Home | History | Annotate | Line # | Download | only in kern
sys_sched.c revision 1.16
      1 /*	$NetBSD: sys_sched.c,v 1.16 2008/02/22 22:32:49 rmind Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 2008, Mindaugas Rasiukevicius <rmind at NetBSD org>
      5  * All rights reserved.
      6  *
      7  * Redistribution and use in source and binary forms, with or without
      8  * modification, are permitted provided that the following conditions
      9  * are met:
     10  * 1. Redistributions of source code must retain the above copyright
     11  *    notice, this list of conditions and the following disclaimer.
     12  * 2. Redistributions in binary form must reproduce the above copyright
     13  *    notice, this list of conditions and the following disclaimer in the
     14  *    documentation and/or other materials provided with the distribution.
     15  *
     16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
     17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
     20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     26  * SUCH DAMAGE.
     27  */
     28 
     29 /*
     30  * TODO:
     31  *  - Handle pthread_setschedprio() as defined by POSIX;
     32  *  - Handle sched_yield() case for SCHED_FIFO as defined by POSIX;
     33  */
     34 
     35 #include <sys/cdefs.h>
     36 __KERNEL_RCSID(0, "$NetBSD: sys_sched.c,v 1.16 2008/02/22 22:32:49 rmind Exp $");
     37 
     38 #include <sys/param.h>
     39 
     40 #include <sys/cpu.h>
     41 #include <sys/kauth.h>
     42 #include <sys/kmem.h>
     43 #include <sys/lwp.h>
     44 #include <sys/mutex.h>
     45 #include <sys/proc.h>
     46 #include <sys/pset.h>
     47 #include <sys/sched.h>
     48 #include <sys/syscallargs.h>
     49 #include <sys/sysctl.h>
     50 #include <sys/systm.h>
     51 #include <sys/types.h>
     52 #include <sys/unistd.h>
     53 
     54 /*
     55  * Convert user priority or the in-kernel priority or convert the current
     56  * priority to the appropriate range according to the policy change.
     57  */
     58 static pri_t
     59 convert_pri(lwp_t *l, int policy, pri_t pri)
     60 {
     61 	int delta = 0;
     62 
     63 	switch (policy) {
     64 	case SCHED_OTHER:
     65 		delta = PRI_USER;
     66 		break;
     67 	case SCHED_FIFO:
     68 	case SCHED_RR:
     69 		delta = PRI_USER_RT;
     70 		break;
     71 	default:
     72 		panic("upri_to_kpri");
     73 	}
     74 
     75 	if (pri != PRI_NONE) {
     76 		/* Convert user priority to the in-kernel */
     77 		KASSERT(pri >= SCHED_PRI_MIN && pri <= SCHED_PRI_MAX);
     78 		return pri + delta;
     79 	}
     80 	if (l->l_class == policy)
     81 		return l->l_priority;
     82 
     83 	/* Change the current priority to the appropriate range */
     84 	if (l->l_class == SCHED_OTHER) {
     85 		KASSERT(policy == SCHED_FIFO || policy == SCHED_RR);
     86 		return l->l_priority + delta;
     87 	}
     88 	if (policy == SCHED_OTHER) {
     89 		KASSERT(l->l_class == SCHED_FIFO || l->l_class == SCHED_RR);
     90 		return l->l_priority - delta;
     91 	}
     92 	KASSERT(l->l_class != SCHED_OTHER && policy != SCHED_OTHER);
     93 	return l->l_class;
     94 }
     95 
     96 /*
     97  * Set scheduling parameters.
     98  */
     99 int
    100 sys__sched_setparam(struct lwp *l, const struct sys__sched_setparam_args *uap,
    101     register_t *retval)
    102 {
    103 	/* {
    104 		syscallarg(pid_t) pid;
    105 		syscallarg(lwpid_t) lid;
    106 		syscallarg(int) policy;
    107 		syscallarg(const struct sched_param *) params;
    108 	} */
    109 	struct sched_param param;
    110 	struct proc *p;
    111 	struct lwp *t;
    112 	lwpid_t lid;
    113 	u_int lcnt;
    114 	int policy;
    115 	pri_t pri;
    116 	int error;
    117 
    118 	/* Get the parameters from the user-space */
    119 	error = copyin(SCARG(uap, params), &param, sizeof(param));
    120 	if (error) {
    121 		return error;
    122 	}
    123 	pri = param.sched_priority;
    124 	policy = SCARG(uap, policy);
    125 
    126 	/* If no parameters specified, just return (this should not happen) */
    127 	if (pri == PRI_NONE && policy == SCHED_NONE)
    128 		return 0;
    129 
    130 	/* Validate scheduling class */
    131 	if (policy != SCHED_NONE && (policy < SCHED_OTHER || policy > SCHED_RR))
    132 		return EINVAL;
    133 
    134 	/* Validate priority */
    135 	if (pri != PRI_NONE && (pri < SCHED_PRI_MIN || pri > SCHED_PRI_MAX))
    136 		return EINVAL;
    137 
    138 	if (SCARG(uap, pid) != 0) {
    139 		/* Find the process */
    140 		p = p_find(SCARG(uap, pid), PFIND_UNLOCK_FAIL);
    141 		if (p == NULL)
    142 			return ESRCH;
    143 		mutex_enter(&p->p_smutex);
    144 		mutex_exit(&proclist_lock);
    145 		/* Disallow modification of system processes */
    146 		if (p->p_flag & PK_SYSTEM) {
    147 			mutex_exit(&p->p_smutex);
    148 			return EPERM;
    149 		}
    150 	} else {
    151 		/* Use the calling process */
    152 		p = l->l_proc;
    153 		mutex_enter(&p->p_smutex);
    154 	}
    155 
    156 	/* Find the LWP(s) */
    157 	lcnt = 0;
    158 	lid = SCARG(uap, lid);
    159 	LIST_FOREACH(t, &p->p_lwps, l_sibling) {
    160 		pri_t kpri;
    161 		int lpolicy;
    162 
    163 		if (lid && lid != t->l_lid)
    164 			continue;
    165 		lcnt++;
    166 		KASSERT(pri != PRI_NONE || policy != SCHED_NONE);
    167 		lwp_lock(t);
    168 
    169 		if (policy == SCHED_NONE)
    170 			lpolicy = t->l_class;
    171 		else
    172 			lpolicy = policy;
    173 
    174 		/*
    175 		 * Note that, priority may need to be changed to get into
    176 		 * the correct priority range of the new scheduling class.
    177 		 */
    178 		kpri = convert_pri(t, lpolicy, pri);
    179 
    180 		/* Check the permission */
    181 		error = kauth_authorize_process(l->l_cred,
    182 		    KAUTH_PROCESS_SCHEDULER_SETPARAM, p, t, KAUTH_ARG(lpolicy),
    183 		    KAUTH_ARG(kpri));
    184 		if (error) {
    185 			lwp_unlock(t);
    186 			break;
    187 		}
    188 
    189 		/* Set the scheduling class */
    190 		if (policy != SCHED_NONE)
    191 			t->l_class = policy;
    192 
    193 		/* Change the priority */
    194 		if (t->l_priority != kpri)
    195 			lwp_changepri(t, kpri);
    196 
    197 		lwp_unlock(t);
    198 	}
    199 	mutex_exit(&p->p_smutex);
    200 	return (lcnt == 0) ? ESRCH : error;
    201 }
    202 
    203 /*
    204  * Get scheduling parameters.
    205  */
    206 int
    207 sys__sched_getparam(struct lwp *l, const struct sys__sched_getparam_args *uap,
    208     register_t *retval)
    209 {
    210 	/* {
    211 		syscallarg(pid_t) pid;
    212 		syscallarg(lwpid_t) lid;
    213 		syscallarg(int *) policy;
    214 		syscallarg(struct sched_param *) params;
    215 	} */
    216 	struct sched_param param;
    217 	struct lwp *t;
    218 	int error, policy;
    219 
    220 	/* Locks the LWP */
    221 	t = lwp_find2(SCARG(uap, pid), SCARG(uap, lid));
    222 	if (t == NULL)
    223 		return ESRCH;
    224 
    225 	/* Check the permission */
    226 	error = kauth_authorize_process(l->l_cred,
    227 	    KAUTH_PROCESS_SCHEDULER_GETPARAM, t->l_proc, NULL, NULL, NULL);
    228 	if (error != 0) {
    229 		lwp_unlock(t);
    230 		return error;
    231 	}
    232 
    233 	param.sched_priority = t->l_priority;
    234 	policy = t->l_class;
    235 	lwp_unlock(t);
    236 
    237 	switch (policy) {
    238 	case SCHED_OTHER:
    239 		param.sched_priority -= PRI_USER;
    240 		break;
    241 	case SCHED_RR:
    242 	case SCHED_FIFO:
    243 		param.sched_priority -= PRI_USER_RT;
    244 		break;
    245 	}
    246 	error = copyout(&param, SCARG(uap, params), sizeof(param));
    247 	if (error == 0 && SCARG(uap, policy) != NULL)
    248 		error = copyout(&policy, SCARG(uap, policy), sizeof(int));
    249 	return error;
    250 }
    251 
    252 /*
    253  * Set affinity.
    254  */
    255 int
    256 sys__sched_setaffinity(struct lwp *l,
    257     const struct sys__sched_setaffinity_args *uap, register_t *retval)
    258 {
    259 	/* {
    260 		syscallarg(pid_t) pid;
    261 		syscallarg(lwpid_t) lid;
    262 		syscallarg(size_t) size;
    263 		syscallarg(void *) cpuset;
    264 	} */
    265 	cpuset_t *cpuset;
    266 	struct cpu_info *ci = NULL;
    267 	struct proc *p;
    268 	struct lwp *t;
    269 	CPU_INFO_ITERATOR cii;
    270 	lwpid_t lid;
    271 	u_int lcnt;
    272 	int error;
    273 
    274 	/* Allocate the CPU set, and get it from userspace */
    275 	cpuset = kmem_zalloc(sizeof(cpuset_t), KM_SLEEP);
    276 	error = copyin(SCARG(uap, cpuset), cpuset,
    277 	    min(SCARG(uap, size), sizeof(cpuset_t)));
    278 	if (error)
    279 		goto error;
    280 
    281 	/* Look for a CPU in the set */
    282 	for (CPU_INFO_FOREACH(cii, ci))
    283 		if (CPU_ISSET(cpu_index(ci), cpuset))
    284 			break;
    285 	if (ci == NULL) {
    286 		/* Empty set */
    287 		kmem_free(cpuset, sizeof(cpuset_t));
    288 		cpuset = NULL;
    289 	}
    290 
    291 	if (SCARG(uap, pid) != 0) {
    292 		/* Find the process */
    293 		p = p_find(SCARG(uap, pid), PFIND_UNLOCK_FAIL);
    294 		if (p == NULL) {
    295 			error = ESRCH;
    296 			goto error;
    297 		}
    298 		mutex_enter(&p->p_smutex);
    299 		mutex_exit(&proclist_lock);
    300 	} else {
    301 		/* Use the calling process */
    302 		p = l->l_proc;
    303 		mutex_enter(&p->p_smutex);
    304 	}
    305 
    306 	/*
    307 	 * Check the permission.
    308 	 * Disallow modification of system processes.
    309 	 */
    310 	error = kauth_authorize_process(l->l_cred,
    311 	    KAUTH_PROCESS_SCHEDULER_SETAFFINITY, p, NULL, NULL, NULL);
    312 	if (error != 0) {
    313 		mutex_exit(&p->p_smutex);
    314 		goto error;
    315 	}
    316 	if ((p->p_flag & PK_SYSTEM) != 0) {
    317 		mutex_exit(&p->p_smutex);
    318 		error = EPERM;
    319 		goto error;
    320 	}
    321 
    322 	/* Find the LWP(s) */
    323 	lcnt = 0;
    324 	lid = SCARG(uap, lid);
    325 	LIST_FOREACH(t, &p->p_lwps, l_sibling) {
    326 		if (lid && lid != t->l_lid)
    327 			continue;
    328 		lwp_lock(t);
    329 		if (cpuset) {
    330 			/* Set the affinity flag and new CPU set */
    331 			t->l_flag |= LW_AFFINITY;
    332 			memcpy(&t->l_affinity, cpuset, sizeof(cpuset_t));
    333 			/* Migrate to another CPU, unlocks LWP */
    334 			lwp_migrate(t, ci);
    335 		} else {
    336 			/* Unset the affinity flag */
    337 			t->l_flag &= ~LW_AFFINITY;
    338 			lwp_unlock(t);
    339 		}
    340 		lcnt++;
    341 	}
    342 	mutex_exit(&p->p_smutex);
    343 	if (lcnt == 0)
    344 		error = ESRCH;
    345 error:
    346 	if (cpuset != NULL)
    347 		kmem_free(cpuset, sizeof(cpuset_t));
    348 	return error;
    349 }
    350 
    351 /*
    352  * Get affinity.
    353  */
    354 int
    355 sys__sched_getaffinity(struct lwp *l,
    356     const struct sys__sched_getaffinity_args *uap, register_t *retval)
    357 {
    358 	/* {
    359 		syscallarg(pid_t) pid;
    360 		syscallarg(lwpid_t) lid;
    361 		syscallarg(size_t) size;
    362 		syscallarg(void *) cpuset;
    363 	} */
    364 	struct lwp *t;
    365 	void *cpuset;
    366 	int error;
    367 
    368 	if (SCARG(uap, size) <= 0)
    369 		return EINVAL;
    370 	cpuset = kmem_zalloc(sizeof(cpuset_t), KM_SLEEP);
    371 
    372 	/* Locks the LWP */
    373 	t = lwp_find2(SCARG(uap, pid), SCARG(uap, lid));
    374 	if (t == NULL) {
    375 		kmem_free(cpuset, sizeof(cpuset_t));
    376 		return ESRCH;
    377 	}
    378 	/* Check the permission */
    379 	if (kauth_authorize_process(l->l_cred,
    380 	    KAUTH_PROCESS_SCHEDULER_GETAFFINITY, t->l_proc, NULL, NULL, NULL)) {
    381 		lwp_unlock(t);
    382 		kmem_free(cpuset, sizeof(cpuset_t));
    383 		return EPERM;
    384 	}
    385 	if (t->l_flag & LW_AFFINITY)
    386 		memcpy(cpuset, &t->l_affinity, sizeof(cpuset_t));
    387 	lwp_unlock(t);
    388 
    389 	error = copyout(cpuset, SCARG(uap, cpuset),
    390 	    min(SCARG(uap, size), sizeof(cpuset_t)));
    391 
    392 	kmem_free(cpuset, sizeof(cpuset_t));
    393 	return error;
    394 }
    395 
    396 /*
    397  * Yield.
    398  */
    399 int
    400 sys_sched_yield(struct lwp *l, const void *v, register_t *retval)
    401 {
    402 
    403 	yield();
    404 	return 0;
    405 }
    406 
    407 /*
    408  * Sysctl nodes and initialization.
    409  */
    410 SYSCTL_SETUP(sysctl_sched_setup, "sysctl sched setup")
    411 {
    412 	const struct sysctlnode *node = NULL;
    413 
    414 	sysctl_createv(clog, 0, NULL, NULL,
    415 		CTLFLAG_PERMANENT,
    416 		CTLTYPE_NODE, "kern", NULL,
    417 		NULL, 0, NULL, 0,
    418 		CTL_KERN, CTL_EOL);
    419 	sysctl_createv(clog, 0, NULL, NULL,
    420 		CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE,
    421 		CTLTYPE_INT, "posix_sched",
    422 		SYSCTL_DESCR("Version of IEEE Std 1003.1 and its "
    423 			     "Process Scheduling option to which the "
    424 			     "system attempts to conform"),
    425 		NULL, _POSIX_PRIORITY_SCHEDULING, NULL, 0,
    426 		CTL_KERN, CTL_CREATE, CTL_EOL);
    427 	sysctl_createv(clog, 0, NULL, &node,
    428 		CTLFLAG_PERMANENT,
    429 		CTLTYPE_NODE, "sched",
    430 		SYSCTL_DESCR("Scheduler options"),
    431 		NULL, 0, NULL, 0,
    432 		CTL_KERN, CTL_CREATE, CTL_EOL);
    433 
    434 	if (node == NULL)
    435 		return;
    436 
    437 	sysctl_createv(clog, 0, &node, NULL,
    438 		CTLFLAG_PERMANENT | CTLFLAG_IMMEDIATE,
    439 		CTLTYPE_INT, "pri_min",
    440 		SYSCTL_DESCR("Minimal POSIX real-time priority"),
    441 		NULL, SCHED_PRI_MIN, NULL, 0,
    442 		CTL_CREATE, CTL_EOL);
    443 	sysctl_createv(clog, 0, &node, NULL,
    444 		CTLFLAG_PERMANENT | CTLFLAG_IMMEDIATE,
    445 		CTLTYPE_INT, "pri_max",
    446 		SYSCTL_DESCR("Minimal POSIX real-time priority"),
    447 		NULL, SCHED_PRI_MAX, NULL, 0,
    448 		CTL_CREATE, CTL_EOL);
    449 }
    450