Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* rwsem.h: R/W semaphores implemented using XADD/CMPXCHG for i486+ |
| 2 | * |
| 3 | * Written by David Howells (dhowells@redhat.com). |
| 4 | * |
| 5 | * Derived from asm-i386/semaphore.h |
| 6 | * |
| 7 | * |
| 8 | * The MSW of the count is the negated number of active writers and waiting |
| 9 | * lockers, and the LSW is the total number of active locks |
| 10 | * |
| 11 | * The lock count is initialized to 0 (no active and no waiting lockers). |
| 12 | * |
| 13 | * When a writer subtracts WRITE_BIAS, it'll get 0xffff0001 for the case of an |
| 14 | * uncontended lock. This can be determined because XADD returns the old value. |
| 15 | * Readers increment by 1 and see a positive value when uncontended, negative |
| 16 | * if there are writers (and maybe) readers waiting (in which case it goes to |
| 17 | * sleep). |
| 18 | * |
| 19 | * The value of WAITING_BIAS supports up to 32766 waiting processes. This can |
| 20 | * be extended to 65534 by manually checking the whole MSW rather than relying |
| 21 | * on the S flag. |
| 22 | * |
| 23 | * The value of ACTIVE_BIAS supports up to 65535 active processes. |
| 24 | * |
| 25 | * This should be totally fair - if anything is waiting, a process that wants a |
| 26 | * lock will go to the back of the queue. When the currently active lock is |
| 27 | * released, if there's a writer at the front of the queue, then that and only |
| 28 | * that will be woken up; if there's a bunch of consequtive readers at the |
| 29 | * front, then they'll all be woken up, but no other readers will be. |
| 30 | */ |
| 31 | |
| 32 | #ifndef _I386_RWSEM_H |
| 33 | #define _I386_RWSEM_H |
| 34 | |
| 35 | #ifndef _LINUX_RWSEM_H |
| 36 | #error "please don't include asm/rwsem.h directly, use linux/rwsem.h instead" |
| 37 | #endif |
| 38 | |
| 39 | #ifdef __KERNEL__ |
| 40 | |
| 41 | #include <linux/list.h> |
| 42 | #include <linux/spinlock.h> |
| 43 | |
| 44 | struct rwsem_waiter; |
| 45 | |
| 46 | extern struct rw_semaphore *FASTCALL(rwsem_down_read_failed(struct rw_semaphore *sem)); |
| 47 | extern struct rw_semaphore *FASTCALL(rwsem_down_write_failed(struct rw_semaphore *sem)); |
| 48 | extern struct rw_semaphore *FASTCALL(rwsem_wake(struct rw_semaphore *)); |
| 49 | extern struct rw_semaphore *FASTCALL(rwsem_downgrade_wake(struct rw_semaphore *sem)); |
| 50 | |
| 51 | /* |
| 52 | * the semaphore definition |
| 53 | */ |
| 54 | struct rw_semaphore { |
| 55 | signed long count; |
| 56 | #define RWSEM_UNLOCKED_VALUE 0x00000000 |
| 57 | #define RWSEM_ACTIVE_BIAS 0x00000001 |
| 58 | #define RWSEM_ACTIVE_MASK 0x0000ffff |
| 59 | #define RWSEM_WAITING_BIAS (-0x00010000) |
| 60 | #define RWSEM_ACTIVE_READ_BIAS RWSEM_ACTIVE_BIAS |
| 61 | #define RWSEM_ACTIVE_WRITE_BIAS (RWSEM_WAITING_BIAS + RWSEM_ACTIVE_BIAS) |
| 62 | spinlock_t wait_lock; |
| 63 | struct list_head wait_list; |
| 64 | #if RWSEM_DEBUG |
| 65 | int debug; |
| 66 | #endif |
| 67 | }; |
| 68 | |
| 69 | /* |
| 70 | * initialisation |
| 71 | */ |
| 72 | #if RWSEM_DEBUG |
| 73 | #define __RWSEM_DEBUG_INIT , 0 |
| 74 | #else |
| 75 | #define __RWSEM_DEBUG_INIT /* */ |
| 76 | #endif |
| 77 | |
| 78 | #define __RWSEM_INITIALIZER(name) \ |
| 79 | { RWSEM_UNLOCKED_VALUE, SPIN_LOCK_UNLOCKED, LIST_HEAD_INIT((name).wait_list) \ |
| 80 | __RWSEM_DEBUG_INIT } |
| 81 | |
| 82 | #define DECLARE_RWSEM(name) \ |
| 83 | struct rw_semaphore name = __RWSEM_INITIALIZER(name) |
| 84 | |
| 85 | static inline void init_rwsem(struct rw_semaphore *sem) |
| 86 | { |
| 87 | sem->count = RWSEM_UNLOCKED_VALUE; |
| 88 | spin_lock_init(&sem->wait_lock); |
| 89 | INIT_LIST_HEAD(&sem->wait_list); |
| 90 | #if RWSEM_DEBUG |
| 91 | sem->debug = 0; |
| 92 | #endif |
| 93 | } |
| 94 | |
| 95 | /* |
| 96 | * lock for reading |
| 97 | */ |
| 98 | static inline void __down_read(struct rw_semaphore *sem) |
| 99 | { |
| 100 | __asm__ __volatile__( |
| 101 | "# beginning down_read\n\t" |
| 102 | LOCK_PREFIX " incl (%%eax)\n\t" /* adds 0x00000001, returns the old value */ |
| 103 | " js 2f\n\t" /* jump if we weren't granted the lock */ |
| 104 | "1:\n\t" |
| 105 | LOCK_SECTION_START("") |
| 106 | "2:\n\t" |
| 107 | " pushl %%ecx\n\t" |
| 108 | " pushl %%edx\n\t" |
| 109 | " call rwsem_down_read_failed\n\t" |
| 110 | " popl %%edx\n\t" |
| 111 | " popl %%ecx\n\t" |
| 112 | " jmp 1b\n" |
| 113 | LOCK_SECTION_END |
| 114 | "# ending down_read\n\t" |
| 115 | : "=m"(sem->count) |
| 116 | : "a"(sem), "m"(sem->count) |
| 117 | : "memory", "cc"); |
| 118 | } |
| 119 | |
| 120 | /* |
| 121 | * trylock for reading -- returns 1 if successful, 0 if contention |
| 122 | */ |
| 123 | static inline int __down_read_trylock(struct rw_semaphore *sem) |
| 124 | { |
| 125 | __s32 result, tmp; |
| 126 | __asm__ __volatile__( |
| 127 | "# beginning __down_read_trylock\n\t" |
| 128 | " movl %0,%1\n\t" |
| 129 | "1:\n\t" |
| 130 | " movl %1,%2\n\t" |
| 131 | " addl %3,%2\n\t" |
| 132 | " jle 2f\n\t" |
| 133 | LOCK_PREFIX " cmpxchgl %2,%0\n\t" |
| 134 | " jnz 1b\n\t" |
| 135 | "2:\n\t" |
| 136 | "# ending __down_read_trylock\n\t" |
| 137 | : "+m"(sem->count), "=&a"(result), "=&r"(tmp) |
| 138 | : "i"(RWSEM_ACTIVE_READ_BIAS) |
| 139 | : "memory", "cc"); |
| 140 | return result>=0 ? 1 : 0; |
| 141 | } |
| 142 | |
| 143 | /* |
| 144 | * lock for writing |
| 145 | */ |
| 146 | static inline void __down_write(struct rw_semaphore *sem) |
| 147 | { |
| 148 | int tmp; |
| 149 | |
| 150 | tmp = RWSEM_ACTIVE_WRITE_BIAS; |
| 151 | __asm__ __volatile__( |
| 152 | "# beginning down_write\n\t" |
| 153 | LOCK_PREFIX " xadd %%edx,(%%eax)\n\t" /* subtract 0x0000ffff, returns the old value */ |
| 154 | " testl %%edx,%%edx\n\t" /* was the count 0 before? */ |
| 155 | " jnz 2f\n\t" /* jump if we weren't granted the lock */ |
| 156 | "1:\n\t" |
| 157 | LOCK_SECTION_START("") |
| 158 | "2:\n\t" |
| 159 | " pushl %%ecx\n\t" |
| 160 | " call rwsem_down_write_failed\n\t" |
| 161 | " popl %%ecx\n\t" |
| 162 | " jmp 1b\n" |
| 163 | LOCK_SECTION_END |
| 164 | "# ending down_write" |
| 165 | : "=m"(sem->count), "=d"(tmp) |
| 166 | : "a"(sem), "1"(tmp), "m"(sem->count) |
| 167 | : "memory", "cc"); |
| 168 | } |
| 169 | |
| 170 | /* |
| 171 | * trylock for writing -- returns 1 if successful, 0 if contention |
| 172 | */ |
| 173 | static inline int __down_write_trylock(struct rw_semaphore *sem) |
| 174 | { |
| 175 | signed long ret = cmpxchg(&sem->count, |
| 176 | RWSEM_UNLOCKED_VALUE, |
| 177 | RWSEM_ACTIVE_WRITE_BIAS); |
| 178 | if (ret == RWSEM_UNLOCKED_VALUE) |
| 179 | return 1; |
| 180 | return 0; |
| 181 | } |
| 182 | |
| 183 | /* |
| 184 | * unlock after reading |
| 185 | */ |
| 186 | static inline void __up_read(struct rw_semaphore *sem) |
| 187 | { |
| 188 | __s32 tmp = -RWSEM_ACTIVE_READ_BIAS; |
| 189 | __asm__ __volatile__( |
| 190 | "# beginning __up_read\n\t" |
| 191 | LOCK_PREFIX " xadd %%edx,(%%eax)\n\t" /* subtracts 1, returns the old value */ |
| 192 | " js 2f\n\t" /* jump if the lock is being waited upon */ |
| 193 | "1:\n\t" |
| 194 | LOCK_SECTION_START("") |
| 195 | "2:\n\t" |
| 196 | " decw %%dx\n\t" /* do nothing if still outstanding active readers */ |
| 197 | " jnz 1b\n\t" |
| 198 | " pushl %%ecx\n\t" |
| 199 | " call rwsem_wake\n\t" |
| 200 | " popl %%ecx\n\t" |
| 201 | " jmp 1b\n" |
| 202 | LOCK_SECTION_END |
| 203 | "# ending __up_read\n" |
| 204 | : "=m"(sem->count), "=d"(tmp) |
| 205 | : "a"(sem), "1"(tmp), "m"(sem->count) |
| 206 | : "memory", "cc"); |
| 207 | } |
| 208 | |
| 209 | /* |
| 210 | * unlock after writing |
| 211 | */ |
| 212 | static inline void __up_write(struct rw_semaphore *sem) |
| 213 | { |
| 214 | __asm__ __volatile__( |
| 215 | "# beginning __up_write\n\t" |
| 216 | " movl %2,%%edx\n\t" |
| 217 | LOCK_PREFIX " xaddl %%edx,(%%eax)\n\t" /* tries to transition 0xffff0001 -> 0x00000000 */ |
| 218 | " jnz 2f\n\t" /* jump if the lock is being waited upon */ |
| 219 | "1:\n\t" |
| 220 | LOCK_SECTION_START("") |
| 221 | "2:\n\t" |
| 222 | " decw %%dx\n\t" /* did the active count reduce to 0? */ |
| 223 | " jnz 1b\n\t" /* jump back if not */ |
| 224 | " pushl %%ecx\n\t" |
| 225 | " call rwsem_wake\n\t" |
| 226 | " popl %%ecx\n\t" |
| 227 | " jmp 1b\n" |
| 228 | LOCK_SECTION_END |
| 229 | "# ending __up_write\n" |
| 230 | : "=m"(sem->count) |
| 231 | : "a"(sem), "i"(-RWSEM_ACTIVE_WRITE_BIAS), "m"(sem->count) |
| 232 | : "memory", "cc", "edx"); |
| 233 | } |
| 234 | |
| 235 | /* |
| 236 | * downgrade write lock to read lock |
| 237 | */ |
| 238 | static inline void __downgrade_write(struct rw_semaphore *sem) |
| 239 | { |
| 240 | __asm__ __volatile__( |
| 241 | "# beginning __downgrade_write\n\t" |
| 242 | LOCK_PREFIX " addl %2,(%%eax)\n\t" /* transitions 0xZZZZ0001 -> 0xYYYY0001 */ |
| 243 | " js 2f\n\t" /* jump if the lock is being waited upon */ |
| 244 | "1:\n\t" |
| 245 | LOCK_SECTION_START("") |
| 246 | "2:\n\t" |
| 247 | " pushl %%ecx\n\t" |
| 248 | " pushl %%edx\n\t" |
| 249 | " call rwsem_downgrade_wake\n\t" |
| 250 | " popl %%edx\n\t" |
| 251 | " popl %%ecx\n\t" |
| 252 | " jmp 1b\n" |
| 253 | LOCK_SECTION_END |
| 254 | "# ending __downgrade_write\n" |
| 255 | : "=m"(sem->count) |
| 256 | : "a"(sem), "i"(-RWSEM_WAITING_BIAS), "m"(sem->count) |
| 257 | : "memory", "cc"); |
| 258 | } |
| 259 | |
| 260 | /* |
| 261 | * implement atomic add functionality |
| 262 | */ |
| 263 | static inline void rwsem_atomic_add(int delta, struct rw_semaphore *sem) |
| 264 | { |
| 265 | __asm__ __volatile__( |
| 266 | LOCK_PREFIX "addl %1,%0" |
| 267 | : "=m"(sem->count) |
| 268 | : "ir"(delta), "m"(sem->count)); |
| 269 | } |
| 270 | |
| 271 | /* |
| 272 | * implement exchange and add functionality |
| 273 | */ |
| 274 | static inline int rwsem_atomic_update(int delta, struct rw_semaphore *sem) |
| 275 | { |
| 276 | int tmp = delta; |
| 277 | |
| 278 | __asm__ __volatile__( |
| 279 | LOCK_PREFIX "xadd %0,(%2)" |
| 280 | : "+r"(tmp), "=m"(sem->count) |
| 281 | : "r"(sem), "m"(sem->count) |
| 282 | : "memory"); |
| 283 | |
| 284 | return tmp+delta; |
| 285 | } |
| 286 | |
Rik Van Riel | eb92f4e | 2005-10-29 18:15:44 -0700 | [diff] [blame] | 287 | static inline int rwsem_is_locked(struct rw_semaphore *sem) |
| 288 | { |
| 289 | return (sem->count != 0); |
| 290 | } |
| 291 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 292 | #endif /* __KERNEL__ */ |
| 293 | #endif /* _I386_RWSEM_H */ |