xref: /freebsd/sys/powerpc/powerpc/mp_machdep.c (revision 2e620256bd76c449c835c604e404483437743011)
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 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/kernel.h>
33 #include <sys/ktr.h>
34 #include <sys/bus.h>
35 #include <sys/cpuset.h>
36 #include <sys/domainset.h>
37 #include <sys/lock.h>
38 #include <sys/malloc.h>
39 #include <sys/mutex.h>
40 #include <sys/pcpu.h>
41 #include <sys/proc.h>
42 #include <sys/sched.h>
43 #include <sys/smp.h>
44 
45 #include <vm/vm.h>
46 #include <vm/vm_param.h>
47 #include <vm/pmap.h>
48 #include <vm/vm_map.h>
49 #include <vm/vm_extern.h>
50 #include <vm/vm_kern.h>
51 
52 #include <machine/bus.h>
53 #include <machine/cpu.h>
54 #include <machine/intr_machdep.h>
55 #include <machine/pcb.h>
56 #include <machine/platform.h>
57 #include <machine/md_var.h>
58 #include <machine/setjmp.h>
59 #include <machine/smp.h>
60 
61 #include "pic_if.h"
62 
63 volatile static int ap_awake;
64 volatile static u_int ap_letgo;
65 volatile static u_quad_t ap_timebase;
66 static struct mtx ap_boot_mtx;
67 
68 void
69 machdep_ap_bootstrap(void)
70 {
71 
72 	PCPU_SET(awake, 1);
73 	__asm __volatile("msync; isync");
74 
75 	while (ap_letgo == 0)
76 		nop_prio_vlow();
77 	nop_prio_medium();
78 
79 	/*
80 	 * Set timebase as soon as possible to meet an implicit rendezvous
81 	 * from cpu_mp_unleash(), which sets ap_letgo and then immediately
82 	 * sets timebase.
83 	 *
84 	 * Note that this is instrinsically racy and is only relevant on
85 	 * platforms that do not support better mechanisms.
86 	 */
87 	platform_smp_timebase_sync(ap_timebase, 1);
88 
89 	/* Give platform code a chance to do anything else necessary */
90 	platform_smp_ap_init();
91 
92 	/* Initialize decrementer */
93 	decr_ap_init();
94 
95 	/* Serialize console output and AP count increment */
96 	mtx_lock_spin(&ap_boot_mtx);
97 	ap_awake++;
98 	if (bootverbose)
99 		printf("SMP: AP CPU #%d launched\n", PCPU_GET(cpuid));
100 	else
101 		printf("%s%d%s", ap_awake == 2 ? "Launching APs: " : "",
102 		    PCPU_GET(cpuid), ap_awake == mp_ncpus ? "\n" : " ");
103 	mtx_unlock_spin(&ap_boot_mtx);
104 
105 	while(smp_started == 0)
106 		;
107 
108 	/* Start per-CPU event timers. */
109 	cpu_initclocks_ap();
110 
111 	/* Announce ourselves awake, and enter the scheduler */
112 	sched_ap_entry();
113 }
114 
115 void
116 cpu_mp_setmaxid(void)
117 {
118 	struct cpuref cpuref;
119 	int error;
120 
121 	mp_ncpus = 0;
122 	mp_maxid = 0;
123 	error = platform_smp_first_cpu(&cpuref);
124 	while (!error) {
125 		mp_ncpus++;
126 		mp_maxid = max(cpuref.cr_cpuid, mp_maxid);
127 		error = platform_smp_next_cpu(&cpuref);
128 	}
129 	/* Sanity. */
130 	if (mp_ncpus == 0)
131 		mp_ncpus = 1;
132 }
133 
134 int
135 cpu_mp_probe(void)
136 {
137 
138 	/*
139 	 * We're not going to enable SMP if there's only 1 processor.
140 	 */
141 	return (mp_ncpus > 1);
142 }
143 
144 void
145 cpu_mp_start(void)
146 {
147 	struct cpuref bsp, cpu;
148 	struct pcpu *pc;
149 	int domain, error;
150 
151 	error = platform_smp_get_bsp(&bsp);
152 	KASSERT(error == 0, ("Don't know BSP"));
153 
154 	error = platform_smp_first_cpu(&cpu);
155 	while (!error) {
156 		if (cpu.cr_cpuid >= MAXCPU) {
157 			printf("SMP: cpu%d: skipped -- ID out of range\n",
158 			    cpu.cr_cpuid);
159 			goto next;
160 		}
161 		if (CPU_ISSET(cpu.cr_cpuid, &all_cpus)) {
162 			printf("SMP: cpu%d: skipped - duplicate ID\n",
163 			    cpu.cr_cpuid);
164 			goto next;
165 		}
166 
167 		if (vm_ndomains > 1)
168 			domain = cpu.cr_domain;
169 		else
170 			domain = 0;
171 
172 		if (cpu.cr_cpuid != bsp.cr_cpuid) {
173 			void *dpcpu;
174 
175 			pc = &__pcpu[cpu.cr_cpuid];
176 			dpcpu = kmem_malloc_domainset(DOMAINSET_PREF(domain),
177 			    DPCPU_SIZE, M_WAITOK | M_ZERO);
178 			pcpu_init(pc, cpu.cr_cpuid, sizeof(*pc));
179 			dpcpu_init(dpcpu, cpu.cr_cpuid);
180 		} else {
181 			pc = pcpup;
182 			pc->pc_cpuid = bsp.cr_cpuid;
183 			pc->pc_bsp = 1;
184 		}
185 		pc->pc_domain = domain;
186 		pc->pc_hwref = cpu.cr_hwref;
187 
188 		CPU_SET(pc->pc_cpuid, &cpuset_domain[pc->pc_domain]);
189 		KASSERT(pc->pc_domain < MAXMEMDOM, ("bad domain value %d\n",
190 		    pc->pc_domain));
191 		CPU_SET(pc->pc_cpuid, &all_cpus);
192 next:
193 		error = platform_smp_next_cpu(&cpu);
194 	}
195 
196 #ifdef SMP
197 	platform_smp_probe_threads();
198 #endif
199 }
200 
201 void
202 cpu_mp_announce(void)
203 {
204 	struct pcpu *pc;
205 	int i;
206 
207 	if (!bootverbose)
208 		return;
209 
210 	CPU_FOREACH(i) {
211 		pc = pcpu_find(i);
212 		if (pc == NULL)
213 			continue;
214 		printf("cpu%d: dev=%x domain=%d ", i, (int)pc->pc_hwref, pc->pc_domain);
215 		if (pc->pc_bsp)
216 			printf(" (BSP)");
217 		printf("\n");
218 	}
219 }
220 
221 static void
222 cpu_mp_unleash(void *dummy)
223 {
224 	struct pcpu *pc;
225 	int cpus, timeout;
226 	int ret;
227 
228 	if (mp_ncpus <= 1)
229 		return;
230 
231 	mtx_init(&ap_boot_mtx, "ap boot", NULL, MTX_SPIN);
232 
233 	cpus = 0;
234 	smp_cpus = 0;
235 #ifdef BOOKE
236 	tlb1_ap_prep();
237 #endif
238 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
239 		cpus++;
240 		if (!pc->pc_bsp) {
241 			if (bootverbose)
242 				printf("Waking up CPU %d (dev=%x)\n",
243 				    pc->pc_cpuid, (int)pc->pc_hwref);
244 
245 			pc->pc_flags = PCPU_GET(flags); /* Copy cached CPU flags */
246 			ret = platform_smp_start_cpu(pc);
247 			if (ret == 0) {
248 				timeout = 2000;	/* wait 2sec for the AP */
249 				while (!pc->pc_awake && --timeout > 0)
250 					DELAY(1000);
251 			}
252 		} else {
253 			pc->pc_awake = 1;
254 		}
255 		if (pc->pc_awake) {
256 			if (bootverbose)
257 				printf("Adding CPU %d, hwref=%jx, awake=%x\n",
258 				    pc->pc_cpuid, (uintmax_t)pc->pc_hwref,
259 				    pc->pc_awake);
260 			smp_cpus++;
261 		} else
262 			CPU_SET(pc->pc_cpuid, &stopped_cpus);
263 	}
264 
265 	ap_awake = 1;
266 
267 	/* Provide our current DEC and TB values for APs */
268 	ap_timebase = mftb() + 10;
269 	__asm __volatile("msync; isync");
270 
271 	/* Let APs continue */
272 	atomic_store_rel_int(&ap_letgo, 1);
273 
274 	platform_smp_timebase_sync(ap_timebase, 0);
275 
276 	while (ap_awake < smp_cpus)
277 		;
278 
279 	if (smp_cpus != cpus || cpus != mp_ncpus) {
280 		printf("SMP: %d CPUs found; %d CPUs usable; %d CPUs woken\n",
281 		    mp_ncpus, cpus, smp_cpus);
282 	}
283 
284 	if (smp_cpus > 1)
285 		atomic_store_rel_int(&smp_started, 1);
286 
287 	/* Let the APs get into the scheduler */
288 	DELAY(10000);
289 
290 }
291 
292 SYSINIT(start_aps, SI_SUB_SMP, SI_ORDER_FIRST, cpu_mp_unleash, NULL);
293 
294 int
295 powerpc_ipi_handler(void *arg)
296 {
297 	u_int cpuid;
298 	uint32_t ipimask;
299 
300 	CTR2(KTR_SMP, "%s: MSR 0x%08x", __func__, mfmsr());
301 
302 	ipimask = atomic_readandclear_32(&(pcpup->pc_ipimask));
303 	if (ipimask == 0)
304 		return (FILTER_STRAY);
305 	if (ipimask & (1 << IPI_AST)) {
306 		CTR1(KTR_SMP, "%s: IPI_AST", __func__);
307 	}
308 	if (ipimask & (1 << IPI_PREEMPT)) {
309 		CTR1(KTR_SMP, "%s: IPI_PREEMPT", __func__);
310 		sched_preempt(curthread);
311 	}
312 	if (ipimask & (1 << IPI_RENDEZVOUS)) {
313 		CTR1(KTR_SMP, "%s: IPI_RENDEZVOUS", __func__);
314 		smp_rendezvous_action();
315 	}
316 	if (ipimask & (1 << IPI_STOP)) {
317 
318 		/*
319 		 * IPI_STOP_HARD is mapped to IPI_STOP so it is not
320 		 * necessary to add such case.
321 		 */
322 		CTR1(KTR_SMP, "%s: IPI_STOP or IPI_STOP_HARD (stop)",
323 				__func__);
324 		cpuid = PCPU_GET(cpuid);
325 		savectx(&stoppcbs[cpuid]);
326 		CPU_SET_ATOMIC(cpuid, &stopped_cpus);
327 		while (!CPU_ISSET(cpuid, &started_cpus))
328 			cpu_spinwait();
329 		CPU_CLR_ATOMIC(cpuid, &stopped_cpus);
330 		CPU_CLR_ATOMIC(cpuid, &started_cpus);
331 		CTR1(KTR_SMP, "%s: IPI_STOP (restart)", __func__);
332 	}
333 	if (ipimask & (1 << IPI_HARDCLOCK)) {
334 		CTR1(KTR_SMP, "%s: IPI_HARDCLOCK", __func__);
335 		hardclockintr();
336 	}
337 
338 	return (FILTER_HANDLED);
339 }
340 
341 static void
342 ipi_send(struct pcpu *pc, int ipi)
343 {
344 
345 	CTR4(KTR_SMP, "%s: pc=%p, targetcpu=%d, IPI=%d", __func__,
346 	    pc, pc->pc_cpuid, ipi);
347 
348 	atomic_set_32(&pc->pc_ipimask, (1 << ipi));
349 	powerpc_sync();
350 	PIC_IPI(root_pic, pc->pc_cpuid);
351 
352 	CTR1(KTR_SMP, "%s: sent", __func__);
353 }
354 
355 /* Send an IPI to a set of cpus. */
356 void
357 ipi_selected(cpuset_t cpus, int ipi)
358 {
359 	struct pcpu *pc;
360 
361 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
362 		if (CPU_ISSET(pc->pc_cpuid, &cpus))
363 			ipi_send(pc, ipi);
364 	}
365 }
366 
367 /* Send an IPI to a specific CPU. */
368 void
369 ipi_cpu(int cpu, u_int ipi)
370 {
371 
372 	ipi_send(cpuid_to_pcpu[cpu], ipi);
373 }
374 
375 /* Send an IPI to all CPUs EXCEPT myself. */
376 void
377 ipi_all_but_self(int ipi)
378 {
379 	struct pcpu *pc;
380 
381 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
382 		if (pc != pcpup)
383 			ipi_send(pc, ipi);
384 	}
385 }
386