xref: /linux/arch/arm64/include/asm/io.h (revision f4cdf7ca9a1fdcca413157df19753f388a5a224e)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  * Based on arch/arm/include/asm/io.h
4  *
5  * Copyright (C) 1996-2000 Russell King
6  * Copyright (C) 2012 ARM Ltd.
7  */
8 #ifndef __ASM_IO_H
9 #define __ASM_IO_H
10 
11 #include <linux/types.h>
12 #include <linux/pgtable.h>
13 
14 #include <asm/byteorder.h>
15 #include <asm/barrier.h>
16 #include <asm/memory.h>
17 #include <asm/early_ioremap.h>
18 #include <asm/alternative.h>
19 #include <asm/cpufeature.h>
20 #include <asm/rsi.h>
21 
22 /*
23  * Generic IO read/write.  These perform native-endian accesses.
24  */
25 #define __raw_writeb __raw_writeb
26 static __always_inline void __raw_writeb(u8 val, volatile void __iomem *addr)
27 {
28 	volatile u8 __iomem *ptr = addr;
29 	asm volatile("strb %w0, %1" : : "rZ" (val), "Qo" (*ptr));
30 }
31 
32 #define __raw_writew __raw_writew
33 static __always_inline void __raw_writew(u16 val, volatile void __iomem *addr)
34 {
35 	volatile u16 __iomem *ptr = addr;
36 	asm volatile("strh %w0, %1" : : "rZ" (val), "Qo" (*ptr));
37 }
38 
39 #define __raw_writel __raw_writel
40 static __always_inline void __raw_writel(u32 val, volatile void __iomem *addr)
41 {
42 	volatile u32 __iomem *ptr = addr;
43 	asm volatile("str %w0, %1" : : "rZ" (val), "Qo" (*ptr));
44 }
45 
46 #define __raw_writeq __raw_writeq
47 static __always_inline void __raw_writeq(u64 val, volatile void __iomem *addr)
48 {
49 	volatile u64 __iomem *ptr = addr;
50 	asm volatile("str %x0, %1" : : "rZ" (val), "Qo" (*ptr));
51 }
52 
53 #define __raw_readb __raw_readb
54 static __always_inline u8 __raw_readb(const volatile void __iomem *addr)
55 {
56 	u8 val;
57 	asm volatile(ALTERNATIVE("nop", "dmb osh",
58 				 ARM64_WORKAROUND_NVIDIA_OLYMPUS_1027)
59 		     ALTERNATIVE("ldrb %w0, [%1]",
60 				 "ldarb %w0, [%1]",
61 				 ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE)
62 		     : "=r" (val) : "r" (addr));
63 	return val;
64 }
65 
66 #define __raw_readw __raw_readw
67 static __always_inline u16 __raw_readw(const volatile void __iomem *addr)
68 {
69 	u16 val;
70 
71 	asm volatile(ALTERNATIVE("nop", "dmb osh",
72 				 ARM64_WORKAROUND_NVIDIA_OLYMPUS_1027)
73 		     ALTERNATIVE("ldrh %w0, [%1]",
74 				 "ldarh %w0, [%1]",
75 				 ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE)
76 		     : "=r" (val) : "r" (addr));
77 	return val;
78 }
79 
80 #define __raw_readl __raw_readl
81 static __always_inline u32 __raw_readl(const volatile void __iomem *addr)
82 {
83 	u32 val;
84 	asm volatile(ALTERNATIVE("nop", "dmb osh",
85 				 ARM64_WORKAROUND_NVIDIA_OLYMPUS_1027)
86 		     ALTERNATIVE("ldr %w0, [%1]",
87 				 "ldar %w0, [%1]",
88 				 ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE)
89 		     : "=r" (val) : "r" (addr));
90 	return val;
91 }
92 
93 #define __raw_readq __raw_readq
94 static __always_inline u64 __raw_readq(const volatile void __iomem *addr)
95 {
96 	u64 val;
97 	asm volatile(ALTERNATIVE("nop", "dmb osh",
98 				 ARM64_WORKAROUND_NVIDIA_OLYMPUS_1027)
99 		     ALTERNATIVE("ldr %0, [%1]",
100 				 "ldar %0, [%1]",
101 				 ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE)
102 		     : "=r" (val) : "r" (addr));
103 	return val;
104 }
105 
106 /* IO barriers */
107 #define __io_ar(v)							\
108 ({									\
109 	unsigned long tmp;						\
110 									\
111 	dma_rmb();								\
112 									\
113 	/*								\
114 	 * Create a dummy control dependency from the IO read to any	\
115 	 * later instructions. This ensures that a subsequent call to	\
116 	 * udelay() will be ordered due to the ISB in get_cycles().	\
117 	 */								\
118 	asm volatile("eor	%0, %1, %1\n"				\
119 		     "cbnz	%0, ."					\
120 		     : "=r" (tmp) : "r" ((unsigned long)(v))		\
121 		     : "memory");					\
122 })
123 
124 #define __io_bw()		dma_wmb()
125 #define __io_br(v)
126 #define __io_aw(v)
127 
128 /* arm64-specific, don't use in portable drivers */
129 #define __iormb(v)		__io_ar(v)
130 #define __iowmb()		__io_bw()
131 #define __iomb()		dma_mb()
132 
133 /*
134  *  I/O port access primitives.
135  */
136 #define arch_has_dev_port()	(1)
137 #define IO_SPACE_LIMIT		(PCI_IO_SIZE - 1)
138 #define PCI_IOBASE		((void __iomem *)PCI_IO_START)
139 
140 /*
141  * The ARM64 iowrite implementation is intended to support drivers that want to
142  * use write combining. For instance PCI drivers using write combining with a 64
143  * byte __iowrite64_copy() expect to get a 64 byte MemWr TLP on the PCIe bus.
144  *
145  * Newer ARM core have sensitive write combining buffers, it is important that
146  * the stores be contiguous blocks of store instructions. Normal memcpy
147  * approaches have a very low chance to generate write combining.
148  *
149  * Since this is the only API on ARM64 that should be used with write combining
150  * it also integrates the DGH hint which is supposed to lower the latency to
151  * emit the large TLP from the CPU.
152  */
153 
154 static __always_inline void
155 __const_memcpy_toio_aligned32(volatile u32 __iomem *to, const u32 *from,
156 			      size_t count)
157 {
158 	switch (count) {
159 	case 8:
160 		asm volatile("str %w0, [%8, #4 * 0]\n"
161 			     "str %w1, [%8, #4 * 1]\n"
162 			     "str %w2, [%8, #4 * 2]\n"
163 			     "str %w3, [%8, #4 * 3]\n"
164 			     "str %w4, [%8, #4 * 4]\n"
165 			     "str %w5, [%8, #4 * 5]\n"
166 			     "str %w6, [%8, #4 * 6]\n"
167 			     "str %w7, [%8, #4 * 7]\n"
168 			     :
169 			     : "rZ"(from[0]), "rZ"(from[1]), "rZ"(from[2]),
170 			       "rZ"(from[3]), "rZ"(from[4]), "rZ"(from[5]),
171 			       "rZ"(from[6]), "rZ"(from[7]), "r"(to));
172 		break;
173 	case 4:
174 		asm volatile("str %w0, [%4, #4 * 0]\n"
175 			     "str %w1, [%4, #4 * 1]\n"
176 			     "str %w2, [%4, #4 * 2]\n"
177 			     "str %w3, [%4, #4 * 3]\n"
178 			     :
179 			     : "rZ"(from[0]), "rZ"(from[1]), "rZ"(from[2]),
180 			       "rZ"(from[3]), "r"(to));
181 		break;
182 	case 2:
183 		asm volatile("str %w0, [%2, #4 * 0]\n"
184 			     "str %w1, [%2, #4 * 1]\n"
185 			     :
186 			     : "rZ"(from[0]), "rZ"(from[1]), "r"(to));
187 		break;
188 	case 1:
189 		__raw_writel(*from, to);
190 		break;
191 	default:
192 		BUILD_BUG();
193 	}
194 }
195 
196 void __iowrite32_copy_full(void __iomem *to, const void *from, size_t count);
197 
198 static __always_inline void
199 __iowrite32_copy(void __iomem *to, const void *from, size_t count)
200 {
201 	if (__builtin_constant_p(count) &&
202 	    (count == 8 || count == 4 || count == 2 || count == 1)) {
203 		__const_memcpy_toio_aligned32(to, from, count);
204 		dgh();
205 	} else {
206 		__iowrite32_copy_full(to, from, count);
207 	}
208 }
209 #define __iowrite32_copy __iowrite32_copy
210 
211 static __always_inline void
212 __const_memcpy_toio_aligned64(volatile u64 __iomem *to, const u64 *from,
213 			      size_t count)
214 {
215 	switch (count) {
216 	case 8:
217 		asm volatile("str %x0, [%8, #8 * 0]\n"
218 			     "str %x1, [%8, #8 * 1]\n"
219 			     "str %x2, [%8, #8 * 2]\n"
220 			     "str %x3, [%8, #8 * 3]\n"
221 			     "str %x4, [%8, #8 * 4]\n"
222 			     "str %x5, [%8, #8 * 5]\n"
223 			     "str %x6, [%8, #8 * 6]\n"
224 			     "str %x7, [%8, #8 * 7]\n"
225 			     :
226 			     : "rZ"(from[0]), "rZ"(from[1]), "rZ"(from[2]),
227 			       "rZ"(from[3]), "rZ"(from[4]), "rZ"(from[5]),
228 			       "rZ"(from[6]), "rZ"(from[7]), "r"(to));
229 		break;
230 	case 4:
231 		asm volatile("str %x0, [%4, #8 * 0]\n"
232 			     "str %x1, [%4, #8 * 1]\n"
233 			     "str %x2, [%4, #8 * 2]\n"
234 			     "str %x3, [%4, #8 * 3]\n"
235 			     :
236 			     : "rZ"(from[0]), "rZ"(from[1]), "rZ"(from[2]),
237 			       "rZ"(from[3]), "r"(to));
238 		break;
239 	case 2:
240 		asm volatile("str %x0, [%2, #8 * 0]\n"
241 			     "str %x1, [%2, #8 * 1]\n"
242 			     :
243 			     : "rZ"(from[0]), "rZ"(from[1]), "r"(to));
244 		break;
245 	case 1:
246 		__raw_writeq(*from, to);
247 		break;
248 	default:
249 		BUILD_BUG();
250 	}
251 }
252 
253 void __iowrite64_copy_full(void __iomem *to, const void *from, size_t count);
254 
255 static __always_inline void
256 __iowrite64_copy(void __iomem *to, const void *from, size_t count)
257 {
258 	if (__builtin_constant_p(count) &&
259 	    (count == 8 || count == 4 || count == 2 || count == 1)) {
260 		__const_memcpy_toio_aligned64(to, from, count);
261 		dgh();
262 	} else {
263 		__iowrite64_copy_full(to, from, count);
264 	}
265 }
266 #define __iowrite64_copy __iowrite64_copy
267 
268 /*
269  * I/O memory mapping functions.
270  */
271 
272 typedef int (*ioremap_prot_hook_t)(phys_addr_t phys_addr, size_t size,
273 				   pgprot_t *prot);
274 int arm64_ioremap_prot_hook_register(const ioremap_prot_hook_t hook);
275 void __iomem *__ioremap_prot(phys_addr_t phys, size_t size, pgprot_t prot);
276 
277 static inline void __iomem *ioremap_prot(phys_addr_t phys, size_t size,
278 					 pgprot_t user_prot)
279 {
280 	pgprot_t prot;
281 	ptval_t user_prot_val = pgprot_val(user_prot);
282 
283 	if (WARN_ON_ONCE(!(user_prot_val & PTE_USER)))
284 		return NULL;
285 
286 	prot = __pgprot_modify(PAGE_KERNEL, PTE_ATTRINDX_MASK,
287 			       user_prot_val & PTE_ATTRINDX_MASK);
288 	return __ioremap_prot(phys, size, prot);
289 }
290 #define ioremap_prot ioremap_prot
291 
292 #define ioremap(addr, size)	\
293 	__ioremap_prot((addr), (size), __pgprot(PROT_DEVICE_nGnRE))
294 #define ioremap_wc(addr, size)	\
295 	__ioremap_prot((addr), (size), __pgprot(PROT_NORMAL_NC))
296 #define ioremap_np(addr, size)	\
297 	__ioremap_prot((addr), (size), __pgprot(PROT_DEVICE_nGnRnE))
298 
299 
300 #define ioremap_encrypted(addr, size)	\
301 	__ioremap_prot((addr), (size), PAGE_KERNEL)
302 
303 /*
304  * io{read,write}{16,32,64}be() macros
305  */
306 #define ioread16be(p)		({ __u16 __v = be16_to_cpu((__force __be16)__raw_readw(p)); __iormb(__v); __v; })
307 #define ioread32be(p)		({ __u32 __v = be32_to_cpu((__force __be32)__raw_readl(p)); __iormb(__v); __v; })
308 #define ioread64be(p)		({ __u64 __v = be64_to_cpu((__force __be64)__raw_readq(p)); __iormb(__v); __v; })
309 
310 #define iowrite16be(v,p)	({ __iowmb(); __raw_writew((__force __u16)cpu_to_be16(v), p); })
311 #define iowrite32be(v,p)	({ __iowmb(); __raw_writel((__force __u32)cpu_to_be32(v), p); })
312 #define iowrite64be(v,p)	({ __iowmb(); __raw_writeq((__force __u64)cpu_to_be64(v), p); })
313 
314 #include <asm-generic/io.h>
315 
316 #define ioremap_cache ioremap_cache
317 static inline void __iomem *ioremap_cache(phys_addr_t addr, size_t size)
318 {
319 	if (pfn_is_map_memory(__phys_to_pfn(addr)))
320 		return (void __iomem *)__phys_to_virt(addr);
321 
322 	return __ioremap_prot(addr, size, __pgprot(PROT_NORMAL));
323 }
324 
325 /*
326  * More restrictive address range checking than the default implementation
327  * (PHYS_OFFSET and PHYS_MASK taken into account).
328  */
329 #define ARCH_HAS_VALID_PHYS_ADDR_RANGE
330 extern int valid_phys_addr_range(phys_addr_t addr, size_t size);
331 extern int valid_mmap_phys_addr_range(unsigned long pfn, size_t size);
332 
333 extern bool arch_memremap_can_ram_remap(resource_size_t offset, size_t size,
334 					unsigned long flags);
335 #define arch_memremap_can_ram_remap arch_memremap_can_ram_remap
336 
337 static inline bool arm64_is_protected_mmio(phys_addr_t phys_addr, size_t size)
338 {
339 	if (unlikely(is_realm_world()))
340 		return arm64_rsi_is_protected(phys_addr, size);
341 	return false;
342 }
343 
344 #endif	/* __ASM_IO_H */
345