xref: /freebsd/sys/powerpc/pseries/xics.c (revision 22cf89c938886d14f5796fc49f9f020c23ea8eaf)
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 #include "opt_platform.h"
30 
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/module.h>
34 #include <sys/bus.h>
35 #include <sys/conf.h>
36 #include <sys/kernel.h>
37 #include <sys/lock.h>
38 #include <sys/malloc.h>
39 #include <sys/mutex.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 EARLY_DRIVER_MODULE(xicp, ofwbus, xicp_driver, 0, 0, BUS_PASS_INTERRUPT - 1);
157 EARLY_DRIVER_MODULE(xics, ofwbus, xics_driver, 0, 0, BUS_PASS_INTERRUPT);
158 
159 #ifdef POWERNV
160 static struct resource *
161 xicp_mem_for_cpu(int cpu)
162 {
163 	devclass_t dc;
164 	device_t dev;
165 	struct xicp_softc *sc;
166 	int i;
167 
168 	dc = devclass_find(xicp_driver.name);
169 	for (i = 0; (dev = devclass_get_device(dc, i)) != NULL; i++){
170 		sc = device_get_softc(dev);
171 		if (cpu >= sc->cpu_range[0] && cpu < sc->cpu_range[1])
172 			return (sc->mem[cpu - sc->cpu_range[0]]);
173 	}
174 
175 	return (NULL);
176 }
177 #endif
178 
179 static int
180 xicp_probe(device_t dev)
181 {
182 
183 	if (!ofw_bus_is_compatible(dev, "ibm,ppc-xicp") &&
184 	    !ofw_bus_is_compatible(dev, "ibm,opal-intc"))
185 		return (ENXIO);
186 
187 	device_set_desc(dev, "External Interrupt Presentation Controller");
188 	return (BUS_PROBE_GENERIC);
189 }
190 
191 static int
192 xics_probe(device_t dev)
193 {
194 
195 	if (!ofw_bus_is_compatible(dev, "ibm,ppc-xics") &&
196 	    !ofw_bus_is_compatible(dev, "IBM,opal-xics"))
197 		return (ENXIO);
198 
199 	device_set_desc(dev, "External Interrupt Source Controller");
200 	return (BUS_PROBE_GENERIC);
201 }
202 
203 static int
204 xicp_attach(device_t dev)
205 {
206 	struct xicp_softc *sc = device_get_softc(dev);
207 	phandle_t phandle = ofw_bus_get_node(dev);
208 
209 	if (rtas_exists()) {
210 		sc->ibm_int_on = rtas_token_lookup("ibm,int-on");
211 		sc->ibm_int_off = rtas_token_lookup("ibm,int-off");
212 		sc->ibm_set_xive = rtas_token_lookup("ibm,set-xive");
213 		sc->ibm_get_xive = rtas_token_lookup("ibm,get-xive");
214 #ifdef POWERNV
215 	} else if (opal_check() == 0) {
216 		/* No init needed */
217 #endif
218 	} else {
219 		device_printf(dev, "Cannot attach without RTAS or OPAL\n");
220 		return (ENXIO);
221 	}
222 
223 	if (OF_hasprop(phandle, "ibm,interrupt-server-ranges")) {
224 		OF_getencprop(phandle, "ibm,interrupt-server-ranges",
225 		    sc->cpu_range, sizeof(sc->cpu_range));
226 		sc->cpu_range[1] += sc->cpu_range[0];
227 		device_printf(dev, "Handling CPUs %d-%d\n", sc->cpu_range[0],
228 		    sc->cpu_range[1]-1);
229 #ifdef POWERNV
230 	} else if (ofw_bus_is_compatible(dev, "ibm,opal-intc")) {
231 			/*
232 			 * For now run POWER9 XIVE interrupt controller in XICS
233 			 * compatibility mode.
234 			 */
235 			sc->xics_emu = true;
236 			opal_call(OPAL_XIVE_RESET, OPAL_XIVE_XICS_MODE_EMU);
237 #endif
238 	} else {
239 		sc->cpu_range[0] = 0;
240 		sc->cpu_range[1] = mp_ncpus;
241 	}
242 
243 #ifdef POWERNV
244 	if (mfmsr() & PSL_HV) {
245 		int i;
246 
247 		if (sc->xics_emu) {
248 			opal_call(OPAL_INT_SET_CPPR, 0xff);
249 			for (i = 0; i < mp_ncpus; i++) {
250 				opal_call(OPAL_INT_SET_MFRR,
251 				    pcpu_find(i)->pc_hwref, 0xff);
252 			}
253 		} else {
254 			for (i = 0; i < sc->cpu_range[1] - sc->cpu_range[0]; i++) {
255 				sc->mem[i] = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
256 				    &i, RF_ACTIVE);
257 				if (sc->mem[i] == NULL) {
258 					device_printf(dev, "Could not alloc mem "
259 					    "resource %d\n", i);
260 					return (ENXIO);
261 				}
262 
263 				/* Unmask interrupts on all cores */
264 				bus_write_1(sc->mem[i], 4, 0xff);
265 				bus_write_1(sc->mem[i], 12, 0xff);
266 			}
267 		}
268 	}
269 #endif
270 
271 	mtx_init(&sc->sc_mtx, "XICP", NULL, MTX_DEF);
272 	sc->nintvecs = 0;
273 
274 	powerpc_register_pic(dev, OF_xref_from_node(phandle), MAX_XICP_IRQS,
275 	    1 /* Number of IPIs */, FALSE);
276 	root_pic = dev;
277 
278 #ifdef POWERNV
279 	if (sc->xics_emu)
280 		powernv_smp_ap_extra_init = xicp_smp_cpu_startup;
281 #endif
282 
283 	return (0);
284 }
285 
286 static int
287 xics_attach(device_t dev)
288 {
289 	phandle_t phandle = ofw_bus_get_node(dev);
290 
291 	/* The XICP (root PIC) will handle all our interrupts */
292 	powerpc_register_pic(root_pic, OF_xref_from_node(phandle),
293 	    MAX_XICP_IRQS, 1 /* Number of IPIs */, FALSE);
294 
295 	return (0);
296 }
297 
298 static __inline struct xicp_intvec *
299 xicp_setup_priv(struct xicp_softc *sc, u_int irq, void **priv)
300 {
301 	if (*priv == NULL) {
302 		KASSERT(sc->nintvecs + 1 < nitems(sc->intvecs),
303 			("Too many XICP interrupts"));
304 		mtx_lock(&sc->sc_mtx);
305 		*priv = &sc->intvecs[sc->nintvecs++];
306 		mtx_unlock(&sc->sc_mtx);
307 	}
308 
309 	return (*priv);
310 }
311 
312 /*
313  * PIC I/F methods.
314  */
315 
316 static void
317 xicp_bind(device_t dev, u_int irq, cpuset_t cpumask, void **priv)
318 {
319 	struct xicp_softc *sc = device_get_softc(dev);
320 	struct xicp_intvec *iv;
321 	cell_t status, cpu;
322 	int ncpus, i, error = -1;
323 
324 	/* Ignore IPIs */
325 	if (irq == MAX_XICP_IRQS)
326 		return;
327 
328 	iv = xicp_setup_priv(sc, irq, priv);
329 
330 	/*
331 	 * This doesn't appear to actually support affinity groups, so pick a
332 	 * random CPU.
333 	 */
334 	ncpus = 0;
335 	CPU_FOREACH(cpu)
336 		if (CPU_ISSET(cpu, &cpumask)) ncpus++;
337 
338 	i = mftb() % ncpus;
339 	ncpus = 0;
340 	CPU_FOREACH(cpu) {
341 		if (!CPU_ISSET(cpu, &cpumask))
342 			continue;
343 		if (ncpus == i)
344 			break;
345 		ncpus++;
346 	}
347 
348 	cpu = pcpu_find(cpu)->pc_hwref;
349 	iv->cpu = cpu;
350 
351 	if (rtas_exists())
352 		error = rtas_call_method(sc->ibm_set_xive, 3, 1, irq, cpu,
353 		    XICP_PRIORITY, &status);
354 #ifdef POWERNV
355 	else
356 		error = opal_call(OPAL_SET_XIVE, irq, cpu << 2, XICP_PRIORITY);
357 #endif
358 
359 	if (error < 0)
360 		panic("Cannot bind interrupt %d to CPU %d", irq, cpu);
361 }
362 
363 static void
364 xicp_dispatch(device_t dev, struct trapframe *tf)
365 {
366 	struct xicp_softc *sc;
367 	struct resource *regs = NULL;
368 	uint64_t xirr, junk;
369 	int i;
370 
371 	sc = device_get_softc(dev);
372 #ifdef POWERNV
373 	if ((mfmsr() & PSL_HV) && !sc->xics_emu) {
374 		regs = xicp_mem_for_cpu(PCPU_GET(hwref));
375 		KASSERT(regs != NULL,
376 		    ("Can't find regs for CPU %ld", (uintptr_t)PCPU_GET(hwref)));
377 	}
378 #endif
379 
380 	for (;;) {
381 		/* Return value in R4, use the PFT call */
382 		if (regs) {
383 			xirr = bus_read_4(regs, 4);
384 #ifdef POWERNV
385 		} else if (sc->xics_emu) {
386 			opal_call(OPAL_INT_GET_XIRR, &cpu_xirr[PCPU_GET(cpuid)],
387 			    false);
388 			xirr = cpu_xirr[PCPU_GET(cpuid)];
389 #endif
390 		} else {
391 			/* Return value in R4, use the PFT call */
392 			phyp_pft_hcall(H_XIRR, 0, 0, 0, 0, &xirr, &junk, &junk);
393 		}
394 		xirr &= 0x00ffffff;
395 
396 		if (xirr == 0) /* No more pending interrupts? */
397 			break;
398 
399 		if (xirr == XICP_IPI) {		/* Magic number for IPIs */
400 			xirr = MAX_XICP_IRQS;	/* Map to FreeBSD magic */
401 
402 			/* Clear IPI */
403 			if (regs)
404 				bus_write_1(regs, 12, 0xff);
405 #ifdef POWERNV
406 			else if (sc->xics_emu)
407 				opal_call(OPAL_INT_SET_MFRR,
408 				    PCPU_GET(hwref), 0xff);
409 #endif
410 			else
411 				phyp_hcall(H_IPI, (uint64_t)(PCPU_GET(hwref)),
412 				    0xff);
413 			i = sc->ipi_vec;
414 		} else {
415 			/* XXX: super inefficient */
416 			for (i = 0; i < sc->nintvecs; i++) {
417 				if (sc->intvecs[i].irq == xirr)
418 					break;
419 			}
420 			KASSERT(i < sc->nintvecs, ("Unmapped XIRR"));
421 		}
422 
423 		powerpc_dispatch_intr(sc->intvecs[i].vector, tf);
424 	}
425 }
426 
427 static void
428 xicp_enable(device_t dev, u_int irq, u_int vector, void **priv)
429 {
430 	struct xicp_softc *sc;
431 	struct xicp_intvec *intr;
432 	cell_t status, cpu;
433 
434 	sc = device_get_softc(dev);
435 
436 	/* Bind to this CPU to start: distrib. ID is last entry in gserver# */
437 	cpu = PCPU_GET(hwref);
438 
439 	intr = xicp_setup_priv(sc, irq, priv);
440 
441 	intr->irq = irq;
442 	intr->vector = vector;
443 	intr->cpu = cpu;
444 	mb();
445 
446 	/* IPIs are also enabled.  Stash off the vector index */
447 	if (irq == MAX_XICP_IRQS) {
448 		sc->ipi_vec = intr - sc->intvecs;
449 		return;
450 	}
451 
452 	if (rtas_exists()) {
453 		rtas_call_method(sc->ibm_set_xive, 3, 1, irq, cpu,
454 		    XICP_PRIORITY, &status);
455 		xicp_unmask(dev, irq, intr);
456 #ifdef POWERNV
457 	} else {
458 		status = opal_call(OPAL_SET_XIVE, irq, cpu << 2, XICP_PRIORITY);
459 		/* Unmask implicit for OPAL */
460 
461 		if (status != 0)
462 			panic("OPAL_SET_XIVE IRQ %d -> cpu %d failed: %d", irq,
463 			    cpu, status);
464 #endif
465 	}
466 }
467 
468 static void
469 xicp_eoi(device_t dev, u_int irq, void *priv)
470 {
471 #ifdef POWERNV
472 	struct xicp_softc *sc;
473 #endif
474 	uint64_t xirr;
475 
476 	if (irq == MAX_XICP_IRQS) /* Remap IPI interrupt to internal value */
477 		irq = XICP_IPI;
478 	xirr = irq | (0xff << 24);
479 
480 #ifdef POWERNV
481 	if (mfmsr() & PSL_HV) {
482 		sc = device_get_softc(dev);
483 		if (sc->xics_emu)
484 			opal_call(OPAL_INT_EOI, xirr);
485 		else
486 			bus_write_4(xicp_mem_for_cpu(PCPU_GET(hwref)), 4, xirr);
487 	} else
488 #endif
489 		phyp_hcall(H_EOI, xirr);
490 }
491 
492 static void
493 xicp_ipi(device_t dev, u_int cpu)
494 {
495 
496 #ifdef POWERNV
497 	struct xicp_softc *sc;
498 	cpu = pcpu_find(cpu)->pc_hwref;
499 
500 	if (mfmsr() & PSL_HV) {
501 		sc = device_get_softc(dev);
502 		if (sc->xics_emu) {
503 			int64_t rv;
504 			rv = opal_call(OPAL_INT_SET_MFRR, cpu, XICP_PRIORITY);
505 			if (rv != 0)
506 			    device_printf(dev, "IPI SET_MFRR result: %ld\n", rv);
507 		} else
508 			bus_write_1(xicp_mem_for_cpu(cpu), 12, XICP_PRIORITY);
509 	} else
510 #endif
511 		phyp_hcall(H_IPI, (uint64_t)cpu, XICP_PRIORITY);
512 }
513 
514 static void
515 xicp_mask(device_t dev, u_int irq, void *priv)
516 {
517 	struct xicp_softc *sc = device_get_softc(dev);
518 	cell_t status;
519 
520 	if (irq == MAX_XICP_IRQS)
521 		return;
522 
523 	if (rtas_exists()) {
524 		rtas_call_method(sc->ibm_int_off, 1, 1, irq, &status);
525 #ifdef POWERNV
526 	} else {
527 		struct xicp_intvec *ivec = priv;
528 
529 		KASSERT(ivec != NULL, ("Masking unconfigured interrupt"));
530 		opal_call(OPAL_SET_XIVE, irq, ivec->cpu << 2, 0xff);
531 #endif
532 	}
533 }
534 
535 static void
536 xicp_unmask(device_t dev, u_int irq, void *priv)
537 {
538 	struct xicp_softc *sc = device_get_softc(dev);
539 	cell_t status;
540 
541 	if (irq == MAX_XICP_IRQS)
542 		return;
543 
544 	if (rtas_exists()) {
545 		rtas_call_method(sc->ibm_int_on, 1, 1, irq, &status);
546 #ifdef POWERNV
547 	} else {
548 		struct xicp_intvec *ivec = priv;
549 
550 		KASSERT(ivec != NULL, ("Unmasking unconfigured interrupt"));
551 		opal_call(OPAL_SET_XIVE, irq, ivec->cpu << 2, XICP_PRIORITY);
552 #endif
553 	}
554 }
555 
556 #ifdef POWERNV
557 /* This is only used on POWER9 systems with the XIVE's XICS emulation. */
558 static void
559 xicp_smp_cpu_startup(void)
560 {
561 	struct xicp_softc *sc;
562 
563 	if (mfmsr() & PSL_HV) {
564 		sc = device_get_softc(root_pic);
565 
566 		if (sc->xics_emu)
567 			opal_call(OPAL_INT_SET_CPPR, 0xff);
568 	}
569 }
570 #endif
571