xref: /openbmc/linux/arch/powerpc/sysdev/xive/native.c (revision 86edee97)
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/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/opal.h>
30 #include <asm/kvm_ppc.h>
31 
32 #include "xive-internal.h"
33 
34 
35 static u32 xive_provision_size;
36 static u32 *xive_provision_chips;
37 static u32 xive_provision_chip_count;
38 static u32 xive_queue_shift;
39 static u32 xive_pool_vps = XIVE_INVALID_VP;
40 static struct kmem_cache *xive_provision_cache;
41 static bool xive_has_single_esc;
42 
43 int xive_native_populate_irq_data(u32 hw_irq, struct xive_irq_data *data)
44 {
45 	__be64 flags, eoi_page, trig_page;
46 	__be32 esb_shift, src_chip;
47 	u64 opal_flags;
48 	s64 rc;
49 
50 	memset(data, 0, sizeof(*data));
51 
52 	rc = opal_xive_get_irq_info(hw_irq, &flags, &eoi_page, &trig_page,
53 				    &esb_shift, &src_chip);
54 	if (rc) {
55 		pr_err("opal_xive_get_irq_info(0x%x) returned %lld\n",
56 		       hw_irq, rc);
57 		return -EINVAL;
58 	}
59 
60 	opal_flags = be64_to_cpu(flags);
61 	if (opal_flags & OPAL_XIVE_IRQ_STORE_EOI)
62 		data->flags |= XIVE_IRQ_FLAG_STORE_EOI;
63 	if (opal_flags & OPAL_XIVE_IRQ_LSI)
64 		data->flags |= XIVE_IRQ_FLAG_LSI;
65 	if (opal_flags & OPAL_XIVE_IRQ_SHIFT_BUG)
66 		data->flags |= XIVE_IRQ_FLAG_SHIFT_BUG;
67 	if (opal_flags & OPAL_XIVE_IRQ_MASK_VIA_FW)
68 		data->flags |= XIVE_IRQ_FLAG_MASK_FW;
69 	if (opal_flags & OPAL_XIVE_IRQ_EOI_VIA_FW)
70 		data->flags |= XIVE_IRQ_FLAG_EOI_FW;
71 	data->eoi_page = be64_to_cpu(eoi_page);
72 	data->trig_page = be64_to_cpu(trig_page);
73 	data->esb_shift = be32_to_cpu(esb_shift);
74 	data->src_chip = be32_to_cpu(src_chip);
75 
76 	data->eoi_mmio = ioremap(data->eoi_page, 1u << data->esb_shift);
77 	if (!data->eoi_mmio) {
78 		pr_err("Failed to map EOI page for irq 0x%x\n", hw_irq);
79 		return -ENOMEM;
80 	}
81 
82 	data->hw_irq = hw_irq;
83 
84 	if (!data->trig_page)
85 		return 0;
86 	if (data->trig_page == data->eoi_page) {
87 		data->trig_mmio = data->eoi_mmio;
88 		return 0;
89 	}
90 
91 	data->trig_mmio = ioremap(data->trig_page, 1u << data->esb_shift);
92 	if (!data->trig_mmio) {
93 		pr_err("Failed to map trigger page for irq 0x%x\n", hw_irq);
94 		return -ENOMEM;
95 	}
96 	return 0;
97 }
98 EXPORT_SYMBOL_GPL(xive_native_populate_irq_data);
99 
100 int xive_native_configure_irq(u32 hw_irq, u32 target, u8 prio, u32 sw_irq)
101 {
102 	s64 rc;
103 
104 	for (;;) {
105 		rc = opal_xive_set_irq_config(hw_irq, target, prio, sw_irq);
106 		if (rc != OPAL_BUSY)
107 			break;
108 		msleep(OPAL_BUSY_DELAY_MS);
109 	}
110 	return rc == 0 ? 0 : -ENXIO;
111 }
112 EXPORT_SYMBOL_GPL(xive_native_configure_irq);
113 
114 static int xive_native_get_irq_config(u32 hw_irq, u32 *target, u8 *prio,
115 				      u32 *sw_irq)
116 {
117 	s64 rc;
118 	__be64 vp;
119 	__be32 lirq;
120 
121 	rc = opal_xive_get_irq_config(hw_irq, &vp, prio, &lirq);
122 
123 	*target = be64_to_cpu(vp);
124 	*sw_irq = be32_to_cpu(lirq);
125 
126 	return rc == 0 ? 0 : -ENXIO;
127 }
128 
129 /* This can be called multiple time to change a queue configuration */
130 int xive_native_configure_queue(u32 vp_id, struct xive_q *q, u8 prio,
131 				__be32 *qpage, u32 order, bool can_escalate)
132 {
133 	s64 rc = 0;
134 	__be64 qeoi_page_be;
135 	__be32 esc_irq_be;
136 	u64 flags, qpage_phys;
137 
138 	/* If there's an actual queue page, clean it */
139 	if (order) {
140 		if (WARN_ON(!qpage))
141 			return -EINVAL;
142 		qpage_phys = __pa(qpage);
143 	} else
144 		qpage_phys = 0;
145 
146 	/* Initialize the rest of the fields */
147 	q->msk = order ? ((1u << (order - 2)) - 1) : 0;
148 	q->idx = 0;
149 	q->toggle = 0;
150 
151 	rc = opal_xive_get_queue_info(vp_id, prio, NULL, NULL,
152 				      &qeoi_page_be,
153 				      &esc_irq_be,
154 				      NULL);
155 	if (rc) {
156 		pr_err("Error %lld getting queue info prio %d\n", rc, prio);
157 		rc = -EIO;
158 		goto fail;
159 	}
160 	q->eoi_phys = be64_to_cpu(qeoi_page_be);
161 
162 	/* Default flags */
163 	flags = OPAL_XIVE_EQ_ALWAYS_NOTIFY | OPAL_XIVE_EQ_ENABLED;
164 
165 	/* Escalation needed ? */
166 	if (can_escalate) {
167 		q->esc_irq = be32_to_cpu(esc_irq_be);
168 		flags |= OPAL_XIVE_EQ_ESCALATE;
169 	}
170 
171 	/* Configure and enable the queue in HW */
172 	for (;;) {
173 		rc = opal_xive_set_queue_info(vp_id, prio, qpage_phys, order, flags);
174 		if (rc != OPAL_BUSY)
175 			break;
176 		msleep(OPAL_BUSY_DELAY_MS);
177 	}
178 	if (rc) {
179 		pr_err("Error %lld setting queue for prio %d\n", rc, prio);
180 		rc = -EIO;
181 	} else {
182 		/*
183 		 * KVM code requires all of the above to be visible before
184 		 * q->qpage is set due to how it manages IPI EOIs
185 		 */
186 		wmb();
187 		q->qpage = qpage;
188 	}
189 fail:
190 	return rc;
191 }
192 EXPORT_SYMBOL_GPL(xive_native_configure_queue);
193 
194 static void __xive_native_disable_queue(u32 vp_id, struct xive_q *q, u8 prio)
195 {
196 	s64 rc;
197 
198 	/* Disable the queue in HW */
199 	for (;;) {
200 		rc = opal_xive_set_queue_info(vp_id, prio, 0, 0, 0);
201 		if (rc != OPAL_BUSY)
202 			break;
203 		msleep(OPAL_BUSY_DELAY_MS);
204 	}
205 	if (rc)
206 		pr_err("Error %lld disabling queue for prio %d\n", rc, prio);
207 }
208 
209 void xive_native_disable_queue(u32 vp_id, struct xive_q *q, u8 prio)
210 {
211 	__xive_native_disable_queue(vp_id, q, prio);
212 }
213 EXPORT_SYMBOL_GPL(xive_native_disable_queue);
214 
215 static int xive_native_setup_queue(unsigned int cpu, struct xive_cpu *xc, u8 prio)
216 {
217 	struct xive_q *q = &xc->queue[prio];
218 	__be32 *qpage;
219 
220 	qpage = xive_queue_page_alloc(cpu, xive_queue_shift);
221 	if (IS_ERR(qpage))
222 		return PTR_ERR(qpage);
223 
224 	return xive_native_configure_queue(get_hard_smp_processor_id(cpu),
225 					   q, prio, qpage, xive_queue_shift, false);
226 }
227 
228 static void xive_native_cleanup_queue(unsigned int cpu, struct xive_cpu *xc, u8 prio)
229 {
230 	struct xive_q *q = &xc->queue[prio];
231 	unsigned int alloc_order;
232 
233 	/*
234 	 * We use the variant with no iounmap as this is called on exec
235 	 * from an IPI and iounmap isn't safe
236 	 */
237 	__xive_native_disable_queue(get_hard_smp_processor_id(cpu), q, prio);
238 	alloc_order = xive_alloc_order(xive_queue_shift);
239 	free_pages((unsigned long)q->qpage, alloc_order);
240 	q->qpage = NULL;
241 }
242 
243 static bool xive_native_match(struct device_node *node)
244 {
245 	return of_device_is_compatible(node, "ibm,opal-xive-vc");
246 }
247 
248 static s64 opal_xive_allocate_irq(u32 chip_id)
249 {
250 	s64 irq = opal_xive_allocate_irq_raw(chip_id);
251 
252 	/*
253 	 * Old versions of skiboot can incorrectly return 0xffffffff to
254 	 * indicate no space, fix it up here.
255 	 */
256 	return irq == 0xffffffff ? OPAL_RESOURCE : irq;
257 }
258 
259 #ifdef CONFIG_SMP
260 static int xive_native_get_ipi(unsigned int cpu, struct xive_cpu *xc)
261 {
262 	s64 irq;
263 
264 	/* Allocate an IPI and populate info about it */
265 	for (;;) {
266 		irq = opal_xive_allocate_irq(xc->chip_id);
267 		if (irq == OPAL_BUSY) {
268 			msleep(OPAL_BUSY_DELAY_MS);
269 			continue;
270 		}
271 		if (irq < 0) {
272 			pr_err("Failed to allocate IPI on CPU %d\n", cpu);
273 			return -ENXIO;
274 		}
275 		xc->hw_ipi = irq;
276 		break;
277 	}
278 	return 0;
279 }
280 #endif /* CONFIG_SMP */
281 
282 u32 xive_native_alloc_irq(void)
283 {
284 	s64 rc;
285 
286 	for (;;) {
287 		rc = opal_xive_allocate_irq(OPAL_XIVE_ANY_CHIP);
288 		if (rc != OPAL_BUSY)
289 			break;
290 		msleep(OPAL_BUSY_DELAY_MS);
291 	}
292 	if (rc < 0)
293 		return 0;
294 	return rc;
295 }
296 EXPORT_SYMBOL_GPL(xive_native_alloc_irq);
297 
298 void xive_native_free_irq(u32 irq)
299 {
300 	for (;;) {
301 		s64 rc = opal_xive_free_irq(irq);
302 		if (rc != OPAL_BUSY)
303 			break;
304 		msleep(OPAL_BUSY_DELAY_MS);
305 	}
306 }
307 EXPORT_SYMBOL_GPL(xive_native_free_irq);
308 
309 #ifdef CONFIG_SMP
310 static void xive_native_put_ipi(unsigned int cpu, struct xive_cpu *xc)
311 {
312 	s64 rc;
313 
314 	/* Free the IPI */
315 	if (!xc->hw_ipi)
316 		return;
317 	for (;;) {
318 		rc = opal_xive_free_irq(xc->hw_ipi);
319 		if (rc == OPAL_BUSY) {
320 			msleep(OPAL_BUSY_DELAY_MS);
321 			continue;
322 		}
323 		xc->hw_ipi = 0;
324 		break;
325 	}
326 }
327 #endif /* CONFIG_SMP */
328 
329 static void xive_native_shutdown(void)
330 {
331 	/* Switch the XIVE to emulation mode */
332 	opal_xive_reset(OPAL_XIVE_MODE_EMU);
333 }
334 
335 /*
336  * Perform an "ack" cycle on the current thread, thus
337  * grabbing the pending active priorities and updating
338  * the CPPR to the most favored one.
339  */
340 static void xive_native_update_pending(struct xive_cpu *xc)
341 {
342 	u8 he, cppr;
343 	u16 ack;
344 
345 	/* Perform the acknowledge hypervisor to register cycle */
346 	ack = be16_to_cpu(__raw_readw(xive_tima + TM_SPC_ACK_HV_REG));
347 
348 	/* Synchronize subsequent queue accesses */
349 	mb();
350 
351 	/*
352 	 * Grab the CPPR and the "HE" field which indicates the source
353 	 * of the hypervisor interrupt (if any)
354 	 */
355 	cppr = ack & 0xff;
356 	he = (ack >> 8) >> 6;
357 	switch(he) {
358 	case TM_QW3_NSR_HE_NONE: /* Nothing to see here */
359 		break;
360 	case TM_QW3_NSR_HE_PHYS: /* Physical thread interrupt */
361 		if (cppr == 0xff)
362 			return;
363 		/* Mark the priority pending */
364 		xc->pending_prio |= 1 << cppr;
365 
366 		/*
367 		 * A new interrupt should never have a CPPR less favored
368 		 * than our current one.
369 		 */
370 		if (cppr >= xc->cppr)
371 			pr_err("CPU %d odd ack CPPR, got %d at %d\n",
372 			       smp_processor_id(), cppr, xc->cppr);
373 
374 		/* Update our idea of what the CPPR is */
375 		xc->cppr = cppr;
376 		break;
377 	case TM_QW3_NSR_HE_POOL: /* HV Pool interrupt (unused) */
378 	case TM_QW3_NSR_HE_LSI:  /* Legacy FW LSI (unused) */
379 		pr_err("CPU %d got unexpected interrupt type HE=%d\n",
380 		       smp_processor_id(), he);
381 		return;
382 	}
383 }
384 
385 static void xive_native_eoi(u32 hw_irq)
386 {
387 	/*
388 	 * Not normally used except if specific interrupts need
389 	 * a workaround on EOI.
390 	 */
391 	opal_int_eoi(hw_irq);
392 }
393 
394 static void xive_native_setup_cpu(unsigned int cpu, struct xive_cpu *xc)
395 {
396 	s64 rc;
397 	u32 vp;
398 	__be64 vp_cam_be;
399 	u64 vp_cam;
400 
401 	if (xive_pool_vps == XIVE_INVALID_VP)
402 		return;
403 
404 	/* Check if pool VP already active, if it is, pull it */
405 	if (in_be32(xive_tima + TM_QW2_HV_POOL + TM_WORD2) & TM_QW2W2_VP)
406 		in_be64(xive_tima + TM_SPC_PULL_POOL_CTX);
407 
408 	/* Enable the pool VP */
409 	vp = xive_pool_vps + cpu;
410 	for (;;) {
411 		rc = opal_xive_set_vp_info(vp, OPAL_XIVE_VP_ENABLED, 0);
412 		if (rc != OPAL_BUSY)
413 			break;
414 		msleep(OPAL_BUSY_DELAY_MS);
415 	}
416 	if (rc) {
417 		pr_err("Failed to enable pool VP on CPU %d\n", cpu);
418 		return;
419 	}
420 
421 	/* Grab it's CAM value */
422 	rc = opal_xive_get_vp_info(vp, NULL, &vp_cam_be, NULL, NULL);
423 	if (rc) {
424 		pr_err("Failed to get pool VP info CPU %d\n", cpu);
425 		return;
426 	}
427 	vp_cam = be64_to_cpu(vp_cam_be);
428 
429 	/* Push it on the CPU (set LSMFB to 0xff to skip backlog scan) */
430 	out_be32(xive_tima + TM_QW2_HV_POOL + TM_WORD0, 0xff);
431 	out_be32(xive_tima + TM_QW2_HV_POOL + TM_WORD2, TM_QW2W2_VP | vp_cam);
432 }
433 
434 static void xive_native_teardown_cpu(unsigned int cpu, struct xive_cpu *xc)
435 {
436 	s64 rc;
437 	u32 vp;
438 
439 	if (xive_pool_vps == XIVE_INVALID_VP)
440 		return;
441 
442 	/* Pull the pool VP from the CPU */
443 	in_be64(xive_tima + TM_SPC_PULL_POOL_CTX);
444 
445 	/* Disable it */
446 	vp = xive_pool_vps + cpu;
447 	for (;;) {
448 		rc = opal_xive_set_vp_info(vp, 0, 0);
449 		if (rc != OPAL_BUSY)
450 			break;
451 		msleep(OPAL_BUSY_DELAY_MS);
452 	}
453 }
454 
455 void xive_native_sync_source(u32 hw_irq)
456 {
457 	opal_xive_sync(XIVE_SYNC_EAS, hw_irq);
458 }
459 EXPORT_SYMBOL_GPL(xive_native_sync_source);
460 
461 void xive_native_sync_queue(u32 hw_irq)
462 {
463 	opal_xive_sync(XIVE_SYNC_QUEUE, hw_irq);
464 }
465 EXPORT_SYMBOL_GPL(xive_native_sync_queue);
466 
467 static const struct xive_ops xive_native_ops = {
468 	.populate_irq_data	= xive_native_populate_irq_data,
469 	.configure_irq		= xive_native_configure_irq,
470 	.get_irq_config		= xive_native_get_irq_config,
471 	.setup_queue		= xive_native_setup_queue,
472 	.cleanup_queue		= xive_native_cleanup_queue,
473 	.match			= xive_native_match,
474 	.shutdown		= xive_native_shutdown,
475 	.update_pending		= xive_native_update_pending,
476 	.eoi			= xive_native_eoi,
477 	.setup_cpu		= xive_native_setup_cpu,
478 	.teardown_cpu		= xive_native_teardown_cpu,
479 	.sync_source		= xive_native_sync_source,
480 #ifdef CONFIG_SMP
481 	.get_ipi		= xive_native_get_ipi,
482 	.put_ipi		= xive_native_put_ipi,
483 #endif /* CONFIG_SMP */
484 	.name			= "native",
485 };
486 
487 static bool xive_parse_provisioning(struct device_node *np)
488 {
489 	int rc;
490 
491 	if (of_property_read_u32(np, "ibm,xive-provision-page-size",
492 				 &xive_provision_size) < 0)
493 		return true;
494 	rc = of_property_count_elems_of_size(np, "ibm,xive-provision-chips", 4);
495 	if (rc < 0) {
496 		pr_err("Error %d getting provision chips array\n", rc);
497 		return false;
498 	}
499 	xive_provision_chip_count = rc;
500 	if (rc == 0)
501 		return true;
502 
503 	xive_provision_chips = kcalloc(4, xive_provision_chip_count,
504 				       GFP_KERNEL);
505 	if (WARN_ON(!xive_provision_chips))
506 		return false;
507 
508 	rc = of_property_read_u32_array(np, "ibm,xive-provision-chips",
509 					xive_provision_chips,
510 					xive_provision_chip_count);
511 	if (rc < 0) {
512 		pr_err("Error %d reading provision chips array\n", rc);
513 		return false;
514 	}
515 
516 	xive_provision_cache = kmem_cache_create("xive-provision",
517 						 xive_provision_size,
518 						 xive_provision_size,
519 						 0, NULL);
520 	if (!xive_provision_cache) {
521 		pr_err("Failed to allocate provision cache\n");
522 		return false;
523 	}
524 	return true;
525 }
526 
527 static void xive_native_setup_pools(void)
528 {
529 	/* Allocate a pool big enough */
530 	pr_debug("XIVE: Allocating VP block for pool size %u\n", nr_cpu_ids);
531 
532 	xive_pool_vps = xive_native_alloc_vp_block(nr_cpu_ids);
533 	if (WARN_ON(xive_pool_vps == XIVE_INVALID_VP))
534 		pr_err("XIVE: Failed to allocate pool VP, KVM might not function\n");
535 
536 	pr_debug("XIVE: Pool VPs allocated at 0x%x for %u max CPUs\n",
537 		 xive_pool_vps, nr_cpu_ids);
538 }
539 
540 u32 xive_native_default_eq_shift(void)
541 {
542 	return xive_queue_shift;
543 }
544 EXPORT_SYMBOL_GPL(xive_native_default_eq_shift);
545 
546 unsigned long xive_tima_os;
547 EXPORT_SYMBOL_GPL(xive_tima_os);
548 
549 bool __init xive_native_init(void)
550 {
551 	struct device_node *np;
552 	struct resource r;
553 	void __iomem *tima;
554 	struct property *prop;
555 	u8 max_prio = 7;
556 	const __be32 *p;
557 	u32 val, cpu;
558 	s64 rc;
559 
560 	if (xive_cmdline_disabled)
561 		return false;
562 
563 	pr_devel("xive_native_init()\n");
564 	np = of_find_compatible_node(NULL, NULL, "ibm,opal-xive-pe");
565 	if (!np) {
566 		pr_devel("not found !\n");
567 		return false;
568 	}
569 	pr_devel("Found %pOF\n", np);
570 
571 	/* Resource 1 is HV window */
572 	if (of_address_to_resource(np, 1, &r)) {
573 		pr_err("Failed to get thread mgmnt area resource\n");
574 		return false;
575 	}
576 	tima = ioremap(r.start, resource_size(&r));
577 	if (!tima) {
578 		pr_err("Failed to map thread mgmnt area\n");
579 		return false;
580 	}
581 
582 	/* Read number of priorities */
583 	if (of_property_read_u32(np, "ibm,xive-#priorities", &val) == 0)
584 		max_prio = val - 1;
585 
586 	/* Iterate the EQ sizes and pick one */
587 	of_property_for_each_u32(np, "ibm,xive-eq-sizes", prop, p, val) {
588 		xive_queue_shift = val;
589 		if (val == PAGE_SHIFT)
590 			break;
591 	}
592 
593 	/* Do we support single escalation */
594 	if (of_get_property(np, "single-escalation-support", NULL) != NULL)
595 		xive_has_single_esc = true;
596 
597 	/* Configure Thread Management areas for KVM */
598 	for_each_possible_cpu(cpu)
599 		kvmppc_set_xive_tima(cpu, r.start, tima);
600 
601 	/* Resource 2 is OS window */
602 	if (of_address_to_resource(np, 2, &r)) {
603 		pr_err("Failed to get thread mgmnt area resource\n");
604 		return false;
605 	}
606 
607 	xive_tima_os = r.start;
608 
609 	/* Grab size of provisionning pages */
610 	xive_parse_provisioning(np);
611 
612 	/* Switch the XIVE to exploitation mode */
613 	rc = opal_xive_reset(OPAL_XIVE_MODE_EXPL);
614 	if (rc) {
615 		pr_err("Switch to exploitation mode failed with error %lld\n", rc);
616 		return false;
617 	}
618 
619 	/* Setup some dummy HV pool VPs */
620 	xive_native_setup_pools();
621 
622 	/* Initialize XIVE core with our backend */
623 	if (!xive_core_init(&xive_native_ops, tima, TM_QW3_HV_PHYS,
624 			    max_prio)) {
625 		opal_xive_reset(OPAL_XIVE_MODE_EMU);
626 		return false;
627 	}
628 	pr_info("Using %dkB queues\n", 1 << (xive_queue_shift - 10));
629 	return true;
630 }
631 
632 static bool xive_native_provision_pages(void)
633 {
634 	u32 i;
635 	void *p;
636 
637 	for (i = 0; i < xive_provision_chip_count; i++) {
638 		u32 chip = xive_provision_chips[i];
639 
640 		/*
641 		 * XXX TODO: Try to make the allocation local to the node where
642 		 * the chip resides.
643 		 */
644 		p = kmem_cache_alloc(xive_provision_cache, GFP_KERNEL);
645 		if (!p) {
646 			pr_err("Failed to allocate provisioning page\n");
647 			return false;
648 		}
649 		opal_xive_donate_page(chip, __pa(p));
650 	}
651 	return true;
652 }
653 
654 u32 xive_native_alloc_vp_block(u32 max_vcpus)
655 {
656 	s64 rc;
657 	u32 order;
658 
659 	order = fls(max_vcpus) - 1;
660 	if (max_vcpus > (1 << order))
661 		order++;
662 
663 	pr_debug("VP block alloc, for max VCPUs %d use order %d\n",
664 		 max_vcpus, order);
665 
666 	for (;;) {
667 		rc = opal_xive_alloc_vp_block(order);
668 		switch (rc) {
669 		case OPAL_BUSY:
670 			msleep(OPAL_BUSY_DELAY_MS);
671 			break;
672 		case OPAL_XIVE_PROVISIONING:
673 			if (!xive_native_provision_pages())
674 				return XIVE_INVALID_VP;
675 			break;
676 		default:
677 			if (rc < 0) {
678 				pr_err("OPAL failed to allocate VCPUs order %d, err %lld\n",
679 				       order, rc);
680 				return XIVE_INVALID_VP;
681 			}
682 			return rc;
683 		}
684 	}
685 }
686 EXPORT_SYMBOL_GPL(xive_native_alloc_vp_block);
687 
688 void xive_native_free_vp_block(u32 vp_base)
689 {
690 	s64 rc;
691 
692 	if (vp_base == XIVE_INVALID_VP)
693 		return;
694 
695 	rc = opal_xive_free_vp_block(vp_base);
696 	if (rc < 0)
697 		pr_warn("OPAL error %lld freeing VP block\n", rc);
698 }
699 EXPORT_SYMBOL_GPL(xive_native_free_vp_block);
700 
701 int xive_native_enable_vp(u32 vp_id, bool single_escalation)
702 {
703 	s64 rc;
704 	u64 flags = OPAL_XIVE_VP_ENABLED;
705 
706 	if (single_escalation)
707 		flags |= OPAL_XIVE_VP_SINGLE_ESCALATION;
708 	for (;;) {
709 		rc = opal_xive_set_vp_info(vp_id, flags, 0);
710 		if (rc != OPAL_BUSY)
711 			break;
712 		msleep(OPAL_BUSY_DELAY_MS);
713 	}
714 	return rc ? -EIO : 0;
715 }
716 EXPORT_SYMBOL_GPL(xive_native_enable_vp);
717 
718 int xive_native_disable_vp(u32 vp_id)
719 {
720 	s64 rc;
721 
722 	for (;;) {
723 		rc = opal_xive_set_vp_info(vp_id, 0, 0);
724 		if (rc != OPAL_BUSY)
725 			break;
726 		msleep(OPAL_BUSY_DELAY_MS);
727 	}
728 	return rc ? -EIO : 0;
729 }
730 EXPORT_SYMBOL_GPL(xive_native_disable_vp);
731 
732 int xive_native_get_vp_info(u32 vp_id, u32 *out_cam_id, u32 *out_chip_id)
733 {
734 	__be64 vp_cam_be;
735 	__be32 vp_chip_id_be;
736 	s64 rc;
737 
738 	rc = opal_xive_get_vp_info(vp_id, NULL, &vp_cam_be, NULL, &vp_chip_id_be);
739 	if (rc)
740 		return -EIO;
741 	*out_cam_id = be64_to_cpu(vp_cam_be) & 0xffffffffu;
742 	*out_chip_id = be32_to_cpu(vp_chip_id_be);
743 
744 	return 0;
745 }
746 EXPORT_SYMBOL_GPL(xive_native_get_vp_info);
747 
748 bool xive_native_has_single_escalation(void)
749 {
750 	return xive_has_single_esc;
751 }
752 EXPORT_SYMBOL_GPL(xive_native_has_single_escalation);
753 
754 int xive_native_get_queue_info(u32 vp_id, u32 prio,
755 			       u64 *out_qpage,
756 			       u64 *out_qsize,
757 			       u64 *out_qeoi_page,
758 			       u32 *out_escalate_irq,
759 			       u64 *out_qflags)
760 {
761 	__be64 qpage;
762 	__be64 qsize;
763 	__be64 qeoi_page;
764 	__be32 escalate_irq;
765 	__be64 qflags;
766 	s64 rc;
767 
768 	rc = opal_xive_get_queue_info(vp_id, prio, &qpage, &qsize,
769 				      &qeoi_page, &escalate_irq, &qflags);
770 	if (rc) {
771 		pr_err("OPAL failed to get queue info for VCPU %d/%d : %lld\n",
772 		       vp_id, prio, rc);
773 		return -EIO;
774 	}
775 
776 	if (out_qpage)
777 		*out_qpage = be64_to_cpu(qpage);
778 	if (out_qsize)
779 		*out_qsize = be32_to_cpu(qsize);
780 	if (out_qeoi_page)
781 		*out_qeoi_page = be64_to_cpu(qeoi_page);
782 	if (out_escalate_irq)
783 		*out_escalate_irq = be32_to_cpu(escalate_irq);
784 	if (out_qflags)
785 		*out_qflags = be64_to_cpu(qflags);
786 
787 	return 0;
788 }
789 EXPORT_SYMBOL_GPL(xive_native_get_queue_info);
790 
791 int xive_native_get_queue_state(u32 vp_id, u32 prio, u32 *qtoggle, u32 *qindex)
792 {
793 	__be32 opal_qtoggle;
794 	__be32 opal_qindex;
795 	s64 rc;
796 
797 	rc = opal_xive_get_queue_state(vp_id, prio, &opal_qtoggle,
798 				       &opal_qindex);
799 	if (rc) {
800 		pr_err("OPAL failed to get queue state for VCPU %d/%d : %lld\n",
801 		       vp_id, prio, rc);
802 		return -EIO;
803 	}
804 
805 	if (qtoggle)
806 		*qtoggle = be32_to_cpu(opal_qtoggle);
807 	if (qindex)
808 		*qindex = be32_to_cpu(opal_qindex);
809 
810 	return 0;
811 }
812 EXPORT_SYMBOL_GPL(xive_native_get_queue_state);
813 
814 int xive_native_set_queue_state(u32 vp_id, u32 prio, u32 qtoggle, u32 qindex)
815 {
816 	s64 rc;
817 
818 	rc = opal_xive_set_queue_state(vp_id, prio, qtoggle, qindex);
819 	if (rc) {
820 		pr_err("OPAL failed to set queue state for VCPU %d/%d : %lld\n",
821 		       vp_id, prio, rc);
822 		return -EIO;
823 	}
824 
825 	return 0;
826 }
827 EXPORT_SYMBOL_GPL(xive_native_set_queue_state);
828 
829 bool xive_native_has_queue_state_support(void)
830 {
831 	return opal_check_token(OPAL_XIVE_GET_QUEUE_STATE) &&
832 		opal_check_token(OPAL_XIVE_SET_QUEUE_STATE);
833 }
834 EXPORT_SYMBOL_GPL(xive_native_has_queue_state_support);
835 
836 int xive_native_get_vp_state(u32 vp_id, u64 *out_state)
837 {
838 	__be64 state;
839 	s64 rc;
840 
841 	rc = opal_xive_get_vp_state(vp_id, &state);
842 	if (rc) {
843 		pr_err("OPAL failed to get vp state for VCPU %d : %lld\n",
844 		       vp_id, rc);
845 		return -EIO;
846 	}
847 
848 	if (out_state)
849 		*out_state = be64_to_cpu(state);
850 	return 0;
851 }
852 EXPORT_SYMBOL_GPL(xive_native_get_vp_state);
853