xref: /freebsd/sys/powerpc/pseries/xics.c (revision ba3c1f5972d7b90feb6e6da47905ff2757e0fe57)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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/lock.h>
40 #include <sys/malloc.h>
41 #include <sys/mutex.h>
42 #include <sys/smp.h>
43 
44 #include <vm/vm.h>
45 #include <vm/pmap.h>
46 
47 #include <machine/bus.h>
48 #include <machine/intr_machdep.h>
49 #include <machine/md_var.h>
50 #include <machine/rtas.h>
51 
52 #include <dev/ofw/ofw_bus.h>
53 #include <dev/ofw/ofw_bus_subr.h>
54 
55 #ifdef POWERNV
56 #include <powerpc/powernv/opal.h>
57 #endif
58 
59 #include "phyp-hvcall.h"
60 #include "pic_if.h"
61 
62 #define XICP_PRIORITY	5	/* Random non-zero number */
63 #define XICP_IPI	2
64 #define MAX_XICP_IRQS	(1<<24)	/* 24-bit XIRR field */
65 
66 static int	xicp_probe(device_t);
67 static int	xicp_attach(device_t);
68 static int	xics_probe(device_t);
69 static int	xics_attach(device_t);
70 
71 static void	xicp_bind(device_t dev, u_int irq, cpuset_t cpumask, void **priv);
72 static void	xicp_dispatch(device_t, struct trapframe *);
73 static void	xicp_enable(device_t, u_int, u_int, void **priv);
74 static void	xicp_eoi(device_t, u_int, void *priv);
75 static void	xicp_ipi(device_t, u_int);
76 static void	xicp_mask(device_t, u_int, void *priv);
77 static void	xicp_unmask(device_t, u_int, void *priv);
78 
79 #ifdef POWERNV
80 extern void (*powernv_smp_ap_extra_init)(void);
81 static void	xicp_smp_cpu_startup(void);
82 #endif
83 
84 static device_method_t  xicp_methods[] = {
85 	/* Device interface */
86 	DEVMETHOD(device_probe,		xicp_probe),
87 	DEVMETHOD(device_attach,	xicp_attach),
88 
89 	/* PIC interface */
90 	DEVMETHOD(pic_bind,		xicp_bind),
91 	DEVMETHOD(pic_dispatch,		xicp_dispatch),
92 	DEVMETHOD(pic_enable,		xicp_enable),
93 	DEVMETHOD(pic_eoi,		xicp_eoi),
94 	DEVMETHOD(pic_ipi,		xicp_ipi),
95 	DEVMETHOD(pic_mask,		xicp_mask),
96 	DEVMETHOD(pic_unmask,		xicp_unmask),
97 
98 	DEVMETHOD_END
99 };
100 
101 static device_method_t  xics_methods[] = {
102 	/* Device interface */
103 	DEVMETHOD(device_probe,		xics_probe),
104 	DEVMETHOD(device_attach,	xics_attach),
105 
106 	DEVMETHOD_END
107 };
108 
109 struct xicp_intvec {
110 	int irq;
111 	int vector;
112 	int cpu;
113 };
114 
115 struct xicp_softc {
116 	struct mtx sc_mtx;
117 	struct resource *mem[MAXCPU];
118 
119 	int cpu_range[2];
120 
121 	int ibm_int_on;
122 	int ibm_int_off;
123 	int ibm_get_xive;
124 	int ibm_set_xive;
125 
126 	/* XXX: inefficient -- hash table? tree? */
127 	struct xicp_intvec intvecs[256];
128 	int nintvecs;
129 	int ipi_vec;
130 	bool xics_emu;
131 };
132 
133 static driver_t xicp_driver = {
134 	"xicp",
135 	xicp_methods,
136 	sizeof(struct xicp_softc)
137 };
138 
139 static driver_t xics_driver = {
140 	"xics",
141 	xics_methods,
142 	0
143 };
144 
145 #ifdef POWERNV
146 /* We can only pass physical addresses into OPAL.  Kernel stacks are in the KVA,
147  * not in the direct map, so we need to somehow extract the physical address.
148  * However, pmap_kextract() takes locks, which is forbidden in a critical region
149  * (which PIC_DISPATCH() operates in).  The kernel is mapped into the Direct
150  * Map (0xc000....), and the CPU implicitly drops the top two bits when doing
151  * real address by nature that the bus width is smaller than 64-bits.  Placing
152  * cpu_xirr into the DMAP lets us take advantage of this and avoids the
153  * pmap_kextract() that would otherwise be needed if using the stack variable.
154  */
155 static uint32_t cpu_xirr[MAXCPU];
156 #endif
157 
158 EARLY_DRIVER_MODULE(xicp, ofwbus, xicp_driver, 0, 0, BUS_PASS_INTERRUPT - 1);
159 EARLY_DRIVER_MODULE(xics, ofwbus, xics_driver, 0, 0, BUS_PASS_INTERRUPT);
160 
161 #ifdef POWERNV
162 static struct resource *
163 xicp_mem_for_cpu(int cpu)
164 {
165 	devclass_t dc;
166 	device_t dev;
167 	struct xicp_softc *sc;
168 	int i;
169 
170 	dc = devclass_find(xicp_driver.name);
171 	for (i = 0; (dev = devclass_get_device(dc, i)) != NULL; i++){
172 		sc = device_get_softc(dev);
173 		if (cpu >= sc->cpu_range[0] && cpu < sc->cpu_range[1])
174 			return (sc->mem[cpu - sc->cpu_range[0]]);
175 	}
176 
177 	return (NULL);
178 }
179 #endif
180 
181 static int
182 xicp_probe(device_t dev)
183 {
184 
185 	if (!ofw_bus_is_compatible(dev, "ibm,ppc-xicp") &&
186 	    !ofw_bus_is_compatible(dev, "ibm,opal-intc"))
187 		return (ENXIO);
188 
189 	device_set_desc(dev, "External Interrupt Presentation Controller");
190 	return (BUS_PROBE_GENERIC);
191 }
192 
193 static int
194 xics_probe(device_t dev)
195 {
196 
197 	if (!ofw_bus_is_compatible(dev, "ibm,ppc-xics") &&
198 	    !ofw_bus_is_compatible(dev, "IBM,opal-xics"))
199 		return (ENXIO);
200 
201 	device_set_desc(dev, "External Interrupt Source Controller");
202 	return (BUS_PROBE_GENERIC);
203 }
204 
205 static int
206 xicp_attach(device_t dev)
207 {
208 	struct xicp_softc *sc = device_get_softc(dev);
209 	phandle_t phandle = ofw_bus_get_node(dev);
210 
211 	if (rtas_exists()) {
212 		sc->ibm_int_on = rtas_token_lookup("ibm,int-on");
213 		sc->ibm_int_off = rtas_token_lookup("ibm,int-off");
214 		sc->ibm_set_xive = rtas_token_lookup("ibm,set-xive");
215 		sc->ibm_get_xive = rtas_token_lookup("ibm,get-xive");
216 #ifdef POWERNV
217 	} else if (opal_check() == 0) {
218 		/* No init needed */
219 #endif
220 	} else {
221 		device_printf(dev, "Cannot attach without RTAS or OPAL\n");
222 		return (ENXIO);
223 	}
224 
225 	if (OF_hasprop(phandle, "ibm,interrupt-server-ranges")) {
226 		OF_getencprop(phandle, "ibm,interrupt-server-ranges",
227 		    sc->cpu_range, sizeof(sc->cpu_range));
228 		sc->cpu_range[1] += sc->cpu_range[0];
229 		device_printf(dev, "Handling CPUs %d-%d\n", sc->cpu_range[0],
230 		    sc->cpu_range[1]-1);
231 #ifdef POWERNV
232 	} else if (ofw_bus_is_compatible(dev, "ibm,opal-intc")) {
233 			/*
234 			 * For now run POWER9 XIVE interrupt controller in XICS
235 			 * compatibility mode.
236 			 */
237 			sc->xics_emu = true;
238 			opal_call(OPAL_XIVE_RESET, OPAL_XIVE_XICS_MODE_EMU);
239 #endif
240 	} else {
241 		sc->cpu_range[0] = 0;
242 		sc->cpu_range[1] = mp_ncpus;
243 	}
244 
245 #ifdef POWERNV
246 	if (mfmsr() & PSL_HV) {
247 		int i;
248 
249 		if (sc->xics_emu) {
250 			opal_call(OPAL_INT_SET_CPPR, 0xff);
251 			for (i = 0; i < mp_ncpus; i++) {
252 				opal_call(OPAL_INT_SET_MFRR,
253 				    pcpu_find(i)->pc_hwref, 0xff);
254 			}
255 		} else {
256 			for (i = 0; i < sc->cpu_range[1] - sc->cpu_range[0]; i++) {
257 				sc->mem[i] = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
258 				    &i, RF_ACTIVE);
259 				if (sc->mem[i] == NULL) {
260 					device_printf(dev, "Could not alloc mem "
261 					    "resource %d\n", i);
262 					return (ENXIO);
263 				}
264 
265 				/* Unmask interrupts on all cores */
266 				bus_write_1(sc->mem[i], 4, 0xff);
267 				bus_write_1(sc->mem[i], 12, 0xff);
268 			}
269 		}
270 	}
271 #endif
272 
273 	mtx_init(&sc->sc_mtx, "XICP", NULL, MTX_DEF);
274 	sc->nintvecs = 0;
275 
276 	powerpc_register_pic(dev, OF_xref_from_node(phandle), MAX_XICP_IRQS,
277 	    1 /* Number of IPIs */, FALSE);
278 	root_pic = dev;
279 
280 #ifdef POWERNV
281 	if (sc->xics_emu)
282 		powernv_smp_ap_extra_init = xicp_smp_cpu_startup;
283 #endif
284 
285 	return (0);
286 }
287 
288 static int
289 xics_attach(device_t dev)
290 {
291 	phandle_t phandle = ofw_bus_get_node(dev);
292 
293 	/* The XICP (root PIC) will handle all our interrupts */
294 	powerpc_register_pic(root_pic, OF_xref_from_node(phandle),
295 	    MAX_XICP_IRQS, 1 /* Number of IPIs */, FALSE);
296 
297 	return (0);
298 }
299 
300 static __inline struct xicp_intvec *
301 xicp_setup_priv(struct xicp_softc *sc, u_int irq, void **priv)
302 {
303 	if (*priv == NULL) {
304 		KASSERT(sc->nintvecs + 1 < nitems(sc->intvecs),
305 			("Too many XICP interrupts"));
306 		mtx_lock(&sc->sc_mtx);
307 		*priv = &sc->intvecs[sc->nintvecs++];
308 		mtx_unlock(&sc->sc_mtx);
309 	}
310 
311 	return (*priv);
312 }
313 
314 /*
315  * PIC I/F methods.
316  */
317 
318 static void
319 xicp_bind(device_t dev, u_int irq, cpuset_t cpumask, void **priv)
320 {
321 	struct xicp_softc *sc = device_get_softc(dev);
322 	struct xicp_intvec *iv;
323 	cell_t status, cpu;
324 	int ncpus, i, error = -1;
325 
326 	/* Ignore IPIs */
327 	if (irq == MAX_XICP_IRQS)
328 		return;
329 
330 	iv = xicp_setup_priv(sc, irq, priv);
331 
332 	/*
333 	 * This doesn't appear to actually support affinity groups, so pick a
334 	 * random CPU.
335 	 */
336 	ncpus = 0;
337 	CPU_FOREACH(cpu)
338 		if (CPU_ISSET(cpu, &cpumask)) ncpus++;
339 
340 	i = mftb() % ncpus;
341 	ncpus = 0;
342 	CPU_FOREACH(cpu) {
343 		if (!CPU_ISSET(cpu, &cpumask))
344 			continue;
345 		if (ncpus == i)
346 			break;
347 		ncpus++;
348 	}
349 
350 	cpu = pcpu_find(cpu)->pc_hwref;
351 	iv->cpu = cpu;
352 
353 	if (rtas_exists())
354 		error = rtas_call_method(sc->ibm_set_xive, 3, 1, irq, cpu,
355 		    XICP_PRIORITY, &status);
356 #ifdef POWERNV
357 	else
358 		error = opal_call(OPAL_SET_XIVE, irq, cpu << 2, XICP_PRIORITY);
359 #endif
360 
361 	if (error < 0)
362 		panic("Cannot bind interrupt %d to CPU %d", irq, cpu);
363 }
364 
365 static void
366 xicp_dispatch(device_t dev, struct trapframe *tf)
367 {
368 	struct xicp_softc *sc;
369 	struct resource *regs = NULL;
370 	uint64_t xirr, junk;
371 	int i;
372 
373 	sc = device_get_softc(dev);
374 #ifdef POWERNV
375 	if ((mfmsr() & PSL_HV) && !sc->xics_emu) {
376 		regs = xicp_mem_for_cpu(PCPU_GET(hwref));
377 		KASSERT(regs != NULL,
378 		    ("Can't find regs for CPU %ld", (uintptr_t)PCPU_GET(hwref)));
379 	}
380 #endif
381 
382 	for (;;) {
383 		/* Return value in R4, use the PFT call */
384 		if (regs) {
385 			xirr = bus_read_4(regs, 4);
386 #ifdef POWERNV
387 		} else if (sc->xics_emu) {
388 			opal_call(OPAL_INT_GET_XIRR, &cpu_xirr[PCPU_GET(cpuid)],
389 			    false);
390 			xirr = cpu_xirr[PCPU_GET(cpuid)];
391 #endif
392 		} else {
393 			/* Return value in R4, use the PFT call */
394 			phyp_pft_hcall(H_XIRR, 0, 0, 0, 0, &xirr, &junk, &junk);
395 		}
396 		xirr &= 0x00ffffff;
397 
398 		if (xirr == 0) /* No more pending interrupts? */
399 			break;
400 
401 		if (xirr == XICP_IPI) {		/* Magic number for IPIs */
402 			xirr = MAX_XICP_IRQS;	/* Map to FreeBSD magic */
403 
404 			/* Clear IPI */
405 			if (regs)
406 				bus_write_1(regs, 12, 0xff);
407 #ifdef POWERNV
408 			else if (sc->xics_emu)
409 				opal_call(OPAL_INT_SET_MFRR,
410 				    PCPU_GET(hwref), 0xff);
411 #endif
412 			else
413 				phyp_hcall(H_IPI, (uint64_t)(PCPU_GET(hwref)),
414 				    0xff);
415 			i = sc->ipi_vec;
416 		} else {
417 			/* XXX: super inefficient */
418 			for (i = 0; i < sc->nintvecs; i++) {
419 				if (sc->intvecs[i].irq == xirr)
420 					break;
421 			}
422 			KASSERT(i < sc->nintvecs, ("Unmapped XIRR"));
423 		}
424 
425 		powerpc_dispatch_intr(sc->intvecs[i].vector, tf);
426 	}
427 }
428 
429 static void
430 xicp_enable(device_t dev, u_int irq, u_int vector, void **priv)
431 {
432 	struct xicp_softc *sc;
433 	struct xicp_intvec *intr;
434 	cell_t status, cpu;
435 
436 	sc = device_get_softc(dev);
437 
438 	/* Bind to this CPU to start: distrib. ID is last entry in gserver# */
439 	cpu = PCPU_GET(hwref);
440 
441 	intr = xicp_setup_priv(sc, irq, priv);
442 
443 	intr->irq = irq;
444 	intr->vector = vector;
445 	intr->cpu = cpu;
446 	mb();
447 
448 	/* IPIs are also enabled.  Stash off the vector index */
449 	if (irq == MAX_XICP_IRQS) {
450 		sc->ipi_vec = intr - sc->intvecs;
451 		return;
452 	}
453 
454 	if (rtas_exists()) {
455 		rtas_call_method(sc->ibm_set_xive, 3, 1, irq, cpu,
456 		    XICP_PRIORITY, &status);
457 		xicp_unmask(dev, irq, intr);
458 #ifdef POWERNV
459 	} else {
460 		status = opal_call(OPAL_SET_XIVE, irq, cpu << 2, XICP_PRIORITY);
461 		/* Unmask implicit for OPAL */
462 
463 		if (status != 0)
464 			panic("OPAL_SET_XIVE IRQ %d -> cpu %d failed: %d", irq,
465 			    cpu, status);
466 #endif
467 	}
468 }
469 
470 static void
471 xicp_eoi(device_t dev, u_int irq, void *priv)
472 {
473 #ifdef POWERNV
474 	struct xicp_softc *sc;
475 #endif
476 	uint64_t xirr;
477 
478 	if (irq == MAX_XICP_IRQS) /* Remap IPI interrupt to internal value */
479 		irq = XICP_IPI;
480 	xirr = irq | (0xff << 24);
481 
482 #ifdef POWERNV
483 	if (mfmsr() & PSL_HV) {
484 		sc = device_get_softc(dev);
485 		if (sc->xics_emu)
486 			opal_call(OPAL_INT_EOI, xirr);
487 		else
488 			bus_write_4(xicp_mem_for_cpu(PCPU_GET(hwref)), 4, xirr);
489 	} else
490 #endif
491 		phyp_hcall(H_EOI, xirr);
492 }
493 
494 static void
495 xicp_ipi(device_t dev, u_int cpu)
496 {
497 
498 #ifdef POWERNV
499 	struct xicp_softc *sc;
500 	cpu = pcpu_find(cpu)->pc_hwref;
501 
502 	if (mfmsr() & PSL_HV) {
503 		sc = device_get_softc(dev);
504 		if (sc->xics_emu) {
505 			int64_t rv;
506 			rv = opal_call(OPAL_INT_SET_MFRR, cpu, XICP_PRIORITY);
507 			if (rv != 0)
508 			    device_printf(dev, "IPI SET_MFRR result: %ld\n", rv);
509 		} else
510 			bus_write_1(xicp_mem_for_cpu(cpu), 12, XICP_PRIORITY);
511 	} else
512 #endif
513 		phyp_hcall(H_IPI, (uint64_t)cpu, XICP_PRIORITY);
514 }
515 
516 static void
517 xicp_mask(device_t dev, u_int irq, void *priv)
518 {
519 	struct xicp_softc *sc = device_get_softc(dev);
520 	cell_t status;
521 
522 	if (irq == MAX_XICP_IRQS)
523 		return;
524 
525 	if (rtas_exists()) {
526 		rtas_call_method(sc->ibm_int_off, 1, 1, irq, &status);
527 #ifdef POWERNV
528 	} else {
529 		struct xicp_intvec *ivec = priv;
530 
531 		KASSERT(ivec != NULL, ("Masking unconfigured interrupt"));
532 		opal_call(OPAL_SET_XIVE, irq, ivec->cpu << 2, 0xff);
533 #endif
534 	}
535 }
536 
537 static void
538 xicp_unmask(device_t dev, u_int irq, void *priv)
539 {
540 	struct xicp_softc *sc = device_get_softc(dev);
541 	cell_t status;
542 
543 	if (irq == MAX_XICP_IRQS)
544 		return;
545 
546 	if (rtas_exists()) {
547 		rtas_call_method(sc->ibm_int_on, 1, 1, irq, &status);
548 #ifdef POWERNV
549 	} else {
550 		struct xicp_intvec *ivec = priv;
551 
552 		KASSERT(ivec != NULL, ("Unmasking unconfigured interrupt"));
553 		opal_call(OPAL_SET_XIVE, irq, ivec->cpu << 2, XICP_PRIORITY);
554 #endif
555 	}
556 }
557 
558 #ifdef POWERNV
559 /* This is only used on POWER9 systems with the XIVE's XICS emulation. */
560 static void
561 xicp_smp_cpu_startup(void)
562 {
563 	struct xicp_softc *sc;
564 
565 	if (mfmsr() & PSL_HV) {
566 		sc = device_get_softc(root_pic);
567 
568 		if (sc->xics_emu)
569 			opal_call(OPAL_INT_SET_CPPR, 0xff);
570 	}
571 }
572 #endif
573