xref: /freebsd/sys/riscv/riscv/mp_machdep.c (revision c697fb7f)
1 /*-
2  * Copyright (c) 2015 The FreeBSD Foundation
3  * Copyright (c) 2016 Ruslan Bukin <br@bsdpad.com>
4  * All rights reserved.
5  *
6  * Portions of this software were developed by Andrew Turner under
7  * sponsorship from the FreeBSD Foundation.
8  *
9  * Portions of this software were developed by SRI International and the
10  * University of Cambridge Computer Laboratory under DARPA/AFRL contract
11  * FA8750-10-C-0237 ("CTSRD"), as part of the DARPA CRASH research programme.
12  *
13  * Portions of this software were developed by the University of Cambridge
14  * Computer Laboratory as part of the CTSRD Project, with support from the
15  * UK Higher Education Innovation Fund (HEIF).
16  *
17  * Redistribution and use in source and binary forms, with or without
18  * modification, are permitted provided that the following conditions
19  * are met:
20  * 1. Redistributions of source code must retain the above copyright
21  *    notice, this list of conditions and the following disclaimer.
22  * 2. Redistributions in binary form must reproduce the above copyright
23  *    notice, this list of conditions and the following disclaimer in the
24  *    documentation and/or other materials provided with the distribution.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  */
38 
39 #include "opt_kstack_pages.h"
40 #include "opt_platform.h"
41 
42 #include <sys/cdefs.h>
43 __FBSDID("$FreeBSD$");
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/bus.h>
48 #include <sys/cpu.h>
49 #include <sys/kernel.h>
50 #include <sys/ktr.h>
51 #include <sys/malloc.h>
52 #include <sys/module.h>
53 #include <sys/mutex.h>
54 #include <sys/proc.h>
55 #include <sys/sched.h>
56 #include <sys/smp.h>
57 
58 #include <vm/vm.h>
59 #include <vm/pmap.h>
60 #include <vm/vm_extern.h>
61 #include <vm/vm_kern.h>
62 #include <vm/vm_map.h>
63 
64 #include <machine/intr.h>
65 #include <machine/smp.h>
66 #include <machine/sbi.h>
67 
68 #ifdef FDT
69 #include <dev/ofw/openfirm.h>
70 #include <dev/ofw/ofw_cpu.h>
71 #endif
72 
73 boolean_t ofw_cpu_reg(phandle_t node, u_int, cell_t *);
74 
75 uint32_t __riscv_boot_ap[MAXCPU];
76 
77 static enum {
78 	CPUS_UNKNOWN,
79 #ifdef FDT
80 	CPUS_FDT,
81 #endif
82 } cpu_enum_method;
83 
84 static device_identify_t riscv64_cpu_identify;
85 static device_probe_t riscv64_cpu_probe;
86 static device_attach_t riscv64_cpu_attach;
87 
88 static int ipi_handler(void *);
89 
90 struct mtx ap_boot_mtx;
91 struct pcb stoppcbs[MAXCPU];
92 
93 extern uint32_t boot_hart;
94 extern cpuset_t all_harts;
95 
96 #ifdef INVARIANTS
97 static uint32_t cpu_reg[MAXCPU][2];
98 #endif
99 static device_t cpu_list[MAXCPU];
100 
101 void mpentry(unsigned long cpuid);
102 void init_secondary(uint64_t);
103 
104 uint8_t secondary_stacks[MAXCPU][PAGE_SIZE * KSTACK_PAGES] __aligned(16);
105 
106 /* Set to 1 once we're ready to let the APs out of the pen. */
107 volatile int aps_ready = 0;
108 
109 /* Temporary variables for init_secondary()  */
110 void *dpcpu[MAXCPU - 1];
111 
112 static device_method_t riscv64_cpu_methods[] = {
113 	/* Device interface */
114 	DEVMETHOD(device_identify,	riscv64_cpu_identify),
115 	DEVMETHOD(device_probe,		riscv64_cpu_probe),
116 	DEVMETHOD(device_attach,	riscv64_cpu_attach),
117 
118 	DEVMETHOD_END
119 };
120 
121 static devclass_t riscv64_cpu_devclass;
122 static driver_t riscv64_cpu_driver = {
123 	"riscv64_cpu",
124 	riscv64_cpu_methods,
125 	0
126 };
127 
128 DRIVER_MODULE(riscv64_cpu, cpu, riscv64_cpu_driver, riscv64_cpu_devclass, 0, 0);
129 
130 static void
131 riscv64_cpu_identify(driver_t *driver, device_t parent)
132 {
133 
134 	if (device_find_child(parent, "riscv64_cpu", -1) != NULL)
135 		return;
136 	if (BUS_ADD_CHILD(parent, 0, "riscv64_cpu", -1) == NULL)
137 		device_printf(parent, "add child failed\n");
138 }
139 
140 static int
141 riscv64_cpu_probe(device_t dev)
142 {
143 	u_int cpuid;
144 
145 	cpuid = device_get_unit(dev);
146 	if (cpuid >= MAXCPU || cpuid > mp_maxid)
147 		return (EINVAL);
148 
149 	device_quiet(dev);
150 	return (0);
151 }
152 
153 static int
154 riscv64_cpu_attach(device_t dev)
155 {
156 	const uint32_t *reg;
157 	size_t reg_size;
158 	u_int cpuid;
159 	int i;
160 
161 	cpuid = device_get_unit(dev);
162 
163 	if (cpuid >= MAXCPU || cpuid > mp_maxid)
164 		return (EINVAL);
165 	KASSERT(cpu_list[cpuid] == NULL, ("Already have cpu %u", cpuid));
166 
167 	reg = cpu_get_cpuid(dev, &reg_size);
168 	if (reg == NULL)
169 		return (EINVAL);
170 
171 	if (bootverbose) {
172 		device_printf(dev, "register <");
173 		for (i = 0; i < reg_size; i++)
174 			printf("%s%x", (i == 0) ? "" : " ", reg[i]);
175 		printf(">\n");
176 	}
177 
178 	/* Set the device to start it later */
179 	cpu_list[cpuid] = dev;
180 
181 	return (0);
182 }
183 
184 static void
185 release_aps(void *dummy __unused)
186 {
187 	cpuset_t mask;
188 	int i;
189 
190 	if (mp_ncpus == 1)
191 		return;
192 
193 	/* Setup the IPI handler */
194 	riscv_setup_ipihandler(ipi_handler);
195 
196 	atomic_store_rel_int(&aps_ready, 1);
197 
198 	/* Wake up the other CPUs */
199 	mask = all_harts;
200 	CPU_CLR(boot_hart, &mask);
201 
202 	printf("Release APs\n");
203 
204 	sbi_send_ipi(mask.__bits);
205 
206 	for (i = 0; i < 2000; i++) {
207 		if (smp_started)
208 			return;
209 		DELAY(1000);
210 	}
211 
212 	printf("APs not started\n");
213 }
214 SYSINIT(start_aps, SI_SUB_SMP, SI_ORDER_FIRST, release_aps, NULL);
215 
216 void
217 init_secondary(uint64_t hart)
218 {
219 	struct pcpu *pcpup;
220 	u_int cpuid;
221 
222 	/* Renumber this cpu */
223 	cpuid = hart;
224 	if (cpuid < boot_hart)
225 		cpuid += mp_maxid + 1;
226 	cpuid -= boot_hart;
227 
228 	/* Setup the pcpu pointer */
229 	pcpup = &__pcpu[cpuid];
230 	__asm __volatile("mv tp, %0" :: "r"(pcpup));
231 
232 	/* Workaround: make sure wfi doesn't halt the hart */
233 	csr_set(sie, SIE_SSIE);
234 	csr_set(sip, SIE_SSIE);
235 
236 	/* Spin until the BSP releases the APs */
237 	while (!aps_ready)
238 		__asm __volatile("wfi");
239 
240 	/* Initialize curthread */
241 	KASSERT(PCPU_GET(idlethread) != NULL, ("no idle thread"));
242 	pcpup->pc_curthread = pcpup->pc_idlethread;
243 	pcpup->pc_curpcb = pcpup->pc_idlethread->td_pcb;
244 
245 	/*
246 	 * Identify current CPU. This is necessary to setup
247 	 * affinity registers and to provide support for
248 	 * runtime chip identification.
249 	 */
250 	identify_cpu();
251 
252 	/* Enable software interrupts */
253 	riscv_unmask_ipi();
254 
255 #ifndef EARLY_AP_STARTUP
256 	/* Start per-CPU event timers. */
257 	cpu_initclocks_ap();
258 #endif
259 
260 	/* Enable external (PLIC) interrupts */
261 	csr_set(sie, SIE_SEIE);
262 
263 	/* Activate process 0's pmap. */
264 	pmap_activate_boot(vmspace_pmap(proc0.p_vmspace));
265 
266 	mtx_lock_spin(&ap_boot_mtx);
267 
268 	atomic_add_rel_32(&smp_cpus, 1);
269 
270 	if (smp_cpus == mp_ncpus) {
271 		/* enable IPI's, tlb shootdown, freezes etc */
272 		atomic_store_rel_int(&smp_started, 1);
273 	}
274 
275 	mtx_unlock_spin(&ap_boot_mtx);
276 
277 	/* Enter the scheduler */
278 	sched_throw(NULL);
279 
280 	panic("scheduler returned us to init_secondary");
281 	/* NOTREACHED */
282 }
283 
284 static int
285 ipi_handler(void *arg)
286 {
287 	u_int ipi_bitmap;
288 	u_int cpu, ipi;
289 	int bit;
290 
291 	sbi_clear_ipi();
292 
293 	cpu = PCPU_GET(cpuid);
294 
295 	mb();
296 
297 	ipi_bitmap = atomic_readandclear_int(PCPU_PTR(pending_ipis));
298 	if (ipi_bitmap == 0)
299 		return (FILTER_HANDLED);
300 
301 	while ((bit = ffs(ipi_bitmap))) {
302 		bit = (bit - 1);
303 		ipi = (1 << bit);
304 		ipi_bitmap &= ~ipi;
305 
306 		mb();
307 
308 		switch (ipi) {
309 		case IPI_AST:
310 			CTR0(KTR_SMP, "IPI_AST");
311 			break;
312 		case IPI_PREEMPT:
313 			CTR1(KTR_SMP, "%s: IPI_PREEMPT", __func__);
314 			sched_preempt(curthread);
315 			break;
316 		case IPI_RENDEZVOUS:
317 			CTR0(KTR_SMP, "IPI_RENDEZVOUS");
318 			smp_rendezvous_action();
319 			break;
320 		case IPI_STOP:
321 		case IPI_STOP_HARD:
322 			CTR0(KTR_SMP, (ipi == IPI_STOP) ? "IPI_STOP" : "IPI_STOP_HARD");
323 			savectx(&stoppcbs[cpu]);
324 
325 			/* Indicate we are stopped */
326 			CPU_SET_ATOMIC(cpu, &stopped_cpus);
327 
328 			/* Wait for restart */
329 			while (!CPU_ISSET(cpu, &started_cpus))
330 				cpu_spinwait();
331 
332 			CPU_CLR_ATOMIC(cpu, &started_cpus);
333 			CPU_CLR_ATOMIC(cpu, &stopped_cpus);
334 			CTR0(KTR_SMP, "IPI_STOP (restart)");
335 
336 			/*
337 			 * The kernel debugger might have set a breakpoint,
338 			 * so flush the instruction cache.
339 			 */
340 			fence_i();
341 			break;
342 		case IPI_HARDCLOCK:
343 			CTR1(KTR_SMP, "%s: IPI_HARDCLOCK", __func__);
344 			hardclockintr();
345 			break;
346 		default:
347 			panic("Unknown IPI %#0x on cpu %d", ipi, curcpu);
348 		}
349 	}
350 
351 	return (FILTER_HANDLED);
352 }
353 
354 struct cpu_group *
355 cpu_topo(void)
356 {
357 
358 	return (smp_topo_none());
359 }
360 
361 /* Determine if we running MP machine */
362 int
363 cpu_mp_probe(void)
364 {
365 
366 	return (mp_ncpus > 1);
367 }
368 
369 #ifdef FDT
370 static boolean_t
371 cpu_init_fdt(u_int id, phandle_t node, u_int addr_size, pcell_t *reg)
372 {
373 	struct pcpu *pcpup;
374 	uint64_t hart;
375 	u_int cpuid;
376 
377 	/* Check if this hart supports MMU. */
378 	if (OF_getproplen(node, "mmu-type") < 0)
379 		return (0);
380 
381 	KASSERT(id < MAXCPU, ("Too many CPUs"));
382 
383 	KASSERT(addr_size == 1 || addr_size == 2, ("Invalid register size"));
384 #ifdef INVARIANTS
385 	cpu_reg[id][0] = reg[0];
386 	if (addr_size == 2)
387 		cpu_reg[id][1] = reg[1];
388 #endif
389 
390 	hart = reg[0];
391 	if (addr_size == 2) {
392 		hart <<= 32;
393 		hart |= reg[1];
394 	}
395 
396 	KASSERT(hart < MAXCPU, ("Too many harts."));
397 
398 	/* We are already running on this cpu */
399 	if (hart == boot_hart)
400 		return (1);
401 
402 	/*
403 	 * Rotate the CPU IDs to put the boot CPU as CPU 0.
404 	 * We keep the other CPUs ordered.
405 	 */
406 	cpuid = hart;
407 	if (cpuid < boot_hart)
408 		cpuid += mp_maxid + 1;
409 	cpuid -= boot_hart;
410 
411 	/* Check if we are able to start this cpu */
412 	if (cpuid > mp_maxid)
413 		return (0);
414 
415 	pcpup = &__pcpu[cpuid];
416 	pcpu_init(pcpup, cpuid, sizeof(struct pcpu));
417 	pcpup->pc_hart = hart;
418 
419 	dpcpu[cpuid - 1] = (void *)kmem_malloc(DPCPU_SIZE, M_WAITOK | M_ZERO);
420 	dpcpu_init(dpcpu[cpuid - 1], cpuid);
421 
422 	printf("Starting CPU %u (hart %lx)\n", cpuid, hart);
423 	__riscv_boot_ap[hart] = 1;
424 
425 	CPU_SET(cpuid, &all_cpus);
426 	CPU_SET(hart, &all_harts);
427 
428 	return (1);
429 }
430 #endif
431 
432 /* Initialize and fire up non-boot processors */
433 void
434 cpu_mp_start(void)
435 {
436 
437 	mtx_init(&ap_boot_mtx, "ap boot", NULL, MTX_SPIN);
438 
439 	CPU_SET(0, &all_cpus);
440 	CPU_SET(boot_hart, &all_harts);
441 
442 	switch(cpu_enum_method) {
443 #ifdef FDT
444 	case CPUS_FDT:
445 		ofw_cpu_early_foreach(cpu_init_fdt, true);
446 		break;
447 #endif
448 	case CPUS_UNKNOWN:
449 		break;
450 	}
451 }
452 
453 /* Introduce rest of cores to the world */
454 void
455 cpu_mp_announce(void)
456 {
457 }
458 
459 static boolean_t
460 cpu_check_mmu(u_int id, phandle_t node, u_int addr_size, pcell_t *reg)
461 {
462 
463 	/* Check if this hart supports MMU. */
464 	if (OF_getproplen(node, "mmu-type") < 0)
465 		return (0);
466 
467 	return (1);
468 }
469 
470 void
471 cpu_mp_setmaxid(void)
472 {
473 #ifdef FDT
474 	int cores;
475 
476 	cores = ofw_cpu_early_foreach(cpu_check_mmu, true);
477 	if (cores > 0) {
478 		cores = MIN(cores, MAXCPU);
479 		if (bootverbose)
480 			printf("Found %d CPUs in the device tree\n", cores);
481 		mp_ncpus = cores;
482 		mp_maxid = cores - 1;
483 		cpu_enum_method = CPUS_FDT;
484 		return;
485 	}
486 #endif
487 
488 	if (bootverbose)
489 		printf("No CPU data, limiting to 1 core\n");
490 	mp_ncpus = 1;
491 	mp_maxid = 0;
492 }
493