xref: /freebsd/sys/powerpc/pseries/xics.c (revision 0957b409)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright 2011 Nathan Whitehorn
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
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,
20  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
21  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
22  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
23  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include "opt_platform.h"
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/module.h>
36 #include <sys/bus.h>
37 #include <sys/conf.h>
38 #include <sys/kernel.h>
39 #include <sys/malloc.h>
40 #include <sys/smp.h>
41 
42 #include <vm/vm.h>
43 #include <vm/pmap.h>
44 
45 #include <machine/bus.h>
46 #include <machine/intr_machdep.h>
47 #include <machine/md_var.h>
48 #include <machine/rtas.h>
49 
50 #include <dev/ofw/ofw_bus.h>
51 #include <dev/ofw/ofw_bus_subr.h>
52 
53 #ifdef POWERNV
54 #include <powerpc/powernv/opal.h>
55 #endif
56 
57 #include "phyp-hvcall.h"
58 #include "pic_if.h"
59 
60 #define XICP_PRIORITY	5	/* Random non-zero number */
61 #define XICP_IPI	2
62 #define MAX_XICP_IRQS	(1<<24)	/* 24-bit XIRR field */
63 
64 static int	xicp_probe(device_t);
65 static int	xicp_attach(device_t);
66 static int	xics_probe(device_t);
67 static int	xics_attach(device_t);
68 
69 static void	xicp_bind(device_t dev, u_int irq, cpuset_t cpumask, void **priv);
70 static void	xicp_dispatch(device_t, struct trapframe *);
71 static void	xicp_enable(device_t, u_int, u_int, void **priv);
72 static void	xicp_eoi(device_t, u_int, void *priv);
73 static void	xicp_ipi(device_t, u_int);
74 static void	xicp_mask(device_t, u_int, void *priv);
75 static void	xicp_unmask(device_t, u_int, void *priv);
76 
77 #ifdef POWERNV
78 extern void (*powernv_smp_ap_extra_init)(void);
79 static void	xicp_smp_cpu_startup(void);
80 #endif
81 
82 static device_method_t  xicp_methods[] = {
83 	/* Device interface */
84 	DEVMETHOD(device_probe,		xicp_probe),
85 	DEVMETHOD(device_attach,	xicp_attach),
86 
87 	/* PIC interface */
88 	DEVMETHOD(pic_bind,		xicp_bind),
89 	DEVMETHOD(pic_dispatch,		xicp_dispatch),
90 	DEVMETHOD(pic_enable,		xicp_enable),
91 	DEVMETHOD(pic_eoi,		xicp_eoi),
92 	DEVMETHOD(pic_ipi,		xicp_ipi),
93 	DEVMETHOD(pic_mask,		xicp_mask),
94 	DEVMETHOD(pic_unmask,		xicp_unmask),
95 
96 	DEVMETHOD_END
97 };
98 
99 static device_method_t  xics_methods[] = {
100 	/* Device interface */
101 	DEVMETHOD(device_probe,		xics_probe),
102 	DEVMETHOD(device_attach,	xics_attach),
103 
104 	DEVMETHOD_END
105 };
106 
107 struct xicp_intvec {
108 	int irq;
109 	int vector;
110 	int cpu;
111 };
112 
113 struct xicp_softc {
114 	struct mtx sc_mtx;
115 	struct resource *mem[MAXCPU];
116 
117 	int cpu_range[2];
118 
119 	int ibm_int_on;
120 	int ibm_int_off;
121 	int ibm_get_xive;
122 	int ibm_set_xive;
123 
124 	/* XXX: inefficient -- hash table? tree? */
125 	struct xicp_intvec intvecs[256];
126 	int nintvecs;
127 	int ipi_vec;
128 	bool xics_emu;
129 };
130 
131 static driver_t xicp_driver = {
132 	"xicp",
133 	xicp_methods,
134 	sizeof(struct xicp_softc)
135 };
136 
137 static driver_t xics_driver = {
138 	"xics",
139 	xics_methods,
140 	0
141 };
142 
143 #ifdef POWERNV
144 /* We can only pass physical addresses into OPAL.  Kernel stacks are in the KVA,
145  * not in the direct map, so we need to somehow extract the physical address.
146  * However, pmap_kextract() takes locks, which is forbidden in a critical region
147  * (which PIC_DISPATCH() operates in).  The kernel is mapped into the Direct
148  * Map (0xc000....), and the CPU implicitly drops the top two bits when doing
149  * real address by nature that the bus width is smaller than 64-bits.  Placing
150  * cpu_xirr into the DMAP lets us take advantage of this and avoids the
151  * pmap_kextract() that would otherwise be needed if using the stack variable.
152  */
153 static uint32_t cpu_xirr[MAXCPU];
154 #endif
155 
156 static devclass_t xicp_devclass;
157 static devclass_t xics_devclass;
158 
159 EARLY_DRIVER_MODULE(xicp, ofwbus, xicp_driver, xicp_devclass, 0, 0,
160     BUS_PASS_INTERRUPT-1);
161 EARLY_DRIVER_MODULE(xics, ofwbus, xics_driver, xics_devclass, 0, 0,
162     BUS_PASS_INTERRUPT);
163 
164 #ifdef POWERNV
165 static struct resource *
166 xicp_mem_for_cpu(int cpu)
167 {
168 	device_t dev;
169 	struct xicp_softc *sc;
170 	int i;
171 
172 	for (i = 0; (dev = devclass_get_device(xicp_devclass, i)) != NULL; i++){
173 		sc = device_get_softc(dev);
174 		if (cpu >= sc->cpu_range[0] && cpu < sc->cpu_range[1])
175 			return (sc->mem[cpu - sc->cpu_range[0]]);
176 	}
177 
178 	return (NULL);
179 }
180 #endif
181 
182 static int
183 xicp_probe(device_t dev)
184 {
185 
186 	if (!ofw_bus_is_compatible(dev, "ibm,ppc-xicp") &&
187 	    !ofw_bus_is_compatible(dev, "ibm,opal-intc"))
188 		return (ENXIO);
189 
190 	device_set_desc(dev, "External Interrupt Presentation Controller");
191 	return (BUS_PROBE_GENERIC);
192 }
193 
194 static int
195 xics_probe(device_t dev)
196 {
197 
198 	if (!ofw_bus_is_compatible(dev, "ibm,ppc-xics") &&
199 	    !ofw_bus_is_compatible(dev, "IBM,opal-xics"))
200 		return (ENXIO);
201 
202 	device_set_desc(dev, "External Interrupt Source Controller");
203 	return (BUS_PROBE_GENERIC);
204 }
205 
206 static int
207 xicp_attach(device_t dev)
208 {
209 	struct xicp_softc *sc = device_get_softc(dev);
210 	phandle_t phandle = ofw_bus_get_node(dev);
211 
212 	if (rtas_exists()) {
213 		sc->ibm_int_on = rtas_token_lookup("ibm,int-on");
214 		sc->ibm_int_off = rtas_token_lookup("ibm,int-off");
215 		sc->ibm_set_xive = rtas_token_lookup("ibm,set-xive");
216 		sc->ibm_get_xive = rtas_token_lookup("ibm,get-xive");
217 #ifdef POWERNV
218 	} else if (opal_check() == 0) {
219 		/* No init needed */
220 #endif
221 	} else {
222 		device_printf(dev, "Cannot attach without RTAS or OPAL\n");
223 		return (ENXIO);
224 	}
225 
226 	if (OF_hasprop(phandle, "ibm,interrupt-server-ranges")) {
227 		OF_getencprop(phandle, "ibm,interrupt-server-ranges",
228 		    sc->cpu_range, sizeof(sc->cpu_range));
229 		sc->cpu_range[1] += sc->cpu_range[0];
230 		device_printf(dev, "Handling CPUs %d-%d\n", sc->cpu_range[0],
231 		    sc->cpu_range[1]-1);
232 #ifdef POWERNV
233 	} else if (ofw_bus_is_compatible(dev, "ibm,opal-intc")) {
234 			/*
235 			 * For now run POWER9 XIVE interrupt controller in XICS
236 			 * compatibility mode.
237 			 */
238 			sc->xics_emu = true;
239 			opal_call(OPAL_XIVE_RESET, OPAL_XIVE_XICS_MODE_EMU);
240 #endif
241 	} else {
242 		sc->cpu_range[0] = 0;
243 		sc->cpu_range[1] = mp_ncpus;
244 	}
245 
246 #ifdef POWERNV
247 	if (mfmsr() & PSL_HV) {
248 		int i;
249 
250 		if (sc->xics_emu) {
251 			opal_call(OPAL_INT_SET_CPPR, 0xff);
252 			for (i = 0; i < mp_ncpus; i++) {
253 				opal_call(OPAL_INT_SET_MFRR,
254 				    pcpu_find(i)->pc_hwref, 0xff);
255 			}
256 		} else {
257 			for (i = 0; i < sc->cpu_range[1] - sc->cpu_range[0]; i++) {
258 				sc->mem[i] = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
259 				    &i, RF_ACTIVE);
260 				if (sc->mem[i] == NULL) {
261 					device_printf(dev, "Could not alloc mem "
262 					    "resource %d\n", i);
263 					return (ENXIO);
264 				}
265 
266 				/* Unmask interrupts on all cores */
267 				bus_write_1(sc->mem[i], 4, 0xff);
268 				bus_write_1(sc->mem[i], 12, 0xff);
269 			}
270 		}
271 	}
272 #endif
273 
274 	mtx_init(&sc->sc_mtx, "XICP", NULL, MTX_DEF);
275 	sc->nintvecs = 0;
276 
277 	powerpc_register_pic(dev, OF_xref_from_node(phandle), MAX_XICP_IRQS,
278 	    1 /* Number of IPIs */, FALSE);
279 	root_pic = dev;
280 
281 #ifdef POWERNV
282 	if (sc->xics_emu)
283 		powernv_smp_ap_extra_init = xicp_smp_cpu_startup;
284 #endif
285 
286 	return (0);
287 }
288 
289 static int
290 xics_attach(device_t dev)
291 {
292 	phandle_t phandle = ofw_bus_get_node(dev);
293 
294 	/* The XICP (root PIC) will handle all our interrupts */
295 	powerpc_register_pic(root_pic, OF_xref_from_node(phandle),
296 	    MAX_XICP_IRQS, 1 /* Number of IPIs */, FALSE);
297 
298 	return (0);
299 }
300 
301 static __inline struct xicp_intvec *
302 xicp_setup_priv(struct xicp_softc *sc, u_int irq, void **priv)
303 {
304 	if (*priv == NULL) {
305 		KASSERT(sc->nintvecs + 1 < nitems(sc->intvecs),
306 			("Too many XICP interrupts"));
307 		mtx_lock(&sc->sc_mtx);
308 		*priv = &sc->intvecs[sc->nintvecs++];
309 		mtx_unlock(&sc->sc_mtx);
310 	}
311 
312 	return (*priv);
313 }
314 
315 /*
316  * PIC I/F methods.
317  */
318 
319 static void
320 xicp_bind(device_t dev, u_int irq, cpuset_t cpumask, void **priv)
321 {
322 	struct xicp_softc *sc = device_get_softc(dev);
323 	struct xicp_intvec *iv;
324 	cell_t status, cpu;
325 	int ncpus, i, error;
326 
327 	/* Ignore IPIs */
328 	if (irq == MAX_XICP_IRQS)
329 		return;
330 
331 	iv = xicp_setup_priv(sc, irq, priv);
332 
333 	/*
334 	 * This doesn't appear to actually support affinity groups, so pick a
335 	 * random CPU.
336 	 */
337 	ncpus = 0;
338 	CPU_FOREACH(cpu)
339 		if (CPU_ISSET(cpu, &cpumask)) ncpus++;
340 
341 	i = mftb() % ncpus;
342 	ncpus = 0;
343 	CPU_FOREACH(cpu) {
344 		if (!CPU_ISSET(cpu, &cpumask))
345 			continue;
346 		if (ncpus == i)
347 			break;
348 		ncpus++;
349 	}
350 
351 	cpu = pcpu_find(cpu)->pc_hwref;
352 	iv->cpu = cpu;
353 
354 	if (rtas_exists())
355 		error = rtas_call_method(sc->ibm_set_xive, 3, 1, irq, cpu,
356 		    XICP_PRIORITY, &status);
357 #ifdef POWERNV
358 	else
359 		error = opal_call(OPAL_SET_XIVE, irq, cpu << 2, XICP_PRIORITY);
360 #endif
361 
362 	if (error < 0)
363 		panic("Cannot bind interrupt %d to CPU %d", irq, cpu);
364 }
365 
366 static void
367 xicp_dispatch(device_t dev, struct trapframe *tf)
368 {
369 	struct xicp_softc *sc;
370 	struct resource *regs = NULL;
371 	uint64_t xirr, junk;
372 	int i;
373 
374 	sc = device_get_softc(dev);
375 #ifdef POWERNV
376 	if ((mfmsr() & PSL_HV) && !sc->xics_emu) {
377 		regs = xicp_mem_for_cpu(PCPU_GET(hwref));
378 		KASSERT(regs != NULL,
379 		    ("Can't find regs for CPU %ld", (uintptr_t)PCPU_GET(hwref)));
380 	}
381 #endif
382 
383 	for (;;) {
384 		/* Return value in R4, use the PFT call */
385 		if (regs) {
386 			xirr = bus_read_4(regs, 4);
387 #ifdef POWERNV
388 		} else if (sc->xics_emu) {
389 			opal_call(OPAL_INT_GET_XIRR, &cpu_xirr[PCPU_GET(cpuid)],
390 			    false);
391 			xirr = cpu_xirr[PCPU_GET(cpuid)];
392 #endif
393 		} else {
394 			/* Return value in R4, use the PFT call */
395 			phyp_pft_hcall(H_XIRR, 0, 0, 0, 0, &xirr, &junk, &junk);
396 		}
397 		xirr &= 0x00ffffff;
398 
399 		if (xirr == 0) /* No more pending interrupts? */
400 			break;
401 
402 		if (xirr == XICP_IPI) {		/* Magic number for IPIs */
403 			xirr = MAX_XICP_IRQS;	/* Map to FreeBSD magic */
404 
405 			/* Clear IPI */
406 			if (regs)
407 				bus_write_1(regs, 12, 0xff);
408 #ifdef POWERNV
409 			else if (sc->xics_emu)
410 				opal_call(OPAL_INT_SET_MFRR,
411 				    PCPU_GET(hwref), 0xff);
412 #endif
413 			else
414 				phyp_hcall(H_IPI, (uint64_t)(PCPU_GET(hwref)),
415 				    0xff);
416 			i = sc->ipi_vec;
417 		} else {
418 
419 			/* XXX: super inefficient */
420 			for (i = 0; i < sc->nintvecs; i++) {
421 				if (sc->intvecs[i].irq == xirr)
422 					break;
423 			}
424 			KASSERT(i < sc->nintvecs, ("Unmapped XIRR"));
425 		}
426 
427 		powerpc_dispatch_intr(sc->intvecs[i].vector, tf);
428 	}
429 }
430 
431 static void
432 xicp_enable(device_t dev, u_int irq, u_int vector, void **priv)
433 {
434 	struct xicp_softc *sc;
435 	struct xicp_intvec *intr;
436 	cell_t status, cpu;
437 
438 	sc = device_get_softc(dev);
439 
440 	/* Bind to this CPU to start: distrib. ID is last entry in gserver# */
441 	cpu = PCPU_GET(hwref);
442 
443 	intr = xicp_setup_priv(sc, irq, priv);
444 
445 	intr->irq = irq;
446 	intr->vector = vector;
447 	intr->cpu = cpu;
448 	mb();
449 
450 	/* IPIs are also enabled.  Stash off the vector index */
451 	if (irq == MAX_XICP_IRQS) {
452 		sc->ipi_vec = intr - sc->intvecs;
453 		return;
454 	}
455 
456 	if (rtas_exists()) {
457 		rtas_call_method(sc->ibm_set_xive, 3, 1, irq, cpu,
458 		    XICP_PRIORITY, &status);
459 		xicp_unmask(dev, irq, intr);
460 #ifdef POWERNV
461 	} else {
462 		status = opal_call(OPAL_SET_XIVE, irq, cpu << 2, XICP_PRIORITY);
463 		/* Unmask implicit for OPAL */
464 
465 		if (status != 0)
466 			panic("OPAL_SET_XIVE IRQ %d -> cpu %d failed: %d", irq,
467 			    cpu, status);
468 #endif
469 	}
470 }
471 
472 static void
473 xicp_eoi(device_t dev, u_int irq, void *priv)
474 {
475 #ifdef POWERNV
476 	struct xicp_softc *sc;
477 #endif
478 	uint64_t xirr;
479 
480 	if (irq == MAX_XICP_IRQS) /* Remap IPI interrupt to internal value */
481 		irq = XICP_IPI;
482 	xirr = irq | (0xff << 24);
483 
484 #ifdef POWERNV
485 	if (mfmsr() & PSL_HV) {
486 		sc = device_get_softc(dev);
487 		if (sc->xics_emu)
488 			opal_call(OPAL_INT_EOI, xirr);
489 		else
490 			bus_write_4(xicp_mem_for_cpu(PCPU_GET(hwref)), 4, xirr);
491 	} else
492 #endif
493 		phyp_hcall(H_EOI, xirr);
494 }
495 
496 static void
497 xicp_ipi(device_t dev, u_int cpu)
498 {
499 
500 #ifdef POWERNV
501 	struct xicp_softc *sc;
502 	cpu = pcpu_find(cpu)->pc_hwref;
503 
504 	if (mfmsr() & PSL_HV) {
505 		sc = device_get_softc(dev);
506 		if (sc->xics_emu) {
507 			int64_t rv;
508 			rv = opal_call(OPAL_INT_SET_MFRR, cpu, XICP_PRIORITY);
509 			if (rv != 0)
510 			    device_printf(dev, "IPI SET_MFRR result: %ld\n", rv);
511 		} else
512 			bus_write_1(xicp_mem_for_cpu(cpu), 12, XICP_PRIORITY);
513 	} else
514 #endif
515 		phyp_hcall(H_IPI, (uint64_t)cpu, XICP_PRIORITY);
516 }
517 
518 static void
519 xicp_mask(device_t dev, u_int irq, void *priv)
520 {
521 	struct xicp_softc *sc = device_get_softc(dev);
522 	cell_t status;
523 
524 	if (irq == MAX_XICP_IRQS)
525 		return;
526 
527 	if (rtas_exists()) {
528 		rtas_call_method(sc->ibm_int_off, 1, 1, irq, &status);
529 #ifdef POWERNV
530 	} else {
531 		struct xicp_intvec *ivec = priv;
532 
533 		KASSERT(ivec != NULL, ("Masking unconfigured interrupt"));
534 		opal_call(OPAL_SET_XIVE, irq, ivec->cpu << 2, 0xff);
535 #endif
536 	}
537 }
538 
539 static void
540 xicp_unmask(device_t dev, u_int irq, void *priv)
541 {
542 	struct xicp_softc *sc = device_get_softc(dev);
543 	cell_t status;
544 
545 	if (irq == MAX_XICP_IRQS)
546 		return;
547 
548 	if (rtas_exists()) {
549 		rtas_call_method(sc->ibm_int_on, 1, 1, irq, &status);
550 #ifdef POWERNV
551 	} else {
552 		struct xicp_intvec *ivec = priv;
553 
554 		KASSERT(ivec != NULL, ("Unmasking unconfigured interrupt"));
555 		opal_call(OPAL_SET_XIVE, irq, ivec->cpu << 2, XICP_PRIORITY);
556 #endif
557 	}
558 }
559 
560 #ifdef POWERNV
561 /* This is only used on POWER9 systems with the XIVE's XICS emulation. */
562 static void
563 xicp_smp_cpu_startup(void)
564 {
565 	struct xicp_softc *sc;
566 
567 	if (mfmsr() & PSL_HV) {
568 		sc = device_get_softc(root_pic);
569 
570 		if (sc->xics_emu)
571 			opal_call(OPAL_INT_SET_CPPR, 0xff);
572 	}
573 }
574 #endif
575