xref: /openbmc/linux/arch/powerpc/sysdev/xive/spapr.c (revision c4e78957)
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/cpumask.h>
19 #include <linux/mm.h>
20 #include <linux/delay.h>
21 #include <linux/libfdt.h>
22 
23 #include <asm/machdep.h>
24 #include <asm/prom.h>
25 #include <asm/io.h>
26 #include <asm/smp.h>
27 #include <asm/irq.h>
28 #include <asm/errno.h>
29 #include <asm/xive.h>
30 #include <asm/xive-regs.h>
31 #include <asm/hvcall.h>
32 #include <asm/svm.h>
33 #include <asm/ultravisor.h>
34 
35 #include "xive-internal.h"
36 
37 static u32 xive_queue_shift;
38 
39 struct xive_irq_bitmap {
40 	unsigned long		*bitmap;
41 	unsigned int		base;
42 	unsigned int		count;
43 	spinlock_t		lock;
44 	struct list_head	list;
45 };
46 
47 static LIST_HEAD(xive_irq_bitmaps);
48 
49 static int __init xive_irq_bitmap_add(int base, int count)
50 {
51 	struct xive_irq_bitmap *xibm;
52 
53 	xibm = kzalloc(sizeof(*xibm), GFP_KERNEL);
54 	if (!xibm)
55 		return -ENOMEM;
56 
57 	spin_lock_init(&xibm->lock);
58 	xibm->base = base;
59 	xibm->count = count;
60 	xibm->bitmap = kzalloc(xibm->count, GFP_KERNEL);
61 	if (!xibm->bitmap) {
62 		kfree(xibm);
63 		return -ENOMEM;
64 	}
65 	list_add(&xibm->list, &xive_irq_bitmaps);
66 
67 	pr_info("Using IRQ range [%x-%x]", xibm->base,
68 		xibm->base + xibm->count - 1);
69 	return 0;
70 }
71 
72 static void xive_irq_bitmap_remove_all(void)
73 {
74 	struct xive_irq_bitmap *xibm, *tmp;
75 
76 	list_for_each_entry_safe(xibm, tmp, &xive_irq_bitmaps, list) {
77 		list_del(&xibm->list);
78 		kfree(xibm->bitmap);
79 		kfree(xibm);
80 	}
81 }
82 
83 static int __xive_irq_bitmap_alloc(struct xive_irq_bitmap *xibm)
84 {
85 	int irq;
86 
87 	irq = find_first_zero_bit(xibm->bitmap, xibm->count);
88 	if (irq != xibm->count) {
89 		set_bit(irq, xibm->bitmap);
90 		irq += xibm->base;
91 	} else {
92 		irq = -ENOMEM;
93 	}
94 
95 	return irq;
96 }
97 
98 static int xive_irq_bitmap_alloc(void)
99 {
100 	struct xive_irq_bitmap *xibm;
101 	unsigned long flags;
102 	int irq = -ENOENT;
103 
104 	list_for_each_entry(xibm, &xive_irq_bitmaps, list) {
105 		spin_lock_irqsave(&xibm->lock, flags);
106 		irq = __xive_irq_bitmap_alloc(xibm);
107 		spin_unlock_irqrestore(&xibm->lock, flags);
108 		if (irq >= 0)
109 			break;
110 	}
111 	return irq;
112 }
113 
114 static void xive_irq_bitmap_free(int irq)
115 {
116 	unsigned long flags;
117 	struct xive_irq_bitmap *xibm;
118 
119 	list_for_each_entry(xibm, &xive_irq_bitmaps, list) {
120 		if ((irq >= xibm->base) && (irq < xibm->base + xibm->count)) {
121 			spin_lock_irqsave(&xibm->lock, flags);
122 			clear_bit(irq - xibm->base, xibm->bitmap);
123 			spin_unlock_irqrestore(&xibm->lock, flags);
124 			break;
125 		}
126 	}
127 }
128 
129 
130 /* Based on the similar routines in RTAS */
131 static unsigned int plpar_busy_delay_time(long rc)
132 {
133 	unsigned int ms = 0;
134 
135 	if (H_IS_LONG_BUSY(rc)) {
136 		ms = get_longbusy_msecs(rc);
137 	} else if (rc == H_BUSY) {
138 		ms = 10; /* seems appropriate for XIVE hcalls */
139 	}
140 
141 	return ms;
142 }
143 
144 static unsigned int plpar_busy_delay(int rc)
145 {
146 	unsigned int ms;
147 
148 	ms = plpar_busy_delay_time(rc);
149 	if (ms)
150 		mdelay(ms);
151 
152 	return ms;
153 }
154 
155 /*
156  * Note: this call has a partition wide scope and can take a while to
157  * complete. If it returns H_LONG_BUSY_* it should be retried
158  * periodically.
159  */
160 static long plpar_int_reset(unsigned long flags)
161 {
162 	long rc;
163 
164 	do {
165 		rc = plpar_hcall_norets(H_INT_RESET, flags);
166 	} while (plpar_busy_delay(rc));
167 
168 	if (rc)
169 		pr_err("H_INT_RESET failed %ld\n", rc);
170 
171 	return rc;
172 }
173 
174 static long plpar_int_get_source_info(unsigned long flags,
175 				      unsigned long lisn,
176 				      unsigned long *src_flags,
177 				      unsigned long *eoi_page,
178 				      unsigned long *trig_page,
179 				      unsigned long *esb_shift)
180 {
181 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
182 	long rc;
183 
184 	do {
185 		rc = plpar_hcall(H_INT_GET_SOURCE_INFO, retbuf, flags, lisn);
186 	} while (plpar_busy_delay(rc));
187 
188 	if (rc) {
189 		pr_err("H_INT_GET_SOURCE_INFO lisn=0x%lx failed %ld\n", lisn, rc);
190 		return rc;
191 	}
192 
193 	*src_flags = retbuf[0];
194 	*eoi_page  = retbuf[1];
195 	*trig_page = retbuf[2];
196 	*esb_shift = retbuf[3];
197 
198 	pr_debug("H_INT_GET_SOURCE_INFO lisn=0x%lx flags=0x%lx eoi=0x%lx trig=0x%lx shift=0x%lx\n",
199 		 lisn, retbuf[0], retbuf[1], retbuf[2], retbuf[3]);
200 
201 	return 0;
202 }
203 
204 #define XIVE_SRC_SET_EISN (1ull << (63 - 62))
205 #define XIVE_SRC_MASK     (1ull << (63 - 63)) /* unused */
206 
207 static long plpar_int_set_source_config(unsigned long flags,
208 					unsigned long lisn,
209 					unsigned long target,
210 					unsigned long prio,
211 					unsigned long sw_irq)
212 {
213 	long rc;
214 
215 
216 	pr_debug("H_INT_SET_SOURCE_CONFIG flags=0x%lx lisn=0x%lx target=%ld prio=%ld sw_irq=%ld\n",
217 		 flags, lisn, target, prio, sw_irq);
218 
219 
220 	do {
221 		rc = plpar_hcall_norets(H_INT_SET_SOURCE_CONFIG, flags, lisn,
222 					target, prio, sw_irq);
223 	} while (plpar_busy_delay(rc));
224 
225 	if (rc) {
226 		pr_err("H_INT_SET_SOURCE_CONFIG lisn=0x%lx target=%ld prio=%ld failed %ld\n",
227 		       lisn, target, prio, rc);
228 		return rc;
229 	}
230 
231 	return 0;
232 }
233 
234 static long plpar_int_get_source_config(unsigned long flags,
235 					unsigned long lisn,
236 					unsigned long *target,
237 					unsigned long *prio,
238 					unsigned long *sw_irq)
239 {
240 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
241 	long rc;
242 
243 	pr_debug("H_INT_GET_SOURCE_CONFIG flags=0x%lx lisn=0x%lx\n", flags, lisn);
244 
245 	do {
246 		rc = plpar_hcall(H_INT_GET_SOURCE_CONFIG, retbuf, flags, lisn,
247 				 target, prio, sw_irq);
248 	} while (plpar_busy_delay(rc));
249 
250 	if (rc) {
251 		pr_err("H_INT_GET_SOURCE_CONFIG lisn=0x%lx failed %ld\n",
252 		       lisn, rc);
253 		return rc;
254 	}
255 
256 	*target = retbuf[0];
257 	*prio   = retbuf[1];
258 	*sw_irq = retbuf[2];
259 
260 	pr_debug("H_INT_GET_SOURCE_CONFIG target=%ld prio=%ld sw_irq=%ld\n",
261 		 retbuf[0], retbuf[1], retbuf[2]);
262 
263 	return 0;
264 }
265 
266 static long plpar_int_get_queue_info(unsigned long flags,
267 				     unsigned long target,
268 				     unsigned long priority,
269 				     unsigned long *esn_page,
270 				     unsigned long *esn_size)
271 {
272 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
273 	long rc;
274 
275 	do {
276 		rc = plpar_hcall(H_INT_GET_QUEUE_INFO, retbuf, flags, target,
277 				 priority);
278 	} while (plpar_busy_delay(rc));
279 
280 	if (rc) {
281 		pr_err("H_INT_GET_QUEUE_INFO cpu=%ld prio=%ld failed %ld\n",
282 		       target, priority, rc);
283 		return rc;
284 	}
285 
286 	*esn_page = retbuf[0];
287 	*esn_size = retbuf[1];
288 
289 	pr_debug("H_INT_GET_QUEUE_INFO cpu=%ld prio=%ld page=0x%lx size=0x%lx\n",
290 		 target, priority, retbuf[0], retbuf[1]);
291 
292 	return 0;
293 }
294 
295 #define XIVE_EQ_ALWAYS_NOTIFY (1ull << (63 - 63))
296 
297 static long plpar_int_set_queue_config(unsigned long flags,
298 				       unsigned long target,
299 				       unsigned long priority,
300 				       unsigned long qpage,
301 				       unsigned long qsize)
302 {
303 	long rc;
304 
305 	pr_debug("H_INT_SET_QUEUE_CONFIG flags=0x%lx target=%ld priority=0x%lx qpage=0x%lx qsize=0x%lx\n",
306 		 flags,  target, priority, qpage, qsize);
307 
308 	do {
309 		rc = plpar_hcall_norets(H_INT_SET_QUEUE_CONFIG, flags, target,
310 					priority, qpage, qsize);
311 	} while (plpar_busy_delay(rc));
312 
313 	if (rc) {
314 		pr_err("H_INT_SET_QUEUE_CONFIG cpu=%ld prio=%ld qpage=0x%lx returned %ld\n",
315 		       target, priority, qpage, rc);
316 		return  rc;
317 	}
318 
319 	return 0;
320 }
321 
322 static long plpar_int_sync(unsigned long flags, unsigned long lisn)
323 {
324 	long rc;
325 
326 	do {
327 		rc = plpar_hcall_norets(H_INT_SYNC, flags, lisn);
328 	} while (plpar_busy_delay(rc));
329 
330 	if (rc) {
331 		pr_err("H_INT_SYNC lisn=0x%lx returned %ld\n", lisn, rc);
332 		return  rc;
333 	}
334 
335 	return 0;
336 }
337 
338 #define XIVE_ESB_FLAG_STORE (1ull << (63 - 63))
339 
340 static long plpar_int_esb(unsigned long flags,
341 			  unsigned long lisn,
342 			  unsigned long offset,
343 			  unsigned long in_data,
344 			  unsigned long *out_data)
345 {
346 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
347 	long rc;
348 
349 	pr_debug("H_INT_ESB flags=0x%lx lisn=0x%lx offset=0x%lx in=0x%lx\n",
350 		 flags,  lisn, offset, in_data);
351 
352 	do {
353 		rc = plpar_hcall(H_INT_ESB, retbuf, flags, lisn, offset,
354 				 in_data);
355 	} while (plpar_busy_delay(rc));
356 
357 	if (rc) {
358 		pr_err("H_INT_ESB lisn=0x%lx offset=0x%lx returned %ld\n",
359 		       lisn, offset, rc);
360 		return  rc;
361 	}
362 
363 	*out_data = retbuf[0];
364 
365 	return 0;
366 }
367 
368 static u64 xive_spapr_esb_rw(u32 lisn, u32 offset, u64 data, bool write)
369 {
370 	unsigned long read_data;
371 	long rc;
372 
373 	rc = plpar_int_esb(write ? XIVE_ESB_FLAG_STORE : 0,
374 			   lisn, offset, data, &read_data);
375 	if (rc)
376 		return -1;
377 
378 	return write ? 0 : read_data;
379 }
380 
381 #define XIVE_SRC_H_INT_ESB     (1ull << (63 - 60))
382 #define XIVE_SRC_LSI           (1ull << (63 - 61))
383 #define XIVE_SRC_TRIGGER       (1ull << (63 - 62))
384 #define XIVE_SRC_STORE_EOI     (1ull << (63 - 63))
385 
386 static int xive_spapr_populate_irq_data(u32 hw_irq, struct xive_irq_data *data)
387 {
388 	long rc;
389 	unsigned long flags;
390 	unsigned long eoi_page;
391 	unsigned long trig_page;
392 	unsigned long esb_shift;
393 
394 	memset(data, 0, sizeof(*data));
395 
396 	rc = plpar_int_get_source_info(0, hw_irq, &flags, &eoi_page, &trig_page,
397 				       &esb_shift);
398 	if (rc)
399 		return  -EINVAL;
400 
401 	if (flags & XIVE_SRC_H_INT_ESB)
402 		data->flags  |= XIVE_IRQ_FLAG_H_INT_ESB;
403 	if (flags & XIVE_SRC_STORE_EOI)
404 		data->flags  |= XIVE_IRQ_FLAG_STORE_EOI;
405 	if (flags & XIVE_SRC_LSI)
406 		data->flags  |= XIVE_IRQ_FLAG_LSI;
407 	data->eoi_page  = eoi_page;
408 	data->esb_shift = esb_shift;
409 	data->trig_page = trig_page;
410 
411 	data->hw_irq = hw_irq;
412 
413 	/*
414 	 * No chip-id for the sPAPR backend. This has an impact how we
415 	 * pick a target. See xive_pick_irq_target().
416 	 */
417 	data->src_chip = XIVE_INVALID_CHIP_ID;
418 
419 	/*
420 	 * When the H_INT_ESB flag is set, the H_INT_ESB hcall should
421 	 * be used for interrupt management. Skip the remapping of the
422 	 * ESB pages which are not available.
423 	 */
424 	if (data->flags & XIVE_IRQ_FLAG_H_INT_ESB)
425 		return 0;
426 
427 	data->eoi_mmio = ioremap(data->eoi_page, 1u << data->esb_shift);
428 	if (!data->eoi_mmio) {
429 		pr_err("Failed to map EOI page for irq 0x%x\n", hw_irq);
430 		return -ENOMEM;
431 	}
432 
433 	/* Full function page supports trigger */
434 	if (flags & XIVE_SRC_TRIGGER) {
435 		data->trig_mmio = data->eoi_mmio;
436 		return 0;
437 	}
438 
439 	data->trig_mmio = ioremap(data->trig_page, 1u << data->esb_shift);
440 	if (!data->trig_mmio) {
441 		pr_err("Failed to map trigger page for irq 0x%x\n", hw_irq);
442 		return -ENOMEM;
443 	}
444 	return 0;
445 }
446 
447 static int xive_spapr_configure_irq(u32 hw_irq, u32 target, u8 prio, u32 sw_irq)
448 {
449 	long rc;
450 
451 	rc = plpar_int_set_source_config(XIVE_SRC_SET_EISN, hw_irq, target,
452 					 prio, sw_irq);
453 
454 	return rc == 0 ? 0 : -ENXIO;
455 }
456 
457 static int xive_spapr_get_irq_config(u32 hw_irq, u32 *target, u8 *prio,
458 				     u32 *sw_irq)
459 {
460 	long rc;
461 	unsigned long h_target;
462 	unsigned long h_prio;
463 	unsigned long h_sw_irq;
464 
465 	rc = plpar_int_get_source_config(0, hw_irq, &h_target, &h_prio,
466 					 &h_sw_irq);
467 
468 	*target = h_target;
469 	*prio = h_prio;
470 	*sw_irq = h_sw_irq;
471 
472 	return rc == 0 ? 0 : -ENXIO;
473 }
474 
475 /* This can be called multiple time to change a queue configuration */
476 static int xive_spapr_configure_queue(u32 target, struct xive_q *q, u8 prio,
477 				   __be32 *qpage, u32 order)
478 {
479 	s64 rc = 0;
480 	unsigned long esn_page;
481 	unsigned long esn_size;
482 	u64 flags, qpage_phys;
483 
484 	/* If there's an actual queue page, clean it */
485 	if (order) {
486 		if (WARN_ON(!qpage))
487 			return -EINVAL;
488 		qpage_phys = __pa(qpage);
489 	} else {
490 		qpage_phys = 0;
491 	}
492 
493 	/* Initialize the rest of the fields */
494 	q->msk = order ? ((1u << (order - 2)) - 1) : 0;
495 	q->idx = 0;
496 	q->toggle = 0;
497 
498 	rc = plpar_int_get_queue_info(0, target, prio, &esn_page, &esn_size);
499 	if (rc) {
500 		pr_err("Error %lld getting queue info CPU %d prio %d\n", rc,
501 		       target, prio);
502 		rc = -EIO;
503 		goto fail;
504 	}
505 
506 	/* TODO: add support for the notification page */
507 	q->eoi_phys = esn_page;
508 
509 	/* Default is to always notify */
510 	flags = XIVE_EQ_ALWAYS_NOTIFY;
511 
512 	/* Configure and enable the queue in HW */
513 	rc = plpar_int_set_queue_config(flags, target, prio, qpage_phys, order);
514 	if (rc) {
515 		pr_err("Error %lld setting queue for CPU %d prio %d\n", rc,
516 		       target, prio);
517 		rc = -EIO;
518 	} else {
519 		q->qpage = qpage;
520 		if (is_secure_guest())
521 			uv_share_page(PHYS_PFN(qpage_phys),
522 					1 << xive_alloc_order(order));
523 	}
524 fail:
525 	return rc;
526 }
527 
528 static int xive_spapr_setup_queue(unsigned int cpu, struct xive_cpu *xc,
529 				  u8 prio)
530 {
531 	struct xive_q *q = &xc->queue[prio];
532 	__be32 *qpage;
533 
534 	qpage = xive_queue_page_alloc(cpu, xive_queue_shift);
535 	if (IS_ERR(qpage))
536 		return PTR_ERR(qpage);
537 
538 	return xive_spapr_configure_queue(get_hard_smp_processor_id(cpu),
539 					  q, prio, qpage, xive_queue_shift);
540 }
541 
542 static void xive_spapr_cleanup_queue(unsigned int cpu, struct xive_cpu *xc,
543 				  u8 prio)
544 {
545 	struct xive_q *q = &xc->queue[prio];
546 	unsigned int alloc_order;
547 	long rc;
548 	int hw_cpu = get_hard_smp_processor_id(cpu);
549 
550 	rc = plpar_int_set_queue_config(0, hw_cpu, prio, 0, 0);
551 	if (rc)
552 		pr_err("Error %ld setting queue for CPU %d prio %d\n", rc,
553 		       hw_cpu, prio);
554 
555 	alloc_order = xive_alloc_order(xive_queue_shift);
556 	if (is_secure_guest())
557 		uv_unshare_page(PHYS_PFN(__pa(q->qpage)), 1 << alloc_order);
558 	free_pages((unsigned long)q->qpage, alloc_order);
559 	q->qpage = NULL;
560 }
561 
562 static bool xive_spapr_match(struct device_node *node)
563 {
564 	/* Ignore cascaded controllers for the moment */
565 	return true;
566 }
567 
568 #ifdef CONFIG_SMP
569 static int xive_spapr_get_ipi(unsigned int cpu, struct xive_cpu *xc)
570 {
571 	int irq = xive_irq_bitmap_alloc();
572 
573 	if (irq < 0) {
574 		pr_err("Failed to allocate IPI on CPU %d\n", cpu);
575 		return -ENXIO;
576 	}
577 
578 	xc->hw_ipi = irq;
579 	return 0;
580 }
581 
582 static void xive_spapr_put_ipi(unsigned int cpu, struct xive_cpu *xc)
583 {
584 	if (xc->hw_ipi == XIVE_BAD_IRQ)
585 		return;
586 
587 	xive_irq_bitmap_free(xc->hw_ipi);
588 	xc->hw_ipi = XIVE_BAD_IRQ;
589 }
590 #endif /* CONFIG_SMP */
591 
592 static void xive_spapr_shutdown(void)
593 {
594 	plpar_int_reset(0);
595 }
596 
597 /*
598  * Perform an "ack" cycle on the current thread. Grab the pending
599  * active priorities and update the CPPR to the most favored one.
600  */
601 static void xive_spapr_update_pending(struct xive_cpu *xc)
602 {
603 	u8 nsr, cppr;
604 	u16 ack;
605 
606 	/*
607 	 * Perform the "Acknowledge O/S to Register" cycle.
608 	 *
609 	 * Let's speedup the access to the TIMA using the raw I/O
610 	 * accessor as we don't need the synchronisation routine of
611 	 * the higher level ones
612 	 */
613 	ack = be16_to_cpu(__raw_readw(xive_tima + TM_SPC_ACK_OS_REG));
614 
615 	/* Synchronize subsequent queue accesses */
616 	mb();
617 
618 	/*
619 	 * Grab the CPPR and the "NSR" field which indicates the source
620 	 * of the interrupt (if any)
621 	 */
622 	cppr = ack & 0xff;
623 	nsr = ack >> 8;
624 
625 	if (nsr & TM_QW1_NSR_EO) {
626 		if (cppr == 0xff)
627 			return;
628 		/* Mark the priority pending */
629 		xc->pending_prio |= 1 << cppr;
630 
631 		/*
632 		 * A new interrupt should never have a CPPR less favored
633 		 * than our current one.
634 		 */
635 		if (cppr >= xc->cppr)
636 			pr_err("CPU %d odd ack CPPR, got %d at %d\n",
637 			       smp_processor_id(), cppr, xc->cppr);
638 
639 		/* Update our idea of what the CPPR is */
640 		xc->cppr = cppr;
641 	}
642 }
643 
644 static void xive_spapr_setup_cpu(unsigned int cpu, struct xive_cpu *xc)
645 {
646 	/* Only some debug on the TIMA settings */
647 	pr_debug("(HW value: %08x %08x %08x)\n",
648 		 in_be32(xive_tima + TM_QW1_OS + TM_WORD0),
649 		 in_be32(xive_tima + TM_QW1_OS + TM_WORD1),
650 		 in_be32(xive_tima + TM_QW1_OS + TM_WORD2));
651 }
652 
653 static void xive_spapr_teardown_cpu(unsigned int cpu, struct xive_cpu *xc)
654 {
655 	/* Nothing to do */;
656 }
657 
658 static void xive_spapr_sync_source(u32 hw_irq)
659 {
660 	/* Specs are unclear on what this is doing */
661 	plpar_int_sync(0, hw_irq);
662 }
663 
664 static int xive_spapr_debug_show(struct seq_file *m, void *private)
665 {
666 	struct xive_irq_bitmap *xibm;
667 	char *buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
668 
669 	if (!buf)
670 		return -ENOMEM;
671 
672 	list_for_each_entry(xibm, &xive_irq_bitmaps, list) {
673 		memset(buf, 0, PAGE_SIZE);
674 		bitmap_print_to_pagebuf(true, buf, xibm->bitmap, xibm->count);
675 		seq_printf(m, "bitmap #%d: %s", xibm->count, buf);
676 	}
677 	kfree(buf);
678 
679 	return 0;
680 }
681 
682 static const struct xive_ops xive_spapr_ops = {
683 	.populate_irq_data	= xive_spapr_populate_irq_data,
684 	.configure_irq		= xive_spapr_configure_irq,
685 	.get_irq_config		= xive_spapr_get_irq_config,
686 	.setup_queue		= xive_spapr_setup_queue,
687 	.cleanup_queue		= xive_spapr_cleanup_queue,
688 	.match			= xive_spapr_match,
689 	.shutdown		= xive_spapr_shutdown,
690 	.update_pending		= xive_spapr_update_pending,
691 	.setup_cpu		= xive_spapr_setup_cpu,
692 	.teardown_cpu		= xive_spapr_teardown_cpu,
693 	.sync_source		= xive_spapr_sync_source,
694 	.esb_rw			= xive_spapr_esb_rw,
695 #ifdef CONFIG_SMP
696 	.get_ipi		= xive_spapr_get_ipi,
697 	.put_ipi		= xive_spapr_put_ipi,
698 	.debug_show		= xive_spapr_debug_show,
699 #endif /* CONFIG_SMP */
700 	.name			= "spapr",
701 };
702 
703 /*
704  * get max priority from "/ibm,plat-res-int-priorities"
705  */
706 static bool __init xive_get_max_prio(u8 *max_prio)
707 {
708 	struct device_node *rootdn;
709 	const __be32 *reg;
710 	u32 len;
711 	int prio, found;
712 
713 	rootdn = of_find_node_by_path("/");
714 	if (!rootdn) {
715 		pr_err("not root node found !\n");
716 		return false;
717 	}
718 
719 	reg = of_get_property(rootdn, "ibm,plat-res-int-priorities", &len);
720 	if (!reg) {
721 		pr_err("Failed to read 'ibm,plat-res-int-priorities' property\n");
722 		return false;
723 	}
724 
725 	if (len % (2 * sizeof(u32)) != 0) {
726 		pr_err("invalid 'ibm,plat-res-int-priorities' property\n");
727 		return false;
728 	}
729 
730 	/* HW supports priorities in the range [0-7] and 0xFF is a
731 	 * wildcard priority used to mask. We scan the ranges reserved
732 	 * by the hypervisor to find the lowest priority we can use.
733 	 */
734 	found = 0xFF;
735 	for (prio = 0; prio < 8; prio++) {
736 		int reserved = 0;
737 		int i;
738 
739 		for (i = 0; i < len / (2 * sizeof(u32)); i++) {
740 			int base  = be32_to_cpu(reg[2 * i]);
741 			int range = be32_to_cpu(reg[2 * i + 1]);
742 
743 			if (prio >= base && prio < base + range)
744 				reserved++;
745 		}
746 
747 		if (!reserved)
748 			found = prio;
749 	}
750 
751 	if (found == 0xFF) {
752 		pr_err("no valid priority found in 'ibm,plat-res-int-priorities'\n");
753 		return false;
754 	}
755 
756 	*max_prio = found;
757 	return true;
758 }
759 
760 static const u8 *__init get_vec5_feature(unsigned int index)
761 {
762 	unsigned long root, chosen;
763 	int size;
764 	const u8 *vec5;
765 
766 	root = of_get_flat_dt_root();
767 	chosen = of_get_flat_dt_subnode_by_name(root, "chosen");
768 	if (chosen == -FDT_ERR_NOTFOUND)
769 		return NULL;
770 
771 	vec5 = of_get_flat_dt_prop(chosen, "ibm,architecture-vec-5", &size);
772 	if (!vec5)
773 		return NULL;
774 
775 	if (size <= index)
776 		return NULL;
777 
778 	return vec5 + index;
779 }
780 
781 static bool __init xive_spapr_disabled(void)
782 {
783 	const u8 *vec5_xive;
784 
785 	vec5_xive = get_vec5_feature(OV5_INDX(OV5_XIVE_SUPPORT));
786 	if (vec5_xive) {
787 		u8 val;
788 
789 		val = *vec5_xive & OV5_FEAT(OV5_XIVE_SUPPORT);
790 		switch (val) {
791 		case OV5_FEAT(OV5_XIVE_EITHER):
792 		case OV5_FEAT(OV5_XIVE_LEGACY):
793 			break;
794 		case OV5_FEAT(OV5_XIVE_EXPLOIT):
795 			/* Hypervisor only supports XIVE */
796 			if (xive_cmdline_disabled)
797 				pr_warn("WARNING: Ignoring cmdline option xive=off\n");
798 			return false;
799 		default:
800 			pr_warn("%s: Unknown xive support option: 0x%x\n",
801 				__func__, val);
802 			break;
803 		}
804 	}
805 
806 	return xive_cmdline_disabled;
807 }
808 
809 bool __init xive_spapr_init(void)
810 {
811 	struct device_node *np;
812 	struct resource r;
813 	void __iomem *tima;
814 	struct property *prop;
815 	u8 max_prio;
816 	u32 val;
817 	u32 len;
818 	const __be32 *reg;
819 	int i, err;
820 
821 	if (xive_spapr_disabled())
822 		return false;
823 
824 	pr_devel("%s()\n", __func__);
825 	np = of_find_compatible_node(NULL, NULL, "ibm,power-ivpe");
826 	if (!np) {
827 		pr_devel("not found !\n");
828 		return false;
829 	}
830 	pr_devel("Found %s\n", np->full_name);
831 
832 	/* Resource 1 is the OS ring TIMA */
833 	if (of_address_to_resource(np, 1, &r)) {
834 		pr_err("Failed to get thread mgmnt area resource\n");
835 		goto err_put;
836 	}
837 	tima = ioremap(r.start, resource_size(&r));
838 	if (!tima) {
839 		pr_err("Failed to map thread mgmnt area\n");
840 		goto err_put;
841 	}
842 
843 	if (!xive_get_max_prio(&max_prio))
844 		goto err_unmap;
845 
846 	/* Feed the IRQ number allocator with the ranges given in the DT */
847 	reg = of_get_property(np, "ibm,xive-lisn-ranges", &len);
848 	if (!reg) {
849 		pr_err("Failed to read 'ibm,xive-lisn-ranges' property\n");
850 		goto err_unmap;
851 	}
852 
853 	if (len % (2 * sizeof(u32)) != 0) {
854 		pr_err("invalid 'ibm,xive-lisn-ranges' property\n");
855 		goto err_unmap;
856 	}
857 
858 	for (i = 0; i < len / (2 * sizeof(u32)); i++, reg += 2) {
859 		err = xive_irq_bitmap_add(be32_to_cpu(reg[0]),
860 					  be32_to_cpu(reg[1]));
861 		if (err < 0)
862 			goto err_mem_free;
863 	}
864 
865 	/* Iterate the EQ sizes and pick one */
866 	of_property_for_each_u32(np, "ibm,xive-eq-sizes", prop, reg, val) {
867 		xive_queue_shift = val;
868 		if (val == PAGE_SHIFT)
869 			break;
870 	}
871 
872 	/* Initialize XIVE core with our backend */
873 	if (!xive_core_init(np, &xive_spapr_ops, tima, TM_QW1_OS, max_prio))
874 		goto err_mem_free;
875 
876 	of_node_put(np);
877 	pr_info("Using %dkB queues\n", 1 << (xive_queue_shift - 10));
878 	return true;
879 
880 err_mem_free:
881 	xive_irq_bitmap_remove_all();
882 err_unmap:
883 	iounmap(tima);
884 err_put:
885 	of_node_put(np);
886 	return false;
887 }
888 
889 machine_arch_initcall(pseries, xive_core_debug_init);
890