Kyle Swenson | 8d8f654 | 2021-03-15 11:02:55 -0600 | [diff] [blame] | 1 | /* |
| 2 | * async.c: Asynchronous function calls for boot performance |
| 3 | * |
| 4 | * (C) Copyright 2009 Intel Corporation |
| 5 | * Author: Arjan van de Ven <arjan@linux.intel.com> |
| 6 | * |
| 7 | * This program is free software; you can redistribute it and/or |
| 8 | * modify it under the terms of the GNU General Public License |
| 9 | * as published by the Free Software Foundation; version 2 |
| 10 | * of the License. |
| 11 | */ |
| 12 | |
| 13 | |
| 14 | /* |
| 15 | |
| 16 | Goals and Theory of Operation |
| 17 | |
| 18 | The primary goal of this feature is to reduce the kernel boot time, |
| 19 | by doing various independent hardware delays and discovery operations |
| 20 | decoupled and not strictly serialized. |
| 21 | |
| 22 | More specifically, the asynchronous function call concept allows |
| 23 | certain operations (primarily during system boot) to happen |
| 24 | asynchronously, out of order, while these operations still |
| 25 | have their externally visible parts happen sequentially and in-order. |
| 26 | (not unlike how out-of-order CPUs retire their instructions in order) |
| 27 | |
| 28 | Key to the asynchronous function call implementation is the concept of |
| 29 | a "sequence cookie" (which, although it has an abstracted type, can be |
| 30 | thought of as a monotonically incrementing number). |
| 31 | |
| 32 | The async core will assign each scheduled event such a sequence cookie and |
| 33 | pass this to the called functions. |
| 34 | |
| 35 | The asynchronously called function should before doing a globally visible |
| 36 | operation, such as registering device numbers, call the |
| 37 | async_synchronize_cookie() function and pass in its own cookie. The |
| 38 | async_synchronize_cookie() function will make sure that all asynchronous |
| 39 | operations that were scheduled prior to the operation corresponding with the |
| 40 | cookie have completed. |
| 41 | |
| 42 | Subsystem/driver initialization code that scheduled asynchronous probe |
| 43 | functions, but which shares global resources with other drivers/subsystems |
| 44 | that do not use the asynchronous call feature, need to do a full |
| 45 | synchronization with the async_synchronize_full() function, before returning |
| 46 | from their init function. This is to maintain strict ordering between the |
| 47 | asynchronous and synchronous parts of the kernel. |
| 48 | |
| 49 | */ |
| 50 | |
| 51 | #include <linux/async.h> |
| 52 | #include <linux/atomic.h> |
| 53 | #include <linux/ktime.h> |
| 54 | #include <linux/export.h> |
| 55 | #include <linux/wait.h> |
| 56 | #include <linux/sched.h> |
| 57 | #include <linux/slab.h> |
| 58 | #include <linux/workqueue.h> |
| 59 | |
| 60 | #include "workqueue_internal.h" |
| 61 | |
| 62 | static async_cookie_t next_cookie = 1; |
| 63 | |
| 64 | #define MAX_WORK 32768 |
| 65 | #define ASYNC_COOKIE_MAX ULLONG_MAX /* infinity cookie */ |
| 66 | |
| 67 | static LIST_HEAD(async_global_pending); /* pending from all registered doms */ |
| 68 | static ASYNC_DOMAIN(async_dfl_domain); |
| 69 | static DEFINE_SPINLOCK(async_lock); |
| 70 | |
| 71 | struct async_entry { |
| 72 | struct list_head domain_list; |
| 73 | struct list_head global_list; |
| 74 | struct work_struct work; |
| 75 | async_cookie_t cookie; |
| 76 | async_func_t func; |
| 77 | void *data; |
| 78 | struct async_domain *domain; |
| 79 | }; |
| 80 | |
| 81 | static DECLARE_WAIT_QUEUE_HEAD(async_done); |
| 82 | |
| 83 | static atomic_t entry_count; |
| 84 | |
| 85 | static async_cookie_t lowest_in_progress(struct async_domain *domain) |
| 86 | { |
| 87 | struct list_head *pending; |
| 88 | async_cookie_t ret = ASYNC_COOKIE_MAX; |
| 89 | unsigned long flags; |
| 90 | |
| 91 | spin_lock_irqsave(&async_lock, flags); |
| 92 | |
| 93 | if (domain) |
| 94 | pending = &domain->pending; |
| 95 | else |
| 96 | pending = &async_global_pending; |
| 97 | |
| 98 | if (!list_empty(pending)) |
| 99 | ret = list_first_entry(pending, struct async_entry, |
| 100 | domain_list)->cookie; |
| 101 | |
| 102 | spin_unlock_irqrestore(&async_lock, flags); |
| 103 | return ret; |
| 104 | } |
| 105 | |
| 106 | /* |
| 107 | * pick the first pending entry and run it |
| 108 | */ |
| 109 | static void async_run_entry_fn(struct work_struct *work) |
| 110 | { |
| 111 | struct async_entry *entry = |
| 112 | container_of(work, struct async_entry, work); |
| 113 | unsigned long flags; |
| 114 | ktime_t uninitialized_var(calltime), delta, rettime; |
| 115 | |
| 116 | /* 1) run (and print duration) */ |
| 117 | if (initcall_debug && system_state == SYSTEM_BOOTING) { |
| 118 | pr_debug("calling %lli_%pF @ %i\n", |
| 119 | (long long)entry->cookie, |
| 120 | entry->func, task_pid_nr(current)); |
| 121 | calltime = ktime_get(); |
| 122 | } |
| 123 | entry->func(entry->data, entry->cookie); |
| 124 | if (initcall_debug && system_state == SYSTEM_BOOTING) { |
| 125 | rettime = ktime_get(); |
| 126 | delta = ktime_sub(rettime, calltime); |
| 127 | pr_debug("initcall %lli_%pF returned 0 after %lld usecs\n", |
| 128 | (long long)entry->cookie, |
| 129 | entry->func, |
| 130 | (long long)ktime_to_ns(delta) >> 10); |
| 131 | } |
| 132 | |
| 133 | /* 2) remove self from the pending queues */ |
| 134 | spin_lock_irqsave(&async_lock, flags); |
| 135 | list_del_init(&entry->domain_list); |
| 136 | list_del_init(&entry->global_list); |
| 137 | |
| 138 | /* 3) free the entry */ |
| 139 | kfree(entry); |
| 140 | atomic_dec(&entry_count); |
| 141 | |
| 142 | spin_unlock_irqrestore(&async_lock, flags); |
| 143 | |
| 144 | /* 4) wake up any waiters */ |
| 145 | wake_up(&async_done); |
| 146 | } |
| 147 | |
| 148 | static async_cookie_t __async_schedule(async_func_t func, void *data, struct async_domain *domain) |
| 149 | { |
| 150 | struct async_entry *entry; |
| 151 | unsigned long flags; |
| 152 | async_cookie_t newcookie; |
| 153 | |
| 154 | /* allow irq-off callers */ |
| 155 | entry = kzalloc(sizeof(struct async_entry), GFP_ATOMIC); |
| 156 | |
| 157 | /* |
| 158 | * If we're out of memory or if there's too much work |
| 159 | * pending already, we execute synchronously. |
| 160 | */ |
| 161 | if (!entry || atomic_read(&entry_count) > MAX_WORK) { |
| 162 | kfree(entry); |
| 163 | spin_lock_irqsave(&async_lock, flags); |
| 164 | newcookie = next_cookie++; |
| 165 | spin_unlock_irqrestore(&async_lock, flags); |
| 166 | |
| 167 | /* low on memory.. run synchronously */ |
| 168 | func(data, newcookie); |
| 169 | return newcookie; |
| 170 | } |
| 171 | INIT_LIST_HEAD(&entry->domain_list); |
| 172 | INIT_LIST_HEAD(&entry->global_list); |
| 173 | INIT_WORK(&entry->work, async_run_entry_fn); |
| 174 | entry->func = func; |
| 175 | entry->data = data; |
| 176 | entry->domain = domain; |
| 177 | |
| 178 | spin_lock_irqsave(&async_lock, flags); |
| 179 | |
| 180 | /* allocate cookie and queue */ |
| 181 | newcookie = entry->cookie = next_cookie++; |
| 182 | |
| 183 | list_add_tail(&entry->domain_list, &domain->pending); |
| 184 | if (domain->registered) |
| 185 | list_add_tail(&entry->global_list, &async_global_pending); |
| 186 | |
| 187 | atomic_inc(&entry_count); |
| 188 | spin_unlock_irqrestore(&async_lock, flags); |
| 189 | |
| 190 | /* mark that this task has queued an async job, used by module init */ |
| 191 | current->flags |= PF_USED_ASYNC; |
| 192 | |
| 193 | /* schedule for execution */ |
| 194 | queue_work(system_unbound_wq, &entry->work); |
| 195 | |
| 196 | return newcookie; |
| 197 | } |
| 198 | |
| 199 | /** |
| 200 | * async_schedule - schedule a function for asynchronous execution |
| 201 | * @func: function to execute asynchronously |
| 202 | * @data: data pointer to pass to the function |
| 203 | * |
| 204 | * Returns an async_cookie_t that may be used for checkpointing later. |
| 205 | * Note: This function may be called from atomic or non-atomic contexts. |
| 206 | */ |
| 207 | async_cookie_t async_schedule(async_func_t func, void *data) |
| 208 | { |
| 209 | return __async_schedule(func, data, &async_dfl_domain); |
| 210 | } |
| 211 | EXPORT_SYMBOL_GPL(async_schedule); |
| 212 | |
| 213 | /** |
| 214 | * async_schedule_domain - schedule a function for asynchronous execution within a certain domain |
| 215 | * @func: function to execute asynchronously |
| 216 | * @data: data pointer to pass to the function |
| 217 | * @domain: the domain |
| 218 | * |
| 219 | * Returns an async_cookie_t that may be used for checkpointing later. |
| 220 | * @domain may be used in the async_synchronize_*_domain() functions to |
| 221 | * wait within a certain synchronization domain rather than globally. A |
| 222 | * synchronization domain is specified via @domain. Note: This function |
| 223 | * may be called from atomic or non-atomic contexts. |
| 224 | */ |
| 225 | async_cookie_t async_schedule_domain(async_func_t func, void *data, |
| 226 | struct async_domain *domain) |
| 227 | { |
| 228 | return __async_schedule(func, data, domain); |
| 229 | } |
| 230 | EXPORT_SYMBOL_GPL(async_schedule_domain); |
| 231 | |
| 232 | /** |
| 233 | * async_synchronize_full - synchronize all asynchronous function calls |
| 234 | * |
| 235 | * This function waits until all asynchronous function calls have been done. |
| 236 | */ |
| 237 | void async_synchronize_full(void) |
| 238 | { |
| 239 | async_synchronize_full_domain(NULL); |
| 240 | } |
| 241 | EXPORT_SYMBOL_GPL(async_synchronize_full); |
| 242 | |
| 243 | /** |
| 244 | * async_unregister_domain - ensure no more anonymous waiters on this domain |
| 245 | * @domain: idle domain to flush out of any async_synchronize_full instances |
| 246 | * |
| 247 | * async_synchronize_{cookie|full}_domain() are not flushed since callers |
| 248 | * of these routines should know the lifetime of @domain |
| 249 | * |
| 250 | * Prefer ASYNC_DOMAIN_EXCLUSIVE() declarations over flushing |
| 251 | */ |
| 252 | void async_unregister_domain(struct async_domain *domain) |
| 253 | { |
| 254 | spin_lock_irq(&async_lock); |
| 255 | WARN_ON(!domain->registered || !list_empty(&domain->pending)); |
| 256 | domain->registered = 0; |
| 257 | spin_unlock_irq(&async_lock); |
| 258 | } |
| 259 | EXPORT_SYMBOL_GPL(async_unregister_domain); |
| 260 | |
| 261 | /** |
| 262 | * async_synchronize_full_domain - synchronize all asynchronous function within a certain domain |
| 263 | * @domain: the domain to synchronize |
| 264 | * |
| 265 | * This function waits until all asynchronous function calls for the |
| 266 | * synchronization domain specified by @domain have been done. |
| 267 | */ |
| 268 | void async_synchronize_full_domain(struct async_domain *domain) |
| 269 | { |
| 270 | async_synchronize_cookie_domain(ASYNC_COOKIE_MAX, domain); |
| 271 | } |
| 272 | EXPORT_SYMBOL_GPL(async_synchronize_full_domain); |
| 273 | |
| 274 | /** |
| 275 | * async_synchronize_cookie_domain - synchronize asynchronous function calls within a certain domain with cookie checkpointing |
| 276 | * @cookie: async_cookie_t to use as checkpoint |
| 277 | * @domain: the domain to synchronize (%NULL for all registered domains) |
| 278 | * |
| 279 | * This function waits until all asynchronous function calls for the |
| 280 | * synchronization domain specified by @domain submitted prior to @cookie |
| 281 | * have been done. |
| 282 | */ |
| 283 | void async_synchronize_cookie_domain(async_cookie_t cookie, struct async_domain *domain) |
| 284 | { |
| 285 | ktime_t uninitialized_var(starttime), delta, endtime; |
| 286 | |
| 287 | if (initcall_debug && system_state == SYSTEM_BOOTING) { |
| 288 | pr_debug("async_waiting @ %i\n", task_pid_nr(current)); |
| 289 | starttime = ktime_get(); |
| 290 | } |
| 291 | |
| 292 | wait_event(async_done, lowest_in_progress(domain) >= cookie); |
| 293 | |
| 294 | if (initcall_debug && system_state == SYSTEM_BOOTING) { |
| 295 | endtime = ktime_get(); |
| 296 | delta = ktime_sub(endtime, starttime); |
| 297 | |
| 298 | pr_debug("async_continuing @ %i after %lli usec\n", |
| 299 | task_pid_nr(current), |
| 300 | (long long)ktime_to_ns(delta) >> 10); |
| 301 | } |
| 302 | } |
| 303 | EXPORT_SYMBOL_GPL(async_synchronize_cookie_domain); |
| 304 | |
| 305 | /** |
| 306 | * async_synchronize_cookie - synchronize asynchronous function calls with cookie checkpointing |
| 307 | * @cookie: async_cookie_t to use as checkpoint |
| 308 | * |
| 309 | * This function waits until all asynchronous function calls prior to @cookie |
| 310 | * have been done. |
| 311 | */ |
| 312 | void async_synchronize_cookie(async_cookie_t cookie) |
| 313 | { |
| 314 | async_synchronize_cookie_domain(cookie, &async_dfl_domain); |
| 315 | } |
| 316 | EXPORT_SYMBOL_GPL(async_synchronize_cookie); |
| 317 | |
| 318 | /** |
| 319 | * current_is_async - is %current an async worker task? |
| 320 | * |
| 321 | * Returns %true if %current is an async worker task. |
| 322 | */ |
| 323 | bool current_is_async(void) |
| 324 | { |
| 325 | struct worker *worker = current_wq_worker(); |
| 326 | |
| 327 | return worker && worker->current_func == async_run_entry_fn; |
| 328 | } |