1 /*
2 * Copyright (c) 2006, 2025, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "utilities/copy.hpp"
26 #include "runtime/sharedRuntime.hpp"
27 #include "utilities/align.hpp"
28 #include "utilities/byteswap.hpp"
29 #include "utilities/copy.hpp"
30
31
32 // Copy bytes; larger units are filled atomically if everything is aligned.
33 void Copy::conjoint_memory_atomic(const void* from, void* to, size_t size) {
34 uintptr_t bits = (uintptr_t) from | (uintptr_t) to | (uintptr_t) size;
35
36 // (Note: We could improve performance by ignoring the low bits of size,
37 // and putting a short cleanup loop after each bulk copy loop.
38 // There are plenty of other ways to make this faster also,
39 // and it's a slippery slope. For now, let's keep this code simple
40 // since the simplicity helps clarify the atomicity semantics of
41 // this operation. There are also CPU-specific assembly versions
42 // which may or may not want to include such optimizations.)
43
44 if (bits % sizeof(jlong) == 0) {
45 Copy::conjoint_jlongs_atomic((const jlong*) from, (jlong*) to, size / sizeof(jlong));
46 } else if (bits % sizeof(jint) == 0) {
47 Copy::conjoint_jints_atomic((const jint*) from, (jint*) to, size / sizeof(jint));
48 } else if (bits % sizeof(jshort) == 0) {
49 Copy::conjoint_jshorts_atomic((const jshort*) from, (jshort*) to, size / sizeof(jshort));
50 } else {
51 // Not aligned, so no need to be atomic.
52 Copy::conjoint_jbytes((const void*) from, (void*) to, size);
53 }
54 }
55
56 #define COPY_ALIGNED_SEGMENT(t) \
57 if (bits % sizeof(t) == 0) { \
58 size_t segment = remain / sizeof(t); \
59 if (segment > 0) { \
60 Copy::conjoint_##t##s_atomic((const t*) cursor_from, (t*) cursor_to, segment); \
61 remain -= segment * sizeof(t); \
62 cursor_from = (void*)(((char*)cursor_from) + segment * sizeof(t)); \
63 cursor_to = (void*)(((char*)cursor_to) + segment * sizeof(t)); \
64 } \
65 } \
66
67 void Copy::copy_value_content(const void* from, void* to, size_t size) {
68 // Simple cases first
69 uintptr_t bits = (uintptr_t) from | (uintptr_t) to | (uintptr_t) size;
70 if (bits % sizeof(jlong) == 0) {
71 Copy::conjoint_jlongs_atomic((const jlong*) from, (jlong*) to, size / sizeof(jlong));
72 return;
73 } else if (bits % sizeof(jint) == 0) {
74 Copy::conjoint_jints_atomic((const jint*) from, (jint*) to, size / sizeof(jint));
75 return;
76 } else if (bits % sizeof(jshort) == 0) {
77 Copy::conjoint_jshorts_atomic((const jshort*) from, (jshort*) to, size / sizeof(jshort));
78 return;
79 }
80
81 // Complex cases
82 bits = (uintptr_t) from | (uintptr_t) to;
83 const void* cursor_from = from;
84 void* cursor_to = to;
85 size_t remain = size;
86 COPY_ALIGNED_SEGMENT(jlong)
87 COPY_ALIGNED_SEGMENT(jint)
88 COPY_ALIGNED_SEGMENT(jshort)
89 if (remain > 0) {
90 Copy::conjoint_jbytes((const void*) cursor_from, (void*) cursor_to, remain);
91 }
92 }
93
94 #undef COPY_ALIGNED_SEGMENT
95
96 class CopySwap : AllStatic {
97 public:
98 /**
99 * Copy and optionally byte swap elements
100 *
101 * <swap> - true if elements should be byte swapped
102 *
103 * @param src address of source
104 * @param dst address of destination
105 * @param byte_count number of bytes to copy
106 * @param elem_size size of the elements to copy-swap
107 */
108 template<bool swap>
109 static void conjoint_swap_if_needed(const void* src, void* dst, size_t byte_count, size_t elem_size) {
110 assert(src != nullptr, "address must not be null");
111 assert(dst != nullptr, "address must not be null");
112 assert(elem_size == 2 || elem_size == 4 || elem_size == 8,
113 "incorrect element size: %zu", elem_size);
114 assert(is_aligned(byte_count, elem_size),
115 "byte_count %zu must be multiple of element size %zu", byte_count, elem_size);
116
117 address src_end = (address)src + byte_count;
118
119 if (dst <= src || dst >= src_end) {
120 do_conjoint_swap<RIGHT,swap>(src, dst, byte_count, elem_size);
121 } else {
122 do_conjoint_swap<LEFT,swap>(src, dst, byte_count, elem_size);
123 }
124 }
125
126 private:
127 enum CopyDirection {
128 RIGHT, // lower -> higher address
129 LEFT // higher -> lower address
130 };
131
132 /**
133 * Copy and byte swap elements
134 *
135 * <T> - type of element to copy
136 * <D> - copy direction
137 * <is_src_aligned> - true if src argument is aligned to element size
138 * <is_dst_aligned> - true if dst argument is aligned to element size
139 *
140 * @param src address of source
141 * @param dst address of destination
142 * @param byte_count number of bytes to copy
143 */
144 template <typename T, CopyDirection D, bool swap, bool is_src_aligned, bool is_dst_aligned>
145 static void do_conjoint_swap(const void* src, void* dst, size_t byte_count) {
146 const char* cur_src;
147 char* cur_dst;
148
149 switch (D) {
150 case RIGHT:
151 cur_src = (const char*)src;
152 cur_dst = (char*)dst;
153 break;
154 case LEFT:
155 cur_src = (const char*)src + byte_count - sizeof(T);
156 cur_dst = (char*)dst + byte_count - sizeof(T);
157 break;
158 }
159
160 for (size_t i = 0; i < byte_count / sizeof(T); i++) {
161 T tmp;
162
163 if (is_src_aligned) {
164 tmp = *(T*)cur_src;
165 } else {
166 memcpy(&tmp, cur_src, sizeof(T));
167 }
168
169 if (swap) {
170 tmp = byteswap(tmp);
171 }
172
173 if (is_dst_aligned) {
174 *(T*)cur_dst = tmp;
175 } else {
176 memcpy(cur_dst, &tmp, sizeof(T));
177 }
178
179 switch (D) {
180 case RIGHT:
181 cur_src += sizeof(T);
182 cur_dst += sizeof(T);
183 break;
184 case LEFT:
185 cur_src -= sizeof(T);
186 cur_dst -= sizeof(T);
187 break;
188 }
189 }
190 }
191
192 /**
193 * Copy and byte swap elements
194 *
195 * <T> - type of element to copy
196 * <D> - copy direction
197 * <swap> - true if elements should be byte swapped
198 *
199 * @param src address of source
200 * @param dst address of destination
201 * @param byte_count number of bytes to copy
202 */
203 template <typename T, CopyDirection direction, bool swap>
204 static void do_conjoint_swap(const void* src, void* dst, size_t byte_count) {
205 if (is_aligned(src, sizeof(T))) {
206 if (is_aligned(dst, sizeof(T))) {
207 do_conjoint_swap<T,direction,swap,true,true>(src, dst, byte_count);
208 } else {
209 do_conjoint_swap<T,direction,swap,true,false>(src, dst, byte_count);
210 }
211 } else {
212 if (is_aligned(dst, sizeof(T))) {
213 do_conjoint_swap<T,direction,swap,false,true>(src, dst, byte_count);
214 } else {
215 do_conjoint_swap<T,direction,swap,false,false>(src, dst, byte_count);
216 }
217 }
218 }
219
220
221 /**
222 * Copy and byte swap elements
223 *
224 * <D> - copy direction
225 * <swap> - true if elements should be byte swapped
226 *
227 * @param src address of source
228 * @param dst address of destination
229 * @param byte_count number of bytes to copy
230 * @param elem_size size of the elements to copy-swap
231 */
232 template <CopyDirection D, bool swap>
233 static void do_conjoint_swap(const void* src, void* dst, size_t byte_count, size_t elem_size) {
234 switch (elem_size) {
235 case 2: do_conjoint_swap<uint16_t,D,swap>(src, dst, byte_count); break;
236 case 4: do_conjoint_swap<uint32_t,D,swap>(src, dst, byte_count); break;
237 case 8: do_conjoint_swap<uint64_t,D,swap>(src, dst, byte_count); break;
238 default: guarantee(false, "do_conjoint_swap: Invalid elem_size %zu\n", elem_size);
239 }
240 }
241 };
242
243 void Copy::conjoint_copy(const void* src, void* dst, size_t byte_count, size_t elem_size) {
244 CopySwap::conjoint_swap_if_needed<false>(src, dst, byte_count, elem_size);
245 }
246
247 void Copy::conjoint_swap(const void* src, void* dst, size_t byte_count, size_t elem_size) {
248 CopySwap::conjoint_swap_if_needed<true>(src, dst, byte_count, elem_size);
249 }
250
251 // Fill bytes; larger units are filled atomically if everything is aligned.
252 void Copy::fill_to_memory_atomic(void* to, size_t size, jubyte value) {
253 address dst = (address)to;
254 uintptr_t bits = (uintptr_t)to | (uintptr_t)size;
255 if (bits % sizeof(jlong) == 0) {
256 jlong fill = (julong)((jubyte)value); // zero-extend
257 if (fill != 0) {
258 fill += fill << 8;
259 fill += fill << 16;
260 fill += fill << 32;
261 }
262 // Copy::fill_to_jlongs_atomic((jlong*) dst, size / sizeof(jlong));
263 for (uintptr_t off = 0; off < size; off += sizeof(jlong)) {
264 *(jlong*)(dst + off) = fill;
265 }
266 } else if (bits % sizeof(jint) == 0) {
267 jint fill = (juint)((jubyte)value); // zero-extend
268 if (fill != 0) {
269 fill += fill << 8;
270 fill += fill << 16;
271 }
272 // Copy::fill_to_jints_atomic((jint*) dst, size / sizeof(jint));
273 for (uintptr_t off = 0; off < size; off += sizeof(jint)) {
274 *(jint*)(dst + off) = fill;
275 }
276 } else if (bits % sizeof(jshort) == 0) {
277 jshort fill = (jushort)((jubyte)value); // zero-extend
278 fill += (jshort)(fill << 8);
279 // Copy::fill_to_jshorts_atomic((jshort*) dst, size / sizeof(jshort));
280 for (uintptr_t off = 0; off < size; off += sizeof(jshort)) {
281 *(jshort*)(dst + off) = fill;
282 }
283 } else {
284 // Not aligned, so no need to be atomic.
285 #ifdef MUSL_LIBC
286 // This code is used by Unsafe and may hit the next page after truncation
287 // of mapped memory. Therefore, we use volatile to prevent compilers from
288 // replacing the loop by memset which may not trigger SIGBUS as needed
289 // (observed on Alpine Linux x86_64)
290 jbyte fill = value;
291 for (uintptr_t off = 0; off < size; off += sizeof(jbyte)) {
292 *(volatile jbyte*)(dst + off) = fill;
293 }
294 #else
295 Copy::fill_to_bytes(dst, size, value);
296 #endif
297 }
298 }