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-asm.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 */
36 103: addl %ecx,%edx /* ecx is zerorest also */
37 jmp .Lcopy_user_handle_tail
40 _ASM_EXTABLE_CPY(100b, 103b)
41 _ASM_EXTABLE_CPY(101b, 103b)
45 * copy_user_generic_unrolled - memory copy with exception handling.
46 * This version is for CPUs like P4 that don't have efficient micro
55 * eax uncopied bytes or 0 if successful.
57 SYM_FUNC_START(copy_user_generic_unrolled)
60 jb 20f /* less then 8 bytes, go to byte copy loop */
65 jz .L_copy_short_string
68 3: movq 2*8(%rsi),%r10
69 4: movq 3*8(%rsi),%r11
72 7: movq %r10,2*8(%rdi)
73 8: movq %r11,3*8(%rdi)
75 10: movq 5*8(%rsi),%r9
76 11: movq 6*8(%rsi),%r10
77 12: movq 7*8(%rsi),%r11
78 13: movq %r8,4*8(%rdi)
79 14: movq %r9,5*8(%rdi)
80 15: movq %r10,6*8(%rdi)
81 16: movq %r11,7*8(%rdi)
114 40: leal (%rdx,%rcx,8),%edx
117 60: jmp .Lcopy_user_handle_tail /* ecx is zerorest also */
120 _ASM_EXTABLE_CPY(1b, 30b)
121 _ASM_EXTABLE_CPY(2b, 30b)
122 _ASM_EXTABLE_CPY(3b, 30b)
123 _ASM_EXTABLE_CPY(4b, 30b)
124 _ASM_EXTABLE_CPY(5b, 30b)
125 _ASM_EXTABLE_CPY(6b, 30b)
126 _ASM_EXTABLE_CPY(7b, 30b)
127 _ASM_EXTABLE_CPY(8b, 30b)
128 _ASM_EXTABLE_CPY(9b, 30b)
129 _ASM_EXTABLE_CPY(10b, 30b)
130 _ASM_EXTABLE_CPY(11b, 30b)
131 _ASM_EXTABLE_CPY(12b, 30b)
132 _ASM_EXTABLE_CPY(13b, 30b)
133 _ASM_EXTABLE_CPY(14b, 30b)
134 _ASM_EXTABLE_CPY(15b, 30b)
135 _ASM_EXTABLE_CPY(16b, 30b)
136 _ASM_EXTABLE_CPY(18b, 40b)
137 _ASM_EXTABLE_CPY(19b, 40b)
138 _ASM_EXTABLE_CPY(21b, 50b)
139 _ASM_EXTABLE_CPY(22b, 50b)
140 SYM_FUNC_END(copy_user_generic_unrolled)
141 EXPORT_SYMBOL(copy_user_generic_unrolled)
143 /* Some CPUs run faster using the string copy instructions.
144 * This is also a lot simpler. Use them when possible.
146 * Only 4GB of copy is supported. This shouldn't be a problem
147 * because the kernel normally only writes from/to page sized chunks
148 * even if user space passed a longer buffer.
149 * And more would be dangerous because both Intel and AMD have
150 * errata with rep movsq > 4GB. If someone feels the need to fix
151 * this please consider this.
159 * eax uncopied bytes or 0 if successful.
161 SYM_FUNC_START(copy_user_generic_string)
164 jb 2f /* less than 8 bytes, go to byte copy loop */
179 11: leal (%rdx,%rcx,8),%ecx
180 12: movl %ecx,%edx /* ecx is zerorest also */
181 jmp .Lcopy_user_handle_tail
184 _ASM_EXTABLE_CPY(1b, 11b)
185 _ASM_EXTABLE_CPY(3b, 12b)
186 SYM_FUNC_END(copy_user_generic_string)
187 EXPORT_SYMBOL(copy_user_generic_string)
190 * Some CPUs are adding enhanced REP MOVSB/STOSB instructions.
191 * It's recommended to use enhanced REP MOVSB/STOSB if it's enabled.
199 * eax uncopied bytes or 0 if successful.
201 SYM_FUNC_START(copy_user_enhanced_fast_string)
204 jb .L_copy_short_string /* less then 64 bytes, avoid the costly 'rep' */
213 12: movl %ecx,%edx /* ecx is zerorest also */
214 jmp .Lcopy_user_handle_tail
217 _ASM_EXTABLE_CPY(1b, 12b)
218 SYM_FUNC_END(copy_user_enhanced_fast_string)
219 EXPORT_SYMBOL(copy_user_enhanced_fast_string)
222 * Try to copy last bytes and clear the rest if needed.
223 * Since protection fault in copy_from/to_user is not a normal situation,
224 * it is not necessary to optimize tail handling.
225 * Don't try to copy the tail if machine check happened
233 * eax uncopied bytes or 0 if successful.
235 SYM_CODE_START_LOCAL(.Lcopy_user_handle_tail)
237 cmp $X86_TRAP_MC,%eax /* check if X86_TRAP_MC */
245 * Return zero to pretend that this copy succeeded. This
246 * is counter-intuitive, but needed to prevent the code
247 * in lib/iov_iter.c from retrying and running back into
248 * the poison cache line again. The machine check handler
249 * will ensure that a SIGBUS is sent to the task.
255 _ASM_EXTABLE_CPY(1b, 2b)
256 SYM_CODE_END(.Lcopy_user_handle_tail)
259 * copy_user_nocache - Uncached memory copy with exception handling
260 * This will force destination out of cache for more performance.
262 * Note: Cached memory copy is used when destination or size is not
263 * naturally aligned. That is:
264 * - Require 8-byte alignment when size is 8 bytes or larger.
265 * - Require 4-byte alignment when size is 4 bytes.
267 SYM_FUNC_START(__copy_user_nocache)
270 /* If size is less than 8 bytes, go to 4-byte copy */
272 jb .L_4b_nocache_copy_entry
274 /* If destination is not 8-byte aligned, "cache" copy to align it */
277 /* Set 4x8-byte copy count and remainder */
281 jz .L_8b_nocache_copy_entry /* jump if count is 0 */
283 /* Perform 4x8-byte nocache loop-copy */
284 .L_4x8b_nocache_copy_loop:
286 2: movq 1*8(%rsi),%r9
287 3: movq 2*8(%rsi),%r10
288 4: movq 3*8(%rsi),%r11
290 6: movnti %r9,1*8(%rdi)
291 7: movnti %r10,2*8(%rdi)
292 8: movnti %r11,3*8(%rdi)
293 9: movq 4*8(%rsi),%r8
294 10: movq 5*8(%rsi),%r9
295 11: movq 6*8(%rsi),%r10
296 12: movq 7*8(%rsi),%r11
297 13: movnti %r8,4*8(%rdi)
298 14: movnti %r9,5*8(%rdi)
299 15: movnti %r10,6*8(%rdi)
300 16: movnti %r11,7*8(%rdi)
304 jnz .L_4x8b_nocache_copy_loop
306 /* Set 8-byte copy count and remainder */
307 .L_8b_nocache_copy_entry:
311 jz .L_4b_nocache_copy_entry /* jump if count is 0 */
313 /* Perform 8-byte nocache loop-copy */
314 .L_8b_nocache_copy_loop:
316 21: movnti %r8,(%rdi)
320 jnz .L_8b_nocache_copy_loop
322 /* If no byte left, we're done */
323 .L_4b_nocache_copy_entry:
327 /* If destination is not 4-byte aligned, go to byte copy: */
330 jnz .L_1b_cache_copy_entry
332 /* Set 4-byte copy count (1 or 0) and remainder */
336 jz .L_1b_cache_copy_entry /* jump if count is 0 */
338 /* Perform 4-byte nocache copy: */
340 31: movnti %r8d,(%rdi)
344 /* If no bytes left, we're done: */
348 /* Perform byte "cache" loop-copy for the remainder */
349 .L_1b_cache_copy_entry:
351 .L_1b_cache_copy_loop:
357 jnz .L_1b_cache_copy_loop
359 /* Finished copying; fence the prior stores */
370 jmp .L_fixup_handle_tail
372 lea (%rdx,%rcx,8),%rdx
373 jmp .L_fixup_handle_tail
375 lea (%rdx,%rcx,4),%rdx
376 jmp .L_fixup_handle_tail
379 .L_fixup_handle_tail:
381 jmp .Lcopy_user_handle_tail
384 _ASM_EXTABLE_CPY(1b, .L_fixup_4x8b_copy)
385 _ASM_EXTABLE_CPY(2b, .L_fixup_4x8b_copy)
386 _ASM_EXTABLE_CPY(3b, .L_fixup_4x8b_copy)
387 _ASM_EXTABLE_CPY(4b, .L_fixup_4x8b_copy)
388 _ASM_EXTABLE_CPY(5b, .L_fixup_4x8b_copy)
389 _ASM_EXTABLE_CPY(6b, .L_fixup_4x8b_copy)
390 _ASM_EXTABLE_CPY(7b, .L_fixup_4x8b_copy)
391 _ASM_EXTABLE_CPY(8b, .L_fixup_4x8b_copy)
392 _ASM_EXTABLE_CPY(9b, .L_fixup_4x8b_copy)
393 _ASM_EXTABLE_CPY(10b, .L_fixup_4x8b_copy)
394 _ASM_EXTABLE_CPY(11b, .L_fixup_4x8b_copy)
395 _ASM_EXTABLE_CPY(12b, .L_fixup_4x8b_copy)
396 _ASM_EXTABLE_CPY(13b, .L_fixup_4x8b_copy)
397 _ASM_EXTABLE_CPY(14b, .L_fixup_4x8b_copy)
398 _ASM_EXTABLE_CPY(15b, .L_fixup_4x8b_copy)
399 _ASM_EXTABLE_CPY(16b, .L_fixup_4x8b_copy)
400 _ASM_EXTABLE_CPY(20b, .L_fixup_8b_copy)
401 _ASM_EXTABLE_CPY(21b, .L_fixup_8b_copy)
402 _ASM_EXTABLE_CPY(30b, .L_fixup_4b_copy)
403 _ASM_EXTABLE_CPY(31b, .L_fixup_4b_copy)
404 _ASM_EXTABLE_CPY(40b, .L_fixup_1b_copy)
405 _ASM_EXTABLE_CPY(41b, .L_fixup_1b_copy)
406 SYM_FUNC_END(__copy_user_nocache)
407 EXPORT_SYMBOL(__copy_user_nocache)