xref: /linux/arch/x86/lib/usercopy_64.c (revision d198b34f3855eee2571dda03eea75a09c7c31480)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * User address space access functions.
4  *
5  * Copyright 1997 Andi Kleen <ak@muc.de>
6  * Copyright 1997 Linus Torvalds
7  * Copyright 2002 Andi Kleen <ak@suse.de>
8  */
9 #include <linux/export.h>
10 #include <linux/uaccess.h>
11 #include <linux/highmem.h>
12 
13 /*
14  * Zero Userspace
15  */
16 
17 unsigned long __clear_user(void __user *addr, unsigned long size)
18 {
19 	long __d0;
20 	might_fault();
21 	/* no memory constraint because it doesn't change any memory gcc knows
22 	   about */
23 	stac();
24 	asm volatile(
25 		"	testq  %[size8],%[size8]\n"
26 		"	jz     4f\n"
27 		"0:	movq $0,(%[dst])\n"
28 		"	addq   $8,%[dst]\n"
29 		"	decl %%ecx ; jnz   0b\n"
30 		"4:	movq  %[size1],%%rcx\n"
31 		"	testl %%ecx,%%ecx\n"
32 		"	jz     2f\n"
33 		"1:	movb   $0,(%[dst])\n"
34 		"	incq   %[dst]\n"
35 		"	decl %%ecx ; jnz  1b\n"
36 		"2:\n"
37 		".section .fixup,\"ax\"\n"
38 		"3:	lea 0(%[size1],%[size8],8),%[size8]\n"
39 		"	jmp 2b\n"
40 		".previous\n"
41 		_ASM_EXTABLE_UA(0b, 3b)
42 		_ASM_EXTABLE_UA(1b, 2b)
43 		: [size8] "=&c"(size), [dst] "=&D" (__d0)
44 		: [size1] "r"(size & 7), "[size8]" (size / 8), "[dst]"(addr));
45 	clac();
46 	return size;
47 }
48 EXPORT_SYMBOL(__clear_user);
49 
50 unsigned long clear_user(void __user *to, unsigned long n)
51 {
52 	if (access_ok(to, n))
53 		return __clear_user(to, n);
54 	return n;
55 }
56 EXPORT_SYMBOL(clear_user);
57 
58 /*
59  * Similar to copy_user_handle_tail, probe for the write fault point,
60  * but reuse __memcpy_mcsafe in case a new read error is encountered.
61  * clac() is handled in _copy_to_iter_mcsafe().
62  */
63 __visible notrace unsigned long
64 mcsafe_handle_tail(char *to, char *from, unsigned len)
65 {
66 	for (; len; --len, to++, from++) {
67 		/*
68 		 * Call the assembly routine back directly since
69 		 * memcpy_mcsafe() may silently fallback to memcpy.
70 		 */
71 		unsigned long rem = __memcpy_mcsafe(to, from, 1);
72 
73 		if (rem)
74 			break;
75 	}
76 	return len;
77 }
78 
79 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
80 /**
81  * clean_cache_range - write back a cache range with CLWB
82  * @vaddr:	virtual start address
83  * @size:	number of bytes to write back
84  *
85  * Write back a cache range using the CLWB (cache line write back)
86  * instruction. Note that @size is internally rounded up to be cache
87  * line size aligned.
88  */
89 static void clean_cache_range(void *addr, size_t size)
90 {
91 	u16 x86_clflush_size = boot_cpu_data.x86_clflush_size;
92 	unsigned long clflush_mask = x86_clflush_size - 1;
93 	void *vend = addr + size;
94 	void *p;
95 
96 	for (p = (void *)((unsigned long)addr & ~clflush_mask);
97 	     p < vend; p += x86_clflush_size)
98 		clwb(p);
99 }
100 
101 void arch_wb_cache_pmem(void *addr, size_t size)
102 {
103 	clean_cache_range(addr, size);
104 }
105 EXPORT_SYMBOL_GPL(arch_wb_cache_pmem);
106 
107 long __copy_user_flushcache(void *dst, const void __user *src, unsigned size)
108 {
109 	unsigned long flushed, dest = (unsigned long) dst;
110 	long rc = __copy_user_nocache(dst, src, size, 0);
111 
112 	/*
113 	 * __copy_user_nocache() uses non-temporal stores for the bulk
114 	 * of the transfer, but we need to manually flush if the
115 	 * transfer is unaligned. A cached memory copy is used when
116 	 * destination or size is not naturally aligned. That is:
117 	 *   - Require 8-byte alignment when size is 8 bytes or larger.
118 	 *   - Require 4-byte alignment when size is 4 bytes.
119 	 */
120 	if (size < 8) {
121 		if (!IS_ALIGNED(dest, 4) || size != 4)
122 			clean_cache_range(dst, 1);
123 	} else {
124 		if (!IS_ALIGNED(dest, 8)) {
125 			dest = ALIGN(dest, boot_cpu_data.x86_clflush_size);
126 			clean_cache_range(dst, 1);
127 		}
128 
129 		flushed = dest - (unsigned long) dst;
130 		if (size > flushed && !IS_ALIGNED(size - flushed, 8))
131 			clean_cache_range(dst + size - 1, 1);
132 	}
133 
134 	return rc;
135 }
136 
137 void __memcpy_flushcache(void *_dst, const void *_src, size_t size)
138 {
139 	unsigned long dest = (unsigned long) _dst;
140 	unsigned long source = (unsigned long) _src;
141 
142 	/* cache copy and flush to align dest */
143 	if (!IS_ALIGNED(dest, 8)) {
144 		unsigned len = min_t(unsigned, size, ALIGN(dest, 8) - dest);
145 
146 		memcpy((void *) dest, (void *) source, len);
147 		clean_cache_range((void *) dest, len);
148 		dest += len;
149 		source += len;
150 		size -= len;
151 		if (!size)
152 			return;
153 	}
154 
155 	/* 4x8 movnti loop */
156 	while (size >= 32) {
157 		asm("movq    (%0), %%r8\n"
158 		    "movq   8(%0), %%r9\n"
159 		    "movq  16(%0), %%r10\n"
160 		    "movq  24(%0), %%r11\n"
161 		    "movnti  %%r8,   (%1)\n"
162 		    "movnti  %%r9,  8(%1)\n"
163 		    "movnti %%r10, 16(%1)\n"
164 		    "movnti %%r11, 24(%1)\n"
165 		    :: "r" (source), "r" (dest)
166 		    : "memory", "r8", "r9", "r10", "r11");
167 		dest += 32;
168 		source += 32;
169 		size -= 32;
170 	}
171 
172 	/* 1x8 movnti loop */
173 	while (size >= 8) {
174 		asm("movq    (%0), %%r8\n"
175 		    "movnti  %%r8,   (%1)\n"
176 		    :: "r" (source), "r" (dest)
177 		    : "memory", "r8");
178 		dest += 8;
179 		source += 8;
180 		size -= 8;
181 	}
182 
183 	/* 1x4 movnti loop */
184 	while (size >= 4) {
185 		asm("movl    (%0), %%r8d\n"
186 		    "movnti  %%r8d,   (%1)\n"
187 		    :: "r" (source), "r" (dest)
188 		    : "memory", "r8");
189 		dest += 4;
190 		source += 4;
191 		size -= 4;
192 	}
193 
194 	/* cache copy for remaining bytes */
195 	if (size) {
196 		memcpy((void *) dest, (void *) source, size);
197 		clean_cache_range((void *) dest, size);
198 	}
199 }
200 EXPORT_SYMBOL_GPL(__memcpy_flushcache);
201 
202 void memcpy_page_flushcache(char *to, struct page *page, size_t offset,
203 		size_t len)
204 {
205 	char *from = kmap_atomic(page);
206 
207 	memcpy_flushcache(to, from + offset, len);
208 	kunmap_atomic(from);
209 }
210 #endif
211