1 /* $NetBSD: linux_sched.c,v 1.63 2010/07/07 01:30:35 chs Exp $ */ 2 3 /*- 4 * Copyright (c) 1999 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; by Matthias Scheler. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 /* 34 * Linux compatibility module. Try to deal with scheduler related syscalls. 35 */ 36 37 #include <sys/cdefs.h> 38 __KERNEL_RCSID(0, "$NetBSD: linux_sched.c,v 1.63 2010/07/07 01:30:35 chs Exp $"); 39 40 #include <sys/param.h> 41 #include <sys/mount.h> 42 #include <sys/proc.h> 43 #include <sys/systm.h> 44 #include <sys/sysctl.h> 45 #include <sys/malloc.h> 46 #include <sys/syscallargs.h> 47 #include <sys/wait.h> 48 #include <sys/kauth.h> 49 #include <sys/ptrace.h> 50 #include <sys/atomic.h> 51 52 #include <sys/cpu.h> 53 54 #include <compat/linux/common/linux_types.h> 55 #include <compat/linux/common/linux_signal.h> 56 #include <compat/linux/common/linux_emuldata.h> 57 #include <compat/linux/common/linux_ipc.h> 58 #include <compat/linux/common/linux_sem.h> 59 #include <compat/linux/common/linux_exec.h> 60 #include <compat/linux/common/linux_machdep.h> 61 62 #include <compat/linux/linux_syscallargs.h> 63 64 #include <compat/linux/common/linux_sched.h> 65 66 static int linux_clone_nptl(struct lwp *, const struct linux_sys_clone_args *, register_t *); 67 68 static void 69 linux_child_return(void *arg) 70 { 71 struct lwp *l = arg; 72 struct proc *p = l->l_proc; 73 struct linux_emuldata *led = l->l_emuldata; 74 void *ctp = led->led_child_tidptr; 75 76 if (ctp) { 77 if (copyout(&p->p_pid, ctp, sizeof(p->p_pid)) != 0) 78 printf("%s: LINUX_CLONE_CHILD_SETTID " 79 "failed (child_tidptr = %p, tid = %d)\n", 80 __func__, ctp, p->p_pid); 81 } 82 child_return(arg); 83 } 84 85 int 86 linux_sys_clone(struct lwp *l, const struct linux_sys_clone_args *uap, register_t *retval) 87 { 88 /* { 89 syscallarg(int) flags; 90 syscallarg(void *) stack; 91 syscallarg(void *) parent_tidptr; 92 syscallarg(void *) tls; 93 syscallarg(void *) child_tidptr; 94 } */ 95 struct proc *p; 96 struct linux_emuldata *led; 97 int flags, sig, error; 98 99 /* 100 * We don't support the Linux CLONE_PID or CLONE_PTRACE flags. 101 */ 102 if (SCARG(uap, flags) & (LINUX_CLONE_PID|LINUX_CLONE_PTRACE)) 103 return (EINVAL); 104 105 /* 106 * Thread group implies shared signals. Shared signals 107 * imply shared VM. This matches what Linux kernel does. 108 */ 109 if (SCARG(uap, flags) & LINUX_CLONE_THREAD 110 && (SCARG(uap, flags) & LINUX_CLONE_SIGHAND) == 0) 111 return (EINVAL); 112 if (SCARG(uap, flags) & LINUX_CLONE_SIGHAND 113 && (SCARG(uap, flags) & LINUX_CLONE_VM) == 0) 114 return (EINVAL); 115 116 /* 117 * The thread group flavor is implemented totally differently. 118 */ 119 if (SCARG(uap, flags) & LINUX_CLONE_THREAD) { 120 return linux_clone_nptl(l, uap, retval); 121 } 122 123 flags = 0; 124 if (SCARG(uap, flags) & LINUX_CLONE_VM) 125 flags |= FORK_SHAREVM; 126 if (SCARG(uap, flags) & LINUX_CLONE_FS) 127 flags |= FORK_SHARECWD; 128 if (SCARG(uap, flags) & LINUX_CLONE_FILES) 129 flags |= FORK_SHAREFILES; 130 if (SCARG(uap, flags) & LINUX_CLONE_SIGHAND) 131 flags |= FORK_SHARESIGS; 132 if (SCARG(uap, flags) & LINUX_CLONE_VFORK) 133 flags |= FORK_PPWAIT; 134 135 sig = SCARG(uap, flags) & LINUX_CLONE_CSIGNAL; 136 if (sig < 0 || sig >= LINUX__NSIG) 137 return (EINVAL); 138 sig = linux_to_native_signo[sig]; 139 140 if (SCARG(uap, flags) & LINUX_CLONE_CHILD_SETTID) { 141 led = l->l_emuldata; 142 led->led_child_tidptr = SCARG(uap, child_tidptr); 143 } 144 145 /* 146 * Note that Linux does not provide a portable way of specifying 147 * the stack area; the caller must know if the stack grows up 148 * or down. So, we pass a stack size of 0, so that the code 149 * that makes this adjustment is a noop. 150 */ 151 if ((error = fork1(l, flags, sig, SCARG(uap, stack), 0, 152 linux_child_return, NULL, retval, &p)) != 0) 153 return error; 154 155 return 0; 156 } 157 158 static int 159 linux_clone_nptl(struct lwp *l, const struct linux_sys_clone_args *uap, register_t *retval) 160 { 161 /* { 162 syscallarg(int) flags; 163 syscallarg(void *) stack; 164 syscallarg(void *) parent_tidptr; 165 syscallarg(void *) tls; 166 syscallarg(void *) child_tidptr; 167 } */ 168 struct proc *p; 169 struct lwp *l2; 170 struct linux_emuldata *led; 171 void *parent_tidptr, *tls, *child_tidptr; 172 struct schedstate_percpu *spc; 173 vaddr_t uaddr; 174 lwpid_t lid; 175 int flags, tnprocs, error; 176 177 p = l->l_proc; 178 flags = SCARG(uap, flags); 179 parent_tidptr = SCARG(uap, parent_tidptr); 180 tls = SCARG(uap, tls); 181 child_tidptr = SCARG(uap, child_tidptr); 182 183 tnprocs = atomic_inc_uint_nv(&nprocs); 184 if (__predict_false(tnprocs >= maxproc) || 185 kauth_authorize_process(l->l_cred, KAUTH_PROCESS_FORK, p, 186 KAUTH_ARG(tnprocs), NULL, NULL) != 0) { 187 atomic_dec_uint(&nprocs); 188 return EAGAIN; 189 } 190 191 uaddr = uvm_uarea_alloc(); 192 if (__predict_false(uaddr == 0)) { 193 atomic_dec_uint(&nprocs); 194 return ENOMEM; 195 } 196 197 error = lwp_create(l, p, uaddr, LWP_DETACHED | LWP_PIDLID, 198 SCARG(uap, stack), 0, child_return, NULL, &l2, 199 l->l_class); 200 if (__predict_false(error)) { 201 atomic_dec_uint(&nprocs); 202 uvm_uarea_free(uaddr); 203 return error; 204 } 205 lid = l2->l_lid; 206 207 /* LINUX_CLONE_CHILD_CLEARTID: clear TID in child's memory on exit() */ 208 if (flags & LINUX_CLONE_CHILD_CLEARTID) { 209 led = l2->l_emuldata; 210 led->led_clear_tid = child_tidptr; 211 } 212 213 /* LINUX_CLONE_PARENT_SETTID: store child's TID in parent's memory */ 214 if (flags & LINUX_CLONE_PARENT_SETTID) { 215 if (copyout(&lid, parent_tidptr, sizeof(lid)) != 0) 216 printf("%s: LINUX_CLONE_PARENT_SETTID " 217 "failed (parent_tidptr = %p tid = %d)\n", 218 __func__, parent_tidptr, lid); 219 } 220 221 /* LINUX_CLONE_CHILD_SETTID: store child's TID in child's memory */ 222 if (flags & LINUX_CLONE_CHILD_SETTID) { 223 if (copyout(&lid, child_tidptr, sizeof(lid)) != 0) 224 printf("%s: LINUX_CLONE_CHILD_SETTID " 225 "failed (child_tidptr = %p, tid = %d)\n", 226 __func__, child_tidptr, lid); 227 } 228 229 if (flags & LINUX_CLONE_SETTLS) { 230 error = LINUX_LWP_SETPRIVATE(l2, tls); 231 if (error) { 232 lwp_exit(l2); 233 return error; 234 } 235 } 236 237 /* 238 * Set the new LWP running, unless the process is stopping, 239 * then the LWP is created stopped. 240 */ 241 mutex_enter(p->p_lock); 242 lwp_lock(l2); 243 spc = &l2->l_cpu->ci_schedstate; 244 if ((l->l_flag & (LW_WREBOOT | LW_WSUSPEND | LW_WEXIT)) == 0) { 245 if (p->p_stat == SSTOP || (p->p_sflag & PS_STOPPING) != 0) { 246 KASSERT(l2->l_wchan == NULL); 247 l2->l_stat = LSSTOP; 248 p->p_nrlwps--; 249 lwp_unlock_to(l2, spc->spc_lwplock); 250 } else { 251 KASSERT(lwp_locked(l2, spc->spc_mutex)); 252 l2->l_stat = LSRUN; 253 sched_enqueue(l2, false); 254 lwp_unlock(l2); 255 } 256 } else { 257 l2->l_stat = LSSUSPENDED; 258 p->p_nrlwps--; 259 lwp_unlock_to(l2, spc->spc_lwplock); 260 } 261 mutex_exit(p->p_lock); 262 263 retval[0] = lid; 264 retval[1] = 0; 265 return 0; 266 } 267 268 /* 269 * linux realtime priority 270 * 271 * - SCHED_RR and SCHED_FIFO tasks have priorities [1,99]. 272 * 273 * - SCHED_OTHER tasks don't have realtime priorities. 274 * in particular, sched_param::sched_priority is always 0. 275 */ 276 277 #define LINUX_SCHED_RTPRIO_MIN 1 278 #define LINUX_SCHED_RTPRIO_MAX 99 279 280 static int 281 sched_linux2native(int linux_policy, struct linux_sched_param *linux_params, 282 int *native_policy, struct sched_param *native_params) 283 { 284 285 switch (linux_policy) { 286 case LINUX_SCHED_OTHER: 287 if (native_policy != NULL) { 288 *native_policy = SCHED_OTHER; 289 } 290 break; 291 292 case LINUX_SCHED_FIFO: 293 if (native_policy != NULL) { 294 *native_policy = SCHED_FIFO; 295 } 296 break; 297 298 case LINUX_SCHED_RR: 299 if (native_policy != NULL) { 300 *native_policy = SCHED_RR; 301 } 302 break; 303 304 default: 305 return EINVAL; 306 } 307 308 if (linux_params != NULL) { 309 int prio = linux_params->sched_priority; 310 311 KASSERT(native_params != NULL); 312 313 if (linux_policy == LINUX_SCHED_OTHER) { 314 if (prio != 0) { 315 return EINVAL; 316 } 317 native_params->sched_priority = PRI_NONE; /* XXX */ 318 } else { 319 if (prio < LINUX_SCHED_RTPRIO_MIN || 320 prio > LINUX_SCHED_RTPRIO_MAX) { 321 return EINVAL; 322 } 323 native_params->sched_priority = 324 (prio - LINUX_SCHED_RTPRIO_MIN) 325 * (SCHED_PRI_MAX - SCHED_PRI_MIN) 326 / (LINUX_SCHED_RTPRIO_MAX - LINUX_SCHED_RTPRIO_MIN) 327 + SCHED_PRI_MIN; 328 } 329 } 330 331 return 0; 332 } 333 334 static int 335 sched_native2linux(int native_policy, struct sched_param *native_params, 336 int *linux_policy, struct linux_sched_param *linux_params) 337 { 338 339 switch (native_policy) { 340 case SCHED_OTHER: 341 if (linux_policy != NULL) { 342 *linux_policy = LINUX_SCHED_OTHER; 343 } 344 break; 345 346 case SCHED_FIFO: 347 if (linux_policy != NULL) { 348 *linux_policy = LINUX_SCHED_FIFO; 349 } 350 break; 351 352 case SCHED_RR: 353 if (linux_policy != NULL) { 354 *linux_policy = LINUX_SCHED_RR; 355 } 356 break; 357 358 default: 359 panic("%s: unknown policy %d\n", __func__, native_policy); 360 } 361 362 if (native_params != NULL) { 363 int prio = native_params->sched_priority; 364 365 KASSERT(prio >= SCHED_PRI_MIN); 366 KASSERT(prio <= SCHED_PRI_MAX); 367 KASSERT(linux_params != NULL); 368 369 #ifdef DEBUG_LINUX 370 printf("native2linux: native: policy %d, priority %d\n", 371 native_policy, prio); 372 #endif 373 374 if (native_policy == SCHED_OTHER) { 375 linux_params->sched_priority = 0; 376 } else { 377 linux_params->sched_priority = 378 (prio - SCHED_PRI_MIN) 379 * (LINUX_SCHED_RTPRIO_MAX - LINUX_SCHED_RTPRIO_MIN) 380 / (SCHED_PRI_MAX - SCHED_PRI_MIN) 381 + LINUX_SCHED_RTPRIO_MIN; 382 } 383 #ifdef DEBUG_LINUX 384 printf("native2linux: linux: policy %d, priority %d\n", 385 -1, linux_params->sched_priority); 386 #endif 387 } 388 389 return 0; 390 } 391 392 int 393 linux_sys_sched_setparam(struct lwp *l, const struct linux_sys_sched_setparam_args *uap, register_t *retval) 394 { 395 /* { 396 syscallarg(linux_pid_t) pid; 397 syscallarg(const struct linux_sched_param *) sp; 398 } */ 399 int error, policy; 400 struct linux_sched_param lp; 401 struct sched_param sp; 402 403 if (SCARG(uap, pid) < 0 || SCARG(uap, sp) == NULL) { 404 error = EINVAL; 405 goto out; 406 } 407 408 error = copyin(SCARG(uap, sp), &lp, sizeof(lp)); 409 if (error) 410 goto out; 411 412 /* We need the current policy in Linux terms. */ 413 error = do_sched_getparam(0, SCARG(uap, pid), &policy, NULL); 414 if (error) 415 goto out; 416 error = sched_native2linux(policy, NULL, &policy, NULL); 417 if (error) 418 goto out; 419 420 error = sched_linux2native(policy, &lp, &policy, &sp); 421 if (error) 422 goto out; 423 424 error = do_sched_setparam(0, SCARG(uap, pid), policy, &sp); 425 if (error) 426 goto out; 427 428 out: 429 return error; 430 } 431 432 int 433 linux_sys_sched_getparam(struct lwp *l, const struct linux_sys_sched_getparam_args *uap, register_t *retval) 434 { 435 /* { 436 syscallarg(linux_pid_t) pid; 437 syscallarg(struct linux_sched_param *) sp; 438 } */ 439 struct linux_sched_param lp; 440 struct sched_param sp; 441 int error, policy; 442 443 if (SCARG(uap, pid) < 0 || SCARG(uap, sp) == NULL) { 444 error = EINVAL; 445 goto out; 446 } 447 448 error = do_sched_getparam(0, SCARG(uap, pid), &policy, &sp); 449 if (error) 450 goto out; 451 #ifdef DEBUG_LINUX 452 printf("getparam: native: policy %d, priority %d\n", 453 policy, sp.sched_priority); 454 #endif 455 456 error = sched_native2linux(policy, &sp, NULL, &lp); 457 if (error) 458 goto out; 459 #ifdef DEBUG_LINUX 460 printf("getparam: linux: policy %d, priority %d\n", 461 policy, lp.sched_priority); 462 #endif 463 464 error = copyout(&lp, SCARG(uap, sp), sizeof(lp)); 465 if (error) 466 goto out; 467 468 out: 469 return error; 470 } 471 472 int 473 linux_sys_sched_setscheduler(struct lwp *l, const struct linux_sys_sched_setscheduler_args *uap, register_t *retval) 474 { 475 /* { 476 syscallarg(linux_pid_t) pid; 477 syscallarg(int) policy; 478 syscallarg(cont struct linux_sched_param *) sp; 479 } */ 480 int error, policy; 481 struct linux_sched_param lp; 482 struct sched_param sp; 483 484 if (SCARG(uap, pid) < 0 || SCARG(uap, sp) == NULL) { 485 error = EINVAL; 486 goto out; 487 } 488 489 error = copyin(SCARG(uap, sp), &lp, sizeof(lp)); 490 if (error) 491 goto out; 492 #ifdef DEBUG_LINUX 493 printf("setscheduler: linux: policy %d, priority %d\n", 494 SCARG(uap, policy), lp.sched_priority); 495 #endif 496 497 error = sched_linux2native(SCARG(uap, policy), &lp, &policy, &sp); 498 if (error) 499 goto out; 500 #ifdef DEBUG_LINUX 501 printf("setscheduler: native: policy %d, priority %d\n", 502 policy, sp.sched_priority); 503 #endif 504 505 error = do_sched_setparam(0, SCARG(uap, pid), policy, &sp); 506 if (error) 507 goto out; 508 509 out: 510 return error; 511 } 512 513 int 514 linux_sys_sched_getscheduler(struct lwp *l, const struct linux_sys_sched_getscheduler_args *uap, register_t *retval) 515 { 516 /* { 517 syscallarg(linux_pid_t) pid; 518 } */ 519 int error, policy; 520 521 *retval = -1; 522 523 error = do_sched_getparam(0, SCARG(uap, pid), &policy, NULL); 524 if (error) 525 goto out; 526 527 error = sched_native2linux(policy, NULL, &policy, NULL); 528 if (error) 529 goto out; 530 531 *retval = policy; 532 533 out: 534 return error; 535 } 536 537 int 538 linux_sys_sched_yield(struct lwp *l, const void *v, register_t *retval) 539 { 540 541 yield(); 542 return 0; 543 } 544 545 int 546 linux_sys_sched_get_priority_max(struct lwp *l, const struct linux_sys_sched_get_priority_max_args *uap, register_t *retval) 547 { 548 /* { 549 syscallarg(int) policy; 550 } */ 551 552 switch (SCARG(uap, policy)) { 553 case LINUX_SCHED_OTHER: 554 *retval = 0; 555 break; 556 case LINUX_SCHED_FIFO: 557 case LINUX_SCHED_RR: 558 *retval = LINUX_SCHED_RTPRIO_MAX; 559 break; 560 default: 561 return EINVAL; 562 } 563 564 return 0; 565 } 566 567 int 568 linux_sys_sched_get_priority_min(struct lwp *l, const struct linux_sys_sched_get_priority_min_args *uap, register_t *retval) 569 { 570 /* { 571 syscallarg(int) policy; 572 } */ 573 574 switch (SCARG(uap, policy)) { 575 case LINUX_SCHED_OTHER: 576 *retval = 0; 577 break; 578 case LINUX_SCHED_FIFO: 579 case LINUX_SCHED_RR: 580 *retval = LINUX_SCHED_RTPRIO_MIN; 581 break; 582 default: 583 return EINVAL; 584 } 585 586 return 0; 587 } 588 589 int 590 linux_sys_exit(struct lwp *l, const struct linux_sys_exit_args *uap, register_t *retval) 591 { 592 593 lwp_exit(l); 594 return 0; 595 } 596 597 #ifndef __m68k__ 598 /* Present on everything but m68k */ 599 int 600 linux_sys_exit_group(struct lwp *l, const struct linux_sys_exit_group_args *uap, register_t *retval) 601 { 602 603 return sys_exit(l, (const void *)uap, retval); 604 } 605 #endif /* !__m68k__ */ 606 607 int 608 linux_sys_set_tid_address(struct lwp *l, const struct linux_sys_set_tid_address_args *uap, register_t *retval) 609 { 610 /* { 611 syscallarg(int *) tidptr; 612 } */ 613 struct linux_emuldata *led; 614 615 led = (struct linux_emuldata *)l->l_emuldata; 616 led->led_clear_tid = SCARG(uap, tid); 617 *retval = l->l_lid; 618 619 return 0; 620 } 621 622 /* ARGUSED1 */ 623 int 624 linux_sys_gettid(struct lwp *l, const void *v, register_t *retval) 625 { 626 627 *retval = l->l_lid; 628 return 0; 629 } 630 631 int 632 linux_sys_sched_getaffinity(struct lwp *l, const struct linux_sys_sched_getaffinity_args *uap, register_t *retval) 633 { 634 /* { 635 syscallarg(linux_pid_t) pid; 636 syscallarg(unsigned int) len; 637 syscallarg(unsigned long *) mask; 638 } */ 639 proc_t *p; 640 unsigned long *lp, *data; 641 int error, size, nb = ncpu; 642 643 /* Unlike Linux, dynamically calculate cpu mask size */ 644 size = sizeof(long) * ((ncpu + LONG_BIT - 1) / LONG_BIT); 645 if (SCARG(uap, len) < size) 646 return EINVAL; 647 648 /* XXX: Pointless check. TODO: Actually implement this. */ 649 mutex_enter(proc_lock); 650 p = proc_find(SCARG(uap, pid)); 651 mutex_exit(proc_lock); 652 if (p == NULL) { 653 return ESRCH; 654 } 655 656 /* 657 * return the actual number of CPU, tag all of them as available 658 * The result is a mask, the first CPU being in the least significant 659 * bit. 660 */ 661 data = kmem_zalloc(size, KM_SLEEP); 662 lp = data; 663 while (nb > LONG_BIT) { 664 *lp++ = ~0UL; 665 nb -= LONG_BIT; 666 } 667 if (nb) 668 *lp = (1 << ncpu) - 1; 669 670 error = copyout(data, SCARG(uap, mask), size); 671 kmem_free(data, size); 672 *retval = size; 673 return error; 674 } 675 676 int 677 linux_sys_sched_setaffinity(struct lwp *l, const struct linux_sys_sched_setaffinity_args *uap, register_t *retval) 678 { 679 /* { 680 syscallarg(linux_pid_t) pid; 681 syscallarg(unsigned int) len; 682 syscallarg(unsigned long *) mask; 683 } */ 684 proc_t *p; 685 686 /* XXX: Pointless check. TODO: Actually implement this. */ 687 mutex_enter(proc_lock); 688 p = proc_find(SCARG(uap, pid)); 689 mutex_exit(proc_lock); 690 if (p == NULL) { 691 return ESRCH; 692 } 693 694 /* Let's ignore it */ 695 #ifdef DEBUG_LINUX 696 printf("linux_sys_sched_setaffinity\n"); 697 #endif 698 return 0; 699 }; 700