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 "runtime/sharedRuntime.hpp"
26 #include "utilities/align.hpp"
27 #include "utilities/byteswap.hpp"
28 #include "utilities/copy.hpp"
29
30
31 // Copy bytes; larger units are filled atomically if everything is aligned.
32 void Copy::conjoint_memory_atomic(const void* from, void* to, size_t size) {
33 uintptr_t bits = (uintptr_t) from | (uintptr_t) to | (uintptr_t) size;
34
35 // (Note: We could improve performance by ignoring the low bits of size,
36 // and putting a short cleanup loop after each bulk copy loop.
37 // There are plenty of other ways to make this faster also,
38 // and it's a slippery slope. For now, let's keep this code simple
39 // since the simplicity helps clarify the atomicity semantics of
40 // this operation. There are also CPU-specific assembly versions
41 // which may or may not want to include such optimizations.)
42
43 if (bits % sizeof(jlong) == 0) {
44 Copy::conjoint_jlongs_atomic((const jlong*) from, (jlong*) to, size / sizeof(jlong));
45 } else if (bits % sizeof(jint) == 0) {
46 Copy::conjoint_jints_atomic((const jint*) from, (jint*) to, size / sizeof(jint));
47 } else if (bits % sizeof(jshort) == 0) {
48 Copy::conjoint_jshorts_atomic((const jshort*) from, (jshort*) to, size / sizeof(jshort));
49 } else {
50 // Not aligned, so no need to be atomic.
51 Copy::conjoint_jbytes((const void*) from, (void*) to, size);
52 }
53 }
54
55 class CopySwap : AllStatic {
56 public:
57 /**
58 * Copy and optionally byte swap elements
59 *
60 * <swap> - true if elements should be byte swapped
61 *
62 * @param src address of source
63 * @param dst address of destination
64 * @param byte_count number of bytes to copy
65 * @param elem_size size of the elements to copy-swap
66 */
67 template<bool swap>
68 static void conjoint_swap_if_needed(const void* src, void* dst, size_t byte_count, size_t elem_size) {
69 assert(src != nullptr, "address must not be null");
70 assert(dst != nullptr, "address must not be null");
71 assert(elem_size == 2 || elem_size == 4 || elem_size == 8,
72 "incorrect element size: %zu", elem_size);
73 assert(is_aligned(byte_count, elem_size),
74 "byte_count %zu must be multiple of element size %zu", byte_count, elem_size);
75
76 address src_end = (address)src + byte_count;
77
78 if (dst <= src || dst >= src_end) {
79 do_conjoint_swap<RIGHT,swap>(src, dst, byte_count, elem_size);
80 } else {
81 do_conjoint_swap<LEFT,swap>(src, dst, byte_count, elem_size);
82 }
83 }
84
85 private:
86 enum CopyDirection {
87 RIGHT, // lower -> higher address
88 LEFT // higher -> lower address
89 };
90
91 /**
92 * Copy and byte swap elements
93 *
94 * <T> - type of element to copy
95 * <D> - copy direction
96 * <is_src_aligned> - true if src argument is aligned to element size
97 * <is_dst_aligned> - true if dst argument is aligned to element size
98 *
99 * @param src address of source
100 * @param dst address of destination
101 * @param byte_count number of bytes to copy
102 */
103 template <typename T, CopyDirection D, bool swap, bool is_src_aligned, bool is_dst_aligned>
104 static void do_conjoint_swap(const void* src, void* dst, size_t byte_count) {
105 const char* cur_src;
106 char* cur_dst;
107
108 switch (D) {
109 case RIGHT:
110 cur_src = (const char*)src;
111 cur_dst = (char*)dst;
112 break;
113 case LEFT:
114 cur_src = (const char*)src + byte_count - sizeof(T);
115 cur_dst = (char*)dst + byte_count - sizeof(T);
116 break;
117 }
118
119 for (size_t i = 0; i < byte_count / sizeof(T); i++) {
120 T tmp;
121
122 if (is_src_aligned) {
123 tmp = *(T*)cur_src;
124 } else {
125 memcpy(&tmp, cur_src, sizeof(T));
126 }
127
128 if (swap) {
129 tmp = byteswap(tmp);
130 }
131
132 if (is_dst_aligned) {
133 *(T*)cur_dst = tmp;
134 } else {
135 memcpy(cur_dst, &tmp, sizeof(T));
136 }
137
138 switch (D) {
139 case RIGHT:
140 cur_src += sizeof(T);
141 cur_dst += sizeof(T);
142 break;
143 case LEFT:
144 cur_src -= sizeof(T);
145 cur_dst -= sizeof(T);
146 break;
147 }
148 }
149 }
150
151 /**
152 * Copy and byte swap elements
153 *
154 * <T> - type of element to copy
155 * <D> - copy direction
156 * <swap> - true if elements should be byte swapped
157 *
158 * @param src address of source
159 * @param dst address of destination
160 * @param byte_count number of bytes to copy
161 */
162 template <typename T, CopyDirection direction, bool swap>
163 static void do_conjoint_swap(const void* src, void* dst, size_t byte_count) {
164 if (is_aligned(src, sizeof(T))) {
165 if (is_aligned(dst, sizeof(T))) {
166 do_conjoint_swap<T,direction,swap,true,true>(src, dst, byte_count);
167 } else {
168 do_conjoint_swap<T,direction,swap,true,false>(src, dst, byte_count);
169 }
170 } else {
171 if (is_aligned(dst, sizeof(T))) {
172 do_conjoint_swap<T,direction,swap,false,true>(src, dst, byte_count);
173 } else {
174 do_conjoint_swap<T,direction,swap,false,false>(src, dst, byte_count);
175 }
176 }
177 }
178
179
180 /**
181 * Copy and byte swap elements
182 *
183 * <D> - copy direction
184 * <swap> - true if elements should be byte swapped
185 *
186 * @param src address of source
187 * @param dst address of destination
188 * @param byte_count number of bytes to copy
189 * @param elem_size size of the elements to copy-swap
190 */
191 template <CopyDirection D, bool swap>
192 static void do_conjoint_swap(const void* src, void* dst, size_t byte_count, size_t elem_size) {
193 switch (elem_size) {
194 case 2: do_conjoint_swap<uint16_t,D,swap>(src, dst, byte_count); break;
195 case 4: do_conjoint_swap<uint32_t,D,swap>(src, dst, byte_count); break;
196 case 8: do_conjoint_swap<uint64_t,D,swap>(src, dst, byte_count); break;
197 default: guarantee(false, "do_conjoint_swap: Invalid elem_size %zu\n", elem_size);
198 }
199 }
200 };
201
202 void Copy::conjoint_copy(const void* src, void* dst, size_t byte_count, size_t elem_size) {
203 CopySwap::conjoint_swap_if_needed<false>(src, dst, byte_count, elem_size);
204 }
205
206 void Copy::conjoint_swap(const void* src, void* dst, size_t byte_count, size_t elem_size) {
207 CopySwap::conjoint_swap_if_needed<true>(src, dst, byte_count, elem_size);
208 }
209
210 // Fill bytes; larger units are filled atomically if everything is aligned.
211 void Copy::fill_to_memory_atomic(void* to, size_t size, jubyte value) {
212 address dst = (address)to;
213 uintptr_t bits = (uintptr_t)to | (uintptr_t)size;
214 if (bits % sizeof(jlong) == 0) {
215 jlong fill = (julong)((jubyte)value); // zero-extend
216 if (fill != 0) {
217 fill += fill << 8;
218 fill += fill << 16;
219 fill += fill << 32;
220 }
221 // Copy::fill_to_jlongs_atomic((jlong*) dst, size / sizeof(jlong));
222 for (uintptr_t off = 0; off < size; off += sizeof(jlong)) {
223 *(jlong*)(dst + off) = fill;
224 }
225 } else if (bits % sizeof(jint) == 0) {
226 jint fill = (juint)((jubyte)value); // zero-extend
227 if (fill != 0) {
228 fill += fill << 8;
229 fill += fill << 16;
230 }
231 // Copy::fill_to_jints_atomic((jint*) dst, size / sizeof(jint));
232 for (uintptr_t off = 0; off < size; off += sizeof(jint)) {
233 *(jint*)(dst + off) = fill;
234 }
235 } else if (bits % sizeof(jshort) == 0) {
236 jshort fill = (jushort)((jubyte)value); // zero-extend
237 fill += (jshort)(fill << 8);
238 // Copy::fill_to_jshorts_atomic((jshort*) dst, size / sizeof(jshort));
239 for (uintptr_t off = 0; off < size; off += sizeof(jshort)) {
240 *(jshort*)(dst + off) = fill;
241 }
242 } else {
243 // Not aligned, so no need to be atomic.
244 #ifdef MUSL_LIBC
245 // This code is used by Unsafe and may hit the next page after truncation
246 // of mapped memory. Therefore, we use volatile to prevent compilers from
247 // replacing the loop by memset which may not trigger SIGBUS as needed
248 // (observed on Alpine Linux x86_64)
249 jbyte fill = value;
250 for (uintptr_t off = 0; off < size; off += sizeof(jbyte)) {
251 *(volatile jbyte*)(dst + off) = fill;
252 }
253 #else
254 Copy::fill_to_bytes(dst, size, value);
255 #endif
256 }
257 }