projects/linux·blob master · refs

452 lines · 14.4 KiB

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 DIRECT_MAP_PHYSMEM_END __pa(PAGE_END - 1)
114
115#define MIN_THREAD_SHIFT (14 + KASAN_THREAD_SHIFT)
116
117/*
118 * VMAP'd stacks are allocated at page granularity, so we must ensure that such
119 * stacks are a multiple of page size.
120 */
121#if (MIN_THREAD_SHIFT < PAGE_SHIFT)
122#define THREAD_SHIFT PAGE_SHIFT
123#else
124#define THREAD_SHIFT MIN_THREAD_SHIFT
125#endif
126
127#if THREAD_SHIFT >= PAGE_SHIFT
128#define THREAD_SIZE_ORDER (THREAD_SHIFT - PAGE_SHIFT)
129#endif
130
131#define THREAD_SIZE (UL(1) << THREAD_SHIFT)
132
133/*
134 * By aligning VMAP'd stacks to 2 * THREAD_SIZE, we can detect overflow by
135 * checking sp & (1 << THREAD_SHIFT), which we can do cheaply in the entry
136 * assembly.
137 */
138#define THREAD_ALIGN (2 * THREAD_SIZE)
139
140#define IRQ_STACK_SIZE THREAD_SIZE
141
142#define OVERFLOW_STACK_SIZE SZ_4K
143
144#define NVHE_STACK_SHIFT PAGE_SHIFT
145#define NVHE_STACK_SIZE (UL(1) << NVHE_STACK_SHIFT)
146
147/*
148 * With the minimum frame size of [x29, x30], exactly half the combined
149 * sizes of the hyp and overflow stacks is the maximum size needed to
150 * save the unwinded stacktrace; plus an additional entry to delimit the
151 * end.
152 */
153#define NVHE_STACKTRACE_SIZE ((OVERFLOW_STACK_SIZE + NVHE_STACK_SIZE) / 2 + sizeof(long))
154
155/*
156 * Alignment of kernel segments (e.g. .text, .data).
157 *
158 * 4 KB granule: 16 level 3 entries, with contiguous bit
159 * 16 KB granule: 4 level 3 entries, without contiguous bit
160 * 64 KB granule: 1 level 3 entry
161 */
162#define SEGMENT_ALIGN SZ_64K
163
164/*
165 * Memory types available.
166 *
167 * IMPORTANT: MT_NORMAL must be index 0 since vm_get_page_prot() may 'or' in
168 * the MT_NORMAL_TAGGED memory type for PROT_MTE mappings. Note
169 * that protection_map[] only contains MT_NORMAL attributes.
170 */
171#define MT_NORMAL 0
172#define MT_NORMAL_TAGGED 1
173#define MT_NORMAL_NC 2
174#define MT_DEVICE_nGnRnE 3
175#define MT_DEVICE_nGnRE 4
176
177/*
178 * Memory types for Stage-2 translation when HCR_EL2.FWB=0. See R_HMNDG,
179 * R_TNHFM, R_GQFSF and I_MCQKW for the details on how these attributes get
180 * combined with Stage-1.
181 */
182#define MT_S2_NORMAL 0xf
183#define MT_S2_NORMAL_NC 0x5
184#define MT_S2_DEVICE_nGnRE 0x1
185#define MT_S2_AS_S1 MT_S2_NORMAL
186
187/*
188 * Memory types for Stage-2 translation when HCR_EL2.FWB=1. Stage-2 enforces
189 * Normal-WB and Device-nGnRE, unless we actively say that S1 wins. See
190 * R_VRJSW and R_RHWZM for details.
191 */
192#define MT_S2_FWB_NORMAL 6
193#define MT_S2_FWB_NORMAL_NC 5
194#define MT_S2_FWB_DEVICE_nGnRE 1
195#define MT_S2_FWB_AS_S1 7
196
197#ifdef CONFIG_ARM64_4K_PAGES
198#define IOREMAP_MAX_ORDER (PUD_SHIFT)
199#else
200#define IOREMAP_MAX_ORDER (PMD_SHIFT)
201#endif
202
203/*
204 * Open-coded (swapper_pg_dir - reserved_pg_dir) as this cannot be calculated
205 * until link time.
206 */
207#define RESERVED_SWAPPER_OFFSET (PAGE_SIZE)
208
209/*
210 * Open-coded (swapper_pg_dir - tramp_pg_dir) as this cannot be calculated
211 * until link time.
212 */
213#define TRAMP_SWAPPER_OFFSET (2 * PAGE_SIZE)
214
215#ifndef __ASSEMBLER__
216
217#include <linux/bitops.h>
218#include <linux/compiler.h>
219#include <linux/mmdebug.h>
220#include <linux/types.h>
221#include <asm/boot.h>
222#include <asm/bug.h>
223#include <asm/sections.h>
224#include <asm/sysreg.h>
225
226static inline u64 __pure read_tcr(void)
227{
228 u64 tcr;
229
230 // read_sysreg() uses asm volatile, so avoid it here
231 asm("mrs %0, tcr_el1" : "=r"(tcr));
232 return tcr;
233}
234
235#if VA_BITS > 48
236// For reasons of #include hell, we can't use TCR_T1SZ_OFFSET/TCR_T1SZ_MASK here
237#define vabits_actual (64 - ((read_tcr() >> 16) & 63))
238#else
239#define vabits_actual ((u64)VA_BITS)
240#endif
241
242extern s64 memstart_addr;
243/* PHYS_OFFSET - the physical address of the start of memory. */
244#define PHYS_OFFSET ({ VM_BUG_ON(memstart_addr & 1); memstart_addr; })
245
246/* the offset between the kernel virtual and physical mappings */
247extern u64 kimage_voffset;
248
249static inline unsigned long kaslr_offset(void)
250{
251 return (u64)&_text - KIMAGE_VADDR;
252}
253
254#ifdef CONFIG_RANDOMIZE_BASE
255void kaslr_init(void);
256static inline bool kaslr_enabled(void)
257{
258 extern bool __kaslr_is_enabled;
259 return __kaslr_is_enabled;
260}
261#else
262static inline void kaslr_init(void) { }
263static inline bool kaslr_enabled(void) { return false; }
264#endif
265
266/*
267 * Allow all memory at the discovery stage. We will clip it later.
268 */
269#define MIN_MEMBLOCK_ADDR 0
270#define MAX_MEMBLOCK_ADDR U64_MAX
271
272/*
273 * PFNs are used to describe any physical page; this means
274 * PFN 0 == physical address 0.
275 *
276 * This is the PFN of the first RAM page in the kernel
277 * direct-mapped view. We assume this is the first page
278 * of RAM in the mem_map as well.
279 */
280#define PHYS_PFN_OFFSET (PHYS_OFFSET >> PAGE_SHIFT)
281
282/*
283 * When dealing with data aborts, watchpoints, or instruction traps we may end
284 * up with a tagged userland pointer. Clear the tag to get a sane pointer to
285 * pass on to access_ok(), for instance.
286 */
287#define __untagged_addr(addr) \
288 ((__force __typeof__(addr))sign_extend64((__force u64)(addr), 55))
289
290#define untagged_addr(addr) ({ \
291 u64 __addr = (__force u64)(addr); \
292 __addr &= __untagged_addr(__addr); \
293 (__force __typeof__(addr))__addr; \
294})
295
296#if defined(CONFIG_KASAN_SW_TAGS) || defined(CONFIG_KASAN_HW_TAGS)
297#define __tag_shifted(tag) ((u64)(tag) << 56)
298#define __tag_reset(addr) __untagged_addr(addr)
299#define __tag_get(addr) (__u8)((u64)(addr) >> 56)
300#else
301#define __tag_shifted(tag) 0UL
302#define __tag_reset(addr) (addr)
303#define __tag_get(addr) 0
304#endif /* CONFIG_KASAN_SW_TAGS || CONFIG_KASAN_HW_TAGS */
305
306static inline const void *__tag_set(const void *addr, u8 tag)
307{
308 u64 __addr = (u64)addr & ~__tag_shifted(0xff);
309 return (const void *)(__addr | __tag_shifted(tag));
310}
311
312#ifdef CONFIG_KASAN_HW_TAGS
313#define arch_enable_tag_checks_sync() mte_enable_kernel_sync()
314#define arch_enable_tag_checks_async() mte_enable_kernel_async()
315#define arch_enable_tag_checks_asymm() mte_enable_kernel_asymm()
316#define arch_enable_tag_checks_write_only() mte_enable_kernel_store_only()
317#define arch_suppress_tag_checks_start() mte_enable_tco()
318#define arch_suppress_tag_checks_stop() mte_disable_tco()
319#define arch_force_async_tag_fault() mte_check_tfsr_exit()
320#define arch_get_random_tag() mte_get_random_tag()
321#define arch_get_mem_tag(addr) mte_get_mem_tag(addr)
322#define arch_set_mem_tag_range(addr, size, tag, init) \
323 mte_set_mem_tag_range((addr), (size), (tag), (init))
324#endif /* CONFIG_KASAN_HW_TAGS */
325
326/*
327 * Physical vs virtual RAM address space conversion. These are
328 * private definitions which should NOT be used outside memory.h
329 * files. Use virt_to_phys/phys_to_virt/__pa/__va instead.
330 */
331
332
333/*
334 * Check whether an arbitrary address is within the linear map, which
335 * lives in the [PAGE_OFFSET, PAGE_END) interval at the bottom of the
336 * kernel's TTBR1 address range.
337 */
338#define __is_lm_address(addr) (((u64)(addr) - PAGE_OFFSET) < (PAGE_END - PAGE_OFFSET))
339
340#define __lm_to_phys(addr) (((addr) - PAGE_OFFSET) + PHYS_OFFSET)
341#define __kimg_to_phys(addr) ((addr) - kimage_voffset)
342
343#define __virt_to_phys_nodebug(x) ({ \
344 phys_addr_t __x = (phys_addr_t)(__tag_reset(x)); \
345 __is_lm_address(__x) ? __lm_to_phys(__x) : __kimg_to_phys(__x); \
346})
347
348#define __pa_symbol_nodebug(x) __kimg_to_phys((phys_addr_t)(x))
349
350#ifdef CONFIG_DEBUG_VIRTUAL
351extern phys_addr_t __virt_to_phys(unsigned long x);
352extern phys_addr_t __phys_addr_symbol(unsigned long x);
353#else
354#define __virt_to_phys(x) __virt_to_phys_nodebug(x)
355#define __phys_addr_symbol(x) __pa_symbol_nodebug(x)
356#endif /* CONFIG_DEBUG_VIRTUAL */
357
358#define __phys_to_virt(x) ((unsigned long)((x) - PHYS_OFFSET) | PAGE_OFFSET)
359#define __phys_to_kimg(x) ((unsigned long)((x) + kimage_voffset))
360
361/*
362 * Note: Drivers should NOT use these. They are the wrong
363 * translation for translating DMA addresses. Use the driver
364 * DMA support - see dma-mapping.h.
365 */
366#define virt_to_phys virt_to_phys
367static inline phys_addr_t virt_to_phys(const volatile void *x)
368{
369 return __virt_to_phys((unsigned long)(x));
370}
371
372#define phys_to_virt phys_to_virt
373static inline void *phys_to_virt(phys_addr_t x)
374{
375 return (void *)(__phys_to_virt(x));
376}
377
378/* Needed already here for resolving __phys_to_pfn() in virt_to_pfn() */
379#include <asm-generic/memory_model.h>
380
381static inline unsigned long virt_to_pfn(const void *kaddr)
382{
383 return __phys_to_pfn(virt_to_phys(kaddr));
384}
385
386/*
387 * Drivers should NOT use these either.
388 */
389#define __pa(x) __virt_to_phys((unsigned long)(x))
390#define __pa_symbol(x) __phys_addr_symbol(RELOC_HIDE((unsigned long)(x), 0))
391#define __pa_nodebug(x) __virt_to_phys_nodebug((unsigned long)(x))
392#define __va(x) ((void *)__phys_to_virt((phys_addr_t)(x)))
393#define pfn_to_kaddr(pfn) __va((pfn) << PAGE_SHIFT)
394#define sym_to_pfn(x) __phys_to_pfn(__pa_symbol(x))
395
396/*
397 * virt_to_page(x) convert a _valid_ virtual address to struct page *
398 * virt_addr_valid(x) indicates whether a virtual address is valid
399 */
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
428void dump_mem_limit(void);
429#endif /* !__ASSEMBLER__ */
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 */