xref: /openbmc/linux/arch/x86/lib/copy_user_64.S (revision 75016ca3) 
1/* SPDX-License-Identifier: GPL-2.0-only */
2/*
3 * Copyright 2008 Vitaly Mayatskikh <vmayatsk@redhat.com>
4 * Copyright 2002 Andi Kleen, SuSE Labs.
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.h>
15#include <asm/asm.h>
16#include <asm/smap.h>
17#include <asm/export.h>
18#include <asm/trapnr.h>
19
20.macro ALIGN_DESTINATION
21	/* check for bad alignment of destination */
22	movl %edi,%ecx
23	andl $7,%ecx
24	jz 102f				/* already aligned */
25	subl $8,%ecx
26	negl %ecx
27	subl %ecx,%edx
28100:	movb (%rsi),%al
29101:	movb %al,(%rdi)
30	incq %rsi
31	incq %rdi
32	decl %ecx
33	jnz 100b
34102:
35
36	_ASM_EXTABLE_CPY(100b, .Lcopy_user_handle_align)
37	_ASM_EXTABLE_CPY(101b, .Lcopy_user_handle_align)
38.endm
39
40/*
41 * copy_user_generic_unrolled - memory copy with exception handling.
42 * This version is for CPUs like P4 that don't have efficient micro
43 * code for rep movsq
44 *
45 * Input:
46 * rdi destination
47 * rsi source
48 * rdx count
49 *
50 * Output:
51 * eax uncopied bytes or 0 if successful.
52 */
53SYM_FUNC_START(copy_user_generic_unrolled)
54	ASM_STAC
55	cmpl $8,%edx
56	jb 20f		/* less then 8 bytes, go to byte copy loop */
57	ALIGN_DESTINATION
58	movl %edx,%ecx
59	andl $63,%edx
60	shrl $6,%ecx
61	jz .L_copy_short_string
621:	movq (%rsi),%r8
632:	movq 1*8(%rsi),%r9
643:	movq 2*8(%rsi),%r10
654:	movq 3*8(%rsi),%r11
665:	movq %r8,(%rdi)
676:	movq %r9,1*8(%rdi)
687:	movq %r10,2*8(%rdi)
698:	movq %r11,3*8(%rdi)
709:	movq 4*8(%rsi),%r8
7110:	movq 5*8(%rsi),%r9
7211:	movq 6*8(%rsi),%r10
7312:	movq 7*8(%rsi),%r11
7413:	movq %r8,4*8(%rdi)
7514:	movq %r9,5*8(%rdi)
7615:	movq %r10,6*8(%rdi)
7716:	movq %r11,7*8(%rdi)
78	leaq 64(%rsi),%rsi
79	leaq 64(%rdi),%rdi
80	decl %ecx
81	jnz 1b
82.L_copy_short_string:
83	movl %edx,%ecx
84	andl $7,%edx
85	shrl $3,%ecx
86	jz 20f
8718:	movq (%rsi),%r8
8819:	movq %r8,(%rdi)
89	leaq 8(%rsi),%rsi
90	leaq 8(%rdi),%rdi
91	decl %ecx
92	jnz 18b
9320:	andl %edx,%edx
94	jz 23f
95	movl %edx,%ecx
9621:	movb (%rsi),%al
9722:	movb %al,(%rdi)
98	incq %rsi
99	incq %rdi
100	decl %ecx
101	jnz 21b
10223:	xor %eax,%eax
103	ASM_CLAC
104	RET
105
10630:	shll $6,%ecx
107	addl %ecx,%edx
108	jmp 60f
10940:	leal (%rdx,%rcx,8),%edx
110	jmp 60f
11150:	movl %ecx,%edx
11260:	jmp .Lcopy_user_handle_tail /* ecx is zerorest also */
113
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)
134SYM_FUNC_END(copy_user_generic_unrolled)
135EXPORT_SYMBOL(copy_user_generic_unrolled)
136
137/* Some CPUs run faster using the string copy instructions.
138 * This is also a lot simpler. Use them when possible.
139 *
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.
146 *
147 * Input:
148 * rdi destination
149 * rsi source
150 * rdx count
151 *
152 * Output:
153 * eax uncopied bytes or 0 if successful.
154 */
155SYM_FUNC_START(copy_user_generic_string)
156	ASM_STAC
157	cmpl $8,%edx
158	jb 2f		/* less than 8 bytes, go to byte copy loop */
159	ALIGN_DESTINATION
160	movl %edx,%ecx
161	shrl $3,%ecx
162	andl $7,%edx
1631:	rep movsq
1642:	movl %edx,%ecx
1653:	rep movsb
166	xorl %eax,%eax
167	ASM_CLAC
168	RET
169
17011:	leal (%rdx,%rcx,8),%ecx
17112:	movl %ecx,%edx		/* ecx is zerorest also */
172	jmp .Lcopy_user_handle_tail
173
174	_ASM_EXTABLE_CPY(1b, 11b)
175	_ASM_EXTABLE_CPY(3b, 12b)
176SYM_FUNC_END(copy_user_generic_string)
177EXPORT_SYMBOL(copy_user_generic_string)
178
179/*
180 * Some CPUs are adding enhanced REP MOVSB/STOSB instructions.
181 * It's recommended to use enhanced REP MOVSB/STOSB if it's enabled.
182 *
183 * Input:
184 * rdi destination
185 * rsi source
186 * rdx count
187 *
188 * Output:
189 * eax uncopied bytes or 0 if successful.
190 */
191SYM_FUNC_START(copy_user_enhanced_fast_string)
192	ASM_STAC
193	/* CPUs without FSRM should avoid rep movsb for short copies */
194	ALTERNATIVE "cmpl $64, %edx; jb .L_copy_short_string", "", X86_FEATURE_FSRM
195	movl %edx,%ecx
1961:	rep movsb
197	xorl %eax,%eax
198	ASM_CLAC
199	RET
200
20112:	movl %ecx,%edx		/* ecx is zerorest also */
202	jmp .Lcopy_user_handle_tail
203
204	_ASM_EXTABLE_CPY(1b, 12b)
205SYM_FUNC_END(copy_user_enhanced_fast_string)
206EXPORT_SYMBOL(copy_user_enhanced_fast_string)
207
208/*
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
213 *
214 * Input:
215 * eax trap number written by ex_handler_copy()
216 * rdi destination
217 * rsi source
218 * rdx count
219 *
220 * Output:
221 * eax uncopied bytes or 0 if successful.
222 */
223SYM_CODE_START_LOCAL(.Lcopy_user_handle_tail)
224	cmp $X86_TRAP_MC,%eax
225	je 3f
226
227	movl %edx,%ecx
2281:	rep movsb
2292:	mov %ecx,%eax
230	ASM_CLAC
231	RET
232
2333:
234	movl %edx,%eax
235	ASM_CLAC
236	RET
237
238	_ASM_EXTABLE_CPY(1b, 2b)
239
240.Lcopy_user_handle_align:
241	addl %ecx,%edx			/* ecx is zerorest also */
242	jmp .Lcopy_user_handle_tail
243
244SYM_CODE_END(.Lcopy_user_handle_tail)
245
246/*
247 * copy_user_nocache - Uncached memory copy with exception handling
248 * This will force destination out of cache for more performance.
249 *
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.
254 */
255SYM_FUNC_START(__copy_user_nocache)
256	ASM_STAC
257
258	/* If size is less than 8 bytes, go to 4-byte copy */
259	cmpl $8,%edx
260	jb .L_4b_nocache_copy_entry
261
262	/* If destination is not 8-byte aligned, "cache" copy to align it */
263	ALIGN_DESTINATION
264
265	/* Set 4x8-byte copy count and remainder */
266	movl %edx,%ecx
267	andl $63,%edx
268	shrl $6,%ecx
269	jz .L_8b_nocache_copy_entry	/* jump if count is 0 */
270
271	/* Perform 4x8-byte nocache loop-copy */
272.L_4x8b_nocache_copy_loop:
2731:	movq (%rsi),%r8
2742:	movq 1*8(%rsi),%r9
2753:	movq 2*8(%rsi),%r10
2764:	movq 3*8(%rsi),%r11
2775:	movnti %r8,(%rdi)
2786:	movnti %r9,1*8(%rdi)
2797:	movnti %r10,2*8(%rdi)
2808:	movnti %r11,3*8(%rdi)
2819:	movq 4*8(%rsi),%r8
28210:	movq 5*8(%rsi),%r9
28311:	movq 6*8(%rsi),%r10
28412:	movq 7*8(%rsi),%r11
28513:	movnti %r8,4*8(%rdi)
28614:	movnti %r9,5*8(%rdi)
28715:	movnti %r10,6*8(%rdi)
28816:	movnti %r11,7*8(%rdi)
289	leaq 64(%rsi),%rsi
290	leaq 64(%rdi),%rdi
291	decl %ecx
292	jnz .L_4x8b_nocache_copy_loop
293
294	/* Set 8-byte copy count and remainder */
295.L_8b_nocache_copy_entry:
296	movl %edx,%ecx
297	andl $7,%edx
298	shrl $3,%ecx
299	jz .L_4b_nocache_copy_entry	/* jump if count is 0 */
300
301	/* Perform 8-byte nocache loop-copy */
302.L_8b_nocache_copy_loop:
30320:	movq (%rsi),%r8
30421:	movnti %r8,(%rdi)
305	leaq 8(%rsi),%rsi
306	leaq 8(%rdi),%rdi
307	decl %ecx
308	jnz .L_8b_nocache_copy_loop
309
310	/* If no byte left, we're done */
311.L_4b_nocache_copy_entry:
312	andl %edx,%edx
313	jz .L_finish_copy
314
315	/* If destination is not 4-byte aligned, go to byte copy: */
316	movl %edi,%ecx
317	andl $3,%ecx
318	jnz .L_1b_cache_copy_entry
319
320	/* Set 4-byte copy count (1 or 0) and remainder */
321	movl %edx,%ecx
322	andl $3,%edx
323	shrl $2,%ecx
324	jz .L_1b_cache_copy_entry	/* jump if count is 0 */
325
326	/* Perform 4-byte nocache copy: */
32730:	movl (%rsi),%r8d
32831:	movnti %r8d,(%rdi)
329	leaq 4(%rsi),%rsi
330	leaq 4(%rdi),%rdi
331
332	/* If no bytes left, we're done: */
333	andl %edx,%edx
334	jz .L_finish_copy
335
336	/* Perform byte "cache" loop-copy for the remainder */
337.L_1b_cache_copy_entry:
338	movl %edx,%ecx
339.L_1b_cache_copy_loop:
34040:	movb (%rsi),%al
34141:	movb %al,(%rdi)
342	incq %rsi
343	incq %rdi
344	decl %ecx
345	jnz .L_1b_cache_copy_loop
346
347	/* Finished copying; fence the prior stores */
348.L_finish_copy:
349	xorl %eax,%eax
350	ASM_CLAC
351	sfence
352	RET
353
354.L_fixup_4x8b_copy:
355	shll $6,%ecx
356	addl %ecx,%edx
357	jmp .L_fixup_handle_tail
358.L_fixup_8b_copy:
359	lea (%rdx,%rcx,8),%rdx
360	jmp .L_fixup_handle_tail
361.L_fixup_4b_copy:
362	lea (%rdx,%rcx,4),%rdx
363	jmp .L_fixup_handle_tail
364.L_fixup_1b_copy:
365	movl %ecx,%edx
366.L_fixup_handle_tail:
367	sfence
368	jmp .Lcopy_user_handle_tail
369
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)
392SYM_FUNC_END(__copy_user_nocache)
393EXPORT_SYMBOL(__copy_user_nocache)
394