xref: /openbmc/linux/arch/x86/lib/copy_user_64.S (revision 8e7ae8ba) 
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 .Lcopy_user_short_string_bytes
57	ALIGN_DESTINATION
58	movl %edx,%ecx
59	andl $63,%edx
60	shrl $6,%ecx
61	jz copy_user_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	jmp copy_user_short_string
83
8430:	shll $6,%ecx
85	addl %ecx,%edx
86	jmp .Lcopy_user_handle_tail
87
88	_ASM_EXTABLE_CPY(1b, 30b)
89	_ASM_EXTABLE_CPY(2b, 30b)
90	_ASM_EXTABLE_CPY(3b, 30b)
91	_ASM_EXTABLE_CPY(4b, 30b)
92	_ASM_EXTABLE_CPY(5b, 30b)
93	_ASM_EXTABLE_CPY(6b, 30b)
94	_ASM_EXTABLE_CPY(7b, 30b)
95	_ASM_EXTABLE_CPY(8b, 30b)
96	_ASM_EXTABLE_CPY(9b, 30b)
97	_ASM_EXTABLE_CPY(10b, 30b)
98	_ASM_EXTABLE_CPY(11b, 30b)
99	_ASM_EXTABLE_CPY(12b, 30b)
100	_ASM_EXTABLE_CPY(13b, 30b)
101	_ASM_EXTABLE_CPY(14b, 30b)
102	_ASM_EXTABLE_CPY(15b, 30b)
103	_ASM_EXTABLE_CPY(16b, 30b)
104SYM_FUNC_END(copy_user_generic_unrolled)
105EXPORT_SYMBOL(copy_user_generic_unrolled)
106
107/* Some CPUs run faster using the string copy instructions.
108 * This is also a lot simpler. Use them when possible.
109 *
110 * Only 4GB of copy is supported. This shouldn't be a problem
111 * because the kernel normally only writes from/to page sized chunks
112 * even if user space passed a longer buffer.
113 * And more would be dangerous because both Intel and AMD have
114 * errata with rep movsq > 4GB. If someone feels the need to fix
115 * this please consider this.
116 *
117 * Input:
118 * rdi destination
119 * rsi source
120 * rdx count
121 *
122 * Output:
123 * eax uncopied bytes or 0 if successful.
124 */
125SYM_FUNC_START(copy_user_generic_string)
126	ASM_STAC
127	cmpl $8,%edx
128	jb 2f		/* less than 8 bytes, go to byte copy loop */
129	ALIGN_DESTINATION
130	movl %edx,%ecx
131	shrl $3,%ecx
132	andl $7,%edx
1331:	rep movsq
1342:	movl %edx,%ecx
1353:	rep movsb
136	xorl %eax,%eax
137	ASM_CLAC
138	RET
139
14011:	leal (%rdx,%rcx,8),%ecx
14112:	movl %ecx,%edx		/* ecx is zerorest also */
142	jmp .Lcopy_user_handle_tail
143
144	_ASM_EXTABLE_CPY(1b, 11b)
145	_ASM_EXTABLE_CPY(3b, 12b)
146SYM_FUNC_END(copy_user_generic_string)
147EXPORT_SYMBOL(copy_user_generic_string)
148
149/*
150 * Some CPUs are adding enhanced REP MOVSB/STOSB instructions.
151 * It's recommended to use enhanced REP MOVSB/STOSB if it's enabled.
152 *
153 * Input:
154 * rdi destination
155 * rsi source
156 * rdx count
157 *
158 * Output:
159 * eax uncopied bytes or 0 if successful.
160 */
161SYM_FUNC_START(copy_user_enhanced_fast_string)
162	ASM_STAC
163	/* CPUs without FSRM should avoid rep movsb for short copies */
164	ALTERNATIVE "cmpl $64, %edx; jb copy_user_short_string", "", X86_FEATURE_FSRM
165	movl %edx,%ecx
1661:	rep movsb
167	xorl %eax,%eax
168	ASM_CLAC
169	RET
170
17112:	movl %ecx,%edx		/* ecx is zerorest also */
172	jmp .Lcopy_user_handle_tail
173
174	_ASM_EXTABLE_CPY(1b, 12b)
175SYM_FUNC_END(copy_user_enhanced_fast_string)
176EXPORT_SYMBOL(copy_user_enhanced_fast_string)
177
178/*
179 * Try to copy last bytes and clear the rest if needed.
180 * Since protection fault in copy_from/to_user is not a normal situation,
181 * it is not necessary to optimize tail handling.
182 * Don't try to copy the tail if machine check happened
183 *
184 * Input:
185 * eax trap number written by ex_handler_copy()
186 * rdi destination
187 * rsi source
188 * rdx count
189 *
190 * Output:
191 * eax uncopied bytes or 0 if successful.
192 */
193SYM_CODE_START_LOCAL(.Lcopy_user_handle_tail)
194	cmp $X86_TRAP_MC,%eax
195	je 3f
196
197	movl %edx,%ecx
1981:	rep movsb
1992:	mov %ecx,%eax
200	ASM_CLAC
201	RET
202
2033:
204	movl %edx,%eax
205	ASM_CLAC
206	RET
207
208	_ASM_EXTABLE_CPY(1b, 2b)
209
210.Lcopy_user_handle_align:
211	addl %ecx,%edx			/* ecx is zerorest also */
212	jmp .Lcopy_user_handle_tail
213
214SYM_CODE_END(.Lcopy_user_handle_tail)
215
216/*
217 * Finish memcpy of less than 64 bytes.  #AC should already be set.
218 *
219 * Input:
220 * rdi destination
221 * rsi source
222 * rdx count (< 64)
223 *
224 * Output:
225 * eax uncopied bytes or 0 if successful.
226 */
227SYM_CODE_START_LOCAL(copy_user_short_string)
228	movl %edx,%ecx
229	andl $7,%edx
230	shrl $3,%ecx
231	jz .Lcopy_user_short_string_bytes
23218:	movq (%rsi),%r8
23319:	movq %r8,(%rdi)
234	leaq 8(%rsi),%rsi
235	leaq 8(%rdi),%rdi
236	decl %ecx
237	jnz 18b
238.Lcopy_user_short_string_bytes:
239	andl %edx,%edx
240	jz 23f
241	movl %edx,%ecx
24221:	movb (%rsi),%al
24322:	movb %al,(%rdi)
244	incq %rsi
245	incq %rdi
246	decl %ecx
247	jnz 21b
24823:	xor %eax,%eax
249	ASM_CLAC
250	RET
251
25240:	leal (%rdx,%rcx,8),%edx
253	jmp 60f
25450:	movl %ecx,%edx		/* ecx is zerorest also */
25560:	jmp .Lcopy_user_handle_tail
256
257	_ASM_EXTABLE_CPY(18b, 40b)
258	_ASM_EXTABLE_CPY(19b, 40b)
259	_ASM_EXTABLE_CPY(21b, 50b)
260	_ASM_EXTABLE_CPY(22b, 50b)
261SYM_CODE_END(copy_user_short_string)
262
263/*
264 * copy_user_nocache - Uncached memory copy with exception handling
265 * This will force destination out of cache for more performance.
266 *
267 * Note: Cached memory copy is used when destination or size is not
268 * naturally aligned. That is:
269 *  - Require 8-byte alignment when size is 8 bytes or larger.
270 *  - Require 4-byte alignment when size is 4 bytes.
271 */
272SYM_FUNC_START(__copy_user_nocache)
273	ASM_STAC
274
275	/* If size is less than 8 bytes, go to 4-byte copy */
276	cmpl $8,%edx
277	jb .L_4b_nocache_copy_entry
278
279	/* If destination is not 8-byte aligned, "cache" copy to align it */
280	ALIGN_DESTINATION
281
282	/* Set 4x8-byte copy count and remainder */
283	movl %edx,%ecx
284	andl $63,%edx
285	shrl $6,%ecx
286	jz .L_8b_nocache_copy_entry	/* jump if count is 0 */
287
288	/* Perform 4x8-byte nocache loop-copy */
289.L_4x8b_nocache_copy_loop:
2901:	movq (%rsi),%r8
2912:	movq 1*8(%rsi),%r9
2923:	movq 2*8(%rsi),%r10
2934:	movq 3*8(%rsi),%r11
2945:	movnti %r8,(%rdi)
2956:	movnti %r9,1*8(%rdi)
2967:	movnti %r10,2*8(%rdi)
2978:	movnti %r11,3*8(%rdi)
2989:	movq 4*8(%rsi),%r8
29910:	movq 5*8(%rsi),%r9
30011:	movq 6*8(%rsi),%r10
30112:	movq 7*8(%rsi),%r11
30213:	movnti %r8,4*8(%rdi)
30314:	movnti %r9,5*8(%rdi)
30415:	movnti %r10,6*8(%rdi)
30516:	movnti %r11,7*8(%rdi)
306	leaq 64(%rsi),%rsi
307	leaq 64(%rdi),%rdi
308	decl %ecx
309	jnz .L_4x8b_nocache_copy_loop
310
311	/* Set 8-byte copy count and remainder */
312.L_8b_nocache_copy_entry:
313	movl %edx,%ecx
314	andl $7,%edx
315	shrl $3,%ecx
316	jz .L_4b_nocache_copy_entry	/* jump if count is 0 */
317
318	/* Perform 8-byte nocache loop-copy */
319.L_8b_nocache_copy_loop:
32020:	movq (%rsi),%r8
32121:	movnti %r8,(%rdi)
322	leaq 8(%rsi),%rsi
323	leaq 8(%rdi),%rdi
324	decl %ecx
325	jnz .L_8b_nocache_copy_loop
326
327	/* If no byte left, we're done */
328.L_4b_nocache_copy_entry:
329	andl %edx,%edx
330	jz .L_finish_copy
331
332	/* If destination is not 4-byte aligned, go to byte copy: */
333	movl %edi,%ecx
334	andl $3,%ecx
335	jnz .L_1b_cache_copy_entry
336
337	/* Set 4-byte copy count (1 or 0) and remainder */
338	movl %edx,%ecx
339	andl $3,%edx
340	shrl $2,%ecx
341	jz .L_1b_cache_copy_entry	/* jump if count is 0 */
342
343	/* Perform 4-byte nocache copy: */
34430:	movl (%rsi),%r8d
34531:	movnti %r8d,(%rdi)
346	leaq 4(%rsi),%rsi
347	leaq 4(%rdi),%rdi
348
349	/* If no bytes left, we're done: */
350	andl %edx,%edx
351	jz .L_finish_copy
352
353	/* Perform byte "cache" loop-copy for the remainder */
354.L_1b_cache_copy_entry:
355	movl %edx,%ecx
356.L_1b_cache_copy_loop:
35740:	movb (%rsi),%al
35841:	movb %al,(%rdi)
359	incq %rsi
360	incq %rdi
361	decl %ecx
362	jnz .L_1b_cache_copy_loop
363
364	/* Finished copying; fence the prior stores */
365.L_finish_copy:
366	xorl %eax,%eax
367	ASM_CLAC
368	sfence
369	RET
370
371.L_fixup_4x8b_copy:
372	shll $6,%ecx
373	addl %ecx,%edx
374	jmp .L_fixup_handle_tail
375.L_fixup_8b_copy:
376	lea (%rdx,%rcx,8),%rdx
377	jmp .L_fixup_handle_tail
378.L_fixup_4b_copy:
379	lea (%rdx,%rcx,4),%rdx
380	jmp .L_fixup_handle_tail
381.L_fixup_1b_copy:
382	movl %ecx,%edx
383.L_fixup_handle_tail:
384	sfence
385	jmp .Lcopy_user_handle_tail
386
387	_ASM_EXTABLE_CPY(1b, .L_fixup_4x8b_copy)
388	_ASM_EXTABLE_CPY(2b, .L_fixup_4x8b_copy)
389	_ASM_EXTABLE_CPY(3b, .L_fixup_4x8b_copy)
390	_ASM_EXTABLE_CPY(4b, .L_fixup_4x8b_copy)
391	_ASM_EXTABLE_CPY(5b, .L_fixup_4x8b_copy)
392	_ASM_EXTABLE_CPY(6b, .L_fixup_4x8b_copy)
393	_ASM_EXTABLE_CPY(7b, .L_fixup_4x8b_copy)
394	_ASM_EXTABLE_CPY(8b, .L_fixup_4x8b_copy)
395	_ASM_EXTABLE_CPY(9b, .L_fixup_4x8b_copy)
396	_ASM_EXTABLE_CPY(10b, .L_fixup_4x8b_copy)
397	_ASM_EXTABLE_CPY(11b, .L_fixup_4x8b_copy)
398	_ASM_EXTABLE_CPY(12b, .L_fixup_4x8b_copy)
399	_ASM_EXTABLE_CPY(13b, .L_fixup_4x8b_copy)
400	_ASM_EXTABLE_CPY(14b, .L_fixup_4x8b_copy)
401	_ASM_EXTABLE_CPY(15b, .L_fixup_4x8b_copy)
402	_ASM_EXTABLE_CPY(16b, .L_fixup_4x8b_copy)
403	_ASM_EXTABLE_CPY(20b, .L_fixup_8b_copy)
404	_ASM_EXTABLE_CPY(21b, .L_fixup_8b_copy)
405	_ASM_EXTABLE_CPY(30b, .L_fixup_4b_copy)
406	_ASM_EXTABLE_CPY(31b, .L_fixup_4b_copy)
407	_ASM_EXTABLE_CPY(40b, .L_fixup_1b_copy)
408	_ASM_EXTABLE_CPY(41b, .L_fixup_1b_copy)
409SYM_FUNC_END(__copy_user_nocache)
410EXPORT_SYMBOL(__copy_user_nocache)
411