xref: /openbmc/linux/arch/powerpc/sysdev/xive/spapr.c (revision cef69974)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright 2016,2017 IBM Corporation.
4  */
5 
6 #define pr_fmt(fmt) "xive: " fmt
7 
8 #include <linux/types.h>
9 #include <linux/irq.h>
10 #include <linux/smp.h>
11 #include <linux/interrupt.h>
12 #include <linux/init.h>
13 #include <linux/of.h>
14 #include <linux/slab.h>
15 #include <linux/spinlock.h>
16 #include <linux/cpumask.h>
17 #include <linux/mm.h>
18 #include <linux/delay.h>
19 #include <linux/libfdt.h>
20 
21 #include <asm/machdep.h>
22 #include <asm/prom.h>
23 #include <asm/io.h>
24 #include <asm/smp.h>
25 #include <asm/irq.h>
26 #include <asm/errno.h>
27 #include <asm/xive.h>
28 #include <asm/xive-regs.h>
29 #include <asm/hvcall.h>
30 #include <asm/svm.h>
31 #include <asm/ultravisor.h>
32 
33 #include "xive-internal.h"
34 
35 static u32 xive_queue_shift;
36 
37 struct xive_irq_bitmap {
38 	unsigned long		*bitmap;
39 	unsigned int		base;
40 	unsigned int		count;
41 	spinlock_t		lock;
42 	struct list_head	list;
43 };
44 
45 static LIST_HEAD(xive_irq_bitmaps);
46 
47 static int __init xive_irq_bitmap_add(int base, int count)
48 {
49 	struct xive_irq_bitmap *xibm;
50 
51 	xibm = kzalloc(sizeof(*xibm), GFP_KERNEL);
52 	if (!xibm)
53 		return -ENOMEM;
54 
55 	spin_lock_init(&xibm->lock);
56 	xibm->base = base;
57 	xibm->count = count;
58 	xibm->bitmap = kzalloc(xibm->count, GFP_KERNEL);
59 	if (!xibm->bitmap) {
60 		kfree(xibm);
61 		return -ENOMEM;
62 	}
63 	list_add(&xibm->list, &xive_irq_bitmaps);
64 
65 	pr_info("Using IRQ range [%x-%x]", xibm->base,
66 		xibm->base + xibm->count - 1);
67 	return 0;
68 }
69 
70 static void xive_irq_bitmap_remove_all(void)
71 {
72 	struct xive_irq_bitmap *xibm, *tmp;
73 
74 	list_for_each_entry_safe(xibm, tmp, &xive_irq_bitmaps, list) {
75 		list_del(&xibm->list);
76 		kfree(xibm->bitmap);
77 		kfree(xibm);
78 	}
79 }
80 
81 static int __xive_irq_bitmap_alloc(struct xive_irq_bitmap *xibm)
82 {
83 	int irq;
84 
85 	irq = find_first_zero_bit(xibm->bitmap, xibm->count);
86 	if (irq != xibm->count) {
87 		set_bit(irq, xibm->bitmap);
88 		irq += xibm->base;
89 	} else {
90 		irq = -ENOMEM;
91 	}
92 
93 	return irq;
94 }
95 
96 static int xive_irq_bitmap_alloc(void)
97 {
98 	struct xive_irq_bitmap *xibm;
99 	unsigned long flags;
100 	int irq = -ENOENT;
101 
102 	list_for_each_entry(xibm, &xive_irq_bitmaps, list) {
103 		spin_lock_irqsave(&xibm->lock, flags);
104 		irq = __xive_irq_bitmap_alloc(xibm);
105 		spin_unlock_irqrestore(&xibm->lock, flags);
106 		if (irq >= 0)
107 			break;
108 	}
109 	return irq;
110 }
111 
112 static void xive_irq_bitmap_free(int irq)
113 {
114 	unsigned long flags;
115 	struct xive_irq_bitmap *xibm;
116 
117 	list_for_each_entry(xibm, &xive_irq_bitmaps, list) {
118 		if ((irq >= xibm->base) && (irq < xibm->base + xibm->count)) {
119 			spin_lock_irqsave(&xibm->lock, flags);
120 			clear_bit(irq - xibm->base, xibm->bitmap);
121 			spin_unlock_irqrestore(&xibm->lock, flags);
122 			break;
123 		}
124 	}
125 }
126 
127 
128 /* Based on the similar routines in RTAS */
129 static unsigned int plpar_busy_delay_time(long rc)
130 {
131 	unsigned int ms = 0;
132 
133 	if (H_IS_LONG_BUSY(rc)) {
134 		ms = get_longbusy_msecs(rc);
135 	} else if (rc == H_BUSY) {
136 		ms = 10; /* seems appropriate for XIVE hcalls */
137 	}
138 
139 	return ms;
140 }
141 
142 static unsigned int plpar_busy_delay(int rc)
143 {
144 	unsigned int ms;
145 
146 	ms = plpar_busy_delay_time(rc);
147 	if (ms)
148 		mdelay(ms);
149 
150 	return ms;
151 }
152 
153 /*
154  * Note: this call has a partition wide scope and can take a while to
155  * complete. If it returns H_LONG_BUSY_* it should be retried
156  * periodically.
157  */
158 static long plpar_int_reset(unsigned long flags)
159 {
160 	long rc;
161 
162 	do {
163 		rc = plpar_hcall_norets(H_INT_RESET, flags);
164 	} while (plpar_busy_delay(rc));
165 
166 	if (rc)
167 		pr_err("H_INT_RESET failed %ld\n", rc);
168 
169 	return rc;
170 }
171 
172 static long plpar_int_get_source_info(unsigned long flags,
173 				      unsigned long lisn,
174 				      unsigned long *src_flags,
175 				      unsigned long *eoi_page,
176 				      unsigned long *trig_page,
177 				      unsigned long *esb_shift)
178 {
179 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
180 	long rc;
181 
182 	do {
183 		rc = plpar_hcall(H_INT_GET_SOURCE_INFO, retbuf, flags, lisn);
184 	} while (plpar_busy_delay(rc));
185 
186 	if (rc) {
187 		pr_err("H_INT_GET_SOURCE_INFO lisn=0x%lx failed %ld\n", lisn, rc);
188 		return rc;
189 	}
190 
191 	*src_flags = retbuf[0];
192 	*eoi_page  = retbuf[1];
193 	*trig_page = retbuf[2];
194 	*esb_shift = retbuf[3];
195 
196 	pr_debug("H_INT_GET_SOURCE_INFO lisn=0x%lx flags=0x%lx eoi=0x%lx trig=0x%lx shift=0x%lx\n",
197 		 lisn, retbuf[0], retbuf[1], retbuf[2], retbuf[3]);
198 
199 	return 0;
200 }
201 
202 #define XIVE_SRC_SET_EISN (1ull << (63 - 62))
203 #define XIVE_SRC_MASK     (1ull << (63 - 63)) /* unused */
204 
205 static long plpar_int_set_source_config(unsigned long flags,
206 					unsigned long lisn,
207 					unsigned long target,
208 					unsigned long prio,
209 					unsigned long sw_irq)
210 {
211 	long rc;
212 
213 
214 	pr_debug("H_INT_SET_SOURCE_CONFIG flags=0x%lx lisn=0x%lx target=%ld prio=%ld sw_irq=%ld\n",
215 		 flags, lisn, target, prio, sw_irq);
216 
217 
218 	do {
219 		rc = plpar_hcall_norets(H_INT_SET_SOURCE_CONFIG, flags, lisn,
220 					target, prio, sw_irq);
221 	} while (plpar_busy_delay(rc));
222 
223 	if (rc) {
224 		pr_err("H_INT_SET_SOURCE_CONFIG lisn=0x%lx target=%ld prio=%ld failed %ld\n",
225 		       lisn, target, prio, rc);
226 		return rc;
227 	}
228 
229 	return 0;
230 }
231 
232 static long plpar_int_get_source_config(unsigned long flags,
233 					unsigned long lisn,
234 					unsigned long *target,
235 					unsigned long *prio,
236 					unsigned long *sw_irq)
237 {
238 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
239 	long rc;
240 
241 	pr_debug("H_INT_GET_SOURCE_CONFIG flags=0x%lx lisn=0x%lx\n", flags, lisn);
242 
243 	do {
244 		rc = plpar_hcall(H_INT_GET_SOURCE_CONFIG, retbuf, flags, lisn,
245 				 target, prio, sw_irq);
246 	} while (plpar_busy_delay(rc));
247 
248 	if (rc) {
249 		pr_err("H_INT_GET_SOURCE_CONFIG lisn=0x%lx failed %ld\n",
250 		       lisn, rc);
251 		return rc;
252 	}
253 
254 	*target = retbuf[0];
255 	*prio   = retbuf[1];
256 	*sw_irq = retbuf[2];
257 
258 	pr_debug("H_INT_GET_SOURCE_CONFIG target=%ld prio=%ld sw_irq=%ld\n",
259 		 retbuf[0], retbuf[1], retbuf[2]);
260 
261 	return 0;
262 }
263 
264 static long plpar_int_get_queue_info(unsigned long flags,
265 				     unsigned long target,
266 				     unsigned long priority,
267 				     unsigned long *esn_page,
268 				     unsigned long *esn_size)
269 {
270 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
271 	long rc;
272 
273 	do {
274 		rc = plpar_hcall(H_INT_GET_QUEUE_INFO, retbuf, flags, target,
275 				 priority);
276 	} while (plpar_busy_delay(rc));
277 
278 	if (rc) {
279 		pr_err("H_INT_GET_QUEUE_INFO cpu=%ld prio=%ld failed %ld\n",
280 		       target, priority, rc);
281 		return rc;
282 	}
283 
284 	*esn_page = retbuf[0];
285 	*esn_size = retbuf[1];
286 
287 	pr_debug("H_INT_GET_QUEUE_INFO cpu=%ld prio=%ld page=0x%lx size=0x%lx\n",
288 		 target, priority, retbuf[0], retbuf[1]);
289 
290 	return 0;
291 }
292 
293 #define XIVE_EQ_ALWAYS_NOTIFY (1ull << (63 - 63))
294 
295 static long plpar_int_set_queue_config(unsigned long flags,
296 				       unsigned long target,
297 				       unsigned long priority,
298 				       unsigned long qpage,
299 				       unsigned long qsize)
300 {
301 	long rc;
302 
303 	pr_debug("H_INT_SET_QUEUE_CONFIG flags=0x%lx target=%ld priority=0x%lx qpage=0x%lx qsize=0x%lx\n",
304 		 flags,  target, priority, qpage, qsize);
305 
306 	do {
307 		rc = plpar_hcall_norets(H_INT_SET_QUEUE_CONFIG, flags, target,
308 					priority, qpage, qsize);
309 	} while (plpar_busy_delay(rc));
310 
311 	if (rc) {
312 		pr_err("H_INT_SET_QUEUE_CONFIG cpu=%ld prio=%ld qpage=0x%lx returned %ld\n",
313 		       target, priority, qpage, rc);
314 		return  rc;
315 	}
316 
317 	return 0;
318 }
319 
320 static long plpar_int_sync(unsigned long flags, unsigned long lisn)
321 {
322 	long rc;
323 
324 	do {
325 		rc = plpar_hcall_norets(H_INT_SYNC, flags, lisn);
326 	} while (plpar_busy_delay(rc));
327 
328 	if (rc) {
329 		pr_err("H_INT_SYNC lisn=0x%lx returned %ld\n", lisn, rc);
330 		return  rc;
331 	}
332 
333 	return 0;
334 }
335 
336 #define XIVE_ESB_FLAG_STORE (1ull << (63 - 63))
337 
338 static long plpar_int_esb(unsigned long flags,
339 			  unsigned long lisn,
340 			  unsigned long offset,
341 			  unsigned long in_data,
342 			  unsigned long *out_data)
343 {
344 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
345 	long rc;
346 
347 	pr_debug("H_INT_ESB flags=0x%lx lisn=0x%lx offset=0x%lx in=0x%lx\n",
348 		 flags,  lisn, offset, in_data);
349 
350 	do {
351 		rc = plpar_hcall(H_INT_ESB, retbuf, flags, lisn, offset,
352 				 in_data);
353 	} while (plpar_busy_delay(rc));
354 
355 	if (rc) {
356 		pr_err("H_INT_ESB lisn=0x%lx offset=0x%lx returned %ld\n",
357 		       lisn, offset, rc);
358 		return  rc;
359 	}
360 
361 	*out_data = retbuf[0];
362 
363 	return 0;
364 }
365 
366 static u64 xive_spapr_esb_rw(u32 lisn, u32 offset, u64 data, bool write)
367 {
368 	unsigned long read_data;
369 	long rc;
370 
371 	rc = plpar_int_esb(write ? XIVE_ESB_FLAG_STORE : 0,
372 			   lisn, offset, data, &read_data);
373 	if (rc)
374 		return -1;
375 
376 	return write ? 0 : read_data;
377 }
378 
379 #define XIVE_SRC_H_INT_ESB     (1ull << (63 - 60))
380 #define XIVE_SRC_LSI           (1ull << (63 - 61))
381 #define XIVE_SRC_TRIGGER       (1ull << (63 - 62))
382 #define XIVE_SRC_STORE_EOI     (1ull << (63 - 63))
383 
384 static int xive_spapr_populate_irq_data(u32 hw_irq, struct xive_irq_data *data)
385 {
386 	long rc;
387 	unsigned long flags;
388 	unsigned long eoi_page;
389 	unsigned long trig_page;
390 	unsigned long esb_shift;
391 
392 	memset(data, 0, sizeof(*data));
393 
394 	rc = plpar_int_get_source_info(0, hw_irq, &flags, &eoi_page, &trig_page,
395 				       &esb_shift);
396 	if (rc)
397 		return  -EINVAL;
398 
399 	if (flags & XIVE_SRC_H_INT_ESB)
400 		data->flags  |= XIVE_IRQ_FLAG_H_INT_ESB;
401 	if (flags & XIVE_SRC_STORE_EOI)
402 		data->flags  |= XIVE_IRQ_FLAG_STORE_EOI;
403 	if (flags & XIVE_SRC_LSI)
404 		data->flags  |= XIVE_IRQ_FLAG_LSI;
405 	data->eoi_page  = eoi_page;
406 	data->esb_shift = esb_shift;
407 	data->trig_page = trig_page;
408 
409 	data->hw_irq = hw_irq;
410 
411 	/*
412 	 * No chip-id for the sPAPR backend. This has an impact how we
413 	 * pick a target. See xive_pick_irq_target().
414 	 */
415 	data->src_chip = XIVE_INVALID_CHIP_ID;
416 
417 	/*
418 	 * When the H_INT_ESB flag is set, the H_INT_ESB hcall should
419 	 * be used for interrupt management. Skip the remapping of the
420 	 * ESB pages which are not available.
421 	 */
422 	if (data->flags & XIVE_IRQ_FLAG_H_INT_ESB)
423 		return 0;
424 
425 	data->eoi_mmio = ioremap(data->eoi_page, 1u << data->esb_shift);
426 	if (!data->eoi_mmio) {
427 		pr_err("Failed to map EOI page for irq 0x%x\n", hw_irq);
428 		return -ENOMEM;
429 	}
430 
431 	/* Full function page supports trigger */
432 	if (flags & XIVE_SRC_TRIGGER) {
433 		data->trig_mmio = data->eoi_mmio;
434 		return 0;
435 	}
436 
437 	data->trig_mmio = ioremap(data->trig_page, 1u << data->esb_shift);
438 	if (!data->trig_mmio) {
439 		pr_err("Failed to map trigger page for irq 0x%x\n", hw_irq);
440 		return -ENOMEM;
441 	}
442 	return 0;
443 }
444 
445 static int xive_spapr_configure_irq(u32 hw_irq, u32 target, u8 prio, u32 sw_irq)
446 {
447 	long rc;
448 
449 	rc = plpar_int_set_source_config(XIVE_SRC_SET_EISN, hw_irq, target,
450 					 prio, sw_irq);
451 
452 	return rc == 0 ? 0 : -ENXIO;
453 }
454 
455 static int xive_spapr_get_irq_config(u32 hw_irq, u32 *target, u8 *prio,
456 				     u32 *sw_irq)
457 {
458 	long rc;
459 	unsigned long h_target;
460 	unsigned long h_prio;
461 	unsigned long h_sw_irq;
462 
463 	rc = plpar_int_get_source_config(0, hw_irq, &h_target, &h_prio,
464 					 &h_sw_irq);
465 
466 	*target = h_target;
467 	*prio = h_prio;
468 	*sw_irq = h_sw_irq;
469 
470 	return rc == 0 ? 0 : -ENXIO;
471 }
472 
473 /* This can be called multiple time to change a queue configuration */
474 static int xive_spapr_configure_queue(u32 target, struct xive_q *q, u8 prio,
475 				   __be32 *qpage, u32 order)
476 {
477 	s64 rc = 0;
478 	unsigned long esn_page;
479 	unsigned long esn_size;
480 	u64 flags, qpage_phys;
481 
482 	/* If there's an actual queue page, clean it */
483 	if (order) {
484 		if (WARN_ON(!qpage))
485 			return -EINVAL;
486 		qpage_phys = __pa(qpage);
487 	} else {
488 		qpage_phys = 0;
489 	}
490 
491 	/* Initialize the rest of the fields */
492 	q->msk = order ? ((1u << (order - 2)) - 1) : 0;
493 	q->idx = 0;
494 	q->toggle = 0;
495 
496 	rc = plpar_int_get_queue_info(0, target, prio, &esn_page, &esn_size);
497 	if (rc) {
498 		pr_err("Error %lld getting queue info CPU %d prio %d\n", rc,
499 		       target, prio);
500 		rc = -EIO;
501 		goto fail;
502 	}
503 
504 	/* TODO: add support for the notification page */
505 	q->eoi_phys = esn_page;
506 
507 	/* Default is to always notify */
508 	flags = XIVE_EQ_ALWAYS_NOTIFY;
509 
510 	/* Configure and enable the queue in HW */
511 	rc = plpar_int_set_queue_config(flags, target, prio, qpage_phys, order);
512 	if (rc) {
513 		pr_err("Error %lld setting queue for CPU %d prio %d\n", rc,
514 		       target, prio);
515 		rc = -EIO;
516 	} else {
517 		q->qpage = qpage;
518 		if (is_secure_guest())
519 			uv_share_page(PHYS_PFN(qpage_phys),
520 					1 << xive_alloc_order(order));
521 	}
522 fail:
523 	return rc;
524 }
525 
526 static int xive_spapr_setup_queue(unsigned int cpu, struct xive_cpu *xc,
527 				  u8 prio)
528 {
529 	struct xive_q *q = &xc->queue[prio];
530 	__be32 *qpage;
531 
532 	qpage = xive_queue_page_alloc(cpu, xive_queue_shift);
533 	if (IS_ERR(qpage))
534 		return PTR_ERR(qpage);
535 
536 	return xive_spapr_configure_queue(get_hard_smp_processor_id(cpu),
537 					  q, prio, qpage, xive_queue_shift);
538 }
539 
540 static void xive_spapr_cleanup_queue(unsigned int cpu, struct xive_cpu *xc,
541 				  u8 prio)
542 {
543 	struct xive_q *q = &xc->queue[prio];
544 	unsigned int alloc_order;
545 	long rc;
546 	int hw_cpu = get_hard_smp_processor_id(cpu);
547 
548 	rc = plpar_int_set_queue_config(0, hw_cpu, prio, 0, 0);
549 	if (rc)
550 		pr_err("Error %ld setting queue for CPU %d prio %d\n", rc,
551 		       hw_cpu, prio);
552 
553 	alloc_order = xive_alloc_order(xive_queue_shift);
554 	if (is_secure_guest())
555 		uv_unshare_page(PHYS_PFN(__pa(q->qpage)), 1 << alloc_order);
556 	free_pages((unsigned long)q->qpage, alloc_order);
557 	q->qpage = NULL;
558 }
559 
560 static bool xive_spapr_match(struct device_node *node)
561 {
562 	/* Ignore cascaded controllers for the moment */
563 	return true;
564 }
565 
566 #ifdef CONFIG_SMP
567 static int xive_spapr_get_ipi(unsigned int cpu, struct xive_cpu *xc)
568 {
569 	int irq = xive_irq_bitmap_alloc();
570 
571 	if (irq < 0) {
572 		pr_err("Failed to allocate IPI on CPU %d\n", cpu);
573 		return -ENXIO;
574 	}
575 
576 	xc->hw_ipi = irq;
577 	return 0;
578 }
579 
580 static void xive_spapr_put_ipi(unsigned int cpu, struct xive_cpu *xc)
581 {
582 	if (xc->hw_ipi == XIVE_BAD_IRQ)
583 		return;
584 
585 	xive_irq_bitmap_free(xc->hw_ipi);
586 	xc->hw_ipi = XIVE_BAD_IRQ;
587 }
588 #endif /* CONFIG_SMP */
589 
590 static void xive_spapr_shutdown(void)
591 {
592 	plpar_int_reset(0);
593 }
594 
595 /*
596  * Perform an "ack" cycle on the current thread. Grab the pending
597  * active priorities and update the CPPR to the most favored one.
598  */
599 static void xive_spapr_update_pending(struct xive_cpu *xc)
600 {
601 	u8 nsr, cppr;
602 	u16 ack;
603 
604 	/*
605 	 * Perform the "Acknowledge O/S to Register" cycle.
606 	 *
607 	 * Let's speedup the access to the TIMA using the raw I/O
608 	 * accessor as we don't need the synchronisation routine of
609 	 * the higher level ones
610 	 */
611 	ack = be16_to_cpu(__raw_readw(xive_tima + TM_SPC_ACK_OS_REG));
612 
613 	/* Synchronize subsequent queue accesses */
614 	mb();
615 
616 	/*
617 	 * Grab the CPPR and the "NSR" field which indicates the source
618 	 * of the interrupt (if any)
619 	 */
620 	cppr = ack & 0xff;
621 	nsr = ack >> 8;
622 
623 	if (nsr & TM_QW1_NSR_EO) {
624 		if (cppr == 0xff)
625 			return;
626 		/* Mark the priority pending */
627 		xc->pending_prio |= 1 << cppr;
628 
629 		/*
630 		 * A new interrupt should never have a CPPR less favored
631 		 * than our current one.
632 		 */
633 		if (cppr >= xc->cppr)
634 			pr_err("CPU %d odd ack CPPR, got %d at %d\n",
635 			       smp_processor_id(), cppr, xc->cppr);
636 
637 		/* Update our idea of what the CPPR is */
638 		xc->cppr = cppr;
639 	}
640 }
641 
642 static void xive_spapr_setup_cpu(unsigned int cpu, struct xive_cpu *xc)
643 {
644 	/* Only some debug on the TIMA settings */
645 	pr_debug("(HW value: %08x %08x %08x)\n",
646 		 in_be32(xive_tima + TM_QW1_OS + TM_WORD0),
647 		 in_be32(xive_tima + TM_QW1_OS + TM_WORD1),
648 		 in_be32(xive_tima + TM_QW1_OS + TM_WORD2));
649 }
650 
651 static void xive_spapr_teardown_cpu(unsigned int cpu, struct xive_cpu *xc)
652 {
653 	/* Nothing to do */;
654 }
655 
656 static void xive_spapr_sync_source(u32 hw_irq)
657 {
658 	/* Specs are unclear on what this is doing */
659 	plpar_int_sync(0, hw_irq);
660 }
661 
662 static int xive_spapr_debug_show(struct seq_file *m, void *private)
663 {
664 	struct xive_irq_bitmap *xibm;
665 	char *buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
666 
667 	if (!buf)
668 		return -ENOMEM;
669 
670 	list_for_each_entry(xibm, &xive_irq_bitmaps, list) {
671 		memset(buf, 0, PAGE_SIZE);
672 		bitmap_print_to_pagebuf(true, buf, xibm->bitmap, xibm->count);
673 		seq_printf(m, "bitmap #%d: %s", xibm->count, buf);
674 	}
675 	kfree(buf);
676 
677 	return 0;
678 }
679 
680 static const struct xive_ops xive_spapr_ops = {
681 	.populate_irq_data	= xive_spapr_populate_irq_data,
682 	.configure_irq		= xive_spapr_configure_irq,
683 	.get_irq_config		= xive_spapr_get_irq_config,
684 	.setup_queue		= xive_spapr_setup_queue,
685 	.cleanup_queue		= xive_spapr_cleanup_queue,
686 	.match			= xive_spapr_match,
687 	.shutdown		= xive_spapr_shutdown,
688 	.update_pending		= xive_spapr_update_pending,
689 	.setup_cpu		= xive_spapr_setup_cpu,
690 	.teardown_cpu		= xive_spapr_teardown_cpu,
691 	.sync_source		= xive_spapr_sync_source,
692 	.esb_rw			= xive_spapr_esb_rw,
693 #ifdef CONFIG_SMP
694 	.get_ipi		= xive_spapr_get_ipi,
695 	.put_ipi		= xive_spapr_put_ipi,
696 	.debug_show		= xive_spapr_debug_show,
697 #endif /* CONFIG_SMP */
698 	.name			= "spapr",
699 };
700 
701 /*
702  * get max priority from "/ibm,plat-res-int-priorities"
703  */
704 static bool __init xive_get_max_prio(u8 *max_prio)
705 {
706 	struct device_node *rootdn;
707 	const __be32 *reg;
708 	u32 len;
709 	int prio, found;
710 
711 	rootdn = of_find_node_by_path("/");
712 	if (!rootdn) {
713 		pr_err("not root node found !\n");
714 		return false;
715 	}
716 
717 	reg = of_get_property(rootdn, "ibm,plat-res-int-priorities", &len);
718 	if (!reg) {
719 		pr_err("Failed to read 'ibm,plat-res-int-priorities' property\n");
720 		return false;
721 	}
722 
723 	if (len % (2 * sizeof(u32)) != 0) {
724 		pr_err("invalid 'ibm,plat-res-int-priorities' property\n");
725 		return false;
726 	}
727 
728 	/* HW supports priorities in the range [0-7] and 0xFF is a
729 	 * wildcard priority used to mask. We scan the ranges reserved
730 	 * by the hypervisor to find the lowest priority we can use.
731 	 */
732 	found = 0xFF;
733 	for (prio = 0; prio < 8; prio++) {
734 		int reserved = 0;
735 		int i;
736 
737 		for (i = 0; i < len / (2 * sizeof(u32)); i++) {
738 			int base  = be32_to_cpu(reg[2 * i]);
739 			int range = be32_to_cpu(reg[2 * i + 1]);
740 
741 			if (prio >= base && prio < base + range)
742 				reserved++;
743 		}
744 
745 		if (!reserved)
746 			found = prio;
747 	}
748 
749 	if (found == 0xFF) {
750 		pr_err("no valid priority found in 'ibm,plat-res-int-priorities'\n");
751 		return false;
752 	}
753 
754 	*max_prio = found;
755 	return true;
756 }
757 
758 static const u8 *__init get_vec5_feature(unsigned int index)
759 {
760 	unsigned long root, chosen;
761 	int size;
762 	const u8 *vec5;
763 
764 	root = of_get_flat_dt_root();
765 	chosen = of_get_flat_dt_subnode_by_name(root, "chosen");
766 	if (chosen == -FDT_ERR_NOTFOUND)
767 		return NULL;
768 
769 	vec5 = of_get_flat_dt_prop(chosen, "ibm,architecture-vec-5", &size);
770 	if (!vec5)
771 		return NULL;
772 
773 	if (size <= index)
774 		return NULL;
775 
776 	return vec5 + index;
777 }
778 
779 static bool __init xive_spapr_disabled(void)
780 {
781 	const u8 *vec5_xive;
782 
783 	vec5_xive = get_vec5_feature(OV5_INDX(OV5_XIVE_SUPPORT));
784 	if (vec5_xive) {
785 		u8 val;
786 
787 		val = *vec5_xive & OV5_FEAT(OV5_XIVE_SUPPORT);
788 		switch (val) {
789 		case OV5_FEAT(OV5_XIVE_EITHER):
790 		case OV5_FEAT(OV5_XIVE_LEGACY):
791 			break;
792 		case OV5_FEAT(OV5_XIVE_EXPLOIT):
793 			/* Hypervisor only supports XIVE */
794 			if (xive_cmdline_disabled)
795 				pr_warn("WARNING: Ignoring cmdline option xive=off\n");
796 			return false;
797 		default:
798 			pr_warn("%s: Unknown xive support option: 0x%x\n",
799 				__func__, val);
800 			break;
801 		}
802 	}
803 
804 	return xive_cmdline_disabled;
805 }
806 
807 bool __init xive_spapr_init(void)
808 {
809 	struct device_node *np;
810 	struct resource r;
811 	void __iomem *tima;
812 	struct property *prop;
813 	u8 max_prio;
814 	u32 val;
815 	u32 len;
816 	const __be32 *reg;
817 	int i, err;
818 
819 	if (xive_spapr_disabled())
820 		return false;
821 
822 	pr_devel("%s()\n", __func__);
823 	np = of_find_compatible_node(NULL, NULL, "ibm,power-ivpe");
824 	if (!np) {
825 		pr_devel("not found !\n");
826 		return false;
827 	}
828 	pr_devel("Found %s\n", np->full_name);
829 
830 	/* Resource 1 is the OS ring TIMA */
831 	if (of_address_to_resource(np, 1, &r)) {
832 		pr_err("Failed to get thread mgmnt area resource\n");
833 		return false;
834 	}
835 	tima = ioremap(r.start, resource_size(&r));
836 	if (!tima) {
837 		pr_err("Failed to map thread mgmnt area\n");
838 		return false;
839 	}
840 
841 	if (!xive_get_max_prio(&max_prio))
842 		goto err_unmap;
843 
844 	/* Feed the IRQ number allocator with the ranges given in the DT */
845 	reg = of_get_property(np, "ibm,xive-lisn-ranges", &len);
846 	if (!reg) {
847 		pr_err("Failed to read 'ibm,xive-lisn-ranges' property\n");
848 		goto err_unmap;
849 	}
850 
851 	if (len % (2 * sizeof(u32)) != 0) {
852 		pr_err("invalid 'ibm,xive-lisn-ranges' property\n");
853 		goto err_unmap;
854 	}
855 
856 	for (i = 0; i < len / (2 * sizeof(u32)); i++, reg += 2) {
857 		err = xive_irq_bitmap_add(be32_to_cpu(reg[0]),
858 					  be32_to_cpu(reg[1]));
859 		if (err < 0)
860 			goto err_mem_free;
861 	}
862 
863 	/* Iterate the EQ sizes and pick one */
864 	of_property_for_each_u32(np, "ibm,xive-eq-sizes", prop, reg, val) {
865 		xive_queue_shift = val;
866 		if (val == PAGE_SHIFT)
867 			break;
868 	}
869 
870 	/* Initialize XIVE core with our backend */
871 	if (!xive_core_init(np, &xive_spapr_ops, tima, TM_QW1_OS, max_prio))
872 		goto err_mem_free;
873 
874 	pr_info("Using %dkB queues\n", 1 << (xive_queue_shift - 10));
875 	return true;
876 
877 err_mem_free:
878 	xive_irq_bitmap_remove_all();
879 err_unmap:
880 	iounmap(tima);
881 	return false;
882 }
883 
884 machine_arch_initcall(pseries, xive_core_debug_init);
885