Thomas Gleixner | c942fdd | 2019-05-27 08:55:06 +0200 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 2 | /* |
| 3 | * Queued spinlock |
| 4 | * |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 5 | * (C) Copyright 2013-2015 Hewlett-Packard Development Company, L.P. |
Waiman Long | 81d3dc9 | 2018-04-26 11:34:27 +0100 | [diff] [blame] | 6 | * (C) Copyright 2013-2014,2018 Red Hat, Inc. |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 7 | * (C) Copyright 2015 Intel Corp. |
Waiman Long | 64d816c | 2015-11-09 19:09:21 -0500 | [diff] [blame] | 8 | * (C) Copyright 2015 Hewlett-Packard Enterprise Development LP |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 9 | * |
Waiman Long | 81d3dc9 | 2018-04-26 11:34:27 +0100 | [diff] [blame] | 10 | * Authors: Waiman Long <longman@redhat.com> |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 11 | * Peter Zijlstra <peterz@infradead.org> |
| 12 | */ |
Waiman Long | a23db28 | 2015-04-24 14:56:37 -0400 | [diff] [blame] | 13 | |
| 14 | #ifndef _GEN_PV_LOCK_SLOWPATH |
| 15 | |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 16 | #include <linux/smp.h> |
| 17 | #include <linux/bug.h> |
| 18 | #include <linux/cpumask.h> |
| 19 | #include <linux/percpu.h> |
| 20 | #include <linux/hardirq.h> |
| 21 | #include <linux/mutex.h> |
Stafford Horne | 5671360 | 2017-07-08 04:56:58 +0900 | [diff] [blame] | 22 | #include <linux/prefetch.h> |
Peter Zijlstra (Intel) | 69f9cae | 2015-04-24 14:56:34 -0400 | [diff] [blame] | 23 | #include <asm/byteorder.h> |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 24 | #include <asm/qspinlock.h> |
| 25 | |
| 26 | /* |
Waiman Long | 81d3dc9 | 2018-04-26 11:34:27 +0100 | [diff] [blame] | 27 | * Include queued spinlock statistics code |
| 28 | */ |
| 29 | #include "qspinlock_stat.h" |
| 30 | |
| 31 | /* |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 32 | * The basic principle of a queue-based spinlock can best be understood |
| 33 | * by studying a classic queue-based spinlock implementation called the |
Waiman Long | 5709712 | 2020-01-07 12:49:14 -0500 | [diff] [blame] | 34 | * MCS lock. A copy of the original MCS lock paper ("Algorithms for Scalable |
| 35 | * Synchronization on Shared-Memory Multiprocessors by Mellor-Crummey and |
| 36 | * Scott") is available at |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 37 | * |
Waiman Long | 5709712 | 2020-01-07 12:49:14 -0500 | [diff] [blame] | 38 | * https://bugzilla.kernel.org/show_bug.cgi?id=206115 |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 39 | * |
Waiman Long | 5709712 | 2020-01-07 12:49:14 -0500 | [diff] [blame] | 40 | * This queued spinlock implementation is based on the MCS lock, however to |
| 41 | * make it fit the 4 bytes we assume spinlock_t to be, and preserve its |
| 42 | * existing API, we must modify it somehow. |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 43 | * |
| 44 | * In particular; where the traditional MCS lock consists of a tail pointer |
| 45 | * (8 bytes) and needs the next pointer (another 8 bytes) of its own node to |
| 46 | * unlock the next pending (next->locked), we compress both these: {tail, |
| 47 | * next->locked} into a single u32 value. |
| 48 | * |
| 49 | * Since a spinlock disables recursion of its own context and there is a limit |
| 50 | * to the contexts that can nest; namely: task, softirq, hardirq, nmi. As there |
| 51 | * are at most 4 nesting levels, it can be encoded by a 2-bit number. Now |
| 52 | * we can encode the tail by combining the 2-bit nesting level with the cpu |
| 53 | * number. With one byte for the lock value and 3 bytes for the tail, only a |
| 54 | * 32-bit word is now needed. Even though we only need 1 bit for the lock, |
| 55 | * we extend it to a full byte to achieve better performance for architectures |
| 56 | * that support atomic byte write. |
| 57 | * |
| 58 | * We also change the first spinner to spin on the lock bit instead of its |
| 59 | * node; whereby avoiding the need to carry a node from lock to unlock, and |
| 60 | * preserving existing lock API. This also makes the unlock code simpler and |
| 61 | * faster. |
Peter Zijlstra (Intel) | 69f9cae | 2015-04-24 14:56:34 -0400 | [diff] [blame] | 62 | * |
| 63 | * N.B. The current implementation only supports architectures that allow |
| 64 | * atomic operations on smaller 8-bit and 16-bit data types. |
| 65 | * |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 66 | */ |
| 67 | |
| 68 | #include "mcs_spinlock.h" |
Waiman Long | a23db28 | 2015-04-24 14:56:37 -0400 | [diff] [blame] | 69 | #define MAX_NODES 4 |
Waiman Long | 0fa809c | 2018-10-16 09:45:07 -0400 | [diff] [blame] | 70 | |
| 71 | /* |
| 72 | * On 64-bit architectures, the mcs_spinlock structure will be 16 bytes in |
| 73 | * size and four of them will fit nicely in one 64-byte cacheline. For |
| 74 | * pvqspinlock, however, we need more space for extra data. To accommodate |
| 75 | * that, we insert two more long words to pad it up to 32 bytes. IOW, only |
| 76 | * two of them can fit in a cacheline in this case. That is OK as it is rare |
| 77 | * to have more than 2 levels of slowpath nesting in actual use. We don't |
| 78 | * want to penalize pvqspinlocks to optimize for a rare case in native |
| 79 | * qspinlocks. |
| 80 | */ |
| 81 | struct qnode { |
| 82 | struct mcs_spinlock mcs; |
| 83 | #ifdef CONFIG_PARAVIRT_SPINLOCKS |
| 84 | long reserved[2]; |
Waiman Long | a23db28 | 2015-04-24 14:56:37 -0400 | [diff] [blame] | 85 | #endif |
Waiman Long | 0fa809c | 2018-10-16 09:45:07 -0400 | [diff] [blame] | 86 | }; |
Waiman Long | a23db28 | 2015-04-24 14:56:37 -0400 | [diff] [blame] | 87 | |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 88 | /* |
Will Deacon | 6512276 | 2018-04-26 11:34:17 +0100 | [diff] [blame] | 89 | * The pending bit spinning loop count. |
| 90 | * This heuristic is used to limit the number of lockword accesses |
| 91 | * made by atomic_cond_read_relaxed when waiting for the lock to |
| 92 | * transition out of the "== _Q_PENDING_VAL" state. We don't spin |
| 93 | * indefinitely because there's no guarantee that we'll make forward |
| 94 | * progress. |
| 95 | */ |
| 96 | #ifndef _Q_PENDING_LOOPS |
| 97 | #define _Q_PENDING_LOOPS 1 |
| 98 | #endif |
| 99 | |
| 100 | /* |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 101 | * Per-CPU queue node structures; we can never have more than 4 nested |
| 102 | * contexts: task, softirq, hardirq, nmi. |
| 103 | * |
| 104 | * Exactly fits one 64-byte cacheline on a 64-bit architecture. |
Waiman Long | a23db28 | 2015-04-24 14:56:37 -0400 | [diff] [blame] | 105 | * |
| 106 | * PV doubles the storage and uses the second cacheline for PV state. |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 107 | */ |
Waiman Long | 0fa809c | 2018-10-16 09:45:07 -0400 | [diff] [blame] | 108 | static DEFINE_PER_CPU_ALIGNED(struct qnode, qnodes[MAX_NODES]); |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 109 | |
| 110 | /* |
| 111 | * We must be able to distinguish between no-tail and the tail at 0:0, |
| 112 | * therefore increment the cpu number by one. |
| 113 | */ |
| 114 | |
Peter Zijlstra | 8d53fa1 | 2016-06-08 09:12:30 +0200 | [diff] [blame] | 115 | static inline __pure u32 encode_tail(int cpu, int idx) |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 116 | { |
| 117 | u32 tail; |
| 118 | |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 119 | tail = (cpu + 1) << _Q_TAIL_CPU_OFFSET; |
| 120 | tail |= idx << _Q_TAIL_IDX_OFFSET; /* assume < 4 */ |
| 121 | |
| 122 | return tail; |
| 123 | } |
| 124 | |
Peter Zijlstra | 8d53fa1 | 2016-06-08 09:12:30 +0200 | [diff] [blame] | 125 | static inline __pure struct mcs_spinlock *decode_tail(u32 tail) |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 126 | { |
| 127 | int cpu = (tail >> _Q_TAIL_CPU_OFFSET) - 1; |
| 128 | int idx = (tail & _Q_TAIL_IDX_MASK) >> _Q_TAIL_IDX_OFFSET; |
| 129 | |
Waiman Long | 0fa809c | 2018-10-16 09:45:07 -0400 | [diff] [blame] | 130 | return per_cpu_ptr(&qnodes[idx].mcs, cpu); |
| 131 | } |
| 132 | |
| 133 | static inline __pure |
| 134 | struct mcs_spinlock *grab_mcs_node(struct mcs_spinlock *base, int idx) |
| 135 | { |
| 136 | return &((struct qnode *)base + idx)->mcs; |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 137 | } |
| 138 | |
Peter Zijlstra (Intel) | c1fb159 | 2015-04-24 14:56:32 -0400 | [diff] [blame] | 139 | #define _Q_LOCKED_PENDING_MASK (_Q_LOCKED_MASK | _Q_PENDING_MASK) |
| 140 | |
Waiman Long | 2c83e8e | 2015-04-24 14:56:35 -0400 | [diff] [blame] | 141 | #if _Q_PENDING_BITS == 8 |
Peter Zijlstra (Intel) | 69f9cae | 2015-04-24 14:56:34 -0400 | [diff] [blame] | 142 | /** |
Will Deacon | 59fb586 | 2018-04-26 11:34:19 +0100 | [diff] [blame] | 143 | * clear_pending - clear the pending bit. |
| 144 | * @lock: Pointer to queued spinlock structure |
| 145 | * |
| 146 | * *,1,* -> *,0,* |
| 147 | */ |
| 148 | static __always_inline void clear_pending(struct qspinlock *lock) |
| 149 | { |
| 150 | WRITE_ONCE(lock->pending, 0); |
| 151 | } |
| 152 | |
| 153 | /** |
Peter Zijlstra (Intel) | 69f9cae | 2015-04-24 14:56:34 -0400 | [diff] [blame] | 154 | * clear_pending_set_locked - take ownership and clear the pending bit. |
| 155 | * @lock: Pointer to queued spinlock structure |
| 156 | * |
| 157 | * *,1,0 -> *,0,1 |
| 158 | * |
| 159 | * Lock stealing is not allowed if this function is used. |
| 160 | */ |
| 161 | static __always_inline void clear_pending_set_locked(struct qspinlock *lock) |
| 162 | { |
Will Deacon | 625e88b | 2018-04-26 11:34:16 +0100 | [diff] [blame] | 163 | WRITE_ONCE(lock->locked_pending, _Q_LOCKED_VAL); |
Peter Zijlstra (Intel) | 69f9cae | 2015-04-24 14:56:34 -0400 | [diff] [blame] | 164 | } |
| 165 | |
| 166 | /* |
| 167 | * xchg_tail - Put in the new queue tail code word & retrieve previous one |
| 168 | * @lock : Pointer to queued spinlock structure |
| 169 | * @tail : The new queue tail code word |
| 170 | * Return: The previous queue tail code word |
| 171 | * |
Paul E. McKenney | 548095d | 2017-10-09 11:22:50 -0700 | [diff] [blame] | 172 | * xchg(lock, tail), which heads an address dependency |
Peter Zijlstra (Intel) | 69f9cae | 2015-04-24 14:56:34 -0400 | [diff] [blame] | 173 | * |
| 174 | * p,*,* -> n,*,* ; prev = xchg(lock, node) |
| 175 | */ |
| 176 | static __always_inline u32 xchg_tail(struct qspinlock *lock, u32 tail) |
| 177 | { |
Waiman Long | 64d816c | 2015-11-09 19:09:21 -0500 | [diff] [blame] | 178 | /* |
Will Deacon | 9d4646d | 2018-04-26 11:34:25 +0100 | [diff] [blame] | 179 | * We can use relaxed semantics since the caller ensures that the |
| 180 | * MCS node is properly initialized before updating the tail. |
Waiman Long | 64d816c | 2015-11-09 19:09:21 -0500 | [diff] [blame] | 181 | */ |
Will Deacon | 9d4646d | 2018-04-26 11:34:25 +0100 | [diff] [blame] | 182 | return (u32)xchg_relaxed(&lock->tail, |
Waiman Long | 64d816c | 2015-11-09 19:09:21 -0500 | [diff] [blame] | 183 | tail >> _Q_TAIL_OFFSET) << _Q_TAIL_OFFSET; |
Peter Zijlstra (Intel) | 69f9cae | 2015-04-24 14:56:34 -0400 | [diff] [blame] | 184 | } |
| 185 | |
| 186 | #else /* _Q_PENDING_BITS == 8 */ |
| 187 | |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 188 | /** |
Will Deacon | 59fb586 | 2018-04-26 11:34:19 +0100 | [diff] [blame] | 189 | * clear_pending - clear the pending bit. |
| 190 | * @lock: Pointer to queued spinlock structure |
| 191 | * |
| 192 | * *,1,* -> *,0,* |
| 193 | */ |
| 194 | static __always_inline void clear_pending(struct qspinlock *lock) |
| 195 | { |
| 196 | atomic_andnot(_Q_PENDING_VAL, &lock->val); |
| 197 | } |
| 198 | |
| 199 | /** |
Waiman Long | 6403bd7 | 2015-04-24 14:56:33 -0400 | [diff] [blame] | 200 | * clear_pending_set_locked - take ownership and clear the pending bit. |
| 201 | * @lock: Pointer to queued spinlock structure |
| 202 | * |
| 203 | * *,1,0 -> *,0,1 |
| 204 | */ |
| 205 | static __always_inline void clear_pending_set_locked(struct qspinlock *lock) |
| 206 | { |
| 207 | atomic_add(-_Q_PENDING_VAL + _Q_LOCKED_VAL, &lock->val); |
| 208 | } |
| 209 | |
| 210 | /** |
| 211 | * xchg_tail - Put in the new queue tail code word & retrieve previous one |
| 212 | * @lock : Pointer to queued spinlock structure |
| 213 | * @tail : The new queue tail code word |
| 214 | * Return: The previous queue tail code word |
| 215 | * |
| 216 | * xchg(lock, tail) |
| 217 | * |
| 218 | * p,*,* -> n,*,* ; prev = xchg(lock, node) |
| 219 | */ |
| 220 | static __always_inline u32 xchg_tail(struct qspinlock *lock, u32 tail) |
| 221 | { |
| 222 | u32 old, new, val = atomic_read(&lock->val); |
| 223 | |
| 224 | for (;;) { |
| 225 | new = (val & _Q_LOCKED_PENDING_MASK) | tail; |
Waiman Long | 64d816c | 2015-11-09 19:09:21 -0500 | [diff] [blame] | 226 | /* |
Will Deacon | 9d4646d | 2018-04-26 11:34:25 +0100 | [diff] [blame] | 227 | * We can use relaxed semantics since the caller ensures that |
| 228 | * the MCS node is properly initialized before updating the |
| 229 | * tail. |
Waiman Long | 64d816c | 2015-11-09 19:09:21 -0500 | [diff] [blame] | 230 | */ |
Will Deacon | 9d4646d | 2018-04-26 11:34:25 +0100 | [diff] [blame] | 231 | old = atomic_cmpxchg_relaxed(&lock->val, val, new); |
Waiman Long | 6403bd7 | 2015-04-24 14:56:33 -0400 | [diff] [blame] | 232 | if (old == val) |
| 233 | break; |
| 234 | |
| 235 | val = old; |
| 236 | } |
| 237 | return old; |
| 238 | } |
Peter Zijlstra (Intel) | 69f9cae | 2015-04-24 14:56:34 -0400 | [diff] [blame] | 239 | #endif /* _Q_PENDING_BITS == 8 */ |
Waiman Long | 6403bd7 | 2015-04-24 14:56:33 -0400 | [diff] [blame] | 240 | |
| 241 | /** |
Peter Zijlstra | 7aa54be | 2018-09-26 13:01:20 +0200 | [diff] [blame] | 242 | * queued_fetch_set_pending_acquire - fetch the whole lock value and set pending |
| 243 | * @lock : Pointer to queued spinlock structure |
| 244 | * Return: The previous lock value |
| 245 | * |
| 246 | * *,*,* -> *,1,* |
| 247 | */ |
| 248 | #ifndef queued_fetch_set_pending_acquire |
| 249 | static __always_inline u32 queued_fetch_set_pending_acquire(struct qspinlock *lock) |
| 250 | { |
| 251 | return atomic_fetch_or_acquire(_Q_PENDING_VAL, &lock->val); |
| 252 | } |
| 253 | #endif |
| 254 | |
| 255 | /** |
Waiman Long | 2c83e8e | 2015-04-24 14:56:35 -0400 | [diff] [blame] | 256 | * set_locked - Set the lock bit and own the lock |
| 257 | * @lock: Pointer to queued spinlock structure |
| 258 | * |
| 259 | * *,*,0 -> *,0,1 |
| 260 | */ |
| 261 | static __always_inline void set_locked(struct qspinlock *lock) |
| 262 | { |
Will Deacon | 625e88b | 2018-04-26 11:34:16 +0100 | [diff] [blame] | 263 | WRITE_ONCE(lock->locked, _Q_LOCKED_VAL); |
Waiman Long | 2c83e8e | 2015-04-24 14:56:35 -0400 | [diff] [blame] | 264 | } |
| 265 | |
Waiman Long | a23db28 | 2015-04-24 14:56:37 -0400 | [diff] [blame] | 266 | |
| 267 | /* |
| 268 | * Generate the native code for queued_spin_unlock_slowpath(); provide NOPs for |
| 269 | * all the PV callbacks. |
| 270 | */ |
| 271 | |
| 272 | static __always_inline void __pv_init_node(struct mcs_spinlock *node) { } |
Waiman Long | cd0272f | 2015-11-09 19:09:27 -0500 | [diff] [blame] | 273 | static __always_inline void __pv_wait_node(struct mcs_spinlock *node, |
| 274 | struct mcs_spinlock *prev) { } |
Waiman Long | 75d2270 | 2015-07-11 16:36:52 -0400 | [diff] [blame] | 275 | static __always_inline void __pv_kick_node(struct qspinlock *lock, |
| 276 | struct mcs_spinlock *node) { } |
Waiman Long | 1c4941f | 2015-11-10 16:18:56 -0500 | [diff] [blame] | 277 | static __always_inline u32 __pv_wait_head_or_lock(struct qspinlock *lock, |
| 278 | struct mcs_spinlock *node) |
| 279 | { return 0; } |
Waiman Long | a23db28 | 2015-04-24 14:56:37 -0400 | [diff] [blame] | 280 | |
| 281 | #define pv_enabled() false |
| 282 | |
| 283 | #define pv_init_node __pv_init_node |
| 284 | #define pv_wait_node __pv_wait_node |
| 285 | #define pv_kick_node __pv_kick_node |
Waiman Long | 1c4941f | 2015-11-10 16:18:56 -0500 | [diff] [blame] | 286 | #define pv_wait_head_or_lock __pv_wait_head_or_lock |
Waiman Long | a23db28 | 2015-04-24 14:56:37 -0400 | [diff] [blame] | 287 | |
| 288 | #ifdef CONFIG_PARAVIRT_SPINLOCKS |
| 289 | #define queued_spin_lock_slowpath native_queued_spin_lock_slowpath |
| 290 | #endif |
| 291 | |
| 292 | #endif /* _GEN_PV_LOCK_SLOWPATH */ |
| 293 | |
Waiman Long | 2c83e8e | 2015-04-24 14:56:35 -0400 | [diff] [blame] | 294 | /** |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 295 | * queued_spin_lock_slowpath - acquire the queued spinlock |
| 296 | * @lock: Pointer to queued spinlock structure |
| 297 | * @val: Current value of the queued spinlock 32-bit word |
| 298 | * |
Peter Zijlstra (Intel) | c1fb159 | 2015-04-24 14:56:32 -0400 | [diff] [blame] | 299 | * (queue tail, pending bit, lock value) |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 300 | * |
Peter Zijlstra (Intel) | c1fb159 | 2015-04-24 14:56:32 -0400 | [diff] [blame] | 301 | * fast : slow : unlock |
| 302 | * : : |
| 303 | * uncontended (0,0,0) -:--> (0,0,1) ------------------------------:--> (*,*,0) |
| 304 | * : | ^--------.------. / : |
| 305 | * : v \ \ | : |
| 306 | * pending : (0,1,1) +--> (0,1,0) \ | : |
| 307 | * : | ^--' | | : |
| 308 | * : v | | : |
| 309 | * uncontended : (n,x,y) +--> (n,0,0) --' | : |
| 310 | * queue : | ^--' | : |
| 311 | * : v | : |
| 312 | * contended : (*,x,y) +--> (*,0,0) ---> (*,0,1) -' : |
| 313 | * queue : ^--' : |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 314 | */ |
| 315 | void queued_spin_lock_slowpath(struct qspinlock *lock, u32 val) |
| 316 | { |
| 317 | struct mcs_spinlock *prev, *next, *node; |
Will Deacon | 59fb586 | 2018-04-26 11:34:19 +0100 | [diff] [blame] | 318 | u32 old, tail; |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 319 | int idx; |
| 320 | |
| 321 | BUILD_BUG_ON(CONFIG_NR_CPUS >= (1U << _Q_TAIL_CPU_BITS)); |
| 322 | |
Waiman Long | a23db28 | 2015-04-24 14:56:37 -0400 | [diff] [blame] | 323 | if (pv_enabled()) |
Waiman Long | 81d3dc9 | 2018-04-26 11:34:27 +0100 | [diff] [blame] | 324 | goto pv_queue; |
Waiman Long | a23db28 | 2015-04-24 14:56:37 -0400 | [diff] [blame] | 325 | |
Peter Zijlstra | 43b3f02 | 2015-09-04 17:25:23 +0200 | [diff] [blame] | 326 | if (virt_spin_lock(lock)) |
Peter Zijlstra (Intel) | 2aa79af | 2015-04-24 14:56:36 -0400 | [diff] [blame] | 327 | return; |
| 328 | |
Peter Zijlstra (Intel) | c1fb159 | 2015-04-24 14:56:32 -0400 | [diff] [blame] | 329 | /* |
Will Deacon | 6512276 | 2018-04-26 11:34:17 +0100 | [diff] [blame] | 330 | * Wait for in-progress pending->locked hand-overs with a bounded |
| 331 | * number of spins so that we guarantee forward progress. |
Peter Zijlstra (Intel) | c1fb159 | 2015-04-24 14:56:32 -0400 | [diff] [blame] | 332 | * |
| 333 | * 0,1,0 -> 0,0,1 |
| 334 | */ |
| 335 | if (val == _Q_PENDING_VAL) { |
Will Deacon | 6512276 | 2018-04-26 11:34:17 +0100 | [diff] [blame] | 336 | int cnt = _Q_PENDING_LOOPS; |
| 337 | val = atomic_cond_read_relaxed(&lock->val, |
| 338 | (VAL != _Q_PENDING_VAL) || !cnt--); |
Peter Zijlstra (Intel) | c1fb159 | 2015-04-24 14:56:32 -0400 | [diff] [blame] | 339 | } |
| 340 | |
| 341 | /* |
Will Deacon | 59fb586 | 2018-04-26 11:34:19 +0100 | [diff] [blame] | 342 | * If we observe any contention; queue. |
| 343 | */ |
| 344 | if (val & ~_Q_LOCKED_MASK) |
| 345 | goto queue; |
| 346 | |
| 347 | /* |
Peter Zijlstra (Intel) | c1fb159 | 2015-04-24 14:56:32 -0400 | [diff] [blame] | 348 | * trylock || pending |
| 349 | * |
Peter Zijlstra | 756b1df | 2018-09-26 13:01:19 +0200 | [diff] [blame] | 350 | * 0,0,* -> 0,1,* -> 0,0,1 pending, trylock |
Peter Zijlstra (Intel) | c1fb159 | 2015-04-24 14:56:32 -0400 | [diff] [blame] | 351 | */ |
Peter Zijlstra | 7aa54be | 2018-09-26 13:01:20 +0200 | [diff] [blame] | 352 | val = queued_fetch_set_pending_acquire(lock); |
Peter Zijlstra | 756b1df | 2018-09-26 13:01:19 +0200 | [diff] [blame] | 353 | |
Peter Zijlstra | 53bf57f | 2018-09-26 13:01:18 +0200 | [diff] [blame] | 354 | /* |
Peter Zijlstra | 756b1df | 2018-09-26 13:01:19 +0200 | [diff] [blame] | 355 | * If we observe contention, there is a concurrent locker. |
| 356 | * |
| 357 | * Undo and queue; our setting of PENDING might have made the |
| 358 | * n,0,0 -> 0,0,0 transition fail and it will now be waiting |
| 359 | * on @next to become !NULL. |
Peter Zijlstra | 53bf57f | 2018-09-26 13:01:18 +0200 | [diff] [blame] | 360 | */ |
| 361 | if (unlikely(val & ~_Q_LOCKED_MASK)) { |
Peter Zijlstra | 756b1df | 2018-09-26 13:01:19 +0200 | [diff] [blame] | 362 | |
| 363 | /* Undo PENDING if we set it. */ |
Peter Zijlstra | 53bf57f | 2018-09-26 13:01:18 +0200 | [diff] [blame] | 364 | if (!(val & _Q_PENDING_MASK)) |
| 365 | clear_pending(lock); |
Peter Zijlstra | 756b1df | 2018-09-26 13:01:19 +0200 | [diff] [blame] | 366 | |
Peter Zijlstra | 53bf57f | 2018-09-26 13:01:18 +0200 | [diff] [blame] | 367 | goto queue; |
Peter Zijlstra (Intel) | c1fb159 | 2015-04-24 14:56:32 -0400 | [diff] [blame] | 368 | } |
| 369 | |
| 370 | /* |
Peter Zijlstra | 53bf57f | 2018-09-26 13:01:18 +0200 | [diff] [blame] | 371 | * We're pending, wait for the owner to go away. |
| 372 | * |
| 373 | * 0,1,1 -> 0,1,0 |
| 374 | * |
| 375 | * this wait loop must be a load-acquire such that we match the |
| 376 | * store-release that clears the locked bit and create lock |
| 377 | * sequentiality; this is because not all |
| 378 | * clear_pending_set_locked() implementations imply full |
| 379 | * barriers. |
Peter Zijlstra (Intel) | c1fb159 | 2015-04-24 14:56:32 -0400 | [diff] [blame] | 380 | */ |
Peter Zijlstra | 53bf57f | 2018-09-26 13:01:18 +0200 | [diff] [blame] | 381 | if (val & _Q_LOCKED_MASK) |
| 382 | atomic_cond_read_acquire(&lock->val, !(VAL & _Q_LOCKED_MASK)); |
| 383 | |
| 384 | /* |
| 385 | * take ownership and clear the pending bit. |
| 386 | * |
| 387 | * 0,1,0 -> 0,0,1 |
| 388 | */ |
| 389 | clear_pending_set_locked(lock); |
Waiman Long | ad53fa1 | 2019-04-04 13:43:16 -0400 | [diff] [blame] | 390 | lockevent_inc(lock_pending); |
Peter Zijlstra | 53bf57f | 2018-09-26 13:01:18 +0200 | [diff] [blame] | 391 | return; |
Peter Zijlstra (Intel) | c1fb159 | 2015-04-24 14:56:32 -0400 | [diff] [blame] | 392 | |
| 393 | /* |
| 394 | * End of pending bit optimistic spinning and beginning of MCS |
| 395 | * queuing. |
| 396 | */ |
| 397 | queue: |
Waiman Long | ad53fa1 | 2019-04-04 13:43:16 -0400 | [diff] [blame] | 398 | lockevent_inc(lock_slowpath); |
Waiman Long | 81d3dc9 | 2018-04-26 11:34:27 +0100 | [diff] [blame] | 399 | pv_queue: |
Waiman Long | 0fa809c | 2018-10-16 09:45:07 -0400 | [diff] [blame] | 400 | node = this_cpu_ptr(&qnodes[0].mcs); |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 401 | idx = node->count++; |
| 402 | tail = encode_tail(smp_processor_id(), idx); |
| 403 | |
Waiman Long | d682b59 | 2019-01-29 22:53:45 +0100 | [diff] [blame] | 404 | /* |
| 405 | * 4 nodes are allocated based on the assumption that there will |
| 406 | * not be nested NMIs taking spinlocks. That may not be true in |
| 407 | * some architectures even though the chance of needing more than |
| 408 | * 4 nodes will still be extremely unlikely. When that happens, |
| 409 | * we fall back to spinning on the lock directly without using |
| 410 | * any MCS node. This is not the most elegant solution, but is |
| 411 | * simple enough. |
| 412 | */ |
| 413 | if (unlikely(idx >= MAX_NODES)) { |
Waiman Long | ad53fa1 | 2019-04-04 13:43:16 -0400 | [diff] [blame] | 414 | lockevent_inc(lock_no_node); |
Waiman Long | d682b59 | 2019-01-29 22:53:45 +0100 | [diff] [blame] | 415 | while (!queued_spin_trylock(lock)) |
| 416 | cpu_relax(); |
| 417 | goto release; |
| 418 | } |
| 419 | |
Waiman Long | 0fa809c | 2018-10-16 09:45:07 -0400 | [diff] [blame] | 420 | node = grab_mcs_node(node, idx); |
Will Deacon | 11dc132 | 2018-02-13 13:22:57 +0000 | [diff] [blame] | 421 | |
| 422 | /* |
Waiman Long | 1222109 | 2018-10-16 09:45:06 -0400 | [diff] [blame] | 423 | * Keep counts of non-zero index values: |
| 424 | */ |
Waiman Long | ad53fa1 | 2019-04-04 13:43:16 -0400 | [diff] [blame] | 425 | lockevent_cond_inc(lock_use_node2 + idx - 1, idx); |
Waiman Long | 1222109 | 2018-10-16 09:45:06 -0400 | [diff] [blame] | 426 | |
| 427 | /* |
Will Deacon | 11dc132 | 2018-02-13 13:22:57 +0000 | [diff] [blame] | 428 | * Ensure that we increment the head node->count before initialising |
| 429 | * the actual node. If the compiler is kind enough to reorder these |
| 430 | * stores, then an IRQ could overwrite our assignments. |
| 431 | */ |
| 432 | barrier(); |
| 433 | |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 434 | node->locked = 0; |
| 435 | node->next = NULL; |
Waiman Long | a23db28 | 2015-04-24 14:56:37 -0400 | [diff] [blame] | 436 | pv_init_node(node); |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 437 | |
| 438 | /* |
Waiman Long | 6403bd7 | 2015-04-24 14:56:33 -0400 | [diff] [blame] | 439 | * We touched a (possibly) cold cacheline in the per-cpu queue node; |
| 440 | * attempt the trylock once more in the hope someone let go while we |
| 441 | * weren't watching. |
| 442 | */ |
| 443 | if (queued_spin_trylock(lock)) |
| 444 | goto release; |
| 445 | |
| 446 | /* |
Will Deacon | 9d4646d | 2018-04-26 11:34:25 +0100 | [diff] [blame] | 447 | * Ensure that the initialisation of @node is complete before we |
| 448 | * publish the updated tail via xchg_tail() and potentially link |
| 449 | * @node into the waitqueue via WRITE_ONCE(prev->next, node) below. |
| 450 | */ |
| 451 | smp_wmb(); |
| 452 | |
| 453 | /* |
| 454 | * Publish the updated tail. |
Peter Zijlstra (Intel) | c1fb159 | 2015-04-24 14:56:32 -0400 | [diff] [blame] | 455 | * We have already touched the queueing cacheline; don't bother with |
| 456 | * pending stuff. |
| 457 | * |
Waiman Long | 6403bd7 | 2015-04-24 14:56:33 -0400 | [diff] [blame] | 458 | * p,*,* -> n,*,* |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 459 | */ |
Waiman Long | 6403bd7 | 2015-04-24 14:56:33 -0400 | [diff] [blame] | 460 | old = xchg_tail(lock, tail); |
Waiman Long | aa68744 | 2015-11-09 19:09:23 -0500 | [diff] [blame] | 461 | next = NULL; |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 462 | |
| 463 | /* |
| 464 | * if there was a previous node; link it and wait until reaching the |
| 465 | * head of the waitqueue. |
| 466 | */ |
Waiman Long | 6403bd7 | 2015-04-24 14:56:33 -0400 | [diff] [blame] | 467 | if (old & _Q_TAIL_MASK) { |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 468 | prev = decode_tail(old); |
Peter Zijlstra | 8d53fa1 | 2016-06-08 09:12:30 +0200 | [diff] [blame] | 469 | |
Will Deacon | 9d4646d | 2018-04-26 11:34:25 +0100 | [diff] [blame] | 470 | /* Link @node into the waitqueue. */ |
| 471 | WRITE_ONCE(prev->next, node); |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 472 | |
Waiman Long | cd0272f | 2015-11-09 19:09:27 -0500 | [diff] [blame] | 473 | pv_wait_node(node, prev); |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 474 | arch_mcs_spin_lock_contended(&node->locked); |
Waiman Long | 81b5598 | 2015-11-09 19:09:22 -0500 | [diff] [blame] | 475 | |
| 476 | /* |
| 477 | * While waiting for the MCS lock, the next pointer may have |
| 478 | * been set by another lock waiter. We optimistically load |
| 479 | * the next pointer & prefetch the cacheline for writing |
| 480 | * to reduce latency in the upcoming MCS unlock operation. |
| 481 | */ |
| 482 | next = READ_ONCE(node->next); |
| 483 | if (next) |
| 484 | prefetchw(next); |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 485 | } |
| 486 | |
| 487 | /* |
Peter Zijlstra (Intel) | c1fb159 | 2015-04-24 14:56:32 -0400 | [diff] [blame] | 488 | * we're at the head of the waitqueue, wait for the owner & pending to |
| 489 | * go away. |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 490 | * |
Peter Zijlstra (Intel) | c1fb159 | 2015-04-24 14:56:32 -0400 | [diff] [blame] | 491 | * *,x,y -> *,0,0 |
Waiman Long | 2c83e8e | 2015-04-24 14:56:35 -0400 | [diff] [blame] | 492 | * |
| 493 | * this wait loop must use a load-acquire such that we match the |
| 494 | * store-release that clears the locked bit and create lock |
| 495 | * sequentiality; this is because the set_locked() function below |
| 496 | * does not imply a full barrier. |
| 497 | * |
Waiman Long | 1c4941f | 2015-11-10 16:18:56 -0500 | [diff] [blame] | 498 | * The PV pv_wait_head_or_lock function, if active, will acquire |
| 499 | * the lock and return a non-zero value. So we have to skip the |
Will Deacon | f9c811fa | 2018-04-26 11:34:21 +0100 | [diff] [blame] | 500 | * atomic_cond_read_acquire() call. As the next PV queue head hasn't |
| 501 | * been designated yet, there is no way for the locked value to become |
Waiman Long | 1c4941f | 2015-11-10 16:18:56 -0500 | [diff] [blame] | 502 | * _Q_SLOW_VAL. So both the set_locked() and the |
| 503 | * atomic_cmpxchg_relaxed() calls will be safe. |
| 504 | * |
| 505 | * If PV isn't active, 0 will be returned instead. |
| 506 | * |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 507 | */ |
Waiman Long | 1c4941f | 2015-11-10 16:18:56 -0500 | [diff] [blame] | 508 | if ((val = pv_wait_head_or_lock(lock, node))) |
| 509 | goto locked; |
| 510 | |
Will Deacon | f9c811fa | 2018-04-26 11:34:21 +0100 | [diff] [blame] | 511 | val = atomic_cond_read_acquire(&lock->val, !(VAL & _Q_LOCKED_PENDING_MASK)); |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 512 | |
Waiman Long | 1c4941f | 2015-11-10 16:18:56 -0500 | [diff] [blame] | 513 | locked: |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 514 | /* |
| 515 | * claim the lock: |
| 516 | * |
Peter Zijlstra (Intel) | c1fb159 | 2015-04-24 14:56:32 -0400 | [diff] [blame] | 517 | * n,0,0 -> 0,0,1 : lock, uncontended |
Will Deacon | 59fb586 | 2018-04-26 11:34:19 +0100 | [diff] [blame] | 518 | * *,*,0 -> *,*,1 : lock, contended |
Waiman Long | 2c83e8e | 2015-04-24 14:56:35 -0400 | [diff] [blame] | 519 | * |
Will Deacon | 59fb586 | 2018-04-26 11:34:19 +0100 | [diff] [blame] | 520 | * If the queue head is the only one in the queue (lock value == tail) |
| 521 | * and nobody is pending, clear the tail code and grab the lock. |
| 522 | * Otherwise, we only need to grab the lock. |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 523 | */ |
Will Deacon | c61da58 | 2018-04-26 11:34:20 +0100 | [diff] [blame] | 524 | |
Will Deacon | ae75d90 | 2018-04-26 11:34:26 +0100 | [diff] [blame] | 525 | /* |
Peter Zijlstra | 756b1df | 2018-09-26 13:01:19 +0200 | [diff] [blame] | 526 | * In the PV case we might already have _Q_LOCKED_VAL set, because |
| 527 | * of lock stealing; therefore we must also allow: |
Will Deacon | ae75d90 | 2018-04-26 11:34:26 +0100 | [diff] [blame] | 528 | * |
Peter Zijlstra | 756b1df | 2018-09-26 13:01:19 +0200 | [diff] [blame] | 529 | * n,0,1 -> 0,0,1 |
| 530 | * |
| 531 | * Note: at this point: (val & _Q_PENDING_MASK) == 0, because of the |
| 532 | * above wait condition, therefore any concurrent setting of |
| 533 | * PENDING will make the uncontended transition fail. |
Will Deacon | ae75d90 | 2018-04-26 11:34:26 +0100 | [diff] [blame] | 534 | */ |
Peter Zijlstra | 756b1df | 2018-09-26 13:01:19 +0200 | [diff] [blame] | 535 | if ((val & _Q_TAIL_MASK) == tail) { |
| 536 | if (atomic_try_cmpxchg_relaxed(&lock->val, &val, _Q_LOCKED_VAL)) |
| 537 | goto release; /* No contention */ |
| 538 | } |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 539 | |
Peter Zijlstra | 756b1df | 2018-09-26 13:01:19 +0200 | [diff] [blame] | 540 | /* |
| 541 | * Either somebody is queued behind us or _Q_PENDING_VAL got set |
| 542 | * which will then detect the remaining tail and queue behind us |
| 543 | * ensuring we'll see a @next. |
| 544 | */ |
Will Deacon | c61da58 | 2018-04-26 11:34:20 +0100 | [diff] [blame] | 545 | set_locked(lock); |
| 546 | |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 547 | /* |
Waiman Long | aa68744 | 2015-11-09 19:09:23 -0500 | [diff] [blame] | 548 | * contended path; wait for next if not observed yet, release. |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 549 | */ |
Will Deacon | c131a19 | 2018-04-26 11:34:23 +0100 | [diff] [blame] | 550 | if (!next) |
| 551 | next = smp_cond_load_relaxed(&node->next, (VAL)); |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 552 | |
Waiman Long | 2c83e8e | 2015-04-24 14:56:35 -0400 | [diff] [blame] | 553 | arch_mcs_spin_unlock_contended(&next->locked); |
Waiman Long | 75d2270 | 2015-07-11 16:36:52 -0400 | [diff] [blame] | 554 | pv_kick_node(lock, next); |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 555 | |
| 556 | release: |
| 557 | /* |
| 558 | * release the node |
| 559 | */ |
Waiman Long | 0fa809c | 2018-10-16 09:45:07 -0400 | [diff] [blame] | 560 | __this_cpu_dec(qnodes[0].mcs.count); |
Waiman Long | a33fda3 | 2015-04-24 14:56:30 -0400 | [diff] [blame] | 561 | } |
| 562 | EXPORT_SYMBOL(queued_spin_lock_slowpath); |
Waiman Long | a23db28 | 2015-04-24 14:56:37 -0400 | [diff] [blame] | 563 | |
| 564 | /* |
| 565 | * Generate the paravirt code for queued_spin_unlock_slowpath(). |
| 566 | */ |
| 567 | #if !defined(_GEN_PV_LOCK_SLOWPATH) && defined(CONFIG_PARAVIRT_SPINLOCKS) |
| 568 | #define _GEN_PV_LOCK_SLOWPATH |
| 569 | |
| 570 | #undef pv_enabled |
| 571 | #define pv_enabled() true |
| 572 | |
| 573 | #undef pv_init_node |
| 574 | #undef pv_wait_node |
| 575 | #undef pv_kick_node |
Waiman Long | 1c4941f | 2015-11-10 16:18:56 -0500 | [diff] [blame] | 576 | #undef pv_wait_head_or_lock |
Waiman Long | a23db28 | 2015-04-24 14:56:37 -0400 | [diff] [blame] | 577 | |
| 578 | #undef queued_spin_lock_slowpath |
| 579 | #define queued_spin_lock_slowpath __pv_queued_spin_lock_slowpath |
| 580 | |
| 581 | #include "qspinlock_paravirt.h" |
| 582 | #include "qspinlock.c" |
| 583 | |
Zhenzhong Duan | 05eee61 | 2019-10-23 19:16:22 +0800 | [diff] [blame] | 584 | bool nopvspin __initdata; |
| 585 | static __init int parse_nopvspin(char *arg) |
| 586 | { |
| 587 | nopvspin = true; |
| 588 | return 0; |
| 589 | } |
| 590 | early_param("nopvspin", parse_nopvspin); |
Waiman Long | a23db28 | 2015-04-24 14:56:37 -0400 | [diff] [blame] | 591 | #endif |