Restartable sequences: don't clear rseq_cs after c.s.
[deliverable/linux.git] / tools / testing / selftests / rseq / rseq-x86.h
CommitLineData
b54c5158
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
27#define smp_mb() __asm__ __volatile__ ("mfence" : : : "memory")
28#define smp_rmb() barrier()
29#define smp_wmb() barrier()
30
31#define smp_load_acquire(p) \
32__extension__ ({ \
33 __typeof(*p) ____p1 = READ_ONCE(*p); \
34 barrier(); \
35 ____p1; \
36})
37
38#define smp_acquire__after_ctrl_dep() smp_rmb()
39
40#define smp_store_release(p, v) \
41do { \
42 barrier(); \
43 WRITE_ONCE(*p, v); \
44} while (0)
45
46#define has_fast_acquire_release() 1
47#define has_single_copy_load_64() 1
48
49/*
50 * The __rseq_table section can be used by debuggers to better handle
51 * single-stepping through the restartable critical sections.
52 */
53#define RSEQ_FINISH_ASM(_target_final, _to_write_final, _start_value, \
54 _failure, _spec_store, _spec_input, \
55 _final_store, _final_input, _extra_clobber, \
56 _setup, _teardown, _scratch) \
57do { \
58 _scratch \
59 __asm__ __volatile__ goto ( \
60 ".pushsection __rseq_table, \"aw\"\n\t" \
61 ".balign 32\n\t" \
62 "3:\n\t" \
63 ".quad 1f, 2f, 4f, 0x0\n\t" \
64 ".popsection\n\t" \
65 "1:\n\t" \
66 _setup \
67 RSEQ_INJECT_ASM(1) \
f88ff702
MD
68 "leaq 3b(%%rip), %%rax\n\t" \
69 "movq %%rax, %[rseq_cs]\n\t" \
b54c5158
MD
70 RSEQ_INJECT_ASM(2) \
71 "cmpl %[start_event_counter], %[current_event_counter]\n\t" \
72 "jnz 4f\n\t" \
73 RSEQ_INJECT_ASM(3) \
74 _spec_store \
75 _final_store \
76 "2:\n\t" \
77 RSEQ_INJECT_ASM(5) \
b54c5158
MD
78 _teardown \
79 ".pushsection __rseq_failure, \"a\"\n\t" \
80 "4:\n\t" \
b54c5158
MD
81 _teardown \
82 "jmp %l[failure]\n\t" \
83 ".popsection\n\t" \
84 : /* gcc asm goto does not allow outputs */ \
85 : [start_event_counter]"r"((_start_value).event_counter), \
86 [current_event_counter]"m"((_start_value).rseqp->u.e.event_counter), \
87 [rseq_cs]"m"((_start_value).rseqp->rseq_cs) \
88 _spec_input \
89 _final_input \
90 RSEQ_INJECT_INPUT \
f88ff702 91 : "memory", "cc", "rax" \
b54c5158
MD
92 _extra_clobber \
93 RSEQ_INJECT_CLOBBER \
94 : _failure \
95 ); \
96} while (0)
97
98#define RSEQ_FINISH_FINAL_STORE_ASM() \
99 "movq %[to_write_final], %[target_final]\n\t"
100
101/* x86-64 is TSO */
102#define RSEQ_FINISH_FINAL_STORE_RELEASE_ASM() \
103 RSEQ_FINISH_FINAL_STORE_ASM()
104
105#define RSEQ_FINISH_FINAL_STORE_INPUT(_target_final, _to_write_final) \
106 , [to_write_final]"r"(_to_write_final), \
107 [target_final]"m"(*(_target_final))
108
109#define RSEQ_FINISH_SPECULATIVE_STORE_ASM() \
110 "movq %[to_write_spec], %[target_spec]\n\t" \
111 RSEQ_INJECT_ASM(4)
112
113#define RSEQ_FINISH_SPECULATIVE_STORE_INPUT(_target_spec, _to_write_spec) \
114 , [to_write_spec]"r"(_to_write_spec), \
115 [target_spec]"m"(*(_target_spec))
116
117/* TODO: implement a faster memcpy. */
118#define RSEQ_FINISH_MEMCPY_STORE_ASM() \
119 "test %[len_memcpy], %[len_memcpy]\n\t" \
120 "jz 333f\n\t" \
121 "222:\n\t" \
122 "movb (%[to_write_memcpy]), %%al\n\t" \
123 "movb %%al, (%[target_memcpy])\n\t" \
124 "inc %[to_write_memcpy]\n\t" \
125 "inc %[target_memcpy]\n\t" \
126 "dec %[len_memcpy]\n\t" \
127 "jnz 222b\n\t" \
128 "333:\n\t" \
129 RSEQ_INJECT_ASM(4)
130
131#define RSEQ_FINISH_MEMCPY_STORE_INPUT(_target_memcpy, _to_write_memcpy, _len_memcpy) \
132 , [to_write_memcpy]"r"(_to_write_memcpy), \
133 [target_memcpy]"r"(_target_memcpy), \
134 [len_memcpy]"r"(_len_memcpy), \
135 [rseq_scratch0]"m"(rseq_scratch[0]), \
136 [rseq_scratch1]"m"(rseq_scratch[1]), \
137 [rseq_scratch2]"m"(rseq_scratch[2])
138
139#define RSEQ_FINISH_MEMCPY_CLOBBER() \
140 , "rax"
141
142#define RSEQ_FINISH_MEMCPY_SCRATCH() \
143 uint64_t rseq_scratch[3];
144
145/*
146 * We need to save and restore those input registers so they can be
147 * modified within the assembly.
148 */
149#define RSEQ_FINISH_MEMCPY_SETUP() \
150 "movq %[to_write_memcpy], %[rseq_scratch0]\n\t" \
151 "movq %[target_memcpy], %[rseq_scratch1]\n\t" \
152 "movq %[len_memcpy], %[rseq_scratch2]\n\t"
153
154#define RSEQ_FINISH_MEMCPY_TEARDOWN() \
155 "movq %[rseq_scratch2], %[len_memcpy]\n\t" \
156 "movq %[rseq_scratch1], %[target_memcpy]\n\t" \
157 "movq %[rseq_scratch0], %[to_write_memcpy]\n\t"
158
159#elif __i386__
160
161/*
162 * Support older 32-bit architectures that do not implement fence
163 * instructions.
164 */
165#define smp_mb() \
166 __asm__ __volatile__ ("lock; addl $0,0(%%esp)" : : : "memory")
167#define smp_rmb() \
168 __asm__ __volatile__ ("lock; addl $0,0(%%esp)" : : : "memory")
169#define smp_wmb() \
170 __asm__ __volatile__ ("lock; addl $0,0(%%esp)" : : : "memory")
171
172#define smp_load_acquire(p) \
173__extension__ ({ \
174 __typeof(*p) ____p1 = READ_ONCE(*p); \
175 smp_mb(); \
176 ____p1; \
177})
178
179#define smp_acquire__after_ctrl_dep() smp_rmb()
180
181#define smp_store_release(p, v) \
182do { \
183 smp_mb(); \
184 WRITE_ONCE(*p, v); \
185} while (0)
186
187#define has_fast_acquire_release() 0
188#define has_single_copy_load_64() 0
189
190/*
191 * Use eax as scratch register and take memory operands as input to
192 * lessen register pressure. Especially needed when compiling
193 * do_rseq_memcpy() in O0.
194 */
195#define RSEQ_FINISH_ASM(_target_final, _to_write_final, _start_value, \
196 _failure, _spec_store, _spec_input, \
197 _final_store, _final_input, _extra_clobber, \
198 _setup, _teardown, _scratch) \
199do { \
200 _scratch \
201 __asm__ __volatile__ goto ( \
202 ".pushsection __rseq_table, \"aw\"\n\t" \
203 ".balign 32\n\t" \
204 "3:\n\t" \
205 ".long 1f, 0x0, 2f, 0x0, 4f, 0x0, 0x0, 0x0\n\t" \
206 ".popsection\n\t" \
207 "1:\n\t" \
208 _setup \
209 RSEQ_INJECT_ASM(1) \
210 "movl $3b, %[rseq_cs]\n\t" \
211 RSEQ_INJECT_ASM(2) \
212 "movl %[start_event_counter], %%eax\n\t" \
213 "cmpl %%eax, %[current_event_counter]\n\t" \
214 "jnz 4f\n\t" \
215 RSEQ_INJECT_ASM(3) \
216 _spec_store \
217 _final_store \
218 "2:\n\t" \
219 RSEQ_INJECT_ASM(5) \
b54c5158
MD
220 _teardown \
221 ".pushsection __rseq_failure, \"a\"\n\t" \
222 "4:\n\t" \
b54c5158
MD
223 _teardown \
224 "jmp %l[failure]\n\t" \
225 ".popsection\n\t" \
226 : /* gcc asm goto does not allow outputs */ \
227 : [start_event_counter]"m"((_start_value).event_counter), \
228 [current_event_counter]"m"((_start_value).rseqp->u.e.event_counter), \
229 [rseq_cs]"m"((_start_value).rseqp->rseq_cs) \
230 _spec_input \
231 _final_input \
232 RSEQ_INJECT_INPUT \
233 : "memory", "cc", "eax" \
234 _extra_clobber \
235 RSEQ_INJECT_CLOBBER \
236 : _failure \
237 ); \
238} while (0)
239
240#define RSEQ_FINISH_FINAL_STORE_ASM() \
241 "movl %[to_write_final], %%eax\n\t" \
242 "movl %%eax, %[target_final]\n\t"
243
244#define RSEQ_FINISH_FINAL_STORE_RELEASE_ASM() \
245 "lock; addl $0,0(%%esp)\n\t" \
246 RSEQ_FINISH_FINAL_STORE_ASM()
247
248#define RSEQ_FINISH_FINAL_STORE_INPUT(_target_final, _to_write_final) \
249 , [to_write_final]"m"(_to_write_final), \
250 [target_final]"m"(*(_target_final))
251
252#define RSEQ_FINISH_SPECULATIVE_STORE_ASM() \
253 "movl %[to_write_spec], %%eax\n\t" \
254 "movl %%eax, %[target_spec]\n\t" \
255 RSEQ_INJECT_ASM(4)
256
257#define RSEQ_FINISH_SPECULATIVE_STORE_INPUT(_target_spec, _to_write_spec) \
258 , [to_write_spec]"m"(_to_write_spec), \
259 [target_spec]"m"(*(_target_spec))
260
261/* TODO: implement a faster memcpy. */
262#define RSEQ_FINISH_MEMCPY_STORE_ASM() \
263 "movl %[len_memcpy], %%eax\n\t" \
264 "test %%eax, %%eax\n\t" \
265 "jz 333f\n\t" \
266 "222:\n\t" \
267 "movb (%[to_write_memcpy]), %%al\n\t" \
268 "movb %%al, (%[target_memcpy])\n\t" \
269 "inc %[to_write_memcpy]\n\t" \
270 "inc %[target_memcpy]\n\t" \
271 "decl %[rseq_scratch2]\n\t" \
272 "jnz 222b\n\t" \
273 "333:\n\t" \
274 RSEQ_INJECT_ASM(4)
275
276#define RSEQ_FINISH_MEMCPY_STORE_INPUT(_target_memcpy, _to_write_memcpy, _len_memcpy) \
277 , [to_write_memcpy]"r"(_to_write_memcpy), \
278 [target_memcpy]"r"(_target_memcpy), \
279 [len_memcpy]"m"(_len_memcpy), \
280 [rseq_scratch0]"m"(rseq_scratch[0]), \
281 [rseq_scratch1]"m"(rseq_scratch[1]), \
282 [rseq_scratch2]"m"(rseq_scratch[2])
283
284#define RSEQ_FINISH_MEMCPY_CLOBBER()
285
286#define RSEQ_FINISH_MEMCPY_SCRATCH() \
287 uint32_t rseq_scratch[3];
288
289/*
290 * We need to save and restore those input registers so they can be
291 * modified within the assembly.
292 */
293#define RSEQ_FINISH_MEMCPY_SETUP() \
294 "movl %[to_write_memcpy], %[rseq_scratch0]\n\t" \
295 "movl %[target_memcpy], %[rseq_scratch1]\n\t" \
296 "movl %[len_memcpy], %%eax\n\t" \
297 "movl %%eax, %[rseq_scratch2]\n\t"
298
299#define RSEQ_FINISH_MEMCPY_TEARDOWN() \
300 "movl %[rseq_scratch1], %[target_memcpy]\n\t" \
301 "movl %[rseq_scratch0], %[to_write_memcpy]\n\t"
302
303#endif
This page took 0.036025 seconds and 5 git commands to generate.