xref: /freebsd/sys/powerpc/powerpc/mp_machdep.c (revision c697fb7f)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2008 Marcel Moolenaar
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  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/kernel.h>
35 #include <sys/ktr.h>
36 #include <sys/bus.h>
37 #include <sys/cpuset.h>
38 #include <sys/domainset.h>
39 #include <sys/lock.h>
40 #include <sys/malloc.h>
41 #include <sys/mutex.h>
42 #include <sys/pcpu.h>
43 #include <sys/proc.h>
44 #include <sys/sched.h>
45 #include <sys/smp.h>
46 
47 #include <vm/vm.h>
48 #include <vm/vm_param.h>
49 #include <vm/pmap.h>
50 #include <vm/vm_map.h>
51 #include <vm/vm_extern.h>
52 #include <vm/vm_kern.h>
53 
54 #include <machine/bus.h>
55 #include <machine/cpu.h>
56 #include <machine/intr_machdep.h>
57 #include <machine/pcb.h>
58 #include <machine/platform.h>
59 #include <machine/md_var.h>
60 #include <machine/setjmp.h>
61 #include <machine/smp.h>
62 
63 #include "pic_if.h"
64 
65 volatile static int ap_awake;
66 volatile static u_int ap_letgo;
67 volatile static u_quad_t ap_timebase;
68 static u_int ipi_msg_cnt[32];
69 static struct mtx ap_boot_mtx;
70 struct pcb stoppcbs[MAXCPU];
71 
72 void
73 machdep_ap_bootstrap(void)
74 {
75 
76 	PCPU_SET(awake, 1);
77 	__asm __volatile("msync; isync");
78 
79 	while (ap_letgo == 0)
80 		nop_prio_vlow();
81 	nop_prio_medium();
82 
83 	/*
84 	 * Set timebase as soon as possible to meet an implicit rendezvous
85 	 * from cpu_mp_unleash(), which sets ap_letgo and then immediately
86 	 * sets timebase.
87 	 *
88 	 * Note that this is instrinsically racy and is only relevant on
89 	 * platforms that do not support better mechanisms.
90 	 */
91 	platform_smp_timebase_sync(ap_timebase, 1);
92 
93 	/* Give platform code a chance to do anything else necessary */
94 	platform_smp_ap_init();
95 
96 	/* Initialize decrementer */
97 	decr_ap_init();
98 
99 	/* Serialize console output and AP count increment */
100 	mtx_lock_spin(&ap_boot_mtx);
101 	ap_awake++;
102 	if (bootverbose)
103 		printf("SMP: AP CPU #%d launched\n", PCPU_GET(cpuid));
104 	else
105 		printf("%s%d%s", ap_awake == 2 ? "Launching APs: " : "",
106 		    PCPU_GET(cpuid), ap_awake == mp_ncpus ? "\n" : " ");
107 	mtx_unlock_spin(&ap_boot_mtx);
108 
109 	while(smp_started == 0)
110 		;
111 
112 	/* Start per-CPU event timers. */
113 	cpu_initclocks_ap();
114 
115 	/* Announce ourselves awake, and enter the scheduler */
116 	sched_throw(NULL);
117 }
118 
119 void
120 cpu_mp_setmaxid(void)
121 {
122 	struct cpuref cpuref;
123 	int error;
124 
125 	mp_ncpus = 0;
126 	mp_maxid = 0;
127 	error = platform_smp_first_cpu(&cpuref);
128 	while (!error) {
129 		mp_ncpus++;
130 		mp_maxid = max(cpuref.cr_cpuid, mp_maxid);
131 		error = platform_smp_next_cpu(&cpuref);
132 	}
133 	/* Sanity. */
134 	if (mp_ncpus == 0)
135 		mp_ncpus = 1;
136 }
137 
138 int
139 cpu_mp_probe(void)
140 {
141 
142 	/*
143 	 * We're not going to enable SMP if there's only 1 processor.
144 	 */
145 	return (mp_ncpus > 1);
146 }
147 
148 void
149 cpu_mp_start(void)
150 {
151 	struct cpuref bsp, cpu;
152 	struct pcpu *pc;
153 	int domain, error;
154 
155 	error = platform_smp_get_bsp(&bsp);
156 	KASSERT(error == 0, ("Don't know BSP"));
157 
158 	error = platform_smp_first_cpu(&cpu);
159 	while (!error) {
160 		if (cpu.cr_cpuid >= MAXCPU) {
161 			printf("SMP: cpu%d: skipped -- ID out of range\n",
162 			    cpu.cr_cpuid);
163 			goto next;
164 		}
165 		if (CPU_ISSET(cpu.cr_cpuid, &all_cpus)) {
166 			printf("SMP: cpu%d: skipped - duplicate ID\n",
167 			    cpu.cr_cpuid);
168 			goto next;
169 		}
170 
171 		if (vm_ndomains > 1)
172 			domain = cpu.cr_domain;
173 		else
174 			domain = 0;
175 
176 		if (cpu.cr_cpuid != bsp.cr_cpuid) {
177 			void *dpcpu;
178 
179 			pc = &__pcpu[cpu.cr_cpuid];
180 			dpcpu = (void *)kmem_malloc_domainset(DOMAINSET_PREF(domain),
181 			    DPCPU_SIZE, M_WAITOK | M_ZERO);
182 			pcpu_init(pc, cpu.cr_cpuid, sizeof(*pc));
183 			dpcpu_init(dpcpu, cpu.cr_cpuid);
184 		} else {
185 			pc = pcpup;
186 			pc->pc_cpuid = bsp.cr_cpuid;
187 			pc->pc_bsp = 1;
188 		}
189 		pc->pc_domain = domain;
190 		pc->pc_hwref = cpu.cr_hwref;
191 
192 		CPU_SET(pc->pc_cpuid, &cpuset_domain[pc->pc_domain]);
193 		KASSERT(pc->pc_domain < MAXMEMDOM, ("bad domain value %d\n",
194 		    pc->pc_domain));
195 		CPU_SET(pc->pc_cpuid, &all_cpus);
196 next:
197 		error = platform_smp_next_cpu(&cpu);
198 	}
199 
200 #ifdef SMP
201 	platform_smp_probe_threads();
202 #endif
203 }
204 
205 void
206 cpu_mp_announce(void)
207 {
208 	struct pcpu *pc;
209 	int i;
210 
211 	if (!bootverbose)
212 		return;
213 
214 	CPU_FOREACH(i) {
215 		pc = pcpu_find(i);
216 		if (pc == NULL)
217 			continue;
218 		printf("cpu%d: dev=%x domain=%d ", i, (int)pc->pc_hwref, pc->pc_domain);
219 		if (pc->pc_bsp)
220 			printf(" (BSP)");
221 		printf("\n");
222 	}
223 }
224 
225 static void
226 cpu_mp_unleash(void *dummy)
227 {
228 	struct pcpu *pc;
229 	int cpus, timeout;
230 	int ret;
231 
232 	if (mp_ncpus <= 1)
233 		return;
234 
235 	mtx_init(&ap_boot_mtx, "ap boot", NULL, MTX_SPIN);
236 
237 	cpus = 0;
238 	smp_cpus = 0;
239 #ifdef BOOKE
240 	tlb1_ap_prep();
241 #endif
242 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
243 		cpus++;
244 		if (!pc->pc_bsp) {
245 			if (bootverbose)
246 				printf("Waking up CPU %d (dev=%x)\n",
247 				    pc->pc_cpuid, (int)pc->pc_hwref);
248 
249 			ret = platform_smp_start_cpu(pc);
250 			if (ret == 0) {
251 				timeout = 2000;	/* wait 2sec for the AP */
252 				while (!pc->pc_awake && --timeout > 0)
253 					DELAY(1000);
254 			}
255 		} else {
256 			pc->pc_awake = 1;
257 		}
258 		if (pc->pc_awake) {
259 			if (bootverbose)
260 				printf("Adding CPU %d, hwref=%jx, awake=%x\n",
261 				    pc->pc_cpuid, (uintmax_t)pc->pc_hwref,
262 				    pc->pc_awake);
263 			smp_cpus++;
264 		} else
265 			CPU_SET(pc->pc_cpuid, &stopped_cpus);
266 	}
267 
268 	ap_awake = 1;
269 
270 	/* Provide our current DEC and TB values for APs */
271 	ap_timebase = mftb() + 10;
272 	__asm __volatile("msync; isync");
273 
274 	/* Let APs continue */
275 	atomic_store_rel_int(&ap_letgo, 1);
276 
277 	platform_smp_timebase_sync(ap_timebase, 0);
278 
279 	while (ap_awake < smp_cpus)
280 		;
281 
282 	if (smp_cpus != cpus || cpus != mp_ncpus) {
283 		printf("SMP: %d CPUs found; %d CPUs usable; %d CPUs woken\n",
284 		    mp_ncpus, cpus, smp_cpus);
285 	}
286 
287 	if (smp_cpus > 1)
288 		atomic_store_rel_int(&smp_started, 1);
289 
290 	/* Let the APs get into the scheduler */
291 	DELAY(10000);
292 
293 }
294 
295 SYSINIT(start_aps, SI_SUB_SMP, SI_ORDER_FIRST, cpu_mp_unleash, NULL);
296 
297 int
298 powerpc_ipi_handler(void *arg)
299 {
300 	u_int cpuid;
301 	uint32_t ipimask;
302 	int msg;
303 
304 	CTR2(KTR_SMP, "%s: MSR 0x%08x", __func__, mfmsr());
305 
306 	ipimask = atomic_readandclear_32(&(pcpup->pc_ipimask));
307 	if (ipimask == 0)
308 		return (FILTER_STRAY);
309 	while ((msg = ffs(ipimask) - 1) != -1) {
310 		ipimask &= ~(1u << msg);
311 		ipi_msg_cnt[msg]++;
312 		switch (msg) {
313 		case IPI_AST:
314 			CTR1(KTR_SMP, "%s: IPI_AST", __func__);
315 			break;
316 		case IPI_PREEMPT:
317 			CTR1(KTR_SMP, "%s: IPI_PREEMPT", __func__);
318 			sched_preempt(curthread);
319 			break;
320 		case IPI_RENDEZVOUS:
321 			CTR1(KTR_SMP, "%s: IPI_RENDEZVOUS", __func__);
322 			smp_rendezvous_action();
323 			break;
324 		case IPI_STOP:
325 
326 			/*
327 			 * IPI_STOP_HARD is mapped to IPI_STOP so it is not
328 			 * necessary to add such case in the switch.
329 			 */
330 			CTR1(KTR_SMP, "%s: IPI_STOP or IPI_STOP_HARD (stop)",
331 			    __func__);
332 			cpuid = PCPU_GET(cpuid);
333 			savectx(&stoppcbs[cpuid]);
334 			CPU_SET_ATOMIC(cpuid, &stopped_cpus);
335 			while (!CPU_ISSET(cpuid, &started_cpus))
336 				cpu_spinwait();
337 			CPU_CLR_ATOMIC(cpuid, &stopped_cpus);
338 			CPU_CLR_ATOMIC(cpuid, &started_cpus);
339 			CTR1(KTR_SMP, "%s: IPI_STOP (restart)", __func__);
340 			break;
341 		case IPI_HARDCLOCK:
342 			CTR1(KTR_SMP, "%s: IPI_HARDCLOCK", __func__);
343 			hardclockintr();
344 			break;
345 		}
346 	}
347 
348 	return (FILTER_HANDLED);
349 }
350 
351 static void
352 ipi_send(struct pcpu *pc, int ipi)
353 {
354 
355 	CTR4(KTR_SMP, "%s: pc=%p, targetcpu=%d, IPI=%d", __func__,
356 	    pc, pc->pc_cpuid, ipi);
357 
358 	atomic_set_32(&pc->pc_ipimask, (1 << ipi));
359 	powerpc_sync();
360 	PIC_IPI(root_pic, pc->pc_cpuid);
361 
362 	CTR1(KTR_SMP, "%s: sent", __func__);
363 }
364 
365 /* Send an IPI to a set of cpus. */
366 void
367 ipi_selected(cpuset_t cpus, int ipi)
368 {
369 	struct pcpu *pc;
370 
371 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
372 		if (CPU_ISSET(pc->pc_cpuid, &cpus))
373 			ipi_send(pc, ipi);
374 	}
375 }
376 
377 /* Send an IPI to a specific CPU. */
378 void
379 ipi_cpu(int cpu, u_int ipi)
380 {
381 
382 	ipi_send(cpuid_to_pcpu[cpu], ipi);
383 }
384 
385 /* Send an IPI to all CPUs EXCEPT myself. */
386 void
387 ipi_all_but_self(int ipi)
388 {
389 	struct pcpu *pc;
390 
391 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
392 		if (pc != pcpup)
393 			ipi_send(pc, ipi);
394 	}
395 }
396