Home | History | Annotate | Line # | Download | only in kern
kern_cpu.c revision 1.37
      1 /*	$NetBSD: kern_cpu.c,v 1.37 2008/10/31 00:36:22 rmind Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 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 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 #include <sys/cdefs.h>
     59 __KERNEL_RCSID(0, "$NetBSD: kern_cpu.c,v 1.37 2008/10/31 00:36:22 rmind Exp $");
     60 
     61 #include <sys/param.h>
     62 #include <sys/systm.h>
     63 #include <sys/idle.h>
     64 #include <sys/sched.h>
     65 #include <sys/intr.h>
     66 #include <sys/conf.h>
     67 #include <sys/cpu.h>
     68 #include <sys/cpuio.h>
     69 #include <sys/proc.h>
     70 #include <sys/percpu.h>
     71 #include <sys/kernel.h>
     72 #include <sys/kauth.h>
     73 #include <sys/xcall.h>
     74 #include <sys/pool.h>
     75 #include <sys/kmem.h>
     76 #include <sys/select.h>
     77 #include <sys/namei.h>
     78 #include <sys/callout.h>
     79 
     80 #include <uvm/uvm_extern.h>
     81 
     82 void	cpuctlattach(int);
     83 
     84 static void	cpu_xc_online(struct cpu_info *);
     85 static void	cpu_xc_offline(struct cpu_info *);
     86 
     87 dev_type_ioctl(cpuctl_ioctl);
     88 
     89 const struct cdevsw cpuctl_cdevsw = {
     90 	nullopen, nullclose, nullread, nullwrite, cpuctl_ioctl,
     91 	nullstop, notty, nopoll, nommap, nokqfilter,
     92 	D_OTHER | D_MPSAFE
     93 };
     94 
     95 kmutex_t cpu_lock;
     96 int	ncpu;
     97 int	ncpuonline;
     98 bool	mp_online;
     99 struct	cpuqueue cpu_queue = CIRCLEQ_HEAD_INITIALIZER(cpu_queue);
    100 
    101 static struct cpu_info *cpu_infos[MAXCPUS];
    102 
    103 int
    104 mi_cpu_attach(struct cpu_info *ci)
    105 {
    106 	int error;
    107 
    108 	ci->ci_index = ncpu;
    109 	cpu_infos[cpu_index(ci)] = ci;
    110 	CIRCLEQ_INSERT_TAIL(&cpu_queue, ci, ci_data.cpu_qchain);
    111 	TAILQ_INIT(&ci->ci_data.cpu_ld_locks);
    112 	__cpu_simple_lock_init(&ci->ci_data.cpu_ld_lock);
    113 
    114 	sched_cpuattach(ci);
    115 
    116 	error = create_idle_lwp(ci);
    117 	if (error != 0) {
    118 		/* XXX revert sched_cpuattach */
    119 		return error;
    120 	}
    121 
    122 	if (ci == curcpu())
    123 		ci->ci_data.cpu_onproc = curlwp;
    124 	else
    125 		ci->ci_data.cpu_onproc = ci->ci_data.cpu_idlelwp;
    126 
    127 	percpu_init_cpu(ci);
    128 	softint_init(ci);
    129 	callout_init_cpu(ci);
    130 	xc_init_cpu(ci);
    131 	pool_cache_cpu_init(ci);
    132 	selsysinit(ci);
    133 	cache_cpu_init(ci);
    134 	TAILQ_INIT(&ci->ci_data.cpu_biodone);
    135 	ncpu++;
    136 	ncpuonline++;
    137 
    138 	return 0;
    139 }
    140 
    141 void
    142 cpuctlattach(int dummy)
    143 {
    144 
    145 }
    146 
    147 int
    148 cpuctl_ioctl(dev_t dev, u_long cmd, void *data, int flag, lwp_t *l)
    149 {
    150 	CPU_INFO_ITERATOR cii;
    151 	cpustate_t *cs;
    152 	struct cpu_info *ci;
    153 	int error, i;
    154 	u_int id;
    155 
    156 	error = 0;
    157 
    158 	mutex_enter(&cpu_lock);
    159 	switch (cmd) {
    160 	case IOC_CPU_SETSTATE:
    161 		cs = data;
    162 		error = kauth_authorize_system(l->l_cred,
    163 		    KAUTH_SYSTEM_CPU, KAUTH_REQ_SYSTEM_CPU_SETSTATE, cs, NULL,
    164 		    NULL);
    165 		if (error != 0)
    166 			break;
    167 		if (cs->cs_id >= __arraycount(cpu_infos) ||
    168 		    (ci = cpu_lookup(cs->cs_id)) == NULL) {
    169 			error = ESRCH;
    170 			break;
    171 		}
    172 		if (!cs->cs_intr) {
    173 			error = EOPNOTSUPP;
    174 			break;
    175 		}
    176 		error = cpu_setstate(ci, cs->cs_online);
    177 		break;
    178 
    179 	case IOC_CPU_GETSTATE:
    180 		cs = data;
    181 		id = cs->cs_id;
    182 		memset(cs, 0, sizeof(*cs));
    183 		cs->cs_id = id;
    184 		if (cs->cs_id >= __arraycount(cpu_infos) ||
    185 		    (ci = cpu_lookup(id)) == NULL) {
    186 			error = ESRCH;
    187 			break;
    188 		}
    189 		if ((ci->ci_schedstate.spc_flags & SPCF_OFFLINE) != 0)
    190 			cs->cs_online = false;
    191 		else
    192 			cs->cs_online = true;
    193 		cs->cs_intr = true;
    194 		cs->cs_lastmod = ci->ci_schedstate.spc_lastmod;
    195 		break;
    196 
    197 	case IOC_CPU_MAPID:
    198 		i = 0;
    199 		for (CPU_INFO_FOREACH(cii, ci)) {
    200 			if (i++ == *(int *)data)
    201 				break;
    202 		}
    203 		if (ci == NULL)
    204 			error = ESRCH;
    205 		else
    206 			*(int *)data = ci->ci_cpuid;
    207 		break;
    208 
    209 	case IOC_CPU_GETCOUNT:
    210 		*(int *)data = ncpu;
    211 		break;
    212 
    213 	default:
    214 		error = ENOTTY;
    215 		break;
    216 	}
    217 	mutex_exit(&cpu_lock);
    218 
    219 	return error;
    220 }
    221 
    222 struct cpu_info *
    223 cpu_lookup(u_int idx)
    224 {
    225 	struct cpu_info *ci = cpu_infos[idx];
    226 
    227 	KASSERT(idx < __arraycount(cpu_infos));
    228 	KASSERT(ci == NULL || cpu_index(ci) == idx);
    229 
    230 	return ci;
    231 }
    232 
    233 static void
    234 cpu_xc_offline(struct cpu_info *ci)
    235 {
    236 	struct schedstate_percpu *spc, *mspc = NULL;
    237 	struct cpu_info *target_ci;
    238 	struct lwp *l;
    239 	CPU_INFO_ITERATOR cii;
    240 	int s;
    241 
    242 	/*
    243 	 * Thread which sent unicast (separate context) is holding
    244 	 * the cpu_lock for us.
    245 	 */
    246 	spc = &ci->ci_schedstate;
    247 	s = splsched();
    248 	spc->spc_flags |= SPCF_OFFLINE;
    249 	splx(s);
    250 
    251 	/* Take the first available CPU for the migration */
    252 	for (CPU_INFO_FOREACH(cii, target_ci)) {
    253 		mspc = &target_ci->ci_schedstate;
    254 		if ((mspc->spc_flags & SPCF_OFFLINE) == 0)
    255 			break;
    256 	}
    257 	KASSERT(target_ci != NULL);
    258 
    259 	/*
    260 	 * Migrate all non-bound threads to the other CPU.  Note that this
    261 	 * runs from the xcall thread, thus handling of LSONPROC is not needed.
    262 	 */
    263 	mutex_enter(proc_lock);
    264 	LIST_FOREACH(l, &alllwp, l_list) {
    265 		struct cpu_info *mci;
    266 
    267 		lwp_lock(l);
    268 		if (l->l_cpu != ci || (l->l_pflag & (LP_BOUND | LP_INTR))) {
    269 			lwp_unlock(l);
    270 			continue;
    271 		}
    272 		/* Normal case - no affinity */
    273 		if ((l->l_flag & LW_AFFINITY) == 0) {
    274 			lwp_migrate(l, target_ci);
    275 			continue;
    276 		}
    277 		/* Affinity is set, find an online CPU in the set */
    278 		KASSERT(l->l_affinity != NULL);
    279 		for (CPU_INFO_FOREACH(cii, mci)) {
    280 			mspc = &mci->ci_schedstate;
    281 			if ((mspc->spc_flags & SPCF_OFFLINE) == 0 &&
    282 			    kcpuset_isset(cpu_index(mci), l->l_affinity))
    283 				break;
    284 		}
    285 		if (mci == NULL) {
    286 			lwp_unlock(l);
    287 			mutex_exit(proc_lock);
    288 			goto fail;
    289 		}
    290 		lwp_migrate(l, mci);
    291 	}
    292 	mutex_exit(proc_lock);
    293 
    294 #ifdef __HAVE_MD_CPU_OFFLINE
    295 	cpu_offline_md();
    296 #endif
    297 	return;
    298 fail:
    299 	/* Just unset the SPCF_OFFLINE flag, caller will check */
    300 	s = splsched();
    301 	spc->spc_flags &= ~SPCF_OFFLINE;
    302 	splx(s);
    303 }
    304 
    305 static void
    306 cpu_xc_online(struct cpu_info *ci)
    307 {
    308 	struct schedstate_percpu *spc;
    309 	int s;
    310 
    311 	spc = &ci->ci_schedstate;
    312 	s = splsched();
    313 	spc->spc_flags &= ~SPCF_OFFLINE;
    314 	splx(s);
    315 }
    316 
    317 int
    318 cpu_setstate(struct cpu_info *ci, bool online)
    319 {
    320 	struct schedstate_percpu *spc;
    321 	CPU_INFO_ITERATOR cii;
    322 	struct cpu_info *ci2;
    323 	uint64_t where;
    324 	xcfunc_t func;
    325 	int nonline;
    326 
    327 	spc = &ci->ci_schedstate;
    328 
    329 	KASSERT(mutex_owned(&cpu_lock));
    330 
    331 	if (online) {
    332 		if ((spc->spc_flags & SPCF_OFFLINE) == 0)
    333 			return 0;
    334 		func = (xcfunc_t)cpu_xc_online;
    335 		ncpuonline++;
    336 	} else {
    337 		if ((spc->spc_flags & SPCF_OFFLINE) != 0)
    338 			return 0;
    339 		nonline = 0;
    340 		/*
    341 		 * Ensure that at least one CPU within the processor set
    342 		 * stays online.  Revisit this later.
    343 		 */
    344 		for (CPU_INFO_FOREACH(cii, ci2)) {
    345 			if ((ci2->ci_schedstate.spc_flags & SPCF_OFFLINE) != 0)
    346 				continue;
    347 			if (ci2->ci_schedstate.spc_psid != spc->spc_psid)
    348 				continue;
    349 			nonline++;
    350 		}
    351 		if (nonline == 1)
    352 			return EBUSY;
    353 		func = (xcfunc_t)cpu_xc_offline;
    354 		ncpuonline--;
    355 	}
    356 
    357 	where = xc_unicast(0, func, ci, NULL, ci);
    358 	xc_wait(where);
    359 	if (online) {
    360 		KASSERT((spc->spc_flags & SPCF_OFFLINE) == 0);
    361 	} else if ((spc->spc_flags & SPCF_OFFLINE) == 0) {
    362 		/* If was not set offline, then it is busy */
    363 		return EBUSY;
    364 	}
    365 
    366 	spc->spc_lastmod = time_second;
    367 	return 0;
    368 }
    369