xref: /freebsd/sys/powerpc/powerpc/mp_machdep.c (revision 731d06abf2105cc0873fa84e972178f9f37ca760)
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/lock.h>
39 #include <sys/malloc.h>
40 #include <sys/mutex.h>
41 #include <sys/pcpu.h>
42 #include <sys/proc.h>
43 #include <sys/sched.h>
44 #include <sys/smp.h>
45 
46 #include <vm/vm.h>
47 #include <vm/vm_param.h>
48 #include <vm/pmap.h>
49 #include <vm/vm_map.h>
50 #include <vm/vm_extern.h>
51 #include <vm/vm_kern.h>
52 
53 #include <machine/bus.h>
54 #include <machine/cpu.h>
55 #include <machine/intr_machdep.h>
56 #include <machine/pcb.h>
57 #include <machine/platform.h>
58 #include <machine/md_var.h>
59 #include <machine/setjmp.h>
60 #include <machine/smp.h>
61 
62 #include "pic_if.h"
63 
64 extern struct pcpu __pcpu[MAXCPU];
65 
66 volatile static int ap_awake;
67 volatile static u_int ap_letgo;
68 volatile static u_quad_t ap_timebase;
69 static u_int ipi_msg_cnt[32];
70 static struct mtx ap_boot_mtx;
71 struct pcb stoppcbs[MAXCPU];
72 
73 void
74 machdep_ap_bootstrap(void)
75 {
76 
77 	PCPU_SET(awake, 1);
78 	__asm __volatile("msync; isync");
79 
80 	while (ap_letgo == 0)
81 		nop_prio_vlow();
82 	nop_prio_medium();
83 
84 	/*
85 	 * Set timebase as soon as possible to meet an implicit rendezvous
86 	 * from cpu_mp_unleash(), which sets ap_letgo and then immediately
87 	 * sets timebase.
88 	 *
89 	 * Note that this is instrinsically racy and is only relevant on
90 	 * platforms that do not support better mechanisms.
91 	 */
92 	platform_smp_timebase_sync(ap_timebase, 1);
93 
94 	/* Give platform code a chance to do anything else necessary */
95 	platform_smp_ap_init();
96 
97 	/* Initialize decrementer */
98 	decr_ap_init();
99 
100 	/* Serialize console output and AP count increment */
101 	mtx_lock_spin(&ap_boot_mtx);
102 	ap_awake++;
103 	if (bootverbose)
104 		printf("SMP: AP CPU #%d launched\n", PCPU_GET(cpuid));
105 	else
106 		printf("%s%d%s", ap_awake == 2 ? "Launching APs: " : "",
107 		    PCPU_GET(cpuid), ap_awake == mp_ncpus ? "\n" : " ");
108 	mtx_unlock_spin(&ap_boot_mtx);
109 
110 	while(smp_started == 0)
111 		;
112 
113 	/* Start per-CPU event timers. */
114 	cpu_initclocks_ap();
115 
116 	/* Announce ourselves awake, and enter the scheduler */
117 	sched_throw(NULL);
118 }
119 
120 void
121 cpu_mp_setmaxid(void)
122 {
123 	struct cpuref cpuref;
124 	int error;
125 
126 	mp_ncpus = 0;
127 	mp_maxid = 0;
128 	error = platform_smp_first_cpu(&cpuref);
129 	while (!error) {
130 		mp_ncpus++;
131 		mp_maxid = max(cpuref.cr_cpuid, mp_maxid);
132 		error = platform_smp_next_cpu(&cpuref);
133 	}
134 	/* Sanity. */
135 	if (mp_ncpus == 0)
136 		mp_ncpus = 1;
137 }
138 
139 int
140 cpu_mp_probe(void)
141 {
142 
143 	/*
144 	 * We're not going to enable SMP if there's only 1 processor.
145 	 */
146 	return (mp_ncpus > 1);
147 }
148 
149 void
150 cpu_mp_start(void)
151 {
152 	struct cpuref bsp, cpu;
153 	struct pcpu *pc;
154 	int error;
155 
156 	error = platform_smp_get_bsp(&bsp);
157 	KASSERT(error == 0, ("Don't know BSP"));
158 
159 	error = platform_smp_first_cpu(&cpu);
160 	while (!error) {
161 		if (cpu.cr_cpuid >= MAXCPU) {
162 			printf("SMP: cpu%d: skipped -- ID out of range\n",
163 			    cpu.cr_cpuid);
164 			goto next;
165 		}
166 		if (CPU_ISSET(cpu.cr_cpuid, &all_cpus)) {
167 			printf("SMP: cpu%d: skipped - duplicate ID\n",
168 			    cpu.cr_cpuid);
169 			goto next;
170 		}
171 		if (cpu.cr_cpuid != bsp.cr_cpuid) {
172 			void *dpcpu;
173 
174 			pc = &__pcpu[cpu.cr_cpuid];
175 			dpcpu = (void *)kmem_malloc(DPCPU_SIZE, M_WAITOK |
176 			    M_ZERO);
177 			pcpu_init(pc, cpu.cr_cpuid, sizeof(*pc));
178 			dpcpu_init(dpcpu, cpu.cr_cpuid);
179 		} else {
180 			pc = pcpup;
181 			pc->pc_cpuid = bsp.cr_cpuid;
182 			pc->pc_bsp = 1;
183 		}
184 		pc->pc_hwref = cpu.cr_hwref;
185 
186 		if (vm_ndomains > 1)
187 			pc->pc_domain = cpu.cr_domain;
188 		else
189 			pc->pc_domain = 0;
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 			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 	int msg;
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 	while ((msg = ffs(ipimask) - 1) != -1) {
309 		ipimask &= ~(1u << msg);
310 		ipi_msg_cnt[msg]++;
311 		switch (msg) {
312 		case IPI_AST:
313 			CTR1(KTR_SMP, "%s: IPI_AST", __func__);
314 			break;
315 		case IPI_PREEMPT:
316 			CTR1(KTR_SMP, "%s: IPI_PREEMPT", __func__);
317 			sched_preempt(curthread);
318 			break;
319 		case IPI_RENDEZVOUS:
320 			CTR1(KTR_SMP, "%s: IPI_RENDEZVOUS", __func__);
321 			smp_rendezvous_action();
322 			break;
323 		case IPI_STOP:
324 
325 			/*
326 			 * IPI_STOP_HARD is mapped to IPI_STOP so it is not
327 			 * necessary to add such case in the switch.
328 			 */
329 			CTR1(KTR_SMP, "%s: IPI_STOP or IPI_STOP_HARD (stop)",
330 			    __func__);
331 			cpuid = PCPU_GET(cpuid);
332 			savectx(&stoppcbs[cpuid]);
333 			savectx(PCPU_GET(curpcb));
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