xref: /freebsd/sys/powerpc/powernv/xive.c (revision 3ff01b231dfa83d518854c63e7c9cd1debd1139e)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright 2019 Justin Hibbits
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/endian.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/malloc.h>
42 #include <sys/mutex.h>
43 #include <sys/smp.h>
44 
45 #include <vm/vm.h>
46 #include <vm/pmap.h>
47 
48 #include <machine/bus.h>
49 #include <machine/intr_machdep.h>
50 #include <machine/md_var.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 "pic_if.h"
60 
61 #define XIVE_PRIORITY	7	/* Random non-zero number */
62 #define MAX_XIVE_IRQS	(1<<24)	/* 24-bit XIRR field */
63 
64 /* Registers */
65 #define	XIVE_TM_QW1_OS		0x010	/* Guest OS registers */
66 #define	XIVE_TM_QW2_HV_POOL	0x020	/* Hypervisor pool registers */
67 #define	XIVE_TM_QW3_HV		0x030	/* Hypervisor registers */
68 
69 #define	XIVE_TM_NSR	0x00
70 #define	XIVE_TM_CPPR	0x01
71 #define	XIVE_TM_IPB	0x02
72 #define	XIVE_TM_LSMFB	0x03
73 #define	XIVE_TM_ACK_CNT	0x04
74 #define	XIVE_TM_INC	0x05
75 #define	XIVE_TM_AGE	0x06
76 #define	XIVE_TM_PIPR	0x07
77 
78 #define	TM_WORD0	0x0
79 #define	TM_WORD2	0x8
80 #define	  TM_QW2W2_VP	  0x80000000
81 
82 #define	XIVE_TM_SPC_ACK			0x800
83 #define	  TM_QW3NSR_HE_SHIFT		  14
84 #define	  TM_QW3_NSR_HE_NONE		  0
85 #define	  TM_QW3_NSR_HE_POOL		  1
86 #define	  TM_QW3_NSR_HE_PHYS		  2
87 #define	  TM_QW3_NSR_HE_LSI		  3
88 #define	XIVE_TM_SPC_PULL_POOL_CTX	0x828
89 
90 #define	XIVE_IRQ_LOAD_EOI	0x000
91 #define	XIVE_IRQ_STORE_EOI	0x400
92 #define	XIVE_IRQ_PQ_00		0xc00
93 #define	XIVE_IRQ_PQ_01		0xd00
94 
95 #define	XIVE_IRQ_VAL_P		0x02
96 #define	XIVE_IRQ_VAL_Q		0x01
97 
98 struct xive_softc;
99 struct xive_irq;
100 
101 extern void (*powernv_smp_ap_extra_init)(void);
102 
103 /* Private support */
104 static void	xive_setup_cpu(void);
105 static void	xive_smp_cpu_startup(void);
106 static void	xive_init_irq(struct xive_irq *irqd, u_int irq);
107 static struct xive_irq	*xive_configure_irq(u_int irq);
108 static int	xive_provision_page(struct xive_softc *sc);
109 
110 /* Interfaces */
111 static int	xive_probe(device_t);
112 static int	xive_attach(device_t);
113 static int	xics_probe(device_t);
114 static int	xics_attach(device_t);
115 
116 static void	xive_bind(device_t, u_int, cpuset_t, void **);
117 static void	xive_dispatch(device_t, struct trapframe *);
118 static void	xive_enable(device_t, u_int, u_int, void **);
119 static void	xive_eoi(device_t, u_int, void *);
120 static void	xive_ipi(device_t, u_int);
121 static void	xive_mask(device_t, u_int, void *);
122 static void	xive_unmask(device_t, u_int, void *);
123 static void	xive_translate_code(device_t dev, u_int irq, int code,
124 		    enum intr_trigger *trig, enum intr_polarity *pol);
125 
126 static device_method_t  xive_methods[] = {
127 	/* Device interface */
128 	DEVMETHOD(device_probe,		xive_probe),
129 	DEVMETHOD(device_attach,	xive_attach),
130 
131 	/* PIC interface */
132 	DEVMETHOD(pic_bind,		xive_bind),
133 	DEVMETHOD(pic_dispatch,		xive_dispatch),
134 	DEVMETHOD(pic_enable,		xive_enable),
135 	DEVMETHOD(pic_eoi,		xive_eoi),
136 	DEVMETHOD(pic_ipi,		xive_ipi),
137 	DEVMETHOD(pic_mask,		xive_mask),
138 	DEVMETHOD(pic_unmask,		xive_unmask),
139 	DEVMETHOD(pic_translate_code,	xive_translate_code),
140 
141 	DEVMETHOD_END
142 };
143 
144 static device_method_t  xics_methods[] = {
145 	/* Device interface */
146 	DEVMETHOD(device_probe,		xics_probe),
147 	DEVMETHOD(device_attach,	xics_attach),
148 
149 	DEVMETHOD_END
150 };
151 
152 struct xive_softc {
153 	struct mtx sc_mtx;
154 	struct resource *sc_mem;
155 	vm_size_t	sc_prov_page_size;
156 	uint32_t	sc_offset;
157 };
158 
159 struct xive_queue {
160 	uint32_t	*q_page;
161 	uint32_t	*q_eoi_page;
162 	uint32_t	 q_toggle;
163 	uint32_t	 q_size;
164 	uint32_t	 q_index;
165 	uint32_t	 q_mask;
166 };
167 
168 struct xive_irq {
169 	uint32_t	girq;
170 	uint32_t	lirq;
171 	uint64_t	vp;
172 	uint64_t	flags;
173 #define	OPAL_XIVE_IRQ_SHIFT_BUG		0x00000008
174 #define	OPAL_XIVE_IRQ_LSI		0x00000004
175 #define	OPAL_XIVE_IRQ_STORE_EOI		0x00000002
176 #define	OPAL_XIVE_IRQ_TRIGGER_PAGE	0x00000001
177 	uint8_t	prio;
178 	vm_offset_t	eoi_page;
179 	vm_offset_t	trig_page;
180 	vm_size_t	esb_size;
181 	int		chip;
182 };
183 
184 struct xive_cpu {
185 	uint64_t	vp;
186 	uint64_t	flags;
187 	struct xive_irq	ipi_data;
188 	struct xive_queue	queue; /* We only use a single queue for now. */
189 	uint64_t	cam;
190 	uint32_t	chip;
191 };
192 
193 static driver_t xive_driver = {
194 	"xive",
195 	xive_methods,
196 	sizeof(struct xive_softc)
197 };
198 
199 static driver_t xics_driver = {
200 	"xivevc",
201 	xics_methods,
202 	0
203 };
204 
205 static devclass_t xive_devclass;
206 static devclass_t xics_devclass;
207 
208 EARLY_DRIVER_MODULE(xive, ofwbus, xive_driver, xive_devclass, 0, 0,
209     BUS_PASS_INTERRUPT-1);
210 EARLY_DRIVER_MODULE(xivevc, ofwbus, xics_driver, xics_devclass, 0, 0,
211     BUS_PASS_INTERRUPT);
212 
213 MALLOC_DEFINE(M_XIVE, "xive", "XIVE Memory");
214 
215 DPCPU_DEFINE_STATIC(struct xive_cpu, xive_cpu_data);
216 
217 static int xive_ipi_vector = -1;
218 
219 /*
220  * XIVE Exploitation mode driver.
221  *
222  * The XIVE, present in the POWER9 CPU, can run in two modes: XICS emulation
223  * mode, and "Exploitation mode".  XICS emulation mode is compatible with the
224  * POWER8 and earlier XICS interrupt controller, using OPAL calls to emulate
225  * hypervisor calls and memory accesses.  Exploitation mode gives us raw access
226  * to the XIVE MMIO, improving performance significantly.
227  *
228  * The XIVE controller is a very bizarre interrupt controller.  It uses queues
229  * in memory to pass interrupts around, and maps itself into 512GB of physical
230  * device address space, giving each interrupt in the system one or more pages
231  * of address space.  An IRQ is tied to a virtual processor, which could be a
232  * physical CPU thread, or a guest CPU thread (LPAR running on a physical
233  * thread).  Thus, the controller can route interrupts directly to guest OSes
234  * bypassing processing by the hypervisor, thereby improving performance of the
235  * guest OS.
236  *
237  * An IRQ, in addition to being tied to a virtual processor, has one or two
238  * page mappings: an EOI page, and an optional trigger page.  The trigger page
239  * could be the same as the EOI page.  Level-sensitive interrupts (LSIs) don't
240  * have a trigger page, as they're external interrupts controlled by physical
241  * lines.  MSIs and IPIs have trigger pages.  An IPI is really just another IRQ
242  * in the XIVE, which is triggered by software.
243  *
244  * An interesting behavior of the XIVE controller is that oftentimes the
245  * contents of an address location don't actually matter, but the direction of
246  * the action is the signifier (read vs write), and the address is significant.
247  * Hence, masking and unmasking an interrupt is done by reading different
248  * addresses in the EOI page, and triggering an interrupt consists of writing to
249  * the trigger page.
250  *
251  * Additionally, the MMIO region mapped is CPU-sensitive, just like the
252  * per-processor register space (private access) in OpenPIC.  In order for a CPU
253  * to receive interrupts it must itself configure its CPPR (Current Processor
254  * Priority Register), it cannot be set by any other processor.  This
255  * necessitates the xive_smp_cpu_startup() function.
256  *
257  * Queues are pages of memory, sized powers-of-two, that are shared with the
258  * XIVE.  The XIVE writes into the queue with an alternating polarity bit, which
259  * flips when the queue wraps.
260  */
261 
262 /*
263  * Offset-based read/write interfaces.
264  */
265 static uint16_t
266 xive_read_2(struct xive_softc *sc, bus_size_t offset)
267 {
268 
269 	return (bus_read_2(sc->sc_mem, sc->sc_offset + offset));
270 }
271 
272 static void
273 xive_write_1(struct xive_softc *sc, bus_size_t offset, uint8_t val)
274 {
275 
276 	bus_write_1(sc->sc_mem, sc->sc_offset + offset, val);
277 }
278 
279 /* EOI and Trigger page access interfaces. */
280 static uint64_t
281 xive_read_mmap8(vm_offset_t addr)
282 {
283 	return (*(volatile uint64_t *)addr);
284 }
285 
286 static void
287 xive_write_mmap8(vm_offset_t addr, uint64_t val)
288 {
289 	*(uint64_t *)(addr) = val;
290 }
291 
292 /* Device interfaces. */
293 static int
294 xive_probe(device_t dev)
295 {
296 
297 	if (!ofw_bus_is_compatible(dev, "ibm,opal-xive-pe"))
298 		return (ENXIO);
299 
300 	device_set_desc(dev, "External Interrupt Virtualization Engine");
301 
302 	/* Make sure we always win against the xicp driver. */
303 	return (BUS_PROBE_DEFAULT);
304 }
305 
306 static int
307 xics_probe(device_t dev)
308 {
309 
310 	if (!ofw_bus_is_compatible(dev, "ibm,opal-xive-vc"))
311 		return (ENXIO);
312 
313 	device_set_desc(dev, "External Interrupt Virtualization Engine Root");
314 	return (BUS_PROBE_DEFAULT);
315 }
316 
317 static int
318 xive_attach(device_t dev)
319 {
320 	struct xive_softc *sc = device_get_softc(dev);
321 	struct xive_cpu *xive_cpud;
322 	phandle_t phandle = ofw_bus_get_node(dev);
323 	int64_t vp_block;
324 	int error;
325 	int rid;
326 	int i, order;
327 	uint64_t vp_id;
328 	int64_t ipi_irq;
329 
330 	opal_call(OPAL_XIVE_RESET, OPAL_XIVE_XICS_MODE_EXP);
331 
332 	error = OF_getencprop(phandle, "ibm,xive-provision-page-size",
333 	    (pcell_t *)&sc->sc_prov_page_size, sizeof(sc->sc_prov_page_size));
334 
335 	rid = 1;	/* Get the Hypervisor-level register set. */
336 	sc->sc_mem = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
337 	    &rid, RF_ACTIVE);
338 	sc->sc_offset = XIVE_TM_QW3_HV;
339 
340 	mtx_init(&sc->sc_mtx, "XIVE", NULL, MTX_DEF);
341 
342 	/* Workaround for qemu single-thread powernv */
343 	if (mp_maxid == 0)
344 		order = 1;
345 	else
346 		order = fls(mp_maxid + (mp_maxid - 1)) - 1;
347 
348 	do {
349 		vp_block = opal_call(OPAL_XIVE_ALLOCATE_VP_BLOCK, order);
350 		if (vp_block == OPAL_BUSY)
351 			DELAY(10);
352 		else if (vp_block == OPAL_XIVE_PROVISIONING)
353 			xive_provision_page(sc);
354 		else
355 			break;
356 	} while (1);
357 
358 	if (vp_block < 0) {
359 		device_printf(dev,
360 		    "Unable to allocate VP block.  Opal error %d\n",
361 		    (int)vp_block);
362 		bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->sc_mem);
363 		return (ENXIO);
364 	}
365 
366 	/*
367 	 * Set up the VPs.  Try to do as much as we can in attach, to lessen
368 	 * what's needed at AP spawn time.
369 	 */
370 	CPU_FOREACH(i) {
371 		vp_id = pcpu_find(i)->pc_hwref;
372 
373 		xive_cpud = DPCPU_ID_PTR(i, xive_cpu_data);
374 		xive_cpud->vp = vp_id + vp_block;
375 		opal_call(OPAL_XIVE_GET_VP_INFO, xive_cpud->vp, NULL,
376 		    vtophys(&xive_cpud->cam), NULL, vtophys(&xive_cpud->chip));
377 
378 		xive_cpud->cam = be64toh(xive_cpud->cam);
379 		xive_cpud->chip = be64toh(xive_cpud->chip);
380 
381 		/* Allocate the queue page and populate the queue state data. */
382 		xive_cpud->queue.q_page = contigmalloc(PAGE_SIZE, M_XIVE,
383 		    M_ZERO | M_WAITOK, 0, BUS_SPACE_MAXADDR, PAGE_SIZE, 0);
384 		xive_cpud->queue.q_size = 1 << PAGE_SHIFT;
385 		xive_cpud->queue.q_mask =
386 		    ((xive_cpud->queue.q_size / sizeof(int)) - 1);
387 		xive_cpud->queue.q_toggle = 0;
388 		xive_cpud->queue.q_index = 0;
389 		do {
390 			error = opal_call(OPAL_XIVE_SET_VP_INFO, xive_cpud->vp,
391 			    OPAL_XIVE_VP_ENABLED, 0);
392 		} while (error == OPAL_BUSY);
393 		error = opal_call(OPAL_XIVE_SET_QUEUE_INFO, vp_id,
394 		    XIVE_PRIORITY, vtophys(xive_cpud->queue.q_page), PAGE_SHIFT,
395 		    OPAL_XIVE_EQ_ALWAYS_NOTIFY | OPAL_XIVE_EQ_ENABLED);
396 
397 		do {
398 			ipi_irq = opal_call(OPAL_XIVE_ALLOCATE_IRQ,
399 			    xive_cpud->chip);
400 		} while (ipi_irq == OPAL_BUSY);
401 
402 		if (ipi_irq < 0)
403 			device_printf(root_pic,
404 			    "Failed allocating IPI.  OPAL error %d\n",
405 			    (int)ipi_irq);
406 		else {
407 			xive_init_irq(&xive_cpud->ipi_data, ipi_irq);
408 			xive_cpud->ipi_data.vp = vp_id;
409 			xive_cpud->ipi_data.lirq = MAX_XIVE_IRQS;
410 			opal_call(OPAL_XIVE_SET_IRQ_CONFIG, ipi_irq,
411 			    xive_cpud->ipi_data.vp, XIVE_PRIORITY,
412 			    MAX_XIVE_IRQS);
413 		}
414 	}
415 
416 	powerpc_register_pic(dev, OF_xref_from_node(phandle), MAX_XIVE_IRQS,
417 	    1 /* Number of IPIs */, FALSE);
418 	root_pic = dev;
419 
420 	xive_setup_cpu();
421 	powernv_smp_ap_extra_init = xive_smp_cpu_startup;
422 
423 	return (0);
424 }
425 
426 static int
427 xics_attach(device_t dev)
428 {
429 	phandle_t phandle = ofw_bus_get_node(dev);
430 
431 	/* The XIVE (root PIC) will handle all our interrupts */
432 	powerpc_register_pic(root_pic, OF_xref_from_node(phandle),
433 	    MAX_XIVE_IRQS, 1 /* Number of IPIs */, FALSE);
434 
435 	return (0);
436 }
437 
438 /*
439  * PIC I/F methods.
440  */
441 
442 static void
443 xive_bind(device_t dev, u_int irq, cpuset_t cpumask, void **priv)
444 {
445 	struct xive_irq *irqd;
446 	int cpu;
447 	int ncpus, i, error;
448 
449 	if (*priv == NULL)
450 		*priv = xive_configure_irq(irq);
451 
452 	irqd = *priv;
453 
454 	/*
455 	 * This doesn't appear to actually support affinity groups, so pick a
456 	 * random CPU.
457 	 */
458 	ncpus = 0;
459 	CPU_FOREACH(cpu)
460 		if (CPU_ISSET(cpu, &cpumask)) ncpus++;
461 
462 	i = mftb() % ncpus;
463 	ncpus = 0;
464 	CPU_FOREACH(cpu) {
465 		if (!CPU_ISSET(cpu, &cpumask))
466 			continue;
467 		if (ncpus == i)
468 			break;
469 		ncpus++;
470 	}
471 
472 	opal_call(OPAL_XIVE_SYNC, OPAL_XIVE_SYNC_QUEUE, irq);
473 
474 	irqd->vp = pcpu_find(cpu)->pc_hwref;
475 	error = opal_call(OPAL_XIVE_SET_IRQ_CONFIG, irq, irqd->vp,
476 	    XIVE_PRIORITY, irqd->lirq);
477 
478 	if (error < 0)
479 		panic("Cannot bind interrupt %d to CPU %d", irq, cpu);
480 
481 	xive_eoi(dev, irq, irqd);
482 }
483 
484 /* Read the next entry in the queue page and update the index. */
485 static int
486 xive_read_eq(struct xive_queue *q)
487 {
488 	uint32_t i = be32toh(q->q_page[q->q_index]);
489 
490 	/* Check validity, using current queue polarity. */
491 	if ((i >> 31) == q->q_toggle)
492 		return (0);
493 
494 	q->q_index = (q->q_index + 1) & q->q_mask;
495 
496 	if (q->q_index == 0)
497 		q->q_toggle ^= 1;
498 
499 	return (i & 0x7fffffff);
500 }
501 
502 static void
503 xive_dispatch(device_t dev, struct trapframe *tf)
504 {
505 	struct xive_softc *sc;
506 	struct xive_cpu *xive_cpud;
507 	uint32_t vector;
508 	uint16_t ack;
509 	uint8_t cppr, he;
510 
511 	sc = device_get_softc(dev);
512 
513 	xive_cpud = DPCPU_PTR(xive_cpu_data);
514 	for (;;) {
515 		ack = xive_read_2(sc, XIVE_TM_SPC_ACK);
516 		cppr = (ack & 0xff);
517 
518 		he = ack >> TM_QW3NSR_HE_SHIFT;
519 
520 		if (he == TM_QW3_NSR_HE_NONE)
521 			break;
522 
523 		else if (__predict_false(he != TM_QW3_NSR_HE_PHYS)) {
524 			/*
525 			 * We don't support TM_QW3_NSR_HE_POOL or
526 			 * TM_QW3_NSR_HE_LSI interrupts.
527 			 */
528 			device_printf(dev,
529 			    "Unexpected interrupt he type: %d\n", he);
530 			goto end;
531 		}
532 
533 		xive_write_1(sc, XIVE_TM_CPPR, cppr);
534 
535 		for (;;) {
536 			vector = xive_read_eq(&xive_cpud->queue);
537 
538 			if (vector == 0)
539 				break;
540 
541 			if (vector == MAX_XIVE_IRQS)
542 				vector = xive_ipi_vector;
543 
544 			powerpc_dispatch_intr(vector, tf);
545 		}
546 	}
547 end:
548 	xive_write_1(sc, XIVE_TM_CPPR, 0xff);
549 }
550 
551 static void
552 xive_enable(device_t dev, u_int irq, u_int vector, void **priv)
553 {
554 	struct xive_irq *irqd;
555 	cell_t status, cpu;
556 
557 	if (irq == MAX_XIVE_IRQS) {
558 		if (xive_ipi_vector == -1)
559 			xive_ipi_vector = vector;
560 		return;
561 	}
562 	if (*priv == NULL)
563 		*priv = xive_configure_irq(irq);
564 
565 	irqd = *priv;
566 
567 	/* Bind to this CPU to start */
568 	cpu = PCPU_GET(hwref);
569 	irqd->lirq = vector;
570 
571 	for (;;) {
572 		status = opal_call(OPAL_XIVE_SET_IRQ_CONFIG, irq, cpu,
573 		    XIVE_PRIORITY, vector);
574 		if (status != OPAL_BUSY)
575 			break;
576 		DELAY(10);
577 	}
578 
579 	if (status != 0)
580 		panic("OPAL_SET_XIVE IRQ %d -> cpu %d failed: %d", irq,
581 		    cpu, status);
582 
583 	xive_unmask(dev, irq, *priv);
584 }
585 
586 static void
587 xive_eoi(device_t dev, u_int irq, void *priv)
588 {
589 	struct xive_irq *rirq;
590 	struct xive_cpu *cpud;
591 	uint8_t eoi_val;
592 
593 	if (irq == MAX_XIVE_IRQS) {
594 		cpud = DPCPU_PTR(xive_cpu_data);
595 		rirq = &cpud->ipi_data;
596 	} else
597 		rirq = priv;
598 
599 	if (rirq->flags & OPAL_XIVE_IRQ_STORE_EOI)
600 		xive_write_mmap8(rirq->eoi_page + XIVE_IRQ_STORE_EOI, 0);
601 	else if (rirq->flags & OPAL_XIVE_IRQ_LSI)
602 		xive_read_mmap8(rirq->eoi_page + XIVE_IRQ_LOAD_EOI);
603 	else {
604 		eoi_val = xive_read_mmap8(rirq->eoi_page + XIVE_IRQ_PQ_00);
605 		if ((eoi_val & XIVE_IRQ_VAL_Q) && rirq->trig_page != 0)
606 			xive_write_mmap8(rirq->trig_page, 0);
607 	}
608 }
609 
610 static void
611 xive_ipi(device_t dev, u_int cpu)
612 {
613 	struct xive_cpu *xive_cpud;
614 
615 	xive_cpud = DPCPU_ID_PTR(cpu, xive_cpu_data);
616 
617 	if (xive_cpud->ipi_data.trig_page == 0)
618 		return;
619 	xive_write_mmap8(xive_cpud->ipi_data.trig_page, 0);
620 }
621 
622 static void
623 xive_mask(device_t dev, u_int irq, void *priv)
624 {
625 	struct xive_irq *rirq;
626 
627 	/* Never mask IPIs */
628 	if (irq == MAX_XIVE_IRQS)
629 		return;
630 
631 	rirq = priv;
632 
633 	if (!(rirq->flags & OPAL_XIVE_IRQ_LSI))
634 		return;
635 	xive_read_mmap8(rirq->eoi_page + XIVE_IRQ_PQ_01);
636 }
637 
638 static void
639 xive_unmask(device_t dev, u_int irq, void *priv)
640 {
641 	struct xive_irq *rirq;
642 
643 	rirq = priv;
644 
645 	xive_read_mmap8(rirq->eoi_page + XIVE_IRQ_PQ_00);
646 }
647 
648 static void
649 xive_translate_code(device_t dev, u_int irq, int code,
650     enum intr_trigger *trig, enum intr_polarity *pol)
651 {
652 	switch (code) {
653 	case 0:
654 		/* L to H edge */
655 		*trig = INTR_TRIGGER_EDGE;
656 		*pol = INTR_POLARITY_HIGH;
657 		break;
658 	case 1:
659 		/* Active L level */
660 		*trig = INTR_TRIGGER_LEVEL;
661 		*pol = INTR_POLARITY_LOW;
662 		break;
663 	default:
664 		*trig = INTR_TRIGGER_CONFORM;
665 		*pol = INTR_POLARITY_CONFORM;
666 	}
667 }
668 
669 /* Private functions. */
670 /*
671  * Setup the current CPU.  Called by the BSP at driver attachment, and by each
672  * AP at wakeup (via xive_smp_cpu_startup()).
673  */
674 static void
675 xive_setup_cpu(void)
676 {
677 	struct xive_softc *sc;
678 	struct xive_cpu *cpup;
679 	uint32_t val;
680 
681 	cpup = DPCPU_PTR(xive_cpu_data);
682 
683 	sc = device_get_softc(root_pic);
684 
685 	val = bus_read_4(sc->sc_mem, XIVE_TM_QW2_HV_POOL + TM_WORD2);
686 	if (val & TM_QW2W2_VP)
687 		bus_read_8(sc->sc_mem, XIVE_TM_SPC_PULL_POOL_CTX);
688 
689 	bus_write_4(sc->sc_mem, XIVE_TM_QW2_HV_POOL + TM_WORD0, 0xff);
690 	bus_write_4(sc->sc_mem, XIVE_TM_QW2_HV_POOL + TM_WORD2,
691 	    TM_QW2W2_VP | cpup->cam);
692 
693 	xive_unmask(root_pic, cpup->ipi_data.girq, &cpup->ipi_data);
694 	xive_write_1(sc, XIVE_TM_CPPR, 0xff);
695 }
696 
697 /* Populate an IRQ structure, mapping the EOI and trigger pages. */
698 static void
699 xive_init_irq(struct xive_irq *irqd, u_int irq)
700 {
701 	uint64_t eoi_phys, trig_phys;
702 	uint32_t esb_shift;
703 
704 	opal_call(OPAL_XIVE_GET_IRQ_INFO, irq,
705 	    vtophys(&irqd->flags), vtophys(&eoi_phys),
706 	    vtophys(&trig_phys), vtophys(&esb_shift),
707 	    vtophys(&irqd->chip));
708 
709 	irqd->flags = be64toh(irqd->flags);
710 	eoi_phys = be64toh(eoi_phys);
711 	trig_phys = be64toh(trig_phys);
712 	esb_shift = be32toh(esb_shift);
713 	irqd->chip = be32toh(irqd->chip);
714 
715 	irqd->girq = irq;
716 	irqd->esb_size = 1 << esb_shift;
717 	irqd->eoi_page = (vm_offset_t)pmap_mapdev(eoi_phys, irqd->esb_size);
718 
719 	if (eoi_phys == trig_phys)
720 		irqd->trig_page = irqd->eoi_page;
721 	else if (trig_phys != 0)
722 		irqd->trig_page = (vm_offset_t)pmap_mapdev(trig_phys,
723 		    irqd->esb_size);
724 	else
725 		irqd->trig_page = 0;
726 
727 	opal_call(OPAL_XIVE_GET_IRQ_CONFIG, irq, vtophys(&irqd->vp),
728 	    vtophys(&irqd->prio), vtophys(&irqd->lirq));
729 
730 	irqd->vp = be64toh(irqd->vp);
731 	irqd->prio = be64toh(irqd->prio);
732 	irqd->lirq = be32toh(irqd->lirq);
733 }
734 
735 /* Allocate an IRQ struct before populating it. */
736 static struct xive_irq *
737 xive_configure_irq(u_int irq)
738 {
739 	struct xive_irq *irqd;
740 
741 	irqd = malloc(sizeof(struct xive_irq), M_XIVE, M_WAITOK);
742 
743 	xive_init_irq(irqd, irq);
744 
745 	return (irqd);
746 }
747 
748 /*
749  * Part of the OPAL API.  OPAL_XIVE_ALLOCATE_VP_BLOCK might require more pages,
750  * provisioned through this call.
751  */
752 static int
753 xive_provision_page(struct xive_softc *sc)
754 {
755 	void *prov_page;
756 	int error;
757 
758 	do {
759 		prov_page = contigmalloc(sc->sc_prov_page_size, M_XIVE, 0,
760 		    0, BUS_SPACE_MAXADDR,
761 		    sc->sc_prov_page_size, sc->sc_prov_page_size);
762 
763 		error = opal_call(OPAL_XIVE_DONATE_PAGE, -1,
764 		    vtophys(prov_page));
765 	} while (error == OPAL_XIVE_PROVISIONING);
766 
767 	return (0);
768 }
769 
770 /* The XIVE_TM_CPPR register must be set by each thread */
771 static void
772 xive_smp_cpu_startup(void)
773 {
774 
775 	xive_setup_cpu();
776 }
777