1 /* $NetBSD: kern_cpu.c,v 1.92 2020/07/13 13:16:07 jruoho Exp $ */ 2 3 /*- 4 * Copyright (c) 2007, 2008, 2009, 2010, 2012, 2019 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Andrew Doran. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /*- 33 * Copyright (c)2007 YAMAMOTO Takashi, 34 * All rights reserved. 35 * 36 * Redistribution and use in source and binary forms, with or without 37 * modification, are permitted provided that the following conditions 38 * are met: 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 45 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 46 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 47 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 48 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 49 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 50 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 51 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 52 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 53 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 54 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 55 * SUCH DAMAGE. 56 */ 57 58 /* 59 * CPU related routines not shared with rump. 60 */ 61 62 #include <sys/cdefs.h> 63 __KERNEL_RCSID(0, "$NetBSD: kern_cpu.c,v 1.92 2020/07/13 13:16:07 jruoho Exp $"); 64 65 #ifdef _KERNEL_OPT 66 #include "opt_cpu_ucode.h" 67 #endif 68 69 #include <sys/param.h> 70 #include <sys/systm.h> 71 #include <sys/idle.h> 72 #include <sys/sched.h> 73 #include <sys/intr.h> 74 #include <sys/conf.h> 75 #include <sys/cpu.h> 76 #include <sys/cpuio.h> 77 #include <sys/proc.h> 78 #include <sys/percpu.h> 79 #include <sys/kernel.h> 80 #include <sys/kauth.h> 81 #include <sys/xcall.h> 82 #include <sys/pool.h> 83 #include <sys/kmem.h> 84 #include <sys/select.h> 85 #include <sys/namei.h> 86 #include <sys/callout.h> 87 #include <sys/pcu.h> 88 89 #include <uvm/uvm_extern.h> 90 91 #include "ioconf.h" 92 93 /* 94 * If the port has stated that cpu_data is the first thing in cpu_info, 95 * verify that the claim is true. This will prevent them from getting out 96 * of sync. 97 */ 98 #ifdef __HAVE_CPU_DATA_FIRST 99 CTASSERT(offsetof(struct cpu_info, ci_data) == 0); 100 #else 101 CTASSERT(offsetof(struct cpu_info, ci_data) != 0); 102 #endif 103 104 int (*compat_cpuctl_ioctl)(struct lwp *, u_long, void *) = (void *)enosys; 105 106 static void cpu_xc_online(struct cpu_info *, void *); 107 static void cpu_xc_offline(struct cpu_info *, void *); 108 109 dev_type_ioctl(cpuctl_ioctl); 110 111 const struct cdevsw cpuctl_cdevsw = { 112 .d_open = nullopen, 113 .d_close = nullclose, 114 .d_read = nullread, 115 .d_write = nullwrite, 116 .d_ioctl = cpuctl_ioctl, 117 .d_stop = nullstop, 118 .d_tty = notty, 119 .d_poll = nopoll, 120 .d_mmap = nommap, 121 .d_kqfilter = nokqfilter, 122 .d_discard = nodiscard, 123 .d_flag = D_OTHER | D_MPSAFE 124 }; 125 126 int 127 mi_cpu_attach(struct cpu_info *ci) 128 { 129 int error; 130 131 KASSERT(maxcpus > 0); 132 133 if ((ci->ci_index = ncpu) >= maxcpus) 134 panic("Too many CPUs. Increase MAXCPUS?"); 135 kcpuset_set(kcpuset_attached, cpu_index(ci)); 136 137 /* 138 * Create a convenience cpuset of just ourselves. 139 */ 140 kcpuset_create(&ci->ci_data.cpu_kcpuset, true); 141 kcpuset_set(ci->ci_data.cpu_kcpuset, cpu_index(ci)); 142 143 TAILQ_INIT(&ci->ci_data.cpu_ld_locks); 144 __cpu_simple_lock_init(&ci->ci_data.cpu_ld_lock); 145 146 /* This is useful for eg, per-cpu evcnt */ 147 snprintf(ci->ci_data.cpu_name, sizeof(ci->ci_data.cpu_name), "cpu%d", 148 cpu_index(ci)); 149 150 if (__predict_false(cpu_infos == NULL)) { 151 size_t ci_bufsize = (maxcpus + 1) * sizeof(struct cpu_info *); 152 cpu_infos = kmem_zalloc(ci_bufsize, KM_SLEEP); 153 } 154 cpu_infos[cpu_index(ci)] = ci; 155 156 sched_cpuattach(ci); 157 158 error = create_idle_lwp(ci); 159 if (error != 0) { 160 /* XXX revert sched_cpuattach */ 161 return error; 162 } 163 164 if (ci == curcpu()) 165 ci->ci_onproc = curlwp; 166 else 167 ci->ci_onproc = ci->ci_data.cpu_idlelwp; 168 169 percpu_init_cpu(ci); 170 softint_init(ci); 171 callout_init_cpu(ci); 172 xc_init_cpu(ci); 173 pool_cache_cpu_init(ci); 174 selsysinit(ci); 175 cache_cpu_init(ci); 176 TAILQ_INIT(&ci->ci_data.cpu_biodone); 177 ncpu++; 178 ncpuonline++; 179 180 return 0; 181 } 182 183 void 184 cpuctlattach(int dummy __unused) 185 { 186 187 KASSERT(cpu_infos != NULL); 188 } 189 190 int 191 cpuctl_ioctl(dev_t dev, u_long cmd, void *data, int flag, lwp_t *l) 192 { 193 CPU_INFO_ITERATOR cii; 194 cpustate_t *cs; 195 struct cpu_info *ci; 196 int error, i; 197 u_int id; 198 199 error = 0; 200 201 mutex_enter(&cpu_lock); 202 switch (cmd) { 203 case IOC_CPU_SETSTATE: 204 cs = data; 205 error = kauth_authorize_system(l->l_cred, 206 KAUTH_SYSTEM_CPU, KAUTH_REQ_SYSTEM_CPU_SETSTATE, cs, NULL, 207 NULL); 208 if (error != 0) 209 break; 210 if (cs->cs_id >= maxcpus || 211 (ci = cpu_lookup(cs->cs_id)) == NULL) { 212 error = ESRCH; 213 break; 214 } 215 error = cpu_setintr(ci, cs->cs_intr); 216 if (error) 217 break; 218 error = cpu_setstate(ci, cs->cs_online); 219 break; 220 221 case IOC_CPU_GETSTATE: 222 cs = data; 223 id = cs->cs_id; 224 memset(cs, 0, sizeof(*cs)); 225 cs->cs_id = id; 226 if (cs->cs_id >= maxcpus || 227 (ci = cpu_lookup(id)) == NULL) { 228 error = ESRCH; 229 break; 230 } 231 if ((ci->ci_schedstate.spc_flags & SPCF_OFFLINE) != 0) 232 cs->cs_online = false; 233 else 234 cs->cs_online = true; 235 if ((ci->ci_schedstate.spc_flags & SPCF_NOINTR) != 0) 236 cs->cs_intr = false; 237 else 238 cs->cs_intr = true; 239 cs->cs_lastmod = (int32_t)ci->ci_schedstate.spc_lastmod; 240 cs->cs_lastmodhi = (int32_t) 241 (ci->ci_schedstate.spc_lastmod >> 32); 242 cs->cs_intrcnt = cpu_intr_count(ci) + 1; 243 cs->cs_hwid = ci->ci_cpuid; 244 break; 245 246 case IOC_CPU_MAPID: 247 i = 0; 248 for (CPU_INFO_FOREACH(cii, ci)) { 249 if (i++ == *(int *)data) 250 break; 251 } 252 if (ci == NULL) 253 error = ESRCH; 254 else 255 *(int *)data = cpu_index(ci); 256 break; 257 258 case IOC_CPU_GETCOUNT: 259 *(int *)data = ncpu; 260 break; 261 262 #ifdef CPU_UCODE 263 case IOC_CPU_UCODE_GET_VERSION: 264 error = cpu_ucode_get_version((struct cpu_ucode_version *)data); 265 break; 266 267 case IOC_CPU_UCODE_APPLY: 268 error = kauth_authorize_machdep(l->l_cred, 269 KAUTH_MACHDEP_CPU_UCODE_APPLY, 270 NULL, NULL, NULL, NULL); 271 if (error != 0) 272 break; 273 error = cpu_ucode_apply((const struct cpu_ucode *)data); 274 break; 275 #endif 276 277 default: 278 error = (*compat_cpuctl_ioctl)(l, cmd, data); 279 break; 280 } 281 mutex_exit(&cpu_lock); 282 283 return error; 284 } 285 286 struct cpu_info * 287 cpu_lookup(u_int idx) 288 { 289 struct cpu_info *ci; 290 291 /* 292 * cpu_infos is a NULL terminated array of MAXCPUS + 1 entries, 293 * so an index of MAXCPUS here is ok. See mi_cpu_attach. 294 */ 295 KASSERT(idx <= maxcpus); 296 297 if (__predict_false(cpu_infos == NULL)) { 298 KASSERT(idx == 0); 299 return curcpu(); 300 } 301 302 ci = cpu_infos[idx]; 303 KASSERT(ci == NULL || cpu_index(ci) == idx); 304 KASSERTMSG(idx < maxcpus || ci == NULL, "idx %d ci %p", idx, ci); 305 306 return ci; 307 } 308 309 static void 310 cpu_xc_offline(struct cpu_info *ci, void *unused) 311 { 312 struct schedstate_percpu *spc, *mspc = NULL; 313 struct cpu_info *target_ci; 314 struct lwp *l; 315 CPU_INFO_ITERATOR cii; 316 int s; 317 318 /* 319 * Thread that made the cross call (separate context) holds 320 * cpu_lock on our behalf. 321 */ 322 spc = &ci->ci_schedstate; 323 s = splsched(); 324 spc->spc_flags |= SPCF_OFFLINE; 325 splx(s); 326 327 /* Take the first available CPU for the migration. */ 328 for (CPU_INFO_FOREACH(cii, target_ci)) { 329 mspc = &target_ci->ci_schedstate; 330 if ((mspc->spc_flags & SPCF_OFFLINE) == 0) 331 break; 332 } 333 KASSERT(target_ci != NULL); 334 335 /* 336 * Migrate all non-bound threads to the other CPU. Note that this 337 * runs from the xcall thread, thus handling of LSONPROC is not needed. 338 */ 339 mutex_enter(&proc_lock); 340 LIST_FOREACH(l, &alllwp, l_list) { 341 struct cpu_info *mci; 342 343 lwp_lock(l); 344 if (l->l_cpu != ci || (l->l_pflag & (LP_BOUND | LP_INTR))) { 345 lwp_unlock(l); 346 continue; 347 } 348 /* Regular case - no affinity. */ 349 if (l->l_affinity == NULL) { 350 lwp_migrate(l, target_ci); 351 continue; 352 } 353 /* Affinity is set, find an online CPU in the set. */ 354 for (CPU_INFO_FOREACH(cii, mci)) { 355 mspc = &mci->ci_schedstate; 356 if ((mspc->spc_flags & SPCF_OFFLINE) == 0 && 357 kcpuset_isset(l->l_affinity, cpu_index(mci))) 358 break; 359 } 360 if (mci == NULL) { 361 lwp_unlock(l); 362 mutex_exit(&proc_lock); 363 goto fail; 364 } 365 lwp_migrate(l, mci); 366 } 367 mutex_exit(&proc_lock); 368 369 #if PCU_UNIT_COUNT > 0 370 pcu_save_all_on_cpu(); 371 #endif 372 373 #ifdef __HAVE_MD_CPU_OFFLINE 374 cpu_offline_md(); 375 #endif 376 return; 377 fail: 378 /* Just unset the SPCF_OFFLINE flag, caller will check */ 379 s = splsched(); 380 spc->spc_flags &= ~SPCF_OFFLINE; 381 splx(s); 382 } 383 384 static void 385 cpu_xc_online(struct cpu_info *ci, void *unused) 386 { 387 struct schedstate_percpu *spc; 388 int s; 389 390 spc = &ci->ci_schedstate; 391 s = splsched(); 392 spc->spc_flags &= ~SPCF_OFFLINE; 393 splx(s); 394 } 395 396 int 397 cpu_setstate(struct cpu_info *ci, bool online) 398 { 399 struct schedstate_percpu *spc; 400 CPU_INFO_ITERATOR cii; 401 struct cpu_info *ci2; 402 uint64_t where; 403 xcfunc_t func; 404 int nonline; 405 406 spc = &ci->ci_schedstate; 407 408 KASSERT(mutex_owned(&cpu_lock)); 409 410 if (online) { 411 if ((spc->spc_flags & SPCF_OFFLINE) == 0) 412 return 0; 413 func = (xcfunc_t)cpu_xc_online; 414 } else { 415 if ((spc->spc_flags & SPCF_OFFLINE) != 0) 416 return 0; 417 nonline = 0; 418 /* 419 * Ensure that at least one CPU within the processor set 420 * stays online. Revisit this later. 421 */ 422 for (CPU_INFO_FOREACH(cii, ci2)) { 423 if ((ci2->ci_schedstate.spc_flags & SPCF_OFFLINE) != 0) 424 continue; 425 if (ci2->ci_schedstate.spc_psid != spc->spc_psid) 426 continue; 427 nonline++; 428 } 429 if (nonline == 1) 430 return EBUSY; 431 func = (xcfunc_t)cpu_xc_offline; 432 } 433 434 where = xc_unicast(0, func, ci, NULL, ci); 435 xc_wait(where); 436 if (online) { 437 KASSERT((spc->spc_flags & SPCF_OFFLINE) == 0); 438 ncpuonline++; 439 } else { 440 if ((spc->spc_flags & SPCF_OFFLINE) == 0) { 441 /* If was not set offline, then it is busy */ 442 return EBUSY; 443 } 444 ncpuonline--; 445 } 446 447 spc->spc_lastmod = time_second; 448 return 0; 449 } 450 451 #if defined(__HAVE_INTR_CONTROL) 452 static void 453 cpu_xc_intr(struct cpu_info *ci, void *unused) 454 { 455 struct schedstate_percpu *spc; 456 int s; 457 458 spc = &ci->ci_schedstate; 459 s = splsched(); 460 spc->spc_flags &= ~SPCF_NOINTR; 461 splx(s); 462 } 463 464 static void 465 cpu_xc_nointr(struct cpu_info *ci, void *unused) 466 { 467 struct schedstate_percpu *spc; 468 int s; 469 470 spc = &ci->ci_schedstate; 471 s = splsched(); 472 spc->spc_flags |= SPCF_NOINTR; 473 splx(s); 474 } 475 476 int 477 cpu_setintr(struct cpu_info *ci, bool intr) 478 { 479 struct schedstate_percpu *spc; 480 CPU_INFO_ITERATOR cii; 481 struct cpu_info *ci2; 482 uint64_t where; 483 xcfunc_t func; 484 int nintr; 485 486 spc = &ci->ci_schedstate; 487 488 KASSERT(mutex_owned(&cpu_lock)); 489 490 if (intr) { 491 if ((spc->spc_flags & SPCF_NOINTR) == 0) 492 return 0; 493 func = (xcfunc_t)cpu_xc_intr; 494 } else { 495 if (CPU_IS_PRIMARY(ci)) 496 return EINVAL; 497 if ((spc->spc_flags & SPCF_NOINTR) != 0) 498 return 0; 499 /* 500 * Ensure that at least one CPU within the system 501 * is handing device interrupts. 502 */ 503 nintr = 0; 504 for (CPU_INFO_FOREACH(cii, ci2)) { 505 if ((ci2->ci_schedstate.spc_flags & SPCF_NOINTR) != 0) 506 continue; 507 if (ci2 == ci) 508 continue; 509 nintr++; 510 } 511 if (nintr == 0) 512 return EBUSY; 513 func = (xcfunc_t)cpu_xc_nointr; 514 } 515 516 where = xc_unicast(0, func, ci, NULL, ci); 517 xc_wait(where); 518 if (intr) { 519 KASSERT((spc->spc_flags & SPCF_NOINTR) == 0); 520 } else if ((spc->spc_flags & SPCF_NOINTR) == 0) { 521 /* If was not set offline, then it is busy */ 522 return EBUSY; 523 } 524 525 /* Direct interrupts away from the CPU and record the change. */ 526 cpu_intr_redistribute(); 527 spc->spc_lastmod = time_second; 528 return 0; 529 } 530 #else /* __HAVE_INTR_CONTROL */ 531 int 532 cpu_setintr(struct cpu_info *ci, bool intr) 533 { 534 535 return EOPNOTSUPP; 536 } 537 538 u_int 539 cpu_intr_count(struct cpu_info *ci) 540 { 541 542 return 0; /* 0 == "don't know" */ 543 } 544 #endif /* __HAVE_INTR_CONTROL */ 545 546 #ifdef CPU_UCODE 547 int 548 cpu_ucode_load(struct cpu_ucode_softc *sc, const char *fwname) 549 { 550 firmware_handle_t fwh; 551 int error; 552 553 if (sc->sc_blob != NULL) { 554 firmware_free(sc->sc_blob, sc->sc_blobsize); 555 sc->sc_blob = NULL; 556 sc->sc_blobsize = 0; 557 } 558 559 error = cpu_ucode_md_open(&fwh, sc->loader_version, fwname); 560 if (error != 0) { 561 #ifdef DEBUG 562 printf("ucode: firmware_open(%s) failed: %i\n", fwname, error); 563 #endif 564 goto err0; 565 } 566 567 sc->sc_blobsize = firmware_get_size(fwh); 568 if (sc->sc_blobsize == 0) { 569 error = EFTYPE; 570 firmware_close(fwh); 571 goto err0; 572 } 573 sc->sc_blob = firmware_malloc(sc->sc_blobsize); 574 if (sc->sc_blob == NULL) { 575 error = ENOMEM; 576 firmware_close(fwh); 577 goto err0; 578 } 579 580 error = firmware_read(fwh, 0, sc->sc_blob, sc->sc_blobsize); 581 firmware_close(fwh); 582 if (error != 0) 583 goto err1; 584 585 return 0; 586 587 err1: 588 firmware_free(sc->sc_blob, sc->sc_blobsize); 589 sc->sc_blob = NULL; 590 sc->sc_blobsize = 0; 591 err0: 592 return error; 593 } 594 #endif 595