alias(__init_end),
409 			   POISON_FREE_INITMEM, "unused kernel");
410 	/*
411 	 * Unmap the __init region but leave the VM area in place. This
412 	 * prevents the region from being reused for kernel modules, which
413 	 * is not supported by kallsyms.
414 	 */
415 	vunmap_range((u64)__init_begin, (u64)__init_end);
416 }
417 
418 void dump_mem_limit(void)
419 {
420 	if (memory_limit != PHYS_ADDR_MAX) {
421 		pr_emerg("Memory Limit: %llu MB\n", memory_limit >> 20);
422 	} else {
423 		pr_emerg("Memory Limit: none\n");
424 	}
425 }
426