xref: /freebsd/sys/powerpc/powerpc/mp_machdep.c (revision bdd1243df58e60e85101c09001d9812a789b6bc4)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 
70 void
71 machdep_ap_bootstrap(void)
72 {
73 
74 	PCPU_SET(awake, 1);
75 	__asm __volatile("msync; isync");
76 
77 	while (ap_letgo == 0)
78 		nop_prio_vlow();
79 	nop_prio_medium();
80 
81 	/*
82 	 * Set timebase as soon as possible to meet an implicit rendezvous
83 	 * from cpu_mp_unleash(), which sets ap_letgo and then immediately
84 	 * sets timebase.
85 	 *
86 	 * Note that this is instrinsically racy and is only relevant on
87 	 * platforms that do not support better mechanisms.
88 	 */
89 	platform_smp_timebase_sync(ap_timebase, 1);
90 
91 	/* Give platform code a chance to do anything else necessary */
92 	platform_smp_ap_init();
93 
94 	/* Initialize decrementer */
95 	decr_ap_init();
96 
97 	/* Serialize console output and AP count increment */
98 	mtx_lock_spin(&ap_boot_mtx);
99 	ap_awake++;
100 	if (bootverbose)
101 		printf("SMP: AP CPU #%d launched\n", PCPU_GET(cpuid));
102 	else
103 		printf("%s%d%s", ap_awake == 2 ? "Launching APs: " : "",
104 		    PCPU_GET(cpuid), ap_awake == mp_ncpus ? "\n" : " ");
105 	mtx_unlock_spin(&ap_boot_mtx);
106 
107 	while(smp_started == 0)
108 		;
109 
110 	/* Start per-CPU event timers. */
111 	cpu_initclocks_ap();
112 
113 	/* Announce ourselves awake, and enter the scheduler */
114 	sched_ap_entry();
115 }
116 
117 void
118 cpu_mp_setmaxid(void)
119 {
120 	struct cpuref cpuref;
121 	int error;
122 
123 	mp_ncpus = 0;
124 	mp_maxid = 0;
125 	error = platform_smp_first_cpu(&cpuref);
126 	while (!error) {
127 		mp_ncpus++;
128 		mp_maxid = max(cpuref.cr_cpuid, mp_maxid);
129 		error = platform_smp_next_cpu(&cpuref);
130 	}
131 	/* Sanity. */
132 	if (mp_ncpus == 0)
133 		mp_ncpus = 1;
134 }
135 
136 int
137 cpu_mp_probe(void)
138 {
139 
140 	/*
141 	 * We're not going to enable SMP if there's only 1 processor.
142 	 */
143 	return (mp_ncpus > 1);
144 }
145 
146 void
147 cpu_mp_start(void)
148 {
149 	struct cpuref bsp, cpu;
150 	struct pcpu *pc;
151 	int domain, error;
152 
153 	error = platform_smp_get_bsp(&bsp);
154 	KASSERT(error == 0, ("Don't know BSP"));
155 
156 	error = platform_smp_first_cpu(&cpu);
157 	while (!error) {
158 		if (cpu.cr_cpuid >= MAXCPU) {
159 			printf("SMP: cpu%d: skipped -- ID out of range\n",
160 			    cpu.cr_cpuid);
161 			goto next;
162 		}
163 		if (CPU_ISSET(cpu.cr_cpuid, &all_cpus)) {
164 			printf("SMP: cpu%d: skipped - duplicate ID\n",
165 			    cpu.cr_cpuid);
166 			goto next;
167 		}
168 
169 		if (vm_ndomains > 1)
170 			domain = cpu.cr_domain;
171 		else
172 			domain = 0;
173 
174 		if (cpu.cr_cpuid != bsp.cr_cpuid) {
175 			void *dpcpu;
176 
177 			pc = &__pcpu[cpu.cr_cpuid];
178 			dpcpu = kmem_malloc_domainset(DOMAINSET_PREF(domain),
179 			    DPCPU_SIZE, M_WAITOK | M_ZERO);
180 			pcpu_init(pc, cpu.cr_cpuid, sizeof(*pc));
181 			dpcpu_init(dpcpu, cpu.cr_cpuid);
182 		} else {
183 			pc = pcpup;
184 			pc->pc_cpuid = bsp.cr_cpuid;
185 			pc->pc_bsp = 1;
186 		}
187 		pc->pc_domain = domain;
188 		pc->pc_hwref = cpu.cr_hwref;
189 
190 		CPU_SET(pc->pc_cpuid, &cpuset_domain[pc->pc_domain]);
191 		KASSERT(pc->pc_domain < MAXMEMDOM, ("bad domain value %d\n",
192 		    pc->pc_domain));
193 		CPU_SET(pc->pc_cpuid, &all_cpus);
194 next:
195 		error = platform_smp_next_cpu(&cpu);
196 	}
197 
198 #ifdef SMP
199 	platform_smp_probe_threads();
200 #endif
201 }
202 
203 void
204 cpu_mp_announce(void)
205 {
206 	struct pcpu *pc;
207 	int i;
208 
209 	if (!bootverbose)
210 		return;
211 
212 	CPU_FOREACH(i) {
213 		pc = pcpu_find(i);
214 		if (pc == NULL)
215 			continue;
216 		printf("cpu%d: dev=%x domain=%d ", i, (int)pc->pc_hwref, pc->pc_domain);
217 		if (pc->pc_bsp)
218 			printf(" (BSP)");
219 		printf("\n");
220 	}
221 }
222 
223 static void
224 cpu_mp_unleash(void *dummy)
225 {
226 	struct pcpu *pc;
227 	int cpus, timeout;
228 	int ret;
229 
230 	if (mp_ncpus <= 1)
231 		return;
232 
233 	mtx_init(&ap_boot_mtx, "ap boot", NULL, MTX_SPIN);
234 
235 	cpus = 0;
236 	smp_cpus = 0;
237 #ifdef BOOKE
238 	tlb1_ap_prep();
239 #endif
240 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
241 		cpus++;
242 		if (!pc->pc_bsp) {
243 			if (bootverbose)
244 				printf("Waking up CPU %d (dev=%x)\n",
245 				    pc->pc_cpuid, (int)pc->pc_hwref);
246 
247 			pc->pc_flags = PCPU_GET(flags); /* Copy cached CPU flags */
248 			ret = platform_smp_start_cpu(pc);
249 			if (ret == 0) {
250 				timeout = 2000;	/* wait 2sec for the AP */
251 				while (!pc->pc_awake && --timeout > 0)
252 					DELAY(1000);
253 			}
254 		} else {
255 			pc->pc_awake = 1;
256 		}
257 		if (pc->pc_awake) {
258 			if (bootverbose)
259 				printf("Adding CPU %d, hwref=%jx, awake=%x\n",
260 				    pc->pc_cpuid, (uintmax_t)pc->pc_hwref,
261 				    pc->pc_awake);
262 			smp_cpus++;
263 		} else
264 			CPU_SET(pc->pc_cpuid, &stopped_cpus);
265 	}
266 
267 	ap_awake = 1;
268 
269 	/* Provide our current DEC and TB values for APs */
270 	ap_timebase = mftb() + 10;
271 	__asm __volatile("msync; isync");
272 
273 	/* Let APs continue */
274 	atomic_store_rel_int(&ap_letgo, 1);
275 
276 	platform_smp_timebase_sync(ap_timebase, 0);
277 
278 	while (ap_awake < smp_cpus)
279 		;
280 
281 	if (smp_cpus != cpus || cpus != mp_ncpus) {
282 		printf("SMP: %d CPUs found; %d CPUs usable; %d CPUs woken\n",
283 		    mp_ncpus, cpus, smp_cpus);
284 	}
285 
286 	if (smp_cpus > 1)
287 		atomic_store_rel_int(&smp_started, 1);
288 
289 	/* Let the APs get into the scheduler */
290 	DELAY(10000);
291 
292 }
293 
294 SYSINIT(start_aps, SI_SUB_SMP, SI_ORDER_FIRST, cpu_mp_unleash, NULL);
295 
296 int
297 powerpc_ipi_handler(void *arg)
298 {
299 	u_int cpuid;
300 	uint32_t ipimask;
301 
302 	CTR2(KTR_SMP, "%s: MSR 0x%08x", __func__, mfmsr());
303 
304 	ipimask = atomic_readandclear_32(&(pcpup->pc_ipimask));
305 	if (ipimask == 0)
306 		return (FILTER_STRAY);
307 	if (ipimask & (1 << IPI_AST)) {
308 		CTR1(KTR_SMP, "%s: IPI_AST", __func__);
309 	}
310 	if (ipimask & (1 << IPI_PREEMPT)) {
311 		CTR1(KTR_SMP, "%s: IPI_PREEMPT", __func__);
312 		sched_preempt(curthread);
313 	}
314 	if (ipimask & (1 << IPI_RENDEZVOUS)) {
315 		CTR1(KTR_SMP, "%s: IPI_RENDEZVOUS", __func__);
316 		smp_rendezvous_action();
317 	}
318 	if (ipimask & (1 << IPI_STOP)) {
319 
320 		/*
321 		 * IPI_STOP_HARD is mapped to IPI_STOP so it is not
322 		 * necessary to add such case.
323 		 */
324 		CTR1(KTR_SMP, "%s: IPI_STOP or IPI_STOP_HARD (stop)",
325 				__func__);
326 		cpuid = PCPU_GET(cpuid);
327 		savectx(&stoppcbs[cpuid]);
328 		CPU_SET_ATOMIC(cpuid, &stopped_cpus);
329 		while (!CPU_ISSET(cpuid, &started_cpus))
330 			cpu_spinwait();
331 		CPU_CLR_ATOMIC(cpuid, &stopped_cpus);
332 		CPU_CLR_ATOMIC(cpuid, &started_cpus);
333 		CTR1(KTR_SMP, "%s: IPI_STOP (restart)", __func__);
334 	}
335 	if (ipimask & (1 << IPI_HARDCLOCK)) {
336 		CTR1(KTR_SMP, "%s: IPI_HARDCLOCK", __func__);
337 		hardclockintr();
338 	}
339 
340 	return (FILTER_HANDLED);
341 }
342 
343 static void
344 ipi_send(struct pcpu *pc, int ipi)
345 {
346 
347 	CTR4(KTR_SMP, "%s: pc=%p, targetcpu=%d, IPI=%d", __func__,
348 	    pc, pc->pc_cpuid, ipi);
349 
350 	atomic_set_32(&pc->pc_ipimask, (1 << ipi));
351 	powerpc_sync();
352 	PIC_IPI(root_pic, pc->pc_cpuid);
353 
354 	CTR1(KTR_SMP, "%s: sent", __func__);
355 }
356 
357 /* Send an IPI to a set of cpus. */
358 void
359 ipi_selected(cpuset_t cpus, int ipi)
360 {
361 	struct pcpu *pc;
362 
363 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
364 		if (CPU_ISSET(pc->pc_cpuid, &cpus))
365 			ipi_send(pc, ipi);
366 	}
367 }
368 
369 /* Send an IPI to a specific CPU. */
370 void
371 ipi_cpu(int cpu, u_int ipi)
372 {
373 
374 	ipi_send(cpuid_to_pcpu[cpu], ipi);
375 }
376 
377 /* Send an IPI to all CPUs EXCEPT myself. */
378 void
379 ipi_all_but_self(int ipi)
380 {
381 	struct pcpu *pc;
382 
383 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
384 		if (pc != pcpup)
385 			ipi_send(pc, ipi);
386 	}
387 }
388