1 /*
2 * Copyright (c) 2006, 2024, 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 "precompiled.hpp"
26 #include "utilities/copy.hpp"
27 #include "runtime/sharedRuntime.hpp"
28 #include "utilities/align.hpp"
29 #include "utilities/byteswap.hpp"
30 #include "utilities/copy.hpp"
31
32
33 // Copy bytes; larger units are filled atomically if everything is aligned.
34 void Copy::conjoint_memory_atomic(const void* from, void* to, size_t size) {
35 uintptr_t bits = (uintptr_t) from | (uintptr_t) to | (uintptr_t) size;
36
37 // (Note: We could improve performance by ignoring the low bits of size,
38 // and putting a short cleanup loop after each bulk copy loop.
39 // There are plenty of other ways to make this faster also,
40 // and it's a slippery slope. For now, let's keep this code simple
41 // since the simplicity helps clarify the atomicity semantics of
42 // this operation. There are also CPU-specific assembly versions
43 // which may or may not want to include such optimizations.)
44
45 if (bits % sizeof(jlong) == 0) {
46 Copy::conjoint_jlongs_atomic((const jlong*) from, (jlong*) to, size / sizeof(jlong));
47 } else if (bits % sizeof(jint) == 0) {
48 Copy::conjoint_jints_atomic((const jint*) from, (jint*) to, size / sizeof(jint));
49 } else if (bits % sizeof(jshort) == 0) {
50 Copy::conjoint_jshorts_atomic((const jshort*) from, (jshort*) to, size / sizeof(jshort));
51 } else {
52 // Not aligned, so no need to be atomic.
53 Copy::conjoint_jbytes((const void*) from, (void*) to, size);
54 }
55 }
56
57 class CopySwap : AllStatic {
58 public:
59 /**
60 * Copy and optionally byte swap elements
61 *
62 * <swap> - true if elements should be byte swapped
63 *
64 * @param src address of source
65 * @param dst address of destination
66 * @param byte_count number of bytes to copy
67 * @param elem_size size of the elements to copy-swap
68 */
69 template<bool swap>
70 static void conjoint_swap_if_needed(const void* src, void* dst, size_t byte_count, size_t elem_size) {
71 assert(src != nullptr, "address must not be null");
72 assert(dst != nullptr, "address must not be null");
73 assert(elem_size == 2 || elem_size == 4 || elem_size == 8,
74 "incorrect element size: " SIZE_FORMAT, elem_size);
75 assert(is_aligned(byte_count, elem_size),
76 "byte_count " SIZE_FORMAT " must be multiple of element size " SIZE_FORMAT, byte_count, elem_size);
77
78 address src_end = (address)src + byte_count;
79
80 if (dst <= src || dst >= src_end) {
81 do_conjoint_swap<RIGHT,swap>(src, dst, byte_count, elem_size);
82 } else {
83 do_conjoint_swap<LEFT,swap>(src, dst, byte_count, elem_size);
84 }
85 }
86
87 private:
88 enum CopyDirection {
89 RIGHT, // lower -> higher address
90 LEFT // higher -> lower address
91 };
92
93 /**
94 * Copy and byte swap elements
95 *
96 * <T> - type of element to copy
97 * <D> - copy direction
98 * <is_src_aligned> - true if src argument is aligned to element size
99 * <is_dst_aligned> - true if dst argument is aligned to element size
100 *
101 * @param src address of source
102 * @param dst address of destination
103 * @param byte_count number of bytes to copy
104 */
105 template <typename T, CopyDirection D, bool swap, bool is_src_aligned, bool is_dst_aligned>
106 static void do_conjoint_swap(const void* src, void* dst, size_t byte_count) {
107 const char* cur_src;
108 char* cur_dst;
109
110 switch (D) {
111 case RIGHT:
112 cur_src = (const char*)src;
113 cur_dst = (char*)dst;
114 break;
115 case LEFT:
116 cur_src = (const char*)src + byte_count - sizeof(T);
117 cur_dst = (char*)dst + byte_count - sizeof(T);
118 break;
119 }
120
121 for (size_t i = 0; i < byte_count / sizeof(T); i++) {
122 T tmp;
123
124 if (is_src_aligned) {
125 tmp = *(T*)cur_src;
126 } else {
127 memcpy(&tmp, cur_src, sizeof(T));
128 }
129
130 if (swap) {
131 tmp = byteswap(tmp);
132 }
133
134 if (is_dst_aligned) {
135 *(T*)cur_dst = tmp;
136 } else {
137 memcpy(cur_dst, &tmp, sizeof(T));
138 }
139
140 switch (D) {
141 case RIGHT:
142 cur_src += sizeof(T);
143 cur_dst += sizeof(T);
144 break;
145 case LEFT:
146 cur_src -= sizeof(T);
147 cur_dst -= sizeof(T);
148 break;
149 }
150 }
151 }
152
153 /**
154 * Copy and byte swap elements
155 *
156 * <T> - type of element to copy
157 * <D> - copy direction
158 * <swap> - true if elements should be byte swapped
159 *
160 * @param src address of source
161 * @param dst address of destination
162 * @param byte_count number of bytes to copy
163 */
164 template <typename T, CopyDirection direction, bool swap>
165 static void do_conjoint_swap(const void* src, void* dst, size_t byte_count) {
166 if (is_aligned(src, sizeof(T))) {
167 if (is_aligned(dst, sizeof(T))) {
168 do_conjoint_swap<T,direction,swap,true,true>(src, dst, byte_count);
169 } else {
170 do_conjoint_swap<T,direction,swap,true,false>(src, dst, byte_count);
171 }
172 } else {
173 if (is_aligned(dst, sizeof(T))) {
174 do_conjoint_swap<T,direction,swap,false,true>(src, dst, byte_count);
175 } else {
176 do_conjoint_swap<T,direction,swap,false,false>(src, dst, byte_count);
177 }
178 }
179 }
180
181
182 /**
183 * Copy and byte swap elements
184 *
185 * <D> - copy direction
186 * <swap> - true if elements should be byte swapped
187 *
188 * @param src address of source
189 * @param dst address of destination
190 * @param byte_count number of bytes to copy
191 * @param elem_size size of the elements to copy-swap
192 */
193 template <CopyDirection D, bool swap>
194 static void do_conjoint_swap(const void* src, void* dst, size_t byte_count, size_t elem_size) {
195 switch (elem_size) {
196 case 2: do_conjoint_swap<uint16_t,D,swap>(src, dst, byte_count); break;
197 case 4: do_conjoint_swap<uint32_t,D,swap>(src, dst, byte_count); break;
198 case 8: do_conjoint_swap<uint64_t,D,swap>(src, dst, byte_count); break;
199 default: guarantee(false, "do_conjoint_swap: Invalid elem_size " SIZE_FORMAT "\n", elem_size);
200 }
201 }
202 };
203
204 void Copy::conjoint_copy(const void* src, void* dst, size_t byte_count, size_t elem_size) {
205 CopySwap::conjoint_swap_if_needed<false>(src, dst, byte_count, elem_size);
206 }
207
208 void Copy::conjoint_swap(const void* src, void* dst, size_t byte_count, size_t elem_size) {
209 CopySwap::conjoint_swap_if_needed<true>(src, dst, byte_count, elem_size);
210 }
211
212 // Fill bytes; larger units are filled atomically if everything is aligned.
213 void Copy::fill_to_memory_atomic(void* to, size_t size, jubyte value) {
214 address dst = (address)to;
215 uintptr_t bits = (uintptr_t)to | (uintptr_t)size;
216 if (bits % sizeof(jlong) == 0) {
217 jlong fill = (julong)((jubyte)value); // zero-extend
218 if (fill != 0) {
219 fill += fill << 8;
220 fill += fill << 16;
221 fill += fill << 32;
222 }
223 // Copy::fill_to_jlongs_atomic((jlong*) dst, size / sizeof(jlong));
224 for (uintptr_t off = 0; off < size; off += sizeof(jlong)) {
225 *(jlong*)(dst + off) = fill;
226 }
227 } else if (bits % sizeof(jint) == 0) {
228 jint fill = (juint)((jubyte)value); // zero-extend
229 if (fill != 0) {
230 fill += fill << 8;
231 fill += fill << 16;
232 }
233 // Copy::fill_to_jints_atomic((jint*) dst, size / sizeof(jint));
234 for (uintptr_t off = 0; off < size; off += sizeof(jint)) {
235 *(jint*)(dst + off) = fill;
236 }
237 } else if (bits % sizeof(jshort) == 0) {
238 jshort fill = (jushort)((jubyte)value); // zero-extend
239 fill += (jshort)(fill << 8);
240 // Copy::fill_to_jshorts_atomic((jshort*) dst, size / sizeof(jshort));
241 for (uintptr_t off = 0; off < size; off += sizeof(jshort)) {
242 *(jshort*)(dst + off) = fill;
243 }
244 } else {
245 // Not aligned, so no need to be atomic.
246 #ifdef MUSL_LIBC
247 // This code is used by Unsafe and may hit the next page after truncation
248 // of mapped memory. Therefore, we use volatile to prevent compilers from
249 // replacing the loop by memset which may not trigger SIGBUS as needed
250 // (observed on Alpine Linux x86_64)
251 jbyte fill = value;
252 for (uintptr_t off = 0; off < size; off += sizeof(jbyte)) {
253 *(volatile jbyte*)(dst + off) = fill;
254 }
255 #else
256 Copy::fill_to_bytes(dst, size, value);
257 #endif
258 }
259 }