sched_m2.c revision 1.3.2.3
1 1.3.2.3 rmind /* $NetBSD: sched_m2.c,v 1.3.2.3 2007/10/11 20:43:32 rmind Exp $ */ 2 1.3.2.2 rmind 3 1.3.2.2 rmind /* 4 1.3.2.2 rmind * Copyright (c) 2007, Mindaugas Rasiukevicius 5 1.3.2.2 rmind * 6 1.3.2.2 rmind * Redistribution and use in source and binary forms, with or without 7 1.3.2.2 rmind * modification, are permitted provided that the following conditions 8 1.3.2.2 rmind * are met: 9 1.3.2.2 rmind * 1. Redistributions of source code must retain the above copyright 10 1.3.2.2 rmind * notice, this list of conditions and the following disclaimer. 11 1.3.2.2 rmind * 2. Redistributions in binary form must reproduce the above copyright 12 1.3.2.2 rmind * notice, this list of conditions and the following disclaimer in the 13 1.3.2.2 rmind * documentation and/or other materials provided with the distribution. 14 1.3.2.2 rmind * 15 1.3.2.2 rmind * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 16 1.3.2.2 rmind * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 17 1.3.2.2 rmind * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 18 1.3.2.2 rmind * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 19 1.3.2.2 rmind * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 20 1.3.2.2 rmind * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 21 1.3.2.2 rmind * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 22 1.3.2.2 rmind * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 23 1.3.2.2 rmind * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 24 1.3.2.2 rmind * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 25 1.3.2.2 rmind * POSSIBILITY OF SUCH DAMAGE. 26 1.3.2.2 rmind */ 27 1.3.2.2 rmind 28 1.3.2.2 rmind /* 29 1.3.2.2 rmind * TODO: 30 1.3.2.2 rmind * - Implementation of fair share queue; 31 1.3.2.2 rmind * - Support for NUMA; 32 1.3.2.2 rmind */ 33 1.3.2.2 rmind 34 1.3.2.2 rmind #include <sys/cdefs.h> 35 1.3.2.3 rmind __KERNEL_RCSID(0, "$NetBSD: sched_m2.c,v 1.3.2.3 2007/10/11 20:43:32 rmind Exp $"); 36 1.3.2.2 rmind 37 1.3.2.2 rmind #include <sys/param.h> 38 1.3.2.2 rmind 39 1.3.2.2 rmind #include <sys/cpu.h> 40 1.3.2.2 rmind #include <sys/callout.h> 41 1.3.2.2 rmind #include <sys/errno.h> 42 1.3.2.2 rmind #include <sys/kernel.h> 43 1.3.2.2 rmind #include <sys/kmem.h> 44 1.3.2.2 rmind #include <sys/lwp.h> 45 1.3.2.2 rmind #include <sys/mutex.h> 46 1.3.2.2 rmind #include <sys/pool.h> 47 1.3.2.2 rmind #include <sys/proc.h> 48 1.3.2.2 rmind #include <sys/resource.h> 49 1.3.2.2 rmind #include <sys/resourcevar.h> 50 1.3.2.2 rmind #include <sys/sched.h> 51 1.3.2.2 rmind #include <sys/syscallargs.h> 52 1.3.2.2 rmind #include <sys/sysctl.h> 53 1.3.2.2 rmind #include <sys/types.h> 54 1.3.2.2 rmind 55 1.3.2.2 rmind #include <machine/cpu.h> 56 1.3.2.2 rmind 57 1.3.2.2 rmind /* 58 1.3.2.3 rmind * Priority related defintions. 59 1.3.2.2 rmind */ 60 1.3.2.3 rmind #define PRI_TS_COUNT (NPRI_USER) 61 1.3.2.3 rmind #define PRI_RT_COUNT (PRI_COUNT - PRI_TS_COUNT) 62 1.3.2.3 rmind #define PRI_HTS_RANGE (PRI_TS_COUNT / 10) 63 1.3.2.3 rmind 64 1.3.2.3 rmind #define PRI_HIGHEST_TS (PRI_KERNEL - 1) 65 1.3.2.3 rmind #define PRI_DEFAULT (NPRI_USER >> 1) 66 1.3.2.3 rmind 67 1.3.2.3 rmind const int schedppq = 1; 68 1.3.2.2 rmind 69 1.3.2.2 rmind /* 70 1.3.2.2 rmind * Bits per map. 71 1.3.2.2 rmind */ 72 1.3.2.3 rmind #define BITMAP_BITS (32) 73 1.3.2.3 rmind #define BITMAP_SHIFT (5) 74 1.3.2.3 rmind #define BITMAP_MSB (0x80000000) 75 1.3.2.3 rmind #define BITMAP_MASK (BITMAP_BITS - 1) 76 1.3.2.2 rmind 77 1.3.2.2 rmind /* 78 1.3.2.2 rmind * Time-slices and priorities. 79 1.3.2.2 rmind */ 80 1.3.2.2 rmind static u_int min_ts; /* Minimal time-slice */ 81 1.3.2.2 rmind static u_int max_ts; /* Maximal time-slice */ 82 1.3.2.2 rmind static u_int rt_ts; /* Real-time time-slice */ 83 1.3.2.2 rmind static u_int ts_map[PRI_COUNT]; /* Map of time-slices */ 84 1.3.2.2 rmind static pri_t high_pri[PRI_COUNT]; /* Map for priority increase */ 85 1.3.2.2 rmind 86 1.3.2.2 rmind /* 87 1.3.2.2 rmind * Migration and balancing. 88 1.3.2.2 rmind */ 89 1.3.2.2 rmind #ifdef MULTIPROCESSOR 90 1.3.2.2 rmind static u_int cacheht_time; /* Cache hotness time */ 91 1.3.2.2 rmind static u_int min_catch; /* Minimal LWP count for catching */ 92 1.3.2.2 rmind 93 1.3.2.2 rmind static u_int balance_period; /* Balance period */ 94 1.3.2.2 rmind static struct callout balance_ch; /* Callout of balancer */ 95 1.3.2.2 rmind 96 1.3.2.2 rmind static struct cpu_info * volatile worker_ci; 97 1.3.2.2 rmind 98 1.3.2.2 rmind #define CACHE_HOT(sil) (sil->sl_lrtime && \ 99 1.3.2.2 rmind (hardclock_ticks - sil->sl_lrtime < cacheht_time)) 100 1.3.2.2 rmind 101 1.3.2.2 rmind #endif 102 1.3.2.2 rmind 103 1.3.2.2 rmind /* 104 1.3.2.2 rmind * Structures, runqueue. 105 1.3.2.2 rmind */ 106 1.3.2.2 rmind 107 1.3.2.2 rmind typedef struct { 108 1.3.2.2 rmind TAILQ_HEAD(, lwp) q_head; 109 1.3.2.2 rmind } queue_t; 110 1.3.2.2 rmind 111 1.3.2.2 rmind typedef struct { 112 1.3.2.2 rmind /* Lock and bitmap */ 113 1.3.2.2 rmind kmutex_t r_rq_mutex; 114 1.3.2.3 rmind uint32_t r_bitmap[PRI_COUNT >> BITMAP_SHIFT]; 115 1.3.2.2 rmind /* Counters */ 116 1.3.2.2 rmind u_int r_count; /* Count of the threads */ 117 1.3.2.2 rmind pri_t r_highest_pri; /* Highest priority */ 118 1.3.2.2 rmind u_int r_avgcount; /* Average count of threads */ 119 1.3.2.2 rmind u_int r_mcount; /* Count of migratable threads */ 120 1.3.2.2 rmind /* Runqueues */ 121 1.3.2.2 rmind queue_t r_rt_queue[PRI_RT_COUNT]; 122 1.3.2.2 rmind queue_t r_ts_queue[PRI_TS_COUNT]; 123 1.3.2.2 rmind } runqueue_t; 124 1.3.2.2 rmind 125 1.3.2.2 rmind typedef struct { 126 1.3.2.2 rmind u_int sl_flags; 127 1.3.2.2 rmind u_int sl_timeslice; /* Time-slice of thread */ 128 1.3.2.2 rmind u_int sl_slept; /* Saved sleep time for sleep sum */ 129 1.3.2.2 rmind u_int sl_slpsum; /* Sum of sleep time */ 130 1.3.2.2 rmind u_int sl_rtime; /* Saved start time of run */ 131 1.3.2.2 rmind u_int sl_rtsum; /* Sum of the run time */ 132 1.3.2.2 rmind u_int sl_lrtime; /* Last run time */ 133 1.3.2.2 rmind } sched_info_lwp_t; 134 1.3.2.2 rmind 135 1.3.2.2 rmind /* Flags */ 136 1.3.2.2 rmind #define SL_BATCH 0x01 137 1.3.2.2 rmind 138 1.3.2.2 rmind /* Pool of the scheduler-specific structures for threads */ 139 1.3.2.2 rmind static struct pool sil_pool; 140 1.3.2.2 rmind 141 1.3.2.2 rmind /* 142 1.3.2.2 rmind * Prototypes. 143 1.3.2.2 rmind */ 144 1.3.2.2 rmind 145 1.3.2.2 rmind static inline void * sched_getrq(runqueue_t *, const pri_t); 146 1.3.2.2 rmind static inline void sched_newts(struct lwp *); 147 1.3.2.2 rmind static void sched_precalcts(void); 148 1.3.2.2 rmind 149 1.3.2.2 rmind #ifdef MULTIPROCESSOR 150 1.3.2.2 rmind static struct lwp * sched_catchlwp(void); 151 1.3.2.2 rmind static void sched_balance(void *); 152 1.3.2.2 rmind #endif 153 1.3.2.2 rmind 154 1.3.2.2 rmind /* 155 1.3.2.2 rmind * Initialization and setup. 156 1.3.2.2 rmind */ 157 1.3.2.2 rmind 158 1.3.2.2 rmind void 159 1.3.2.2 rmind sched_rqinit(void) 160 1.3.2.2 rmind { 161 1.3.2.2 rmind struct cpu_info *ci = curcpu(); 162 1.3.2.2 rmind 163 1.3.2.2 rmind if (hz < 100) { 164 1.3.2.2 rmind panic("sched_rqinit: value of HZ is too low\n"); 165 1.3.2.2 rmind } 166 1.3.2.2 rmind 167 1.3.2.2 rmind /* Default timing ranges */ 168 1.3.2.2 rmind min_ts = mstohz(50); /* ~50ms */ 169 1.3.2.2 rmind max_ts = mstohz(150); /* ~150ms */ 170 1.3.2.2 rmind rt_ts = mstohz(100); /* ~100ms */ 171 1.3.2.2 rmind sched_precalcts(); 172 1.3.2.2 rmind 173 1.3.2.2 rmind #ifdef MULTIPROCESSOR 174 1.3.2.2 rmind /* Balancing */ 175 1.3.2.2 rmind worker_ci = ci; 176 1.3.2.2 rmind cacheht_time = mstohz(5); /* ~5 ms */ 177 1.3.2.2 rmind balance_period = mstohz(300); /* ~300ms */ 178 1.3.2.2 rmind min_catch = ~0; 179 1.3.2.2 rmind #endif 180 1.3.2.2 rmind 181 1.3.2.2 rmind /* Pool of the scheduler-specific structures */ 182 1.3.2.2 rmind pool_init(&sil_pool, sizeof(sched_info_lwp_t), 0, 0, 0, 183 1.3.2.2 rmind "lwpsd", &pool_allocator_nointr, IPL_NONE); 184 1.3.2.2 rmind 185 1.3.2.2 rmind /* Attach the primary CPU here */ 186 1.3.2.2 rmind sched_cpuattach(ci); 187 1.3.2.2 rmind 188 1.3.2.2 rmind /* Initialize the scheduler structure of the primary LWP */ 189 1.3.2.2 rmind lwp0.l_mutex = &ci->ci_schedstate.spc_lwplock; 190 1.3.2.2 rmind sched_lwp_fork(&lwp0); 191 1.3.2.2 rmind sched_newts(&lwp0); 192 1.3.2.2 rmind } 193 1.3.2.2 rmind 194 1.3.2.2 rmind void 195 1.3.2.2 rmind sched_setup(void) 196 1.3.2.2 rmind { 197 1.3.2.2 rmind 198 1.3.2.2 rmind #ifdef MULTIPROCESSOR 199 1.3.2.2 rmind /* Minimal count of LWPs for catching: log2(count of CPUs) */ 200 1.3.2.2 rmind min_catch = min(ffs(ncpu) - 1, 4); 201 1.3.2.2 rmind 202 1.3.2.2 rmind /* Initialize balancing callout and run it */ 203 1.3.2.2 rmind callout_init(&balance_ch, CALLOUT_MPSAFE); 204 1.3.2.2 rmind callout_setfunc(&balance_ch, sched_balance, NULL); 205 1.3.2.2 rmind callout_schedule(&balance_ch, balance_period); 206 1.3.2.2 rmind #endif 207 1.3.2.2 rmind } 208 1.3.2.2 rmind 209 1.3.2.2 rmind void 210 1.3.2.2 rmind sched_cpuattach(struct cpu_info *ci) 211 1.3.2.2 rmind { 212 1.3.2.2 rmind runqueue_t *ci_rq; 213 1.3.2.2 rmind void *rq_ptr; 214 1.3.2.2 rmind u_int i, size; 215 1.3.2.2 rmind 216 1.3.2.2 rmind /* 217 1.3.2.2 rmind * Allocate the run queue. 218 1.3.2.2 rmind * XXX: Estimate cache behaviour more.. 219 1.3.2.2 rmind */ 220 1.3.2.2 rmind size = roundup(sizeof(runqueue_t), CACHE_LINE_SIZE) + CACHE_LINE_SIZE; 221 1.3.2.2 rmind rq_ptr = kmem_zalloc(size, KM_NOSLEEP); 222 1.3.2.2 rmind if (rq_ptr == NULL) { 223 1.3.2.2 rmind panic("scheduler: could not allocate the runqueue"); 224 1.3.2.2 rmind } 225 1.3.2.2 rmind /* XXX: Save the original pointer for future.. */ 226 1.3.2.2 rmind ci_rq = (void *)(roundup((intptr_t)(rq_ptr), CACHE_LINE_SIZE)); 227 1.3.2.2 rmind 228 1.3.2.2 rmind /* Initialize run queues */ 229 1.3.2.2 rmind mutex_init(&ci_rq->r_rq_mutex, MUTEX_SPIN, IPL_SCHED); 230 1.3.2.2 rmind for (i = 0; i < PRI_RT_COUNT; i++) 231 1.3.2.2 rmind TAILQ_INIT(&ci_rq->r_rt_queue[i].q_head); 232 1.3.2.2 rmind for (i = 0; i < PRI_TS_COUNT; i++) 233 1.3.2.2 rmind TAILQ_INIT(&ci_rq->r_ts_queue[i].q_head); 234 1.3.2.3 rmind ci_rq->r_highest_pri = 0; 235 1.3.2.2 rmind 236 1.3.2.2 rmind ci->ci_schedstate.spc_sched_info = ci_rq; 237 1.3.2.2 rmind ci->ci_schedstate.spc_mutex = &ci_rq->r_rq_mutex; 238 1.3.2.2 rmind } 239 1.3.2.2 rmind 240 1.3.2.2 rmind /* Pre-calculate the time-slices for the priorities */ 241 1.3.2.2 rmind static void 242 1.3.2.2 rmind sched_precalcts(void) 243 1.3.2.2 rmind { 244 1.3.2.2 rmind pri_t p; 245 1.3.2.2 rmind 246 1.3.2.3 rmind /* Time-sharing range */ 247 1.3.2.3 rmind for (p = 0; p <= PRI_HIGHEST_TS; p++) { 248 1.3.2.3 rmind ts_map[p] = max_ts - 249 1.3.2.3 rmind (p * 100 / (PRI_TS_COUNT - 1) * (max_ts - min_ts) / 100); 250 1.3.2.3 rmind high_pri[p] = (PRI_HIGHEST_TS - PRI_HTS_RANGE) + 251 1.3.2.3 rmind ((p * PRI_HTS_RANGE) / (PRI_TS_COUNT - 1)); 252 1.3.2.2 rmind } 253 1.3.2.2 rmind 254 1.3.2.3 rmind /* Real-time range */ 255 1.3.2.3 rmind for (p = (PRI_HIGHEST_TS + 1); p < PRI_COUNT; p++) { 256 1.3.2.3 rmind ts_map[p] = rt_ts; 257 1.3.2.3 rmind high_pri[p] = p; 258 1.3.2.2 rmind } 259 1.3.2.2 rmind } 260 1.3.2.2 rmind 261 1.3.2.2 rmind /* 262 1.3.2.2 rmind * Hooks. 263 1.3.2.2 rmind */ 264 1.3.2.2 rmind 265 1.3.2.2 rmind void 266 1.3.2.2 rmind sched_proc_fork(struct proc *parent, struct proc *child) 267 1.3.2.2 rmind { 268 1.3.2.2 rmind struct lwp *l; 269 1.3.2.2 rmind 270 1.3.2.2 rmind LIST_FOREACH(l, &child->p_lwps, l_sibling) { 271 1.3.2.2 rmind lwp_lock(l); 272 1.3.2.2 rmind sched_newts(l); 273 1.3.2.2 rmind lwp_unlock(l); 274 1.3.2.2 rmind } 275 1.3.2.2 rmind } 276 1.3.2.2 rmind 277 1.3.2.2 rmind void 278 1.3.2.2 rmind sched_proc_exit(struct proc *child, struct proc *parent) 279 1.3.2.2 rmind { 280 1.3.2.2 rmind 281 1.3.2.2 rmind /* Dummy */ 282 1.3.2.2 rmind } 283 1.3.2.2 rmind 284 1.3.2.2 rmind void 285 1.3.2.2 rmind sched_lwp_fork(struct lwp *l) 286 1.3.2.2 rmind { 287 1.3.2.2 rmind 288 1.3.2.2 rmind KASSERT(l->l_sched_info == NULL); 289 1.3.2.2 rmind l->l_sched_info = pool_get(&sil_pool, PR_WAITOK); 290 1.3.2.2 rmind memset(l->l_sched_info, 0, sizeof(sched_info_lwp_t)); 291 1.3.2.3 rmind if (l->l_usrpri <= PRI_HIGHEST_TS) /* XXX: For now only.. */ 292 1.3.2.2 rmind l->l_usrpri = l->l_priority = PRI_DEFAULT; 293 1.3.2.2 rmind } 294 1.3.2.2 rmind 295 1.3.2.2 rmind void 296 1.3.2.2 rmind sched_lwp_exit(struct lwp *l) 297 1.3.2.2 rmind { 298 1.3.2.2 rmind 299 1.3.2.2 rmind KASSERT(l->l_sched_info != NULL); 300 1.3.2.2 rmind pool_put(&sil_pool, l->l_sched_info); 301 1.3.2.2 rmind l->l_sched_info = NULL; 302 1.3.2.2 rmind } 303 1.3.2.2 rmind 304 1.3.2.2 rmind void 305 1.3.2.2 rmind sched_setrunnable(struct lwp *l) 306 1.3.2.2 rmind { 307 1.3.2.2 rmind 308 1.3.2.2 rmind /* Dummy */ 309 1.3.2.2 rmind } 310 1.3.2.2 rmind 311 1.3.2.2 rmind void 312 1.3.2.2 rmind sched_schedclock(struct lwp *l) 313 1.3.2.2 rmind { 314 1.3.2.2 rmind 315 1.3.2.2 rmind /* Dummy */ 316 1.3.2.2 rmind } 317 1.3.2.2 rmind 318 1.3.2.2 rmind /* 319 1.3.2.2 rmind * Priorities and time-slice. 320 1.3.2.2 rmind */ 321 1.3.2.2 rmind 322 1.3.2.2 rmind void 323 1.3.2.2 rmind sched_nice(struct proc *p, int prio) 324 1.3.2.2 rmind { 325 1.3.2.2 rmind int nprio; 326 1.3.2.2 rmind struct lwp *l; 327 1.3.2.2 rmind 328 1.3.2.2 rmind KASSERT(mutex_owned(&p->p_stmutex)); 329 1.3.2.2 rmind 330 1.3.2.2 rmind p->p_nice = prio; 331 1.3.2.3 rmind nprio = max(min(PRI_DEFAULT + p->p_nice, PRI_HIGHEST_TS), 0); 332 1.3.2.2 rmind 333 1.3.2.2 rmind LIST_FOREACH(l, &p->p_lwps, l_sibling) { 334 1.3.2.2 rmind lwp_lock(l); 335 1.3.2.2 rmind lwp_changepri(l, nprio); 336 1.3.2.2 rmind lwp_unlock(l); 337 1.3.2.2 rmind } 338 1.3.2.2 rmind } 339 1.3.2.2 rmind 340 1.3.2.2 rmind /* Recalculate the time-slice */ 341 1.3.2.2 rmind static inline void 342 1.3.2.2 rmind sched_newts(struct lwp *l) 343 1.3.2.2 rmind { 344 1.3.2.2 rmind sched_info_lwp_t *sil = l->l_sched_info; 345 1.3.2.2 rmind 346 1.3.2.2 rmind sil->sl_timeslice = ts_map[lwp_eprio(l)]; 347 1.3.2.2 rmind } 348 1.3.2.2 rmind 349 1.3.2.2 rmind /* 350 1.3.2.2 rmind * Control of the runqueue. 351 1.3.2.2 rmind */ 352 1.3.2.2 rmind 353 1.3.2.2 rmind static inline void * 354 1.3.2.2 rmind sched_getrq(runqueue_t *ci_rq, const pri_t prio) 355 1.3.2.2 rmind { 356 1.3.2.2 rmind 357 1.3.2.2 rmind KASSERT(prio < PRI_COUNT); 358 1.3.2.3 rmind return (prio <= PRI_HIGHEST_TS) ? 359 1.3.2.3 rmind &ci_rq->r_ts_queue[prio].q_head : 360 1.3.2.3 rmind &ci_rq->r_rt_queue[prio - PRI_HIGHEST_TS - 1].q_head; 361 1.3.2.2 rmind } 362 1.3.2.2 rmind 363 1.3.2.2 rmind void 364 1.3.2.2 rmind sched_enqueue(struct lwp *l, bool swtch) 365 1.3.2.2 rmind { 366 1.3.2.2 rmind runqueue_t *ci_rq; 367 1.3.2.2 rmind sched_info_lwp_t *sil = l->l_sched_info; 368 1.3.2.2 rmind TAILQ_HEAD(, lwp) *q_head; 369 1.3.2.2 rmind const pri_t eprio = lwp_eprio(l); 370 1.3.2.2 rmind 371 1.3.2.2 rmind ci_rq = l->l_cpu->ci_schedstate.spc_sched_info; 372 1.3.2.2 rmind KASSERT(lwp_locked(l, l->l_cpu->ci_schedstate.spc_mutex)); 373 1.3.2.2 rmind 374 1.3.2.2 rmind /* Update the last run time on switch */ 375 1.3.2.2 rmind if (swtch == true) { 376 1.3.2.2 rmind sil->sl_lrtime = hardclock_ticks; 377 1.3.2.2 rmind sil->sl_rtsum += (hardclock_ticks - sil->sl_rtime); 378 1.3.2.2 rmind } else 379 1.3.2.2 rmind sil->sl_lrtime = 0; 380 1.3.2.2 rmind 381 1.3.2.2 rmind /* Enqueue the thread */ 382 1.3.2.2 rmind q_head = sched_getrq(ci_rq, eprio); 383 1.3.2.2 rmind if (TAILQ_EMPTY(q_head)) { 384 1.3.2.2 rmind u_int i; 385 1.3.2.2 rmind uint32_t q; 386 1.3.2.2 rmind 387 1.3.2.2 rmind /* Mark bit */ 388 1.3.2.2 rmind i = eprio >> BITMAP_SHIFT; 389 1.3.2.3 rmind q = BITMAP_MSB >> (eprio & BITMAP_MASK); 390 1.3.2.3 rmind KASSERT((ci_rq->r_bitmap[i] & q) == 0); 391 1.3.2.3 rmind ci_rq->r_bitmap[i] |= q; 392 1.3.2.2 rmind } 393 1.3.2.2 rmind TAILQ_INSERT_TAIL(q_head, l, l_runq); 394 1.3.2.2 rmind ci_rq->r_count++; 395 1.3.2.2 rmind if ((l->l_flag & LW_BOUND) == 0) 396 1.3.2.2 rmind ci_rq->r_mcount++; 397 1.3.2.2 rmind 398 1.3.2.2 rmind /* 399 1.3.2.2 rmind * Update the value of highest priority in the runqueue, 400 1.3.2.2 rmind * if priority of this thread is higher. 401 1.3.2.2 rmind */ 402 1.3.2.3 rmind if (eprio > ci_rq->r_highest_pri) 403 1.3.2.2 rmind ci_rq->r_highest_pri = eprio; 404 1.3.2.2 rmind 405 1.3.2.2 rmind sched_newts(l); 406 1.3.2.2 rmind } 407 1.3.2.2 rmind 408 1.3.2.2 rmind void 409 1.3.2.2 rmind sched_dequeue(struct lwp *l) 410 1.3.2.2 rmind { 411 1.3.2.2 rmind runqueue_t *ci_rq; 412 1.3.2.2 rmind TAILQ_HEAD(, lwp) *q_head; 413 1.3.2.2 rmind const pri_t eprio = lwp_eprio(l); 414 1.3.2.2 rmind 415 1.3.2.2 rmind ci_rq = l->l_cpu->ci_schedstate.spc_sched_info; 416 1.3.2.2 rmind KASSERT(lwp_locked(l, l->l_cpu->ci_schedstate.spc_mutex)); 417 1.3.2.3 rmind KASSERT(eprio <= ci_rq->r_highest_pri); 418 1.3.2.2 rmind KASSERT(ci_rq->r_bitmap[eprio >> BITMAP_SHIFT] != 0); 419 1.3.2.2 rmind KASSERT(ci_rq->r_count > 0); 420 1.3.2.2 rmind 421 1.3.2.2 rmind ci_rq->r_count--; 422 1.3.2.2 rmind if ((l->l_flag & LW_BOUND) == 0) 423 1.3.2.2 rmind ci_rq->r_mcount--; 424 1.3.2.2 rmind 425 1.3.2.2 rmind q_head = sched_getrq(ci_rq, eprio); 426 1.3.2.2 rmind TAILQ_REMOVE(q_head, l, l_runq); 427 1.3.2.2 rmind if (TAILQ_EMPTY(q_head)) { 428 1.3.2.2 rmind u_int i; 429 1.3.2.2 rmind uint32_t q; 430 1.3.2.2 rmind 431 1.3.2.2 rmind /* Unmark bit */ 432 1.3.2.2 rmind i = eprio >> BITMAP_SHIFT; 433 1.3.2.3 rmind q = BITMAP_MSB >> (eprio & BITMAP_MASK); 434 1.3.2.3 rmind KASSERT((ci_rq->r_bitmap[i] & q) != 0); 435 1.3.2.3 rmind ci_rq->r_bitmap[i] &= ~q; 436 1.3.2.2 rmind 437 1.3.2.2 rmind /* 438 1.3.2.2 rmind * Update the value of highest priority in the runqueue, in a 439 1.3.2.2 rmind * case it was a last thread in the queue of highest priority. 440 1.3.2.2 rmind */ 441 1.3.2.2 rmind if (eprio != ci_rq->r_highest_pri) 442 1.3.2.2 rmind return; 443 1.3.2.2 rmind 444 1.3.2.2 rmind do { 445 1.3.2.2 rmind q = ffs(ci_rq->r_bitmap[i]); 446 1.3.2.2 rmind if (q) { 447 1.3.2.2 rmind ci_rq->r_highest_pri = 448 1.3.2.3 rmind (i << BITMAP_SHIFT) + (BITMAP_BITS - q); 449 1.3.2.2 rmind return; 450 1.3.2.2 rmind } 451 1.3.2.3 rmind } while (i--); 452 1.3.2.2 rmind 453 1.3.2.3 rmind /* If not found - set the lowest value */ 454 1.3.2.3 rmind ci_rq->r_highest_pri = 0; 455 1.3.2.2 rmind } 456 1.3.2.2 rmind } 457 1.3.2.2 rmind 458 1.3.2.2 rmind void 459 1.3.2.2 rmind sched_slept(struct lwp *l) 460 1.3.2.2 rmind { 461 1.3.2.2 rmind sched_info_lwp_t *sil = l->l_sched_info; 462 1.3.2.2 rmind 463 1.3.2.2 rmind /* Save the time when thread has slept */ 464 1.3.2.2 rmind sil->sl_slept = hardclock_ticks; 465 1.3.2.2 rmind 466 1.3.2.2 rmind /* 467 1.3.2.3 rmind * If thread is in time-sharing queue and batch flag is not marked, 468 1.3.2.3 rmind * increase the the priority, and run with the lower time-quantum. 469 1.3.2.2 rmind */ 470 1.3.2.3 rmind if (l->l_usrpri < PRI_HIGHEST_TS && (sil->sl_flags & SL_BATCH) == 0) { 471 1.3.2.3 rmind KASSERT(l->l_policy == SCHED_OTHER); 472 1.3.2.3 rmind l->l_usrpri++; 473 1.3.2.3 rmind } 474 1.3.2.2 rmind } 475 1.3.2.2 rmind 476 1.3.2.2 rmind void 477 1.3.2.2 rmind sched_wakeup(struct lwp *l) 478 1.3.2.2 rmind { 479 1.3.2.2 rmind sched_info_lwp_t *sil = l->l_sched_info; 480 1.3.2.2 rmind 481 1.3.2.2 rmind /* Update sleep time delta */ 482 1.3.2.2 rmind sil->sl_slpsum += (l->l_slptime == 0) ? 483 1.3.2.2 rmind (hardclock_ticks - sil->sl_slept) : hz; 484 1.3.2.2 rmind 485 1.3.2.2 rmind /* If thread was sleeping a second or more - set a high priority */ 486 1.3.2.2 rmind if (l->l_slptime > 1 || (hardclock_ticks - sil->sl_slept) >= hz) 487 1.3.2.2 rmind l->l_usrpri = l->l_priority = high_pri[l->l_usrpri]; 488 1.3.2.2 rmind KASSERT(sil->sl_slept > 0); 489 1.3.2.2 rmind 490 1.3.2.2 rmind /* Also, consider looking for a better CPU to wake up */ 491 1.3.2.2 rmind if ((l->l_flag & (LW_BOUND | LW_SYSTEM)) == 0) 492 1.3.2.2 rmind l->l_cpu = sched_takecpu(l); 493 1.3.2.2 rmind } 494 1.3.2.2 rmind 495 1.3.2.2 rmind void 496 1.3.2.2 rmind sched_pstats_hook(struct lwp *l) 497 1.3.2.2 rmind { 498 1.3.2.2 rmind sched_info_lwp_t *sil = l->l_sched_info; 499 1.3.2.3 rmind const bool batch = (sil->sl_rtsum > sil->sl_slpsum); 500 1.3.2.3 rmind 501 1.3.2.3 rmind /* Reset the time sums */ 502 1.3.2.3 rmind sil->sl_slpsum = 0; 503 1.3.2.3 rmind sil->sl_rtsum = 0; 504 1.3.2.3 rmind 505 1.3.2.3 rmind /* Estimate threads on time-sharing queue only */ 506 1.3.2.3 rmind if (l->l_usrpri > PRI_HIGHEST_TS) 507 1.3.2.3 rmind return; 508 1.3.2.3 rmind KASSERT(l->l_policy == SCHED_OTHER); 509 1.3.2.2 rmind 510 1.3.2.2 rmind /* 511 1.3.2.2 rmind * Set that thread is more CPU-bound, if sum of run time exceeds the 512 1.3.2.2 rmind * sum of sleep time. If it is CPU-bound not a first time - decrease 513 1.3.2.2 rmind * the priority. 514 1.3.2.2 rmind */ 515 1.3.2.3 rmind if (batch) { 516 1.3.2.3 rmind if ((sil->sl_flags & SL_BATCH) && l->l_usrpri) 517 1.3.2.3 rmind l->l_usrpri--; 518 1.3.2.2 rmind sil->sl_flags |= SL_BATCH; 519 1.3.2.2 rmind } else { 520 1.3.2.2 rmind sil->sl_flags &= ~SL_BATCH; 521 1.3.2.2 rmind } 522 1.3.2.2 rmind 523 1.3.2.2 rmind /* If thread was not ran a second or more - set a high priority */ 524 1.3.2.3 rmind if (l->l_stat == LSRUN && sil->sl_lrtime && 525 1.3.2.3 rmind (hardclock_ticks - sil->sl_lrtime >= hz)) 526 1.3.2.2 rmind lwp_changepri(l, high_pri[l->l_usrpri]); 527 1.3.2.2 rmind } 528 1.3.2.2 rmind 529 1.3.2.2 rmind /* 530 1.3.2.2 rmind * Migration and balancing. 531 1.3.2.2 rmind */ 532 1.3.2.2 rmind 533 1.3.2.2 rmind #ifdef MULTIPROCESSOR 534 1.3.2.2 rmind 535 1.3.2.2 rmind /* Check if LWP can migrate to the chosen CPU */ 536 1.3.2.2 rmind static inline bool 537 1.3.2.2 rmind sched_migratable(const struct lwp *l, const struct cpu_info *ci) 538 1.3.2.2 rmind { 539 1.3.2.2 rmind 540 1.3.2.2 rmind if (ci->ci_schedstate.spc_flags & SPCF_OFFLINE) 541 1.3.2.2 rmind return false; 542 1.3.2.2 rmind 543 1.3.2.2 rmind if ((l->l_flag & LW_BOUND) == 0) 544 1.3.2.2 rmind return true; 545 1.3.2.2 rmind #if 0 546 1.3.2.2 rmind return cpu_in_pset(ci, l->l_psid); 547 1.3.2.2 rmind #else 548 1.3.2.2 rmind return false; 549 1.3.2.2 rmind #endif 550 1.3.2.2 rmind } 551 1.3.2.2 rmind 552 1.3.2.2 rmind /* 553 1.3.2.2 rmind * Estimate the migration of LWP to the other CPU. 554 1.3.2.2 rmind * Take and return the CPU, if migration is needed. 555 1.3.2.2 rmind */ 556 1.3.2.2 rmind struct cpu_info * 557 1.3.2.2 rmind sched_takecpu(struct lwp *l) 558 1.3.2.2 rmind { 559 1.3.2.2 rmind struct cpu_info *ci, *tci = NULL; 560 1.3.2.2 rmind struct schedstate_percpu *spc; 561 1.3.2.2 rmind runqueue_t *ci_rq; 562 1.3.2.2 rmind sched_info_lwp_t *sil; 563 1.3.2.2 rmind CPU_INFO_ITERATOR cii; 564 1.3.2.2 rmind pri_t eprio, lpri; 565 1.3.2.2 rmind 566 1.3.2.2 rmind ci = l->l_cpu; 567 1.3.2.2 rmind spc = &ci->ci_schedstate; 568 1.3.2.2 rmind ci_rq = spc->spc_sched_info; 569 1.3.2.2 rmind 570 1.3.2.2 rmind /* CPU of this thread is idling - run there */ 571 1.3.2.2 rmind if (ci_rq->r_count == 0) 572 1.3.2.2 rmind return ci; 573 1.3.2.2 rmind 574 1.3.2.2 rmind eprio = lwp_eprio(l); 575 1.3.2.2 rmind sil = l->l_sched_info; 576 1.3.2.2 rmind 577 1.3.2.2 rmind /* Stay if thread is cache-hot */ 578 1.3.2.2 rmind if (l->l_stat == LSSLEEP && l->l_slptime <= 1 && 579 1.3.2.3 rmind CACHE_HOT(sil) && eprio >= spc->spc_curpriority) 580 1.3.2.2 rmind return ci; 581 1.3.2.2 rmind 582 1.3.2.2 rmind /* Run on current CPU if priority of thread is higher */ 583 1.3.2.2 rmind ci = curcpu(); 584 1.3.2.2 rmind spc = &ci->ci_schedstate; 585 1.3.2.3 rmind if (eprio > spc->spc_curpriority && sched_migratable(l, ci)) 586 1.3.2.2 rmind return ci; 587 1.3.2.2 rmind 588 1.3.2.2 rmind /* 589 1.3.2.2 rmind * Look for the CPU with the lowest priority thread. In case of 590 1.3.2.2 rmind * equal the priority - check the lower count of the threads. 591 1.3.2.2 rmind */ 592 1.3.2.3 rmind lpri = PRI_COUNT; 593 1.3.2.2 rmind for (CPU_INFO_FOREACH(cii, ci)) { 594 1.3.2.2 rmind runqueue_t *ici_rq; 595 1.3.2.2 rmind pri_t pri; 596 1.3.2.2 rmind 597 1.3.2.2 rmind spc = &ci->ci_schedstate; 598 1.3.2.2 rmind ici_rq = spc->spc_sched_info; 599 1.3.2.3 rmind pri = max(spc->spc_curpriority, ici_rq->r_highest_pri); 600 1.3.2.3 rmind if (pri > lpri) 601 1.3.2.2 rmind continue; 602 1.3.2.2 rmind 603 1.3.2.3 rmind if (pri == lpri && tci && ci_rq->r_count < ici_rq->r_count) 604 1.3.2.2 rmind continue; 605 1.3.2.2 rmind 606 1.3.2.2 rmind if (sched_migratable(l, ci) == false) 607 1.3.2.2 rmind continue; 608 1.3.2.2 rmind 609 1.3.2.2 rmind lpri = pri; 610 1.3.2.2 rmind tci = ci; 611 1.3.2.2 rmind ci_rq = ici_rq; 612 1.3.2.2 rmind } 613 1.3.2.2 rmind 614 1.3.2.3 rmind KASSERT(tci != NULL); 615 1.3.2.2 rmind return tci; 616 1.3.2.2 rmind } 617 1.3.2.2 rmind 618 1.3.2.2 rmind /* 619 1.3.2.2 rmind * Tries to catch an LWP from the runqueue of other CPU. 620 1.3.2.2 rmind */ 621 1.3.2.2 rmind static struct lwp * 622 1.3.2.2 rmind sched_catchlwp(void) 623 1.3.2.2 rmind { 624 1.3.2.2 rmind struct cpu_info *curci = curcpu(), *ci = worker_ci; 625 1.3.2.2 rmind TAILQ_HEAD(, lwp) *q_head; 626 1.3.2.2 rmind runqueue_t *ci_rq; 627 1.3.2.2 rmind struct lwp *l; 628 1.3.2.2 rmind 629 1.3.2.2 rmind if (curci == ci) 630 1.3.2.2 rmind return NULL; 631 1.3.2.2 rmind 632 1.3.2.2 rmind /* Lockless check */ 633 1.3.2.2 rmind ci_rq = ci->ci_schedstate.spc_sched_info; 634 1.3.2.2 rmind if (ci_rq->r_count < min_catch) 635 1.3.2.2 rmind return NULL; 636 1.3.2.2 rmind 637 1.3.2.2 rmind /* 638 1.3.2.2 rmind * Double-lock the runqueues. 639 1.3.2.2 rmind */ 640 1.3.2.2 rmind if (curci < ci) { 641 1.3.2.2 rmind spc_lock(ci); 642 1.3.2.2 rmind } else if (!mutex_tryenter(ci->ci_schedstate.spc_mutex)) { 643 1.3.2.2 rmind const runqueue_t *cur_rq = curci->ci_schedstate.spc_sched_info; 644 1.3.2.2 rmind 645 1.3.2.2 rmind spc_unlock(curci); 646 1.3.2.2 rmind spc_lock(ci); 647 1.3.2.2 rmind spc_lock(curci); 648 1.3.2.2 rmind 649 1.3.2.2 rmind if (cur_rq->r_count) { 650 1.3.2.2 rmind spc_unlock(ci); 651 1.3.2.2 rmind return NULL; 652 1.3.2.2 rmind } 653 1.3.2.2 rmind } 654 1.3.2.2 rmind 655 1.3.2.2 rmind if (ci_rq->r_count < min_catch) { 656 1.3.2.2 rmind spc_unlock(ci); 657 1.3.2.2 rmind return NULL; 658 1.3.2.2 rmind } 659 1.3.2.2 rmind 660 1.3.2.2 rmind /* Take the highest priority thread */ 661 1.3.2.2 rmind q_head = sched_getrq(ci_rq, ci_rq->r_highest_pri); 662 1.3.2.2 rmind l = TAILQ_FIRST(q_head); 663 1.3.2.2 rmind 664 1.3.2.2 rmind for (;;) { 665 1.3.2.2 rmind sched_info_lwp_t *sil; 666 1.3.2.2 rmind 667 1.3.2.2 rmind /* Check the first and next result from the queue */ 668 1.3.2.2 rmind if (l == NULL) 669 1.3.2.2 rmind break; 670 1.3.2.2 rmind 671 1.3.2.2 rmind /* Look for threads, whose are allowed to migrate */ 672 1.3.2.2 rmind sil = l->l_sched_info; 673 1.3.2.2 rmind if ((l->l_flag & LW_SYSTEM) || CACHE_HOT(sil) || 674 1.3.2.2 rmind sched_migratable(l, curci) == false) { 675 1.3.2.2 rmind l = TAILQ_NEXT(l, l_runq); 676 1.3.2.2 rmind continue; 677 1.3.2.2 rmind } 678 1.3.2.2 rmind /* Recheck if chosen thread is still on the runqueue */ 679 1.3.2.2 rmind if (l->l_stat == LSRUN && (l->l_flag & LW_INMEM)) { 680 1.3.2.2 rmind sched_dequeue(l); 681 1.3.2.2 rmind l->l_cpu = curci; 682 1.3.2.2 rmind lwp_setlock(l, curci->ci_schedstate.spc_mutex); 683 1.3.2.2 rmind sched_enqueue(l, false); 684 1.3.2.2 rmind break; 685 1.3.2.2 rmind } 686 1.3.2.2 rmind l = TAILQ_NEXT(l, l_runq); 687 1.3.2.2 rmind } 688 1.3.2.2 rmind spc_unlock(ci); 689 1.3.2.2 rmind 690 1.3.2.2 rmind return l; 691 1.3.2.2 rmind } 692 1.3.2.2 rmind 693 1.3.2.2 rmind /* 694 1.3.2.2 rmind * Periodical calculations for balancing. 695 1.3.2.2 rmind */ 696 1.3.2.2 rmind static void 697 1.3.2.2 rmind sched_balance(void *nocallout) 698 1.3.2.2 rmind { 699 1.3.2.2 rmind struct cpu_info *ci, *hci; 700 1.3.2.2 rmind runqueue_t *ci_rq; 701 1.3.2.2 rmind CPU_INFO_ITERATOR cii; 702 1.3.2.2 rmind u_int highest; 703 1.3.2.2 rmind 704 1.3.2.2 rmind hci = curcpu(); 705 1.3.2.2 rmind highest = 0; 706 1.3.2.2 rmind 707 1.3.2.2 rmind /* Make lockless countings */ 708 1.3.2.2 rmind for (CPU_INFO_FOREACH(cii, ci)) { 709 1.3.2.2 rmind ci_rq = ci->ci_schedstate.spc_sched_info; 710 1.3.2.2 rmind 711 1.3.2.2 rmind /* Average count of the threads */ 712 1.3.2.2 rmind ci_rq->r_avgcount = (ci_rq->r_avgcount + ci_rq->r_mcount) >> 1; 713 1.3.2.2 rmind 714 1.3.2.2 rmind /* Look for CPU with the highest average */ 715 1.3.2.2 rmind if (ci_rq->r_avgcount > highest) { 716 1.3.2.2 rmind hci = ci; 717 1.3.2.2 rmind highest = ci_rq->r_avgcount; 718 1.3.2.2 rmind } 719 1.3.2.2 rmind } 720 1.3.2.2 rmind 721 1.3.2.2 rmind /* Update the worker */ 722 1.3.2.2 rmind worker_ci = hci; 723 1.3.2.2 rmind 724 1.3.2.2 rmind if (nocallout == NULL) 725 1.3.2.2 rmind callout_schedule(&balance_ch, balance_period); 726 1.3.2.2 rmind } 727 1.3.2.2 rmind 728 1.3.2.2 rmind #else 729 1.3.2.2 rmind 730 1.3.2.2 rmind struct cpu_info * 731 1.3.2.2 rmind sched_takecpu(struct lwp *l) 732 1.3.2.2 rmind { 733 1.3.2.2 rmind 734 1.3.2.2 rmind return l->l_cpu; 735 1.3.2.2 rmind } 736 1.3.2.2 rmind 737 1.3.2.2 rmind #endif /* MULTIPROCESSOR */ 738 1.3.2.2 rmind 739 1.3.2.2 rmind /* 740 1.3.2.2 rmind * Scheduler mill. 741 1.3.2.2 rmind */ 742 1.3.2.2 rmind struct lwp * 743 1.3.2.2 rmind sched_nextlwp(void) 744 1.3.2.2 rmind { 745 1.3.2.2 rmind struct cpu_info *ci = curcpu(); 746 1.3.2.2 rmind struct schedstate_percpu *spc; 747 1.3.2.2 rmind TAILQ_HEAD(, lwp) *q_head; 748 1.3.2.2 rmind sched_info_lwp_t *sil; 749 1.3.2.2 rmind runqueue_t *ci_rq; 750 1.3.2.2 rmind struct lwp *l; 751 1.3.2.2 rmind 752 1.3.2.2 rmind spc = &ci->ci_schedstate; 753 1.3.2.2 rmind ci_rq = ci->ci_schedstate.spc_sched_info; 754 1.3.2.2 rmind 755 1.3.2.2 rmind #ifdef MULTIPROCESSOR 756 1.3.2.2 rmind /* If runqueue is empty, try to catch some thread from other CPU */ 757 1.3.2.2 rmind if (spc->spc_flags & SPCF_OFFLINE) { 758 1.3.2.2 rmind if (ci_rq->r_mcount == 0) 759 1.3.2.2 rmind return NULL; 760 1.3.2.2 rmind } else if (ci_rq->r_count == 0) { 761 1.3.2.2 rmind /* Reset the counter, and call the balancer */ 762 1.3.2.2 rmind ci_rq->r_avgcount = 0; 763 1.3.2.2 rmind sched_balance(ci); 764 1.3.2.2 rmind 765 1.3.2.2 rmind /* The re-locking will be done inside */ 766 1.3.2.2 rmind return sched_catchlwp(); 767 1.3.2.2 rmind } 768 1.3.2.2 rmind #else 769 1.3.2.2 rmind if (ci_rq->r_count == 0) 770 1.3.2.2 rmind return NULL; 771 1.3.2.2 rmind #endif 772 1.3.2.2 rmind 773 1.3.2.2 rmind /* Take the highest priority thread */ 774 1.3.2.2 rmind KASSERT(ci_rq->r_bitmap[ci_rq->r_highest_pri >> BITMAP_SHIFT]); 775 1.3.2.2 rmind q_head = sched_getrq(ci_rq, ci_rq->r_highest_pri); 776 1.3.2.2 rmind l = TAILQ_FIRST(q_head); 777 1.3.2.2 rmind KASSERT(l != NULL); 778 1.3.2.2 rmind 779 1.3.2.2 rmind /* Update the counters */ 780 1.3.2.2 rmind sil = l->l_sched_info; 781 1.3.2.2 rmind KASSERT(sil->sl_timeslice >= min_ts); 782 1.3.2.2 rmind KASSERT(sil->sl_timeslice <= max_ts); 783 1.3.2.2 rmind spc->spc_ticks = sil->sl_timeslice; 784 1.3.2.2 rmind sil->sl_rtime = hardclock_ticks; 785 1.3.2.2 rmind 786 1.3.2.2 rmind return l; 787 1.3.2.2 rmind } 788 1.3.2.2 rmind 789 1.3.2.2 rmind bool 790 1.3.2.2 rmind sched_curcpu_runnable_p(void) 791 1.3.2.2 rmind { 792 1.3.2.2 rmind const struct cpu_info *ci = curcpu(); 793 1.3.2.2 rmind const runqueue_t *ci_rq = ci->ci_schedstate.spc_sched_info; 794 1.3.2.2 rmind 795 1.3.2.2 rmind if (ci->ci_schedstate.spc_flags & SPCF_OFFLINE) 796 1.3.2.2 rmind return ci_rq->r_mcount; 797 1.3.2.2 rmind 798 1.3.2.2 rmind return ci_rq->r_count; 799 1.3.2.2 rmind } 800 1.3.2.2 rmind 801 1.3.2.2 rmind /* 802 1.3.2.2 rmind * Time-driven events. 803 1.3.2.2 rmind */ 804 1.3.2.2 rmind 805 1.3.2.2 rmind /* 806 1.3.2.2 rmind * Called once per time-quantum. This routine is CPU-local and runs at 807 1.3.2.2 rmind * IPL_SCHED, thus the locking is not needed. 808 1.3.2.2 rmind */ 809 1.3.2.2 rmind void 810 1.3.2.2 rmind sched_tick(struct cpu_info *ci) 811 1.3.2.2 rmind { 812 1.3.2.2 rmind const runqueue_t *ci_rq = ci->ci_schedstate.spc_sched_info; 813 1.3.2.2 rmind struct schedstate_percpu *spc = &ci->ci_schedstate; 814 1.3.2.2 rmind struct lwp *l = curlwp; 815 1.3.2.2 rmind sched_info_lwp_t *sil = l->l_sched_info; 816 1.3.2.2 rmind 817 1.3.2.2 rmind if (CURCPU_IDLE_P()) 818 1.3.2.2 rmind return; 819 1.3.2.2 rmind 820 1.3.2.2 rmind switch (l->l_policy) { 821 1.3.2.2 rmind case SCHED_FIFO: 822 1.3.2.2 rmind /* 823 1.3.2.2 rmind * Update the time-quantum, and continue running, 824 1.3.2.2 rmind * if thread runs on FIFO real-time policy. 825 1.3.2.2 rmind */ 826 1.3.2.2 rmind spc->spc_ticks = sil->sl_timeslice; 827 1.3.2.2 rmind return; 828 1.3.2.2 rmind case SCHED_OTHER: 829 1.3.2.3 rmind /* 830 1.3.2.3 rmind * If thread is in time-sharing queue, decrease the priority, 831 1.3.2.3 rmind * and run with a higher time-quantum. 832 1.3.2.3 rmind */ 833 1.3.2.3 rmind if (l->l_usrpri > PRI_HIGHEST_TS || l->l_usrpri == 0) 834 1.3.2.2 rmind break; 835 1.3.2.3 rmind l->l_priority = --l->l_usrpri; 836 1.3.2.2 rmind break; 837 1.3.2.2 rmind } 838 1.3.2.2 rmind 839 1.3.2.2 rmind /* 840 1.3.2.2 rmind * If there are higher priority threads or threads in the same queue, 841 1.3.2.2 rmind * mark that thread should yield, otherwise, continue running. 842 1.3.2.2 rmind */ 843 1.3.2.3 rmind if (lwp_eprio(l) <= ci_rq->r_highest_pri) { 844 1.3.2.2 rmind spc->spc_flags |= SPCF_SHOULDYIELD; 845 1.3.2.2 rmind cpu_need_resched(ci, 0); 846 1.3.2.2 rmind } else 847 1.3.2.2 rmind spc->spc_ticks = sil->sl_timeslice; 848 1.3.2.2 rmind } 849 1.3.2.2 rmind 850 1.3.2.2 rmind /* 851 1.3.2.2 rmind * Sysctl nodes and initialization. 852 1.3.2.2 rmind */ 853 1.3.2.2 rmind 854 1.3.2.2 rmind static int 855 1.3.2.2 rmind sysctl_sched_mints(SYSCTLFN_ARGS) 856 1.3.2.2 rmind { 857 1.3.2.2 rmind struct sysctlnode node; 858 1.3.2.2 rmind struct cpu_info *ci; 859 1.3.2.2 rmind int error, newsize; 860 1.3.2.2 rmind CPU_INFO_ITERATOR cii; 861 1.3.2.2 rmind 862 1.3.2.2 rmind node = *rnode; 863 1.3.2.2 rmind node.sysctl_data = &newsize; 864 1.3.2.2 rmind 865 1.3.2.2 rmind newsize = hztoms(min_ts); 866 1.3.2.2 rmind error = sysctl_lookup(SYSCTLFN_CALL(&node)); 867 1.3.2.2 rmind if (error || newp == NULL) 868 1.3.2.2 rmind return error; 869 1.3.2.2 rmind 870 1.3.2.2 rmind if (newsize < 1 || newsize > hz || newsize >= max_ts) 871 1.3.2.2 rmind return EINVAL; 872 1.3.2.2 rmind 873 1.3.2.2 rmind /* It is safe to do this in such order */ 874 1.3.2.2 rmind for (CPU_INFO_FOREACH(cii, ci)) 875 1.3.2.2 rmind spc_lock(ci); 876 1.3.2.2 rmind 877 1.3.2.2 rmind min_ts = mstohz(newsize); 878 1.3.2.2 rmind sched_precalcts(); 879 1.3.2.2 rmind 880 1.3.2.2 rmind for (CPU_INFO_FOREACH(cii, ci)) 881 1.3.2.2 rmind spc_unlock(ci); 882 1.3.2.2 rmind 883 1.3.2.2 rmind return 0; 884 1.3.2.2 rmind } 885 1.3.2.2 rmind 886 1.3.2.2 rmind static int 887 1.3.2.2 rmind sysctl_sched_maxts(SYSCTLFN_ARGS) 888 1.3.2.2 rmind { 889 1.3.2.2 rmind struct sysctlnode node; 890 1.3.2.2 rmind struct cpu_info *ci; 891 1.3.2.2 rmind int error, newsize; 892 1.3.2.2 rmind CPU_INFO_ITERATOR cii; 893 1.3.2.2 rmind 894 1.3.2.2 rmind node = *rnode; 895 1.3.2.2 rmind node.sysctl_data = &newsize; 896 1.3.2.2 rmind 897 1.3.2.2 rmind newsize = hztoms(max_ts); 898 1.3.2.2 rmind error = sysctl_lookup(SYSCTLFN_CALL(&node)); 899 1.3.2.2 rmind if (error || newp == NULL) 900 1.3.2.2 rmind return error; 901 1.3.2.2 rmind 902 1.3.2.2 rmind if (newsize < 10 || newsize > hz || newsize <= min_ts) 903 1.3.2.2 rmind return EINVAL; 904 1.3.2.2 rmind 905 1.3.2.2 rmind /* It is safe to do this in such order */ 906 1.3.2.2 rmind for (CPU_INFO_FOREACH(cii, ci)) 907 1.3.2.2 rmind spc_lock(ci); 908 1.3.2.2 rmind 909 1.3.2.2 rmind max_ts = mstohz(newsize); 910 1.3.2.2 rmind sched_precalcts(); 911 1.3.2.2 rmind 912 1.3.2.2 rmind for (CPU_INFO_FOREACH(cii, ci)) 913 1.3.2.2 rmind spc_unlock(ci); 914 1.3.2.2 rmind 915 1.3.2.2 rmind return 0; 916 1.3.2.2 rmind } 917 1.3.2.2 rmind 918 1.3.2.2 rmind SYSCTL_SETUP(sysctl_sched_setup, "sysctl kern.sched subtree setup") 919 1.3.2.2 rmind { 920 1.3.2.2 rmind const struct sysctlnode *node = NULL; 921 1.3.2.2 rmind 922 1.3.2.2 rmind sysctl_createv(clog, 0, NULL, NULL, 923 1.3.2.2 rmind CTLFLAG_PERMANENT, 924 1.3.2.2 rmind CTLTYPE_NODE, "kern", NULL, 925 1.3.2.2 rmind NULL, 0, NULL, 0, 926 1.3.2.2 rmind CTL_KERN, CTL_EOL); 927 1.3.2.2 rmind sysctl_createv(clog, 0, NULL, &node, 928 1.3.2.2 rmind CTLFLAG_PERMANENT, 929 1.3.2.2 rmind CTLTYPE_NODE, "sched", 930 1.3.2.2 rmind SYSCTL_DESCR("Scheduler options"), 931 1.3.2.2 rmind NULL, 0, NULL, 0, 932 1.3.2.2 rmind CTL_KERN, CTL_CREATE, CTL_EOL); 933 1.3.2.2 rmind 934 1.3.2.2 rmind if (node == NULL) 935 1.3.2.2 rmind return; 936 1.3.2.2 rmind 937 1.3.2.2 rmind sysctl_createv(clog, 0, &node, NULL, 938 1.3.2.2 rmind CTLFLAG_PERMANENT, 939 1.3.2.2 rmind CTLTYPE_STRING, "name", NULL, 940 1.3.2.2 rmind NULL, 0, __UNCONST("M2"), 0, 941 1.3.2.2 rmind CTL_CREATE, CTL_EOL); 942 1.3.2.2 rmind sysctl_createv(clog, 0, &node, NULL, 943 1.3.2.2 rmind CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 944 1.3.2.2 rmind CTLTYPE_INT, "maxts", 945 1.3.2.2 rmind SYSCTL_DESCR("Maximal time quantum (in microseconds)"), 946 1.3.2.2 rmind sysctl_sched_maxts, 0, &max_ts, 0, 947 1.3.2.2 rmind CTL_CREATE, CTL_EOL); 948 1.3.2.2 rmind sysctl_createv(clog, 0, &node, NULL, 949 1.3.2.2 rmind CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 950 1.3.2.2 rmind CTLTYPE_INT, "mints", 951 1.3.2.2 rmind SYSCTL_DESCR("Minimal time quantum (in microseconds)"), 952 1.3.2.2 rmind sysctl_sched_mints, 0, &min_ts, 0, 953 1.3.2.2 rmind CTL_CREATE, CTL_EOL); 954 1.3.2.2 rmind 955 1.3.2.2 rmind #ifdef MULTIPROCESSOR 956 1.3.2.2 rmind sysctl_createv(clog, 0, &node, NULL, 957 1.3.2.2 rmind CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 958 1.3.2.2 rmind CTLTYPE_INT, "cacheht_time", 959 1.3.2.2 rmind SYSCTL_DESCR("Cache hotness time"), 960 1.3.2.2 rmind NULL, 0, &cacheht_time, 0, 961 1.3.2.2 rmind CTL_CREATE, CTL_EOL); 962 1.3.2.2 rmind sysctl_createv(clog, 0, &node, NULL, 963 1.3.2.2 rmind CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 964 1.3.2.2 rmind CTLTYPE_INT, "balance_period", 965 1.3.2.2 rmind SYSCTL_DESCR("Balance period"), 966 1.3.2.2 rmind NULL, 0, &balance_period, 0, 967 1.3.2.2 rmind CTL_CREATE, CTL_EOL); 968 1.3.2.2 rmind sysctl_createv(clog, 0, &node, NULL, 969 1.3.2.2 rmind CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 970 1.3.2.2 rmind CTLTYPE_INT, "min_catch", 971 1.3.2.2 rmind SYSCTL_DESCR("Minimal count of threads for catching"), 972 1.3.2.2 rmind NULL, 0, &min_catch, 0, 973 1.3.2.2 rmind CTL_CREATE, CTL_EOL); 974 1.3.2.2 rmind #endif 975 1.3.2.2 rmind } 976 1.3.2.2 rmind 977 1.3.2.2 rmind /* 978 1.3.2.2 rmind * Debugging. 979 1.3.2.2 rmind */ 980 1.3.2.2 rmind 981 1.3.2.2 rmind #ifdef DDB 982 1.3.2.2 rmind 983 1.3.2.2 rmind void 984 1.3.2.2 rmind sched_print_runqueue(void (*pr)(const char *, ...)) 985 1.3.2.2 rmind { 986 1.3.2.2 rmind runqueue_t *ci_rq; 987 1.3.2.2 rmind sched_info_lwp_t *sil; 988 1.3.2.2 rmind struct lwp *l; 989 1.3.2.2 rmind struct proc *p; 990 1.3.2.2 rmind int i; 991 1.3.2.2 rmind 992 1.3.2.2 rmind struct cpu_info *ci; 993 1.3.2.2 rmind CPU_INFO_ITERATOR cii; 994 1.3.2.2 rmind 995 1.3.2.2 rmind for (CPU_INFO_FOREACH(cii, ci)) { 996 1.3.2.2 rmind ci_rq = ci->ci_schedstate.spc_sched_info; 997 1.3.2.2 rmind 998 1.3.2.2 rmind (*pr)("Run-queue (CPU = %d):\n", ci->ci_cpuid); 999 1.3.2.2 rmind (*pr)(" pid.lid = %d.%d, threads count = %u, " 1000 1.3.2.2 rmind "avgcount = %u, highest pri = %d\n", 1001 1.3.2.2 rmind ci->ci_curlwp->l_proc->p_pid, ci->ci_curlwp->l_lid, 1002 1.3.2.2 rmind ci_rq->r_count, ci_rq->r_avgcount, ci_rq->r_highest_pri); 1003 1.3.2.3 rmind i = (PRI_COUNT >> BITMAP_SHIFT) - 1; 1004 1.3.2.2 rmind do { 1005 1.3.2.3 rmind uint32_t q; 1006 1.3.2.3 rmind q = ci_rq->r_bitmap[i]; 1007 1.3.2.3 rmind (*pr)(" bitmap[%d] => [ %d (0x%x) ]\n", i, ffs(q), q); 1008 1.3.2.3 rmind } while (i--); 1009 1.3.2.2 rmind } 1010 1.3.2.2 rmind 1011 1.3.2.2 rmind (*pr)(" %5s %4s %4s %10s %3s %4s %11s %3s %s\n", 1012 1.3.2.2 rmind "LID", "PRI", "UPRI", "FL", "ST", "TS", "LWP", "CPU", "LRTIME"); 1013 1.3.2.2 rmind 1014 1.3.2.2 rmind PROCLIST_FOREACH(p, &allproc) { 1015 1.3.2.2 rmind (*pr)(" /- %d (%s)\n", (int)p->p_pid, p->p_comm); 1016 1.3.2.2 rmind LIST_FOREACH(l, &p->p_lwps, l_sibling) { 1017 1.3.2.2 rmind sil = l->l_sched_info; 1018 1.3.2.2 rmind ci = l->l_cpu; 1019 1.3.2.2 rmind (*pr)(" | %5d %4u %4u 0x%8.8x %3s %4u %11p %3d " 1020 1.3.2.2 rmind "%u ST=%d RT=%d %d\n", 1021 1.3.2.2 rmind (int)l->l_lid, l->l_priority, l->l_usrpri, 1022 1.3.2.2 rmind l->l_flag, l->l_stat == LSRUN ? "RQ" : 1023 1.3.2.2 rmind (l->l_stat == LSSLEEP ? "SQ" : "-"), 1024 1.3.2.2 rmind sil->sl_timeslice, l, ci->ci_cpuid, 1025 1.3.2.2 rmind (u_int)(hardclock_ticks - sil->sl_lrtime), 1026 1.3.2.2 rmind sil->sl_slpsum, sil->sl_rtsum, sil->sl_flags); 1027 1.3.2.2 rmind } 1028 1.3.2.2 rmind } 1029 1.3.2.2 rmind } 1030 1.3.2.2 rmind 1031 1.3.2.2 rmind #endif /* defined(DDB) */ 1032
Indexes created Thu Oct 02 10:09:58 GMT 2025