xref: /linux/arch/arm64/include/asm/memory.h (revision 9684ec186f8fadde52d6b6eaf64ca508897d0c71)
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_DEVICE_nGnRE	0x1
181 
182 /*
183  * Memory types for Stage-2 translation when ID_AA64MMFR2_EL1.FWB is 0001
184  * Stage-2 enforces Normal-WB and Device-nGnRE
185  */
186 #define MT_S2_FWB_NORMAL	6
187 #define MT_S2_FWB_DEVICE_nGnRE	1
188 
189 #ifdef CONFIG_ARM64_4K_PAGES
190 #define IOREMAP_MAX_ORDER	(PUD_SHIFT)
191 #else
192 #define IOREMAP_MAX_ORDER	(PMD_SHIFT)
193 #endif
194 
195 /*
196  *  Open-coded (swapper_pg_dir - reserved_pg_dir) as this cannot be calculated
197  *  until link time.
198  */
199 #define RESERVED_SWAPPER_OFFSET	(PAGE_SIZE)
200 
201 /*
202  *  Open-coded (swapper_pg_dir - tramp_pg_dir) as this cannot be calculated
203  *  until link time.
204  */
205 #define TRAMP_SWAPPER_OFFSET	(2 * PAGE_SIZE)
206 
207 #ifndef __ASSEMBLY__
208 
209 #include <linux/bitops.h>
210 #include <linux/compiler.h>
211 #include <linux/mmdebug.h>
212 #include <linux/types.h>
213 #include <asm/boot.h>
214 #include <asm/bug.h>
215 #include <asm/sections.h>
216 #include <asm/sysreg.h>
217 
218 static inline u64 __pure read_tcr(void)
219 {
220 	u64  tcr;
221 
222 	// read_sysreg() uses asm volatile, so avoid it here
223 	asm("mrs %0, tcr_el1" : "=r"(tcr));
224 	return tcr;
225 }
226 
227 #if VA_BITS > 48
228 // For reasons of #include hell, we can't use TCR_T1SZ_OFFSET/TCR_T1SZ_MASK here
229 #define vabits_actual		(64 - ((read_tcr() >> 16) & 63))
230 #else
231 #define vabits_actual		((u64)VA_BITS)
232 #endif
233 
234 extern s64			memstart_addr;
235 /* PHYS_OFFSET - the physical address of the start of memory. */
236 #define PHYS_OFFSET		({ VM_BUG_ON(memstart_addr & 1); memstart_addr; })
237 
238 /* the offset between the kernel virtual and physical mappings */
239 extern u64			kimage_voffset;
240 
241 static inline unsigned long kaslr_offset(void)
242 {
243 	return (u64)&_text - KIMAGE_VADDR;
244 }
245 
246 #ifdef CONFIG_RANDOMIZE_BASE
247 void kaslr_init(void);
248 static inline bool kaslr_enabled(void)
249 {
250 	extern bool __kaslr_is_enabled;
251 	return __kaslr_is_enabled;
252 }
253 #else
254 static inline void kaslr_init(void) { }
255 static inline bool kaslr_enabled(void) { return false; }
256 #endif
257 
258 /*
259  * Allow all memory at the discovery stage. We will clip it later.
260  */
261 #define MIN_MEMBLOCK_ADDR	0
262 #define MAX_MEMBLOCK_ADDR	U64_MAX
263 
264 /*
265  * PFNs are used to describe any physical page; this means
266  * PFN 0 == physical address 0.
267  *
268  * This is the PFN of the first RAM page in the kernel
269  * direct-mapped view.  We assume this is the first page
270  * of RAM in the mem_map as well.
271  */
272 #define PHYS_PFN_OFFSET	(PHYS_OFFSET >> PAGE_SHIFT)
273 
274 /*
275  * When dealing with data aborts, watchpoints, or instruction traps we may end
276  * up with a tagged userland pointer. Clear the tag to get a sane pointer to
277  * pass on to access_ok(), for instance.
278  */
279 #define __untagged_addr(addr)	\
280 	((__force __typeof__(addr))sign_extend64((__force u64)(addr), 55))
281 
282 #define untagged_addr(addr)	({					\
283 	u64 __addr = (__force u64)(addr);					\
284 	__addr &= __untagged_addr(__addr);				\
285 	(__force __typeof__(addr))__addr;				\
286 })
287 
288 #if defined(CONFIG_KASAN_SW_TAGS) || defined(CONFIG_KASAN_HW_TAGS)
289 #define __tag_shifted(tag)	((u64)(tag) << 56)
290 #define __tag_reset(addr)	__untagged_addr(addr)
291 #define __tag_get(addr)		(__u8)((u64)(addr) >> 56)
292 #else
293 #define __tag_shifted(tag)	0UL
294 #define __tag_reset(addr)	(addr)
295 #define __tag_get(addr)		0
296 #endif /* CONFIG_KASAN_SW_TAGS || CONFIG_KASAN_HW_TAGS */
297 
298 static inline const void *__tag_set(const void *addr, u8 tag)
299 {
300 	u64 __addr = (u64)addr & ~__tag_shifted(0xff);
301 	return (const void *)(__addr | __tag_shifted(tag));
302 }
303 
304 #ifdef CONFIG_KASAN_HW_TAGS
305 #define arch_enable_tag_checks_sync()		mte_enable_kernel_sync()
306 #define arch_enable_tag_checks_async()		mte_enable_kernel_async()
307 #define arch_enable_tag_checks_asymm()		mte_enable_kernel_asymm()
308 #define arch_suppress_tag_checks_start()	mte_enable_tco()
309 #define arch_suppress_tag_checks_stop()		mte_disable_tco()
310 #define arch_force_async_tag_fault()		mte_check_tfsr_exit()
311 #define arch_get_random_tag()			mte_get_random_tag()
312 #define arch_get_mem_tag(addr)			mte_get_mem_tag(addr)
313 #define arch_set_mem_tag_range(addr, size, tag, init)	\
314 			mte_set_mem_tag_range((addr), (size), (tag), (init))
315 #endif /* CONFIG_KASAN_HW_TAGS */
316 
317 /*
318  * Physical vs virtual RAM address space conversion.  These are
319  * private definitions which should NOT be used outside memory.h
320  * files.  Use virt_to_phys/phys_to_virt/__pa/__va instead.
321  */
322 
323 
324 /*
325  * Check whether an arbitrary address is within the linear map, which
326  * lives in the [PAGE_OFFSET, PAGE_END) interval at the bottom of the
327  * kernel's TTBR1 address range.
328  */
329 #define __is_lm_address(addr)	(((u64)(addr) - PAGE_OFFSET) < (PAGE_END - PAGE_OFFSET))
330 
331 #define __lm_to_phys(addr)	(((addr) - PAGE_OFFSET) + PHYS_OFFSET)
332 #define __kimg_to_phys(addr)	((addr) - kimage_voffset)
333 
334 #define __virt_to_phys_nodebug(x) ({					\
335 	phys_addr_t __x = (phys_addr_t)(__tag_reset(x));		\
336 	__is_lm_address(__x) ? __lm_to_phys(__x) : __kimg_to_phys(__x);	\
337 })
338 
339 #define __pa_symbol_nodebug(x)	__kimg_to_phys((phys_addr_t)(x))
340 
341 #ifdef CONFIG_DEBUG_VIRTUAL
342 extern phys_addr_t __virt_to_phys(unsigned long x);
343 extern phys_addr_t __phys_addr_symbol(unsigned long x);
344 #else
345 #define __virt_to_phys(x)	__virt_to_phys_nodebug(x)
346 #define __phys_addr_symbol(x)	__pa_symbol_nodebug(x)
347 #endif /* CONFIG_DEBUG_VIRTUAL */
348 
349 #define __phys_to_virt(x)	((unsigned long)((x) - PHYS_OFFSET) | PAGE_OFFSET)
350 #define __phys_to_kimg(x)	((unsigned long)((x) + kimage_voffset))
351 
352 /*
353  * Convert a page to/from a physical address
354  */
355 #define page_to_phys(page)	(__pfn_to_phys(page_to_pfn(page)))
356 #define phys_to_page(phys)	(pfn_to_page(__phys_to_pfn(phys)))
357 
358 /*
359  * Note: Drivers should NOT use these.  They are the wrong
360  * translation for translating DMA addresses.  Use the driver
361  * DMA support - see dma-mapping.h.
362  */
363 #define virt_to_phys virt_to_phys
364 static inline phys_addr_t virt_to_phys(const volatile void *x)
365 {
366 	return __virt_to_phys((unsigned long)(x));
367 }
368 
369 #define phys_to_virt phys_to_virt
370 static inline void *phys_to_virt(phys_addr_t x)
371 {
372 	return (void *)(__phys_to_virt(x));
373 }
374 
375 /* Needed already here for resolving __phys_to_pfn() in virt_to_pfn() */
376 #include <asm-generic/memory_model.h>
377 
378 static inline unsigned long virt_to_pfn(const void *kaddr)
379 {
380 	return __phys_to_pfn(virt_to_phys(kaddr));
381 }
382 
383 /*
384  * Drivers should NOT use these either.
385  */
386 #define __pa(x)			__virt_to_phys((unsigned long)(x))
387 #define __pa_symbol(x)		__phys_addr_symbol(RELOC_HIDE((unsigned long)(x), 0))
388 #define __pa_nodebug(x)		__virt_to_phys_nodebug((unsigned long)(x))
389 #define __va(x)			((void *)__phys_to_virt((phys_addr_t)(x)))
390 #define pfn_to_kaddr(pfn)	__va((pfn) << PAGE_SHIFT)
391 #define sym_to_pfn(x)		__phys_to_pfn(__pa_symbol(x))
392 
393 /*
394  *  virt_to_page(x)	convert a _valid_ virtual address to struct page *
395  *  virt_addr_valid(x)	indicates whether a virtual address is valid
396  */
397 #define ARCH_PFN_OFFSET		((unsigned long)PHYS_PFN_OFFSET)
398 
399 #if defined(CONFIG_DEBUG_VIRTUAL)
400 #define page_to_virt(x)	({						\
401 	__typeof__(x) __page = x;					\
402 	void *__addr = __va(page_to_phys(__page));			\
403 	(void *)__tag_set((const void *)__addr, page_kasan_tag(__page));\
404 })
405 #define virt_to_page(x)		pfn_to_page(virt_to_pfn(x))
406 #else
407 #define page_to_virt(x)	({						\
408 	__typeof__(x) __page = x;					\
409 	u64 __idx = ((u64)__page - VMEMMAP_START) / sizeof(struct page);\
410 	u64 __addr = PAGE_OFFSET + (__idx * PAGE_SIZE);			\
411 	(void *)__tag_set((const void *)__addr, page_kasan_tag(__page));\
412 })
413 
414 #define virt_to_page(x)	({						\
415 	u64 __idx = (__tag_reset((u64)x) - PAGE_OFFSET) / PAGE_SIZE;	\
416 	u64 __addr = VMEMMAP_START + (__idx * sizeof(struct page));	\
417 	(struct page *)__addr;						\
418 })
419 #endif /* CONFIG_DEBUG_VIRTUAL */
420 
421 #define virt_addr_valid(addr)	({					\
422 	__typeof__(addr) __addr = __tag_reset(addr);			\
423 	__is_lm_address(__addr) && pfn_is_map_memory(virt_to_pfn(__addr));	\
424 })
425 
426 void dump_mem_limit(void);
427 #endif /* !ASSEMBLY */
428 
429 /*
430  * Given that the GIC architecture permits ITS implementations that can only be
431  * configured with a LPI table address once, GICv3 systems with many CPUs may
432  * end up reserving a lot of different regions after a kexec for their LPI
433  * tables (one per CPU), as we are forced to reuse the same memory after kexec
434  * (and thus reserve it persistently with EFI beforehand)
435  */
436 #if defined(CONFIG_EFI) && defined(CONFIG_ARM_GIC_V3_ITS)
437 # define INIT_MEMBLOCK_RESERVED_REGIONS	(INIT_MEMBLOCK_REGIONS + NR_CPUS + 1)
438 #endif
439 
440 /*
441  * memory regions which marked with flag MEMBLOCK_NOMAP(for example, the memory
442  * of the EFI_UNUSABLE_MEMORY type) may divide a continuous memory block into
443  * multiple parts. As a result, the number of memory regions is large.
444  */
445 #ifdef CONFIG_EFI
446 #define INIT_MEMBLOCK_MEMORY_REGIONS	(INIT_MEMBLOCK_REGIONS * 8)
447 #endif
448 
449 
450 #endif /* __ASM_MEMORY_H */
451