xref: /openbmc/linux/arch/arm64/include/asm/memory.h (revision e6e8c6c2)
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 MODULES_END		(MODULES_VADDR + MODULES_VSIZE)
48 #define MODULES_VADDR		(_PAGE_END(VA_BITS_MIN))
49 #define MODULES_VSIZE		(SZ_128M)
50 #define VMEMMAP_START		(-(UL(1) << (VA_BITS - VMEMMAP_SHIFT)))
51 #define VMEMMAP_END		(VMEMMAP_START + VMEMMAP_SIZE)
52 #define PCI_IO_END		(VMEMMAP_START - SZ_8M)
53 #define PCI_IO_START		(PCI_IO_END - PCI_IO_SIZE)
54 #define FIXADDR_TOP		(VMEMMAP_START - SZ_32M)
55 
56 #if VA_BITS > 48
57 #define VA_BITS_MIN		(48)
58 #else
59 #define VA_BITS_MIN		(VA_BITS)
60 #endif
61 
62 #define _PAGE_END(va)		(-(UL(1) << ((va) - 1)))
63 
64 #define KERNEL_START		_text
65 #define KERNEL_END		_end
66 
67 /*
68  * Generic and tag-based KASAN require 1/8th and 1/16th of the kernel virtual
69  * address space for the shadow region respectively. They can bloat the stack
70  * significantly, so double the (minimum) stack size when they are in use.
71  */
72 #if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
73 #define KASAN_SHADOW_OFFSET	_AC(CONFIG_KASAN_SHADOW_OFFSET, UL)
74 #define KASAN_SHADOW_END	((UL(1) << (64 - KASAN_SHADOW_SCALE_SHIFT)) \
75 					+ KASAN_SHADOW_OFFSET)
76 #define PAGE_END		(KASAN_SHADOW_END - (1UL << (vabits_actual - KASAN_SHADOW_SCALE_SHIFT)))
77 #define KASAN_THREAD_SHIFT	1
78 #else
79 #define KASAN_THREAD_SHIFT	0
80 #define PAGE_END		(_PAGE_END(VA_BITS_MIN))
81 #endif /* CONFIG_KASAN */
82 
83 #define MIN_THREAD_SHIFT	(14 + KASAN_THREAD_SHIFT)
84 
85 /*
86  * VMAP'd stacks are allocated at page granularity, so we must ensure that such
87  * stacks are a multiple of page size.
88  */
89 #if defined(CONFIG_VMAP_STACK) && (MIN_THREAD_SHIFT < PAGE_SHIFT)
90 #define THREAD_SHIFT		PAGE_SHIFT
91 #else
92 #define THREAD_SHIFT		MIN_THREAD_SHIFT
93 #endif
94 
95 #if THREAD_SHIFT >= PAGE_SHIFT
96 #define THREAD_SIZE_ORDER	(THREAD_SHIFT - PAGE_SHIFT)
97 #endif
98 
99 #define THREAD_SIZE		(UL(1) << THREAD_SHIFT)
100 
101 /*
102  * By aligning VMAP'd stacks to 2 * THREAD_SIZE, we can detect overflow by
103  * checking sp & (1 << THREAD_SHIFT), which we can do cheaply in the entry
104  * assembly.
105  */
106 #ifdef CONFIG_VMAP_STACK
107 #define THREAD_ALIGN		(2 * THREAD_SIZE)
108 #else
109 #define THREAD_ALIGN		THREAD_SIZE
110 #endif
111 
112 #define IRQ_STACK_SIZE		THREAD_SIZE
113 
114 #define OVERFLOW_STACK_SIZE	SZ_4K
115 
116 /*
117  * With the minimum frame size of [x29, x30], exactly half the combined
118  * sizes of the hyp and overflow stacks is the maximum size needed to
119  * save the unwinded stacktrace; plus an additional entry to delimit the
120  * end.
121  */
122 #define NVHE_STACKTRACE_SIZE	((OVERFLOW_STACK_SIZE + PAGE_SIZE) / 2 + sizeof(long))
123 
124 /*
125  * Alignment of kernel segments (e.g. .text, .data).
126  *
127  *  4 KB granule:  16 level 3 entries, with contiguous bit
128  * 16 KB granule:   4 level 3 entries, without contiguous bit
129  * 64 KB granule:   1 level 3 entry
130  */
131 #define SEGMENT_ALIGN		SZ_64K
132 
133 /*
134  * Memory types available.
135  *
136  * IMPORTANT: MT_NORMAL must be index 0 since vm_get_page_prot() may 'or' in
137  *	      the MT_NORMAL_TAGGED memory type for PROT_MTE mappings. Note
138  *	      that protection_map[] only contains MT_NORMAL attributes.
139  */
140 #define MT_NORMAL		0
141 #define MT_NORMAL_TAGGED	1
142 #define MT_NORMAL_NC		2
143 #define MT_DEVICE_nGnRnE	3
144 #define MT_DEVICE_nGnRE		4
145 
146 /*
147  * Memory types for Stage-2 translation
148  */
149 #define MT_S2_NORMAL		0xf
150 #define MT_S2_DEVICE_nGnRE	0x1
151 
152 /*
153  * Memory types for Stage-2 translation when ID_AA64MMFR2_EL1.FWB is 0001
154  * Stage-2 enforces Normal-WB and Device-nGnRE
155  */
156 #define MT_S2_FWB_NORMAL	6
157 #define MT_S2_FWB_DEVICE_nGnRE	1
158 
159 #ifdef CONFIG_ARM64_4K_PAGES
160 #define IOREMAP_MAX_ORDER	(PUD_SHIFT)
161 #else
162 #define IOREMAP_MAX_ORDER	(PMD_SHIFT)
163 #endif
164 
165 /*
166  *  Open-coded (swapper_pg_dir - reserved_pg_dir) as this cannot be calculated
167  *  until link time.
168  */
169 #define RESERVED_SWAPPER_OFFSET	(PAGE_SIZE)
170 
171 /*
172  *  Open-coded (swapper_pg_dir - tramp_pg_dir) as this cannot be calculated
173  *  until link time.
174  */
175 #define TRAMP_SWAPPER_OFFSET	(2 * PAGE_SIZE)
176 
177 #ifndef __ASSEMBLY__
178 
179 #include <linux/bitops.h>
180 #include <linux/compiler.h>
181 #include <linux/mmdebug.h>
182 #include <linux/types.h>
183 #include <asm/bug.h>
184 
185 #if VA_BITS > 48
186 extern u64			vabits_actual;
187 #else
188 #define vabits_actual		((u64)VA_BITS)
189 #endif
190 
191 extern s64			memstart_addr;
192 /* PHYS_OFFSET - the physical address of the start of memory. */
193 #define PHYS_OFFSET		({ VM_BUG_ON(memstart_addr & 1); memstart_addr; })
194 
195 /* the virtual base of the kernel image */
196 extern u64			kimage_vaddr;
197 
198 /* the offset between the kernel virtual and physical mappings */
199 extern u64			kimage_voffset;
200 
201 static inline unsigned long kaslr_offset(void)
202 {
203 	return kimage_vaddr - KIMAGE_VADDR;
204 }
205 
206 /*
207  * Allow all memory at the discovery stage. We will clip it later.
208  */
209 #define MIN_MEMBLOCK_ADDR	0
210 #define MAX_MEMBLOCK_ADDR	U64_MAX
211 
212 /*
213  * PFNs are used to describe any physical page; this means
214  * PFN 0 == physical address 0.
215  *
216  * This is the PFN of the first RAM page in the kernel
217  * direct-mapped view.  We assume this is the first page
218  * of RAM in the mem_map as well.
219  */
220 #define PHYS_PFN_OFFSET	(PHYS_OFFSET >> PAGE_SHIFT)
221 
222 /*
223  * When dealing with data aborts, watchpoints, or instruction traps we may end
224  * up with a tagged userland pointer. Clear the tag to get a sane pointer to
225  * pass on to access_ok(), for instance.
226  */
227 #define __untagged_addr(addr)	\
228 	((__force __typeof__(addr))sign_extend64((__force u64)(addr), 55))
229 
230 #define untagged_addr(addr)	({					\
231 	u64 __addr = (__force u64)(addr);					\
232 	__addr &= __untagged_addr(__addr);				\
233 	(__force __typeof__(addr))__addr;				\
234 })
235 
236 #if defined(CONFIG_KASAN_SW_TAGS) || defined(CONFIG_KASAN_HW_TAGS)
237 #define __tag_shifted(tag)	((u64)(tag) << 56)
238 #define __tag_reset(addr)	__untagged_addr(addr)
239 #define __tag_get(addr)		(__u8)((u64)(addr) >> 56)
240 #else
241 #define __tag_shifted(tag)	0UL
242 #define __tag_reset(addr)	(addr)
243 #define __tag_get(addr)		0
244 #endif /* CONFIG_KASAN_SW_TAGS || CONFIG_KASAN_HW_TAGS */
245 
246 static inline const void *__tag_set(const void *addr, u8 tag)
247 {
248 	u64 __addr = (u64)addr & ~__tag_shifted(0xff);
249 	return (const void *)(__addr | __tag_shifted(tag));
250 }
251 
252 #ifdef CONFIG_KASAN_HW_TAGS
253 #define arch_enable_tagging_sync()		mte_enable_kernel_sync()
254 #define arch_enable_tagging_async()		mte_enable_kernel_async()
255 #define arch_enable_tagging_asymm()		mte_enable_kernel_asymm()
256 #define arch_force_async_tag_fault()		mte_check_tfsr_exit()
257 #define arch_get_random_tag()			mte_get_random_tag()
258 #define arch_get_mem_tag(addr)			mte_get_mem_tag(addr)
259 #define arch_set_mem_tag_range(addr, size, tag, init)	\
260 			mte_set_mem_tag_range((addr), (size), (tag), (init))
261 #endif /* CONFIG_KASAN_HW_TAGS */
262 
263 /*
264  * Physical vs virtual RAM address space conversion.  These are
265  * private definitions which should NOT be used outside memory.h
266  * files.  Use virt_to_phys/phys_to_virt/__pa/__va instead.
267  */
268 
269 
270 /*
271  * Check whether an arbitrary address is within the linear map, which
272  * lives in the [PAGE_OFFSET, PAGE_END) interval at the bottom of the
273  * kernel's TTBR1 address range.
274  */
275 #define __is_lm_address(addr)	(((u64)(addr) - PAGE_OFFSET) < (PAGE_END - PAGE_OFFSET))
276 
277 #define __lm_to_phys(addr)	(((addr) - PAGE_OFFSET) + PHYS_OFFSET)
278 #define __kimg_to_phys(addr)	((addr) - kimage_voffset)
279 
280 #define __virt_to_phys_nodebug(x) ({					\
281 	phys_addr_t __x = (phys_addr_t)(__tag_reset(x));		\
282 	__is_lm_address(__x) ? __lm_to_phys(__x) : __kimg_to_phys(__x);	\
283 })
284 
285 #define __pa_symbol_nodebug(x)	__kimg_to_phys((phys_addr_t)(x))
286 
287 #ifdef CONFIG_DEBUG_VIRTUAL
288 extern phys_addr_t __virt_to_phys(unsigned long x);
289 extern phys_addr_t __phys_addr_symbol(unsigned long x);
290 #else
291 #define __virt_to_phys(x)	__virt_to_phys_nodebug(x)
292 #define __phys_addr_symbol(x)	__pa_symbol_nodebug(x)
293 #endif /* CONFIG_DEBUG_VIRTUAL */
294 
295 #define __phys_to_virt(x)	((unsigned long)((x) - PHYS_OFFSET) | PAGE_OFFSET)
296 #define __phys_to_kimg(x)	((unsigned long)((x) + kimage_voffset))
297 
298 /*
299  * Convert a page to/from a physical address
300  */
301 #define page_to_phys(page)	(__pfn_to_phys(page_to_pfn(page)))
302 #define phys_to_page(phys)	(pfn_to_page(__phys_to_pfn(phys)))
303 
304 /*
305  * Note: Drivers should NOT use these.  They are the wrong
306  * translation for translating DMA addresses.  Use the driver
307  * DMA support - see dma-mapping.h.
308  */
309 #define virt_to_phys virt_to_phys
310 static inline phys_addr_t virt_to_phys(const volatile void *x)
311 {
312 	return __virt_to_phys((unsigned long)(x));
313 }
314 
315 #define phys_to_virt phys_to_virt
316 static inline void *phys_to_virt(phys_addr_t x)
317 {
318 	return (void *)(__phys_to_virt(x));
319 }
320 
321 /*
322  * Drivers should NOT use these either.
323  */
324 #define __pa(x)			__virt_to_phys((unsigned long)(x))
325 #define __pa_symbol(x)		__phys_addr_symbol(RELOC_HIDE((unsigned long)(x), 0))
326 #define __pa_nodebug(x)		__virt_to_phys_nodebug((unsigned long)(x))
327 #define __va(x)			((void *)__phys_to_virt((phys_addr_t)(x)))
328 #define pfn_to_kaddr(pfn)	__va((pfn) << PAGE_SHIFT)
329 #define virt_to_pfn(x)		__phys_to_pfn(__virt_to_phys((unsigned long)(x)))
330 #define sym_to_pfn(x)		__phys_to_pfn(__pa_symbol(x))
331 
332 /*
333  *  virt_to_page(x)	convert a _valid_ virtual address to struct page *
334  *  virt_addr_valid(x)	indicates whether a virtual address is valid
335  */
336 #define ARCH_PFN_OFFSET		((unsigned long)PHYS_PFN_OFFSET)
337 
338 #if defined(CONFIG_DEBUG_VIRTUAL)
339 #define page_to_virt(x)	({						\
340 	__typeof__(x) __page = x;					\
341 	void *__addr = __va(page_to_phys(__page));			\
342 	(void *)__tag_set((const void *)__addr, page_kasan_tag(__page));\
343 })
344 #define virt_to_page(x)		pfn_to_page(virt_to_pfn(x))
345 #else
346 #define page_to_virt(x)	({						\
347 	__typeof__(x) __page = x;					\
348 	u64 __idx = ((u64)__page - VMEMMAP_START) / sizeof(struct page);\
349 	u64 __addr = PAGE_OFFSET + (__idx * PAGE_SIZE);			\
350 	(void *)__tag_set((const void *)__addr, page_kasan_tag(__page));\
351 })
352 
353 #define virt_to_page(x)	({						\
354 	u64 __idx = (__tag_reset((u64)x) - PAGE_OFFSET) / PAGE_SIZE;	\
355 	u64 __addr = VMEMMAP_START + (__idx * sizeof(struct page));	\
356 	(struct page *)__addr;						\
357 })
358 #endif /* CONFIG_DEBUG_VIRTUAL */
359 
360 #define virt_addr_valid(addr)	({					\
361 	__typeof__(addr) __addr = __tag_reset(addr);			\
362 	__is_lm_address(__addr) && pfn_is_map_memory(virt_to_pfn(__addr));	\
363 })
364 
365 void dump_mem_limit(void);
366 
367 static inline bool defer_reserve_crashkernel(void)
368 {
369 	return IS_ENABLED(CONFIG_ZONE_DMA) || IS_ENABLED(CONFIG_ZONE_DMA32);
370 }
371 #endif /* !ASSEMBLY */
372 
373 /*
374  * Given that the GIC architecture permits ITS implementations that can only be
375  * configured with a LPI table address once, GICv3 systems with many CPUs may
376  * end up reserving a lot of different regions after a kexec for their LPI
377  * tables (one per CPU), as we are forced to reuse the same memory after kexec
378  * (and thus reserve it persistently with EFI beforehand)
379  */
380 #if defined(CONFIG_EFI) && defined(CONFIG_ARM_GIC_V3_ITS)
381 # define INIT_MEMBLOCK_RESERVED_REGIONS	(INIT_MEMBLOCK_REGIONS + NR_CPUS + 1)
382 #endif
383 
384 /*
385  * memory regions which marked with flag MEMBLOCK_NOMAP(for example, the memory
386  * of the EFI_UNUSABLE_MEMORY type) may divide a continuous memory block into
387  * multiple parts. As a result, the number of memory regions is large.
388  */
389 #ifdef CONFIG_EFI
390 #define INIT_MEMBLOCK_MEMORY_REGIONS	(INIT_MEMBLOCK_REGIONS * 8)
391 #endif
392 
393 #include <asm-generic/memory_model.h>
394 
395 #endif /* __ASM_MEMORY_H */
396