xref: /freebsd/sys/powerpc/powerpc/mp_machdep.c (revision 81b22a98)
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 struct mtx ap_boot_mtx;
69 struct pcb stoppcbs[MAXCPU];
70 
71 void
72 machdep_ap_bootstrap(void)
73 {
74 
75 	PCPU_SET(awake, 1);
76 	__asm __volatile("msync; isync");
77 
78 	while (ap_letgo == 0)
79 		nop_prio_vlow();
80 	nop_prio_medium();
81 
82 	/*
83 	 * Set timebase as soon as possible to meet an implicit rendezvous
84 	 * from cpu_mp_unleash(), which sets ap_letgo and then immediately
85 	 * sets timebase.
86 	 *
87 	 * Note that this is instrinsically racy and is only relevant on
88 	 * platforms that do not support better mechanisms.
89 	 */
90 	platform_smp_timebase_sync(ap_timebase, 1);
91 
92 	/* Give platform code a chance to do anything else necessary */
93 	platform_smp_ap_init();
94 
95 	/* Initialize decrementer */
96 	decr_ap_init();
97 
98 	/* Serialize console output and AP count increment */
99 	mtx_lock_spin(&ap_boot_mtx);
100 	ap_awake++;
101 	if (bootverbose)
102 		printf("SMP: AP CPU #%d launched\n", PCPU_GET(cpuid));
103 	else
104 		printf("%s%d%s", ap_awake == 2 ? "Launching APs: " : "",
105 		    PCPU_GET(cpuid), ap_awake == mp_ncpus ? "\n" : " ");
106 	mtx_unlock_spin(&ap_boot_mtx);
107 
108 	while(smp_started == 0)
109 		;
110 
111 	/* Start per-CPU event timers. */
112 	cpu_initclocks_ap();
113 
114 	/* Announce ourselves awake, and enter the scheduler */
115 	sched_ap_entry();
116 }
117 
118 void
119 cpu_mp_setmaxid(void)
120 {
121 	struct cpuref cpuref;
122 	int error;
123 
124 	mp_ncpus = 0;
125 	mp_maxid = 0;
126 	error = platform_smp_first_cpu(&cpuref);
127 	while (!error) {
128 		mp_ncpus++;
129 		mp_maxid = max(cpuref.cr_cpuid, mp_maxid);
130 		error = platform_smp_next_cpu(&cpuref);
131 	}
132 	/* Sanity. */
133 	if (mp_ncpus == 0)
134 		mp_ncpus = 1;
135 }
136 
137 int
138 cpu_mp_probe(void)
139 {
140 
141 	/*
142 	 * We're not going to enable SMP if there's only 1 processor.
143 	 */
144 	return (mp_ncpus > 1);
145 }
146 
147 void
148 cpu_mp_start(void)
149 {
150 	struct cpuref bsp, cpu;
151 	struct pcpu *pc;
152 	int domain, error;
153 
154 	error = platform_smp_get_bsp(&bsp);
155 	KASSERT(error == 0, ("Don't know BSP"));
156 
157 	error = platform_smp_first_cpu(&cpu);
158 	while (!error) {
159 		if (cpu.cr_cpuid >= MAXCPU) {
160 			printf("SMP: cpu%d: skipped -- ID out of range\n",
161 			    cpu.cr_cpuid);
162 			goto next;
163 		}
164 		if (CPU_ISSET(cpu.cr_cpuid, &all_cpus)) {
165 			printf("SMP: cpu%d: skipped - duplicate ID\n",
166 			    cpu.cr_cpuid);
167 			goto next;
168 		}
169 
170 		if (vm_ndomains > 1)
171 			domain = cpu.cr_domain;
172 		else
173 			domain = 0;
174 
175 		if (cpu.cr_cpuid != bsp.cr_cpuid) {
176 			void *dpcpu;
177 
178 			pc = &__pcpu[cpu.cr_cpuid];
179 			dpcpu = (void *)kmem_malloc_domainset(DOMAINSET_PREF(domain),
180 			    DPCPU_SIZE, M_WAITOK | M_ZERO);
181 			pcpu_init(pc, cpu.cr_cpuid, sizeof(*pc));
182 			dpcpu_init(dpcpu, cpu.cr_cpuid);
183 		} else {
184 			pc = pcpup;
185 			pc->pc_cpuid = bsp.cr_cpuid;
186 			pc->pc_bsp = 1;
187 		}
188 		pc->pc_domain = domain;
189 		pc->pc_hwref = cpu.cr_hwref;
190 
191 		CPU_SET(pc->pc_cpuid, &cpuset_domain[pc->pc_domain]);
192 		KASSERT(pc->pc_domain < MAXMEMDOM, ("bad domain value %d\n",
193 		    pc->pc_domain));
194 		CPU_SET(pc->pc_cpuid, &all_cpus);
195 next:
196 		error = platform_smp_next_cpu(&cpu);
197 	}
198 
199 #ifdef SMP
200 	platform_smp_probe_threads();
201 #endif
202 }
203 
204 void
205 cpu_mp_announce(void)
206 {
207 	struct pcpu *pc;
208 	int i;
209 
210 	if (!bootverbose)
211 		return;
212 
213 	CPU_FOREACH(i) {
214 		pc = pcpu_find(i);
215 		if (pc == NULL)
216 			continue;
217 		printf("cpu%d: dev=%x domain=%d ", i, (int)pc->pc_hwref, pc->pc_domain);
218 		if (pc->pc_bsp)
219 			printf(" (BSP)");
220 		printf("\n");
221 	}
222 }
223 
224 static void
225 cpu_mp_unleash(void *dummy)
226 {
227 	struct pcpu *pc;
228 	int cpus, timeout;
229 	int ret;
230 
231 	if (mp_ncpus <= 1)
232 		return;
233 
234 	mtx_init(&ap_boot_mtx, "ap boot", NULL, MTX_SPIN);
235 
236 	cpus = 0;
237 	smp_cpus = 0;
238 #ifdef BOOKE
239 	tlb1_ap_prep();
240 #endif
241 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
242 		cpus++;
243 		if (!pc->pc_bsp) {
244 			if (bootverbose)
245 				printf("Waking up CPU %d (dev=%x)\n",
246 				    pc->pc_cpuid, (int)pc->pc_hwref);
247 
248 			pc->pc_flags = PCPU_GET(flags); /* Copy cached CPU flags */
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 
303 	CTR2(KTR_SMP, "%s: MSR 0x%08x", __func__, mfmsr());
304 
305 	ipimask = atomic_readandclear_32(&(pcpup->pc_ipimask));
306 	if (ipimask == 0)
307 		return (FILTER_STRAY);
308 	if (ipimask & (1 << IPI_AST)) {
309 		CTR1(KTR_SMP, "%s: IPI_AST", __func__);
310 	}
311 	if (ipimask & (1 << IPI_PREEMPT)) {
312 		CTR1(KTR_SMP, "%s: IPI_PREEMPT", __func__);
313 		sched_preempt(curthread);
314 	}
315 	if (ipimask & (1 << IPI_RENDEZVOUS)) {
316 		CTR1(KTR_SMP, "%s: IPI_RENDEZVOUS", __func__);
317 		smp_rendezvous_action();
318 	}
319 	if (ipimask & (1 << IPI_STOP)) {
320 
321 		/*
322 		 * IPI_STOP_HARD is mapped to IPI_STOP so it is not
323 		 * necessary to add such case.
324 		 */
325 		CTR1(KTR_SMP, "%s: IPI_STOP or IPI_STOP_HARD (stop)",
326 				__func__);
327 		cpuid = PCPU_GET(cpuid);
328 		savectx(&stoppcbs[cpuid]);
329 		CPU_SET_ATOMIC(cpuid, &stopped_cpus);
330 		while (!CPU_ISSET(cpuid, &started_cpus))
331 			cpu_spinwait();
332 		CPU_CLR_ATOMIC(cpuid, &stopped_cpus);
333 		CPU_CLR_ATOMIC(cpuid, &started_cpus);
334 		CTR1(KTR_SMP, "%s: IPI_STOP (restart)", __func__);
335 	}
336 	if (ipimask & (1 << IPI_HARDCLOCK)) {
337 		CTR1(KTR_SMP, "%s: IPI_HARDCLOCK", __func__);
338 		hardclockintr();
339 	}
340 
341 	return (FILTER_HANDLED);
342 }
343 
344 static void
345 ipi_send(struct pcpu *pc, int ipi)
346 {
347 
348 	CTR4(KTR_SMP, "%s: pc=%p, targetcpu=%d, IPI=%d", __func__,
349 	    pc, pc->pc_cpuid, ipi);
350 
351 	atomic_set_32(&pc->pc_ipimask, (1 << ipi));
352 	powerpc_sync();
353 	PIC_IPI(root_pic, pc->pc_cpuid);
354 
355 	CTR1(KTR_SMP, "%s: sent", __func__);
356 }
357 
358 /* Send an IPI to a set of cpus. */
359 void
360 ipi_selected(cpuset_t cpus, int ipi)
361 {
362 	struct pcpu *pc;
363 
364 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
365 		if (CPU_ISSET(pc->pc_cpuid, &cpus))
366 			ipi_send(pc, ipi);
367 	}
368 }
369 
370 /* Send an IPI to a specific CPU. */
371 void
372 ipi_cpu(int cpu, u_int ipi)
373 {
374 
375 	ipi_send(cpuid_to_pcpu[cpu], ipi);
376 }
377 
378 /* Send an IPI to all CPUs EXCEPT myself. */
379 void
380 ipi_all_but_self(int ipi)
381 {
382 	struct pcpu *pc;
383 
384 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
385 		if (pc != pcpup)
386 			ipi_send(pc, ipi);
387 	}
388 }
389