1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3 * Based on arch/arm/include/asm/memory.h
4 *
5 * Copyright (C) 2000-2002 Russell King
6 * Copyright (C) 2012 ARM Ltd.
7 *
8 * Note: this file should not be included by non-asm/.h files
9 */
10 #ifndef __ASM_MEMORY_H
11 #define __ASM_MEMORY_H
12
13 #include <linux/const.h>
14 #include <linux/sizes.h>
15 #include <asm/page-def.h>
16
17 /*
18 * Size of the PCI I/O space. This must remain a power of two so that
19 * IO_SPACE_LIMIT acts as a mask for the low bits of I/O addresses.
20 */
21 #define PCI_IO_SIZE SZ_16M
22
23 /*
24 * VMEMMAP_SIZE - allows the whole linear region to be covered by
25 * a struct page array
26 *
27 * If we are configured with a 52-bit kernel VA then our VMEMMAP_SIZE
28 * needs to cover the memory region from the beginning of the 52-bit
29 * PAGE_OFFSET all the way to PAGE_END for 48-bit. This allows us to
30 * keep a constant PAGE_OFFSET and "fallback" to using the higher end
31 * of the VMEMMAP where 52-bit support is not available in hardware.
32 */
33 #define VMEMMAP_SHIFT (PAGE_SHIFT - STRUCT_PAGE_MAX_SHIFT)
34 #define VMEMMAP_SIZE ((_PAGE_END(VA_BITS_MIN) - PAGE_OFFSET) >> VMEMMAP_SHIFT)
35
36 /*
37 * PAGE_OFFSET - the virtual address of the start of the linear map, at the
38 * start of the TTBR1 address space.
39 * PAGE_END - the end of the linear map, where all other kernel mappings begin.
40 * KIMAGE_VADDR - the virtual address of the start of the kernel image.
41 * VA_BITS - the maximum number of bits for virtual addresses.
42 */
43 #define VA_BITS (CONFIG_ARM64_VA_BITS)
44 #define _PAGE_OFFSET(va) (-(UL(1) << (va)))
45 #define PAGE_OFFSET (_PAGE_OFFSET(VA_BITS))
46 #define KIMAGE_VADDR (MODULES_END)
47 #define BPF_JIT_REGION_START (_PAGE_END(VA_BITS_MIN))
48 #define BPF_JIT_REGION_SIZE (SZ_128M)
49 #define BPF_JIT_REGION_END (BPF_JIT_REGION_START + BPF_JIT_REGION_SIZE)
50 #define MODULES_END (MODULES_VADDR + MODULES_VSIZE)
51 #define MODULES_VADDR (BPF_JIT_REGION_END)
52 #define MODULES_VSIZE (SZ_128M)
53 #define VMEMMAP_START (-(UL(1) << (VA_BITS - VMEMMAP_SHIFT)))
54 #define VMEMMAP_END (VMEMMAP_START + VMEMMAP_SIZE)
55 #define PCI_IO_END (VMEMMAP_START - SZ_8M)
56 #define PCI_IO_START (PCI_IO_END - PCI_IO_SIZE)
57 #define FIXADDR_TOP (VMEMMAP_START - SZ_32M)
58
59 #if VA_BITS > 48
60 #define VA_BITS_MIN (48)
61 #else
62 #define VA_BITS_MIN (VA_BITS)
63 #endif
64
65 #define _PAGE_END(va) (-(UL(1) << ((va) - 1)))
66
67 #define KERNEL_START _text
68 #define KERNEL_END _end
69
70 /*
71 * Generic and tag-based KASAN require 1/8th and 1/16th of the kernel virtual
72 * address space for the shadow region respectively. They can bloat the stack
73 * significantly, so double the (minimum) stack size when they are in use.
74 */
75 #if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
76 #define KASAN_SHADOW_OFFSET _AC(CONFIG_KASAN_SHADOW_OFFSET, UL)
77 #define KASAN_SHADOW_END ((UL(1) << (64 - KASAN_SHADOW_SCALE_SHIFT)) \
78 + KASAN_SHADOW_OFFSET)
79 #define PAGE_END (KASAN_SHADOW_END - (1UL << (vabits_actual - KASAN_SHADOW_SCALE_SHIFT)))
80 #define KASAN_THREAD_SHIFT 1
81 #else
82 #define KASAN_THREAD_SHIFT 0
83 #define PAGE_END (_PAGE_END(VA_BITS_MIN))
84 #endif /* CONFIG_KASAN */
85
86 #define MIN_THREAD_SHIFT (14 + KASAN_THREAD_SHIFT)
87
88 /*
89 * VMAP'd stacks are allocated at page granularity, so we must ensure that such
90 * stacks are a multiple of page size.
91 */
92 #if defined(CONFIG_VMAP_STACK) && (MIN_THREAD_SHIFT < PAGE_SHIFT)
93 #define THREAD_SHIFT PAGE_SHIFT
94 #else
95 #define THREAD_SHIFT MIN_THREAD_SHIFT
96 #endif
97
98 #if THREAD_SHIFT >= PAGE_SHIFT
99 #define THREAD_SIZE_ORDER (THREAD_SHIFT - PAGE_SHIFT)
100 #endif
101
102 #define THREAD_SIZE (UL(1) << THREAD_SHIFT)
103
104 /*
105 * By aligning VMAP'd stacks to 2 * THREAD_SIZE, we can detect overflow by
106 * checking sp & (1 << THREAD_SHIFT), which we can do cheaply in the entry
107 * assembly.
108 */
109 #ifdef CONFIG_VMAP_STACK
110 #define THREAD_ALIGN (2 * THREAD_SIZE)
111 #else
112 #define THREAD_ALIGN THREAD_SIZE
113 #endif
114
115 #define IRQ_STACK_SIZE THREAD_SIZE
116
117 #define OVERFLOW_STACK_SIZE SZ_4K
118
119 /*
120 * Alignment of kernel segments (e.g. .text, .data).
121 *
122 * 4 KB granule: 16 level 3 entries, with contiguous bit
123 * 16 KB granule: 4 level 3 entries, without contiguous bit
124 * 64 KB granule: 1 level 3 entry
125 */
126 #define SEGMENT_ALIGN SZ_64K
127
128 /*
129 * Memory types available.
130 *
131 * IMPORTANT: MT_NORMAL must be index 0 since vm_get_page_prot() may 'or' in
132 * the MT_NORMAL_TAGGED memory type for PROT_MTE mappings. Note
133 * that protection_map[] only contains MT_NORMAL attributes.
134 */
135 #define MT_NORMAL 0
136 #define MT_NORMAL_TAGGED 1
137 #define MT_NORMAL_NC 2
138 #define MT_NORMAL_WT 3
139 #define MT_DEVICE_nGnRnE 4
140 #define MT_DEVICE_nGnRE 5
141 #define MT_DEVICE_GRE 6
142
143 /*
144 * Memory types for Stage-2 translation
145 */
146 #define MT_S2_NORMAL 0xf
147 #define MT_S2_DEVICE_nGnRE 0x1
148
149 /*
150 * Memory types for Stage-2 translation when ID_AA64MMFR2_EL1.FWB is 0001
151 * Stage-2 enforces Normal-WB and Device-nGnRE
152 */
153 #define MT_S2_FWB_NORMAL 6
154 #define MT_S2_FWB_DEVICE_nGnRE 1
155
156 #ifdef CONFIG_ARM64_4K_PAGES
157 #define IOREMAP_MAX_ORDER (PUD_SHIFT)
158 #else
159 #define IOREMAP_MAX_ORDER (PMD_SHIFT)
160 #endif
161
162 /*
163 * Open-coded (swapper_pg_dir - reserved_pg_dir) as this cannot be calculated
164 * until link time.
165 */
166 #define RESERVED_SWAPPER_OFFSET (PAGE_SIZE)
167
168 /*
169 * Open-coded (swapper_pg_dir - tramp_pg_dir) as this cannot be calculated
170 * until link time.
171 */
172 #define TRAMP_SWAPPER_OFFSET (2 * PAGE_SIZE)
173
174 #ifndef __ASSEMBLY__
175
176 #include <linux/bitops.h>
177 #include <linux/compiler.h>
178 #include <linux/mmdebug.h>
179 #include <linux/types.h>
180 #include <asm/bug.h>
181
182 extern u64 vabits_actual;
183
184 extern s64 memstart_addr;
185 /* PHYS_OFFSET - the physical address of the start of memory. */
186 #define PHYS_OFFSET ({ VM_BUG_ON(memstart_addr & 1); memstart_addr; })
187
188 /* the virtual base of the kernel image */
189 extern u64 kimage_vaddr;
190
191 /* the offset between the kernel virtual and physical mappings */
192 extern u64 kimage_voffset;
193
kaslr_offset(void)194 static inline unsigned long kaslr_offset(void)
195 {
196 return kimage_vaddr - KIMAGE_VADDR;
197 }
198
199 /*
200 * Allow all memory at the discovery stage. We will clip it later.
201 */
202 #define MIN_MEMBLOCK_ADDR 0
203 #define MAX_MEMBLOCK_ADDR U64_MAX
204
205 /*
206 * PFNs are used to describe any physical page; this means
207 * PFN 0 == physical address 0.
208 *
209 * This is the PFN of the first RAM page in the kernel
210 * direct-mapped view. We assume this is the first page
211 * of RAM in the mem_map as well.
212 */
213 #define PHYS_PFN_OFFSET (PHYS_OFFSET >> PAGE_SHIFT)
214
215 /*
216 * When dealing with data aborts, watchpoints, or instruction traps we may end
217 * up with a tagged userland pointer. Clear the tag to get a sane pointer to
218 * pass on to access_ok(), for instance.
219 */
220 #define __untagged_addr(addr) \
221 ((__force __typeof__(addr))sign_extend64((__force u64)(addr), 55))
222
223 #define untagged_addr(addr) ({ \
224 u64 __addr = (__force u64)(addr); \
225 __addr &= __untagged_addr(__addr); \
226 (__force __typeof__(addr))__addr; \
227 })
228
229 #if defined(CONFIG_KASAN_SW_TAGS) || defined(CONFIG_KASAN_HW_TAGS)
230 #define __tag_shifted(tag) ((u64)(tag) << 56)
231 #define __tag_reset(addr) __untagged_addr(addr)
232 #define __tag_get(addr) (__u8)((u64)(addr) >> 56)
233 #else
234 #define __tag_shifted(tag) 0UL
235 #define __tag_reset(addr) (addr)
236 #define __tag_get(addr) 0
237 #endif /* CONFIG_KASAN_SW_TAGS || CONFIG_KASAN_HW_TAGS */
238
__tag_set(const void * addr,u8 tag)239 static inline const void *__tag_set(const void *addr, u8 tag)
240 {
241 u64 __addr = (u64)addr & ~__tag_shifted(0xff);
242 return (const void *)(__addr | __tag_shifted(tag));
243 }
244
245 #ifdef CONFIG_KASAN_HW_TAGS
246 #define arch_enable_tagging_sync() mte_enable_kernel_sync()
247 #define arch_enable_tagging_async() mte_enable_kernel_async()
248 #define arch_set_tagging_report_once(state) mte_set_report_once(state)
249 #define arch_force_async_tag_fault() mte_check_tfsr_exit()
250 #define arch_init_tags(max_tag) mte_init_tags(max_tag)
251 #define arch_get_random_tag() mte_get_random_tag()
252 #define arch_get_mem_tag(addr) mte_get_mem_tag(addr)
253 #define arch_set_mem_tag_range(addr, size, tag, init) \
254 mte_set_mem_tag_range((addr), (size), (tag), (init))
255 #endif /* CONFIG_KASAN_HW_TAGS */
256
257 /*
258 * Physical vs virtual RAM address space conversion. These are
259 * private definitions which should NOT be used outside memory.h
260 * files. Use virt_to_phys/phys_to_virt/__pa/__va instead.
261 */
262
263
264 /*
265 * Check whether an arbitrary address is within the linear map, which
266 * lives in the [PAGE_OFFSET, PAGE_END) interval at the bottom of the
267 * kernel's TTBR1 address range.
268 */
269 #define __is_lm_address(addr) (((u64)(addr) - PAGE_OFFSET) < (PAGE_END - PAGE_OFFSET))
270
271 #define __lm_to_phys(addr) (((addr) - PAGE_OFFSET) + PHYS_OFFSET)
272 #define __kimg_to_phys(addr) ((addr) - kimage_voffset)
273
274 #define __virt_to_phys_nodebug(x) ({ \
275 phys_addr_t __x = (phys_addr_t)(__tag_reset(x)); \
276 __is_lm_address(__x) ? __lm_to_phys(__x) : __kimg_to_phys(__x); \
277 })
278
279 #define __pa_symbol_nodebug(x) __kimg_to_phys((phys_addr_t)(x))
280
281 #ifdef CONFIG_DEBUG_VIRTUAL
282 extern phys_addr_t __virt_to_phys(unsigned long x);
283 extern phys_addr_t __phys_addr_symbol(unsigned long x);
284 #else
285 #define __virt_to_phys(x) __virt_to_phys_nodebug(x)
286 #define __phys_addr_symbol(x) __pa_symbol_nodebug(x)
287 #endif /* CONFIG_DEBUG_VIRTUAL */
288
289 #define __phys_to_virt(x) ((unsigned long)((x) - PHYS_OFFSET) | PAGE_OFFSET)
290 #define __phys_to_kimg(x) ((unsigned long)((x) + kimage_voffset))
291
292 /*
293 * Convert a page to/from a physical address
294 */
295 #define page_to_phys(page) (__pfn_to_phys(page_to_pfn(page)))
296 #define phys_to_page(phys) (pfn_to_page(__phys_to_pfn(phys)))
297
298 /*
299 * Note: Drivers should NOT use these. They are the wrong
300 * translation for translating DMA addresses. Use the driver
301 * DMA support - see dma-mapping.h.
302 */
303 #define virt_to_phys virt_to_phys
virt_to_phys(const volatile void * x)304 static inline phys_addr_t virt_to_phys(const volatile void *x)
305 {
306 return __virt_to_phys((unsigned long)(x));
307 }
308
309 #define phys_to_virt phys_to_virt
phys_to_virt(phys_addr_t x)310 static inline void *phys_to_virt(phys_addr_t x)
311 {
312 return (void *)(__phys_to_virt(x));
313 }
314
315 /*
316 * Drivers should NOT use these either.
317 */
318 #define __pa(x) __virt_to_phys((unsigned long)(x))
319 #define __pa_symbol(x) __phys_addr_symbol(RELOC_HIDE((unsigned long)(x), 0))
320 #define __pa_nodebug(x) __virt_to_phys_nodebug((unsigned long)(x))
321 #define __va(x) ((void *)__phys_to_virt((phys_addr_t)(x)))
322 #define pfn_to_kaddr(pfn) __va((pfn) << PAGE_SHIFT)
323 #define virt_to_pfn(x) __phys_to_pfn(__virt_to_phys((unsigned long)(x)))
324 #define sym_to_pfn(x) __phys_to_pfn(__pa_symbol(x))
325
326 #ifdef CONFIG_CFI_CLANG
327 /*
328 * With CONFIG_CFI_CLANG, the compiler replaces function address
329 * references with the address of the function's CFI jump table
330 * entry. The function_nocfi macro always returns the address of the
331 * actual function instead.
332 */
333 #define function_nocfi(x) ({ \
334 void *addr; \
335 asm("adrp %0, " __stringify(x) "\n\t" \
336 "add %0, %0, :lo12:" __stringify(x) \
337 : "=r" (addr)); \
338 addr; \
339 })
340 #endif
341
342 /*
343 * virt_to_page(x) convert a _valid_ virtual address to struct page *
344 * virt_addr_valid(x) indicates whether a virtual address is valid
345 */
346 #define ARCH_PFN_OFFSET ((unsigned long)PHYS_PFN_OFFSET)
347
348 #if defined(CONFIG_DEBUG_VIRTUAL)
349 #define page_to_virt(x) ({ \
350 __typeof__(x) __page = x; \
351 void *__addr = __va(page_to_phys(__page)); \
352 (void *)__tag_set((const void *)__addr, page_kasan_tag(__page));\
353 })
354 #define virt_to_page(x) pfn_to_page(virt_to_pfn(x))
355 #else
356 #define page_to_virt(x) ({ \
357 __typeof__(x) __page = x; \
358 u64 __idx = ((u64)__page - VMEMMAP_START) / sizeof(struct page);\
359 u64 __addr = PAGE_OFFSET + (__idx * PAGE_SIZE); \
360 (void *)__tag_set((const void *)__addr, page_kasan_tag(__page));\
361 })
362
363 #define virt_to_page(x) ({ \
364 u64 __idx = (__tag_reset((u64)x) - PAGE_OFFSET) / PAGE_SIZE; \
365 u64 __addr = VMEMMAP_START + (__idx * sizeof(struct page)); \
366 (struct page *)__addr; \
367 })
368 #endif /* CONFIG_DEBUG_VIRTUAL */
369
370 #define virt_addr_valid(addr) ({ \
371 __typeof__(addr) __addr = __tag_reset(addr); \
372 __is_lm_address(__addr) && pfn_valid(virt_to_pfn(__addr)); \
373 })
374
375 void dump_mem_limit(void);
376 #endif /* !ASSEMBLY */
377
378 /*
379 * Given that the GIC architecture permits ITS implementations that can only be
380 * configured with a LPI table address once, GICv3 systems with many CPUs may
381 * end up reserving a lot of different regions after a kexec for their LPI
382 * tables (one per CPU), as we are forced to reuse the same memory after kexec
383 * (and thus reserve it persistently with EFI beforehand)
384 */
385 #if defined(CONFIG_EFI) && defined(CONFIG_ARM_GIC_V3_ITS)
386 # define INIT_MEMBLOCK_RESERVED_REGIONS (INIT_MEMBLOCK_REGIONS + NR_CPUS + 1)
387 #endif
388
389 #include <asm-generic/memory_model.h>
390
391 #endif /* __ASM_MEMORY_H */
392