Kyle Swenson | 8d8f654 | 2021-03-15 11:02:55 -0600 | [diff] [blame] | 1 | Wait/Wound Deadlock-Proof Mutex Design |
| 2 | ====================================== |
| 3 | |
| 4 | Please read mutex-design.txt first, as it applies to wait/wound mutexes too. |
| 5 | |
| 6 | Motivation for WW-Mutexes |
| 7 | ------------------------- |
| 8 | |
| 9 | GPU's do operations that commonly involve many buffers. Those buffers |
| 10 | can be shared across contexts/processes, exist in different memory |
| 11 | domains (for example VRAM vs system memory), and so on. And with |
| 12 | PRIME / dmabuf, they can even be shared across devices. So there are |
| 13 | a handful of situations where the driver needs to wait for buffers to |
| 14 | become ready. If you think about this in terms of waiting on a buffer |
| 15 | mutex for it to become available, this presents a problem because |
| 16 | there is no way to guarantee that buffers appear in a execbuf/batch in |
| 17 | the same order in all contexts. That is directly under control of |
| 18 | userspace, and a result of the sequence of GL calls that an application |
| 19 | makes. Which results in the potential for deadlock. The problem gets |
| 20 | more complex when you consider that the kernel may need to migrate the |
| 21 | buffer(s) into VRAM before the GPU operates on the buffer(s), which |
| 22 | may in turn require evicting some other buffers (and you don't want to |
| 23 | evict other buffers which are already queued up to the GPU), but for a |
| 24 | simplified understanding of the problem you can ignore this. |
| 25 | |
| 26 | The algorithm that the TTM graphics subsystem came up with for dealing with |
| 27 | this problem is quite simple. For each group of buffers (execbuf) that need |
| 28 | to be locked, the caller would be assigned a unique reservation id/ticket, |
| 29 | from a global counter. In case of deadlock while locking all the buffers |
| 30 | associated with a execbuf, the one with the lowest reservation ticket (i.e. |
| 31 | the oldest task) wins, and the one with the higher reservation id (i.e. the |
| 32 | younger task) unlocks all of the buffers that it has already locked, and then |
| 33 | tries again. |
| 34 | |
| 35 | In the RDBMS literature this deadlock handling approach is called wait/wound: |
| 36 | The older tasks waits until it can acquire the contended lock. The younger tasks |
| 37 | needs to back off and drop all the locks it is currently holding, i.e. the |
| 38 | younger task is wounded. |
| 39 | |
| 40 | Concepts |
| 41 | -------- |
| 42 | |
| 43 | Compared to normal mutexes two additional concepts/objects show up in the lock |
| 44 | interface for w/w mutexes: |
| 45 | |
| 46 | Acquire context: To ensure eventual forward progress it is important the a task |
| 47 | trying to acquire locks doesn't grab a new reservation id, but keeps the one it |
| 48 | acquired when starting the lock acquisition. This ticket is stored in the |
| 49 | acquire context. Furthermore the acquire context keeps track of debugging state |
| 50 | to catch w/w mutex interface abuse. |
| 51 | |
| 52 | W/w class: In contrast to normal mutexes the lock class needs to be explicit for |
| 53 | w/w mutexes, since it is required to initialize the acquire context. |
| 54 | |
| 55 | Furthermore there are three different class of w/w lock acquire functions: |
| 56 | |
| 57 | * Normal lock acquisition with a context, using ww_mutex_lock. |
| 58 | |
| 59 | * Slowpath lock acquisition on the contending lock, used by the wounded task |
| 60 | after having dropped all already acquired locks. These functions have the |
| 61 | _slow postfix. |
| 62 | |
| 63 | From a simple semantics point-of-view the _slow functions are not strictly |
| 64 | required, since simply calling the normal ww_mutex_lock functions on the |
| 65 | contending lock (after having dropped all other already acquired locks) will |
| 66 | work correctly. After all if no other ww mutex has been acquired yet there's |
| 67 | no deadlock potential and hence the ww_mutex_lock call will block and not |
| 68 | prematurely return -EDEADLK. The advantage of the _slow functions is in |
| 69 | interface safety: |
| 70 | - ww_mutex_lock has a __must_check int return type, whereas ww_mutex_lock_slow |
| 71 | has a void return type. Note that since ww mutex code needs loops/retries |
| 72 | anyway the __must_check doesn't result in spurious warnings, even though the |
| 73 | very first lock operation can never fail. |
| 74 | - When full debugging is enabled ww_mutex_lock_slow checks that all acquired |
| 75 | ww mutex have been released (preventing deadlocks) and makes sure that we |
| 76 | block on the contending lock (preventing spinning through the -EDEADLK |
| 77 | slowpath until the contended lock can be acquired). |
| 78 | |
| 79 | * Functions to only acquire a single w/w mutex, which results in the exact same |
| 80 | semantics as a normal mutex. This is done by calling ww_mutex_lock with a NULL |
| 81 | context. |
| 82 | |
| 83 | Again this is not strictly required. But often you only want to acquire a |
| 84 | single lock in which case it's pointless to set up an acquire context (and so |
| 85 | better to avoid grabbing a deadlock avoidance ticket). |
| 86 | |
| 87 | Of course, all the usual variants for handling wake-ups due to signals are also |
| 88 | provided. |
| 89 | |
| 90 | Usage |
| 91 | ----- |
| 92 | |
| 93 | Three different ways to acquire locks within the same w/w class. Common |
| 94 | definitions for methods #1 and #2: |
| 95 | |
| 96 | static DEFINE_WW_CLASS(ww_class); |
| 97 | |
| 98 | struct obj { |
| 99 | struct ww_mutex lock; |
| 100 | /* obj data */ |
| 101 | }; |
| 102 | |
| 103 | struct obj_entry { |
| 104 | struct list_head head; |
| 105 | struct obj *obj; |
| 106 | }; |
| 107 | |
| 108 | Method 1, using a list in execbuf->buffers that's not allowed to be reordered. |
| 109 | This is useful if a list of required objects is already tracked somewhere. |
| 110 | Furthermore the lock helper can use propagate the -EALREADY return code back to |
| 111 | the caller as a signal that an object is twice on the list. This is useful if |
| 112 | the list is constructed from userspace input and the ABI requires userspace to |
| 113 | not have duplicate entries (e.g. for a gpu commandbuffer submission ioctl). |
| 114 | |
| 115 | int lock_objs(struct list_head *list, struct ww_acquire_ctx *ctx) |
| 116 | { |
| 117 | struct obj *res_obj = NULL; |
| 118 | struct obj_entry *contended_entry = NULL; |
| 119 | struct obj_entry *entry; |
| 120 | |
| 121 | ww_acquire_init(ctx, &ww_class); |
| 122 | |
| 123 | retry: |
| 124 | list_for_each_entry (entry, list, head) { |
| 125 | if (entry->obj == res_obj) { |
| 126 | res_obj = NULL; |
| 127 | continue; |
| 128 | } |
| 129 | ret = ww_mutex_lock(&entry->obj->lock, ctx); |
| 130 | if (ret < 0) { |
| 131 | contended_entry = entry; |
| 132 | goto err; |
| 133 | } |
| 134 | } |
| 135 | |
| 136 | ww_acquire_done(ctx); |
| 137 | return 0; |
| 138 | |
| 139 | err: |
| 140 | list_for_each_entry_continue_reverse (entry, list, head) |
| 141 | ww_mutex_unlock(&entry->obj->lock); |
| 142 | |
| 143 | if (res_obj) |
| 144 | ww_mutex_unlock(&res_obj->lock); |
| 145 | |
| 146 | if (ret == -EDEADLK) { |
| 147 | /* we lost out in a seqno race, lock and retry.. */ |
| 148 | ww_mutex_lock_slow(&contended_entry->obj->lock, ctx); |
| 149 | res_obj = contended_entry->obj; |
| 150 | goto retry; |
| 151 | } |
| 152 | ww_acquire_fini(ctx); |
| 153 | |
| 154 | return ret; |
| 155 | } |
| 156 | |
| 157 | Method 2, using a list in execbuf->buffers that can be reordered. Same semantics |
| 158 | of duplicate entry detection using -EALREADY as method 1 above. But the |
| 159 | list-reordering allows for a bit more idiomatic code. |
| 160 | |
| 161 | int lock_objs(struct list_head *list, struct ww_acquire_ctx *ctx) |
| 162 | { |
| 163 | struct obj_entry *entry, *entry2; |
| 164 | |
| 165 | ww_acquire_init(ctx, &ww_class); |
| 166 | |
| 167 | list_for_each_entry (entry, list, head) { |
| 168 | ret = ww_mutex_lock(&entry->obj->lock, ctx); |
| 169 | if (ret < 0) { |
| 170 | entry2 = entry; |
| 171 | |
| 172 | list_for_each_entry_continue_reverse (entry2, list, head) |
| 173 | ww_mutex_unlock(&entry2->obj->lock); |
| 174 | |
| 175 | if (ret != -EDEADLK) { |
| 176 | ww_acquire_fini(ctx); |
| 177 | return ret; |
| 178 | } |
| 179 | |
| 180 | /* we lost out in a seqno race, lock and retry.. */ |
| 181 | ww_mutex_lock_slow(&entry->obj->lock, ctx); |
| 182 | |
| 183 | /* |
| 184 | * Move buf to head of the list, this will point |
| 185 | * buf->next to the first unlocked entry, |
| 186 | * restarting the for loop. |
| 187 | */ |
| 188 | list_del(&entry->head); |
| 189 | list_add(&entry->head, list); |
| 190 | } |
| 191 | } |
| 192 | |
| 193 | ww_acquire_done(ctx); |
| 194 | return 0; |
| 195 | } |
| 196 | |
| 197 | Unlocking works the same way for both methods #1 and #2: |
| 198 | |
| 199 | void unlock_objs(struct list_head *list, struct ww_acquire_ctx *ctx) |
| 200 | { |
| 201 | struct obj_entry *entry; |
| 202 | |
| 203 | list_for_each_entry (entry, list, head) |
| 204 | ww_mutex_unlock(&entry->obj->lock); |
| 205 | |
| 206 | ww_acquire_fini(ctx); |
| 207 | } |
| 208 | |
| 209 | Method 3 is useful if the list of objects is constructed ad-hoc and not upfront, |
| 210 | e.g. when adjusting edges in a graph where each node has its own ww_mutex lock, |
| 211 | and edges can only be changed when holding the locks of all involved nodes. w/w |
| 212 | mutexes are a natural fit for such a case for two reasons: |
| 213 | - They can handle lock-acquisition in any order which allows us to start walking |
| 214 | a graph from a starting point and then iteratively discovering new edges and |
| 215 | locking down the nodes those edges connect to. |
| 216 | - Due to the -EALREADY return code signalling that a given objects is already |
| 217 | held there's no need for additional book-keeping to break cycles in the graph |
| 218 | or keep track off which looks are already held (when using more than one node |
| 219 | as a starting point). |
| 220 | |
| 221 | Note that this approach differs in two important ways from the above methods: |
| 222 | - Since the list of objects is dynamically constructed (and might very well be |
| 223 | different when retrying due to hitting the -EDEADLK wound condition) there's |
| 224 | no need to keep any object on a persistent list when it's not locked. We can |
| 225 | therefore move the list_head into the object itself. |
| 226 | - On the other hand the dynamic object list construction also means that the -EALREADY return |
| 227 | code can't be propagated. |
| 228 | |
| 229 | Note also that methods #1 and #2 and method #3 can be combined, e.g. to first lock a |
| 230 | list of starting nodes (passed in from userspace) using one of the above |
| 231 | methods. And then lock any additional objects affected by the operations using |
| 232 | method #3 below. The backoff/retry procedure will be a bit more involved, since |
| 233 | when the dynamic locking step hits -EDEADLK we also need to unlock all the |
| 234 | objects acquired with the fixed list. But the w/w mutex debug checks will catch |
| 235 | any interface misuse for these cases. |
| 236 | |
| 237 | Also, method 3 can't fail the lock acquisition step since it doesn't return |
| 238 | -EALREADY. Of course this would be different when using the _interruptible |
| 239 | variants, but that's outside of the scope of these examples here. |
| 240 | |
| 241 | struct obj { |
| 242 | struct ww_mutex ww_mutex; |
| 243 | struct list_head locked_list; |
| 244 | }; |
| 245 | |
| 246 | static DEFINE_WW_CLASS(ww_class); |
| 247 | |
| 248 | void __unlock_objs(struct list_head *list) |
| 249 | { |
| 250 | struct obj *entry, *temp; |
| 251 | |
| 252 | list_for_each_entry_safe (entry, temp, list, locked_list) { |
| 253 | /* need to do that before unlocking, since only the current lock holder is |
| 254 | allowed to use object */ |
| 255 | list_del(&entry->locked_list); |
| 256 | ww_mutex_unlock(entry->ww_mutex) |
| 257 | } |
| 258 | } |
| 259 | |
| 260 | void lock_objs(struct list_head *list, struct ww_acquire_ctx *ctx) |
| 261 | { |
| 262 | struct obj *obj; |
| 263 | |
| 264 | ww_acquire_init(ctx, &ww_class); |
| 265 | |
| 266 | retry: |
| 267 | /* re-init loop start state */ |
| 268 | loop { |
| 269 | /* magic code which walks over a graph and decides which objects |
| 270 | * to lock */ |
| 271 | |
| 272 | ret = ww_mutex_lock(obj->ww_mutex, ctx); |
| 273 | if (ret == -EALREADY) { |
| 274 | /* we have that one already, get to the next object */ |
| 275 | continue; |
| 276 | } |
| 277 | if (ret == -EDEADLK) { |
| 278 | __unlock_objs(list); |
| 279 | |
| 280 | ww_mutex_lock_slow(obj, ctx); |
| 281 | list_add(&entry->locked_list, list); |
| 282 | goto retry; |
| 283 | } |
| 284 | |
| 285 | /* locked a new object, add it to the list */ |
| 286 | list_add_tail(&entry->locked_list, list); |
| 287 | } |
| 288 | |
| 289 | ww_acquire_done(ctx); |
| 290 | return 0; |
| 291 | } |
| 292 | |
| 293 | void unlock_objs(struct list_head *list, struct ww_acquire_ctx *ctx) |
| 294 | { |
| 295 | __unlock_objs(list); |
| 296 | ww_acquire_fini(ctx); |
| 297 | } |
| 298 | |
| 299 | Method 4: Only lock one single objects. In that case deadlock detection and |
| 300 | prevention is obviously overkill, since with grabbing just one lock you can't |
| 301 | produce a deadlock within just one class. To simplify this case the w/w mutex |
| 302 | api can be used with a NULL context. |
| 303 | |
| 304 | Implementation Details |
| 305 | ---------------------- |
| 306 | |
| 307 | Design: |
| 308 | ww_mutex currently encapsulates a struct mutex, this means no extra overhead for |
| 309 | normal mutex locks, which are far more common. As such there is only a small |
| 310 | increase in code size if wait/wound mutexes are not used. |
| 311 | |
| 312 | In general, not much contention is expected. The locks are typically used to |
| 313 | serialize access to resources for devices. The only way to make wakeups |
| 314 | smarter would be at the cost of adding a field to struct mutex_waiter. This |
| 315 | would add overhead to all cases where normal mutexes are used, and |
| 316 | ww_mutexes are generally less performance sensitive. |
| 317 | |
| 318 | Lockdep: |
| 319 | Special care has been taken to warn for as many cases of api abuse |
| 320 | as possible. Some common api abuses will be caught with |
| 321 | CONFIG_DEBUG_MUTEXES, but CONFIG_PROVE_LOCKING is recommended. |
| 322 | |
| 323 | Some of the errors which will be warned about: |
| 324 | - Forgetting to call ww_acquire_fini or ww_acquire_init. |
| 325 | - Attempting to lock more mutexes after ww_acquire_done. |
| 326 | - Attempting to lock the wrong mutex after -EDEADLK and |
| 327 | unlocking all mutexes. |
| 328 | - Attempting to lock the right mutex after -EDEADLK, |
| 329 | before unlocking all mutexes. |
| 330 | |
| 331 | - Calling ww_mutex_lock_slow before -EDEADLK was returned. |
| 332 | |
| 333 | - Unlocking mutexes with the wrong unlock function. |
| 334 | - Calling one of the ww_acquire_* twice on the same context. |
| 335 | - Using a different ww_class for the mutex than for the ww_acquire_ctx. |
| 336 | - Normal lockdep errors that can result in deadlocks. |
| 337 | |
| 338 | Some of the lockdep errors that can result in deadlocks: |
| 339 | - Calling ww_acquire_init to initialize a second ww_acquire_ctx before |
| 340 | having called ww_acquire_fini on the first. |
| 341 | - 'normal' deadlocks that can occur. |
| 342 | |
| 343 | FIXME: Update this section once we have the TASK_DEADLOCK task state flag magic |
| 344 | implemented. |