| 1 | /* Copyright (C) 2002-2019 Free Software Foundation, Inc. |
| 2 | This file is part of the GNU C Library. |
| 3 | Contributed by Ulrich Drepper <drepper@redhat.com>, 2002. |
| 4 | |
| 5 | The GNU C Library is free software; you can redistribute it and/or |
| 6 | modify it under the terms of the GNU Lesser General Public |
| 7 | License as published by the Free Software Foundation; either |
| 8 | version 2.1 of the License, or (at your option) any later version. |
| 9 | |
| 10 | The GNU C Library is distributed in the hope that it will be useful, |
| 11 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 13 | Lesser General Public License for more details. |
| 14 | |
| 15 | You should have received a copy of the GNU Lesser General Public |
| 16 | License along with the GNU C Library; if not, see |
| 17 | <http://www.gnu.org/licenses/>. */ |
| 18 | |
| 19 | #ifndef _LOWLEVELLOCK_H |
| 20 | #define _LOWLEVELLOCK_H 1 |
| 21 | |
| 22 | #include <stap-probe.h> |
| 23 | |
| 24 | #ifndef __ASSEMBLER__ |
| 25 | # include <time.h> |
| 26 | # include <sys/param.h> |
| 27 | # include <bits/pthreadtypes.h> |
| 28 | # include <kernel-features.h> |
| 29 | |
| 30 | # ifndef LOCK_INSTR |
| 31 | # ifdef UP |
| 32 | # define LOCK_INSTR /* nothing */ |
| 33 | # else |
| 34 | # define LOCK_INSTR "lock;" |
| 35 | # endif |
| 36 | # endif |
| 37 | #else |
| 38 | # ifndef LOCK |
| 39 | # ifdef UP |
| 40 | # define LOCK |
| 41 | # else |
| 42 | # define LOCK lock |
| 43 | # endif |
| 44 | # endif |
| 45 | #endif |
| 46 | |
| 47 | #include <lowlevellock-futex.h> |
| 48 | |
| 49 | /* XXX Remove when no assembler code uses futexes anymore. */ |
| 50 | #define SYS_futex __NR_futex |
| 51 | |
| 52 | #ifndef __ASSEMBLER__ |
| 53 | |
| 54 | /* Initializer for lock. */ |
| 55 | #define LLL_LOCK_INITIALIZER (0) |
| 56 | #define LLL_LOCK_INITIALIZER_LOCKED (1) |
| 57 | #define LLL_LOCK_INITIALIZER_WAITERS (2) |
| 58 | |
| 59 | |
| 60 | /* NB: in the lll_trylock macro we simply return the value in %eax |
| 61 | after the cmpxchg instruction. In case the operation succeded this |
| 62 | value is zero. In case the operation failed, the cmpxchg instruction |
| 63 | has loaded the current value of the memory work which is guaranteed |
| 64 | to be nonzero. */ |
| 65 | #if !IS_IN (libc) || defined UP |
| 66 | # define __lll_trylock_asm LOCK_INSTR "cmpxchgl %2, %1" |
| 67 | #else |
| 68 | # define __lll_trylock_asm "cmpl $0, __libc_multiple_threads(%%rip)\n\t" \ |
| 69 | "je 0f\n\t" \ |
| 70 | "lock; cmpxchgl %2, %1\n\t" \ |
| 71 | "jmp 1f\n\t" \ |
| 72 | "0:\tcmpxchgl %2, %1\n\t" \ |
| 73 | "1:" |
| 74 | #endif |
| 75 | |
| 76 | #define lll_trylock(futex) \ |
| 77 | ({ int ret; \ |
| 78 | __asm __volatile (__lll_trylock_asm \ |
| 79 | : "=a" (ret), "=m" (futex) \ |
| 80 | : "r" (LLL_LOCK_INITIALIZER_LOCKED), "m" (futex), \ |
| 81 | "0" (LLL_LOCK_INITIALIZER) \ |
| 82 | : "memory"); \ |
| 83 | ret; }) |
| 84 | |
| 85 | #define lll_cond_trylock(futex) \ |
| 86 | ({ int ret; \ |
| 87 | __asm __volatile (LOCK_INSTR "cmpxchgl %2, %1" \ |
| 88 | : "=a" (ret), "=m" (futex) \ |
| 89 | : "r" (LLL_LOCK_INITIALIZER_WAITERS), \ |
| 90 | "m" (futex), "0" (LLL_LOCK_INITIALIZER) \ |
| 91 | : "memory"); \ |
| 92 | ret; }) |
| 93 | |
| 94 | #if !IS_IN (libc) || defined UP |
| 95 | # define __lll_lock_asm_start LOCK_INSTR "cmpxchgl %4, %2\n\t" \ |
| 96 | "jz 24f\n\t" |
| 97 | #else |
| 98 | # define __lll_lock_asm_start "cmpl $0, __libc_multiple_threads(%%rip)\n\t" \ |
| 99 | "je 0f\n\t" \ |
| 100 | "lock; cmpxchgl %4, %2\n\t" \ |
| 101 | "jnz 1f\n\t" \ |
| 102 | "jmp 24f\n" \ |
| 103 | "0:\tcmpxchgl %4, %2\n\t" \ |
| 104 | "jz 24f\n\t" |
| 105 | #endif |
| 106 | |
| 107 | #define lll_lock(futex, private) \ |
| 108 | (void) \ |
| 109 | ({ int ignore1, ignore2, ignore3; \ |
| 110 | if (__builtin_constant_p (private) && (private) == LLL_PRIVATE) \ |
| 111 | __asm __volatile (__lll_lock_asm_start \ |
| 112 | "1:\tlea %2, %%" RDI_LP "\n" \ |
| 113 | "2:\tsub $128, %%" RSP_LP "\n" \ |
| 114 | ".cfi_adjust_cfa_offset 128\n" \ |
| 115 | "3:\tcallq __lll_lock_wait_private\n" \ |
| 116 | "4:\tadd $128, %%" RSP_LP "\n" \ |
| 117 | ".cfi_adjust_cfa_offset -128\n" \ |
| 118 | "24:" \ |
| 119 | : "=S" (ignore1), "=&D" (ignore2), "=m" (futex), \ |
| 120 | "=a" (ignore3) \ |
| 121 | : "0" (1), "m" (futex), "3" (0) \ |
| 122 | : "cx", "r11", "cc", "memory"); \ |
| 123 | else \ |
| 124 | __asm __volatile (__lll_lock_asm_start \ |
| 125 | "1:\tlea %2, %%" RDI_LP "\n" \ |
| 126 | "2:\tsub $128, %%" RSP_LP "\n" \ |
| 127 | ".cfi_adjust_cfa_offset 128\n" \ |
| 128 | "3:\tcallq __lll_lock_wait\n" \ |
| 129 | "4:\tadd $128, %%" RSP_LP "\n" \ |
| 130 | ".cfi_adjust_cfa_offset -128\n" \ |
| 131 | "24:" \ |
| 132 | : "=S" (ignore1), "=D" (ignore2), "=m" (futex), \ |
| 133 | "=a" (ignore3) \ |
| 134 | : "1" (1), "m" (futex), "3" (0), "0" (private) \ |
| 135 | : "cx", "r11", "cc", "memory"); \ |
| 136 | }) \ |
| 137 | |
| 138 | #define lll_cond_lock(futex, private) \ |
| 139 | (void) \ |
| 140 | ({ int ignore1, ignore2, ignore3; \ |
| 141 | __asm __volatile (LOCK_INSTR "cmpxchgl %4, %2\n\t" \ |
| 142 | "jz 24f\n" \ |
| 143 | "1:\tlea %2, %%" RDI_LP "\n" \ |
| 144 | "2:\tsub $128, %%" RSP_LP "\n" \ |
| 145 | ".cfi_adjust_cfa_offset 128\n" \ |
| 146 | "3:\tcallq __lll_lock_wait\n" \ |
| 147 | "4:\tadd $128, %%" RSP_LP "\n" \ |
| 148 | ".cfi_adjust_cfa_offset -128\n" \ |
| 149 | "24:" \ |
| 150 | : "=S" (ignore1), "=D" (ignore2), "=m" (futex), \ |
| 151 | "=a" (ignore3) \ |
| 152 | : "1" (2), "m" (futex), "3" (0), "0" (private) \ |
| 153 | : "cx", "r11", "cc", "memory"); \ |
| 154 | }) |
| 155 | |
| 156 | #define lll_timedlock(futex, timeout, private) \ |
| 157 | ({ int result, ignore1, ignore2, ignore3; \ |
| 158 | __asm __volatile (LOCK_INSTR "cmpxchgl %1, %4\n\t" \ |
| 159 | "jz 24f\n" \ |
| 160 | "1:\tlea %4, %%" RDI_LP "\n" \ |
| 161 | "0:\tmov %8, %%" RDX_LP "\n" \ |
| 162 | "2:\tsub $128, %%" RSP_LP "\n" \ |
| 163 | ".cfi_adjust_cfa_offset 128\n" \ |
| 164 | "3:\tcallq __lll_timedlock_wait\n" \ |
| 165 | "4:\tadd $128, %%" RSP_LP "\n" \ |
| 166 | ".cfi_adjust_cfa_offset -128\n" \ |
| 167 | "24:" \ |
| 168 | : "=a" (result), "=D" (ignore1), "=S" (ignore2), \ |
| 169 | "=&d" (ignore3), "=m" (futex) \ |
| 170 | : "0" (0), "1" (1), "m" (futex), "m" (timeout), \ |
| 171 | "2" (private) \ |
| 172 | : "memory", "cx", "cc", "r10", "r11"); \ |
| 173 | result; }) |
| 174 | |
| 175 | extern int __lll_timedlock_elision (int *futex, short *adapt_count, |
| 176 | const struct timespec *timeout, |
| 177 | int private) attribute_hidden; |
| 178 | |
| 179 | #define lll_timedlock_elision(futex, adapt_count, timeout, private) \ |
| 180 | __lll_timedlock_elision(&(futex), &(adapt_count), timeout, private) |
| 181 | |
| 182 | #if !IS_IN (libc) || defined UP |
| 183 | # define __lll_unlock_asm_start LOCK_INSTR "decl %0\n\t" \ |
| 184 | "je 24f\n\t" |
| 185 | #else |
| 186 | # define __lll_unlock_asm_start "cmpl $0, __libc_multiple_threads(%%rip)\n\t" \ |
| 187 | "je 0f\n\t" \ |
| 188 | "lock; decl %0\n\t" \ |
| 189 | "jne 1f\n\t" \ |
| 190 | "jmp 24f\n\t" \ |
| 191 | "0:\tdecl %0\n\t" \ |
| 192 | "je 24f\n\t" |
| 193 | #endif |
| 194 | |
| 195 | #define lll_unlock(futex, private) \ |
| 196 | (void) \ |
| 197 | ({ int ignore; \ |
| 198 | if (__builtin_constant_p (private) && (private) == LLL_PRIVATE) \ |
| 199 | __asm __volatile (__lll_unlock_asm_start \ |
| 200 | "1:\tlea %0, %%" RDI_LP "\n" \ |
| 201 | "2:\tsub $128, %%" RSP_LP "\n" \ |
| 202 | ".cfi_adjust_cfa_offset 128\n" \ |
| 203 | "3:\tcallq __lll_unlock_wake_private\n" \ |
| 204 | "4:\tadd $128, %%" RSP_LP "\n" \ |
| 205 | ".cfi_adjust_cfa_offset -128\n" \ |
| 206 | "24:" \ |
| 207 | : "=m" (futex), "=&D" (ignore) \ |
| 208 | : "m" (futex) \ |
| 209 | : "ax", "cx", "r11", "cc", "memory"); \ |
| 210 | else \ |
| 211 | __asm __volatile (__lll_unlock_asm_start \ |
| 212 | "1:\tlea %0, %%" RDI_LP "\n" \ |
| 213 | "2:\tsub $128, %%" RSP_LP "\n" \ |
| 214 | ".cfi_adjust_cfa_offset 128\n" \ |
| 215 | "3:\tcallq __lll_unlock_wake\n" \ |
| 216 | "4:\tadd $128, %%" RSP_LP "\n" \ |
| 217 | ".cfi_adjust_cfa_offset -128\n" \ |
| 218 | "24:" \ |
| 219 | : "=m" (futex), "=&D" (ignore) \ |
| 220 | : "m" (futex), "S" (private) \ |
| 221 | : "ax", "cx", "r11", "cc", "memory"); \ |
| 222 | }) |
| 223 | |
| 224 | #define lll_islocked(futex) \ |
| 225 | (futex != LLL_LOCK_INITIALIZER) |
| 226 | |
| 227 | extern int __lll_timedwait_tid (int *, const struct timespec *) |
| 228 | attribute_hidden; |
| 229 | |
| 230 | /* The kernel notifies a process which uses CLONE_CHILD_CLEARTID via futex |
| 231 | wake-up when the clone terminates. The memory location contains the |
| 232 | thread ID while the clone is running and is reset to zero by the kernel |
| 233 | afterwards. The kernel up to version 3.16.3 does not use the private futex |
| 234 | operations for futex wake-up when the clone terminates. |
| 235 | If ABSTIME is not NULL, is used a timeout for futex call. If the timeout |
| 236 | occurs then return ETIMEOUT, if ABSTIME is invalid, return EINVAL. |
| 237 | The futex operation are issues with cancellable versions. */ |
| 238 | #define lll_wait_tid(tid, abstime) \ |
| 239 | ({ \ |
| 240 | int __res = 0; \ |
| 241 | __typeof (tid) __tid; \ |
| 242 | if (abstime != NULL) \ |
| 243 | __res = __lll_timedwait_tid (&(tid), (abstime)); \ |
| 244 | else \ |
| 245 | /* We need acquire MO here so that we synchronize with the \ |
| 246 | kernel's store to 0 when the clone terminates. (see above) */ \ |
| 247 | while ((__tid = atomic_load_acquire (&(tid))) != 0) \ |
| 248 | lll_futex_wait_cancel (&(tid), __tid, LLL_SHARED); \ |
| 249 | __res; \ |
| 250 | }) |
| 251 | |
| 252 | extern int __lll_lock_elision (int *futex, short *adapt_count, int private) |
| 253 | attribute_hidden; |
| 254 | |
| 255 | extern int __lll_unlock_elision (int *lock, int private) |
| 256 | attribute_hidden; |
| 257 | |
| 258 | extern int __lll_trylock_elision (int *lock, short *adapt_count) |
| 259 | attribute_hidden; |
| 260 | |
| 261 | #define lll_lock_elision(futex, adapt_count, private) \ |
| 262 | __lll_lock_elision (&(futex), &(adapt_count), private) |
| 263 | #define lll_unlock_elision(futex, adapt_count, private) \ |
| 264 | __lll_unlock_elision (&(futex), private) |
| 265 | #define lll_trylock_elision(futex, adapt_count) \ |
| 266 | __lll_trylock_elision (&(futex), &(adapt_count)) |
| 267 | |
| 268 | #endif /* !__ASSEMBLER__ */ |
| 269 | |
| 270 | #endif /* lowlevellock.h */ |
| 271 | |