xref: /linux/arch/x86/lib/usercopy_64.c (revision 4fd18fc38757217c746aa063ba9e4729814dc737)
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 		"	.align 16\n"
28 		"0:	movq $0,(%[dst])\n"
29 		"	addq   $8,%[dst]\n"
30 		"	decl %%ecx ; jnz   0b\n"
31 		"4:	movq  %[size1],%%rcx\n"
32 		"	testl %%ecx,%%ecx\n"
33 		"	jz     2f\n"
34 		"1:	movb   $0,(%[dst])\n"
35 		"	incq   %[dst]\n"
36 		"	decl %%ecx ; jnz  1b\n"
37 		"2:\n"
38 		".section .fixup,\"ax\"\n"
39 		"3:	lea 0(%[size1],%[size8],8),%[size8]\n"
40 		"	jmp 2b\n"
41 		".previous\n"
42 		_ASM_EXTABLE_UA(0b, 3b)
43 		_ASM_EXTABLE_UA(1b, 2b)
44 		: [size8] "=&c"(size), [dst] "=&D" (__d0)
45 		: [size1] "r"(size & 7), "[size8]" (size / 8), "[dst]"(addr));
46 	clac();
47 	return size;
48 }
49 EXPORT_SYMBOL(__clear_user);
50 
51 unsigned long clear_user(void __user *to, unsigned long n)
52 {
53 	if (access_ok(to, n))
54 		return __clear_user(to, n);
55 	return n;
56 }
57 EXPORT_SYMBOL(clear_user);
58 
59 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
60 /**
61  * clean_cache_range - write back a cache range with CLWB
62  * @vaddr:	virtual start address
63  * @size:	number of bytes to write back
64  *
65  * Write back a cache range using the CLWB (cache line write back)
66  * instruction. Note that @size is internally rounded up to be cache
67  * line size aligned.
68  */
69 static void clean_cache_range(void *addr, size_t size)
70 {
71 	u16 x86_clflush_size = boot_cpu_data.x86_clflush_size;
72 	unsigned long clflush_mask = x86_clflush_size - 1;
73 	void *vend = addr + size;
74 	void *p;
75 
76 	for (p = (void *)((unsigned long)addr & ~clflush_mask);
77 	     p < vend; p += x86_clflush_size)
78 		clwb(p);
79 }
80 
81 void arch_wb_cache_pmem(void *addr, size_t size)
82 {
83 	clean_cache_range(addr, size);
84 }
85 EXPORT_SYMBOL_GPL(arch_wb_cache_pmem);
86 
87 long __copy_user_flushcache(void *dst, const void __user *src, unsigned size)
88 {
89 	unsigned long flushed, dest = (unsigned long) dst;
90 	long rc = __copy_user_nocache(dst, src, size, 0);
91 
92 	/*
93 	 * __copy_user_nocache() uses non-temporal stores for the bulk
94 	 * of the transfer, but we need to manually flush if the
95 	 * transfer is unaligned. A cached memory copy is used when
96 	 * destination or size is not naturally aligned. That is:
97 	 *   - Require 8-byte alignment when size is 8 bytes or larger.
98 	 *   - Require 4-byte alignment when size is 4 bytes.
99 	 */
100 	if (size < 8) {
101 		if (!IS_ALIGNED(dest, 4) || size != 4)
102 			clean_cache_range(dst, size);
103 	} else {
104 		if (!IS_ALIGNED(dest, 8)) {
105 			dest = ALIGN(dest, boot_cpu_data.x86_clflush_size);
106 			clean_cache_range(dst, 1);
107 		}
108 
109 		flushed = dest - (unsigned long) dst;
110 		if (size > flushed && !IS_ALIGNED(size - flushed, 8))
111 			clean_cache_range(dst + size - 1, 1);
112 	}
113 
114 	return rc;
115 }
116 
117 void __memcpy_flushcache(void *_dst, const void *_src, size_t size)
118 {
119 	unsigned long dest = (unsigned long) _dst;
120 	unsigned long source = (unsigned long) _src;
121 
122 	/* cache copy and flush to align dest */
123 	if (!IS_ALIGNED(dest, 8)) {
124 		unsigned len = min_t(unsigned, size, ALIGN(dest, 8) - dest);
125 
126 		memcpy((void *) dest, (void *) source, len);
127 		clean_cache_range((void *) dest, len);
128 		dest += len;
129 		source += len;
130 		size -= len;
131 		if (!size)
132 			return;
133 	}
134 
135 	/* 4x8 movnti loop */
136 	while (size >= 32) {
137 		asm("movq    (%0), %%r8\n"
138 		    "movq   8(%0), %%r9\n"
139 		    "movq  16(%0), %%r10\n"
140 		    "movq  24(%0), %%r11\n"
141 		    "movnti  %%r8,   (%1)\n"
142 		    "movnti  %%r9,  8(%1)\n"
143 		    "movnti %%r10, 16(%1)\n"
144 		    "movnti %%r11, 24(%1)\n"
145 		    :: "r" (source), "r" (dest)
146 		    : "memory", "r8", "r9", "r10", "r11");
147 		dest += 32;
148 		source += 32;
149 		size -= 32;
150 	}
151 
152 	/* 1x8 movnti loop */
153 	while (size >= 8) {
154 		asm("movq    (%0), %%r8\n"
155 		    "movnti  %%r8,   (%1)\n"
156 		    :: "r" (source), "r" (dest)
157 		    : "memory", "r8");
158 		dest += 8;
159 		source += 8;
160 		size -= 8;
161 	}
162 
163 	/* 1x4 movnti loop */
164 	while (size >= 4) {
165 		asm("movl    (%0), %%r8d\n"
166 		    "movnti  %%r8d,   (%1)\n"
167 		    :: "r" (source), "r" (dest)
168 		    : "memory", "r8");
169 		dest += 4;
170 		source += 4;
171 		size -= 4;
172 	}
173 
174 	/* cache copy for remaining bytes */
175 	if (size) {
176 		memcpy((void *) dest, (void *) source, size);
177 		clean_cache_range((void *) dest, size);
178 	}
179 }
180 EXPORT_SYMBOL_GPL(__memcpy_flushcache);
181 
182 void memcpy_page_flushcache(char *to, struct page *page, size_t offset,
183 		size_t len)
184 {
185 	char *from = kmap_atomic(page);
186 
187 	memcpy_flushcache(to, from + offset, len);
188 	kunmap_atomic(from);
189 }
190 #endif
191