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 arm64_dma_phys_limit __ro_after_init;
66 
67 #ifdef CONFIG_KEXEC_CORE
68 /*
69  * reserve_crashkernel() - reserves memory for crash kernel
70  *
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.
74  */
reserve_crashkernel(void)75 static void __init reserve_crashkernel(void)
76 {
77 	unsigned long long crash_base, crash_size;
78 	unsigned long long crash_max = arm64_dma_phys_limit;
79 	int ret;
80 
81 	ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
82 				&crash_size, &crash_base);
83 	/* no crashkernel= or invalid value specified */
84 	if (ret || !crash_size)
85 		return;
86 
87 	crash_size = PAGE_ALIGN(crash_size);
88 
89 	/* User specifies base address explicitly. */
90 	if (crash_base)
91 		crash_max = crash_base + crash_size;
92 
93 	/* Current arm64 boot protocol requires 2MB alignment */
94 	crash_base = memblock_phys_alloc_range(crash_size, SZ_2M,
95 					       crash_base, crash_max);
96 	if (!crash_base) {
97 		pr_warn("cannot allocate crashkernel (size:0x%llx)\n",
98 			crash_size);
99 		return;
100 	}
101 
102 	pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n",
103 		crash_base, crash_base + crash_size, crash_size >> 20);
104 
105 	/*
106 	 * The crashkernel memory will be removed from the kernel linear
107 	 * map. Inform kmemleak so that it won't try to access it.
108 	 */
109 	kmemleak_ignore_phys(crash_base);
110 	crashk_res.start = crash_base;
111 	crashk_res.end = crash_base + crash_size - 1;
112 }
113 #else
reserve_crashkernel(void)114 static void __init reserve_crashkernel(void)
115 {
116 }
117 #endif /* CONFIG_KEXEC_CORE */
118 
119 /*
120  * Return the maximum physical address for a zone accessible by the given bits
121  * limit. If DRAM starts above 32-bit, expand the zone to the maximum
122  * available memory, otherwise cap it at 32-bit.
123  */
max_zone_phys(unsigned int zone_bits)124 static phys_addr_t __init max_zone_phys(unsigned int zone_bits)
125 {
126 	phys_addr_t zone_mask = DMA_BIT_MASK(zone_bits);
127 	phys_addr_t phys_start = memblock_start_of_DRAM();
128 
129 	if (phys_start > U32_MAX)
130 		zone_mask = PHYS_ADDR_MAX;
131 	else if (phys_start > zone_mask)
132 		zone_mask = U32_MAX;
133 
134 	return min(zone_mask, memblock_end_of_DRAM() - 1) + 1;
135 }
136 
zone_sizes_init(unsigned long min,unsigned long max)137 static void __init zone_sizes_init(unsigned long min, unsigned long max)
138 {
139 	unsigned long max_zone_pfns[MAX_NR_ZONES]  = {0};
140 	unsigned int __maybe_unused acpi_zone_dma_bits;
141 	unsigned int __maybe_unused dt_zone_dma_bits;
142 	phys_addr_t __maybe_unused dma32_phys_limit = max_zone_phys(32);
143 
144 #ifdef CONFIG_ZONE_DMA
145 	acpi_zone_dma_bits = fls64(acpi_iort_dma_get_max_cpu_address());
146 	dt_zone_dma_bits = fls64(of_dma_get_max_cpu_address(NULL));
147 	zone_dma_bits = min3(32U, dt_zone_dma_bits, acpi_zone_dma_bits);
148 	arm64_dma_phys_limit = max_zone_phys(zone_dma_bits);
149 	max_zone_pfns[ZONE_DMA] = PFN_DOWN(arm64_dma_phys_limit);
150 #endif
151 #ifdef CONFIG_ZONE_DMA32
152 	max_zone_pfns[ZONE_DMA32] = PFN_DOWN(dma32_phys_limit);
153 	if (!arm64_dma_phys_limit)
154 		arm64_dma_phys_limit = dma32_phys_limit;
155 #endif
156 	if (!arm64_dma_phys_limit)
157 		arm64_dma_phys_limit = PHYS_MASK + 1;
158 	max_zone_pfns[ZONE_NORMAL] = max;
159 
160 	free_area_init(max_zone_pfns);
161 }
162 
pfn_is_map_memory(unsigned long pfn)163 int pfn_is_map_memory(unsigned long pfn)
164 {
165 	phys_addr_t addr = PFN_PHYS(pfn);
166 
167 	/* avoid false positives for bogus PFNs, see comment in pfn_valid() */
168 	if (PHYS_PFN(addr) != pfn)
169 		return 0;
170 
171 	return memblock_is_map_memory(addr);
172 }
173 EXPORT_SYMBOL(pfn_is_map_memory);
174 
175 static phys_addr_t memory_limit = PHYS_ADDR_MAX;
176 
177 /*
178  * Limit the memory size that was specified via FDT.
179  */
early_mem(char * p)180 static int __init early_mem(char *p)
181 {
182 	if (!p)
183 		return 1;
184 
185 	memory_limit = memparse(p, &p) & PAGE_MASK;
186 	pr_notice("Memory limited to %lldMB\n", memory_limit >> 20);
187 
188 	return 0;
189 }
190 early_param("mem", early_mem);
191 
arm64_memblock_init(void)192 void __init arm64_memblock_init(void)
193 {
194 	s64 linear_region_size = PAGE_END - _PAGE_OFFSET(vabits_actual);
195 
196 	/*
197 	 * Corner case: 52-bit VA capable systems running KVM in nVHE mode may
198 	 * be limited in their ability to support a linear map that exceeds 51
199 	 * bits of VA space, depending on the placement of the ID map. Given
200 	 * that the placement of the ID map may be randomized, let's simply
201 	 * limit the kernel's linear map to 51 bits as well if we detect this
202 	 * configuration.
203 	 */
204 	if (IS_ENABLED(CONFIG_KVM) && vabits_actual == 52 &&
205 	    is_hyp_mode_available() && !is_kernel_in_hyp_mode()) {
206 		pr_info("Capping linear region to 51 bits for KVM in nVHE mode on LVA capable hardware.\n");
207 		linear_region_size = min_t(u64, linear_region_size, BIT(51));
208 	}
209 
210 	/* Remove memory above our supported physical address size */
211 	memblock_remove(1ULL << PHYS_MASK_SHIFT, ULLONG_MAX);
212 
213 	/*
214 	 * Select a suitable value for the base of physical memory.
215 	 */
216 	memstart_addr = round_down(memblock_start_of_DRAM(),
217 				   ARM64_MEMSTART_ALIGN);
218 
219 	if ((memblock_end_of_DRAM() - memstart_addr) > linear_region_size)
220 		pr_warn("Memory doesn't fit in the linear mapping, VA_BITS too small\n");
221 
222 	/*
223 	 * Remove the memory that we will not be able to cover with the
224 	 * linear mapping. Take care not to clip the kernel which may be
225 	 * high in memory.
226 	 */
227 	memblock_remove(max_t(u64, memstart_addr + linear_region_size,
228 			__pa_symbol(_end)), ULLONG_MAX);
229 	if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) {
230 		/* ensure that memstart_addr remains sufficiently aligned */
231 		memstart_addr = round_up(memblock_end_of_DRAM() - linear_region_size,
232 					 ARM64_MEMSTART_ALIGN);
233 		memblock_remove(0, memstart_addr);
234 	}
235 
236 	/*
237 	 * If we are running with a 52-bit kernel VA config on a system that
238 	 * does not support it, we have to place the available physical
239 	 * memory in the 48-bit addressable part of the linear region, i.e.,
240 	 * we have to move it upward. Since memstart_addr represents the
241 	 * physical address of PAGE_OFFSET, we have to *subtract* from it.
242 	 */
243 	if (IS_ENABLED(CONFIG_ARM64_VA_BITS_52) && (vabits_actual != 52))
244 		memstart_addr -= _PAGE_OFFSET(48) - _PAGE_OFFSET(52);
245 
246 	/*
247 	 * Apply the memory limit if it was set. Since the kernel may be loaded
248 	 * high up in memory, add back the kernel region that must be accessible
249 	 * via the linear mapping.
250 	 */
251 	if (memory_limit != PHYS_ADDR_MAX) {
252 		memblock_mem_limit_remove_map(memory_limit);
253 		memblock_add(__pa_symbol(_text), (u64)(_end - _text));
254 	}
255 
256 	if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
257 		/*
258 		 * Add back the memory we just removed if it results in the
259 		 * initrd to become inaccessible via the linear mapping.
260 		 * Otherwise, this is a no-op
261 		 */
262 		u64 base = phys_initrd_start & PAGE_MASK;
263 		u64 size = PAGE_ALIGN(phys_initrd_start + phys_initrd_size) - base;
264 
265 		/*
266 		 * We can only add back the initrd memory if we don't end up
267 		 * with more memory than we can address via the linear mapping.
268 		 * It is up to the bootloader to position the kernel and the
269 		 * initrd reasonably close to each other (i.e., within 32 GB of
270 		 * each other) so that all granule/#levels combinations can
271 		 * always access both.
272 		 */
273 		if (WARN(base < memblock_start_of_DRAM() ||
274 			 base + size > memblock_start_of_DRAM() +
275 				       linear_region_size,
276 			"initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) {
277 			phys_initrd_size = 0;
278 		} else {
279 			memblock_remove(base, size); /* clear MEMBLOCK_ flags */
280 			memblock_add(base, size);
281 			memblock_reserve(base, size);
282 		}
283 	}
284 
285 	if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
286 		extern u16 memstart_offset_seed;
287 		u64 mmfr0 = read_cpuid(ID_AA64MMFR0_EL1);
288 		int parange = cpuid_feature_extract_unsigned_field(
289 					mmfr0, ID_AA64MMFR0_PARANGE_SHIFT);
290 		s64 range = linear_region_size -
291 			    BIT(id_aa64mmfr0_parange_to_phys_shift(parange));
292 
293 		/*
294 		 * If the size of the linear region exceeds, by a sufficient
295 		 * margin, the size of the region that the physical memory can
296 		 * span, randomize the linear region as well.
297 		 */
298 		if (memstart_offset_seed > 0 && range >= (s64)ARM64_MEMSTART_ALIGN) {
299 			range /= ARM64_MEMSTART_ALIGN;
300 			memstart_addr -= ARM64_MEMSTART_ALIGN *
301 					 ((range * memstart_offset_seed) >> 16);
302 		}
303 	}
304 
305 	/*
306 	 * Register the kernel text, kernel data, initrd, and initial
307 	 * pagetables with memblock.
308 	 */
309 	memblock_reserve(__pa_symbol(_stext), _end - _stext);
310 	if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) {
311 		/* the generic initrd code expects virtual addresses */
312 		initrd_start = __phys_to_virt(phys_initrd_start);
313 		initrd_end = initrd_start + phys_initrd_size;
314 	}
315 
316 	early_init_fdt_scan_reserved_mem();
317 
318 	high_memory = __va(memblock_end_of_DRAM() - 1) + 1;
319 }
320 
bootmem_init(void)321 void __init bootmem_init(void)
322 {
323 	unsigned long min, max;
324 
325 	min = PFN_UP(memblock_start_of_DRAM());
326 	max = PFN_DOWN(memblock_end_of_DRAM());
327 
328 	early_memtest(min << PAGE_SHIFT, max << PAGE_SHIFT);
329 
330 	max_pfn = max_low_pfn = max;
331 	min_low_pfn = min;
332 
333 	arch_numa_init();
334 
335 	/*
336 	 * must be done after arch_numa_init() which calls numa_init() to
337 	 * initialize node_online_map that gets used in hugetlb_cma_reserve()
338 	 * while allocating required CMA size across online nodes.
339 	 */
340 #if defined(CONFIG_HUGETLB_PAGE) && defined(CONFIG_CMA)
341 	arm64_hugetlb_cma_reserve();
342 #endif
343 
344 	dma_pernuma_cma_reserve();
345 
346 	kvm_hyp_reserve();
347 
348 	/*
349 	 * sparse_init() tries to allocate memory from memblock, so must be
350 	 * done after the fixed reservations
351 	 */
352 	sparse_init();
353 	zone_sizes_init(min, max);
354 
355 	/*
356 	 * Reserve the CMA area after arm64_dma_phys_limit was initialised.
357 	 */
358 	dma_contiguous_reserve(arm64_dma_phys_limit);
359 
360 	/*
361 	 * request_standard_resources() depends on crashkernel's memory being
362 	 * reserved, so do it here.
363 	 */
364 	reserve_crashkernel();
365 
366 	memblock_dump_all();
367 }
368 
369 /*
370  * mem_init() marks the free areas in the mem_map and tells us how much memory
371  * is free.  This is done after various parts of the system have claimed their
372  * memory after the kernel image.
373  */
mem_init(void)374 void __init mem_init(void)
375 {
376 	if (swiotlb_force == SWIOTLB_FORCE ||
377 	    max_pfn > PFN_DOWN(arm64_dma_phys_limit))
378 		swiotlb_init(1);
379 	else if (!xen_swiotlb_detect())
380 		swiotlb_force = SWIOTLB_NO_FORCE;
381 
382 	/* this will put all unused low memory onto the freelists */
383 	memblock_free_all();
384 
385 	/*
386 	 * Check boundaries twice: Some fundamental inconsistencies can be
387 	 * detected at build time already.
388 	 */
389 #ifdef CONFIG_COMPAT
390 	BUILD_BUG_ON(TASK_SIZE_32 > DEFAULT_MAP_WINDOW_64);
391 #endif
392 
393 	/*
394 	 * Selected page table levels should match when derived from
395 	 * scratch using the virtual address range and page size.
396 	 */
397 	BUILD_BUG_ON(ARM64_HW_PGTABLE_LEVELS(CONFIG_ARM64_VA_BITS) !=
398 		     CONFIG_PGTABLE_LEVELS);
399 
400 	if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) {
401 		extern int sysctl_overcommit_memory;
402 		/*
403 		 * On a machine this small we won't get anywhere without
404 		 * overcommit, so turn it on by default.
405 		 */
406 		sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
407 	}
408 }
409 
free_initmem(void)410 void free_initmem(void)
411 {
412 	free_reserved_area(lm_alias(__init_begin),
413 			   lm_alias(__init_end),
414 			   POISON_FREE_INITMEM, "unused kernel");
415 	/*
416 	 * Unmap the __init region but leave the VM area in place. This
417 	 * prevents the region from being reused for kernel modules, which
418 	 * is not supported by kallsyms.
419 	 */
420 	vunmap_range((u64)__init_begin, (u64)__init_end);
421 }
422 
dump_mem_limit(void)423 void dump_mem_limit(void)
424 {
425 	if (memory_limit != PHYS_ADDR_MAX) {
426 		pr_emerg("Memory Limit: %llu MB\n", memory_limit >> 20);
427 	} else {
428 		pr_emerg("Memory Limit: none\n");
429 	}
430 }
431