Documentation/hwspinlock.txt

   1 Hardware Spinlock Framework
   2
   3 1. Introduction
   4
   5 Hardware spinlock modules provide hardware assistance for synchronization
   6 and mutual exclusion between heterogeneous processors and those not operating
   7 under a single, shared operating system.
   8
   9 For example, OMAP4 has dual Cortex-A9, dual Cortex-M3 and a C64x+ DSP,
  10 each of which is running a different Operating System (the master, A9,
  11 is usually running Linux and the slave processors, the M3 and the DSP,
  12 are running some flavor of RTOS).
  13
  14 A generic hwspinlock framework allows platform-independent drivers to use
  15 the hwspinlock device in order to access data structures that are shared
  16 between remote processors, that otherwise have no alternative mechanism
  17 to accomplish synchronization and mutual exclusion operations.
  18
  19 This is necessary, for example, for Inter-processor communications:
  20 on OMAP4, cpu-intensive multimedia tasks are offloaded by the host to the
  21 remote M3 and/or C64x+ slave processors (by an IPC subsystem called Syslink).
  22
  23 To achieve fast message-based communications, a minimal kernel support
  24 is needed to deliver messages arriving from a remote processor to the
  25 appropriate user process.
  26
  27 This communication is based on simple data structures that is shared between
  28 the remote processors, and access to it is synchronized using the hwspinlock
  29 module (remote processor directly places new messages in this shared data
  30 structure).
  31
  32 A common hwspinlock interface makes it possible to have generic, platform-
  33 independent, drivers.
  34
  35 2. User API
  36
  37   struct hwspinlock *hwspin_lock_request(void);
  38    - dynamically assign an hwspinlock and return its address, or NULL
  39      in case an unused hwspinlock isn't available. Users of this
  40      API will usually want to communicate the lock's id to the remote core
  41      before it can be used to achieve synchronization.
  42      Should be called from a process context (might sleep).
  43
  44   struct hwspinlock *hwspin_lock_request_specific(unsigned int id);
  45    - assign a specific hwspinlock id and return its address, or NULL
  46      if that hwspinlock is already in use. Usually board code will
  47      be calling this function in order to reserve specific hwspinlock
  48      ids for predefined purposes.
  49      Should be called from a process context (might sleep).
  50
  51   int hwspin_lock_free(struct hwspinlock *hwlock);
  52    - free a previously-assigned hwspinlock; returns 0 on success, or an
  53      appropriate error code on failure (e.g. -EINVAL if the hwspinlock
  54      is already free).
  55      Should be called from a process context (might sleep).
  56
  57   int hwspin_lock_timeout(struct hwspinlock *hwlock, unsigned int timeout);
  58    - lock a previously-assigned hwspinlock with a timeout limit (specified in
  59      msecs). If the hwspinlock is already taken, the function will busy loop
  60      waiting for it to be released, but give up when the timeout elapses.
  61      Upon a successful return from this function, preemption is disabled so
  62      the caller must not sleep, and is advised to release the hwspinlock as
  63      soon as possible, in order to minimize remote cores polling on the
  64      hardware interconnect.
  65      Returns 0 when successful and an appropriate error code otherwise (most
  66      notably -ETIMEDOUT if the hwspinlock is still busy after timeout msecs).
  67      The function will never sleep.
  68
  69   int hwspin_lock_timeout_irq(struct hwspinlock *hwlock, unsigned int timeout);
  70    - lock a previously-assigned hwspinlock with a timeout limit (specified in
  71      msecs). If the hwspinlock is already taken, the function will busy loop
  72      waiting for it to be released, but give up when the timeout elapses.
  73      Upon a successful return from this function, preemption and the local
  74      interrupts are disabled, so the caller must not sleep, and is advised to
  75      release the hwspinlock as soon as possible.
  76      Returns 0 when successful and an appropriate error code otherwise (most
  77      notably -ETIMEDOUT if the hwspinlock is still busy after timeout msecs).
  78      The function will never sleep.
  79
  80   int hwspin_lock_timeout_irqsave(struct hwspinlock *hwlock, unsigned int to,
  81                                                         unsigned long *flags);
  82    - lock a previously-assigned hwspinlock with a timeout limit (specified in
  83      msecs). If the hwspinlock is already taken, the function will busy loop
  84      waiting for it to be released, but give up when the timeout elapses.
  85      Upon a successful return from this function, preemption is disabled,
  86      local interrupts are disabled and their previous state is saved at the
  87      given flags placeholder. The caller must not sleep, and is advised to
  88      release the hwspinlock as soon as possible.
  89      Returns 0 when successful and an appropriate error code otherwise (most
  90      notably -ETIMEDOUT if the hwspinlock is still busy after timeout msecs).
  91      The function will never sleep.
  92
  93   int hwspin_trylock(struct hwspinlock *hwlock);
  94    - attempt to lock a previously-assigned hwspinlock, but immediately fail if
  95      it is already taken.
  96      Upon a successful return from this function, preemption is disabled so
  97      caller must not sleep, and is advised to release the hwspinlock as soon as
  98      possible, in order to minimize remote cores polling on the hardware
  99      interconnect.
 100      Returns 0 on success and an appropriate error code otherwise (most
 101      notably -EBUSY if the hwspinlock was already taken).
 102      The function will never sleep.
 103
 104   int hwspin_trylock_irq(struct hwspinlock *hwlock);
 105    - attempt to lock a previously-assigned hwspinlock, but immediately fail if
 106      it is already taken.
 107      Upon a successful return from this function, preemption and the local
 108      interrupts are disabled so caller must not sleep, and is advised to
 109      release the hwspinlock as soon as possible.
 110      Returns 0 on success and an appropriate error code otherwise (most
 111      notably -EBUSY if the hwspinlock was already taken).
 112      The function will never sleep.
 113
 114   int hwspin_trylock_irqsave(struct hwspinlock *hwlock, unsigned long *flags);
 115    - attempt to lock a previously-assigned hwspinlock, but immediately fail if
 116      it is already taken.
 117      Upon a successful return from this function, preemption is disabled,
 118      the local interrupts are disabled and their previous state is saved
 119      at the given flags placeholder. The caller must not sleep, and is advised
 120      to release the hwspinlock as soon as possible.
 121      Returns 0 on success and an appropriate error code otherwise (most
 122      notably -EBUSY if the hwspinlock was already taken).
 123      The function will never sleep.
 124
 125   void hwspin_unlock(struct hwspinlock *hwlock);
 126    - unlock a previously-locked hwspinlock. Always succeed, and can be called
 127      from any context (the function never sleeps). Note: code should _never_
 128      unlock an hwspinlock which is already unlocked (there is no protection
 129      against this).
 130
 131   void hwspin_unlock_irq(struct hwspinlock *hwlock);
 132    - unlock a previously-locked hwspinlock and enable local interrupts.
 133      The caller should _never_ unlock an hwspinlock which is already unlocked.
 134      Doing so is considered a bug (there is no protection against this).
 135      Upon a successful return from this function, preemption and local
 136      interrupts are enabled. This function will never sleep.
 137
 138   void
 139   hwspin_unlock_irqrestore(struct hwspinlock *hwlock, unsigned long *flags);
 140    - unlock a previously-locked hwspinlock.
 141      The caller should _never_ unlock an hwspinlock which is already unlocked.
 142      Doing so is considered a bug (there is no protection against this).
 143      Upon a successful return from this function, preemption is reenabled,
 144      and the state of the local interrupts is restored to the state saved at
 145      the given flags. This function will never sleep.
 146
 147   int hwspin_lock_get_id(struct hwspinlock *hwlock);
 148    - retrieve id number of a given hwspinlock. This is needed when an
 149      hwspinlock is dynamically assigned: before it can be used to achieve
 150      mutual exclusion with a remote cpu, the id number should be communicated
 151      to the remote task with which we want to synchronize.
 152      Returns the hwspinlock id number, or -EINVAL if hwlock is null.
 153
 154 3. Typical usage
 155
 156 #include <linux/hwspinlock.h>
 157 #include <linux/err.h>
 158
 159 int hwspinlock_example1(void)
 160 {
 161         struct hwspinlock *hwlock;
 162         int ret;
 163
 164         /* dynamically assign a hwspinlock */
 165         hwlock = hwspin_lock_request();
 166         if (!hwlock)
 167                 ...
 168
 169         id = hwspin_lock_get_id(hwlock);
 170         /* probably need to communicate id to a remote processor now */
 171
 172         /* take the lock, spin for 1 sec if it's already taken */
 173         ret = hwspin_lock_timeout(hwlock, 1000);
 174         if (ret)
 175                 ...
 176
 177         /*
 178          * we took the lock, do our thing now, but do NOT sleep
 179          */
 180
 181         /* release the lock */
 182         hwspin_unlock(hwlock);
 183
 184         /* free the lock */
 185         ret = hwspin_lock_free(hwlock);
 186         if (ret)
 187                 ...
 188
 189         return ret;
 190 }
 191
 192 int hwspinlock_example2(void)
 193 {
 194         struct hwspinlock *hwlock;
 195         int ret;
 196
 197         /*
 198          * assign a specific hwspinlock id - this should be called early
 199          * by board init code.
 200          */
 201         hwlock = hwspin_lock_request_specific(PREDEFINED_LOCK_ID);
 202         if (!hwlock)
 203                 ...
 204
 205         /* try to take it, but don't spin on it */
 206         ret = hwspin_trylock(hwlock);
 207         if (!ret) {
 208                 pr_info("lock is already taken\n");
 209                 return -EBUSY;
 210         }
 211
 212         /*
 213          * we took the lock, do our thing now, but do NOT sleep
 214          */
 215
 216         /* release the lock */
 217         hwspin_unlock(hwlock);
 218
 219         /* free the lock */
 220         ret = hwspin_lock_free(hwlock);
 221         if (ret)
 222                 ...
 223
 224         return ret;
 225 }
 226
 227
 228 4. API for implementors
 229
 230   int hwspin_lock_register(struct hwspinlock_device *bank, struct device *dev,
 231                 const struct hwspinlock_ops *ops, int base_id, int num_locks);
 232    - to be called from the underlying platform-specific implementation, in
 233      order to register a new hwspinlock device (which is usually a bank of
 234      numerous locks). Should be called from a process context (this function
 235      might sleep).
 236      Returns 0 on success, or appropriate error code on failure.
 237
 238   int hwspin_lock_unregister(struct hwspinlock_device *bank);
 239    - to be called from the underlying vendor-specific implementation, in order
 240      to unregister an hwspinlock device (which is usually a bank of numerous
 241      locks).
 242      Should be called from a process context (this function might sleep).
 243      Returns the address of hwspinlock on success, or NULL on error (e.g.
 244      if the hwspinlock is sill in use).
 245
 246 5. Important structs
 247
 248 struct hwspinlock_device is a device which usually contains a bank
 249 of hardware locks. It is registered by the underlying hwspinlock
 250 implementation using the hwspin_lock_register() API.
 251
 252 /**
 253  * struct hwspinlock_device - a device which usually spans numerous hwspinlocks
 254  * @dev: underlying device, will be used to invoke runtime PM api
 255  * @ops: platform-specific hwspinlock handlers
 256  * @base_id: id index of the first lock in this device
 257  * @num_locks: number of locks in this device
 258  * @lock: dynamically allocated array of 'struct hwspinlock'
 259  */
 260 struct hwspinlock_device {
 261         struct device *dev;
 262         const struct hwspinlock_ops *ops;
 263         int base_id;
 264         int num_locks;
 265         struct hwspinlock lock[0];
 266 };
 267
 268 struct hwspinlock_device contains an array of hwspinlock structs, each
 269 of which represents a single hardware lock:
 270
 271 /**
 272  * struct hwspinlock - this struct represents a single hwspinlock instance
 273  * @bank: the hwspinlock_device structure which owns this lock
 274  * @lock: initialized and used by hwspinlock core
 275  * @priv: private data, owned by the underlying platform-specific hwspinlock drv
 276  */
 277 struct hwspinlock {
 278         struct hwspinlock_device *bank;
 279         spinlock_t lock;
 280         void *priv;
 281 };
 282
 283 When registering a bank of locks, the hwspinlock driver only needs to
 284 set the priv members of the locks. The rest of the members are set and
 285 initialized by the hwspinlock core itself.
 286
 287 6. Implementation callbacks
 288
 289 There are three possible callbacks defined in 'struct hwspinlock_ops':
 290
 291 struct hwspinlock_ops {
 292         int (*trylock)(struct hwspinlock *lock);
 293         void (*unlock)(struct hwspinlock *lock);
 294         void (*relax)(struct hwspinlock *lock);
 295 };
 296
 297 The first two callbacks are mandatory:
 298
 299 The ->trylock() callback should make a single attempt to take the lock, and
 300 return 0 on failure and 1 on success. This callback may _not_ sleep.
 301
 302 The ->unlock() callback releases the lock. It always succeed, and it, too,
 303 may _not_ sleep.
 304
 305 The ->relax() callback is optional. It is called by hwspinlock core while
 306 spinning on a lock, and can be used by the underlying implementation to force
 307 a delay between two successive invocations of ->trylock(). It may _not_ sleep.