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_RANGE (_PAGE_END(VA_BITS_MIN) - PAGE_OFFSET)
34 #define VMEMMAP_SIZE ((VMEMMAP_RANGE >> PAGE_SHIFT) * sizeof(struct page))
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 MODULES_END (MODULES_VADDR + MODULES_VSIZE)
48 #define MODULES_VADDR (_PAGE_END(VA_BITS_MIN))
49 #define MODULES_VSIZE (SZ_2G)
50 #define VMEMMAP_START (VMEMMAP_END - VMEMMAP_SIZE)
51 #define VMEMMAP_END (-UL(SZ_1G))
52 #define PCI_IO_START (VMEMMAP_END + SZ_8M)
53 #define PCI_IO_END (PCI_IO_START + PCI_IO_SIZE)
54 #define FIXADDR_TOP (-UL(SZ_8M))
55
56 #if VA_BITS > 48
57 #ifdef CONFIG_ARM64_16K_PAGES
58 #define VA_BITS_MIN (47)
59 #else
60 #define VA_BITS_MIN (48)
61 #endif
62 #else
63 #define VA_BITS_MIN (VA_BITS)
64 #endif
65
66 #define _PAGE_END(va) (-(UL(1) << ((va) - 1)))
67
68 #define KERNEL_START _text
69 #define KERNEL_END _end
70
71 /*
72 * Generic and Software Tag-Based KASAN modes require 1/8th and 1/16th of the
73 * kernel virtual address space for storing the shadow memory respectively.
74 *
75 * The mapping between a virtual memory address and its corresponding shadow
76 * memory address is defined based on the formula:
77 *
78 * shadow_addr = (addr >> KASAN_SHADOW_SCALE_SHIFT) + KASAN_SHADOW_OFFSET
79 *
80 * where KASAN_SHADOW_SCALE_SHIFT is the order of the number of bits that map
81 * to a single shadow byte and KASAN_SHADOW_OFFSET is a constant that offsets
82 * the mapping. Note that KASAN_SHADOW_OFFSET does not point to the start of
83 * the shadow memory region.
84 *
85 * Based on this mapping, we define two constants:
86 *
87 * KASAN_SHADOW_START: the start of the shadow memory region;
88 * KASAN_SHADOW_END: the end of the shadow memory region.
89 *
90 * KASAN_SHADOW_END is defined first as the shadow address that corresponds to
91 * the upper bound of possible virtual kernel memory addresses UL(1) << 64
92 * according to the mapping formula.
93 *
94 * KASAN_SHADOW_START is defined second based on KASAN_SHADOW_END. The shadow
95 * memory start must map to the lowest possible kernel virtual memory address
96 * and thus it depends on the actual bitness of the address space.
97 *
98 * As KASAN inserts redzones between stack variables, this increases the stack
99 * memory usage significantly. Thus, we double the (minimum) stack size.
100 */
101 #if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
102 #define KASAN_SHADOW_OFFSET _AC(CONFIG_KASAN_SHADOW_OFFSET, UL)
103 #define KASAN_SHADOW_END ((UL(1) << (64 - KASAN_SHADOW_SCALE_SHIFT)) + KASAN_SHADOW_OFFSET)
104 #define _KASAN_SHADOW_START(va) (KASAN_SHADOW_END - (UL(1) << ((va) - KASAN_SHADOW_SCALE_SHIFT)))
105 #define KASAN_SHADOW_START _KASAN_SHADOW_START(vabits_actual)
106 #define PAGE_END KASAN_SHADOW_START
107 #define KASAN_THREAD_SHIFT 1
108 #else
109 #define KASAN_THREAD_SHIFT 0
110 #define PAGE_END (_PAGE_END(VA_BITS_MIN))
111 #endif /* CONFIG_KASAN */
112
113 #define MIN_THREAD_SHIFT (14 + KASAN_THREAD_SHIFT)
114
115 /*
116 * VMAP'd stacks are allocated at page granularity, so we must ensure that such
117 * stacks are a multiple of page size.
118 */
119 #if defined(CONFIG_VMAP_STACK) && (MIN_THREAD_SHIFT < PAGE_SHIFT)
120 #define THREAD_SHIFT PAGE_SHIFT
121 #else
122 #define THREAD_SHIFT MIN_THREAD_SHIFT
123 #endif
124
125 #if THREAD_SHIFT >= PAGE_SHIFT
126 #define THREAD_SIZE_ORDER (THREAD_SHIFT - PAGE_SHIFT)
127 #endif
128
129 #define THREAD_SIZE (UL(1) << THREAD_SHIFT)
130
131 /*
132 * By aligning VMAP'd stacks to 2 * THREAD_SIZE, we can detect overflow by
133 * checking sp & (1 << THREAD_SHIFT), which we can do cheaply in the entry
134 * assembly.
135 */
136 #ifdef CONFIG_VMAP_STACK
137 #define THREAD_ALIGN (2 * THREAD_SIZE)
138 #else
139 #define THREAD_ALIGN THREAD_SIZE
140 #endif
141
142 #define IRQ_STACK_SIZE THREAD_SIZE
143
144 #define OVERFLOW_STACK_SIZE SZ_4K
145
146 /*
147 * With the minimum frame size of [x29, x30], exactly half the combined
148 * sizes of the hyp and overflow stacks is the maximum size needed to
149 * save the unwinded stacktrace; plus an additional entry to delimit the
150 * end.
151 */
152 #define NVHE_STACKTRACE_SIZE ((OVERFLOW_STACK_SIZE + PAGE_SIZE) / 2 + sizeof(long))
153
154 /*
155 * Alignment of kernel segments (e.g. .text, .data).
156 *
157 * 4 KB granule: 16 level 3 entries, with contiguous bit
158 * 16 KB granule: 4 level 3 entries, without contiguous bit
159 * 64 KB granule: 1 level 3 entry
160 */
161 #define SEGMENT_ALIGN SZ_64K
162
163 /*
164 * Memory types available.
165 *
166 * IMPORTANT: MT_NORMAL must be index 0 since vm_get_page_prot() may 'or' in
167 * the MT_NORMAL_TAGGED memory type for PROT_MTE mappings. Note
168 * that protection_map[] only contains MT_NORMAL attributes.
169 */
170 #define MT_NORMAL 0
171 #define MT_NORMAL_TAGGED 1
172 #define MT_NORMAL_NC 2
173 #define MT_DEVICE_nGnRnE 3
174 #define MT_DEVICE_nGnRE 4
175
176 /*
177 * Memory types for Stage-2 translation
178 */
179 #define MT_S2_NORMAL 0xf
180 #define MT_S2_NORMAL_NC 0x5
181 #define MT_S2_DEVICE_nGnRE 0x1
182
183 /*
184 * Memory types for Stage-2 translation when ID_AA64MMFR2_EL1.FWB is 0001
185 * Stage-2 enforces Normal-WB and Device-nGnRE
186 */
187 #define MT_S2_FWB_NORMAL 6
188 #define MT_S2_FWB_NORMAL_NC 5
189 #define MT_S2_FWB_DEVICE_nGnRE 1
190
191 #ifdef CONFIG_ARM64_4K_PAGES
192 #define IOREMAP_MAX_ORDER (PUD_SHIFT)
193 #else
194 #define IOREMAP_MAX_ORDER (PMD_SHIFT)
195 #endif
196
197 /*
198 * Open-coded (swapper_pg_dir - reserved_pg_dir) as this cannot be calculated
199 * until link time.
200 */
201 #define RESERVED_SWAPPER_OFFSET (PAGE_SIZE)
202
203 /*
204 * Open-coded (swapper_pg_dir - tramp_pg_dir) as this cannot be calculated
205 * until link time.
206 */
207 #define TRAMP_SWAPPER_OFFSET (2 * PAGE_SIZE)
208
209 #ifndef __ASSEMBLY__
210
211 #include <linux/bitops.h>
212 #include <linux/compiler.h>
213 #include <linux/mmdebug.h>
214 #include <linux/types.h>
215 #include <asm/boot.h>
216 #include <asm/bug.h>
217 #include <asm/sections.h>
218 #include <asm/sysreg.h>
219
read_tcr(void)220 static inline u64 __pure read_tcr(void)
221 {
222 u64 tcr;
223
224 // read_sysreg() uses asm volatile, so avoid it here
225 asm("mrs %0, tcr_el1" : "=r"(tcr));
226 return tcr;
227 }
228
229 #if VA_BITS > 48
230 // For reasons of #include hell, we can't use TCR_T1SZ_OFFSET/TCR_T1SZ_MASK here
231 #define vabits_actual (64 - ((read_tcr() >> 16) & 63))
232 #else
233 #define vabits_actual ((u64)VA_BITS)
234 #endif
235
236 extern s64 memstart_addr;
237 /* PHYS_OFFSET - the physical address of the start of memory. */
238 #define PHYS_OFFSET ({ VM_BUG_ON(memstart_addr & 1); memstart_addr; })
239
240 /* the offset between the kernel virtual and physical mappings */
241 extern u64 kimage_voffset;
242
kaslr_offset(void)243 static inline unsigned long kaslr_offset(void)
244 {
245 return (u64)&_text - KIMAGE_VADDR;
246 }
247
248 #ifdef CONFIG_RANDOMIZE_BASE
249 void kaslr_init(void);
kaslr_enabled(void)250 static inline bool kaslr_enabled(void)
251 {
252 extern bool __kaslr_is_enabled;
253 return __kaslr_is_enabled;
254 }
255 #else
kaslr_init(void)256 static inline void kaslr_init(void) { }
kaslr_enabled(void)257 static inline bool kaslr_enabled(void) { return false; }
258 #endif
259
260 /*
261 * Allow all memory at the discovery stage. We will clip it later.
262 */
263 #define MIN_MEMBLOCK_ADDR 0
264 #define MAX_MEMBLOCK_ADDR U64_MAX
265
266 /*
267 * PFNs are used to describe any physical page; this means
268 * PFN 0 == physical address 0.
269 *
270 * This is the PFN of the first RAM page in the kernel
271 * direct-mapped view. We assume this is the first page
272 * of RAM in the mem_map as well.
273 */
274 #define PHYS_PFN_OFFSET (PHYS_OFFSET >> PAGE_SHIFT)
275
276 /*
277 * When dealing with data aborts, watchpoints, or instruction traps we may end
278 * up with a tagged userland pointer. Clear the tag to get a sane pointer to
279 * pass on to access_ok(), for instance.
280 */
281 #define __untagged_addr(addr) \
282 ((__force __typeof__(addr))sign_extend64((__force u64)(addr), 55))
283
284 #define untagged_addr(addr) ({ \
285 u64 __addr = (__force u64)(addr); \
286 __addr &= __untagged_addr(__addr); \
287 (__force __typeof__(addr))__addr; \
288 })
289
290 #if defined(CONFIG_KASAN_SW_TAGS) || defined(CONFIG_KASAN_HW_TAGS)
291 #define __tag_shifted(tag) ((u64)(tag) << 56)
292 #define __tag_reset(addr) __untagged_addr(addr)
293 #define __tag_get(addr) (__u8)((u64)(addr) >> 56)
294 #else
295 #define __tag_shifted(tag) 0UL
296 #define __tag_reset(addr) (addr)
297 #define __tag_get(addr) 0
298 #endif /* CONFIG_KASAN_SW_TAGS || CONFIG_KASAN_HW_TAGS */
299
__tag_set(const void * addr,u8 tag)300 static inline const void *__tag_set(const void *addr, u8 tag)
301 {
302 u64 __addr = (u64)addr & ~__tag_shifted(0xff);
303 return (const void *)(__addr | __tag_shifted(tag));
304 }
305
306 #ifdef CONFIG_KASAN_HW_TAGS
307 #define arch_enable_tag_checks_sync() mte_enable_kernel_sync()
308 #define arch_enable_tag_checks_async() mte_enable_kernel_async()
309 #define arch_enable_tag_checks_asymm() mte_enable_kernel_asymm()
310 #define arch_suppress_tag_checks_start() mte_enable_tco()
311 #define arch_suppress_tag_checks_stop() mte_disable_tco()
312 #define arch_force_async_tag_fault() mte_check_tfsr_exit()
313 #define arch_get_random_tag() mte_get_random_tag()
314 #define arch_get_mem_tag(addr) mte_get_mem_tag(addr)
315 #define arch_set_mem_tag_range(addr, size, tag, init) \
316 mte_set_mem_tag_range((addr), (size), (tag), (init))
317 #endif /* CONFIG_KASAN_HW_TAGS */
318
319 /*
320 * Physical vs virtual RAM address space conversion. These are
321 * private definitions which should NOT be used outside memory.h
322 * files. Use virt_to_phys/phys_to_virt/__pa/__va instead.
323 */
324
325
326 /*
327 * Check whether an arbitrary address is within the linear map, which
328 * lives in the [PAGE_OFFSET, PAGE_END) interval at the bottom of the
329 * kernel's TTBR1 address range.
330 */
331 #define __is_lm_address(addr) (((u64)(addr) - PAGE_OFFSET) < (PAGE_END - PAGE_OFFSET))
332
333 #define __lm_to_phys(addr) (((addr) - PAGE_OFFSET) + PHYS_OFFSET)
334 #define __kimg_to_phys(addr) ((addr) - kimage_voffset)
335
336 #define __virt_to_phys_nodebug(x) ({ \
337 phys_addr_t __x = (phys_addr_t)(__tag_reset(x)); \
338 __is_lm_address(__x) ? __lm_to_phys(__x) : __kimg_to_phys(__x); \
339 })
340
341 #define __pa_symbol_nodebug(x) __kimg_to_phys((phys_addr_t)(x))
342
343 #ifdef CONFIG_DEBUG_VIRTUAL
344 extern phys_addr_t __virt_to_phys(unsigned long x);
345 extern phys_addr_t __phys_addr_symbol(unsigned long x);
346 #else
347 #define __virt_to_phys(x) __virt_to_phys_nodebug(x)
348 #define __phys_addr_symbol(x) __pa_symbol_nodebug(x)
349 #endif /* CONFIG_DEBUG_VIRTUAL */
350
351 #define __phys_to_virt(x) ((unsigned long)((x) - PHYS_OFFSET) | PAGE_OFFSET)
352 #define __phys_to_kimg(x) ((unsigned long)((x) + kimage_voffset))
353
354 /*
355 * Convert a page to/from a physical address
356 */
357 #define page_to_phys(page) (__pfn_to_phys(page_to_pfn(page)))
358 #define phys_to_page(phys) (pfn_to_page(__phys_to_pfn(phys)))
359
360 /*
361 * Note: Drivers should NOT use these. They are the wrong
362 * translation for translating DMA addresses. Use the driver
363 * DMA support - see dma-mapping.h.
364 */
365 #define virt_to_phys virt_to_phys
virt_to_phys(const volatile void * x)366 static inline phys_addr_t virt_to_phys(const volatile void *x)
367 {
368 return __virt_to_phys((unsigned long)(x));
369 }
370
371 #define phys_to_virt phys_to_virt
phys_to_virt(phys_addr_t x)372 static inline void *phys_to_virt(phys_addr_t x)
373 {
374 return (void *)(__phys_to_virt(x));
375 }
376
377 /* Needed already here for resolving __phys_to_pfn() in virt_to_pfn() */
378 #include <asm-generic/memory_model.h>
379
virt_to_pfn(const void * kaddr)380 static inline unsigned long virt_to_pfn(const void *kaddr)
381 {
382 return __phys_to_pfn(virt_to_phys(kaddr));
383 }
384
385 /*
386 * Drivers should NOT use these either.
387 */
388 #define __pa(x) __virt_to_phys((unsigned long)(x))
389 #define __pa_symbol(x) __phys_addr_symbol(RELOC_HIDE((unsigned long)(x), 0))
390 #define __pa_nodebug(x) __virt_to_phys_nodebug((unsigned long)(x))
391 #define __va(x) ((void *)__phys_to_virt((phys_addr_t)(x)))
392 #define pfn_to_kaddr(pfn) __va((pfn) << PAGE_SHIFT)
393 #define sym_to_pfn(x) __phys_to_pfn(__pa_symbol(x))
394
395 /*
396 * virt_to_page(x) convert a _valid_ virtual address to struct page *
397 * virt_addr_valid(x) indicates whether a virtual address is valid
398 */
399 #define ARCH_PFN_OFFSET ((unsigned long)PHYS_PFN_OFFSET)
400
401 #if defined(CONFIG_DEBUG_VIRTUAL)
402 #define page_to_virt(x) ({ \
403 __typeof__(x) __page = x; \
404 void *__addr = __va(page_to_phys(__page)); \
405 (void *)__tag_set((const void *)__addr, page_kasan_tag(__page));\
406 })
407 #define virt_to_page(x) pfn_to_page(virt_to_pfn(x))
408 #else
409 #define page_to_virt(x) ({ \
410 __typeof__(x) __page = x; \
411 u64 __idx = ((u64)__page - VMEMMAP_START) / sizeof(struct page);\
412 u64 __addr = PAGE_OFFSET + (__idx * PAGE_SIZE); \
413 (void *)__tag_set((const void *)__addr, page_kasan_tag(__page));\
414 })
415
416 #define virt_to_page(x) ({ \
417 u64 __idx = (__tag_reset((u64)x) - PAGE_OFFSET) / PAGE_SIZE; \
418 u64 __addr = VMEMMAP_START + (__idx * sizeof(struct page)); \
419 (struct page *)__addr; \
420 })
421 #endif /* CONFIG_DEBUG_VIRTUAL */
422
423 #define virt_addr_valid(addr) ({ \
424 __typeof__(addr) __addr = __tag_reset(addr); \
425 __is_lm_address(__addr) && pfn_is_map_memory(virt_to_pfn(__addr)); \
426 })
427
428 void dump_mem_limit(void);
429 #endif /* !ASSEMBLY */
430
431 /*
432 * Given that the GIC architecture permits ITS implementations that can only be
433 * configured with a LPI table address once, GICv3 systems with many CPUs may
434 * end up reserving a lot of different regions after a kexec for their LPI
435 * tables (one per CPU), as we are forced to reuse the same memory after kexec
436 * (and thus reserve it persistently with EFI beforehand)
437 */
438 #if defined(CONFIG_EFI) && defined(CONFIG_ARM_GIC_V3_ITS)
439 # define INIT_MEMBLOCK_RESERVED_REGIONS (INIT_MEMBLOCK_REGIONS + NR_CPUS + 1)
440 #endif
441
442 /*
443 * memory regions which marked with flag MEMBLOCK_NOMAP(for example, the memory
444 * of the EFI_UNUSABLE_MEMORY type) may divide a continuous memory block into
445 * multiple parts. As a result, the number of memory regions is large.
446 */
447 #ifdef CONFIG_EFI
448 #define INIT_MEMBLOCK_MEMORY_REGIONS (INIT_MEMBLOCK_REGIONS * 8)
449 #endif
450
451
452 #endif /* __ASM_MEMORY_H */
453