xref: /linux/arch/powerpc/sysdev/xive/spapr.c (revision 908fc4c2)
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/of_address.h>
15 #include <linux/of_fdt.h>
16 #include <linux/slab.h>
17 #include <linux/spinlock.h>
18 #include <linux/bitmap.h>
19 #include <linux/cpumask.h>
20 #include <linux/mm.h>
21 #include <linux/delay.h>
22 #include <linux/libfdt.h>
23 
24 #include <asm/machdep.h>
25 #include <asm/prom.h>
26 #include <asm/io.h>
27 #include <asm/smp.h>
28 #include <asm/irq.h>
29 #include <asm/errno.h>
30 #include <asm/xive.h>
31 #include <asm/xive-regs.h>
32 #include <asm/hvcall.h>
33 #include <asm/svm.h>
34 #include <asm/ultravisor.h>
35 
36 #include "xive-internal.h"
37 
38 static u32 xive_queue_shift;
39 
40 struct xive_irq_bitmap {
41 	unsigned long		*bitmap;
42 	unsigned int		base;
43 	unsigned int		count;
44 	spinlock_t		lock;
45 	struct list_head	list;
46 };
47 
48 static LIST_HEAD(xive_irq_bitmaps);
49 
50 static int __init xive_irq_bitmap_add(int base, int count)
51 {
52 	struct xive_irq_bitmap *xibm;
53 
54 	xibm = kzalloc(sizeof(*xibm), GFP_KERNEL);
55 	if (!xibm)
56 		return -ENOMEM;
57 
58 	spin_lock_init(&xibm->lock);
59 	xibm->base = base;
60 	xibm->count = count;
61 	xibm->bitmap = bitmap_zalloc(xibm->count, GFP_KERNEL);
62 	if (!xibm->bitmap) {
63 		kfree(xibm);
64 		return -ENOMEM;
65 	}
66 	list_add(&xibm->list, &xive_irq_bitmaps);
67 
68 	pr_info("Using IRQ range [%x-%x]", xibm->base,
69 		xibm->base + xibm->count - 1);
70 	return 0;
71 }
72 
73 static void xive_irq_bitmap_remove_all(void)
74 {
75 	struct xive_irq_bitmap *xibm, *tmp;
76 
77 	list_for_each_entry_safe(xibm, tmp, &xive_irq_bitmaps, list) {
78 		list_del(&xibm->list);
79 		bitmap_free(xibm->bitmap);
80 		kfree(xibm);
81 	}
82 }
83 
84 static int __xive_irq_bitmap_alloc(struct xive_irq_bitmap *xibm)
85 {
86 	int irq;
87 
88 	irq = find_first_zero_bit(xibm->bitmap, xibm->count);
89 	if (irq != xibm->count) {
90 		set_bit(irq, xibm->bitmap);
91 		irq += xibm->base;
92 	} else {
93 		irq = -ENOMEM;
94 	}
95 
96 	return irq;
97 }
98 
99 static int xive_irq_bitmap_alloc(void)
100 {
101 	struct xive_irq_bitmap *xibm;
102 	unsigned long flags;
103 	int irq = -ENOENT;
104 
105 	list_for_each_entry(xibm, &xive_irq_bitmaps, list) {
106 		spin_lock_irqsave(&xibm->lock, flags);
107 		irq = __xive_irq_bitmap_alloc(xibm);
108 		spin_unlock_irqrestore(&xibm->lock, flags);
109 		if (irq >= 0)
110 			break;
111 	}
112 	return irq;
113 }
114 
115 static void xive_irq_bitmap_free(int irq)
116 {
117 	unsigned long flags;
118 	struct xive_irq_bitmap *xibm;
119 
120 	list_for_each_entry(xibm, &xive_irq_bitmaps, list) {
121 		if ((irq >= xibm->base) && (irq < xibm->base + xibm->count)) {
122 			spin_lock_irqsave(&xibm->lock, flags);
123 			clear_bit(irq - xibm->base, xibm->bitmap);
124 			spin_unlock_irqrestore(&xibm->lock, flags);
125 			break;
126 		}
127 	}
128 }
129 
130 
131 /* Based on the similar routines in RTAS */
132 static unsigned int plpar_busy_delay_time(long rc)
133 {
134 	unsigned int ms = 0;
135 
136 	if (H_IS_LONG_BUSY(rc)) {
137 		ms = get_longbusy_msecs(rc);
138 	} else if (rc == H_BUSY) {
139 		ms = 10; /* seems appropriate for XIVE hcalls */
140 	}
141 
142 	return ms;
143 }
144 
145 static unsigned int plpar_busy_delay(int rc)
146 {
147 	unsigned int ms;
148 
149 	ms = plpar_busy_delay_time(rc);
150 	if (ms)
151 		mdelay(ms);
152 
153 	return ms;
154 }
155 
156 /*
157  * Note: this call has a partition wide scope and can take a while to
158  * complete. If it returns H_LONG_BUSY_* it should be retried
159  * periodically.
160  */
161 static long plpar_int_reset(unsigned long flags)
162 {
163 	long rc;
164 
165 	do {
166 		rc = plpar_hcall_norets(H_INT_RESET, flags);
167 	} while (plpar_busy_delay(rc));
168 
169 	if (rc)
170 		pr_err("H_INT_RESET failed %ld\n", rc);
171 
172 	return rc;
173 }
174 
175 static long plpar_int_get_source_info(unsigned long flags,
176 				      unsigned long lisn,
177 				      unsigned long *src_flags,
178 				      unsigned long *eoi_page,
179 				      unsigned long *trig_page,
180 				      unsigned long *esb_shift)
181 {
182 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
183 	long rc;
184 
185 	do {
186 		rc = plpar_hcall(H_INT_GET_SOURCE_INFO, retbuf, flags, lisn);
187 	} while (plpar_busy_delay(rc));
188 
189 	if (rc) {
190 		pr_err("H_INT_GET_SOURCE_INFO lisn=0x%lx failed %ld\n", lisn, rc);
191 		return rc;
192 	}
193 
194 	*src_flags = retbuf[0];
195 	*eoi_page  = retbuf[1];
196 	*trig_page = retbuf[2];
197 	*esb_shift = retbuf[3];
198 
199 	pr_debug("H_INT_GET_SOURCE_INFO lisn=0x%lx flags=0x%lx eoi=0x%lx trig=0x%lx shift=0x%lx\n",
200 		 lisn, retbuf[0], retbuf[1], retbuf[2], retbuf[3]);
201 
202 	return 0;
203 }
204 
205 #define XIVE_SRC_SET_EISN (1ull << (63 - 62))
206 #define XIVE_SRC_MASK     (1ull << (63 - 63)) /* unused */
207 
208 static long plpar_int_set_source_config(unsigned long flags,
209 					unsigned long lisn,
210 					unsigned long target,
211 					unsigned long prio,
212 					unsigned long sw_irq)
213 {
214 	long rc;
215 
216 
217 	pr_debug("H_INT_SET_SOURCE_CONFIG flags=0x%lx lisn=0x%lx target=%ld prio=%ld sw_irq=%ld\n",
218 		 flags, lisn, target, prio, sw_irq);
219 
220 
221 	do {
222 		rc = plpar_hcall_norets(H_INT_SET_SOURCE_CONFIG, flags, lisn,
223 					target, prio, sw_irq);
224 	} while (plpar_busy_delay(rc));
225 
226 	if (rc) {
227 		pr_err("H_INT_SET_SOURCE_CONFIG lisn=0x%lx target=%ld prio=%ld failed %ld\n",
228 		       lisn, target, prio, rc);
229 		return rc;
230 	}
231 
232 	return 0;
233 }
234 
235 static long plpar_int_get_source_config(unsigned long flags,
236 					unsigned long lisn,
237 					unsigned long *target,
238 					unsigned long *prio,
239 					unsigned long *sw_irq)
240 {
241 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
242 	long rc;
243 
244 	pr_debug("H_INT_GET_SOURCE_CONFIG flags=0x%lx lisn=0x%lx\n", flags, lisn);
245 
246 	do {
247 		rc = plpar_hcall(H_INT_GET_SOURCE_CONFIG, retbuf, flags, lisn,
248 				 target, prio, sw_irq);
249 	} while (plpar_busy_delay(rc));
250 
251 	if (rc) {
252 		pr_err("H_INT_GET_SOURCE_CONFIG lisn=0x%lx failed %ld\n",
253 		       lisn, rc);
254 		return rc;
255 	}
256 
257 	*target = retbuf[0];
258 	*prio   = retbuf[1];
259 	*sw_irq = retbuf[2];
260 
261 	pr_debug("H_INT_GET_SOURCE_CONFIG target=%ld prio=%ld sw_irq=%ld\n",
262 		 retbuf[0], retbuf[1], retbuf[2]);
263 
264 	return 0;
265 }
266 
267 static long plpar_int_get_queue_info(unsigned long flags,
268 				     unsigned long target,
269 				     unsigned long priority,
270 				     unsigned long *esn_page,
271 				     unsigned long *esn_size)
272 {
273 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
274 	long rc;
275 
276 	do {
277 		rc = plpar_hcall(H_INT_GET_QUEUE_INFO, retbuf, flags, target,
278 				 priority);
279 	} while (plpar_busy_delay(rc));
280 
281 	if (rc) {
282 		pr_err("H_INT_GET_QUEUE_INFO cpu=%ld prio=%ld failed %ld\n",
283 		       target, priority, rc);
284 		return rc;
285 	}
286 
287 	*esn_page = retbuf[0];
288 	*esn_size = retbuf[1];
289 
290 	pr_debug("H_INT_GET_QUEUE_INFO cpu=%ld prio=%ld page=0x%lx size=0x%lx\n",
291 		 target, priority, retbuf[0], retbuf[1]);
292 
293 	return 0;
294 }
295 
296 #define XIVE_EQ_ALWAYS_NOTIFY (1ull << (63 - 63))
297 
298 static long plpar_int_set_queue_config(unsigned long flags,
299 				       unsigned long target,
300 				       unsigned long priority,
301 				       unsigned long qpage,
302 				       unsigned long qsize)
303 {
304 	long rc;
305 
306 	pr_debug("H_INT_SET_QUEUE_CONFIG flags=0x%lx target=%ld priority=0x%lx qpage=0x%lx qsize=0x%lx\n",
307 		 flags,  target, priority, qpage, qsize);
308 
309 	do {
310 		rc = plpar_hcall_norets(H_INT_SET_QUEUE_CONFIG, flags, target,
311 					priority, qpage, qsize);
312 	} while (plpar_busy_delay(rc));
313 
314 	if (rc) {
315 		pr_err("H_INT_SET_QUEUE_CONFIG cpu=%ld prio=%ld qpage=0x%lx returned %ld\n",
316 		       target, priority, qpage, rc);
317 		return  rc;
318 	}
319 
320 	return 0;
321 }
322 
323 static long plpar_int_sync(unsigned long flags, unsigned long lisn)
324 {
325 	long rc;
326 
327 	do {
328 		rc = plpar_hcall_norets(H_INT_SYNC, flags, lisn);
329 	} while (plpar_busy_delay(rc));
330 
331 	if (rc) {
332 		pr_err("H_INT_SYNC lisn=0x%lx returned %ld\n", lisn, rc);
333 		return  rc;
334 	}
335 
336 	return 0;
337 }
338 
339 #define XIVE_ESB_FLAG_STORE (1ull << (63 - 63))
340 
341 static long plpar_int_esb(unsigned long flags,
342 			  unsigned long lisn,
343 			  unsigned long offset,
344 			  unsigned long in_data,
345 			  unsigned long *out_data)
346 {
347 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
348 	long rc;
349 
350 	pr_debug("H_INT_ESB flags=0x%lx lisn=0x%lx offset=0x%lx in=0x%lx\n",
351 		 flags,  lisn, offset, in_data);
352 
353 	do {
354 		rc = plpar_hcall(H_INT_ESB, retbuf, flags, lisn, offset,
355 				 in_data);
356 	} while (plpar_busy_delay(rc));
357 
358 	if (rc) {
359 		pr_err("H_INT_ESB lisn=0x%lx offset=0x%lx returned %ld\n",
360 		       lisn, offset, rc);
361 		return  rc;
362 	}
363 
364 	*out_data = retbuf[0];
365 
366 	return 0;
367 }
368 
369 static u64 xive_spapr_esb_rw(u32 lisn, u32 offset, u64 data, bool write)
370 {
371 	unsigned long read_data;
372 	long rc;
373 
374 	rc = plpar_int_esb(write ? XIVE_ESB_FLAG_STORE : 0,
375 			   lisn, offset, data, &read_data);
376 	if (rc)
377 		return -1;
378 
379 	return write ? 0 : read_data;
380 }
381 
382 #define XIVE_SRC_H_INT_ESB     (1ull << (63 - 60))
383 #define XIVE_SRC_LSI           (1ull << (63 - 61))
384 #define XIVE_SRC_TRIGGER       (1ull << (63 - 62))
385 #define XIVE_SRC_STORE_EOI     (1ull << (63 - 63))
386 
387 static int xive_spapr_populate_irq_data(u32 hw_irq, struct xive_irq_data *data)
388 {
389 	long rc;
390 	unsigned long flags;
391 	unsigned long eoi_page;
392 	unsigned long trig_page;
393 	unsigned long esb_shift;
394 
395 	memset(data, 0, sizeof(*data));
396 
397 	rc = plpar_int_get_source_info(0, hw_irq, &flags, &eoi_page, &trig_page,
398 				       &esb_shift);
399 	if (rc)
400 		return  -EINVAL;
401 
402 	if (flags & XIVE_SRC_H_INT_ESB)
403 		data->flags  |= XIVE_IRQ_FLAG_H_INT_ESB;
404 	if (flags & XIVE_SRC_STORE_EOI)
405 		data->flags  |= XIVE_IRQ_FLAG_STORE_EOI;
406 	if (flags & XIVE_SRC_LSI)
407 		data->flags  |= XIVE_IRQ_FLAG_LSI;
408 	data->eoi_page  = eoi_page;
409 	data->esb_shift = esb_shift;
410 	data->trig_page = trig_page;
411 
412 	data->hw_irq = hw_irq;
413 
414 	/*
415 	 * No chip-id for the sPAPR backend. This has an impact how we
416 	 * pick a target. See xive_pick_irq_target().
417 	 */
418 	data->src_chip = XIVE_INVALID_CHIP_ID;
419 
420 	/*
421 	 * When the H_INT_ESB flag is set, the H_INT_ESB hcall should
422 	 * be used for interrupt management. Skip the remapping of the
423 	 * ESB pages which are not available.
424 	 */
425 	if (data->flags & XIVE_IRQ_FLAG_H_INT_ESB)
426 		return 0;
427 
428 	data->eoi_mmio = ioremap(data->eoi_page, 1u << data->esb_shift);
429 	if (!data->eoi_mmio) {
430 		pr_err("Failed to map EOI page for irq 0x%x\n", hw_irq);
431 		return -ENOMEM;
432 	}
433 
434 	/* Full function page supports trigger */
435 	if (flags & XIVE_SRC_TRIGGER) {
436 		data->trig_mmio = data->eoi_mmio;
437 		return 0;
438 	}
439 
440 	data->trig_mmio = ioremap(data->trig_page, 1u << data->esb_shift);
441 	if (!data->trig_mmio) {
442 		pr_err("Failed to map trigger page for irq 0x%x\n", hw_irq);
443 		return -ENOMEM;
444 	}
445 	return 0;
446 }
447 
448 static int xive_spapr_configure_irq(u32 hw_irq, u32 target, u8 prio, u32 sw_irq)
449 {
450 	long rc;
451 
452 	rc = plpar_int_set_source_config(XIVE_SRC_SET_EISN, hw_irq, target,
453 					 prio, sw_irq);
454 
455 	return rc == 0 ? 0 : -ENXIO;
456 }
457 
458 static int xive_spapr_get_irq_config(u32 hw_irq, u32 *target, u8 *prio,
459 				     u32 *sw_irq)
460 {
461 	long rc;
462 	unsigned long h_target;
463 	unsigned long h_prio;
464 	unsigned long h_sw_irq;
465 
466 	rc = plpar_int_get_source_config(0, hw_irq, &h_target, &h_prio,
467 					 &h_sw_irq);
468 
469 	*target = h_target;
470 	*prio = h_prio;
471 	*sw_irq = h_sw_irq;
472 
473 	return rc == 0 ? 0 : -ENXIO;
474 }
475 
476 /* This can be called multiple time to change a queue configuration */
477 static int xive_spapr_configure_queue(u32 target, struct xive_q *q, u8 prio,
478 				   __be32 *qpage, u32 order)
479 {
480 	s64 rc = 0;
481 	unsigned long esn_page;
482 	unsigned long esn_size;
483 	u64 flags, qpage_phys;
484 
485 	/* If there's an actual queue page, clean it */
486 	if (order) {
487 		if (WARN_ON(!qpage))
488 			return -EINVAL;
489 		qpage_phys = __pa(qpage);
490 	} else {
491 		qpage_phys = 0;
492 	}
493 
494 	/* Initialize the rest of the fields */
495 	q->msk = order ? ((1u << (order - 2)) - 1) : 0;
496 	q->idx = 0;
497 	q->toggle = 0;
498 
499 	rc = plpar_int_get_queue_info(0, target, prio, &esn_page, &esn_size);
500 	if (rc) {
501 		pr_err("Error %lld getting queue info CPU %d prio %d\n", rc,
502 		       target, prio);
503 		rc = -EIO;
504 		goto fail;
505 	}
506 
507 	/* TODO: add support for the notification page */
508 	q->eoi_phys = esn_page;
509 
510 	/* Default is to always notify */
511 	flags = XIVE_EQ_ALWAYS_NOTIFY;
512 
513 	/* Configure and enable the queue in HW */
514 	rc = plpar_int_set_queue_config(flags, target, prio, qpage_phys, order);
515 	if (rc) {
516 		pr_err("Error %lld setting queue for CPU %d prio %d\n", rc,
517 		       target, prio);
518 		rc = -EIO;
519 	} else {
520 		q->qpage = qpage;
521 		if (is_secure_guest())
522 			uv_share_page(PHYS_PFN(qpage_phys),
523 					1 << xive_alloc_order(order));
524 	}
525 fail:
526 	return rc;
527 }
528 
529 static int xive_spapr_setup_queue(unsigned int cpu, struct xive_cpu *xc,
530 				  u8 prio)
531 {
532 	struct xive_q *q = &xc->queue[prio];
533 	__be32 *qpage;
534 
535 	qpage = xive_queue_page_alloc(cpu, xive_queue_shift);
536 	if (IS_ERR(qpage))
537 		return PTR_ERR(qpage);
538 
539 	return xive_spapr_configure_queue(get_hard_smp_processor_id(cpu),
540 					  q, prio, qpage, xive_queue_shift);
541 }
542 
543 static void xive_spapr_cleanup_queue(unsigned int cpu, struct xive_cpu *xc,
544 				  u8 prio)
545 {
546 	struct xive_q *q = &xc->queue[prio];
547 	unsigned int alloc_order;
548 	long rc;
549 	int hw_cpu = get_hard_smp_processor_id(cpu);
550 
551 	rc = plpar_int_set_queue_config(0, hw_cpu, prio, 0, 0);
552 	if (rc)
553 		pr_err("Error %ld setting queue for CPU %d prio %d\n", rc,
554 		       hw_cpu, prio);
555 
556 	alloc_order = xive_alloc_order(xive_queue_shift);
557 	if (is_secure_guest())
558 		uv_unshare_page(PHYS_PFN(__pa(q->qpage)), 1 << alloc_order);
559 	free_pages((unsigned long)q->qpage, alloc_order);
560 	q->qpage = NULL;
561 }
562 
563 static bool xive_spapr_match(struct device_node *node)
564 {
565 	/* Ignore cascaded controllers for the moment */
566 	return true;
567 }
568 
569 #ifdef CONFIG_SMP
570 static int xive_spapr_get_ipi(unsigned int cpu, struct xive_cpu *xc)
571 {
572 	int irq = xive_irq_bitmap_alloc();
573 
574 	if (irq < 0) {
575 		pr_err("Failed to allocate IPI on CPU %d\n", cpu);
576 		return -ENXIO;
577 	}
578 
579 	xc->hw_ipi = irq;
580 	return 0;
581 }
582 
583 static void xive_spapr_put_ipi(unsigned int cpu, struct xive_cpu *xc)
584 {
585 	if (xc->hw_ipi == XIVE_BAD_IRQ)
586 		return;
587 
588 	xive_irq_bitmap_free(xc->hw_ipi);
589 	xc->hw_ipi = XIVE_BAD_IRQ;
590 }
591 #endif /* CONFIG_SMP */
592 
593 static void xive_spapr_shutdown(void)
594 {
595 	plpar_int_reset(0);
596 }
597 
598 /*
599  * Perform an "ack" cycle on the current thread. Grab the pending
600  * active priorities and update the CPPR to the most favored one.
601  */
602 static void xive_spapr_update_pending(struct xive_cpu *xc)
603 {
604 	u8 nsr, cppr;
605 	u16 ack;
606 
607 	/*
608 	 * Perform the "Acknowledge O/S to Register" cycle.
609 	 *
610 	 * Let's speedup the access to the TIMA using the raw I/O
611 	 * accessor as we don't need the synchronisation routine of
612 	 * the higher level ones
613 	 */
614 	ack = be16_to_cpu(__raw_readw(xive_tima + TM_SPC_ACK_OS_REG));
615 
616 	/* Synchronize subsequent queue accesses */
617 	mb();
618 
619 	/*
620 	 * Grab the CPPR and the "NSR" field which indicates the source
621 	 * of the interrupt (if any)
622 	 */
623 	cppr = ack & 0xff;
624 	nsr = ack >> 8;
625 
626 	if (nsr & TM_QW1_NSR_EO) {
627 		if (cppr == 0xff)
628 			return;
629 		/* Mark the priority pending */
630 		xc->pending_prio |= 1 << cppr;
631 
632 		/*
633 		 * A new interrupt should never have a CPPR less favored
634 		 * than our current one.
635 		 */
636 		if (cppr >= xc->cppr)
637 			pr_err("CPU %d odd ack CPPR, got %d at %d\n",
638 			       smp_processor_id(), cppr, xc->cppr);
639 
640 		/* Update our idea of what the CPPR is */
641 		xc->cppr = cppr;
642 	}
643 }
644 
645 static void xive_spapr_setup_cpu(unsigned int cpu, struct xive_cpu *xc)
646 {
647 	/* Only some debug on the TIMA settings */
648 	pr_debug("(HW value: %08x %08x %08x)\n",
649 		 in_be32(xive_tima + TM_QW1_OS + TM_WORD0),
650 		 in_be32(xive_tima + TM_QW1_OS + TM_WORD1),
651 		 in_be32(xive_tima + TM_QW1_OS + TM_WORD2));
652 }
653 
654 static void xive_spapr_teardown_cpu(unsigned int cpu, struct xive_cpu *xc)
655 {
656 	/* Nothing to do */;
657 }
658 
659 static void xive_spapr_sync_source(u32 hw_irq)
660 {
661 	/* Specs are unclear on what this is doing */
662 	plpar_int_sync(0, hw_irq);
663 }
664 
665 static int xive_spapr_debug_show(struct seq_file *m, void *private)
666 {
667 	struct xive_irq_bitmap *xibm;
668 	char *buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
669 
670 	if (!buf)
671 		return -ENOMEM;
672 
673 	list_for_each_entry(xibm, &xive_irq_bitmaps, list) {
674 		memset(buf, 0, PAGE_SIZE);
675 		bitmap_print_to_pagebuf(true, buf, xibm->bitmap, xibm->count);
676 		seq_printf(m, "bitmap #%d: %s", xibm->count, buf);
677 	}
678 	kfree(buf);
679 
680 	return 0;
681 }
682 
683 static const struct xive_ops xive_spapr_ops = {
684 	.populate_irq_data	= xive_spapr_populate_irq_data,
685 	.configure_irq		= xive_spapr_configure_irq,
686 	.get_irq_config		= xive_spapr_get_irq_config,
687 	.setup_queue		= xive_spapr_setup_queue,
688 	.cleanup_queue		= xive_spapr_cleanup_queue,
689 	.match			= xive_spapr_match,
690 	.shutdown		= xive_spapr_shutdown,
691 	.update_pending		= xive_spapr_update_pending,
692 	.setup_cpu		= xive_spapr_setup_cpu,
693 	.teardown_cpu		= xive_spapr_teardown_cpu,
694 	.sync_source		= xive_spapr_sync_source,
695 	.esb_rw			= xive_spapr_esb_rw,
696 #ifdef CONFIG_SMP
697 	.get_ipi		= xive_spapr_get_ipi,
698 	.put_ipi		= xive_spapr_put_ipi,
699 	.debug_show		= xive_spapr_debug_show,
700 #endif /* CONFIG_SMP */
701 	.name			= "spapr",
702 };
703 
704 /*
705  * get max priority from "/ibm,plat-res-int-priorities"
706  */
707 static bool __init xive_get_max_prio(u8 *max_prio)
708 {
709 	struct device_node *rootdn;
710 	const __be32 *reg;
711 	u32 len;
712 	int prio, found;
713 
714 	rootdn = of_find_node_by_path("/");
715 	if (!rootdn) {
716 		pr_err("not root node found !\n");
717 		return false;
718 	}
719 
720 	reg = of_get_property(rootdn, "ibm,plat-res-int-priorities", &len);
721 	if (!reg) {
722 		pr_err("Failed to read 'ibm,plat-res-int-priorities' property\n");
723 		return false;
724 	}
725 
726 	if (len % (2 * sizeof(u32)) != 0) {
727 		pr_err("invalid 'ibm,plat-res-int-priorities' property\n");
728 		return false;
729 	}
730 
731 	/* HW supports priorities in the range [0-7] and 0xFF is a
732 	 * wildcard priority used to mask. We scan the ranges reserved
733 	 * by the hypervisor to find the lowest priority we can use.
734 	 */
735 	found = 0xFF;
736 	for (prio = 0; prio < 8; prio++) {
737 		int reserved = 0;
738 		int i;
739 
740 		for (i = 0; i < len / (2 * sizeof(u32)); i++) {
741 			int base  = be32_to_cpu(reg[2 * i]);
742 			int range = be32_to_cpu(reg[2 * i + 1]);
743 
744 			if (prio >= base && prio < base + range)
745 				reserved++;
746 		}
747 
748 		if (!reserved)
749 			found = prio;
750 	}
751 
752 	if (found == 0xFF) {
753 		pr_err("no valid priority found in 'ibm,plat-res-int-priorities'\n");
754 		return false;
755 	}
756 
757 	*max_prio = found;
758 	return true;
759 }
760 
761 static const u8 *__init get_vec5_feature(unsigned int index)
762 {
763 	unsigned long root, chosen;
764 	int size;
765 	const u8 *vec5;
766 
767 	root = of_get_flat_dt_root();
768 	chosen = of_get_flat_dt_subnode_by_name(root, "chosen");
769 	if (chosen == -FDT_ERR_NOTFOUND)
770 		return NULL;
771 
772 	vec5 = of_get_flat_dt_prop(chosen, "ibm,architecture-vec-5", &size);
773 	if (!vec5)
774 		return NULL;
775 
776 	if (size <= index)
777 		return NULL;
778 
779 	return vec5 + index;
780 }
781 
782 static bool __init xive_spapr_disabled(void)
783 {
784 	const u8 *vec5_xive;
785 
786 	vec5_xive = get_vec5_feature(OV5_INDX(OV5_XIVE_SUPPORT));
787 	if (vec5_xive) {
788 		u8 val;
789 
790 		val = *vec5_xive & OV5_FEAT(OV5_XIVE_SUPPORT);
791 		switch (val) {
792 		case OV5_FEAT(OV5_XIVE_EITHER):
793 		case OV5_FEAT(OV5_XIVE_LEGACY):
794 			break;
795 		case OV5_FEAT(OV5_XIVE_EXPLOIT):
796 			/* Hypervisor only supports XIVE */
797 			if (xive_cmdline_disabled)
798 				pr_warn("WARNING: Ignoring cmdline option xive=off\n");
799 			return false;
800 		default:
801 			pr_warn("%s: Unknown xive support option: 0x%x\n",
802 				__func__, val);
803 			break;
804 		}
805 	}
806 
807 	return xive_cmdline_disabled;
808 }
809 
810 bool __init xive_spapr_init(void)
811 {
812 	struct device_node *np;
813 	struct resource r;
814 	void __iomem *tima;
815 	struct property *prop;
816 	u8 max_prio;
817 	u32 val;
818 	u32 len;
819 	const __be32 *reg;
820 	int i, err;
821 
822 	if (xive_spapr_disabled())
823 		return false;
824 
825 	pr_devel("%s()\n", __func__);
826 	np = of_find_compatible_node(NULL, NULL, "ibm,power-ivpe");
827 	if (!np) {
828 		pr_devel("not found !\n");
829 		return false;
830 	}
831 	pr_devel("Found %s\n", np->full_name);
832 
833 	/* Resource 1 is the OS ring TIMA */
834 	if (of_address_to_resource(np, 1, &r)) {
835 		pr_err("Failed to get thread mgmnt area resource\n");
836 		goto err_put;
837 	}
838 	tima = ioremap(r.start, resource_size(&r));
839 	if (!tima) {
840 		pr_err("Failed to map thread mgmnt area\n");
841 		goto err_put;
842 	}
843 
844 	if (!xive_get_max_prio(&max_prio))
845 		goto err_unmap;
846 
847 	/* Feed the IRQ number allocator with the ranges given in the DT */
848 	reg = of_get_property(np, "ibm,xive-lisn-ranges", &len);
849 	if (!reg) {
850 		pr_err("Failed to read 'ibm,xive-lisn-ranges' property\n");
851 		goto err_unmap;
852 	}
853 
854 	if (len % (2 * sizeof(u32)) != 0) {
855 		pr_err("invalid 'ibm,xive-lisn-ranges' property\n");
856 		goto err_unmap;
857 	}
858 
859 	for (i = 0; i < len / (2 * sizeof(u32)); i++, reg += 2) {
860 		err = xive_irq_bitmap_add(be32_to_cpu(reg[0]),
861 					  be32_to_cpu(reg[1]));
862 		if (err < 0)
863 			goto err_mem_free;
864 	}
865 
866 	/* Iterate the EQ sizes and pick one */
867 	of_property_for_each_u32(np, "ibm,xive-eq-sizes", prop, reg, val) {
868 		xive_queue_shift = val;
869 		if (val == PAGE_SHIFT)
870 			break;
871 	}
872 
873 	/* Initialize XIVE core with our backend */
874 	if (!xive_core_init(np, &xive_spapr_ops, tima, TM_QW1_OS, max_prio))
875 		goto err_mem_free;
876 
877 	of_node_put(np);
878 	pr_info("Using %dkB queues\n", 1 << (xive_queue_shift - 10));
879 	return true;
880 
881 err_mem_free:
882 	xive_irq_bitmap_remove_all();
883 err_unmap:
884 	iounmap(tima);
885 err_put:
886 	of_node_put(np);
887 	return false;
888 }
889 
890 machine_arch_initcall(pseries, xive_core_debug_init);
891