1 /*- 2 * Copyright (c) 2008, Jeffrey Roberson <jeff@freebsd.org> 3 * All rights reserved. 4 * 5 * Copyright (c) 2008 Nokia Corporation 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice unmodified, this list of conditions, and the following 13 * disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 23 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 27 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 28 * 29 */ 30 31 #include <sys/cdefs.h> 32 __FBSDID("$FreeBSD$"); 33 34 #include <sys/param.h> 35 #include <sys/systm.h> 36 #include <sys/sysproto.h> 37 #include <sys/kernel.h> 38 #include <sys/lock.h> 39 #include <sys/malloc.h> 40 #include <sys/mutex.h> 41 #include <sys/priv.h> 42 #include <sys/proc.h> 43 #include <sys/refcount.h> 44 #include <sys/sched.h> 45 #include <sys/smp.h> 46 #include <sys/syscallsubr.h> 47 #include <sys/cpuset.h> 48 #include <sys/sx.h> 49 #include <sys/refcount.h> 50 #include <sys/queue.h> 51 #include <sys/limits.h> 52 #include <sys/bus.h> 53 #include <sys/interrupt.h> 54 55 #include <vm/uma.h> 56 57 /* 58 * cpusets provide a mechanism for creating and manipulating sets of 59 * processors for the purpose of constraining the scheduling of threads to 60 * specific processors. 61 * 62 * Each process belongs to an identified set, by default this is set 1. Each 63 * thread may further restrict the cpus it may run on to a subset of this 64 * named set. This creates an anonymous set which other threads and processes 65 * may not join by number. 66 * 67 * The named set is referred to herein as the 'base' set to avoid ambiguity. 68 * This set is usually a child of a 'root' set while the anonymous set may 69 * simply be referred to as a mask. In the syscall api these are referred to 70 * as the ROOT, CPUSET, and MASK levels where CPUSET is called 'base' here. 71 * 72 * Threads inherit their set from their creator whether it be anonymous or 73 * not. This means that anonymous sets are immutable because they may be 74 * shared. To modify an anonymous set a new set is created with the desired 75 * mask and the same parent as the existing anonymous set. This gives the 76 * illusion of each thread having a private mask.A 77 * 78 * Via the syscall apis a user may ask to retrieve or modify the root, base, 79 * or mask that is discovered via a pid, tid, or setid. Modifying a set 80 * modifies all numbered and anonymous child sets to comply with the new mask. 81 * Modifying a pid or tid's mask applies only to that tid but must still 82 * exist within the assigned parent set. 83 * 84 * A thread may not be assigned to a a group seperate from other threads in 85 * the process. This is to remove ambiguity when the setid is queried with 86 * a pid argument. There is no other technical limitation. 87 * 88 * This somewhat complex arrangement is intended to make it easy for 89 * applications to query available processors and bind their threads to 90 * specific processors while also allowing administrators to dynamically 91 * reprovision by changing sets which apply to groups of processes. 92 * 93 * A simple application should not concern itself with sets at all and 94 * rather apply masks to its own threads via CPU_WHICH_TID and a -1 id 95 * meaning 'curthread'. It may query availble cpus for that tid with a 96 * getaffinity call using (CPU_LEVEL_CPUSET, CPU_WHICH_PID, -1, ...). 97 */ 98 static uma_zone_t cpuset_zone; 99 static struct mtx cpuset_lock; 100 static struct setlist cpuset_ids; 101 static struct unrhdr *cpuset_unr; 102 static struct cpuset *cpuset_zero; 103 104 cpuset_t *cpuset_root; 105 106 /* 107 * Acquire a reference to a cpuset, all pointers must be tracked with refs. 108 */ 109 struct cpuset * 110 cpuset_ref(struct cpuset *set) 111 { 112 113 refcount_acquire(&set->cs_ref); 114 return (set); 115 } 116 117 /* 118 * Release a reference in a context where it is safe to allocte. 119 */ 120 void 121 cpuset_rel(struct cpuset *set) 122 { 123 cpusetid_t id; 124 125 if (refcount_release(&set->cs_ref) == 0) 126 return; 127 mtx_lock_spin(&cpuset_lock); 128 LIST_REMOVE(set, cs_siblings); 129 id = set->cs_id; 130 if (id != CPUSET_INVALID) 131 LIST_REMOVE(set, cs_link); 132 mtx_unlock_spin(&cpuset_lock); 133 cpuset_rel(set->cs_parent); 134 uma_zfree(cpuset_zone, set); 135 if (id != CPUSET_INVALID) 136 free_unr(cpuset_unr, id); 137 } 138 139 /* 140 * Deferred release must be used when in a context that is not safe to 141 * allocate/free. This places any unreferenced sets on the list 'head'. 142 */ 143 static void 144 cpuset_rel_defer(struct setlist *head, struct cpuset *set) 145 { 146 147 if (refcount_release(&set->cs_ref) == 0) 148 return; 149 mtx_lock_spin(&cpuset_lock); 150 LIST_REMOVE(set, cs_siblings); 151 if (set->cs_id != CPUSET_INVALID) 152 LIST_REMOVE(set, cs_link); 153 LIST_INSERT_HEAD(head, set, cs_link); 154 mtx_unlock_spin(&cpuset_lock); 155 } 156 157 /* 158 * Complete a deferred release. Removes the set from the list provided to 159 * cpuset_rel_defer. 160 */ 161 static void 162 cpuset_rel_complete(struct cpuset *set) 163 { 164 LIST_REMOVE(set, cs_link); 165 cpuset_rel(set->cs_parent); 166 uma_zfree(cpuset_zone, set); 167 } 168 169 /* 170 * Find a set based on an id. Returns it with a ref. 171 */ 172 static struct cpuset * 173 cpuset_lookup(cpusetid_t setid) 174 { 175 struct cpuset *set; 176 177 if (setid == CPUSET_INVALID) 178 return (NULL); 179 mtx_lock_spin(&cpuset_lock); 180 LIST_FOREACH(set, &cpuset_ids, cs_link) 181 if (set->cs_id == setid) 182 break; 183 if (set) 184 cpuset_ref(set); 185 mtx_unlock_spin(&cpuset_lock); 186 return (set); 187 } 188 189 /* 190 * Create a set in the space provided in 'set' with the provided parameters. 191 * The set is returned with a single ref. May return EDEADLK if the set 192 * will have no valid cpu based on restrictions from the parent. 193 */ 194 static int 195 _cpuset_create(struct cpuset *set, struct cpuset *parent, cpuset_t *mask, 196 cpusetid_t id) 197 { 198 199 if (!CPU_OVERLAP(&parent->cs_mask, mask)) 200 return (EDEADLK); 201 CPU_COPY(mask, &set->cs_mask); 202 LIST_INIT(&set->cs_children); 203 refcount_init(&set->cs_ref, 1); 204 set->cs_flags = 0; 205 mtx_lock_spin(&cpuset_lock); 206 CPU_AND(mask, &parent->cs_mask); 207 set->cs_id = id; 208 set->cs_parent = cpuset_ref(parent); 209 LIST_INSERT_HEAD(&parent->cs_children, set, cs_siblings); 210 if (set->cs_id != CPUSET_INVALID) 211 LIST_INSERT_HEAD(&cpuset_ids, set, cs_link); 212 mtx_unlock_spin(&cpuset_lock); 213 214 return (0); 215 } 216 217 /* 218 * Create a new non-anonymous set with the requested parent and mask. May 219 * return failures if the mask is invalid or a new number can not be 220 * allocated. 221 */ 222 static int 223 cpuset_create(struct cpuset **setp, struct cpuset *parent, cpuset_t *mask) 224 { 225 struct cpuset *set; 226 cpusetid_t id; 227 int error; 228 229 id = alloc_unr(cpuset_unr); 230 if (id == -1) 231 return (ENFILE); 232 *setp = set = uma_zalloc(cpuset_zone, M_WAITOK); 233 error = _cpuset_create(set, parent, mask, id); 234 if (error == 0) 235 return (0); 236 free_unr(cpuset_unr, id); 237 uma_zfree(cpuset_zone, set); 238 239 return (error); 240 } 241 242 /* 243 * Recursively check for errors that would occur from applying mask to 244 * the tree of sets starting at 'set'. Checks for sets that would become 245 * empty as well as RDONLY flags. 246 */ 247 static int 248 cpuset_testupdate(struct cpuset *set, cpuset_t *mask) 249 { 250 struct cpuset *nset; 251 cpuset_t newmask; 252 int error; 253 254 mtx_assert(&cpuset_lock, MA_OWNED); 255 if (set->cs_flags & CPU_SET_RDONLY) 256 return (EPERM); 257 if (!CPU_OVERLAP(&set->cs_mask, mask)) 258 return (EDEADLK); 259 CPU_COPY(&set->cs_mask, &newmask); 260 CPU_AND(&newmask, mask); 261 error = 0; 262 LIST_FOREACH(nset, &set->cs_children, cs_siblings) 263 if ((error = cpuset_testupdate(nset, &newmask)) != 0) 264 break; 265 return (error); 266 } 267 268 /* 269 * Applies the mask 'mask' without checking for empty sets or permissions. 270 */ 271 static void 272 cpuset_update(struct cpuset *set, cpuset_t *mask) 273 { 274 struct cpuset *nset; 275 276 mtx_assert(&cpuset_lock, MA_OWNED); 277 CPU_AND(&set->cs_mask, mask); 278 LIST_FOREACH(nset, &set->cs_children, cs_siblings) 279 cpuset_update(nset, &set->cs_mask); 280 281 return; 282 } 283 284 /* 285 * Modify the set 'set' to use a copy of the mask provided. Apply this new 286 * mask to restrict all children in the tree. Checks for validity before 287 * applying the changes. 288 */ 289 static int 290 cpuset_modify(struct cpuset *set, cpuset_t *mask) 291 { 292 struct cpuset *root; 293 int error; 294 295 error = suser(curthread); 296 if (error) 297 return (error); 298 /* 299 * Verify that we have access to this set of 300 * cpus. 301 */ 302 root = set->cs_parent; 303 if (root && !CPU_SUBSET(&root->cs_mask, mask)) 304 return (EINVAL); 305 mtx_lock_spin(&cpuset_lock); 306 error = cpuset_testupdate(set, mask); 307 if (error) 308 goto out; 309 cpuset_update(set, mask); 310 CPU_COPY(mask, &set->cs_mask); 311 out: 312 mtx_unlock_spin(&cpuset_lock); 313 314 return (error); 315 } 316 317 /* 318 * Walks up the tree from 'set' to find the root. Returns the root 319 * referenced. 320 */ 321 static struct cpuset * 322 cpuset_refroot(struct cpuset *set) 323 { 324 325 for (; set->cs_parent != NULL; set = set->cs_parent) 326 if (set->cs_flags & CPU_SET_ROOT) 327 break; 328 cpuset_ref(set); 329 330 return (set); 331 } 332 333 /* 334 * Find the first non-anonymous set starting from 'set'. Returns this set 335 * referenced. May return the passed in set with an extra ref if it is 336 * not anonymous. 337 */ 338 static struct cpuset * 339 cpuset_refbase(struct cpuset *set) 340 { 341 342 if (set->cs_id == CPUSET_INVALID) 343 set = set->cs_parent; 344 cpuset_ref(set); 345 346 return (set); 347 } 348 349 /* 350 * Resolve the 'which' parameter of several cpuset apis. 351 * 352 * For WHICH_PID and WHICH_TID return a locked proc and valid proc/tid. Also 353 * checks for permission via p_cansched(). 354 * 355 * For WHICH_SET returns a valid set with a new reference. 356 * 357 * -1 may be supplied for any argument to mean the current proc/thread or 358 * the base set of the current thread. May fail with ESRCH/EPERM. 359 */ 360 static int 361 cpuset_which(cpuwhich_t which, id_t id, struct proc **pp, struct thread **tdp, 362 struct cpuset **setp) 363 { 364 struct cpuset *set; 365 struct thread *td; 366 struct proc *p; 367 int error; 368 369 *pp = p = NULL; 370 *tdp = td = NULL; 371 *setp = set = NULL; 372 switch (which) { 373 case CPU_WHICH_PID: 374 if (id == -1) { 375 PROC_LOCK(curproc); 376 p = curproc; 377 break; 378 } 379 if ((p = pfind(id)) == NULL) 380 return (ESRCH); 381 break; 382 case CPU_WHICH_TID: 383 if (id == -1) { 384 PROC_LOCK(curproc); 385 p = curproc; 386 td = curthread; 387 break; 388 } 389 sx_slock(&allproc_lock); 390 FOREACH_PROC_IN_SYSTEM(p) { 391 PROC_LOCK(p); 392 FOREACH_THREAD_IN_PROC(p, td) 393 if (td->td_tid == id) 394 break; 395 if (td != NULL) 396 break; 397 PROC_UNLOCK(p); 398 } 399 sx_sunlock(&allproc_lock); 400 if (td == NULL) 401 return (ESRCH); 402 break; 403 case CPU_WHICH_CPUSET: 404 if (id == -1) { 405 thread_lock(curthread); 406 set = cpuset_refbase(curthread->td_cpuset); 407 thread_unlock(curthread); 408 } else 409 set = cpuset_lookup(id); 410 if (set) { 411 *setp = set; 412 return (0); 413 } 414 return (ESRCH); 415 case CPU_WHICH_IRQ: 416 return (0); 417 default: 418 return (EINVAL); 419 } 420 error = p_cansched(curthread, p); 421 if (error) { 422 PROC_UNLOCK(p); 423 return (error); 424 } 425 if (td == NULL) 426 td = FIRST_THREAD_IN_PROC(p); 427 *pp = p; 428 *tdp = td; 429 return (0); 430 } 431 432 /* 433 * Create an anonymous set with the provided mask in the space provided by 434 * 'fset'. If the passed in set is anonymous we use its parent otherwise 435 * the new set is a child of 'set'. 436 */ 437 static int 438 cpuset_shadow(struct cpuset *set, struct cpuset *fset, cpuset_t *mask) 439 { 440 struct cpuset *parent; 441 442 if (set->cs_id == CPUSET_INVALID) 443 parent = set->cs_parent; 444 else 445 parent = set; 446 if (!CPU_SUBSET(&parent->cs_mask, mask)) 447 return (EDEADLK); 448 return (_cpuset_create(fset, parent, mask, CPUSET_INVALID)); 449 } 450 451 /* 452 * Handle two cases for replacing the base set or mask of an entire process. 453 * 454 * 1) Set is non-null and mask is null. This reparents all anonymous sets 455 * to the provided set and replaces all non-anonymous td_cpusets with the 456 * provided set. 457 * 2) Mask is non-null and set is null. This replaces or creates anonymous 458 * sets for every thread with the existing base as a parent. 459 * 460 * This is overly complicated because we can't allocate while holding a 461 * spinlock and spinlocks must be held while changing and examining thread 462 * state. 463 */ 464 static int 465 cpuset_setproc(pid_t pid, struct cpuset *set, cpuset_t *mask) 466 { 467 struct setlist freelist; 468 struct setlist droplist; 469 struct cpuset *tdset; 470 struct cpuset *nset; 471 struct thread *td; 472 struct proc *p; 473 int threads; 474 int nfree; 475 int error; 476 /* 477 * The algorithm requires two passes due to locking considerations. 478 * 479 * 1) Lookup the process and acquire the locks in the required order. 480 * 2) If enough cpusets have not been allocated release the locks and 481 * allocate them. Loop. 482 */ 483 LIST_INIT(&freelist); 484 LIST_INIT(&droplist); 485 nfree = 0; 486 for (;;) { 487 error = cpuset_which(CPU_WHICH_PID, pid, &p, &td, &nset); 488 if (error) 489 goto out; 490 if (nfree >= p->p_numthreads) 491 break; 492 threads = p->p_numthreads; 493 PROC_UNLOCK(p); 494 for (; nfree < threads; nfree++) { 495 nset = uma_zalloc(cpuset_zone, M_WAITOK); 496 LIST_INSERT_HEAD(&freelist, nset, cs_link); 497 } 498 } 499 PROC_LOCK_ASSERT(p, MA_OWNED); 500 /* 501 * Now that the appropriate locks are held and we have enough cpusets, 502 * make sure the operation will succeed before applying changes. The 503 * proc lock prevents td_cpuset from changing between calls. 504 */ 505 error = 0; 506 FOREACH_THREAD_IN_PROC(p, td) { 507 thread_lock(td); 508 tdset = td->td_cpuset; 509 /* 510 * Verify that a new mask doesn't specify cpus outside of 511 * the set the thread is a member of. 512 */ 513 if (mask) { 514 if (tdset->cs_id == CPUSET_INVALID) 515 tdset = tdset->cs_parent; 516 if (!CPU_SUBSET(&tdset->cs_mask, mask)) 517 error = EDEADLK; 518 /* 519 * Verify that a new set won't leave an existing thread 520 * mask without a cpu to run on. It can, however, restrict 521 * the set. 522 */ 523 } else if (tdset->cs_id == CPUSET_INVALID) { 524 if (!CPU_OVERLAP(&set->cs_mask, &tdset->cs_mask)) 525 error = EDEADLK; 526 } 527 thread_unlock(td); 528 if (error) 529 goto unlock_out; 530 } 531 /* 532 * Replace each thread's cpuset while using deferred release. We 533 * must do this because the thread lock must be held while operating 534 * on the thread and this limits the type of operations allowed. 535 */ 536 FOREACH_THREAD_IN_PROC(p, td) { 537 thread_lock(td); 538 /* 539 * If we presently have an anonymous set or are applying a 540 * mask we must create an anonymous shadow set. That is 541 * either parented to our existing base or the supplied set. 542 * 543 * If we have a base set with no anonymous shadow we simply 544 * replace it outright. 545 */ 546 tdset = td->td_cpuset; 547 if (tdset->cs_id == CPUSET_INVALID || mask) { 548 nset = LIST_FIRST(&freelist); 549 LIST_REMOVE(nset, cs_link); 550 if (mask) 551 error = cpuset_shadow(tdset, nset, mask); 552 else 553 error = _cpuset_create(nset, set, 554 &tdset->cs_mask, CPUSET_INVALID); 555 if (error) { 556 LIST_INSERT_HEAD(&freelist, nset, cs_link); 557 thread_unlock(td); 558 break; 559 } 560 } else 561 nset = cpuset_ref(set); 562 cpuset_rel_defer(&droplist, tdset); 563 td->td_cpuset = nset; 564 sched_affinity(td); 565 thread_unlock(td); 566 } 567 unlock_out: 568 PROC_UNLOCK(p); 569 out: 570 while ((nset = LIST_FIRST(&droplist)) != NULL) 571 cpuset_rel_complete(nset); 572 while ((nset = LIST_FIRST(&freelist)) != NULL) { 573 LIST_REMOVE(nset, cs_link); 574 uma_zfree(cpuset_zone, nset); 575 } 576 return (error); 577 } 578 579 /* 580 * Apply an anonymous mask to a single thread. 581 */ 582 int 583 cpuset_setthread(lwpid_t id, cpuset_t *mask) 584 { 585 struct cpuset *nset; 586 struct cpuset *set; 587 struct thread *td; 588 struct proc *p; 589 int error; 590 591 nset = uma_zalloc(cpuset_zone, M_WAITOK); 592 error = cpuset_which(CPU_WHICH_TID, id, &p, &td, &set); 593 if (error) 594 goto out; 595 set = NULL; 596 thread_lock(td); 597 error = cpuset_shadow(td->td_cpuset, nset, mask); 598 if (error == 0) { 599 set = td->td_cpuset; 600 td->td_cpuset = nset; 601 sched_affinity(td); 602 nset = NULL; 603 } 604 thread_unlock(td); 605 PROC_UNLOCK(p); 606 if (set) 607 cpuset_rel(set); 608 out: 609 if (nset) 610 uma_zfree(cpuset_zone, nset); 611 return (error); 612 } 613 614 /* 615 * Creates the cpuset for thread0. We make two sets: 616 * 617 * 0 - The root set which should represent all valid processors in the 618 * system. It is initially created with a mask of all processors 619 * because we don't know what processors are valid until cpuset_init() 620 * runs. This set is immutable. 621 * 1 - The default set which all processes are a member of until changed. 622 * This allows an administrator to move all threads off of given cpus to 623 * dedicate them to high priority tasks or save power etc. 624 */ 625 struct cpuset * 626 cpuset_thread0(void) 627 { 628 struct cpuset *set; 629 int error; 630 631 cpuset_zone = uma_zcreate("cpuset", sizeof(struct cpuset), NULL, NULL, 632 NULL, NULL, UMA_ALIGN_PTR, 0); 633 mtx_init(&cpuset_lock, "cpuset", NULL, MTX_SPIN | MTX_RECURSE); 634 /* 635 * Create the root system set for the whole machine. Doesn't use 636 * cpuset_create() due to NULL parent. 637 */ 638 set = uma_zalloc(cpuset_zone, M_WAITOK | M_ZERO); 639 set->cs_mask.__bits[0] = -1; 640 LIST_INIT(&set->cs_children); 641 LIST_INSERT_HEAD(&cpuset_ids, set, cs_link); 642 set->cs_ref = 1; 643 set->cs_flags = CPU_SET_ROOT; 644 cpuset_zero = set; 645 cpuset_root = &set->cs_mask; 646 /* 647 * Now derive a default, modifiable set from that to give out. 648 */ 649 set = uma_zalloc(cpuset_zone, M_WAITOK); 650 error = _cpuset_create(set, cpuset_zero, &cpuset_zero->cs_mask, 1); 651 KASSERT(error == 0, ("Error creating default set: %d\n", error)); 652 /* 653 * Initialize the unit allocator. 0 and 1 are allocated above. 654 */ 655 cpuset_unr = new_unrhdr(2, INT_MAX, NULL); 656 657 return (set); 658 } 659 660 /* 661 * This is called once the final set of system cpus is known. Modifies 662 * the root set and all children and mark the root readonly. 663 */ 664 static void 665 cpuset_init(void *arg) 666 { 667 cpuset_t mask; 668 669 CPU_ZERO(&mask); 670 #ifdef SMP 671 mask.__bits[0] = all_cpus; 672 #else 673 mask.__bits[0] = 1; 674 #endif 675 if (cpuset_modify(cpuset_zero, &mask)) 676 panic("Can't set initial cpuset mask.\n"); 677 cpuset_zero->cs_flags |= CPU_SET_RDONLY; 678 } 679 SYSINIT(cpuset, SI_SUB_SMP, SI_ORDER_ANY, cpuset_init, NULL); 680 681 #ifndef _SYS_SYSPROTO_H_ 682 struct cpuset_args { 683 cpusetid_t *setid; 684 }; 685 #endif 686 int 687 cpuset(struct thread *td, struct cpuset_args *uap) 688 { 689 struct cpuset *root; 690 struct cpuset *set; 691 int error; 692 693 thread_lock(td); 694 root = cpuset_refroot(td->td_cpuset); 695 thread_unlock(td); 696 error = cpuset_create(&set, root, &root->cs_mask); 697 cpuset_rel(root); 698 if (error) 699 return (error); 700 error = copyout(&set->cs_id, uap->setid, sizeof(set->cs_id)); 701 if (error == 0) 702 error = cpuset_setproc(-1, set, NULL); 703 cpuset_rel(set); 704 return (error); 705 } 706 707 #ifndef _SYS_SYSPROTO_H_ 708 struct cpuset_setid_args { 709 cpuwhich_t which; 710 id_t id; 711 cpusetid_t setid; 712 }; 713 #endif 714 int 715 cpuset_setid(struct thread *td, struct cpuset_setid_args *uap) 716 { 717 struct cpuset *set; 718 int error; 719 720 /* 721 * Presently we only support per-process sets. 722 */ 723 if (uap->which != CPU_WHICH_PID) 724 return (EINVAL); 725 set = cpuset_lookup(uap->setid); 726 if (set == NULL) 727 return (ESRCH); 728 error = cpuset_setproc(uap->id, set, NULL); 729 cpuset_rel(set); 730 return (error); 731 } 732 733 #ifndef _SYS_SYSPROTO_H_ 734 struct cpuset_getid_args { 735 cpulevel_t level; 736 cpuwhich_t which; 737 id_t id; 738 cpusetid_t *setid; 739 #endif 740 int 741 cpuset_getid(struct thread *td, struct cpuset_getid_args *uap) 742 { 743 struct cpuset *nset; 744 struct cpuset *set; 745 struct thread *ttd; 746 struct proc *p; 747 cpusetid_t id; 748 int error; 749 750 if (uap->level == CPU_LEVEL_WHICH && uap->which != CPU_WHICH_CPUSET) 751 return (EINVAL); 752 error = cpuset_which(uap->which, uap->id, &p, &ttd, &set); 753 if (error) 754 return (error); 755 switch (uap->which) { 756 case CPU_WHICH_TID: 757 case CPU_WHICH_PID: 758 thread_lock(ttd); 759 set = cpuset_refbase(ttd->td_cpuset); 760 thread_unlock(ttd); 761 PROC_UNLOCK(p); 762 break; 763 case CPU_WHICH_CPUSET: 764 break; 765 case CPU_WHICH_IRQ: 766 return (EINVAL); 767 } 768 switch (uap->level) { 769 case CPU_LEVEL_ROOT: 770 nset = cpuset_refroot(set); 771 cpuset_rel(set); 772 set = nset; 773 break; 774 case CPU_LEVEL_CPUSET: 775 break; 776 case CPU_LEVEL_WHICH: 777 break; 778 } 779 id = set->cs_id; 780 cpuset_rel(set); 781 if (error == 0) 782 error = copyout(&id, uap->setid, sizeof(id)); 783 784 return (error); 785 } 786 787 #ifndef _SYS_SYSPROTO_H_ 788 struct cpuset_getaffinity_args { 789 cpulevel_t level; 790 cpuwhich_t which; 791 id_t id; 792 size_t cpusetsize; 793 cpuset_t *mask; 794 }; 795 #endif 796 int 797 cpuset_getaffinity(struct thread *td, struct cpuset_getaffinity_args *uap) 798 { 799 struct thread *ttd; 800 struct cpuset *nset; 801 struct cpuset *set; 802 struct proc *p; 803 cpuset_t *mask; 804 int error; 805 size_t size; 806 807 if (uap->cpusetsize < sizeof(cpuset_t) || 808 uap->cpusetsize > CPU_MAXSIZE / NBBY) 809 return (ERANGE); 810 size = uap->cpusetsize; 811 mask = malloc(size, M_TEMP, M_WAITOK | M_ZERO); 812 error = cpuset_which(uap->which, uap->id, &p, &ttd, &set); 813 if (error) 814 goto out; 815 switch (uap->level) { 816 case CPU_LEVEL_ROOT: 817 case CPU_LEVEL_CPUSET: 818 switch (uap->which) { 819 case CPU_WHICH_TID: 820 case CPU_WHICH_PID: 821 thread_lock(ttd); 822 set = cpuset_ref(ttd->td_cpuset); 823 thread_unlock(ttd); 824 break; 825 case CPU_WHICH_CPUSET: 826 break; 827 case CPU_WHICH_IRQ: 828 error = EINVAL; 829 goto out; 830 } 831 if (uap->level == CPU_LEVEL_ROOT) 832 nset = cpuset_refroot(set); 833 else 834 nset = cpuset_refbase(set); 835 CPU_COPY(&nset->cs_mask, mask); 836 cpuset_rel(nset); 837 break; 838 case CPU_LEVEL_WHICH: 839 switch (uap->which) { 840 case CPU_WHICH_TID: 841 thread_lock(ttd); 842 CPU_COPY(&ttd->td_cpuset->cs_mask, mask); 843 thread_unlock(ttd); 844 break; 845 case CPU_WHICH_PID: 846 FOREACH_THREAD_IN_PROC(p, ttd) { 847 thread_lock(ttd); 848 CPU_OR(mask, &ttd->td_cpuset->cs_mask); 849 thread_unlock(ttd); 850 } 851 break; 852 case CPU_WHICH_CPUSET: 853 CPU_COPY(&set->cs_mask, mask); 854 break; 855 case CPU_WHICH_IRQ: 856 error = intr_getaffinity(uap->id, mask); 857 break; 858 } 859 break; 860 default: 861 error = EINVAL; 862 break; 863 } 864 if (set) 865 cpuset_rel(set); 866 if (p) 867 PROC_UNLOCK(p); 868 if (error == 0) 869 error = copyout(mask, uap->mask, size); 870 out: 871 free(mask, M_TEMP); 872 return (error); 873 } 874 875 #ifndef _SYS_SYSPROTO_H_ 876 struct cpuset_setaffinity_args { 877 cpulevel_t level; 878 cpuwhich_t which; 879 id_t id; 880 size_t cpusetsize; 881 const cpuset_t *mask; 882 }; 883 #endif 884 int 885 cpuset_setaffinity(struct thread *td, struct cpuset_setaffinity_args *uap) 886 { 887 struct cpuset *nset; 888 struct cpuset *set; 889 struct thread *ttd; 890 struct proc *p; 891 cpuset_t *mask; 892 int error; 893 894 if (uap->cpusetsize < sizeof(cpuset_t) || 895 uap->cpusetsize > CPU_MAXSIZE / NBBY) 896 return (ERANGE); 897 mask = malloc(uap->cpusetsize, M_TEMP, M_WAITOK | M_ZERO); 898 error = copyin(uap->mask, mask, uap->cpusetsize); 899 if (error) 900 goto out; 901 /* 902 * Verify that no high bits are set. 903 */ 904 if (uap->cpusetsize > sizeof(cpuset_t)) { 905 char *end; 906 char *cp; 907 908 end = cp = (char *)&mask->__bits; 909 end += uap->cpusetsize; 910 cp += sizeof(cpuset_t); 911 while (cp != end) 912 if (*cp++ != 0) { 913 error = EINVAL; 914 goto out; 915 } 916 917 } 918 switch (uap->level) { 919 case CPU_LEVEL_ROOT: 920 case CPU_LEVEL_CPUSET: 921 error = cpuset_which(uap->which, uap->id, &p, &ttd, &set); 922 if (error) 923 break; 924 switch (uap->which) { 925 case CPU_WHICH_TID: 926 case CPU_WHICH_PID: 927 thread_lock(ttd); 928 set = cpuset_ref(ttd->td_cpuset); 929 thread_unlock(ttd); 930 PROC_UNLOCK(p); 931 break; 932 case CPU_WHICH_CPUSET: 933 break; 934 case CPU_WHICH_IRQ: 935 error = EINVAL; 936 goto out; 937 } 938 if (uap->level == CPU_LEVEL_ROOT) 939 nset = cpuset_refroot(set); 940 else 941 nset = cpuset_refbase(set); 942 error = cpuset_modify(nset, mask); 943 cpuset_rel(nset); 944 cpuset_rel(set); 945 break; 946 case CPU_LEVEL_WHICH: 947 switch (uap->which) { 948 case CPU_WHICH_TID: 949 error = cpuset_setthread(uap->id, mask); 950 break; 951 case CPU_WHICH_PID: 952 error = cpuset_setproc(uap->id, NULL, mask); 953 break; 954 case CPU_WHICH_CPUSET: 955 error = cpuset_which(CPU_WHICH_CPUSET, uap->id, &p, 956 &ttd, &set); 957 if (error == 0) { 958 error = cpuset_modify(set, mask); 959 cpuset_rel(set); 960 } 961 break; 962 case CPU_WHICH_IRQ: 963 error = intr_setaffinity(uap->id, mask); 964 break; 965 default: 966 error = EINVAL; 967 break; 968 } 969 break; 970 default: 971 error = EINVAL; 972 break; 973 } 974 out: 975 free(mask, M_TEMP); 976 return (error); 977 } 978