• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1/* SPDX-License-Identifier: GPL-2.0-only */
2/* Copyright 2002 Andi Kleen */
3
4#include <linux/linkage.h>
5#include <asm/errno.h>
6#include <asm/cpufeatures.h>
7#include <asm/mcsafe_test.h>
8#include <asm/alternative-asm.h>
9#include <asm/export.h>
10
11/*
12 * We build a jump to memcpy_orig by default which gets NOPped out on
13 * the majority of x86 CPUs which set REP_GOOD. In addition, CPUs which
14 * have the enhanced REP MOVSB/STOSB feature (ERMS), change those NOPs
15 * to a jmp to memcpy_erms which does the REP; MOVSB mem copy.
16 */
17
18/*
19 * memcpy - Copy a memory block.
20 *
21 * Input:
22 *  rdi destination
23 *  rsi source
24 *  rdx count
25 *
26 * Output:
27 * rax original destination
28 */
29ENTRY(__memcpy)
30SYM_FUNC_START_WEAK(memcpy)
31	ALTERNATIVE_2 "jmp memcpy_orig", "", X86_FEATURE_REP_GOOD, \
32		      "jmp memcpy_erms", X86_FEATURE_ERMS
33
34	movq %rdi, %rax
35	movq %rdx, %rcx
36	shrq $3, %rcx
37	andl $7, %edx
38	rep movsq
39	movl %edx, %ecx
40	rep movsb
41	ret
42ENDPROC(memcpy)
43ENDPROC(__memcpy)
44EXPORT_SYMBOL(memcpy)
45EXPORT_SYMBOL(__memcpy)
46
47/*
48 * memcpy_erms() - enhanced fast string memcpy. This is faster and
49 * simpler than memcpy. Use memcpy_erms when possible.
50 */
51ENTRY(memcpy_erms)
52	movq %rdi, %rax
53	movq %rdx, %rcx
54	rep movsb
55	ret
56ENDPROC(memcpy_erms)
57
58ENTRY(memcpy_orig)
59	movq %rdi, %rax
60
61	cmpq $0x20, %rdx
62	jb .Lhandle_tail
63
64	/*
65	 * We check whether memory false dependence could occur,
66	 * then jump to corresponding copy mode.
67	 */
68	cmp  %dil, %sil
69	jl .Lcopy_backward
70	subq $0x20, %rdx
71.Lcopy_forward_loop:
72	subq $0x20,	%rdx
73
74	/*
75	 * Move in blocks of 4x8 bytes:
76	 */
77	movq 0*8(%rsi),	%r8
78	movq 1*8(%rsi),	%r9
79	movq 2*8(%rsi),	%r10
80	movq 3*8(%rsi),	%r11
81	leaq 4*8(%rsi),	%rsi
82
83	movq %r8,	0*8(%rdi)
84	movq %r9,	1*8(%rdi)
85	movq %r10,	2*8(%rdi)
86	movq %r11,	3*8(%rdi)
87	leaq 4*8(%rdi),	%rdi
88	jae  .Lcopy_forward_loop
89	addl $0x20,	%edx
90	jmp  .Lhandle_tail
91
92.Lcopy_backward:
93	/*
94	 * Calculate copy position to tail.
95	 */
96	addq %rdx,	%rsi
97	addq %rdx,	%rdi
98	subq $0x20,	%rdx
99	/*
100	 * At most 3 ALU operations in one cycle,
101	 * so append NOPS in the same 16 bytes trunk.
102	 */
103	.p2align 4
104.Lcopy_backward_loop:
105	subq $0x20,	%rdx
106	movq -1*8(%rsi),	%r8
107	movq -2*8(%rsi),	%r9
108	movq -3*8(%rsi),	%r10
109	movq -4*8(%rsi),	%r11
110	leaq -4*8(%rsi),	%rsi
111	movq %r8,		-1*8(%rdi)
112	movq %r9,		-2*8(%rdi)
113	movq %r10,		-3*8(%rdi)
114	movq %r11,		-4*8(%rdi)
115	leaq -4*8(%rdi),	%rdi
116	jae  .Lcopy_backward_loop
117
118	/*
119	 * Calculate copy position to head.
120	 */
121	addl $0x20,	%edx
122	subq %rdx,	%rsi
123	subq %rdx,	%rdi
124.Lhandle_tail:
125	cmpl $16,	%edx
126	jb   .Lless_16bytes
127
128	/*
129	 * Move data from 16 bytes to 31 bytes.
130	 */
131	movq 0*8(%rsi), %r8
132	movq 1*8(%rsi),	%r9
133	movq -2*8(%rsi, %rdx),	%r10
134	movq -1*8(%rsi, %rdx),	%r11
135	movq %r8,	0*8(%rdi)
136	movq %r9,	1*8(%rdi)
137	movq %r10,	-2*8(%rdi, %rdx)
138	movq %r11,	-1*8(%rdi, %rdx)
139	retq
140	.p2align 4
141.Lless_16bytes:
142	cmpl $8,	%edx
143	jb   .Lless_8bytes
144	/*
145	 * Move data from 8 bytes to 15 bytes.
146	 */
147	movq 0*8(%rsi),	%r8
148	movq -1*8(%rsi, %rdx),	%r9
149	movq %r8,	0*8(%rdi)
150	movq %r9,	-1*8(%rdi, %rdx)
151	retq
152	.p2align 4
153.Lless_8bytes:
154	cmpl $4,	%edx
155	jb   .Lless_3bytes
156
157	/*
158	 * Move data from 4 bytes to 7 bytes.
159	 */
160	movl (%rsi), %ecx
161	movl -4(%rsi, %rdx), %r8d
162	movl %ecx, (%rdi)
163	movl %r8d, -4(%rdi, %rdx)
164	retq
165	.p2align 4
166.Lless_3bytes:
167	subl $1, %edx
168	jb .Lend
169	/*
170	 * Move data from 1 bytes to 3 bytes.
171	 */
172	movzbl (%rsi), %ecx
173	jz .Lstore_1byte
174	movzbq 1(%rsi), %r8
175	movzbq (%rsi, %rdx), %r9
176	movb %r8b, 1(%rdi)
177	movb %r9b, (%rdi, %rdx)
178.Lstore_1byte:
179	movb %cl, (%rdi)
180
181.Lend:
182	retq
183ENDPROC(memcpy_orig)
184
185#ifndef CONFIG_UML
186
187MCSAFE_TEST_CTL
188
189/*
190 * __memcpy_mcsafe - memory copy with machine check exception handling
191 * Note that we only catch machine checks when reading the source addresses.
192 * Writes to target are posted and don't generate machine checks.
193 */
194ENTRY(__memcpy_mcsafe)
195	cmpl $8, %edx
196	/* Less than 8 bytes? Go to byte copy loop */
197	jb .L_no_whole_words
198
199	/* Check for bad alignment of source */
200	testl $7, %esi
201	/* Already aligned */
202	jz .L_8byte_aligned
203
204	/* Copy one byte at a time until source is 8-byte aligned */
205	movl %esi, %ecx
206	andl $7, %ecx
207	subl $8, %ecx
208	negl %ecx
209	subl %ecx, %edx
210.L_read_leading_bytes:
211	movb (%rsi), %al
212	MCSAFE_TEST_SRC %rsi 1 .E_leading_bytes
213	MCSAFE_TEST_DST %rdi 1 .E_leading_bytes
214.L_write_leading_bytes:
215	movb %al, (%rdi)
216	incq %rsi
217	incq %rdi
218	decl %ecx
219	jnz .L_read_leading_bytes
220
221.L_8byte_aligned:
222	movl %edx, %ecx
223	andl $7, %edx
224	shrl $3, %ecx
225	jz .L_no_whole_words
226
227.L_read_words:
228	movq (%rsi), %r8
229	MCSAFE_TEST_SRC %rsi 8 .E_read_words
230	MCSAFE_TEST_DST %rdi 8 .E_write_words
231.L_write_words:
232	movq %r8, (%rdi)
233	addq $8, %rsi
234	addq $8, %rdi
235	decl %ecx
236	jnz .L_read_words
237
238	/* Any trailing bytes? */
239.L_no_whole_words:
240	andl %edx, %edx
241	jz .L_done_memcpy_trap
242
243	/* Copy trailing bytes */
244	movl %edx, %ecx
245.L_read_trailing_bytes:
246	movb (%rsi), %al
247	MCSAFE_TEST_SRC %rsi 1 .E_trailing_bytes
248	MCSAFE_TEST_DST %rdi 1 .E_trailing_bytes
249.L_write_trailing_bytes:
250	movb %al, (%rdi)
251	incq %rsi
252	incq %rdi
253	decl %ecx
254	jnz .L_read_trailing_bytes
255
256	/* Copy successful. Return zero */
257.L_done_memcpy_trap:
258	xorl %eax, %eax
259.L_done:
260	ret
261ENDPROC(__memcpy_mcsafe)
262EXPORT_SYMBOL_GPL(__memcpy_mcsafe)
263
264	.section .fixup, "ax"
265	/*
266	 * Return number of bytes not copied for any failure. Note that
267	 * there is no "tail" handling since the source buffer is 8-byte
268	 * aligned and poison is cacheline aligned.
269	 */
270.E_read_words:
271	shll	$3, %ecx
272.E_leading_bytes:
273	addl	%edx, %ecx
274.E_trailing_bytes:
275	mov	%ecx, %eax
276	jmp	.L_done
277
278	/*
279	 * For write fault handling, given the destination is unaligned,
280	 * we handle faults on multi-byte writes with a byte-by-byte
281	 * copy up to the write-protected page.
282	 */
283.E_write_words:
284	shll	$3, %ecx
285	addl	%edx, %ecx
286	movl	%ecx, %edx
287	jmp mcsafe_handle_tail
288
289	.previous
290
291	_ASM_EXTABLE_FAULT(.L_read_leading_bytes, .E_leading_bytes)
292	_ASM_EXTABLE_FAULT(.L_read_words, .E_read_words)
293	_ASM_EXTABLE_FAULT(.L_read_trailing_bytes, .E_trailing_bytes)
294	_ASM_EXTABLE(.L_write_leading_bytes, .E_leading_bytes)
295	_ASM_EXTABLE(.L_write_words, .E_write_words)
296	_ASM_EXTABLE(.L_write_trailing_bytes, .E_trailing_bytes)
297#endif
298