| 1 | /* $NetBSD: kern_clock.c,v 1.134 2015/04/22 16:46:58 pooka Exp $ */ |
| 2 | |
| 3 | /*- |
| 4 | * Copyright (c) 2000, 2004, 2006, 2007, 2008 The NetBSD Foundation, Inc. |
| 5 | * All rights reserved. |
| 6 | * |
| 7 | * This code is derived from software contributed to The NetBSD Foundation |
| 8 | * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, |
| 9 | * NASA Ames Research Center. |
| 10 | * This code is derived from software contributed to The NetBSD Foundation |
| 11 | * by Charles M. Hannum. |
| 12 | * |
| 13 | * Redistribution and use in source and binary forms, with or without |
| 14 | * modification, are permitted provided that the following conditions |
| 15 | * are met: |
| 16 | * 1. Redistributions of source code must retain the above copyright |
| 17 | * notice, this list of conditions and the following disclaimer. |
| 18 | * 2. Redistributions in binary form must reproduce the above copyright |
| 19 | * notice, this list of conditions and the following disclaimer in the |
| 20 | * documentation and/or other materials provided with the distribution. |
| 21 | * |
| 22 | * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS |
| 23 | * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED |
| 24 | * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR |
| 25 | * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS |
| 26 | * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR |
| 27 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF |
| 28 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS |
| 29 | * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN |
| 30 | * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
| 31 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
| 32 | * POSSIBILITY OF SUCH DAMAGE. |
| 33 | */ |
| 34 | |
| 35 | /*- |
| 36 | * Copyright (c) 1982, 1986, 1991, 1993 |
| 37 | * The Regents of the University of California. All rights reserved. |
| 38 | * (c) UNIX System Laboratories, Inc. |
| 39 | * All or some portions of this file are derived from material licensed |
| 40 | * to the University of California by American Telephone and Telegraph |
| 41 | * Co. or Unix System Laboratories, Inc. and are reproduced herein with |
| 42 | * the permission of UNIX System Laboratories, Inc. |
| 43 | * |
| 44 | * Redistribution and use in source and binary forms, with or without |
| 45 | * modification, are permitted provided that the following conditions |
| 46 | * are met: |
| 47 | * 1. Redistributions of source code must retain the above copyright |
| 48 | * notice, this list of conditions and the following disclaimer. |
| 49 | * 2. Redistributions in binary form must reproduce the above copyright |
| 50 | * notice, this list of conditions and the following disclaimer in the |
| 51 | * documentation and/or other materials provided with the distribution. |
| 52 | * 3. Neither the name of the University nor the names of its contributors |
| 53 | * may be used to endorse or promote products derived from this software |
| 54 | * without specific prior written permission. |
| 55 | * |
| 56 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND |
| 57 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 58 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 59 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE |
| 60 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 61 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 62 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 63 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 64 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 65 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 66 | * SUCH DAMAGE. |
| 67 | * |
| 68 | * @(#)kern_clock.c 8.5 (Berkeley) 1/21/94 |
| 69 | */ |
| 70 | |
| 71 | #include <sys/cdefs.h> |
| 72 | __KERNEL_RCSID(0, "$NetBSD: kern_clock.c,v 1.134 2015/04/22 16:46:58 pooka Exp $" ); |
| 73 | |
| 74 | #ifdef _KERNEL_OPT |
| 75 | #include "opt_dtrace.h" |
| 76 | #include "opt_perfctrs.h" |
| 77 | #endif |
| 78 | |
| 79 | #include <sys/param.h> |
| 80 | #include <sys/systm.h> |
| 81 | #include <sys/callout.h> |
| 82 | #include <sys/kernel.h> |
| 83 | #include <sys/proc.h> |
| 84 | #include <sys/resourcevar.h> |
| 85 | #include <sys/signalvar.h> |
| 86 | #include <sys/sysctl.h> |
| 87 | #include <sys/timex.h> |
| 88 | #include <sys/sched.h> |
| 89 | #include <sys/time.h> |
| 90 | #include <sys/timetc.h> |
| 91 | #include <sys/cpu.h> |
| 92 | #include <sys/atomic.h> |
| 93 | |
| 94 | #ifdef GPROF |
| 95 | #include <sys/gmon.h> |
| 96 | #endif |
| 97 | |
| 98 | #ifdef KDTRACE_HOOKS |
| 99 | #include <sys/dtrace_bsd.h> |
| 100 | #include <sys/cpu.h> |
| 101 | |
| 102 | cyclic_clock_func_t cyclic_clock_func[MAXCPUS]; |
| 103 | #endif |
| 104 | |
| 105 | static int sysctl_kern_clockrate(SYSCTLFN_PROTO); |
| 106 | |
| 107 | /* |
| 108 | * Clock handling routines. |
| 109 | * |
| 110 | * This code is written to operate with two timers that run independently of |
| 111 | * each other. The main clock, running hz times per second, is used to keep |
| 112 | * track of real time. The second timer handles kernel and user profiling, |
| 113 | * and does resource use estimation. If the second timer is programmable, |
| 114 | * it is randomized to avoid aliasing between the two clocks. For example, |
| 115 | * the randomization prevents an adversary from always giving up the CPU |
| 116 | * just before its quantum expires. Otherwise, it would never accumulate |
| 117 | * CPU ticks. The mean frequency of the second timer is stathz. |
| 118 | * |
| 119 | * If no second timer exists, stathz will be zero; in this case we drive |
| 120 | * profiling and statistics off the main clock. This WILL NOT be accurate; |
| 121 | * do not do it unless absolutely necessary. |
| 122 | * |
| 123 | * The statistics clock may (or may not) be run at a higher rate while |
| 124 | * profiling. This profile clock runs at profhz. We require that profhz |
| 125 | * be an integral multiple of stathz. |
| 126 | * |
| 127 | * If the statistics clock is running fast, it must be divided by the ratio |
| 128 | * profhz/stathz for statistics. (For profiling, every tick counts.) |
| 129 | */ |
| 130 | |
| 131 | int stathz; |
| 132 | int profhz; |
| 133 | int profsrc; |
| 134 | int schedhz; |
| 135 | int profprocs; |
| 136 | int hardclock_ticks; |
| 137 | static int hardscheddiv; /* hard => sched divider (used if schedhz == 0) */ |
| 138 | static int psdiv; /* prof => stat divider */ |
| 139 | int psratio; /* ratio: prof / stat */ |
| 140 | |
| 141 | static u_int get_intr_timecount(struct timecounter *); |
| 142 | |
| 143 | static struct timecounter intr_timecounter = { |
| 144 | get_intr_timecount, /* get_timecount */ |
| 145 | 0, /* no poll_pps */ |
| 146 | ~0u, /* counter_mask */ |
| 147 | 0, /* frequency */ |
| 148 | "clockinterrupt" , /* name */ |
| 149 | 0, /* quality - minimum implementation level for a clock */ |
| 150 | NULL, /* prev */ |
| 151 | NULL, /* next */ |
| 152 | }; |
| 153 | |
| 154 | static u_int |
| 155 | get_intr_timecount(struct timecounter *tc) |
| 156 | { |
| 157 | |
| 158 | return (u_int)hardclock_ticks; |
| 159 | } |
| 160 | |
| 161 | /* |
| 162 | * Initialize clock frequencies and start both clocks running. |
| 163 | */ |
| 164 | void |
| 165 | initclocks(void) |
| 166 | { |
| 167 | static struct sysctllog *clog; |
| 168 | int i; |
| 169 | |
| 170 | /* |
| 171 | * Set divisors to 1 (normal case) and let the machine-specific |
| 172 | * code do its bit. |
| 173 | */ |
| 174 | psdiv = 1; |
| 175 | /* |
| 176 | * provide minimum default time counter |
| 177 | * will only run at interrupt resolution |
| 178 | */ |
| 179 | intr_timecounter.tc_frequency = hz; |
| 180 | tc_init(&intr_timecounter); |
| 181 | cpu_initclocks(); |
| 182 | |
| 183 | /* |
| 184 | * Compute profhz and stathz, fix profhz if needed. |
| 185 | */ |
| 186 | i = stathz ? stathz : hz; |
| 187 | if (profhz == 0) |
| 188 | profhz = i; |
| 189 | psratio = profhz / i; |
| 190 | if (schedhz == 0) { |
| 191 | /* 16Hz is best */ |
| 192 | hardscheddiv = hz / 16; |
| 193 | if (hardscheddiv <= 0) |
| 194 | panic("hardscheddiv" ); |
| 195 | } |
| 196 | |
| 197 | sysctl_createv(&clog, 0, NULL, NULL, |
| 198 | CTLFLAG_PERMANENT, |
| 199 | CTLTYPE_STRUCT, "clockrate" , |
| 200 | SYSCTL_DESCR("Kernel clock rates" ), |
| 201 | sysctl_kern_clockrate, 0, NULL, |
| 202 | sizeof(struct clockinfo), |
| 203 | CTL_KERN, KERN_CLOCKRATE, CTL_EOL); |
| 204 | sysctl_createv(&clog, 0, NULL, NULL, |
| 205 | CTLFLAG_PERMANENT, |
| 206 | CTLTYPE_INT, "hardclock_ticks" , |
| 207 | SYSCTL_DESCR("Number of hardclock ticks" ), |
| 208 | NULL, 0, &hardclock_ticks, sizeof(hardclock_ticks), |
| 209 | CTL_KERN, KERN_HARDCLOCK_TICKS, CTL_EOL); |
| 210 | } |
| 211 | |
| 212 | /* |
| 213 | * The real-time timer, interrupting hz times per second. |
| 214 | */ |
| 215 | void |
| 216 | hardclock(struct clockframe *frame) |
| 217 | { |
| 218 | struct lwp *l; |
| 219 | struct cpu_info *ci; |
| 220 | |
| 221 | ci = curcpu(); |
| 222 | l = ci->ci_data.cpu_onproc; |
| 223 | |
| 224 | timer_tick(l, CLKF_USERMODE(frame)); |
| 225 | |
| 226 | /* |
| 227 | * If no separate statistics clock is available, run it from here. |
| 228 | */ |
| 229 | if (stathz == 0) |
| 230 | statclock(frame); |
| 231 | /* |
| 232 | * If no separate schedclock is provided, call it here |
| 233 | * at about 16 Hz. |
| 234 | */ |
| 235 | if (schedhz == 0) { |
| 236 | if ((int)(--ci->ci_schedstate.spc_schedticks) <= 0) { |
| 237 | schedclock(l); |
| 238 | ci->ci_schedstate.spc_schedticks = hardscheddiv; |
| 239 | } |
| 240 | } |
| 241 | if ((--ci->ci_schedstate.spc_ticks) <= 0) |
| 242 | sched_tick(ci); |
| 243 | |
| 244 | if (CPU_IS_PRIMARY(ci)) { |
| 245 | hardclock_ticks++; |
| 246 | tc_ticktock(); |
| 247 | } |
| 248 | |
| 249 | /* |
| 250 | * Update real-time timeout queue. |
| 251 | */ |
| 252 | callout_hardclock(); |
| 253 | |
| 254 | #ifdef KDTRACE_HOOKS |
| 255 | cyclic_clock_func_t func = cyclic_clock_func[cpu_index(ci)]; |
| 256 | if (func) { |
| 257 | (*func)((struct clockframe *)frame); |
| 258 | } |
| 259 | #endif |
| 260 | } |
| 261 | |
| 262 | /* |
| 263 | * Start profiling on a process. |
| 264 | * |
| 265 | * Kernel profiling passes proc0 which never exits and hence |
| 266 | * keeps the profile clock running constantly. |
| 267 | */ |
| 268 | void |
| 269 | startprofclock(struct proc *p) |
| 270 | { |
| 271 | |
| 272 | KASSERT(mutex_owned(&p->p_stmutex)); |
| 273 | |
| 274 | if ((p->p_stflag & PST_PROFIL) == 0) { |
| 275 | p->p_stflag |= PST_PROFIL; |
| 276 | /* |
| 277 | * This is only necessary if using the clock as the |
| 278 | * profiling source. |
| 279 | */ |
| 280 | if (++profprocs == 1 && stathz != 0) |
| 281 | psdiv = psratio; |
| 282 | } |
| 283 | } |
| 284 | |
| 285 | /* |
| 286 | * Stop profiling on a process. |
| 287 | */ |
| 288 | void |
| 289 | stopprofclock(struct proc *p) |
| 290 | { |
| 291 | |
| 292 | KASSERT(mutex_owned(&p->p_stmutex)); |
| 293 | |
| 294 | if (p->p_stflag & PST_PROFIL) { |
| 295 | p->p_stflag &= ~PST_PROFIL; |
| 296 | /* |
| 297 | * This is only necessary if using the clock as the |
| 298 | * profiling source. |
| 299 | */ |
| 300 | if (--profprocs == 0 && stathz != 0) |
| 301 | psdiv = 1; |
| 302 | } |
| 303 | } |
| 304 | |
| 305 | #if defined(PERFCTRS) |
| 306 | /* |
| 307 | * Independent profiling "tick" in case we're using a separate |
| 308 | * clock or profiling event source. Currently, that's just |
| 309 | * performance counters--hence the wrapper. |
| 310 | */ |
| 311 | void |
| 312 | proftick(struct clockframe *frame) |
| 313 | { |
| 314 | #ifdef GPROF |
| 315 | struct gmonparam *g; |
| 316 | intptr_t i; |
| 317 | #endif |
| 318 | struct lwp *l; |
| 319 | struct proc *p; |
| 320 | |
| 321 | l = curcpu()->ci_data.cpu_onproc; |
| 322 | p = (l ? l->l_proc : NULL); |
| 323 | if (CLKF_USERMODE(frame)) { |
| 324 | mutex_spin_enter(&p->p_stmutex); |
| 325 | if (p->p_stflag & PST_PROFIL) |
| 326 | addupc_intr(l, CLKF_PC(frame)); |
| 327 | mutex_spin_exit(&p->p_stmutex); |
| 328 | } else { |
| 329 | #ifdef GPROF |
| 330 | g = &_gmonparam; |
| 331 | if (g->state == GMON_PROF_ON) { |
| 332 | i = CLKF_PC(frame) - g->lowpc; |
| 333 | if (i < g->textsize) { |
| 334 | i /= HISTFRACTION * sizeof(*g->kcount); |
| 335 | g->kcount[i]++; |
| 336 | } |
| 337 | } |
| 338 | #endif |
| 339 | #ifdef LWP_PC |
| 340 | if (p != NULL && (p->p_stflag & PST_PROFIL) != 0) |
| 341 | addupc_intr(l, LWP_PC(l)); |
| 342 | #endif |
| 343 | } |
| 344 | } |
| 345 | #endif |
| 346 | |
| 347 | void |
| 348 | schedclock(struct lwp *l) |
| 349 | { |
| 350 | if ((l->l_flag & LW_IDLE) != 0) |
| 351 | return; |
| 352 | |
| 353 | sched_schedclock(l); |
| 354 | } |
| 355 | |
| 356 | /* |
| 357 | * Statistics clock. Grab profile sample, and if divider reaches 0, |
| 358 | * do process and kernel statistics. |
| 359 | */ |
| 360 | void |
| 361 | statclock(struct clockframe *frame) |
| 362 | { |
| 363 | #ifdef GPROF |
| 364 | struct gmonparam *g; |
| 365 | intptr_t i; |
| 366 | #endif |
| 367 | struct cpu_info *ci = curcpu(); |
| 368 | struct schedstate_percpu *spc = &ci->ci_schedstate; |
| 369 | struct proc *p; |
| 370 | struct lwp *l; |
| 371 | |
| 372 | /* |
| 373 | * Notice changes in divisor frequency, and adjust clock |
| 374 | * frequency accordingly. |
| 375 | */ |
| 376 | if (spc->spc_psdiv != psdiv) { |
| 377 | spc->spc_psdiv = psdiv; |
| 378 | spc->spc_pscnt = psdiv; |
| 379 | if (psdiv == 1) { |
| 380 | setstatclockrate(stathz); |
| 381 | } else { |
| 382 | setstatclockrate(profhz); |
| 383 | } |
| 384 | } |
| 385 | l = ci->ci_data.cpu_onproc; |
| 386 | if ((l->l_flag & LW_IDLE) != 0) { |
| 387 | /* |
| 388 | * don't account idle lwps as swapper. |
| 389 | */ |
| 390 | p = NULL; |
| 391 | } else { |
| 392 | p = l->l_proc; |
| 393 | mutex_spin_enter(&p->p_stmutex); |
| 394 | } |
| 395 | |
| 396 | if (CLKF_USERMODE(frame)) { |
| 397 | if ((p->p_stflag & PST_PROFIL) && profsrc == PROFSRC_CLOCK) |
| 398 | addupc_intr(l, CLKF_PC(frame)); |
| 399 | if (--spc->spc_pscnt > 0) { |
| 400 | mutex_spin_exit(&p->p_stmutex); |
| 401 | return; |
| 402 | } |
| 403 | |
| 404 | /* |
| 405 | * Came from user mode; CPU was in user state. |
| 406 | * If this process is being profiled record the tick. |
| 407 | */ |
| 408 | p->p_uticks++; |
| 409 | if (p->p_nice > NZERO) |
| 410 | spc->spc_cp_time[CP_NICE]++; |
| 411 | else |
| 412 | spc->spc_cp_time[CP_USER]++; |
| 413 | } else { |
| 414 | #ifdef GPROF |
| 415 | /* |
| 416 | * Kernel statistics are just like addupc_intr, only easier. |
| 417 | */ |
| 418 | g = &_gmonparam; |
| 419 | if (profsrc == PROFSRC_CLOCK && g->state == GMON_PROF_ON) { |
| 420 | i = CLKF_PC(frame) - g->lowpc; |
| 421 | if (i < g->textsize) { |
| 422 | i /= HISTFRACTION * sizeof(*g->kcount); |
| 423 | g->kcount[i]++; |
| 424 | } |
| 425 | } |
| 426 | #endif |
| 427 | #ifdef LWP_PC |
| 428 | if (p != NULL && profsrc == PROFSRC_CLOCK && |
| 429 | (p->p_stflag & PST_PROFIL)) { |
| 430 | addupc_intr(l, LWP_PC(l)); |
| 431 | } |
| 432 | #endif |
| 433 | if (--spc->spc_pscnt > 0) { |
| 434 | if (p != NULL) |
| 435 | mutex_spin_exit(&p->p_stmutex); |
| 436 | return; |
| 437 | } |
| 438 | /* |
| 439 | * Came from kernel mode, so we were: |
| 440 | * - handling an interrupt, |
| 441 | * - doing syscall or trap work on behalf of the current |
| 442 | * user process, or |
| 443 | * - spinning in the idle loop. |
| 444 | * Whichever it is, charge the time as appropriate. |
| 445 | * Note that we charge interrupts to the current process, |
| 446 | * regardless of whether they are ``for'' that process, |
| 447 | * so that we know how much of its real time was spent |
| 448 | * in ``non-process'' (i.e., interrupt) work. |
| 449 | */ |
| 450 | if (CLKF_INTR(frame) || (curlwp->l_pflag & LP_INTR) != 0) { |
| 451 | if (p != NULL) { |
| 452 | p->p_iticks++; |
| 453 | } |
| 454 | spc->spc_cp_time[CP_INTR]++; |
| 455 | } else if (p != NULL) { |
| 456 | p->p_sticks++; |
| 457 | spc->spc_cp_time[CP_SYS]++; |
| 458 | } else { |
| 459 | spc->spc_cp_time[CP_IDLE]++; |
| 460 | } |
| 461 | } |
| 462 | spc->spc_pscnt = psdiv; |
| 463 | |
| 464 | if (p != NULL) { |
| 465 | atomic_inc_uint(&l->l_cpticks); |
| 466 | mutex_spin_exit(&p->p_stmutex); |
| 467 | } |
| 468 | } |
| 469 | |
| 470 | /* |
| 471 | * sysctl helper routine for kern.clockrate. Assembles a struct on |
| 472 | * the fly to be returned to the caller. |
| 473 | */ |
| 474 | static int |
| 475 | sysctl_kern_clockrate(SYSCTLFN_ARGS) |
| 476 | { |
| 477 | struct clockinfo clkinfo; |
| 478 | struct sysctlnode node; |
| 479 | |
| 480 | clkinfo.tick = tick; |
| 481 | clkinfo.tickadj = tickadj; |
| 482 | clkinfo.hz = hz; |
| 483 | clkinfo.profhz = profhz; |
| 484 | clkinfo.stathz = stathz ? stathz : hz; |
| 485 | |
| 486 | node = *rnode; |
| 487 | node.sysctl_data = &clkinfo; |
| 488 | return (sysctl_lookup(SYSCTLFN_CALL(&node))); |
| 489 | } |
| 490 | |