3 * John Baldwin <jhb@FreeBSD.org>. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the author nor the names of any co-contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY JOHN BALDWIN AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL JOHN BALDWIN OR THE VOICES IN HIS HEAD
21 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
22 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
23 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
24 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
25 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
26 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
27 * THE POSSIBILITY OF SUCH DAMAGE.
31 * This module holds the global variables and machine independent functions
32 * used for the kernel SMP support.
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/kernel.h>
45 #include <sys/mutex.h>
48 #include <sys/sysctl.h>
50 #include <machine/cpu.h>
51 #include <machine/smp.h>
53 #include "opt_sched.h"
56 volatile cpumask_t stopped_cpus;
57 volatile cpumask_t started_cpus;
58 cpumask_t idle_cpus_mask;
59 cpumask_t hlt_cpus_mask;
60 cpumask_t logical_cpus_mask;
62 void (*cpustop_restartfunc)(void);
64 /* This is used in modules that need to work in both SMP and UP. */
68 /* export this for libkvm consumers. */
69 int mp_maxcpus = MAXCPU;
71 struct cpu_top *smp_topology;
72 volatile int smp_started;
75 SYSCTL_NODE(_kern, OID_AUTO, smp, CTLFLAG_RD, NULL, "Kernel SMP");
77 SYSCTL_INT(_kern_smp, OID_AUTO, maxcpus, CTLFLAG_RD, &mp_maxcpus, 0,
78 "Max number of CPUs that the system was compiled for.");
80 int smp_active = 0; /* are the APs allowed to run? */
81 SYSCTL_INT(_kern_smp, OID_AUTO, active, CTLFLAG_RW, &smp_active, 0,
82 "Number of Auxillary Processors (APs) that were successfully started");
84 int smp_disabled = 0; /* has smp been disabled? */
85 SYSCTL_INT(_kern_smp, OID_AUTO, disabled, CTLFLAG_RDTUN, &smp_disabled, 0,
86 "SMP has been disabled from the loader");
87 TUNABLE_INT("kern.smp.disabled", &smp_disabled);
89 int smp_cpus = 1; /* how many cpu's running */
90 SYSCTL_INT(_kern_smp, OID_AUTO, cpus, CTLFLAG_RD, &smp_cpus, 0,
91 "Number of CPUs online");
94 /* Enable forwarding of a signal to a process running on a different CPU */
95 static int forward_signal_enabled = 1;
96 SYSCTL_INT(_kern_smp, OID_AUTO, forward_signal_enabled, CTLFLAG_RW,
97 &forward_signal_enabled, 0,
98 "Forwarding of a signal to a process on a different CPU");
100 /* Enable forwarding of roundrobin to all other cpus */
101 static int forward_roundrobin_enabled = 1;
102 SYSCTL_INT(_kern_smp, OID_AUTO, forward_roundrobin_enabled, CTLFLAG_RW,
103 &forward_roundrobin_enabled, 0,
104 "Forwarding of roundrobin to all other CPUs");
106 /* Variables needed for SMP rendezvous. */
107 static void (*volatile smp_rv_setup_func)(void *arg);
108 static void (*volatile smp_rv_action_func)(void *arg);
109 static void (*volatile smp_rv_teardown_func)(void *arg);
110 static void * volatile smp_rv_func_arg;
111 static volatile int smp_rv_waiters[3];
114 * Shared mutex to restrict busywaits between smp_rendezvous() and
115 * smp(_targeted)_tlb_shootdown(). A deadlock occurs if both of these
116 * functions trigger at once and cause multiple CPUs to busywait with
117 * interrupts disabled.
119 struct mtx smp_ipi_mtx;
122 * Let the MD SMP code initialize mp_maxid very early if it can.
125 mp_setmaxid(void *dummy)
129 SYSINIT(cpu_mp_setmaxid, SI_SUB_TUNABLES, SI_ORDER_FIRST, mp_setmaxid, NULL)
132 * Call the MD SMP initialization code.
135 mp_start(void *dummy)
138 /* Probe for MP hardware. */
139 if (smp_disabled != 0 || cpu_mp_probe() == 0) {
141 all_cpus = PCPU_GET(cpumask);
145 mtx_init(&smp_ipi_mtx, "smp rendezvous", NULL, MTX_SPIN);
147 printf("FreeBSD/SMP: Multiprocessor System Detected: %d CPUs\n",
151 SYSINIT(cpu_mp, SI_SUB_CPU, SI_ORDER_THIRD, mp_start, NULL)
154 forward_signal(struct thread *td)
159 * signotify() has already set TDF_ASTPENDING and TDF_NEEDSIGCHECK on
160 * this thread, so all we need to do is poke it if it is currently
161 * executing so that it executes ast().
163 THREAD_LOCK_ASSERT(td, MA_OWNED);
164 KASSERT(TD_IS_RUNNING(td),
165 ("forward_signal: thread is not TDS_RUNNING"));
167 CTR1(KTR_SMP, "forward_signal(%p)", td->td_proc);
169 if (!smp_started || cold || panicstr)
171 if (!forward_signal_enabled)
174 /* No need to IPI ourself. */
181 ipi_selected(1 << id, IPI_AST);
185 forward_roundrobin(void)
189 cpumask_t id, map, me;
191 CTR0(KTR_SMP, "forward_roundrobin()");
193 if (!smp_started || cold || panicstr)
195 if (!forward_roundrobin_enabled)
198 me = PCPU_GET(cpumask);
199 SLIST_FOREACH(pc, &cpuhead, pc_allcpu) {
200 td = pc->pc_curthread;
202 if (id != me && (id & stopped_cpus) == 0 &&
203 !TD_IS_IDLETHREAD(td)) {
204 td->td_flags |= TDF_NEEDRESCHED;
208 ipi_selected(map, IPI_AST);
212 * When called the executing CPU will send an IPI to all other CPUs
213 * requesting that they halt execution.
215 * Usually (but not necessarily) called with 'other_cpus' as its arg.
217 * - Signals all CPUs in map to stop.
218 * - Waits for each to stop.
225 * XXX FIXME: this is not MP-safe, needs a lock to prevent multiple CPUs
226 * from executing at same time.
229 stop_cpus(cpumask_t map)
236 CTR1(KTR_SMP, "stop_cpus(%x)", map);
238 /* send the stop IPI to all CPUs in map */
239 ipi_selected(map, IPI_STOP);
242 while ((stopped_cpus & map) != map) {
248 printf("timeout stopping cpus\n");
258 * Called by a CPU to restart stopped CPUs.
260 * Usually (but not necessarily) called with 'stopped_cpus' as its arg.
262 * - Signals all CPUs in map to restart.
263 * - Waits for each to restart.
271 restart_cpus(cpumask_t map)
277 CTR1(KTR_SMP, "restart_cpus(%x)", map);
279 /* signal other cpus to restart */
280 atomic_store_rel_int(&started_cpus, map);
282 /* wait for each to clear its bit */
283 while ((stopped_cpus & map) != 0)
290 smp_no_rendevous_barrier(void *dummy)
293 KASSERT((!smp_started),("smp_no_rendevous called and smp is started"));
298 * All-CPU rendezvous. CPUs are signalled, all execute the setup function
299 * (if specified), rendezvous, execute the action function (if specified),
300 * rendezvous again, execute the teardown function (if specified), and then
303 * Note that the supplied external functions _must_ be reentrant and aware
304 * that they are running in parallel and in an unknown lock context.
307 smp_rendezvous_action(void)
309 void* local_func_arg = smp_rv_func_arg;
310 void (*local_setup_func)(void*) = smp_rv_setup_func;
311 void (*local_action_func)(void*) = smp_rv_action_func;
312 void (*local_teardown_func)(void*) = smp_rv_teardown_func;
314 /* Ensure we have up-to-date values. */
315 atomic_add_acq_int(&smp_rv_waiters[0], 1);
316 while (smp_rv_waiters[0] < mp_ncpus)
320 if (local_setup_func != smp_no_rendevous_barrier) {
321 if (smp_rv_setup_func != NULL)
322 smp_rv_setup_func(smp_rv_func_arg);
324 /* spin on entry rendezvous */
325 atomic_add_int(&smp_rv_waiters[1], 1);
326 while (smp_rv_waiters[1] < mp_ncpus)
330 /* action function */
331 if (local_action_func != NULL)
332 local_action_func(local_func_arg);
334 /* spin on exit rendezvous */
335 atomic_add_int(&smp_rv_waiters[2], 1);
336 if (local_teardown_func == smp_no_rendevous_barrier)
338 while (smp_rv_waiters[2] < mp_ncpus)
341 /* teardown function */
342 if (local_teardown_func != NULL)
343 local_teardown_func(local_func_arg);
347 smp_rendezvous(void (* setup_func)(void *),
348 void (* action_func)(void *),
349 void (* teardown_func)(void *),
354 if (setup_func != NULL)
356 if (action_func != NULL)
358 if (teardown_func != NULL)
363 /* obtain rendezvous lock */
364 mtx_lock_spin(&smp_ipi_mtx);
366 /* set static function pointers */
367 smp_rv_setup_func = setup_func;
368 smp_rv_action_func = action_func;
369 smp_rv_teardown_func = teardown_func;
370 smp_rv_func_arg = arg;
371 smp_rv_waiters[1] = 0;
372 smp_rv_waiters[2] = 0;
373 atomic_store_rel_int(&smp_rv_waiters[0], 0);
375 /* signal other processors, which will enter the IPI with interrupts off */
376 ipi_all_but_self(IPI_RENDEZVOUS);
378 /* call executor function */
379 smp_rendezvous_action();
381 if (teardown_func == smp_no_rendevous_barrier)
382 while (atomic_load_acq_int(&smp_rv_waiters[2]) < mp_ncpus)
386 mtx_unlock_spin(&smp_ipi_mtx);
391 * Provide dummy SMP support for UP kernels. Modules that need to use SMP
392 * APIs will still work using this dummy support.
395 mp_setvariables_for_up(void *dummy)
398 mp_maxid = PCPU_GET(cpuid);
399 all_cpus = PCPU_GET(cpumask);
400 KASSERT(PCPU_GET(cpuid) == 0, ("UP must have a CPU ID of zero"));
402 SYSINIT(cpu_mp_setvariables, SI_SUB_TUNABLES, SI_ORDER_FIRST,
403 mp_setvariables_for_up, NULL)
406 smp_rendezvous(void (*setup_func)(void *),
407 void (*action_func)(void *),
408 void (*teardown_func)(void *),
412 if (setup_func != NULL)
414 if (action_func != NULL)
416 if (teardown_func != NULL)