xref: /freebsd-13.1/sys/riscv/riscv/mp_machdep.c (revision 00bc7bbd)
1 /*-
2  * Copyright (c) 2015 The FreeBSD Foundation
3  * Copyright (c) 2016 Ruslan Bukin <[email protected]>
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/cpuset.h>
50 #include <sys/kernel.h>
51 #include <sys/ktr.h>
52 #include <sys/malloc.h>
53 #include <sys/module.h>
54 #include <sys/mutex.h>
55 #include <sys/proc.h>
56 #include <sys/sched.h>
57 #include <sys/smp.h>
58 
59 #include <vm/vm.h>
60 #include <vm/pmap.h>
61 #include <vm/vm_extern.h>
62 #include <vm/vm_kern.h>
63 #include <vm/vm_map.h>
64 
65 #include <machine/intr.h>
66 #include <machine/smp.h>
67 #include <machine/sbi.h>
68 
69 #ifdef FDT
70 #include <dev/ofw/openfirm.h>
71 #include <dev/ofw/ofw_cpu.h>
72 #endif
73 
74 boolean_t ofw_cpu_reg(phandle_t node, u_int, cell_t *);
75 
76 uint32_t __riscv_boot_ap[MAXCPU];
77 
78 static enum {
79 	CPUS_UNKNOWN,
80 #ifdef FDT
81 	CPUS_FDT,
82 #endif
83 } cpu_enum_method;
84 
85 static device_identify_t riscv64_cpu_identify;
86 static device_probe_t riscv64_cpu_probe;
87 static device_attach_t riscv64_cpu_attach;
88 
89 static int ipi_handler(void *);
90 
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(u_long hartid);
102 void init_secondary(uint64_t);
103 
104 static struct mtx ap_boot_mtx;
105 
106 /* Stacks for AP initialization, discarded once idle threads are started. */
107 void *bootstack;
108 static void *bootstacks[MAXCPU];
109 
110 /* Count of started APs, used to synchronize access to bootstack. */
111 static volatile int aps_started;
112 
113 /* Set to 1 once we're ready to let the APs out of the pen. */
114 static volatile int aps_ready;
115 
116 /* Temporary variables for init_secondary()  */
117 void *dpcpu[MAXCPU - 1];
118 
119 static device_method_t riscv64_cpu_methods[] = {
120 	/* Device interface */
121 	DEVMETHOD(device_identify,	riscv64_cpu_identify),
122 	DEVMETHOD(device_probe,		riscv64_cpu_probe),
123 	DEVMETHOD(device_attach,	riscv64_cpu_attach),
124 
125 	DEVMETHOD_END
126 };
127 
128 static devclass_t riscv64_cpu_devclass;
129 static driver_t riscv64_cpu_driver = {
130 	"riscv64_cpu",
131 	riscv64_cpu_methods,
132 	0
133 };
134 
135 DRIVER_MODULE(riscv64_cpu, cpu, riscv64_cpu_driver, riscv64_cpu_devclass, 0, 0);
136 
137 static void
riscv64_cpu_identify(driver_t * driver,device_t parent)138 riscv64_cpu_identify(driver_t *driver, device_t parent)
139 {
140 
141 	if (device_find_child(parent, "riscv64_cpu", -1) != NULL)
142 		return;
143 	if (BUS_ADD_CHILD(parent, 0, "riscv64_cpu", -1) == NULL)
144 		device_printf(parent, "add child failed\n");
145 }
146 
147 static int
riscv64_cpu_probe(device_t dev)148 riscv64_cpu_probe(device_t dev)
149 {
150 	u_int cpuid;
151 
152 	cpuid = device_get_unit(dev);
153 	if (cpuid >= MAXCPU || cpuid > mp_maxid)
154 		return (EINVAL);
155 
156 	device_quiet(dev);
157 	return (0);
158 }
159 
160 static int
riscv64_cpu_attach(device_t dev)161 riscv64_cpu_attach(device_t dev)
162 {
163 	const uint32_t *reg;
164 	size_t reg_size;
165 	u_int cpuid;
166 	int i;
167 
168 	cpuid = device_get_unit(dev);
169 
170 	if (cpuid >= MAXCPU || cpuid > mp_maxid)
171 		return (EINVAL);
172 	KASSERT(cpu_list[cpuid] == NULL, ("Already have cpu %u", cpuid));
173 
174 	reg = cpu_get_cpuid(dev, &reg_size);
175 	if (reg == NULL)
176 		return (EINVAL);
177 
178 	if (bootverbose) {
179 		device_printf(dev, "register <");
180 		for (i = 0; i < reg_size; i++)
181 			printf("%s%x", (i == 0) ? "" : " ", reg[i]);
182 		printf(">\n");
183 	}
184 
185 	/* Set the device to start it later */
186 	cpu_list[cpuid] = dev;
187 
188 	return (0);
189 }
190 
191 static void
release_aps(void * dummy __unused)192 release_aps(void *dummy __unused)
193 {
194 	cpuset_t mask;
195 	int i;
196 
197 	if (mp_ncpus == 1)
198 		return;
199 
200 	/* Setup the IPI handler */
201 	riscv_setup_ipihandler(ipi_handler);
202 
203 	atomic_store_rel_int(&aps_ready, 1);
204 
205 	/* Wake up the other CPUs */
206 	mask = all_harts;
207 	CPU_CLR(boot_hart, &mask);
208 
209 	printf("Release APs\n");
210 
211 	sbi_send_ipi(mask.__bits);
212 
213 	for (i = 0; i < 2000; i++) {
214 		if (smp_started)
215 			return;
216 		DELAY(1000);
217 	}
218 
219 	printf("APs not started\n");
220 }
221 SYSINIT(start_aps, SI_SUB_SMP, SI_ORDER_FIRST, release_aps, NULL);
222 
223 void
init_secondary(uint64_t hart)224 init_secondary(uint64_t hart)
225 {
226 	struct pcpu *pcpup;
227 	u_int cpuid;
228 
229 	/* Renumber this cpu */
230 	cpuid = hart;
231 	if (cpuid < boot_hart)
232 		cpuid += mp_maxid + 1;
233 	cpuid -= boot_hart;
234 
235 	/* Setup the pcpu pointer */
236 	pcpup = &__pcpu[cpuid];
237 	__asm __volatile("mv tp, %0" :: "r"(pcpup));
238 
239 	/* Workaround: make sure wfi doesn't halt the hart */
240 	csr_set(sie, SIE_SSIE);
241 	csr_set(sip, SIE_SSIE);
242 
243 	/* Signal the BSP and spin until it has released all APs. */
244 	atomic_add_int(&aps_started, 1);
245 	while (!atomic_load_int(&aps_ready))
246 		__asm __volatile("wfi");
247 
248 	/* Initialize curthread */
249 	KASSERT(PCPU_GET(idlethread) != NULL, ("no idle thread"));
250 	pcpup->pc_curthread = pcpup->pc_idlethread;
251 	schedinit_ap();
252 
253 	/*
254 	 * Identify current CPU. This is necessary to setup
255 	 * affinity registers and to provide support for
256 	 * runtime chip identification.
257 	 */
258 	identify_cpu();
259 
260 	/* Enable software interrupts */
261 	riscv_unmask_ipi();
262 
263 #ifndef EARLY_AP_STARTUP
264 	/* Start per-CPU event timers. */
265 	cpu_initclocks_ap();
266 #endif
267 
268 	/* Enable external (PLIC) interrupts */
269 	csr_set(sie, SIE_SEIE);
270 
271 	/* Activate this hart in the kernel pmap. */
272 	CPU_SET_ATOMIC(hart, &kernel_pmap->pm_active);
273 
274 	/* Activate process 0's pmap. */
275 	pmap_activate_boot(vmspace_pmap(proc0.p_vmspace));
276 
277 	mtx_lock_spin(&ap_boot_mtx);
278 
279 	atomic_add_rel_32(&smp_cpus, 1);
280 
281 	if (smp_cpus == mp_ncpus) {
282 		/* enable IPI's, tlb shootdown, freezes etc */
283 		atomic_store_rel_int(&smp_started, 1);
284 	}
285 
286 	mtx_unlock_spin(&ap_boot_mtx);
287 
288 	/*
289 	 * Assert that smp_after_idle_runnable condition is reasonable.
290 	 */
291 	MPASS(PCPU_GET(curpcb) == NULL);
292 
293 	/* Enter the scheduler */
294 	sched_throw(NULL);
295 
296 	panic("scheduler returned us to init_secondary");
297 	/* NOTREACHED */
298 }
299 
300 static void
smp_after_idle_runnable(void * arg __unused)301 smp_after_idle_runnable(void *arg __unused)
302 {
303 	struct pcpu *pc;
304 	int cpu;
305 
306 	for (cpu = 1; cpu <= mp_maxid; cpu++) {
307 		if (bootstacks[cpu] != NULL) {
308 			pc = pcpu_find(cpu);
309 			while (atomic_load_ptr(&pc->pc_curpcb) == NULL)
310 				cpu_spinwait();
311 			kmem_free((vm_offset_t)bootstacks[cpu], PAGE_SIZE);
312 		}
313 	}
314 }
315 SYSINIT(smp_after_idle_runnable, SI_SUB_SMP, SI_ORDER_ANY,
316     smp_after_idle_runnable, NULL);
317 
318 static int
ipi_handler(void * arg)319 ipi_handler(void *arg)
320 {
321 	u_int ipi_bitmap;
322 	u_int cpu, ipi;
323 	int bit;
324 
325 	csr_clear(sip, SIP_SSIP);
326 
327 	cpu = PCPU_GET(cpuid);
328 
329 	mb();
330 
331 	ipi_bitmap = atomic_readandclear_int(PCPU_PTR(pending_ipis));
332 	if (ipi_bitmap == 0)
333 		return (FILTER_HANDLED);
334 
335 	while ((bit = ffs(ipi_bitmap))) {
336 		bit = (bit - 1);
337 		ipi = (1 << bit);
338 		ipi_bitmap &= ~ipi;
339 
340 		mb();
341 
342 		switch (ipi) {
343 		case IPI_AST:
344 			CTR0(KTR_SMP, "IPI_AST");
345 			break;
346 		case IPI_PREEMPT:
347 			CTR1(KTR_SMP, "%s: IPI_PREEMPT", __func__);
348 			sched_preempt(curthread);
349 			break;
350 		case IPI_RENDEZVOUS:
351 			CTR0(KTR_SMP, "IPI_RENDEZVOUS");
352 			smp_rendezvous_action();
353 			break;
354 		case IPI_STOP:
355 		case IPI_STOP_HARD:
356 			CTR0(KTR_SMP, (ipi == IPI_STOP) ? "IPI_STOP" : "IPI_STOP_HARD");
357 			savectx(&stoppcbs[cpu]);
358 
359 			/* Indicate we are stopped */
360 			CPU_SET_ATOMIC(cpu, &stopped_cpus);
361 
362 			/* Wait for restart */
363 			while (!CPU_ISSET(cpu, &started_cpus))
364 				cpu_spinwait();
365 
366 			CPU_CLR_ATOMIC(cpu, &started_cpus);
367 			CPU_CLR_ATOMIC(cpu, &stopped_cpus);
368 			CTR0(KTR_SMP, "IPI_STOP (restart)");
369 
370 			/*
371 			 * The kernel debugger might have set a breakpoint,
372 			 * so flush the instruction cache.
373 			 */
374 			fence_i();
375 			break;
376 		case IPI_HARDCLOCK:
377 			CTR1(KTR_SMP, "%s: IPI_HARDCLOCK", __func__);
378 			hardclockintr();
379 			break;
380 		default:
381 			panic("Unknown IPI %#0x on cpu %d", ipi, curcpu);
382 		}
383 	}
384 
385 	return (FILTER_HANDLED);
386 }
387 
388 struct cpu_group *
cpu_topo(void)389 cpu_topo(void)
390 {
391 
392 	return (smp_topo_none());
393 }
394 
395 /* Determine if we running MP machine */
396 int
cpu_mp_probe(void)397 cpu_mp_probe(void)
398 {
399 
400 	return (mp_ncpus > 1);
401 }
402 
403 #ifdef FDT
404 static boolean_t
cpu_init_fdt(u_int id,phandle_t node,u_int addr_size,pcell_t * reg)405 cpu_init_fdt(u_int id, phandle_t node, u_int addr_size, pcell_t *reg)
406 {
407 	struct pcpu *pcpup;
408 	vm_paddr_t start_addr;
409 	uint64_t hart;
410 	u_int cpuid;
411 	int naps;
412 	int error;
413 
414 	/* Check if this hart supports MMU. */
415 	if (OF_getproplen(node, "mmu-type") < 0)
416 		return (0);
417 
418 	KASSERT(id < MAXCPU, ("Too many CPUs"));
419 
420 	KASSERT(addr_size == 1 || addr_size == 2, ("Invalid register size"));
421 #ifdef INVARIANTS
422 	cpu_reg[id][0] = reg[0];
423 	if (addr_size == 2)
424 		cpu_reg[id][1] = reg[1];
425 #endif
426 
427 	hart = reg[0];
428 	if (addr_size == 2) {
429 		hart <<= 32;
430 		hart |= reg[1];
431 	}
432 
433 	KASSERT(hart < MAXCPU, ("Too many harts."));
434 
435 	/* We are already running on this cpu */
436 	if (hart == boot_hart)
437 		return (1);
438 
439 	/*
440 	 * Rotate the CPU IDs to put the boot CPU as CPU 0.
441 	 * We keep the other CPUs ordered.
442 	 */
443 	cpuid = hart;
444 	if (cpuid < boot_hart)
445 		cpuid += mp_maxid + 1;
446 	cpuid -= boot_hart;
447 
448 	/* Check if we are able to start this cpu */
449 	if (cpuid > mp_maxid)
450 		return (0);
451 
452 	/*
453 	 * Depending on the SBI implementation, APs are waiting either in
454 	 * locore.S or to be activated explicitly, via SBI call.
455 	 */
456 	if (sbi_probe_extension(SBI_EXT_ID_HSM) != 0) {
457 		start_addr = pmap_kextract((vm_offset_t)mpentry);
458 		error = sbi_hsm_hart_start(hart, start_addr, 0);
459 		if (error != 0) {
460 			mp_ncpus--;
461 
462 			/* Send a warning to the user and continue. */
463 			printf("AP %u (hart %lu) failed to start, error %d\n",
464 			    cpuid, hart, error);
465 			return (0);
466 		}
467 	}
468 
469 	pcpup = &__pcpu[cpuid];
470 	pcpu_init(pcpup, cpuid, sizeof(struct pcpu));
471 	pcpup->pc_hart = hart;
472 
473 	dpcpu[cpuid - 1] = (void *)kmem_malloc(DPCPU_SIZE, M_WAITOK | M_ZERO);
474 	dpcpu_init(dpcpu[cpuid - 1], cpuid);
475 
476 	bootstacks[cpuid] = (void *)kmem_malloc(PAGE_SIZE, M_WAITOK | M_ZERO);
477 
478 	naps = atomic_load_int(&aps_started);
479 	bootstack = (char *)bootstacks[cpuid] + PAGE_SIZE;
480 
481 	printf("Starting CPU %u (hart %lx)\n", cpuid, hart);
482 	atomic_store_32(&__riscv_boot_ap[hart], 1);
483 
484 	/* Wait for the AP to switch to its boot stack. */
485 	while (atomic_load_int(&aps_started) < naps + 1)
486 		cpu_spinwait();
487 
488 	CPU_SET(cpuid, &all_cpus);
489 	CPU_SET(hart, &all_harts);
490 
491 	return (1);
492 }
493 #endif
494 
495 /* Initialize and fire up non-boot processors */
496 void
cpu_mp_start(void)497 cpu_mp_start(void)
498 {
499 
500 	mtx_init(&ap_boot_mtx, "ap boot", NULL, MTX_SPIN);
501 
502 	CPU_SET(0, &all_cpus);
503 	CPU_SET(boot_hart, &all_harts);
504 
505 	switch(cpu_enum_method) {
506 #ifdef FDT
507 	case CPUS_FDT:
508 		ofw_cpu_early_foreach(cpu_init_fdt, true);
509 		break;
510 #endif
511 	case CPUS_UNKNOWN:
512 		break;
513 	}
514 }
515 
516 /* Introduce rest of cores to the world */
517 void
cpu_mp_announce(void)518 cpu_mp_announce(void)
519 {
520 }
521 
522 static boolean_t
cpu_check_mmu(u_int id,phandle_t node,u_int addr_size,pcell_t * reg)523 cpu_check_mmu(u_int id, phandle_t node, u_int addr_size, pcell_t *reg)
524 {
525 
526 	/* Check if this hart supports MMU. */
527 	if (OF_getproplen(node, "mmu-type") < 0)
528 		return (0);
529 
530 	return (1);
531 }
532 
533 void
cpu_mp_setmaxid(void)534 cpu_mp_setmaxid(void)
535 {
536 #ifdef FDT
537 	int cores;
538 
539 	cores = ofw_cpu_early_foreach(cpu_check_mmu, true);
540 	if (cores > 0) {
541 		cores = MIN(cores, MAXCPU);
542 		if (bootverbose)
543 			printf("Found %d CPUs in the device tree\n", cores);
544 		mp_ncpus = cores;
545 		mp_maxid = cores - 1;
546 		cpu_enum_method = CPUS_FDT;
547 		return;
548 	}
549 #endif
550 
551 	if (bootverbose)
552 		printf("No CPU data, limiting to 1 core\n");
553 	mp_ncpus = 1;
554 	mp_maxid = 0;
555 }
556