1 /* SPDX-License-Identifier: GPL-2.0-only */
3 * Copyright 2008 Vitaly Mayatskikh <vmayatsk@redhat.com>
4 * Copyright 2002 Andi Kleen, SuSE Labs.
6 * Functions to copy from and to user space.
9 #include <linux/linkage.h>
10 #include <asm/current.h>
11 #include <asm/asm-offsets.h>
12 #include <asm/thread_info.h>
13 #include <asm/cpufeatures.h>
14 #include <asm/alternative.h>
17 #include <asm/export.h>
18 #include <asm/trapnr.h>
20 .macro ALIGN_DESTINATION
21 /* check for bad alignment of destination */
24 jz 102f /* already aligned */
36 _ASM_EXTABLE_CPY(100b, .Lcopy_user_handle_align)
37 _ASM_EXTABLE_CPY(101b, .Lcopy_user_handle_align)
41 * copy_user_generic_unrolled - memory copy with exception handling.
42 * This version is for CPUs like P4 that don't have efficient micro
51 * eax uncopied bytes or 0 if successful.
53 SYM_FUNC_START(copy_user_generic_unrolled)
56 jb 20f /* less then 8 bytes, go to byte copy loop */
61 jz .L_copy_short_string
64 3: movq 2*8(%rsi),%r10
65 4: movq 3*8(%rsi),%r11
68 7: movq %r10,2*8(%rdi)
69 8: movq %r11,3*8(%rdi)
71 10: movq 5*8(%rsi),%r9
72 11: movq 6*8(%rsi),%r10
73 12: movq 7*8(%rsi),%r11
74 13: movq %r8,4*8(%rdi)
75 14: movq %r9,5*8(%rdi)
76 15: movq %r10,6*8(%rdi)
77 16: movq %r11,7*8(%rdi)
109 40: leal (%rdx,%rcx,8),%edx
112 60: jmp .Lcopy_user_handle_tail /* ecx is zerorest also */
114 _ASM_EXTABLE_CPY(1b, 30b)
115 _ASM_EXTABLE_CPY(2b, 30b)
116 _ASM_EXTABLE_CPY(3b, 30b)
117 _ASM_EXTABLE_CPY(4b, 30b)
118 _ASM_EXTABLE_CPY(5b, 30b)
119 _ASM_EXTABLE_CPY(6b, 30b)
120 _ASM_EXTABLE_CPY(7b, 30b)
121 _ASM_EXTABLE_CPY(8b, 30b)
122 _ASM_EXTABLE_CPY(9b, 30b)
123 _ASM_EXTABLE_CPY(10b, 30b)
124 _ASM_EXTABLE_CPY(11b, 30b)
125 _ASM_EXTABLE_CPY(12b, 30b)
126 _ASM_EXTABLE_CPY(13b, 30b)
127 _ASM_EXTABLE_CPY(14b, 30b)
128 _ASM_EXTABLE_CPY(15b, 30b)
129 _ASM_EXTABLE_CPY(16b, 30b)
130 _ASM_EXTABLE_CPY(18b, 40b)
131 _ASM_EXTABLE_CPY(19b, 40b)
132 _ASM_EXTABLE_CPY(21b, 50b)
133 _ASM_EXTABLE_CPY(22b, 50b)
134 SYM_FUNC_END(copy_user_generic_unrolled)
135 EXPORT_SYMBOL(copy_user_generic_unrolled)
137 /* Some CPUs run faster using the string copy instructions.
138 * This is also a lot simpler. Use them when possible.
140 * Only 4GB of copy is supported. This shouldn't be a problem
141 * because the kernel normally only writes from/to page sized chunks
142 * even if user space passed a longer buffer.
143 * And more would be dangerous because both Intel and AMD have
144 * errata with rep movsq > 4GB. If someone feels the need to fix
145 * this please consider this.
153 * eax uncopied bytes or 0 if successful.
155 SYM_FUNC_START(copy_user_generic_string)
158 jb 2f /* less than 8 bytes, go to byte copy loop */
170 11: leal (%rdx,%rcx,8),%ecx
171 12: movl %ecx,%edx /* ecx is zerorest also */
172 jmp .Lcopy_user_handle_tail
174 _ASM_EXTABLE_CPY(1b, 11b)
175 _ASM_EXTABLE_CPY(3b, 12b)
176 SYM_FUNC_END(copy_user_generic_string)
177 EXPORT_SYMBOL(copy_user_generic_string)
180 * Some CPUs are adding enhanced REP MOVSB/STOSB instructions.
181 * It's recommended to use enhanced REP MOVSB/STOSB if it's enabled.
189 * eax uncopied bytes or 0 if successful.
191 SYM_FUNC_START(copy_user_enhanced_fast_string)
193 /* CPUs without FSRM should avoid rep movsb for short copies */
194 ALTERNATIVE "cmpl $64, %edx; jb .L_copy_short_string", "", X86_FEATURE_FSRM
201 12: movl %ecx,%edx /* ecx is zerorest also */
202 jmp .Lcopy_user_handle_tail
204 _ASM_EXTABLE_CPY(1b, 12b)
205 SYM_FUNC_END(copy_user_enhanced_fast_string)
206 EXPORT_SYMBOL(copy_user_enhanced_fast_string)
209 * Try to copy last bytes and clear the rest if needed.
210 * Since protection fault in copy_from/to_user is not a normal situation,
211 * it is not necessary to optimize tail handling.
212 * Don't try to copy the tail if machine check happened
215 * eax trap number written by ex_handler_copy()
221 * eax uncopied bytes or 0 if successful.
223 SYM_CODE_START_LOCAL(.Lcopy_user_handle_tail)
224 cmp $X86_TRAP_MC,%eax
238 _ASM_EXTABLE_CPY(1b, 2b)
240 .Lcopy_user_handle_align:
241 addl %ecx,%edx /* ecx is zerorest also */
242 jmp .Lcopy_user_handle_tail
244 SYM_CODE_END(.Lcopy_user_handle_tail)
247 * copy_user_nocache - Uncached memory copy with exception handling
248 * This will force destination out of cache for more performance.
250 * Note: Cached memory copy is used when destination or size is not
251 * naturally aligned. That is:
252 * - Require 8-byte alignment when size is 8 bytes or larger.
253 * - Require 4-byte alignment when size is 4 bytes.
255 SYM_FUNC_START(__copy_user_nocache)
258 /* If size is less than 8 bytes, go to 4-byte copy */
260 jb .L_4b_nocache_copy_entry
262 /* If destination is not 8-byte aligned, "cache" copy to align it */
265 /* Set 4x8-byte copy count and remainder */
269 jz .L_8b_nocache_copy_entry /* jump if count is 0 */
271 /* Perform 4x8-byte nocache loop-copy */
272 .L_4x8b_nocache_copy_loop:
274 2: movq 1*8(%rsi),%r9
275 3: movq 2*8(%rsi),%r10
276 4: movq 3*8(%rsi),%r11
278 6: movnti %r9,1*8(%rdi)
279 7: movnti %r10,2*8(%rdi)
280 8: movnti %r11,3*8(%rdi)
281 9: movq 4*8(%rsi),%r8
282 10: movq 5*8(%rsi),%r9
283 11: movq 6*8(%rsi),%r10
284 12: movq 7*8(%rsi),%r11
285 13: movnti %r8,4*8(%rdi)
286 14: movnti %r9,5*8(%rdi)
287 15: movnti %r10,6*8(%rdi)
288 16: movnti %r11,7*8(%rdi)
292 jnz .L_4x8b_nocache_copy_loop
294 /* Set 8-byte copy count and remainder */
295 .L_8b_nocache_copy_entry:
299 jz .L_4b_nocache_copy_entry /* jump if count is 0 */
301 /* Perform 8-byte nocache loop-copy */
302 .L_8b_nocache_copy_loop:
304 21: movnti %r8,(%rdi)
308 jnz .L_8b_nocache_copy_loop
310 /* If no byte left, we're done */
311 .L_4b_nocache_copy_entry:
315 /* If destination is not 4-byte aligned, go to byte copy: */
318 jnz .L_1b_cache_copy_entry
320 /* Set 4-byte copy count (1 or 0) and remainder */
324 jz .L_1b_cache_copy_entry /* jump if count is 0 */
326 /* Perform 4-byte nocache copy: */
328 31: movnti %r8d,(%rdi)
332 /* If no bytes left, we're done: */
336 /* Perform byte "cache" loop-copy for the remainder */
337 .L_1b_cache_copy_entry:
339 .L_1b_cache_copy_loop:
345 jnz .L_1b_cache_copy_loop
347 /* Finished copying; fence the prior stores */
357 jmp .L_fixup_handle_tail
359 lea (%rdx,%rcx,8),%rdx
360 jmp .L_fixup_handle_tail
362 lea (%rdx,%rcx,4),%rdx
363 jmp .L_fixup_handle_tail
366 .L_fixup_handle_tail:
368 jmp .Lcopy_user_handle_tail
370 _ASM_EXTABLE_CPY(1b, .L_fixup_4x8b_copy)
371 _ASM_EXTABLE_CPY(2b, .L_fixup_4x8b_copy)
372 _ASM_EXTABLE_CPY(3b, .L_fixup_4x8b_copy)
373 _ASM_EXTABLE_CPY(4b, .L_fixup_4x8b_copy)
374 _ASM_EXTABLE_CPY(5b, .L_fixup_4x8b_copy)
375 _ASM_EXTABLE_CPY(6b, .L_fixup_4x8b_copy)
376 _ASM_EXTABLE_CPY(7b, .L_fixup_4x8b_copy)
377 _ASM_EXTABLE_CPY(8b, .L_fixup_4x8b_copy)
378 _ASM_EXTABLE_CPY(9b, .L_fixup_4x8b_copy)
379 _ASM_EXTABLE_CPY(10b, .L_fixup_4x8b_copy)
380 _ASM_EXTABLE_CPY(11b, .L_fixup_4x8b_copy)
381 _ASM_EXTABLE_CPY(12b, .L_fixup_4x8b_copy)
382 _ASM_EXTABLE_CPY(13b, .L_fixup_4x8b_copy)
383 _ASM_EXTABLE_CPY(14b, .L_fixup_4x8b_copy)
384 _ASM_EXTABLE_CPY(15b, .L_fixup_4x8b_copy)
385 _ASM_EXTABLE_CPY(16b, .L_fixup_4x8b_copy)
386 _ASM_EXTABLE_CPY(20b, .L_fixup_8b_copy)
387 _ASM_EXTABLE_CPY(21b, .L_fixup_8b_copy)
388 _ASM_EXTABLE_CPY(30b, .L_fixup_4b_copy)
389 _ASM_EXTABLE_CPY(31b, .L_fixup_4b_copy)
390 _ASM_EXTABLE_CPY(40b, .L_fixup_1b_copy)
391 _ASM_EXTABLE_CPY(41b, .L_fixup_1b_copy)
392 SYM_FUNC_END(__copy_user_nocache)
393 EXPORT_SYMBOL(__copy_user_nocache)