xref: /freebsd/sys/powerpc/powernv/xive.c (revision 22cf89c938886d14f5796fc49f9f020c23ea8eaf)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 #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/endian.h>
37 #include <sys/kernel.h>
38 #include <sys/lock.h>
39 #include <sys/malloc.h>
40 #include <sys/mutex.h>
41 #include <sys/smp.h>
42 
43 #include <vm/vm.h>
44 #include <vm/pmap.h>
45 
46 #include <machine/bus.h>
47 #include <machine/intr_machdep.h>
48 #include <machine/md_var.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 "pic_if.h"
58 
59 #define XIVE_PRIORITY	7	/* Random non-zero number */
60 #define MAX_XIVE_IRQS	(1<<24)	/* 24-bit XIRR field */
61 
62 /* Registers */
63 #define	XIVE_TM_QW1_OS		0x010	/* Guest OS registers */
64 #define	XIVE_TM_QW2_HV_POOL	0x020	/* Hypervisor pool registers */
65 #define	XIVE_TM_QW3_HV		0x030	/* Hypervisor registers */
66 
67 #define	XIVE_TM_NSR	0x00
68 #define	XIVE_TM_CPPR	0x01
69 #define	XIVE_TM_IPB	0x02
70 #define	XIVE_TM_LSMFB	0x03
71 #define	XIVE_TM_ACK_CNT	0x04
72 #define	XIVE_TM_INC	0x05
73 #define	XIVE_TM_AGE	0x06
74 #define	XIVE_TM_PIPR	0x07
75 
76 #define	TM_WORD0	0x0
77 #define	TM_WORD2	0x8
78 #define	  TM_QW2W2_VP	  0x80000000
79 
80 #define	XIVE_TM_SPC_ACK			0x800
81 #define	  TM_QW3NSR_HE_SHIFT		  14
82 #define	  TM_QW3_NSR_HE_NONE		  0
83 #define	  TM_QW3_NSR_HE_POOL		  1
84 #define	  TM_QW3_NSR_HE_PHYS		  2
85 #define	  TM_QW3_NSR_HE_LSI		  3
86 #define	XIVE_TM_SPC_PULL_POOL_CTX	0x828
87 
88 #define	XIVE_IRQ_LOAD_EOI	0x000
89 #define	XIVE_IRQ_STORE_EOI	0x400
90 #define	XIVE_IRQ_PQ_00		0xc00
91 #define	XIVE_IRQ_PQ_01		0xd00
92 
93 #define	XIVE_IRQ_VAL_P		0x02
94 #define	XIVE_IRQ_VAL_Q		0x01
95 
96 struct xive_softc;
97 struct xive_irq;
98 
99 extern void (*powernv_smp_ap_extra_init)(void);
100 
101 /* Private support */
102 static void	xive_setup_cpu(void);
103 static void	xive_smp_cpu_startup(void);
104 static void	xive_init_irq(struct xive_irq *irqd, u_int irq);
105 static struct xive_irq	*xive_configure_irq(u_int irq);
106 static int	xive_provision_page(struct xive_softc *sc);
107 
108 /* Interfaces */
109 static int	xive_probe(device_t);
110 static int	xive_attach(device_t);
111 static int	xics_probe(device_t);
112 static int	xics_attach(device_t);
113 
114 static void	xive_bind(device_t, u_int, cpuset_t, void **);
115 static void	xive_dispatch(device_t, struct trapframe *);
116 static void	xive_enable(device_t, u_int, u_int, void **);
117 static void	xive_eoi(device_t, u_int, void *);
118 static void	xive_ipi(device_t, u_int);
119 static void	xive_mask(device_t, u_int, void *);
120 static void	xive_unmask(device_t, u_int, void *);
121 static void	xive_translate_code(device_t dev, u_int irq, int code,
122 		    enum intr_trigger *trig, enum intr_polarity *pol);
123 
124 static device_method_t  xive_methods[] = {
125 	/* Device interface */
126 	DEVMETHOD(device_probe,		xive_probe),
127 	DEVMETHOD(device_attach,	xive_attach),
128 
129 	/* PIC interface */
130 	DEVMETHOD(pic_bind,		xive_bind),
131 	DEVMETHOD(pic_dispatch,		xive_dispatch),
132 	DEVMETHOD(pic_enable,		xive_enable),
133 	DEVMETHOD(pic_eoi,		xive_eoi),
134 	DEVMETHOD(pic_ipi,		xive_ipi),
135 	DEVMETHOD(pic_mask,		xive_mask),
136 	DEVMETHOD(pic_unmask,		xive_unmask),
137 	DEVMETHOD(pic_translate_code,	xive_translate_code),
138 
139 	DEVMETHOD_END
140 };
141 
142 static device_method_t  xics_methods[] = {
143 	/* Device interface */
144 	DEVMETHOD(device_probe,		xics_probe),
145 	DEVMETHOD(device_attach,	xics_attach),
146 
147 	DEVMETHOD_END
148 };
149 
150 struct xive_softc {
151 	struct mtx sc_mtx;
152 	struct resource *sc_mem;
153 	vm_size_t	sc_prov_page_size;
154 	uint32_t	sc_offset;
155 };
156 
157 struct xive_queue {
158 	uint32_t	*q_page;
159 	uint32_t	*q_eoi_page;
160 	uint32_t	 q_toggle;
161 	uint32_t	 q_size;
162 	uint32_t	 q_index;
163 	uint32_t	 q_mask;
164 };
165 
166 struct xive_irq {
167 	uint32_t	girq;
168 	uint32_t	lirq;
169 	uint64_t	vp;
170 	uint64_t	flags;
171 #define	OPAL_XIVE_IRQ_SHIFT_BUG		0x00000008
172 #define	OPAL_XIVE_IRQ_LSI		0x00000004
173 #define	OPAL_XIVE_IRQ_STORE_EOI		0x00000002
174 #define	OPAL_XIVE_IRQ_TRIGGER_PAGE	0x00000001
175 	uint8_t	prio;
176 	vm_offset_t	eoi_page;
177 	vm_offset_t	trig_page;
178 	vm_size_t	esb_size;
179 	int		chip;
180 };
181 
182 struct xive_cpu {
183 	uint64_t	vp;
184 	uint64_t	flags;
185 	struct xive_irq	ipi_data;
186 	struct xive_queue	queue; /* We only use a single queue for now. */
187 	uint64_t	cam;
188 	uint32_t	chip;
189 };
190 
191 static driver_t xive_driver = {
192 	"xive",
193 	xive_methods,
194 	sizeof(struct xive_softc)
195 };
196 
197 static driver_t xics_driver = {
198 	"xivevc",
199 	xics_methods,
200 	0
201 };
202 
203 EARLY_DRIVER_MODULE(xive, ofwbus, xive_driver, 0, 0, BUS_PASS_INTERRUPT - 1);
204 EARLY_DRIVER_MODULE(xivevc, ofwbus, xics_driver, 0, 0, BUS_PASS_INTERRUPT);
205 
206 MALLOC_DEFINE(M_XIVE, "xive", "XIVE Memory");
207 
208 DPCPU_DEFINE_STATIC(struct xive_cpu, xive_cpu_data);
209 
210 static int xive_ipi_vector = -1;
211 
212 /*
213  * XIVE Exploitation mode driver.
214  *
215  * The XIVE, present in the POWER9 CPU, can run in two modes: XICS emulation
216  * mode, and "Exploitation mode".  XICS emulation mode is compatible with the
217  * POWER8 and earlier XICS interrupt controller, using OPAL calls to emulate
218  * hypervisor calls and memory accesses.  Exploitation mode gives us raw access
219  * to the XIVE MMIO, improving performance significantly.
220  *
221  * The XIVE controller is a very bizarre interrupt controller.  It uses queues
222  * in memory to pass interrupts around, and maps itself into 512GB of physical
223  * device address space, giving each interrupt in the system one or more pages
224  * of address space.  An IRQ is tied to a virtual processor, which could be a
225  * physical CPU thread, or a guest CPU thread (LPAR running on a physical
226  * thread).  Thus, the controller can route interrupts directly to guest OSes
227  * bypassing processing by the hypervisor, thereby improving performance of the
228  * guest OS.
229  *
230  * An IRQ, in addition to being tied to a virtual processor, has one or two
231  * page mappings: an EOI page, and an optional trigger page.  The trigger page
232  * could be the same as the EOI page.  Level-sensitive interrupts (LSIs) don't
233  * have a trigger page, as they're external interrupts controlled by physical
234  * lines.  MSIs and IPIs have trigger pages.  An IPI is really just another IRQ
235  * in the XIVE, which is triggered by software.
236  *
237  * An interesting behavior of the XIVE controller is that oftentimes the
238  * contents of an address location don't actually matter, but the direction of
239  * the action is the signifier (read vs write), and the address is significant.
240  * Hence, masking and unmasking an interrupt is done by reading different
241  * addresses in the EOI page, and triggering an interrupt consists of writing to
242  * the trigger page.
243  *
244  * Additionally, the MMIO region mapped is CPU-sensitive, just like the
245  * per-processor register space (private access) in OpenPIC.  In order for a CPU
246  * to receive interrupts it must itself configure its CPPR (Current Processor
247  * Priority Register), it cannot be set by any other processor.  This
248  * necessitates the xive_smp_cpu_startup() function.
249  *
250  * Queues are pages of memory, sized powers-of-two, that are shared with the
251  * XIVE.  The XIVE writes into the queue with an alternating polarity bit, which
252  * flips when the queue wraps.
253  */
254 
255 /*
256  * Offset-based read/write interfaces.
257  */
258 static uint16_t
259 xive_read_2(struct xive_softc *sc, bus_size_t offset)
260 {
261 
262 	return (bus_read_2(sc->sc_mem, sc->sc_offset + offset));
263 }
264 
265 static void
266 xive_write_1(struct xive_softc *sc, bus_size_t offset, uint8_t val)
267 {
268 
269 	bus_write_1(sc->sc_mem, sc->sc_offset + offset, val);
270 }
271 
272 /* EOI and Trigger page access interfaces. */
273 static uint64_t
274 xive_read_mmap8(vm_offset_t addr)
275 {
276 	return (*(volatile uint64_t *)addr);
277 }
278 
279 static void
280 xive_write_mmap8(vm_offset_t addr, uint64_t val)
281 {
282 	*(uint64_t *)(addr) = val;
283 }
284 
285 /* Device interfaces. */
286 static int
287 xive_probe(device_t dev)
288 {
289 
290 	if (!ofw_bus_is_compatible(dev, "ibm,opal-xive-pe"))
291 		return (ENXIO);
292 
293 	device_set_desc(dev, "External Interrupt Virtualization Engine");
294 
295 	/* Make sure we always win against the xicp driver. */
296 	return (BUS_PROBE_DEFAULT);
297 }
298 
299 static int
300 xics_probe(device_t dev)
301 {
302 
303 	if (!ofw_bus_is_compatible(dev, "ibm,opal-xive-vc"))
304 		return (ENXIO);
305 
306 	device_set_desc(dev, "External Interrupt Virtualization Engine Root");
307 	return (BUS_PROBE_DEFAULT);
308 }
309 
310 static int
311 xive_attach(device_t dev)
312 {
313 	struct xive_softc *sc = device_get_softc(dev);
314 	struct xive_cpu *xive_cpud;
315 	phandle_t phandle = ofw_bus_get_node(dev);
316 	int64_t vp_block;
317 	int error;
318 	int rid;
319 	int i, order;
320 	uint64_t vp_id;
321 	int64_t ipi_irq;
322 
323 	opal_call(OPAL_XIVE_RESET, OPAL_XIVE_XICS_MODE_EXP);
324 
325 	error = OF_getencprop(phandle, "ibm,xive-provision-page-size",
326 	    (pcell_t *)&sc->sc_prov_page_size, sizeof(sc->sc_prov_page_size));
327 
328 	rid = 1;	/* Get the Hypervisor-level register set. */
329 	sc->sc_mem = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
330 	    &rid, RF_ACTIVE);
331 	sc->sc_offset = XIVE_TM_QW3_HV;
332 
333 	mtx_init(&sc->sc_mtx, "XIVE", NULL, MTX_DEF);
334 
335 	/* Workaround for qemu single-thread powernv */
336 	if (mp_maxid == 0)
337 		order = 1;
338 	else
339 		order = fls(mp_maxid + (mp_maxid - 1)) - 1;
340 
341 	do {
342 		vp_block = opal_call(OPAL_XIVE_ALLOCATE_VP_BLOCK, order);
343 		if (vp_block == OPAL_BUSY)
344 			DELAY(10);
345 		else if (vp_block == OPAL_XIVE_PROVISIONING)
346 			xive_provision_page(sc);
347 		else
348 			break;
349 	} while (1);
350 
351 	if (vp_block < 0) {
352 		device_printf(dev,
353 		    "Unable to allocate VP block.  Opal error %d\n",
354 		    (int)vp_block);
355 		bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->sc_mem);
356 		return (ENXIO);
357 	}
358 
359 	/*
360 	 * Set up the VPs.  Try to do as much as we can in attach, to lessen
361 	 * what's needed at AP spawn time.
362 	 */
363 	CPU_FOREACH(i) {
364 		vp_id = pcpu_find(i)->pc_hwref;
365 
366 		xive_cpud = DPCPU_ID_PTR(i, xive_cpu_data);
367 		xive_cpud->vp = vp_id + vp_block;
368 		opal_call(OPAL_XIVE_GET_VP_INFO, xive_cpud->vp, NULL,
369 		    vtophys(&xive_cpud->cam), NULL, vtophys(&xive_cpud->chip));
370 
371 		xive_cpud->cam = be64toh(xive_cpud->cam);
372 		xive_cpud->chip = be64toh(xive_cpud->chip);
373 
374 		/* Allocate the queue page and populate the queue state data. */
375 		xive_cpud->queue.q_page = contigmalloc(PAGE_SIZE, M_XIVE,
376 		    M_ZERO | M_WAITOK, 0, BUS_SPACE_MAXADDR, PAGE_SIZE, 0);
377 		xive_cpud->queue.q_size = 1 << PAGE_SHIFT;
378 		xive_cpud->queue.q_mask =
379 		    ((xive_cpud->queue.q_size / sizeof(int)) - 1);
380 		xive_cpud->queue.q_toggle = 0;
381 		xive_cpud->queue.q_index = 0;
382 		do {
383 			error = opal_call(OPAL_XIVE_SET_VP_INFO, xive_cpud->vp,
384 			    OPAL_XIVE_VP_ENABLED, 0);
385 		} while (error == OPAL_BUSY);
386 		error = opal_call(OPAL_XIVE_SET_QUEUE_INFO, vp_id,
387 		    XIVE_PRIORITY, vtophys(xive_cpud->queue.q_page), PAGE_SHIFT,
388 		    OPAL_XIVE_EQ_ALWAYS_NOTIFY | OPAL_XIVE_EQ_ENABLED);
389 
390 		do {
391 			ipi_irq = opal_call(OPAL_XIVE_ALLOCATE_IRQ,
392 			    xive_cpud->chip);
393 		} while (ipi_irq == OPAL_BUSY);
394 
395 		if (ipi_irq < 0)
396 			device_printf(root_pic,
397 			    "Failed allocating IPI.  OPAL error %d\n",
398 			    (int)ipi_irq);
399 		else {
400 			xive_init_irq(&xive_cpud->ipi_data, ipi_irq);
401 			xive_cpud->ipi_data.vp = vp_id;
402 			xive_cpud->ipi_data.lirq = MAX_XIVE_IRQS;
403 			opal_call(OPAL_XIVE_SET_IRQ_CONFIG, ipi_irq,
404 			    xive_cpud->ipi_data.vp, XIVE_PRIORITY,
405 			    MAX_XIVE_IRQS);
406 		}
407 	}
408 
409 	powerpc_register_pic(dev, OF_xref_from_node(phandle), MAX_XIVE_IRQS,
410 	    1 /* Number of IPIs */, FALSE);
411 	root_pic = dev;
412 
413 	xive_setup_cpu();
414 	powernv_smp_ap_extra_init = xive_smp_cpu_startup;
415 
416 	return (0);
417 }
418 
419 static int
420 xics_attach(device_t dev)
421 {
422 	phandle_t phandle = ofw_bus_get_node(dev);
423 
424 	/* The XIVE (root PIC) will handle all our interrupts */
425 	powerpc_register_pic(root_pic, OF_xref_from_node(phandle),
426 	    MAX_XIVE_IRQS, 1 /* Number of IPIs */, FALSE);
427 
428 	return (0);
429 }
430 
431 /*
432  * PIC I/F methods.
433  */
434 
435 static void
436 xive_bind(device_t dev, u_int irq, cpuset_t cpumask, void **priv)
437 {
438 	struct xive_irq *irqd;
439 	int cpu;
440 	int ncpus, i, error;
441 
442 	if (*priv == NULL)
443 		*priv = xive_configure_irq(irq);
444 
445 	irqd = *priv;
446 
447 	/*
448 	 * This doesn't appear to actually support affinity groups, so pick a
449 	 * random CPU.
450 	 */
451 	ncpus = 0;
452 	CPU_FOREACH(cpu)
453 		if (CPU_ISSET(cpu, &cpumask)) ncpus++;
454 
455 	i = mftb() % ncpus;
456 	ncpus = 0;
457 	CPU_FOREACH(cpu) {
458 		if (!CPU_ISSET(cpu, &cpumask))
459 			continue;
460 		if (ncpus == i)
461 			break;
462 		ncpus++;
463 	}
464 
465 	opal_call(OPAL_XIVE_SYNC, OPAL_XIVE_SYNC_QUEUE, irq);
466 
467 	irqd->vp = pcpu_find(cpu)->pc_hwref;
468 	error = opal_call(OPAL_XIVE_SET_IRQ_CONFIG, irq, irqd->vp,
469 	    XIVE_PRIORITY, irqd->lirq);
470 
471 	if (error < 0)
472 		panic("Cannot bind interrupt %d to CPU %d", irq, cpu);
473 
474 	xive_eoi(dev, irq, irqd);
475 }
476 
477 /* Read the next entry in the queue page and update the index. */
478 static int
479 xive_read_eq(struct xive_queue *q)
480 {
481 	uint32_t i = be32toh(q->q_page[q->q_index]);
482 
483 	/* Check validity, using current queue polarity. */
484 	if ((i >> 31) == q->q_toggle)
485 		return (0);
486 
487 	q->q_index = (q->q_index + 1) & q->q_mask;
488 
489 	if (q->q_index == 0)
490 		q->q_toggle ^= 1;
491 
492 	return (i & 0x7fffffff);
493 }
494 
495 static void
496 xive_dispatch(device_t dev, struct trapframe *tf)
497 {
498 	struct xive_softc *sc;
499 	struct xive_cpu *xive_cpud;
500 	uint32_t vector;
501 	uint16_t ack;
502 	uint8_t cppr, he;
503 
504 	sc = device_get_softc(dev);
505 
506 	xive_cpud = DPCPU_PTR(xive_cpu_data);
507 	for (;;) {
508 		ack = xive_read_2(sc, XIVE_TM_SPC_ACK);
509 		cppr = (ack & 0xff);
510 
511 		he = ack >> TM_QW3NSR_HE_SHIFT;
512 
513 		if (he == TM_QW3_NSR_HE_NONE)
514 			break;
515 
516 		else if (__predict_false(he != TM_QW3_NSR_HE_PHYS)) {
517 			/*
518 			 * We don't support TM_QW3_NSR_HE_POOL or
519 			 * TM_QW3_NSR_HE_LSI interrupts.
520 			 */
521 			device_printf(dev,
522 			    "Unexpected interrupt he type: %d\n", he);
523 			goto end;
524 		}
525 
526 		xive_write_1(sc, XIVE_TM_CPPR, cppr);
527 
528 		for (;;) {
529 			vector = xive_read_eq(&xive_cpud->queue);
530 
531 			if (vector == 0)
532 				break;
533 
534 			if (vector == MAX_XIVE_IRQS)
535 				vector = xive_ipi_vector;
536 
537 			powerpc_dispatch_intr(vector, tf);
538 		}
539 	}
540 end:
541 	xive_write_1(sc, XIVE_TM_CPPR, 0xff);
542 }
543 
544 static void
545 xive_enable(device_t dev, u_int irq, u_int vector, void **priv)
546 {
547 	struct xive_irq *irqd;
548 	cell_t status, cpu;
549 
550 	if (irq == MAX_XIVE_IRQS) {
551 		if (xive_ipi_vector == -1)
552 			xive_ipi_vector = vector;
553 		return;
554 	}
555 	if (*priv == NULL)
556 		*priv = xive_configure_irq(irq);
557 
558 	irqd = *priv;
559 
560 	/* Bind to this CPU to start */
561 	cpu = PCPU_GET(hwref);
562 	irqd->lirq = vector;
563 
564 	for (;;) {
565 		status = opal_call(OPAL_XIVE_SET_IRQ_CONFIG, irq, cpu,
566 		    XIVE_PRIORITY, vector);
567 		if (status != OPAL_BUSY)
568 			break;
569 		DELAY(10);
570 	}
571 
572 	if (status != 0)
573 		panic("OPAL_SET_XIVE IRQ %d -> cpu %d failed: %d", irq,
574 		    cpu, status);
575 
576 	xive_unmask(dev, irq, *priv);
577 }
578 
579 static void
580 xive_eoi(device_t dev, u_int irq, void *priv)
581 {
582 	struct xive_irq *rirq;
583 	struct xive_cpu *cpud;
584 	uint8_t eoi_val;
585 
586 	if (irq == MAX_XIVE_IRQS) {
587 		cpud = DPCPU_PTR(xive_cpu_data);
588 		rirq = &cpud->ipi_data;
589 	} else
590 		rirq = priv;
591 
592 	if (rirq->flags & OPAL_XIVE_IRQ_STORE_EOI)
593 		xive_write_mmap8(rirq->eoi_page + XIVE_IRQ_STORE_EOI, 0);
594 	else if (rirq->flags & OPAL_XIVE_IRQ_LSI)
595 		xive_read_mmap8(rirq->eoi_page + XIVE_IRQ_LOAD_EOI);
596 	else {
597 		eoi_val = xive_read_mmap8(rirq->eoi_page + XIVE_IRQ_PQ_00);
598 		if ((eoi_val & XIVE_IRQ_VAL_Q) && rirq->trig_page != 0)
599 			xive_write_mmap8(rirq->trig_page, 0);
600 	}
601 }
602 
603 static void
604 xive_ipi(device_t dev, u_int cpu)
605 {
606 	struct xive_cpu *xive_cpud;
607 
608 	xive_cpud = DPCPU_ID_PTR(cpu, xive_cpu_data);
609 
610 	if (xive_cpud->ipi_data.trig_page == 0)
611 		return;
612 	xive_write_mmap8(xive_cpud->ipi_data.trig_page, 0);
613 }
614 
615 static void
616 xive_mask(device_t dev, u_int irq, void *priv)
617 {
618 	struct xive_irq *rirq;
619 
620 	/* Never mask IPIs */
621 	if (irq == MAX_XIVE_IRQS)
622 		return;
623 
624 	rirq = priv;
625 
626 	if (!(rirq->flags & OPAL_XIVE_IRQ_LSI))
627 		return;
628 	xive_read_mmap8(rirq->eoi_page + XIVE_IRQ_PQ_01);
629 }
630 
631 static void
632 xive_unmask(device_t dev, u_int irq, void *priv)
633 {
634 	struct xive_irq *rirq;
635 
636 	rirq = priv;
637 
638 	xive_read_mmap8(rirq->eoi_page + XIVE_IRQ_PQ_00);
639 }
640 
641 static void
642 xive_translate_code(device_t dev, u_int irq, int code,
643     enum intr_trigger *trig, enum intr_polarity *pol)
644 {
645 	switch (code) {
646 	case 0:
647 		/* L to H edge */
648 		*trig = INTR_TRIGGER_EDGE;
649 		*pol = INTR_POLARITY_HIGH;
650 		break;
651 	case 1:
652 		/* Active L level */
653 		*trig = INTR_TRIGGER_LEVEL;
654 		*pol = INTR_POLARITY_LOW;
655 		break;
656 	default:
657 		*trig = INTR_TRIGGER_CONFORM;
658 		*pol = INTR_POLARITY_CONFORM;
659 	}
660 }
661 
662 /* Private functions. */
663 /*
664  * Setup the current CPU.  Called by the BSP at driver attachment, and by each
665  * AP at wakeup (via xive_smp_cpu_startup()).
666  */
667 static void
668 xive_setup_cpu(void)
669 {
670 	struct xive_softc *sc;
671 	struct xive_cpu *cpup;
672 	uint32_t val;
673 
674 	cpup = DPCPU_PTR(xive_cpu_data);
675 
676 	sc = device_get_softc(root_pic);
677 
678 	val = bus_read_4(sc->sc_mem, XIVE_TM_QW2_HV_POOL + TM_WORD2);
679 	if (val & TM_QW2W2_VP)
680 		bus_read_8(sc->sc_mem, XIVE_TM_SPC_PULL_POOL_CTX);
681 
682 	bus_write_4(sc->sc_mem, XIVE_TM_QW2_HV_POOL + TM_WORD0, 0xff);
683 	bus_write_4(sc->sc_mem, XIVE_TM_QW2_HV_POOL + TM_WORD2,
684 	    TM_QW2W2_VP | cpup->cam);
685 
686 	xive_unmask(root_pic, cpup->ipi_data.girq, &cpup->ipi_data);
687 	xive_write_1(sc, XIVE_TM_CPPR, 0xff);
688 }
689 
690 /* Populate an IRQ structure, mapping the EOI and trigger pages. */
691 static void
692 xive_init_irq(struct xive_irq *irqd, u_int irq)
693 {
694 	uint64_t eoi_phys, trig_phys;
695 	uint32_t esb_shift;
696 
697 	opal_call(OPAL_XIVE_GET_IRQ_INFO, irq,
698 	    vtophys(&irqd->flags), vtophys(&eoi_phys),
699 	    vtophys(&trig_phys), vtophys(&esb_shift),
700 	    vtophys(&irqd->chip));
701 
702 	irqd->flags = be64toh(irqd->flags);
703 	eoi_phys = be64toh(eoi_phys);
704 	trig_phys = be64toh(trig_phys);
705 	esb_shift = be32toh(esb_shift);
706 	irqd->chip = be32toh(irqd->chip);
707 
708 	irqd->girq = irq;
709 	irqd->esb_size = 1 << esb_shift;
710 	irqd->eoi_page = (vm_offset_t)pmap_mapdev(eoi_phys, irqd->esb_size);
711 
712 	if (eoi_phys == trig_phys)
713 		irqd->trig_page = irqd->eoi_page;
714 	else if (trig_phys != 0)
715 		irqd->trig_page = (vm_offset_t)pmap_mapdev(trig_phys,
716 		    irqd->esb_size);
717 	else
718 		irqd->trig_page = 0;
719 
720 	opal_call(OPAL_XIVE_GET_IRQ_CONFIG, irq, vtophys(&irqd->vp),
721 	    vtophys(&irqd->prio), vtophys(&irqd->lirq));
722 
723 	irqd->vp = be64toh(irqd->vp);
724 	irqd->prio = be64toh(irqd->prio);
725 	irqd->lirq = be32toh(irqd->lirq);
726 }
727 
728 /* Allocate an IRQ struct before populating it. */
729 static struct xive_irq *
730 xive_configure_irq(u_int irq)
731 {
732 	struct xive_irq *irqd;
733 
734 	irqd = malloc(sizeof(struct xive_irq), M_XIVE, M_WAITOK);
735 
736 	xive_init_irq(irqd, irq);
737 
738 	return (irqd);
739 }
740 
741 /*
742  * Part of the OPAL API.  OPAL_XIVE_ALLOCATE_VP_BLOCK might require more pages,
743  * provisioned through this call.
744  */
745 static int
746 xive_provision_page(struct xive_softc *sc)
747 {
748 	void *prov_page;
749 	int error;
750 
751 	do {
752 		prov_page = contigmalloc(sc->sc_prov_page_size, M_XIVE, 0,
753 		    0, BUS_SPACE_MAXADDR,
754 		    sc->sc_prov_page_size, sc->sc_prov_page_size);
755 
756 		error = opal_call(OPAL_XIVE_DONATE_PAGE, -1,
757 		    vtophys(prov_page));
758 	} while (error == OPAL_XIVE_PROVISIONING);
759 
760 	return (0);
761 }
762 
763 /* The XIVE_TM_CPPR register must be set by each thread */
764 static void
765 xive_smp_cpu_startup(void)
766 {
767 
768 	xive_setup_cpu();
769 }
770