1 | /* |
2 | * Copyright (C) 2016-2018 Apple Inc. All rights reserved. |
3 | * |
4 | * Redistribution and use in source and binary forms, with or without |
5 | * modification, are permitted provided that the following conditions |
6 | * are met: |
7 | * 1. Redistributions of source code must retain the above copyright |
8 | * notice, this list of conditions and the following disclaimer. |
9 | * 2. Redistributions in binary form must reproduce the above copyright |
10 | * notice, this list of conditions and the following disclaimer in the |
11 | * documentation and/or other materials provided with the distribution. |
12 | * |
13 | * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY |
14 | * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
15 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR |
16 | * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR |
17 | * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, |
18 | * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
19 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR |
20 | * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY |
21 | * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
22 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
23 | * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
24 | */ |
25 | |
26 | #pragma once |
27 | |
28 | #include "CPU.h" |
29 | #include "JSCJSValue.h" |
30 | |
31 | namespace JSC { |
32 | |
33 | // We use these memory operations when modifying memory that might be scanned by the concurrent collector. |
34 | // We don't call the default operations because they're not guaranteed to store to memory in eight byte aligned |
35 | // chunks. If we happened to fall into the system's normal byte copy loop, we may see a torn JSValue in the |
36 | // concurrent collector. |
37 | |
38 | constexpr size_t smallCutoff = 30 * 8; |
39 | constexpr size_t mediumCutoff = 4 * 1024; |
40 | |
41 | // This is a forwards loop so gcSafeMemmove can rely on the direction. |
42 | template <typename T> |
43 | ALWAYS_INLINE void gcSafeMemcpy(T* dst, T* src, size_t bytes) |
44 | { |
45 | static_assert(sizeof(T) == sizeof(JSValue)); |
46 | RELEASE_ASSERT(bytes % 8 == 0); |
47 | |
48 | #if USE(JSVALUE64) |
49 | |
50 | auto slowPathForwardMemcpy = [&] { |
51 | size_t count = bytes / 8; |
52 | for (unsigned i = 0; i < count; ++i) |
53 | bitwise_cast<volatile uint64_t*>(dst)[i] = bitwise_cast<volatile uint64_t*>(src)[i]; |
54 | }; |
55 | |
56 | #if COMPILER(GCC_COMPATIBLE) && USE(JSVALUE64) |
57 | if (bytes <= smallCutoff) |
58 | slowPathForwardMemcpy(); |
59 | else if (isARM64() || bytes <= mediumCutoff) { |
60 | #if CPU(X86_64) |
61 | size_t alignedBytes = (bytes / 64) * 64; |
62 | size_t tmp; |
63 | size_t offset = 0; |
64 | asm volatile( |
65 | ".balign 32\t\n" |
66 | "1:\t\n" |
67 | "cmpq %q[offset], %q[alignedBytes]\t\n" |
68 | "je 2f\t\n" |
69 | "movups (%q[src], %q[offset], 1), %%xmm0\t\n" |
70 | "movups 16(%q[src], %q[offset], 1), %%xmm1\t\n" |
71 | "movups 32(%q[src], %q[offset], 1), %%xmm2\t\n" |
72 | "movups 48(%q[src], %q[offset], 1), %%xmm3\t\n" |
73 | "movups %%xmm0, (%q[dst], %q[offset], 1)\t\n" |
74 | "movups %%xmm1, 16(%q[dst], %q[offset], 1)\t\n" |
75 | "movups %%xmm2, 32(%q[dst], %q[offset], 1)\t\n" |
76 | "movups %%xmm3, 48(%q[dst], %q[offset], 1)\t\n" |
77 | "addq $64, %q[offset]\t\n" |
78 | "jmp 1b\t\n" |
79 | |
80 | "2:\t\n" |
81 | "cmpq %q[offset], %q[bytes]\t\n" |
82 | "je 3f\t\n" |
83 | "movq (%q[src], %q[offset], 1), %q[tmp]\t\n" |
84 | "movq %q[tmp], (%q[dst], %q[offset], 1)\t\n" |
85 | "addq $8, %q[offset]\t\n" |
86 | "jmp 2b\t\n" |
87 | |
88 | "3:\t\n" |
89 | |
90 | : [alignedBytes] "+r" (alignedBytes), [bytes] "+r" (bytes), [tmp] "+r" (tmp), [offset] "+r" (offset), [dst] "+r" (dst), [src] "+r" (src) |
91 | : |
92 | : "xmm0" , "xmm1" , "xmm2" , "xmm3" , "memory" , "cc" |
93 | ); |
94 | #elif CPU(ARM64) |
95 | uint64_t alignedBytes = (static_cast<uint64_t>(bytes) / 16) * 16; |
96 | size_t offset = 0; |
97 | |
98 | uint64_t dstPtr = static_cast<uint64_t>(bitwise_cast<uintptr_t>(dst)); |
99 | uint64_t srcPtr = static_cast<uint64_t>(bitwise_cast<uintptr_t>(src)); |
100 | |
101 | asm volatile( |
102 | "1:\t\n" |
103 | "cmp %x[offset], %x[alignedBytes]\t\n" |
104 | "b.eq 2f\t\n" |
105 | "ldr q0, [%x[srcPtr], %x[offset]]\t\n" |
106 | "str q0, [%x[dstPtr], %x[offset]]\t\n" |
107 | "add %x[offset], %x[offset], #0x10\t\n" |
108 | "b 1b\t\n" |
109 | |
110 | "2:\t\n" |
111 | "cmp %x[offset], %x[bytes]\t\n" |
112 | "b.eq 3f\t\n" |
113 | "ldr d0, [%x[srcPtr], %x[offset]]\t\n" |
114 | "str d0, [%x[dstPtr], %x[offset]]\t\n" |
115 | "add %x[offset], %x[offset], #0x8\t\n" |
116 | "b 2b\t\n" |
117 | |
118 | "3:\t\n" |
119 | |
120 | : [alignedBytes] "+r" (alignedBytes), [bytes] "+r" (bytes), [offset] "+r" (offset), [dstPtr] "+r" (dstPtr), [srcPtr] "+r" (srcPtr) |
121 | : |
122 | : "d0" , "d1" , "memory" |
123 | ); |
124 | #else |
125 | #error "Unknown architecture." |
126 | #endif // CPU(X86_64) |
127 | } else { |
128 | RELEASE_ASSERT(isX86_64()); |
129 | #if CPU(X86_64) |
130 | size_t count = bytes / 8; |
131 | asm volatile( |
132 | ".balign 16\t\n" |
133 | "cld\t\n" |
134 | "rep movsq\t\n" |
135 | : "+D" (dst), "+S" (src), "+c" (count) |
136 | : |
137 | : "memory" ); |
138 | #endif // CPU(X86_64) |
139 | } |
140 | #else |
141 | slowPathForwardMemcpy(); |
142 | #endif // COMPILER(GCC_COMPATIBLE) |
143 | #else |
144 | memcpy(dst, src, bytes); |
145 | #endif // USE(JSVALUE64) |
146 | } |
147 | |
148 | template <typename T> |
149 | ALWAYS_INLINE void gcSafeMemmove(T* dst, T* src, size_t bytes) |
150 | { |
151 | static_assert(sizeof(T) == sizeof(JSValue)); |
152 | RELEASE_ASSERT(bytes % 8 == 0); |
153 | #if USE(JSVALUE64) |
154 | if (bitwise_cast<uintptr_t>(src) >= bitwise_cast<uintptr_t>(dst)) { |
155 | // This is written to do a forwards loop, so calling it is ok. |
156 | gcSafeMemcpy(dst, src, bytes); |
157 | return; |
158 | } |
159 | |
160 | if ((static_cast<uint64_t>(bitwise_cast<uintptr_t>(src)) + static_cast<uint64_t>(bytes)) <= static_cast<uint64_t>(bitwise_cast<uintptr_t>(dst))) { |
161 | gcSafeMemcpy(dst, src, bytes); |
162 | return; |
163 | } |
164 | |
165 | auto slowPathBackwardsMemmove = [&] { |
166 | size_t count = bytes / 8; |
167 | for (size_t i = count; i--; ) |
168 | bitwise_cast<volatile uint64_t*>(dst)[i] = bitwise_cast<volatile uint64_t*>(src)[i]; |
169 | }; |
170 | |
171 | #if COMPILER(GCC_COMPATIBLE) |
172 | if (bytes <= smallCutoff) |
173 | slowPathBackwardsMemmove(); |
174 | else { |
175 | #if CPU(X86_64) |
176 | size_t alignedBytes = (bytes / 64) * 64; |
177 | |
178 | size_t tail = alignedBytes; |
179 | size_t tmp; |
180 | asm volatile( |
181 | "2:\t\n" |
182 | "cmpq %q[tail], %q[bytes]\t\n" |
183 | "je 1f\t\n" |
184 | "addq $-8, %q[bytes]\t\n" |
185 | "movq (%q[src], %q[bytes], 1), %q[tmp]\t\n" |
186 | "movq %q[tmp], (%q[dst], %q[bytes], 1)\t\n" |
187 | "jmp 2b\t\n" |
188 | |
189 | "1:\t\n" |
190 | "test %q[alignedBytes], %q[alignedBytes]\t\n" |
191 | "jz 3f\t\n" |
192 | |
193 | ".balign 32\t\n" |
194 | "100:\t\n" |
195 | |
196 | "movups -64(%q[src], %q[alignedBytes], 1), %%xmm0\t\n" |
197 | "movups -48(%q[src], %q[alignedBytes], 1), %%xmm1\t\n" |
198 | "movups -32(%q[src], %q[alignedBytes], 1), %%xmm2\t\n" |
199 | "movups -16(%q[src], %q[alignedBytes], 1), %%xmm3\t\n" |
200 | "movups %%xmm0, -64(%q[dst], %q[alignedBytes], 1)\t\n" |
201 | "movups %%xmm1, -48(%q[dst], %q[alignedBytes], 1)\t\n" |
202 | "movups %%xmm2, -32(%q[dst], %q[alignedBytes], 1)\t\n" |
203 | "movups %%xmm3, -16(%q[dst], %q[alignedBytes], 1)\t\n" |
204 | "addq $-64, %q[alignedBytes]\t\n" |
205 | "jnz 100b\t\n" |
206 | |
207 | "3:\t\n" |
208 | |
209 | : [alignedBytes] "+r" (alignedBytes), [tail] "+r" (tail), [bytes] "+r" (bytes), [tmp] "+r" (tmp), [dst] "+r" (dst), [src] "+r" (src) |
210 | : |
211 | : "xmm0" , "xmm1" , "xmm2" , "xmm3" , "memory" , "cc" |
212 | ); |
213 | #elif CPU(ARM64) |
214 | uint64_t alignedBytes = (static_cast<uint64_t>(bytes) / 16) * 16; |
215 | uint64_t dstPtr = static_cast<uint64_t>(bitwise_cast<uintptr_t>(dst)); |
216 | uint64_t srcPtr = static_cast<uint64_t>(bitwise_cast<uintptr_t>(src)); |
217 | |
218 | asm volatile( |
219 | "1:\t\n" |
220 | "cmp %x[alignedBytes], %x[bytes]\t\n" |
221 | "b.eq 2f\t\n" |
222 | "sub %x[bytes], %x[bytes], #0x8\t\n" |
223 | "ldr d0, [%x[srcPtr], %x[bytes]]\t\n" |
224 | "str d0, [%x[dstPtr], %x[bytes]]\t\n" |
225 | "b 1b\t\n" |
226 | |
227 | "2:\t\n" |
228 | "cbz %x[alignedBytes], 3f\t\n" |
229 | "sub %x[alignedBytes], %x[alignedBytes], #0x10\t\n" |
230 | "ldr q0, [%x[srcPtr], %x[alignedBytes]]\t\n" |
231 | "str q0, [%x[dstPtr], %x[alignedBytes]]\t\n" |
232 | "b 2b\t\n" |
233 | |
234 | "3:\t\n" |
235 | |
236 | : [alignedBytes] "+r" (alignedBytes), [bytes] "+r" (bytes), [dstPtr] "+r" (dstPtr), [srcPtr] "+r" (srcPtr) |
237 | : |
238 | : "d0" , "d1" , "memory" |
239 | ); |
240 | #else |
241 | #error "Unknown architecture." |
242 | #endif // CPU(X86_64) |
243 | } |
244 | #else |
245 | slowPathBackwardsMemmove(); |
246 | #endif // COMPILER(GCC_COMPATIBLE) |
247 | #else |
248 | memmove(dst, src, bytes); |
249 | #endif // USE(JSVALUE64) |
250 | } |
251 | |
252 | template <typename T> |
253 | ALWAYS_INLINE void gcSafeZeroMemory(T* dst, size_t bytes) |
254 | { |
255 | static_assert(sizeof(T) == sizeof(JSValue)); |
256 | RELEASE_ASSERT(bytes % 8 == 0); |
257 | #if USE(JSVALUE64) |
258 | #if COMPILER(GCC_COMPATIBLE) |
259 | #if CPU(X86_64) |
260 | uint64_t zero = 0; |
261 | size_t count = bytes / 8; |
262 | asm volatile ( |
263 | "rep stosq\n\t" |
264 | : "+D" (dst), "+c" (count) |
265 | : "a" (zero) |
266 | : "memory" |
267 | ); |
268 | #elif CPU(ARM64) |
269 | uint64_t alignedBytes = (static_cast<uint64_t>(bytes) / 64) * 64; |
270 | uint64_t dstPtr = static_cast<uint64_t>(bitwise_cast<uintptr_t>(dst)); |
271 | uint64_t end = dstPtr + bytes; |
272 | uint64_t alignedEnd = dstPtr + alignedBytes; |
273 | asm volatile( |
274 | "movi d0, #0\t\n" |
275 | "movi d1, #0\t\n" |
276 | |
277 | ".p2align 4\t\n" |
278 | "2:\t\n" |
279 | "cmp %x[dstPtr], %x[alignedEnd]\t\n" |
280 | "b.eq 4f\t\n" |
281 | "stnp q0, q0, [%x[dstPtr]]\t\n" |
282 | "stnp q0, q0, [%x[dstPtr], #0x20]\t\n" |
283 | "add %x[dstPtr], %x[dstPtr], #0x40\t\n" |
284 | "b 2b\t\n" |
285 | |
286 | "4:\t\n" |
287 | "cmp %x[dstPtr], %x[end]\t\n" |
288 | "b.eq 5f\t\n" |
289 | "str d0, [%x[dstPtr]], #0x8\t\n" |
290 | "b 4b\t\n" |
291 | |
292 | "5:\t\n" |
293 | |
294 | : [alignedBytes] "+r" (alignedBytes), [bytes] "+r" (bytes), [dstPtr] "+r" (dstPtr), [end] "+r" (end), [alignedEnd] "+r" (alignedEnd) |
295 | : |
296 | : "d0" , "d1" , "memory" |
297 | ); |
298 | #else |
299 | #error "Unknown architecture." |
300 | #endif // CPU(X86_64) |
301 | #else |
302 | size_t count = bytes / 8; |
303 | for (size_t i = 0; i < count; ++i) |
304 | bitwise_cast<volatile uint64_t*>(dst)[i] = 0; |
305 | #endif // COMPILER(GCC_COMPATIBLE) |
306 | #else |
307 | memset(dst, 0, bytes); |
308 | #endif // USE(JSVALUE64) |
309 | } |
310 | |
311 | } // namespace JSC |
312 | |