xref: /openbmc/linux/arch/powerpc/sysdev/xive/native.c (revision 15e3ae36)
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/debugfs.h>
11 #include <linux/smp.h>
12 #include <linux/interrupt.h>
13 #include <linux/seq_file.h>
14 #include <linux/init.h>
15 #include <linux/of.h>
16 #include <linux/slab.h>
17 #include <linux/spinlock.h>
18 #include <linux/delay.h>
19 #include <linux/cpumask.h>
20 #include <linux/mm.h>
21 
22 #include <asm/machdep.h>
23 #include <asm/prom.h>
24 #include <asm/io.h>
25 #include <asm/smp.h>
26 #include <asm/irq.h>
27 #include <asm/errno.h>
28 #include <asm/xive.h>
29 #include <asm/xive-regs.h>
30 #include <asm/opal.h>
31 #include <asm/kvm_ppc.h>
32 
33 #include "xive-internal.h"
34 
35 
36 static u32 xive_provision_size;
37 static u32 *xive_provision_chips;
38 static u32 xive_provision_chip_count;
39 static u32 xive_queue_shift;
40 static u32 xive_pool_vps = XIVE_INVALID_VP;
41 static struct kmem_cache *xive_provision_cache;
42 static bool xive_has_single_esc;
43 
44 int xive_native_populate_irq_data(u32 hw_irq, struct xive_irq_data *data)
45 {
46 	__be64 flags, eoi_page, trig_page;
47 	__be32 esb_shift, src_chip;
48 	u64 opal_flags;
49 	s64 rc;
50 
51 	memset(data, 0, sizeof(*data));
52 
53 	rc = opal_xive_get_irq_info(hw_irq, &flags, &eoi_page, &trig_page,
54 				    &esb_shift, &src_chip);
55 	if (rc) {
56 		pr_err("opal_xive_get_irq_info(0x%x) returned %lld\n",
57 		       hw_irq, rc);
58 		return -EINVAL;
59 	}
60 
61 	opal_flags = be64_to_cpu(flags);
62 	if (opal_flags & OPAL_XIVE_IRQ_STORE_EOI)
63 		data->flags |= XIVE_IRQ_FLAG_STORE_EOI;
64 	if (opal_flags & OPAL_XIVE_IRQ_LSI)
65 		data->flags |= XIVE_IRQ_FLAG_LSI;
66 	if (opal_flags & OPAL_XIVE_IRQ_SHIFT_BUG)
67 		data->flags |= XIVE_IRQ_FLAG_SHIFT_BUG;
68 	if (opal_flags & OPAL_XIVE_IRQ_MASK_VIA_FW)
69 		data->flags |= XIVE_IRQ_FLAG_MASK_FW;
70 	if (opal_flags & OPAL_XIVE_IRQ_EOI_VIA_FW)
71 		data->flags |= XIVE_IRQ_FLAG_EOI_FW;
72 	data->eoi_page = be64_to_cpu(eoi_page);
73 	data->trig_page = be64_to_cpu(trig_page);
74 	data->esb_shift = be32_to_cpu(esb_shift);
75 	data->src_chip = be32_to_cpu(src_chip);
76 
77 	data->eoi_mmio = ioremap(data->eoi_page, 1u << data->esb_shift);
78 	if (!data->eoi_mmio) {
79 		pr_err("Failed to map EOI page for irq 0x%x\n", hw_irq);
80 		return -ENOMEM;
81 	}
82 
83 	data->hw_irq = hw_irq;
84 
85 	if (!data->trig_page)
86 		return 0;
87 	if (data->trig_page == data->eoi_page) {
88 		data->trig_mmio = data->eoi_mmio;
89 		return 0;
90 	}
91 
92 	data->trig_mmio = ioremap(data->trig_page, 1u << data->esb_shift);
93 	if (!data->trig_mmio) {
94 		pr_err("Failed to map trigger page for irq 0x%x\n", hw_irq);
95 		return -ENOMEM;
96 	}
97 	return 0;
98 }
99 EXPORT_SYMBOL_GPL(xive_native_populate_irq_data);
100 
101 int xive_native_configure_irq(u32 hw_irq, u32 target, u8 prio, u32 sw_irq)
102 {
103 	s64 rc;
104 
105 	for (;;) {
106 		rc = opal_xive_set_irq_config(hw_irq, target, prio, sw_irq);
107 		if (rc != OPAL_BUSY)
108 			break;
109 		msleep(OPAL_BUSY_DELAY_MS);
110 	}
111 	return rc == 0 ? 0 : -ENXIO;
112 }
113 EXPORT_SYMBOL_GPL(xive_native_configure_irq);
114 
115 static int xive_native_get_irq_config(u32 hw_irq, u32 *target, u8 *prio,
116 				      u32 *sw_irq)
117 {
118 	s64 rc;
119 	__be64 vp;
120 	__be32 lirq;
121 
122 	rc = opal_xive_get_irq_config(hw_irq, &vp, prio, &lirq);
123 
124 	*target = be64_to_cpu(vp);
125 	*sw_irq = be32_to_cpu(lirq);
126 
127 	return rc == 0 ? 0 : -ENXIO;
128 }
129 
130 /* This can be called multiple time to change a queue configuration */
131 int xive_native_configure_queue(u32 vp_id, struct xive_q *q, u8 prio,
132 				__be32 *qpage, u32 order, bool can_escalate)
133 {
134 	s64 rc = 0;
135 	__be64 qeoi_page_be;
136 	__be32 esc_irq_be;
137 	u64 flags, qpage_phys;
138 
139 	/* If there's an actual queue page, clean it */
140 	if (order) {
141 		if (WARN_ON(!qpage))
142 			return -EINVAL;
143 		qpage_phys = __pa(qpage);
144 	} else
145 		qpage_phys = 0;
146 
147 	/* Initialize the rest of the fields */
148 	q->msk = order ? ((1u << (order - 2)) - 1) : 0;
149 	q->idx = 0;
150 	q->toggle = 0;
151 
152 	rc = opal_xive_get_queue_info(vp_id, prio, NULL, NULL,
153 				      &qeoi_page_be,
154 				      &esc_irq_be,
155 				      NULL);
156 	if (rc) {
157 		pr_err("Error %lld getting queue info prio %d\n", rc, prio);
158 		rc = -EIO;
159 		goto fail;
160 	}
161 	q->eoi_phys = be64_to_cpu(qeoi_page_be);
162 
163 	/* Default flags */
164 	flags = OPAL_XIVE_EQ_ALWAYS_NOTIFY | OPAL_XIVE_EQ_ENABLED;
165 
166 	/* Escalation needed ? */
167 	if (can_escalate) {
168 		q->esc_irq = be32_to_cpu(esc_irq_be);
169 		flags |= OPAL_XIVE_EQ_ESCALATE;
170 	}
171 
172 	/* Configure and enable the queue in HW */
173 	for (;;) {
174 		rc = opal_xive_set_queue_info(vp_id, prio, qpage_phys, order, flags);
175 		if (rc != OPAL_BUSY)
176 			break;
177 		msleep(OPAL_BUSY_DELAY_MS);
178 	}
179 	if (rc) {
180 		pr_err("Error %lld setting queue for prio %d\n", rc, prio);
181 		rc = -EIO;
182 	} else {
183 		/*
184 		 * KVM code requires all of the above to be visible before
185 		 * q->qpage is set due to how it manages IPI EOIs
186 		 */
187 		wmb();
188 		q->qpage = qpage;
189 	}
190 fail:
191 	return rc;
192 }
193 EXPORT_SYMBOL_GPL(xive_native_configure_queue);
194 
195 static void __xive_native_disable_queue(u32 vp_id, struct xive_q *q, u8 prio)
196 {
197 	s64 rc;
198 
199 	/* Disable the queue in HW */
200 	for (;;) {
201 		rc = opal_xive_set_queue_info(vp_id, prio, 0, 0, 0);
202 		if (rc != OPAL_BUSY)
203 			break;
204 		msleep(OPAL_BUSY_DELAY_MS);
205 	}
206 	if (rc)
207 		pr_err("Error %lld disabling queue for prio %d\n", rc, prio);
208 }
209 
210 void xive_native_disable_queue(u32 vp_id, struct xive_q *q, u8 prio)
211 {
212 	__xive_native_disable_queue(vp_id, q, prio);
213 }
214 EXPORT_SYMBOL_GPL(xive_native_disable_queue);
215 
216 static int xive_native_setup_queue(unsigned int cpu, struct xive_cpu *xc, u8 prio)
217 {
218 	struct xive_q *q = &xc->queue[prio];
219 	__be32 *qpage;
220 
221 	qpage = xive_queue_page_alloc(cpu, xive_queue_shift);
222 	if (IS_ERR(qpage))
223 		return PTR_ERR(qpage);
224 
225 	return xive_native_configure_queue(get_hard_smp_processor_id(cpu),
226 					   q, prio, qpage, xive_queue_shift, false);
227 }
228 
229 static void xive_native_cleanup_queue(unsigned int cpu, struct xive_cpu *xc, u8 prio)
230 {
231 	struct xive_q *q = &xc->queue[prio];
232 	unsigned int alloc_order;
233 
234 	/*
235 	 * We use the variant with no iounmap as this is called on exec
236 	 * from an IPI and iounmap isn't safe
237 	 */
238 	__xive_native_disable_queue(get_hard_smp_processor_id(cpu), q, prio);
239 	alloc_order = xive_alloc_order(xive_queue_shift);
240 	free_pages((unsigned long)q->qpage, alloc_order);
241 	q->qpage = NULL;
242 }
243 
244 static bool xive_native_match(struct device_node *node)
245 {
246 	return of_device_is_compatible(node, "ibm,opal-xive-vc");
247 }
248 
249 static s64 opal_xive_allocate_irq(u32 chip_id)
250 {
251 	s64 irq = opal_xive_allocate_irq_raw(chip_id);
252 
253 	/*
254 	 * Old versions of skiboot can incorrectly return 0xffffffff to
255 	 * indicate no space, fix it up here.
256 	 */
257 	return irq == 0xffffffff ? OPAL_RESOURCE : irq;
258 }
259 
260 #ifdef CONFIG_SMP
261 static int xive_native_get_ipi(unsigned int cpu, struct xive_cpu *xc)
262 {
263 	s64 irq;
264 
265 	/* Allocate an IPI and populate info about it */
266 	for (;;) {
267 		irq = opal_xive_allocate_irq(xc->chip_id);
268 		if (irq == OPAL_BUSY) {
269 			msleep(OPAL_BUSY_DELAY_MS);
270 			continue;
271 		}
272 		if (irq < 0) {
273 			pr_err("Failed to allocate IPI on CPU %d\n", cpu);
274 			return -ENXIO;
275 		}
276 		xc->hw_ipi = irq;
277 		break;
278 	}
279 	return 0;
280 }
281 #endif /* CONFIG_SMP */
282 
283 u32 xive_native_alloc_irq(void)
284 {
285 	s64 rc;
286 
287 	for (;;) {
288 		rc = opal_xive_allocate_irq(OPAL_XIVE_ANY_CHIP);
289 		if (rc != OPAL_BUSY)
290 			break;
291 		msleep(OPAL_BUSY_DELAY_MS);
292 	}
293 	if (rc < 0)
294 		return 0;
295 	return rc;
296 }
297 EXPORT_SYMBOL_GPL(xive_native_alloc_irq);
298 
299 void xive_native_free_irq(u32 irq)
300 {
301 	for (;;) {
302 		s64 rc = opal_xive_free_irq(irq);
303 		if (rc != OPAL_BUSY)
304 			break;
305 		msleep(OPAL_BUSY_DELAY_MS);
306 	}
307 }
308 EXPORT_SYMBOL_GPL(xive_native_free_irq);
309 
310 #ifdef CONFIG_SMP
311 static void xive_native_put_ipi(unsigned int cpu, struct xive_cpu *xc)
312 {
313 	s64 rc;
314 
315 	/* Free the IPI */
316 	if (xc->hw_ipi == XIVE_BAD_IRQ)
317 		return;
318 	for (;;) {
319 		rc = opal_xive_free_irq(xc->hw_ipi);
320 		if (rc == OPAL_BUSY) {
321 			msleep(OPAL_BUSY_DELAY_MS);
322 			continue;
323 		}
324 		xc->hw_ipi = XIVE_BAD_IRQ;
325 		break;
326 	}
327 }
328 #endif /* CONFIG_SMP */
329 
330 static void xive_native_shutdown(void)
331 {
332 	/* Switch the XIVE to emulation mode */
333 	opal_xive_reset(OPAL_XIVE_MODE_EMU);
334 }
335 
336 /*
337  * Perform an "ack" cycle on the current thread, thus
338  * grabbing the pending active priorities and updating
339  * the CPPR to the most favored one.
340  */
341 static void xive_native_update_pending(struct xive_cpu *xc)
342 {
343 	u8 he, cppr;
344 	u16 ack;
345 
346 	/* Perform the acknowledge hypervisor to register cycle */
347 	ack = be16_to_cpu(__raw_readw(xive_tima + TM_SPC_ACK_HV_REG));
348 
349 	/* Synchronize subsequent queue accesses */
350 	mb();
351 
352 	/*
353 	 * Grab the CPPR and the "HE" field which indicates the source
354 	 * of the hypervisor interrupt (if any)
355 	 */
356 	cppr = ack & 0xff;
357 	he = (ack >> 8) >> 6;
358 	switch(he) {
359 	case TM_QW3_NSR_HE_NONE: /* Nothing to see here */
360 		break;
361 	case TM_QW3_NSR_HE_PHYS: /* Physical thread interrupt */
362 		if (cppr == 0xff)
363 			return;
364 		/* Mark the priority pending */
365 		xc->pending_prio |= 1 << cppr;
366 
367 		/*
368 		 * A new interrupt should never have a CPPR less favored
369 		 * than our current one.
370 		 */
371 		if (cppr >= xc->cppr)
372 			pr_err("CPU %d odd ack CPPR, got %d at %d\n",
373 			       smp_processor_id(), cppr, xc->cppr);
374 
375 		/* Update our idea of what the CPPR is */
376 		xc->cppr = cppr;
377 		break;
378 	case TM_QW3_NSR_HE_POOL: /* HV Pool interrupt (unused) */
379 	case TM_QW3_NSR_HE_LSI:  /* Legacy FW LSI (unused) */
380 		pr_err("CPU %d got unexpected interrupt type HE=%d\n",
381 		       smp_processor_id(), he);
382 		return;
383 	}
384 }
385 
386 static void xive_native_eoi(u32 hw_irq)
387 {
388 	/*
389 	 * Not normally used except if specific interrupts need
390 	 * a workaround on EOI.
391 	 */
392 	opal_int_eoi(hw_irq);
393 }
394 
395 static void xive_native_setup_cpu(unsigned int cpu, struct xive_cpu *xc)
396 {
397 	s64 rc;
398 	u32 vp;
399 	__be64 vp_cam_be;
400 	u64 vp_cam;
401 
402 	if (xive_pool_vps == XIVE_INVALID_VP)
403 		return;
404 
405 	/* Check if pool VP already active, if it is, pull it */
406 	if (in_be32(xive_tima + TM_QW2_HV_POOL + TM_WORD2) & TM_QW2W2_VP)
407 		in_be64(xive_tima + TM_SPC_PULL_POOL_CTX);
408 
409 	/* Enable the pool VP */
410 	vp = xive_pool_vps + cpu;
411 	for (;;) {
412 		rc = opal_xive_set_vp_info(vp, OPAL_XIVE_VP_ENABLED, 0);
413 		if (rc != OPAL_BUSY)
414 			break;
415 		msleep(OPAL_BUSY_DELAY_MS);
416 	}
417 	if (rc) {
418 		pr_err("Failed to enable pool VP on CPU %d\n", cpu);
419 		return;
420 	}
421 
422 	/* Grab it's CAM value */
423 	rc = opal_xive_get_vp_info(vp, NULL, &vp_cam_be, NULL, NULL);
424 	if (rc) {
425 		pr_err("Failed to get pool VP info CPU %d\n", cpu);
426 		return;
427 	}
428 	vp_cam = be64_to_cpu(vp_cam_be);
429 
430 	/* Push it on the CPU (set LSMFB to 0xff to skip backlog scan) */
431 	out_be32(xive_tima + TM_QW2_HV_POOL + TM_WORD0, 0xff);
432 	out_be32(xive_tima + TM_QW2_HV_POOL + TM_WORD2, TM_QW2W2_VP | vp_cam);
433 }
434 
435 static void xive_native_teardown_cpu(unsigned int cpu, struct xive_cpu *xc)
436 {
437 	s64 rc;
438 	u32 vp;
439 
440 	if (xive_pool_vps == XIVE_INVALID_VP)
441 		return;
442 
443 	/* Pull the pool VP from the CPU */
444 	in_be64(xive_tima + TM_SPC_PULL_POOL_CTX);
445 
446 	/* Disable it */
447 	vp = xive_pool_vps + cpu;
448 	for (;;) {
449 		rc = opal_xive_set_vp_info(vp, 0, 0);
450 		if (rc != OPAL_BUSY)
451 			break;
452 		msleep(OPAL_BUSY_DELAY_MS);
453 	}
454 }
455 
456 void xive_native_sync_source(u32 hw_irq)
457 {
458 	opal_xive_sync(XIVE_SYNC_EAS, hw_irq);
459 }
460 EXPORT_SYMBOL_GPL(xive_native_sync_source);
461 
462 void xive_native_sync_queue(u32 hw_irq)
463 {
464 	opal_xive_sync(XIVE_SYNC_QUEUE, hw_irq);
465 }
466 EXPORT_SYMBOL_GPL(xive_native_sync_queue);
467 
468 static const struct xive_ops xive_native_ops = {
469 	.populate_irq_data	= xive_native_populate_irq_data,
470 	.configure_irq		= xive_native_configure_irq,
471 	.get_irq_config		= xive_native_get_irq_config,
472 	.setup_queue		= xive_native_setup_queue,
473 	.cleanup_queue		= xive_native_cleanup_queue,
474 	.match			= xive_native_match,
475 	.shutdown		= xive_native_shutdown,
476 	.update_pending		= xive_native_update_pending,
477 	.eoi			= xive_native_eoi,
478 	.setup_cpu		= xive_native_setup_cpu,
479 	.teardown_cpu		= xive_native_teardown_cpu,
480 	.sync_source		= xive_native_sync_source,
481 #ifdef CONFIG_SMP
482 	.get_ipi		= xive_native_get_ipi,
483 	.put_ipi		= xive_native_put_ipi,
484 #endif /* CONFIG_SMP */
485 	.name			= "native",
486 };
487 
488 static bool xive_parse_provisioning(struct device_node *np)
489 {
490 	int rc;
491 
492 	if (of_property_read_u32(np, "ibm,xive-provision-page-size",
493 				 &xive_provision_size) < 0)
494 		return true;
495 	rc = of_property_count_elems_of_size(np, "ibm,xive-provision-chips", 4);
496 	if (rc < 0) {
497 		pr_err("Error %d getting provision chips array\n", rc);
498 		return false;
499 	}
500 	xive_provision_chip_count = rc;
501 	if (rc == 0)
502 		return true;
503 
504 	xive_provision_chips = kcalloc(4, xive_provision_chip_count,
505 				       GFP_KERNEL);
506 	if (WARN_ON(!xive_provision_chips))
507 		return false;
508 
509 	rc = of_property_read_u32_array(np, "ibm,xive-provision-chips",
510 					xive_provision_chips,
511 					xive_provision_chip_count);
512 	if (rc < 0) {
513 		pr_err("Error %d reading provision chips array\n", rc);
514 		return false;
515 	}
516 
517 	xive_provision_cache = kmem_cache_create("xive-provision",
518 						 xive_provision_size,
519 						 xive_provision_size,
520 						 0, NULL);
521 	if (!xive_provision_cache) {
522 		pr_err("Failed to allocate provision cache\n");
523 		return false;
524 	}
525 	return true;
526 }
527 
528 static void xive_native_setup_pools(void)
529 {
530 	/* Allocate a pool big enough */
531 	pr_debug("XIVE: Allocating VP block for pool size %u\n", nr_cpu_ids);
532 
533 	xive_pool_vps = xive_native_alloc_vp_block(nr_cpu_ids);
534 	if (WARN_ON(xive_pool_vps == XIVE_INVALID_VP))
535 		pr_err("XIVE: Failed to allocate pool VP, KVM might not function\n");
536 
537 	pr_debug("XIVE: Pool VPs allocated at 0x%x for %u max CPUs\n",
538 		 xive_pool_vps, nr_cpu_ids);
539 }
540 
541 u32 xive_native_default_eq_shift(void)
542 {
543 	return xive_queue_shift;
544 }
545 EXPORT_SYMBOL_GPL(xive_native_default_eq_shift);
546 
547 unsigned long xive_tima_os;
548 EXPORT_SYMBOL_GPL(xive_tima_os);
549 
550 bool __init xive_native_init(void)
551 {
552 	struct device_node *np;
553 	struct resource r;
554 	void __iomem *tima;
555 	struct property *prop;
556 	u8 max_prio = 7;
557 	const __be32 *p;
558 	u32 val, cpu;
559 	s64 rc;
560 
561 	if (xive_cmdline_disabled)
562 		return false;
563 
564 	pr_devel("xive_native_init()\n");
565 	np = of_find_compatible_node(NULL, NULL, "ibm,opal-xive-pe");
566 	if (!np) {
567 		pr_devel("not found !\n");
568 		return false;
569 	}
570 	pr_devel("Found %pOF\n", np);
571 
572 	/* Resource 1 is HV window */
573 	if (of_address_to_resource(np, 1, &r)) {
574 		pr_err("Failed to get thread mgmnt area resource\n");
575 		return false;
576 	}
577 	tima = ioremap(r.start, resource_size(&r));
578 	if (!tima) {
579 		pr_err("Failed to map thread mgmnt area\n");
580 		return false;
581 	}
582 
583 	/* Read number of priorities */
584 	if (of_property_read_u32(np, "ibm,xive-#priorities", &val) == 0)
585 		max_prio = val - 1;
586 
587 	/* Iterate the EQ sizes and pick one */
588 	of_property_for_each_u32(np, "ibm,xive-eq-sizes", prop, p, val) {
589 		xive_queue_shift = val;
590 		if (val == PAGE_SHIFT)
591 			break;
592 	}
593 
594 	/* Do we support single escalation */
595 	if (of_get_property(np, "single-escalation-support", NULL) != NULL)
596 		xive_has_single_esc = true;
597 
598 	/* Configure Thread Management areas for KVM */
599 	for_each_possible_cpu(cpu)
600 		kvmppc_set_xive_tima(cpu, r.start, tima);
601 
602 	/* Resource 2 is OS window */
603 	if (of_address_to_resource(np, 2, &r)) {
604 		pr_err("Failed to get thread mgmnt area resource\n");
605 		return false;
606 	}
607 
608 	xive_tima_os = r.start;
609 
610 	/* Grab size of provisionning pages */
611 	xive_parse_provisioning(np);
612 
613 	/* Switch the XIVE to exploitation mode */
614 	rc = opal_xive_reset(OPAL_XIVE_MODE_EXPL);
615 	if (rc) {
616 		pr_err("Switch to exploitation mode failed with error %lld\n", rc);
617 		return false;
618 	}
619 
620 	/* Setup some dummy HV pool VPs */
621 	xive_native_setup_pools();
622 
623 	/* Initialize XIVE core with our backend */
624 	if (!xive_core_init(&xive_native_ops, tima, TM_QW3_HV_PHYS,
625 			    max_prio)) {
626 		opal_xive_reset(OPAL_XIVE_MODE_EMU);
627 		return false;
628 	}
629 	pr_info("Using %dkB queues\n", 1 << (xive_queue_shift - 10));
630 	return true;
631 }
632 
633 static bool xive_native_provision_pages(void)
634 {
635 	u32 i;
636 	void *p;
637 
638 	for (i = 0; i < xive_provision_chip_count; i++) {
639 		u32 chip = xive_provision_chips[i];
640 
641 		/*
642 		 * XXX TODO: Try to make the allocation local to the node where
643 		 * the chip resides.
644 		 */
645 		p = kmem_cache_alloc(xive_provision_cache, GFP_KERNEL);
646 		if (!p) {
647 			pr_err("Failed to allocate provisioning page\n");
648 			return false;
649 		}
650 		opal_xive_donate_page(chip, __pa(p));
651 	}
652 	return true;
653 }
654 
655 u32 xive_native_alloc_vp_block(u32 max_vcpus)
656 {
657 	s64 rc;
658 	u32 order;
659 
660 	order = fls(max_vcpus) - 1;
661 	if (max_vcpus > (1 << order))
662 		order++;
663 
664 	pr_debug("VP block alloc, for max VCPUs %d use order %d\n",
665 		 max_vcpus, order);
666 
667 	for (;;) {
668 		rc = opal_xive_alloc_vp_block(order);
669 		switch (rc) {
670 		case OPAL_BUSY:
671 			msleep(OPAL_BUSY_DELAY_MS);
672 			break;
673 		case OPAL_XIVE_PROVISIONING:
674 			if (!xive_native_provision_pages())
675 				return XIVE_INVALID_VP;
676 			break;
677 		default:
678 			if (rc < 0) {
679 				pr_err("OPAL failed to allocate VCPUs order %d, err %lld\n",
680 				       order, rc);
681 				return XIVE_INVALID_VP;
682 			}
683 			return rc;
684 		}
685 	}
686 }
687 EXPORT_SYMBOL_GPL(xive_native_alloc_vp_block);
688 
689 void xive_native_free_vp_block(u32 vp_base)
690 {
691 	s64 rc;
692 
693 	if (vp_base == XIVE_INVALID_VP)
694 		return;
695 
696 	rc = opal_xive_free_vp_block(vp_base);
697 	if (rc < 0)
698 		pr_warn("OPAL error %lld freeing VP block\n", rc);
699 }
700 EXPORT_SYMBOL_GPL(xive_native_free_vp_block);
701 
702 int xive_native_enable_vp(u32 vp_id, bool single_escalation)
703 {
704 	s64 rc;
705 	u64 flags = OPAL_XIVE_VP_ENABLED;
706 
707 	if (single_escalation)
708 		flags |= OPAL_XIVE_VP_SINGLE_ESCALATION;
709 	for (;;) {
710 		rc = opal_xive_set_vp_info(vp_id, flags, 0);
711 		if (rc != OPAL_BUSY)
712 			break;
713 		msleep(OPAL_BUSY_DELAY_MS);
714 	}
715 	return rc ? -EIO : 0;
716 }
717 EXPORT_SYMBOL_GPL(xive_native_enable_vp);
718 
719 int xive_native_disable_vp(u32 vp_id)
720 {
721 	s64 rc;
722 
723 	for (;;) {
724 		rc = opal_xive_set_vp_info(vp_id, 0, 0);
725 		if (rc != OPAL_BUSY)
726 			break;
727 		msleep(OPAL_BUSY_DELAY_MS);
728 	}
729 	return rc ? -EIO : 0;
730 }
731 EXPORT_SYMBOL_GPL(xive_native_disable_vp);
732 
733 int xive_native_get_vp_info(u32 vp_id, u32 *out_cam_id, u32 *out_chip_id)
734 {
735 	__be64 vp_cam_be;
736 	__be32 vp_chip_id_be;
737 	s64 rc;
738 
739 	rc = opal_xive_get_vp_info(vp_id, NULL, &vp_cam_be, NULL, &vp_chip_id_be);
740 	if (rc)
741 		return -EIO;
742 	*out_cam_id = be64_to_cpu(vp_cam_be) & 0xffffffffu;
743 	*out_chip_id = be32_to_cpu(vp_chip_id_be);
744 
745 	return 0;
746 }
747 EXPORT_SYMBOL_GPL(xive_native_get_vp_info);
748 
749 bool xive_native_has_single_escalation(void)
750 {
751 	return xive_has_single_esc;
752 }
753 EXPORT_SYMBOL_GPL(xive_native_has_single_escalation);
754 
755 int xive_native_get_queue_info(u32 vp_id, u32 prio,
756 			       u64 *out_qpage,
757 			       u64 *out_qsize,
758 			       u64 *out_qeoi_page,
759 			       u32 *out_escalate_irq,
760 			       u64 *out_qflags)
761 {
762 	__be64 qpage;
763 	__be64 qsize;
764 	__be64 qeoi_page;
765 	__be32 escalate_irq;
766 	__be64 qflags;
767 	s64 rc;
768 
769 	rc = opal_xive_get_queue_info(vp_id, prio, &qpage, &qsize,
770 				      &qeoi_page, &escalate_irq, &qflags);
771 	if (rc) {
772 		pr_err("OPAL failed to get queue info for VCPU %d/%d : %lld\n",
773 		       vp_id, prio, rc);
774 		return -EIO;
775 	}
776 
777 	if (out_qpage)
778 		*out_qpage = be64_to_cpu(qpage);
779 	if (out_qsize)
780 		*out_qsize = be32_to_cpu(qsize);
781 	if (out_qeoi_page)
782 		*out_qeoi_page = be64_to_cpu(qeoi_page);
783 	if (out_escalate_irq)
784 		*out_escalate_irq = be32_to_cpu(escalate_irq);
785 	if (out_qflags)
786 		*out_qflags = be64_to_cpu(qflags);
787 
788 	return 0;
789 }
790 EXPORT_SYMBOL_GPL(xive_native_get_queue_info);
791 
792 int xive_native_get_queue_state(u32 vp_id, u32 prio, u32 *qtoggle, u32 *qindex)
793 {
794 	__be32 opal_qtoggle;
795 	__be32 opal_qindex;
796 	s64 rc;
797 
798 	rc = opal_xive_get_queue_state(vp_id, prio, &opal_qtoggle,
799 				       &opal_qindex);
800 	if (rc) {
801 		pr_err("OPAL failed to get queue state for VCPU %d/%d : %lld\n",
802 		       vp_id, prio, rc);
803 		return -EIO;
804 	}
805 
806 	if (qtoggle)
807 		*qtoggle = be32_to_cpu(opal_qtoggle);
808 	if (qindex)
809 		*qindex = be32_to_cpu(opal_qindex);
810 
811 	return 0;
812 }
813 EXPORT_SYMBOL_GPL(xive_native_get_queue_state);
814 
815 int xive_native_set_queue_state(u32 vp_id, u32 prio, u32 qtoggle, u32 qindex)
816 {
817 	s64 rc;
818 
819 	rc = opal_xive_set_queue_state(vp_id, prio, qtoggle, qindex);
820 	if (rc) {
821 		pr_err("OPAL failed to set queue state for VCPU %d/%d : %lld\n",
822 		       vp_id, prio, rc);
823 		return -EIO;
824 	}
825 
826 	return 0;
827 }
828 EXPORT_SYMBOL_GPL(xive_native_set_queue_state);
829 
830 bool xive_native_has_queue_state_support(void)
831 {
832 	return opal_check_token(OPAL_XIVE_GET_QUEUE_STATE) &&
833 		opal_check_token(OPAL_XIVE_SET_QUEUE_STATE);
834 }
835 EXPORT_SYMBOL_GPL(xive_native_has_queue_state_support);
836 
837 int xive_native_get_vp_state(u32 vp_id, u64 *out_state)
838 {
839 	__be64 state;
840 	s64 rc;
841 
842 	rc = opal_xive_get_vp_state(vp_id, &state);
843 	if (rc) {
844 		pr_err("OPAL failed to get vp state for VCPU %d : %lld\n",
845 		       vp_id, rc);
846 		return -EIO;
847 	}
848 
849 	if (out_state)
850 		*out_state = be64_to_cpu(state);
851 	return 0;
852 }
853 EXPORT_SYMBOL_GPL(xive_native_get_vp_state);
854 
855 machine_arch_initcall(powernv, xive_core_debug_init);
856