mp_machdep.c revision 222813 
1/*-
2 * Copyright (c) 2008 Marcel Moolenaar
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 *
9 * 1. Redistributions of source code must retain the above copyright
10 *    notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25 */
26
27#include <sys/cdefs.h>
28__FBSDID("$FreeBSD: head/sys/powerpc/powerpc/mp_machdep.c 222813 2011-06-07 08:46:13Z attilio $");
29
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/lock.h>
37#include <sys/mutex.h>
38#include <sys/pcpu.h>
39#include <sys/proc.h>
40#include <sys/sched.h>
41#include <sys/smp.h>
42
43#include <vm/vm.h>
44#include <vm/vm_param.h>
45#include <vm/pmap.h>
46#include <vm/vm_map.h>
47#include <vm/vm_extern.h>
48#include <vm/vm_kern.h>
49
50#include <machine/bus.h>
51#include <machine/cpu.h>
52#include <machine/intr_machdep.h>
53#include <machine/pcb.h>
54#include <machine/platform.h>
55#include <machine/md_var.h>
56#include <machine/smp.h>
57
58#include "pic_if.h"
59
60extern struct pcpu __pcpu[MAXCPU];
61
62volatile static int ap_awake;
63volatile static u_int ap_letgo;
64volatile static u_quad_t ap_timebase;
65static u_int ipi_msg_cnt[32];
66static struct mtx ap_boot_mtx;
67struct pcb stoppcbs[MAXCPU];
68
69void
70machdep_ap_bootstrap(void)
71{
72	/* Set up important bits on the CPU (HID registers, etc.) */
73	cpudep_ap_setup();
74
75	/* Set PIR */
76	PCPU_SET(pir, mfspr(SPR_PIR));
77	PCPU_SET(awake, 1);
78	__asm __volatile("msync; isync");
79
80	while (ap_letgo == 0)
81		;
82
83	/* Initialize DEC and TB, sync with the BSP values */
84#ifdef __powerpc64__
85	/* Writing to the time base register is hypervisor-privileged */
86	if (mfmsr() & PSL_HV)
87		mttb(ap_timebase);
88#else
89	mttb(ap_timebase);
90#endif
91	decr_ap_init();
92
93	/* Serialize console output and AP count increment */
94	mtx_lock_spin(&ap_boot_mtx);
95	ap_awake++;
96	printf("SMP: AP CPU #%d launched\n", PCPU_GET(cpuid));
97	mtx_unlock_spin(&ap_boot_mtx);
98
99	/* Initialize curthread */
100	PCPU_SET(curthread, PCPU_GET(idlethread));
101	PCPU_SET(curpcb, curthread->td_pcb);
102
103	/* Start per-CPU event timers. */
104	cpu_initclocks_ap();
105
106	/* Announce ourselves awake, and enter the scheduler */
107	sched_throw(NULL);
108}
109
110void
111cpu_mp_setmaxid(void)
112{
113	struct cpuref cpuref;
114	int error;
115
116	mp_ncpus = 0;
117	error = platform_smp_first_cpu(&cpuref);
118	while (!error) {
119		mp_ncpus++;
120		error = platform_smp_next_cpu(&cpuref);
121	}
122	/* Sanity. */
123	if (mp_ncpus == 0)
124		mp_ncpus = 1;
125
126	/*
127	 * Set the largest cpuid we're going to use. This is necessary
128	 * for VM initialization.
129	 */
130	mp_maxid = min(mp_ncpus, MAXCPU) - 1;
131}
132
133int
134cpu_mp_probe(void)
135{
136
137	/*
138	 * We're not going to enable SMP if there's only 1 processor.
139	 */
140	return (mp_ncpus > 1);
141}
142
143void
144cpu_mp_start(void)
145{
146	struct cpuref bsp, cpu;
147	struct pcpu *pc;
148	int error;
149
150	error = platform_smp_get_bsp(&bsp);
151	KASSERT(error == 0, ("Don't know BSP"));
152	KASSERT(bsp.cr_cpuid == 0, ("%s: cpuid != 0", __func__));
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		if (cpu.cr_cpuid != bsp.cr_cpuid) {
167			void *dpcpu;
168
169			pc = &__pcpu[cpu.cr_cpuid];
170			dpcpu = (void *)kmem_alloc(kernel_map, DPCPU_SIZE);
171			pcpu_init(pc, cpu.cr_cpuid, sizeof(*pc));
172			dpcpu_init(dpcpu, cpu.cr_cpuid);
173		} else {
174			pc = pcpup;
175			pc->pc_cpuid = bsp.cr_cpuid;
176			pc->pc_bsp = 1;
177		}
178		CPU_SETOF(pc->pc_cpuid, &pc->pc_cpumask);
179		pc->pc_hwref = cpu.cr_hwref;
180		CPU_OR(&all_cpus, &pc->pc_cpumask);
181next:
182		error = platform_smp_next_cpu(&cpu);
183	}
184}
185
186void
187cpu_mp_announce(void)
188{
189	struct pcpu *pc;
190	int i;
191
192	for (i = 0; i <= mp_maxid; i++) {
193		pc = pcpu_find(i);
194		if (pc == NULL)
195			continue;
196		printf("cpu%d: dev=%x", i, (int)pc->pc_hwref);
197		if (pc->pc_bsp)
198			printf(" (BSP)");
199		printf("\n");
200	}
201}
202
203static void
204cpu_mp_unleash(void *dummy)
205{
206	struct pcpu *pc;
207	int cpus, timeout;
208
209	if (mp_ncpus <= 1)
210		return;
211
212	mtx_init(&ap_boot_mtx, "ap boot", NULL, MTX_SPIN);
213
214	cpus = 0;
215	smp_cpus = 0;
216	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
217		cpus++;
218		pc->pc_other_cpus = all_cpus;
219		CPU_NAND(&pc->pc_other_cpus, &pc->pc_cpumask);
220		if (!pc->pc_bsp) {
221			if (bootverbose)
222				printf("Waking up CPU %d (dev=%x)\n",
223				    pc->pc_cpuid, (int)pc->pc_hwref);
224
225			platform_smp_start_cpu(pc);
226
227			timeout = 2000;	/* wait 2sec for the AP */
228			while (!pc->pc_awake && --timeout > 0)
229				DELAY(1000);
230
231		} else {
232			PCPU_SET(pir, mfspr(SPR_PIR));
233			pc->pc_awake = 1;
234		}
235		if (pc->pc_awake) {
236			if (bootverbose)
237				printf("Adding CPU %d, pir=%x, awake=%x\n",
238				    pc->pc_cpuid, pc->pc_pir, pc->pc_awake);
239			smp_cpus++;
240		} else
241			CPU_SET(pc->pc_cpuid, &stopped_cpus);
242	}
243
244	ap_awake = 1;
245
246	/* Provide our current DEC and TB values for APs */
247	ap_timebase = mftb() + 10;
248	__asm __volatile("msync; isync");
249
250	/* Let APs continue */
251	atomic_store_rel_int(&ap_letgo, 1);
252
253#ifdef __powerpc64__
254	/* Writing to the time base register is hypervisor-privileged */
255	if (mfmsr() & PSL_HV)
256		mttb(ap_timebase);
257#else
258	mttb(ap_timebase);
259#endif
260
261	while (ap_awake < smp_cpus)
262		;
263
264	if (smp_cpus != cpus || cpus != mp_ncpus) {
265		printf("SMP: %d CPUs found; %d CPUs usable; %d CPUs woken\n",
266		    mp_ncpus, cpus, smp_cpus);
267	}
268
269	/* Let the APs get into the scheduler */
270	DELAY(10000);
271
272	smp_active = 1;
273	smp_started = 1;
274}
275
276SYSINIT(start_aps, SI_SUB_SMP, SI_ORDER_FIRST, cpu_mp_unleash, NULL);
277
278int
279powerpc_ipi_handler(void *arg)
280{
281	cpuset_t self;
282	uint32_t ipimask;
283	int msg;
284
285	CTR2(KTR_SMP, "%s: MSR 0x%08x", __func__, mfmsr());
286
287	ipimask = atomic_readandclear_32(&(pcpup->pc_ipimask));
288	if (ipimask == 0)
289		return (FILTER_STRAY);
290	while ((msg = ffs(ipimask) - 1) != -1) {
291		ipimask &= ~(1u << msg);
292		ipi_msg_cnt[msg]++;
293		switch (msg) {
294		case IPI_AST:
295			CTR1(KTR_SMP, "%s: IPI_AST", __func__);
296			break;
297		case IPI_PREEMPT:
298			CTR1(KTR_SMP, "%s: IPI_PREEMPT", __func__);
299			sched_preempt(curthread);
300			break;
301		case IPI_RENDEZVOUS:
302			CTR1(KTR_SMP, "%s: IPI_RENDEZVOUS", __func__);
303			smp_rendezvous_action();
304			break;
305		case IPI_STOP:
306
307			/*
308			 * IPI_STOP_HARD is mapped to IPI_STOP so it is not
309			 * necessary to add such case in the switch.
310			 */
311			CTR1(KTR_SMP, "%s: IPI_STOP or IPI_STOP_HARD (stop)",
312			    __func__);
313			savectx(&stoppcbs[PCPU_GET(cpuid)]);
314			self = PCPU_GET(cpumask);
315			savectx(PCPU_GET(curpcb));
316			CPU_OR_ATOMIC(&stopped_cpus, &self);
317			while (!CPU_OVERLAP(&started_cpus, &self))
318				cpu_spinwait();
319			CPU_NAND_ATOMIC(&started_cpus, &self);
320			CPU_NAND_ATOMIC(&stopped_cpus, &self);
321			CTR1(KTR_SMP, "%s: IPI_STOP (restart)", __func__);
322			break;
323		case IPI_HARDCLOCK:
324			CTR1(KTR_SMP, "%s: IPI_HARDCLOCK", __func__);
325			hardclockintr();
326			break;
327		}
328	}
329
330	return (FILTER_HANDLED);
331}
332
333static void
334ipi_send(struct pcpu *pc, int ipi)
335{
336
337	CTR4(KTR_SMP, "%s: pc=%p, targetcpu=%d, IPI=%d", __func__,
338	    pc, pc->pc_cpuid, ipi);
339
340	atomic_set_32(&pc->pc_ipimask, (1 << ipi));
341	PIC_IPI(root_pic, pc->pc_cpuid);
342
343	CTR1(KTR_SMP, "%s: sent", __func__);
344}
345
346/* Send an IPI to a set of cpus. */
347void
348ipi_selected(cpuset_t cpus, int ipi)
349{
350	struct pcpu *pc;
351
352	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
353		if (CPU_OVERLAP(&cpus, &pc->pc_cpumask))
354			ipi_send(pc, ipi);
355	}
356}
357
358/* Send an IPI to a specific CPU. */
359void
360ipi_cpu(int cpu, u_int ipi)
361{
362
363	ipi_send(cpuid_to_pcpu[cpu], ipi);
364}
365
366/* Send an IPI to all CPUs EXCEPT myself. */
367void
368ipi_all_but_self(int ipi)
369{
370	struct pcpu *pc;
371
372	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
373		if (pc != pcpup)
374			ipi_send(pc, ipi);
375	}
376}
377