Commit | Line | Data |
---|---|---|
b76e5200 MD |
1 | /* |
2 | * rseq-x86.h | |
3 | * | |
4 | * (C) Copyright 2016 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com> | |
5 | * | |
6 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
7 | * of this software and associated documentation files (the "Software"), to deal | |
8 | * in the Software without restriction, including without limitation the rights | |
9 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
10 | * copies of the Software, and to permit persons to whom the Software is | |
11 | * furnished to do so, subject to the following conditions: | |
12 | * | |
13 | * The above copyright notice and this permission notice shall be included in | |
14 | * all copies or substantial portions of the Software. | |
15 | * | |
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE | |
19 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE | |
22 | * SOFTWARE. | |
23 | */ | |
24 | ||
25 | #ifdef __x86_64__ | |
26 | ||
b76e5200 MD |
27 | #define has_fast_acquire_release() 1 |
28 | #define has_single_copy_load_64() 1 | |
29 | ||
30 | /* | |
31 | * The __rseq_table section can be used by debuggers to better handle | |
32 | * single-stepping through the restartable critical sections. | |
33 | */ | |
34 | #define RSEQ_FINISH_ASM(_target_final, _to_write_final, _start_value, \ | |
35 | _failure, _spec_store, _spec_input, \ | |
36 | _final_store, _final_input, _extra_clobber, \ | |
37 | _setup, _teardown, _scratch) \ | |
38 | do { \ | |
39 | _scratch \ | |
40 | __asm__ __volatile__ goto ( \ | |
41 | ".pushsection __rseq_table, \"aw\"\n\t" \ | |
42 | ".balign 32\n\t" \ | |
43 | "3:\n\t" \ | |
44 | ".quad 1f, 2f, 4f, 0x0\n\t" \ | |
45 | ".popsection\n\t" \ | |
46 | "1:\n\t" \ | |
47 | _setup \ | |
48 | RSEQ_INJECT_ASM(1) \ | |
49 | "movq $3b, %[rseq_cs]\n\t" \ | |
50 | RSEQ_INJECT_ASM(2) \ | |
51 | "cmpl %[start_event_counter], %[current_event_counter]\n\t" \ | |
52 | "jnz 4f\n\t" \ | |
53 | RSEQ_INJECT_ASM(3) \ | |
54 | _spec_store \ | |
55 | _final_store \ | |
56 | "2:\n\t" \ | |
57 | RSEQ_INJECT_ASM(5) \ | |
58 | "movq $0, %[rseq_cs]\n\t" \ | |
59 | _teardown \ | |
60 | ".pushsection __rseq_failure, \"a\"\n\t" \ | |
61 | "4:\n\t" \ | |
62 | "movq $0, %[rseq_cs]\n\t" \ | |
63 | _teardown \ | |
64 | "jmp %l[failure]\n\t" \ | |
65 | ".popsection\n\t" \ | |
66 | : /* no outputs */ \ | |
67 | : [start_event_counter]"r"((_start_value).event_counter), \ | |
68 | [current_event_counter]"m"((_start_value).rseqp->u.e.event_counter), \ | |
69 | [rseq_cs]"m"((_start_value).rseqp->rseq_cs) \ | |
70 | _spec_input \ | |
71 | _final_input \ | |
72 | RSEQ_INJECT_INPUT \ | |
73 | : "memory", "cc" \ | |
74 | _extra_clobber \ | |
75 | RSEQ_INJECT_CLOBBER \ | |
76 | : _failure \ | |
77 | ); \ | |
78 | } while (0) | |
79 | ||
80 | #define RSEQ_FINISH_FINAL_STORE_ASM() \ | |
81 | "movq %[to_write_final], %[target_final]\n\t" | |
82 | ||
83 | /* x86-64 is TSO */ | |
84 | #define RSEQ_FINISH_FINAL_STORE_RELEASE_ASM() \ | |
85 | RSEQ_FINISH_FINAL_STORE_ASM() | |
86 | ||
87 | #define RSEQ_FINISH_FINAL_STORE_INPUT(_target_final, _to_write_final) \ | |
88 | , [to_write_final]"r"(_to_write_final), \ | |
89 | [target_final]"m"(*(_target_final)) | |
90 | ||
91 | #define RSEQ_FINISH_SPECULATIVE_STORE_ASM() \ | |
92 | "movq %[to_write_spec], %[target_spec]\n\t" \ | |
93 | RSEQ_INJECT_ASM(4) | |
94 | ||
95 | #define RSEQ_FINISH_SPECULATIVE_STORE_INPUT(_target_spec, _to_write_spec) \ | |
96 | , [to_write_spec]"r"(_to_write_spec), \ | |
97 | [target_spec]"m"(*(_target_spec)) | |
98 | ||
99 | /* TODO: implement a faster memcpy. */ | |
100 | #define RSEQ_FINISH_MEMCPY_STORE_ASM() \ | |
101 | "test %[len_memcpy], %[len_memcpy]\n\t" \ | |
102 | "jz 333f\n\t" \ | |
103 | "222:\n\t" \ | |
104 | "movb (%[to_write_memcpy]), %%al\n\t" \ | |
105 | "movb %%al, (%[target_memcpy])\n\t" \ | |
106 | "inc %[to_write_memcpy]\n\t" \ | |
107 | "inc %[target_memcpy]\n\t" \ | |
108 | "dec %[len_memcpy]\n\t" \ | |
109 | "jnz 222b\n\t" \ | |
110 | "333:\n\t" \ | |
111 | RSEQ_INJECT_ASM(4) | |
112 | ||
113 | #define RSEQ_FINISH_MEMCPY_STORE_INPUT(_target_memcpy, _to_write_memcpy, _len_memcpy) \ | |
114 | , [to_write_memcpy]"r"(_to_write_memcpy), \ | |
115 | [target_memcpy]"r"(_target_memcpy), \ | |
116 | [len_memcpy]"r"(_len_memcpy), \ | |
117 | [rseq_scratch0]"m"(rseq_scratch[0]), \ | |
118 | [rseq_scratch1]"m"(rseq_scratch[1]), \ | |
119 | [rseq_scratch2]"m"(rseq_scratch[2]) | |
120 | ||
121 | #define RSEQ_FINISH_MEMCPY_CLOBBER() \ | |
122 | , "rax" | |
123 | ||
124 | #define RSEQ_FINISH_MEMCPY_SCRATCH() \ | |
125 | uint64_t rseq_scratch[3]; | |
126 | ||
127 | /* | |
128 | * We need to save and restore those input registers so they can be | |
129 | * modified within the assembly. | |
130 | */ | |
131 | #define RSEQ_FINISH_MEMCPY_SETUP() \ | |
132 | "movq %[to_write_memcpy], %[rseq_scratch0]\n\t" \ | |
133 | "movq %[target_memcpy], %[rseq_scratch1]\n\t" \ | |
134 | "movq %[len_memcpy], %[rseq_scratch2]\n\t" | |
135 | ||
136 | #define RSEQ_FINISH_MEMCPY_TEARDOWN() \ | |
137 | "movq %[rseq_scratch2], %[len_memcpy]\n\t" \ | |
138 | "movq %[rseq_scratch1], %[target_memcpy]\n\t" \ | |
139 | "movq %[rseq_scratch0], %[to_write_memcpy]\n\t" | |
140 | ||
141 | #elif __i386__ | |
142 | ||
b76e5200 MD |
143 | #define has_fast_acquire_release() 0 |
144 | #define has_single_copy_load_64() 0 | |
145 | ||
146 | /* | |
147 | * Use eax as scratch register and take memory operands as input to | |
148 | * lessen register pressure. Especially needed when compiling | |
149 | * do_rseq_memcpy() in O0. | |
150 | */ | |
151 | #define RSEQ_FINISH_ASM(_target_final, _to_write_final, _start_value, \ | |
152 | _failure, _spec_store, _spec_input, \ | |
153 | _final_store, _final_input, _extra_clobber, \ | |
154 | _setup, _teardown, _scratch) \ | |
155 | do { \ | |
156 | _scratch \ | |
157 | __asm__ __volatile__ goto ( \ | |
158 | ".pushsection __rseq_table, \"aw\"\n\t" \ | |
159 | ".balign 32\n\t" \ | |
160 | "3:\n\t" \ | |
161 | ".long 1f, 0x0, 2f, 0x0, 4f, 0x0, 0x0, 0x0\n\t" \ | |
162 | ".popsection\n\t" \ | |
163 | "1:\n\t" \ | |
164 | _setup \ | |
165 | RSEQ_INJECT_ASM(1) \ | |
166 | "movl $3b, %[rseq_cs]\n\t" \ | |
167 | RSEQ_INJECT_ASM(2) \ | |
168 | "movl %[start_event_counter], %%eax\n\t" \ | |
169 | "cmpl %%eax, %[current_event_counter]\n\t" \ | |
170 | "jnz 4f\n\t" \ | |
171 | RSEQ_INJECT_ASM(3) \ | |
172 | _spec_store \ | |
173 | _final_store \ | |
174 | "2:\n\t" \ | |
175 | RSEQ_INJECT_ASM(5) \ | |
176 | "movl $0, %[rseq_cs]\n\t" \ | |
177 | _teardown \ | |
178 | ".pushsection __rseq_failure, \"a\"\n\t" \ | |
179 | "4:\n\t" \ | |
180 | "movl $0, %[rseq_cs]\n\t" \ | |
181 | _teardown \ | |
182 | "jmp %l[failure]\n\t" \ | |
183 | ".popsection\n\t" \ | |
184 | : /* no outputs */ \ | |
185 | : [start_event_counter]"m"((_start_value).event_counter), \ | |
186 | [current_event_counter]"m"((_start_value).rseqp->u.e.event_counter), \ | |
187 | [rseq_cs]"m"((_start_value).rseqp->rseq_cs) \ | |
188 | _spec_input \ | |
189 | _final_input \ | |
190 | RSEQ_INJECT_INPUT \ | |
191 | : "memory", "cc", "eax" \ | |
192 | _extra_clobber \ | |
193 | RSEQ_INJECT_CLOBBER \ | |
194 | : _failure \ | |
195 | ); \ | |
196 | } while (0) | |
197 | ||
198 | #define RSEQ_FINISH_FINAL_STORE_ASM() \ | |
199 | "movl %[to_write_final], %%eax\n\t" \ | |
200 | "movl %%eax, %[target_final]\n\t" | |
201 | ||
202 | #define RSEQ_FINISH_FINAL_STORE_RELEASE_ASM() \ | |
203 | "lock; addl $0,0(%%esp)\n\t" \ | |
204 | RSEQ_FINISH_FINAL_STORE_ASM() | |
205 | ||
206 | #define RSEQ_FINISH_FINAL_STORE_INPUT(_target_final, _to_write_final) \ | |
207 | , [to_write_final]"m"(_to_write_final), \ | |
208 | [target_final]"m"(*(_target_final)) | |
209 | ||
210 | #define RSEQ_FINISH_SPECULATIVE_STORE_ASM() \ | |
211 | "movl %[to_write_spec], %%eax\n\t" \ | |
212 | "movl %%eax, %[target_spec]\n\t" \ | |
213 | RSEQ_INJECT_ASM(4) | |
214 | ||
215 | #define RSEQ_FINISH_SPECULATIVE_STORE_INPUT(_target_spec, _to_write_spec) \ | |
216 | , [to_write_spec]"m"(_to_write_spec), \ | |
217 | [target_spec]"m"(*(_target_spec)) | |
218 | ||
219 | /* TODO: implement a faster memcpy. */ | |
220 | #define RSEQ_FINISH_MEMCPY_STORE_ASM() \ | |
221 | "movl %[len_memcpy], %%eax\n\t" \ | |
222 | "test %%eax, %%eax\n\t" \ | |
223 | "jz 333f\n\t" \ | |
224 | "222:\n\t" \ | |
225 | "movb (%[to_write_memcpy]), %%al\n\t" \ | |
226 | "movb %%al, (%[target_memcpy])\n\t" \ | |
227 | "inc %[to_write_memcpy]\n\t" \ | |
228 | "inc %[target_memcpy]\n\t" \ | |
229 | "decl %[rseq_scratch2]\n\t" \ | |
230 | "jnz 222b\n\t" \ | |
231 | "333:\n\t" \ | |
232 | RSEQ_INJECT_ASM(4) | |
233 | ||
234 | #define RSEQ_FINISH_MEMCPY_STORE_INPUT(_target_memcpy, _to_write_memcpy, _len_memcpy) \ | |
235 | , [to_write_memcpy]"r"(_to_write_memcpy), \ | |
236 | [target_memcpy]"r"(_target_memcpy), \ | |
237 | [len_memcpy]"m"(_len_memcpy), \ | |
238 | [rseq_scratch0]"m"(rseq_scratch[0]), \ | |
239 | [rseq_scratch1]"m"(rseq_scratch[1]), \ | |
240 | [rseq_scratch2]"m"(rseq_scratch[2]) | |
241 | ||
242 | #define RSEQ_FINISH_MEMCPY_CLOBBER() | |
243 | ||
244 | #define RSEQ_FINISH_MEMCPY_SCRATCH() \ | |
245 | uint32_t rseq_scratch[3]; | |
246 | ||
247 | /* | |
248 | * We need to save and restore those input registers so they can be | |
249 | * modified within the assembly. | |
250 | */ | |
251 | #define RSEQ_FINISH_MEMCPY_SETUP() \ | |
252 | "movl %[to_write_memcpy], %[rseq_scratch0]\n\t" \ | |
253 | "movl %[target_memcpy], %[rseq_scratch1]\n\t" \ | |
254 | "movl %[len_memcpy], %%eax\n\t" \ | |
255 | "movl %%eax, %[rseq_scratch2]\n\t" | |
256 | ||
257 | #define RSEQ_FINISH_MEMCPY_TEARDOWN() \ | |
258 | "movl %[rseq_scratch1], %[target_memcpy]\n\t" \ | |
259 | "movl %[rseq_scratch0], %[to_write_memcpy]\n\t" | |
260 | ||
261 | #endif |