Home | History | Annotate | Line # | Download | only in kern
kern_cpu.c revision 1.20
      1 /*	$NetBSD: kern_cpu.c,v 1.20 2008/02/01 20:01:06 elad Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 2007 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  * 3. All advertising materials mentioning features or use of this software
     19  *    must display the following acknowledgement:
     20  *        This product includes software developed by the NetBSD
     21  *        Foundation, Inc. and its contributors.
     22  * 4. Neither the name of The NetBSD Foundation nor the names of its
     23  *    contributors may be used to endorse or promote products derived
     24  *    from this software without specific prior written permission.
     25  *
     26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     36  * POSSIBILITY OF SUCH DAMAGE.
     37  */
     38 
     39 /*-
     40  * Copyright (c)2007 YAMAMOTO Takashi,
     41  * All rights reserved.
     42  *
     43  * Redistribution and use in source and binary forms, with or without
     44  * modification, are permitted provided that the following conditions
     45  * are met:
     46  * 1. Redistributions of source code must retain the above copyright
     47  *    notice, this list of conditions and the following disclaimer.
     48  * 2. Redistributions in binary form must reproduce the above copyright
     49  *    notice, this list of conditions and the following disclaimer in the
     50  *    documentation and/or other materials provided with the distribution.
     51  *
     52  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
     53  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     54  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     55  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
     56  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     57  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     58  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     59  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     60  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     61  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     62  * SUCH DAMAGE.
     63  */
     64 
     65 #include <sys/cdefs.h>
     66 
     67 __KERNEL_RCSID(0, "$NetBSD: kern_cpu.c,v 1.20 2008/02/01 20:01:06 elad Exp $");
     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 
     84 #include <uvm/uvm_extern.h>
     85 
     86 void	cpuctlattach(int);
     87 
     88 static void	cpu_xc_online(struct cpu_info *);
     89 static void	cpu_xc_offline(struct cpu_info *);
     90 
     91 dev_type_ioctl(cpuctl_ioctl);
     92 
     93 const struct cdevsw cpuctl_cdevsw = {
     94 	nullopen, nullclose, nullread, nullwrite, cpuctl_ioctl,
     95 	nullstop, notty, nopoll, nommap, nokqfilter,
     96 	D_OTHER | D_MPSAFE
     97 };
     98 
     99 kmutex_t cpu_lock;
    100 int	ncpu;
    101 int	ncpuonline;
    102 bool	mp_online;
    103 
    104 static struct cpu_info *cpu_infos[MAXCPUS];
    105 
    106 int
    107 mi_cpu_attach(struct cpu_info *ci)
    108 {
    109 	struct schedstate_percpu *spc = &ci->ci_schedstate;
    110 	int error;
    111 
    112 	ci->ci_index = ncpu;
    113 
    114 	mutex_init(&spc->spc_lwplock, MUTEX_DEFAULT, IPL_SCHED);
    115 	sched_cpuattach(ci);
    116 	uvm_cpu_attach(ci);
    117 
    118 	error = create_idle_lwp(ci);
    119 	if (error != 0) {
    120 		/* XXX revert sched_cpuattach */
    121 		return error;
    122 	}
    123 
    124 	if (ci == curcpu())
    125 		ci->ci_data.cpu_onproc = curlwp;
    126 	else
    127 		ci->ci_data.cpu_onproc = ci->ci_data.cpu_idlelwp;
    128 
    129 	percpu_init_cpu(ci);
    130 	softint_init(ci);
    131 	xc_init_cpu(ci);
    132 	pool_cache_cpu_init(ci);
    133 	TAILQ_INIT(&ci->ci_data.cpu_biodone);
    134 	ncpu++;
    135 	ncpuonline++;
    136 	cpu_infos[cpu_index(ci)] = ci;
    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 ((ci = cpu_lookup(cs->cs_id)) == NULL) {
    168 			error = ESRCH;
    169 			break;
    170 		}
    171 		if (!cs->cs_intr) {
    172 			error = EOPNOTSUPP;
    173 			break;
    174 		}
    175 		error = cpu_setonline(ci, cs->cs_online);
    176 		break;
    177 
    178 	case IOC_CPU_GETSTATE:
    179 		cs = data;
    180 		id = cs->cs_id;
    181 		memset(cs, 0, sizeof(*cs));
    182 		cs->cs_id = id;
    183 		if ((ci = cpu_lookup(id)) == NULL) {
    184 			error = ESRCH;
    185 			break;
    186 		}
    187 		if ((ci->ci_schedstate.spc_flags & SPCF_OFFLINE) != 0)
    188 			cs->cs_online = false;
    189 		else
    190 			cs->cs_online = true;
    191 		cs->cs_intr = true;
    192 		cs->cs_lastmod = ci->ci_schedstate.spc_lastmod;
    193 		break;
    194 
    195 	case IOC_CPU_MAPID:
    196 		i = 0;
    197 		for (CPU_INFO_FOREACH(cii, ci)) {
    198 			if (i++ == *(int *)data)
    199 				break;
    200 		}
    201 		if (ci == NULL)
    202 			error = ESRCH;
    203 		else
    204 			*(int *)data = ci->ci_cpuid;
    205 		break;
    206 
    207 	case IOC_CPU_GETCOUNT:
    208 		*(int *)data = ncpu;
    209 		break;
    210 
    211 	default:
    212 		error = ENOTTY;
    213 		break;
    214 	}
    215 	mutex_exit(&cpu_lock);
    216 
    217 	return error;
    218 }
    219 
    220 struct cpu_info *
    221 cpu_lookup(cpuid_t id)
    222 {
    223 	CPU_INFO_ITERATOR cii;
    224 	struct cpu_info *ci;
    225 
    226 	for (CPU_INFO_FOREACH(cii, ci)) {
    227 		if (ci->ci_cpuid == id)
    228 			return ci;
    229 	}
    230 
    231 	return NULL;
    232 }
    233 
    234 struct cpu_info *
    235 cpu_lookup_byindex(u_int idx)
    236 {
    237 	struct cpu_info *ci = cpu_infos[idx];
    238 
    239 	KASSERT(idx < MAXCPUS);
    240 	KASSERT(ci == NULL || cpu_index(ci) == idx);
    241 
    242 	return ci;
    243 }
    244 
    245 static void
    246 cpu_xc_offline(struct cpu_info *ci)
    247 {
    248 	struct schedstate_percpu *spc, *mspc = NULL;
    249 	struct cpu_info *mci;
    250 	struct lwp *l;
    251 	CPU_INFO_ITERATOR cii;
    252 	int s;
    253 
    254 	spc = &ci->ci_schedstate;
    255 	s = splsched();
    256 	spc->spc_flags |= SPCF_OFFLINE;
    257 	splx(s);
    258 
    259 	/* Take the first available CPU for the migration */
    260 	for (CPU_INFO_FOREACH(cii, mci)) {
    261 		mspc = &mci->ci_schedstate;
    262 		if ((mspc->spc_flags & SPCF_OFFLINE) == 0)
    263 			break;
    264 	}
    265 	KASSERT(mci != NULL);
    266 
    267 	/*
    268 	 * Migrate all non-bound threads to the other CPU.
    269 	 * Please note, that this runs from the xcall thread, thus handling
    270 	 * of LSONPROC is not needed.
    271 	 */
    272 	mutex_enter(&proclist_lock);
    273 
    274 	/*
    275 	 * Note that threads on the runqueue might sleep after this, but
    276 	 * sched_takecpu() would migrate such threads to the appropriate CPU.
    277 	 */
    278 	LIST_FOREACH(l, &alllwp, l_list) {
    279 		lwp_lock(l);
    280 		if (l->l_cpu == ci && (l->l_stat == LSSLEEP ||
    281 		    l->l_stat == LSSTOP || l->l_stat == LSSUSPENDED)) {
    282 			KASSERT((l->l_flag & LW_RUNNING) == 0);
    283 			l->l_cpu = mci;
    284 		}
    285 		lwp_unlock(l);
    286 	}
    287 
    288 	/* Double-lock the run-queues */
    289 	spc_dlock(ci, mci);
    290 
    291 	/* Handle LSRUN and LSIDL cases */
    292 	LIST_FOREACH(l, &alllwp, l_list) {
    293 		if (l->l_cpu != ci || (l->l_flag & LW_BOUND))
    294 			continue;
    295 		if (l->l_stat == LSRUN && (l->l_flag & LW_INMEM) != 0) {
    296 			sched_dequeue(l);
    297 			l->l_cpu = mci;
    298 			lwp_setlock(l, mspc->spc_mutex);
    299 			sched_enqueue(l, false);
    300 		} else if (l->l_stat == LSRUN || l->l_stat == LSIDL) {
    301 			l->l_cpu = mci;
    302 			lwp_setlock(l, mspc->spc_mutex);
    303 		}
    304 	}
    305 	spc_dunlock(ci, mci);
    306 	mutex_exit(&proclist_lock);
    307 
    308 #ifdef __HAVE_MD_CPU_OFFLINE
    309 	cpu_offline_md();
    310 #endif
    311 }
    312 
    313 static void
    314 cpu_xc_online(struct cpu_info *ci)
    315 {
    316 	struct schedstate_percpu *spc;
    317 	int s;
    318 
    319 	spc = &ci->ci_schedstate;
    320 	s = splsched();
    321 	spc->spc_flags &= ~SPCF_OFFLINE;
    322 	splx(s);
    323 }
    324 
    325 int
    326 cpu_setonline(struct cpu_info *ci, bool online)
    327 {
    328 	struct schedstate_percpu *spc;
    329 	CPU_INFO_ITERATOR cii;
    330 	struct cpu_info *ci2;
    331 	uint64_t where;
    332 	xcfunc_t func;
    333 	int nonline;
    334 
    335 	spc = &ci->ci_schedstate;
    336 
    337 	KASSERT(mutex_owned(&cpu_lock));
    338 
    339 	if (online) {
    340 		if ((spc->spc_flags & SPCF_OFFLINE) == 0)
    341 			return 0;
    342 		func = (xcfunc_t)cpu_xc_online;
    343 		ncpuonline++;
    344 	} else {
    345 		if ((spc->spc_flags & SPCF_OFFLINE) != 0)
    346 			return 0;
    347 		nonline = 0;
    348 		for (CPU_INFO_FOREACH(cii, ci2)) {
    349 			nonline += ((ci2->ci_schedstate.spc_flags &
    350 			    SPCF_OFFLINE) == 0);
    351 		}
    352 		if (nonline == 1)
    353 			return EBUSY;
    354 		func = (xcfunc_t)cpu_xc_offline;
    355 		ncpuonline--;
    356 	}
    357 
    358 	where = xc_unicast(0, func, ci, NULL, ci);
    359 	xc_wait(where);
    360 	if (online) {
    361 		KASSERT((spc->spc_flags & SPCF_OFFLINE) == 0);
    362 	} else {
    363 		KASSERT(spc->spc_flags & SPCF_OFFLINE);
    364 	}
    365 	spc->spc_lastmod = time_second;
    366 
    367 	return 0;
    368 }
    369