xref: /linux/arch/x86/lib/copy_user_64.S (revision e5c86679d5e864947a52fb31e45a425dea3e7fa9)
1/*
2 * Copyright 2008 Vitaly Mayatskikh <vmayatsk@redhat.com>
3 * Copyright 2002 Andi Kleen, SuSE Labs.
4 * Subject to the GNU Public License v2.
5 *
6 * Functions to copy from and to user space.
7 */
8
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>
15#include <asm/asm.h>
16#include <asm/smap.h>
17#include <asm/export.h>
18
19/*
20 * copy_user_generic_unrolled - memory copy with exception handling.
21 * This version is for CPUs like P4 that don't have efficient micro
22 * code for rep movsq
23 *
24 * Input:
25 * rdi destination
26 * rsi source
27 * rdx count
28 *
29 * Output:
30 * eax uncopied bytes or 0 if successful.
31 */
32ENTRY(copy_user_generic_unrolled)
33	ASM_STAC
34	cmpl $8,%edx
35	jb 20f		/* less then 8 bytes, go to byte copy loop */
36	ALIGN_DESTINATION
37	movl %edx,%ecx
38	andl $63,%edx
39	shrl $6,%ecx
40	jz 17f
411:	movq (%rsi),%r8
422:	movq 1*8(%rsi),%r9
433:	movq 2*8(%rsi),%r10
444:	movq 3*8(%rsi),%r11
455:	movq %r8,(%rdi)
466:	movq %r9,1*8(%rdi)
477:	movq %r10,2*8(%rdi)
488:	movq %r11,3*8(%rdi)
499:	movq 4*8(%rsi),%r8
5010:	movq 5*8(%rsi),%r9
5111:	movq 6*8(%rsi),%r10
5212:	movq 7*8(%rsi),%r11
5313:	movq %r8,4*8(%rdi)
5414:	movq %r9,5*8(%rdi)
5515:	movq %r10,6*8(%rdi)
5616:	movq %r11,7*8(%rdi)
57	leaq 64(%rsi),%rsi
58	leaq 64(%rdi),%rdi
59	decl %ecx
60	jnz 1b
6117:	movl %edx,%ecx
62	andl $7,%edx
63	shrl $3,%ecx
64	jz 20f
6518:	movq (%rsi),%r8
6619:	movq %r8,(%rdi)
67	leaq 8(%rsi),%rsi
68	leaq 8(%rdi),%rdi
69	decl %ecx
70	jnz 18b
7120:	andl %edx,%edx
72	jz 23f
73	movl %edx,%ecx
7421:	movb (%rsi),%al
7522:	movb %al,(%rdi)
76	incq %rsi
77	incq %rdi
78	decl %ecx
79	jnz 21b
8023:	xor %eax,%eax
81	ASM_CLAC
82	ret
83
84	.section .fixup,"ax"
8530:	shll $6,%ecx
86	addl %ecx,%edx
87	jmp 60f
8840:	leal (%rdx,%rcx,8),%edx
89	jmp 60f
9050:	movl %ecx,%edx
9160:	jmp copy_user_handle_tail /* ecx is zerorest also */
92	.previous
93
94	_ASM_EXTABLE(1b,30b)
95	_ASM_EXTABLE(2b,30b)
96	_ASM_EXTABLE(3b,30b)
97	_ASM_EXTABLE(4b,30b)
98	_ASM_EXTABLE(5b,30b)
99	_ASM_EXTABLE(6b,30b)
100	_ASM_EXTABLE(7b,30b)
101	_ASM_EXTABLE(8b,30b)
102	_ASM_EXTABLE(9b,30b)
103	_ASM_EXTABLE(10b,30b)
104	_ASM_EXTABLE(11b,30b)
105	_ASM_EXTABLE(12b,30b)
106	_ASM_EXTABLE(13b,30b)
107	_ASM_EXTABLE(14b,30b)
108	_ASM_EXTABLE(15b,30b)
109	_ASM_EXTABLE(16b,30b)
110	_ASM_EXTABLE(18b,40b)
111	_ASM_EXTABLE(19b,40b)
112	_ASM_EXTABLE(21b,50b)
113	_ASM_EXTABLE(22b,50b)
114ENDPROC(copy_user_generic_unrolled)
115EXPORT_SYMBOL(copy_user_generic_unrolled)
116
117/* Some CPUs run faster using the string copy instructions.
118 * This is also a lot simpler. Use them when possible.
119 *
120 * Only 4GB of copy is supported. This shouldn't be a problem
121 * because the kernel normally only writes from/to page sized chunks
122 * even if user space passed a longer buffer.
123 * And more would be dangerous because both Intel and AMD have
124 * errata with rep movsq > 4GB. If someone feels the need to fix
125 * this please consider this.
126 *
127 * Input:
128 * rdi destination
129 * rsi source
130 * rdx count
131 *
132 * Output:
133 * eax uncopied bytes or 0 if successful.
134 */
135ENTRY(copy_user_generic_string)
136	ASM_STAC
137	cmpl $8,%edx
138	jb 2f		/* less than 8 bytes, go to byte copy loop */
139	ALIGN_DESTINATION
140	movl %edx,%ecx
141	shrl $3,%ecx
142	andl $7,%edx
1431:	rep
144	movsq
1452:	movl %edx,%ecx
1463:	rep
147	movsb
148	xorl %eax,%eax
149	ASM_CLAC
150	ret
151
152	.section .fixup,"ax"
15311:	leal (%rdx,%rcx,8),%ecx
15412:	movl %ecx,%edx		/* ecx is zerorest also */
155	jmp copy_user_handle_tail
156	.previous
157
158	_ASM_EXTABLE(1b,11b)
159	_ASM_EXTABLE(3b,12b)
160ENDPROC(copy_user_generic_string)
161EXPORT_SYMBOL(copy_user_generic_string)
162
163/*
164 * Some CPUs are adding enhanced REP MOVSB/STOSB instructions.
165 * It's recommended to use enhanced REP MOVSB/STOSB if it's enabled.
166 *
167 * Input:
168 * rdi destination
169 * rsi source
170 * rdx count
171 *
172 * Output:
173 * eax uncopied bytes or 0 if successful.
174 */
175ENTRY(copy_user_enhanced_fast_string)
176	ASM_STAC
177	movl %edx,%ecx
1781:	rep
179	movsb
180	xorl %eax,%eax
181	ASM_CLAC
182	ret
183
184	.section .fixup,"ax"
18512:	movl %ecx,%edx		/* ecx is zerorest also */
186	jmp copy_user_handle_tail
187	.previous
188
189	_ASM_EXTABLE(1b,12b)
190ENDPROC(copy_user_enhanced_fast_string)
191EXPORT_SYMBOL(copy_user_enhanced_fast_string)
192
193/*
194 * copy_user_nocache - Uncached memory copy with exception handling
195 * This will force destination out of cache for more performance.
196 *
197 * Note: Cached memory copy is used when destination or size is not
198 * naturally aligned. That is:
199 *  - Require 8-byte alignment when size is 8 bytes or larger.
200 *  - Require 4-byte alignment when size is 4 bytes.
201 */
202ENTRY(__copy_user_nocache)
203	ASM_STAC
204
205	/* If size is less than 8 bytes, go to 4-byte copy */
206	cmpl $8,%edx
207	jb .L_4b_nocache_copy_entry
208
209	/* If destination is not 8-byte aligned, "cache" copy to align it */
210	ALIGN_DESTINATION
211
212	/* Set 4x8-byte copy count and remainder */
213	movl %edx,%ecx
214	andl $63,%edx
215	shrl $6,%ecx
216	jz .L_8b_nocache_copy_entry	/* jump if count is 0 */
217
218	/* Perform 4x8-byte nocache loop-copy */
219.L_4x8b_nocache_copy_loop:
2201:	movq (%rsi),%r8
2212:	movq 1*8(%rsi),%r9
2223:	movq 2*8(%rsi),%r10
2234:	movq 3*8(%rsi),%r11
2245:	movnti %r8,(%rdi)
2256:	movnti %r9,1*8(%rdi)
2267:	movnti %r10,2*8(%rdi)
2278:	movnti %r11,3*8(%rdi)
2289:	movq 4*8(%rsi),%r8
22910:	movq 5*8(%rsi),%r9
23011:	movq 6*8(%rsi),%r10
23112:	movq 7*8(%rsi),%r11
23213:	movnti %r8,4*8(%rdi)
23314:	movnti %r9,5*8(%rdi)
23415:	movnti %r10,6*8(%rdi)
23516:	movnti %r11,7*8(%rdi)
236	leaq 64(%rsi),%rsi
237	leaq 64(%rdi),%rdi
238	decl %ecx
239	jnz .L_4x8b_nocache_copy_loop
240
241	/* Set 8-byte copy count and remainder */
242.L_8b_nocache_copy_entry:
243	movl %edx,%ecx
244	andl $7,%edx
245	shrl $3,%ecx
246	jz .L_4b_nocache_copy_entry	/* jump if count is 0 */
247
248	/* Perform 8-byte nocache loop-copy */
249.L_8b_nocache_copy_loop:
25020:	movq (%rsi),%r8
25121:	movnti %r8,(%rdi)
252	leaq 8(%rsi),%rsi
253	leaq 8(%rdi),%rdi
254	decl %ecx
255	jnz .L_8b_nocache_copy_loop
256
257	/* If no byte left, we're done */
258.L_4b_nocache_copy_entry:
259	andl %edx,%edx
260	jz .L_finish_copy
261
262	/* If destination is not 4-byte aligned, go to byte copy: */
263	movl %edi,%ecx
264	andl $3,%ecx
265	jnz .L_1b_cache_copy_entry
266
267	/* Set 4-byte copy count (1 or 0) and remainder */
268	movl %edx,%ecx
269	andl $3,%edx
270	shrl $2,%ecx
271	jz .L_1b_cache_copy_entry	/* jump if count is 0 */
272
273	/* Perform 4-byte nocache copy: */
27430:	movl (%rsi),%r8d
27531:	movnti %r8d,(%rdi)
276	leaq 4(%rsi),%rsi
277	leaq 4(%rdi),%rdi
278
279	/* If no bytes left, we're done: */
280	andl %edx,%edx
281	jz .L_finish_copy
282
283	/* Perform byte "cache" loop-copy for the remainder */
284.L_1b_cache_copy_entry:
285	movl %edx,%ecx
286.L_1b_cache_copy_loop:
28740:	movb (%rsi),%al
28841:	movb %al,(%rdi)
289	incq %rsi
290	incq %rdi
291	decl %ecx
292	jnz .L_1b_cache_copy_loop
293
294	/* Finished copying; fence the prior stores */
295.L_finish_copy:
296	xorl %eax,%eax
297	ASM_CLAC
298	sfence
299	ret
300
301	.section .fixup,"ax"
302.L_fixup_4x8b_copy:
303	shll $6,%ecx
304	addl %ecx,%edx
305	jmp .L_fixup_handle_tail
306.L_fixup_8b_copy:
307	lea (%rdx,%rcx,8),%rdx
308	jmp .L_fixup_handle_tail
309.L_fixup_4b_copy:
310	lea (%rdx,%rcx,4),%rdx
311	jmp .L_fixup_handle_tail
312.L_fixup_1b_copy:
313	movl %ecx,%edx
314.L_fixup_handle_tail:
315	sfence
316	jmp copy_user_handle_tail
317	.previous
318
319	_ASM_EXTABLE(1b,.L_fixup_4x8b_copy)
320	_ASM_EXTABLE(2b,.L_fixup_4x8b_copy)
321	_ASM_EXTABLE(3b,.L_fixup_4x8b_copy)
322	_ASM_EXTABLE(4b,.L_fixup_4x8b_copy)
323	_ASM_EXTABLE(5b,.L_fixup_4x8b_copy)
324	_ASM_EXTABLE(6b,.L_fixup_4x8b_copy)
325	_ASM_EXTABLE(7b,.L_fixup_4x8b_copy)
326	_ASM_EXTABLE(8b,.L_fixup_4x8b_copy)
327	_ASM_EXTABLE(9b,.L_fixup_4x8b_copy)
328	_ASM_EXTABLE(10b,.L_fixup_4x8b_copy)
329	_ASM_EXTABLE(11b,.L_fixup_4x8b_copy)
330	_ASM_EXTABLE(12b,.L_fixup_4x8b_copy)
331	_ASM_EXTABLE(13b,.L_fixup_4x8b_copy)
332	_ASM_EXTABLE(14b,.L_fixup_4x8b_copy)
333	_ASM_EXTABLE(15b,.L_fixup_4x8b_copy)
334	_ASM_EXTABLE(16b,.L_fixup_4x8b_copy)
335	_ASM_EXTABLE(20b,.L_fixup_8b_copy)
336	_ASM_EXTABLE(21b,.L_fixup_8b_copy)
337	_ASM_EXTABLE(30b,.L_fixup_4b_copy)
338	_ASM_EXTABLE(31b,.L_fixup_4b_copy)
339	_ASM_EXTABLE(40b,.L_fixup_1b_copy)
340	_ASM_EXTABLE(41b,.L_fixup_1b_copy)
341ENDPROC(__copy_user_nocache)
342EXPORT_SYMBOL(__copy_user_nocache)
343