xref: /openbmc/linux/arch/powerpc/sysdev/xive/native.c (revision 0352f880)
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/of_address.h>
17 #include <linux/slab.h>
18 #include <linux/spinlock.h>
19 #include <linux/delay.h>
20 #include <linux/cpumask.h>
21 #include <linux/mm.h>
22 #include <linux/kmemleak.h>
23 
24 #include <asm/machdep.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/opal.h>
32 #include <asm/kvm_ppc.h>
33 
34 #include "xive-internal.h"
35 
36 
37 static u32 xive_provision_size;
38 static u32 *xive_provision_chips;
39 static u32 xive_provision_chip_count;
40 static u32 xive_queue_shift;
41 static u32 xive_pool_vps = XIVE_INVALID_VP;
42 static struct kmem_cache *xive_provision_cache;
43 static bool xive_has_single_esc;
44 bool xive_has_save_restore;
45 
46 int xive_native_populate_irq_data(u32 hw_irq, struct xive_irq_data *data)
47 {
48 	__be64 flags, eoi_page, trig_page;
49 	__be32 esb_shift, src_chip;
50 	u64 opal_flags;
51 	s64 rc;
52 
53 	memset(data, 0, sizeof(*data));
54 
55 	rc = opal_xive_get_irq_info(hw_irq, &flags, &eoi_page, &trig_page,
56 				    &esb_shift, &src_chip);
57 	if (rc) {
58 		pr_err("opal_xive_get_irq_info(0x%x) returned %lld\n",
59 		       hw_irq, rc);
60 		return -EINVAL;
61 	}
62 
63 	opal_flags = be64_to_cpu(flags);
64 	if (opal_flags & OPAL_XIVE_IRQ_STORE_EOI)
65 		data->flags |= XIVE_IRQ_FLAG_STORE_EOI;
66 	if (opal_flags & OPAL_XIVE_IRQ_STORE_EOI2)
67 		data->flags |= XIVE_IRQ_FLAG_STORE_EOI;
68 	if (opal_flags & OPAL_XIVE_IRQ_LSI)
69 		data->flags |= XIVE_IRQ_FLAG_LSI;
70 	data->eoi_page = be64_to_cpu(eoi_page);
71 	data->trig_page = be64_to_cpu(trig_page);
72 	data->esb_shift = be32_to_cpu(esb_shift);
73 	data->src_chip = be32_to_cpu(src_chip);
74 
75 	data->eoi_mmio = ioremap(data->eoi_page, 1u << data->esb_shift);
76 	if (!data->eoi_mmio) {
77 		pr_err("Failed to map EOI page for irq 0x%x\n", hw_irq);
78 		return -ENOMEM;
79 	}
80 
81 	data->hw_irq = hw_irq;
82 
83 	if (!data->trig_page)
84 		return 0;
85 	if (data->trig_page == data->eoi_page) {
86 		data->trig_mmio = data->eoi_mmio;
87 		return 0;
88 	}
89 
90 	data->trig_mmio = ioremap(data->trig_page, 1u << data->esb_shift);
91 	if (!data->trig_mmio) {
92 		pr_err("Failed to map trigger page for irq 0x%x\n", hw_irq);
93 		return -ENOMEM;
94 	}
95 	return 0;
96 }
97 EXPORT_SYMBOL_GPL(xive_native_populate_irq_data);
98 
99 int xive_native_configure_irq(u32 hw_irq, u32 target, u8 prio, u32 sw_irq)
100 {
101 	s64 rc;
102 
103 	for (;;) {
104 		rc = opal_xive_set_irq_config(hw_irq, target, prio, sw_irq);
105 		if (rc != OPAL_BUSY)
106 			break;
107 		msleep(OPAL_BUSY_DELAY_MS);
108 	}
109 	return rc == 0 ? 0 : -ENXIO;
110 }
111 EXPORT_SYMBOL_GPL(xive_native_configure_irq);
112 
113 static int xive_native_get_irq_config(u32 hw_irq, u32 *target, u8 *prio,
114 				      u32 *sw_irq)
115 {
116 	s64 rc;
117 	__be64 vp;
118 	__be32 lirq;
119 
120 	rc = opal_xive_get_irq_config(hw_irq, &vp, prio, &lirq);
121 
122 	*target = be64_to_cpu(vp);
123 	*sw_irq = be32_to_cpu(lirq);
124 
125 	return rc == 0 ? 0 : -ENXIO;
126 }
127 
128 #define vp_err(vp, fmt, ...) pr_err("VP[0x%x]: " fmt, vp, ##__VA_ARGS__)
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 		vp_err(vp_id, "Failed to get queue %d info : %lld\n", prio, rc);
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 		vp_err(vp_id, "Failed to set queue %d info: %lld\n", prio, rc);
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 		vp_err(vp_id, "Failed to disable queue %d : %lld\n", prio, rc);
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_on_chip(u32 chip_id)
284 {
285 	s64 rc;
286 
287 	for (;;) {
288 		rc = opal_xive_allocate_irq(chip_id);
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_on_chip);
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_prepare_cpu(unsigned int cpu, struct xive_cpu *xc)
387 {
388 	xc->chip_id = cpu_to_chip_id(cpu);
389 }
390 
391 static void xive_native_setup_cpu(unsigned int cpu, struct xive_cpu *xc)
392 {
393 	s64 rc;
394 	u32 vp;
395 	__be64 vp_cam_be;
396 	u64 vp_cam;
397 
398 	if (xive_pool_vps == XIVE_INVALID_VP)
399 		return;
400 
401 	/* Check if pool VP already active, if it is, pull it */
402 	if (in_be32(xive_tima + TM_QW2_HV_POOL + TM_WORD2) & TM_QW2W2_VP)
403 		in_be64(xive_tima + TM_SPC_PULL_POOL_CTX);
404 
405 	/* Enable the pool VP */
406 	vp = xive_pool_vps + cpu;
407 	for (;;) {
408 		rc = opal_xive_set_vp_info(vp, OPAL_XIVE_VP_ENABLED, 0);
409 		if (rc != OPAL_BUSY)
410 			break;
411 		msleep(OPAL_BUSY_DELAY_MS);
412 	}
413 	if (rc) {
414 		pr_err("Failed to enable pool VP on CPU %d\n", cpu);
415 		return;
416 	}
417 
418 	/* Grab it's CAM value */
419 	rc = opal_xive_get_vp_info(vp, NULL, &vp_cam_be, NULL, NULL);
420 	if (rc) {
421 		pr_err("Failed to get pool VP info CPU %d\n", cpu);
422 		return;
423 	}
424 	vp_cam = be64_to_cpu(vp_cam_be);
425 
426 	/* Push it on the CPU (set LSMFB to 0xff to skip backlog scan) */
427 	out_be32(xive_tima + TM_QW2_HV_POOL + TM_WORD0, 0xff);
428 	out_be32(xive_tima + TM_QW2_HV_POOL + TM_WORD2, TM_QW2W2_VP | vp_cam);
429 }
430 
431 static void xive_native_teardown_cpu(unsigned int cpu, struct xive_cpu *xc)
432 {
433 	s64 rc;
434 	u32 vp;
435 
436 	if (xive_pool_vps == XIVE_INVALID_VP)
437 		return;
438 
439 	/* Pull the pool VP from the CPU */
440 	in_be64(xive_tima + TM_SPC_PULL_POOL_CTX);
441 
442 	/* Disable it */
443 	vp = xive_pool_vps + cpu;
444 	for (;;) {
445 		rc = opal_xive_set_vp_info(vp, 0, 0);
446 		if (rc != OPAL_BUSY)
447 			break;
448 		msleep(OPAL_BUSY_DELAY_MS);
449 	}
450 }
451 
452 void xive_native_sync_source(u32 hw_irq)
453 {
454 	opal_xive_sync(XIVE_SYNC_EAS, hw_irq);
455 }
456 EXPORT_SYMBOL_GPL(xive_native_sync_source);
457 
458 void xive_native_sync_queue(u32 hw_irq)
459 {
460 	opal_xive_sync(XIVE_SYNC_QUEUE, hw_irq);
461 }
462 EXPORT_SYMBOL_GPL(xive_native_sync_queue);
463 
464 #ifdef CONFIG_DEBUG_FS
465 static int xive_native_debug_create(struct dentry *xive_dir)
466 {
467 	debugfs_create_bool("save-restore", 0600, xive_dir, &xive_has_save_restore);
468 	return 0;
469 }
470 #endif
471 
472 static const struct xive_ops xive_native_ops = {
473 	.populate_irq_data	= xive_native_populate_irq_data,
474 	.configure_irq		= xive_native_configure_irq,
475 	.get_irq_config		= xive_native_get_irq_config,
476 	.setup_queue		= xive_native_setup_queue,
477 	.cleanup_queue		= xive_native_cleanup_queue,
478 	.match			= xive_native_match,
479 	.shutdown		= xive_native_shutdown,
480 	.update_pending		= xive_native_update_pending,
481 	.prepare_cpu		= xive_native_prepare_cpu,
482 	.setup_cpu		= xive_native_setup_cpu,
483 	.teardown_cpu		= xive_native_teardown_cpu,
484 	.sync_source		= xive_native_sync_source,
485 #ifdef CONFIG_SMP
486 	.get_ipi		= xive_native_get_ipi,
487 	.put_ipi		= xive_native_put_ipi,
488 #endif /* CONFIG_SMP */
489 #ifdef CONFIG_DEBUG_FS
490 	.debug_create		= xive_native_debug_create,
491 #endif /* CONFIG_DEBUG_FS */
492 	.name			= "native",
493 };
494 
495 static bool __init xive_parse_provisioning(struct device_node *np)
496 {
497 	int rc;
498 
499 	if (of_property_read_u32(np, "ibm,xive-provision-page-size",
500 				 &xive_provision_size) < 0)
501 		return true;
502 	rc = of_property_count_elems_of_size(np, "ibm,xive-provision-chips", 4);
503 	if (rc < 0) {
504 		pr_err("Error %d getting provision chips array\n", rc);
505 		return false;
506 	}
507 	xive_provision_chip_count = rc;
508 	if (rc == 0)
509 		return true;
510 
511 	xive_provision_chips = kcalloc(4, xive_provision_chip_count,
512 				       GFP_KERNEL);
513 	if (WARN_ON(!xive_provision_chips))
514 		return false;
515 
516 	rc = of_property_read_u32_array(np, "ibm,xive-provision-chips",
517 					xive_provision_chips,
518 					xive_provision_chip_count);
519 	if (rc < 0) {
520 		pr_err("Error %d reading provision chips array\n", rc);
521 		return false;
522 	}
523 
524 	xive_provision_cache = kmem_cache_create("xive-provision",
525 						 xive_provision_size,
526 						 xive_provision_size,
527 						 0, NULL);
528 	if (!xive_provision_cache) {
529 		pr_err("Failed to allocate provision cache\n");
530 		return false;
531 	}
532 	return true;
533 }
534 
535 static void __init xive_native_setup_pools(void)
536 {
537 	/* Allocate a pool big enough */
538 	pr_debug("XIVE: Allocating VP block for pool size %u\n", nr_cpu_ids);
539 
540 	xive_pool_vps = xive_native_alloc_vp_block(nr_cpu_ids);
541 	if (WARN_ON(xive_pool_vps == XIVE_INVALID_VP))
542 		pr_err("XIVE: Failed to allocate pool VP, KVM might not function\n");
543 
544 	pr_debug("XIVE: Pool VPs allocated at 0x%x for %u max CPUs\n",
545 		 xive_pool_vps, nr_cpu_ids);
546 }
547 
548 u32 xive_native_default_eq_shift(void)
549 {
550 	return xive_queue_shift;
551 }
552 EXPORT_SYMBOL_GPL(xive_native_default_eq_shift);
553 
554 unsigned long xive_tima_os;
555 EXPORT_SYMBOL_GPL(xive_tima_os);
556 
557 bool __init xive_native_init(void)
558 {
559 	struct device_node *np;
560 	struct resource r;
561 	void __iomem *tima;
562 	struct property *prop;
563 	u8 max_prio = 7;
564 	const __be32 *p;
565 	u32 val, cpu;
566 	s64 rc;
567 
568 	if (xive_cmdline_disabled)
569 		return false;
570 
571 	pr_devel("xive_native_init()\n");
572 	np = of_find_compatible_node(NULL, NULL, "ibm,opal-xive-pe");
573 	if (!np) {
574 		pr_devel("not found !\n");
575 		return false;
576 	}
577 	pr_devel("Found %pOF\n", np);
578 
579 	/* Resource 1 is HV window */
580 	if (of_address_to_resource(np, 1, &r)) {
581 		pr_err("Failed to get thread mgmnt area resource\n");
582 		goto err_put;
583 	}
584 	tima = ioremap(r.start, resource_size(&r));
585 	if (!tima) {
586 		pr_err("Failed to map thread mgmnt area\n");
587 		goto err_put;
588 	}
589 
590 	/* Read number of priorities */
591 	if (of_property_read_u32(np, "ibm,xive-#priorities", &val) == 0)
592 		max_prio = val - 1;
593 
594 	/* Iterate the EQ sizes and pick one */
595 	of_property_for_each_u32(np, "ibm,xive-eq-sizes", prop, p, val) {
596 		xive_queue_shift = val;
597 		if (val == PAGE_SHIFT)
598 			break;
599 	}
600 
601 	/* Do we support single escalation */
602 	if (of_get_property(np, "single-escalation-support", NULL) != NULL)
603 		xive_has_single_esc = true;
604 
605 	if (of_get_property(np, "vp-save-restore", NULL))
606 		xive_has_save_restore = true;
607 
608 	/* Configure Thread Management areas for KVM */
609 	for_each_possible_cpu(cpu)
610 		kvmppc_set_xive_tima(cpu, r.start, tima);
611 
612 	/* Resource 2 is OS window */
613 	if (of_address_to_resource(np, 2, &r)) {
614 		pr_err("Failed to get thread mgmnt area resource\n");
615 		goto err_put;
616 	}
617 
618 	xive_tima_os = r.start;
619 
620 	/* Grab size of provisioning pages */
621 	xive_parse_provisioning(np);
622 
623 	/* Switch the XIVE to exploitation mode */
624 	rc = opal_xive_reset(OPAL_XIVE_MODE_EXPL);
625 	if (rc) {
626 		pr_err("Switch to exploitation mode failed with error %lld\n", rc);
627 		goto err_put;
628 	}
629 
630 	/* Setup some dummy HV pool VPs */
631 	xive_native_setup_pools();
632 
633 	/* Initialize XIVE core with our backend */
634 	if (!xive_core_init(np, &xive_native_ops, tima, TM_QW3_HV_PHYS,
635 			    max_prio)) {
636 		opal_xive_reset(OPAL_XIVE_MODE_EMU);
637 		goto err_put;
638 	}
639 	of_node_put(np);
640 	pr_info("Using %dkB queues\n", 1 << (xive_queue_shift - 10));
641 	return true;
642 
643 err_put:
644 	of_node_put(np);
645 	return false;
646 }
647 
648 static bool xive_native_provision_pages(void)
649 {
650 	u32 i;
651 	void *p;
652 
653 	for (i = 0; i < xive_provision_chip_count; i++) {
654 		u32 chip = xive_provision_chips[i];
655 
656 		/*
657 		 * XXX TODO: Try to make the allocation local to the node where
658 		 * the chip resides.
659 		 */
660 		p = kmem_cache_alloc(xive_provision_cache, GFP_KERNEL);
661 		if (!p) {
662 			pr_err("Failed to allocate provisioning page\n");
663 			return false;
664 		}
665 		kmemleak_ignore(p);
666 		opal_xive_donate_page(chip, __pa(p));
667 	}
668 	return true;
669 }
670 
671 u32 xive_native_alloc_vp_block(u32 max_vcpus)
672 {
673 	s64 rc;
674 	u32 order;
675 
676 	order = fls(max_vcpus) - 1;
677 	if (max_vcpus > (1 << order))
678 		order++;
679 
680 	pr_debug("VP block alloc, for max VCPUs %d use order %d\n",
681 		 max_vcpus, order);
682 
683 	for (;;) {
684 		rc = opal_xive_alloc_vp_block(order);
685 		switch (rc) {
686 		case OPAL_BUSY:
687 			msleep(OPAL_BUSY_DELAY_MS);
688 			break;
689 		case OPAL_XIVE_PROVISIONING:
690 			if (!xive_native_provision_pages())
691 				return XIVE_INVALID_VP;
692 			break;
693 		default:
694 			if (rc < 0) {
695 				pr_err("OPAL failed to allocate VCPUs order %d, err %lld\n",
696 				       order, rc);
697 				return XIVE_INVALID_VP;
698 			}
699 			return rc;
700 		}
701 	}
702 }
703 EXPORT_SYMBOL_GPL(xive_native_alloc_vp_block);
704 
705 void xive_native_free_vp_block(u32 vp_base)
706 {
707 	s64 rc;
708 
709 	if (vp_base == XIVE_INVALID_VP)
710 		return;
711 
712 	rc = opal_xive_free_vp_block(vp_base);
713 	if (rc < 0)
714 		pr_warn("OPAL error %lld freeing VP block\n", rc);
715 }
716 EXPORT_SYMBOL_GPL(xive_native_free_vp_block);
717 
718 int xive_native_enable_vp(u32 vp_id, bool single_escalation)
719 {
720 	s64 rc;
721 	u64 flags = OPAL_XIVE_VP_ENABLED;
722 
723 	if (single_escalation)
724 		flags |= OPAL_XIVE_VP_SINGLE_ESCALATION;
725 	for (;;) {
726 		rc = opal_xive_set_vp_info(vp_id, flags, 0);
727 		if (rc != OPAL_BUSY)
728 			break;
729 		msleep(OPAL_BUSY_DELAY_MS);
730 	}
731 	if (rc)
732 		vp_err(vp_id, "Failed to enable VP : %lld\n", rc);
733 	return rc ? -EIO : 0;
734 }
735 EXPORT_SYMBOL_GPL(xive_native_enable_vp);
736 
737 int xive_native_disable_vp(u32 vp_id)
738 {
739 	s64 rc;
740 
741 	for (;;) {
742 		rc = opal_xive_set_vp_info(vp_id, 0, 0);
743 		if (rc != OPAL_BUSY)
744 			break;
745 		msleep(OPAL_BUSY_DELAY_MS);
746 	}
747 	if (rc)
748 		vp_err(vp_id, "Failed to disable VP : %lld\n", rc);
749 	return rc ? -EIO : 0;
750 }
751 EXPORT_SYMBOL_GPL(xive_native_disable_vp);
752 
753 int xive_native_get_vp_info(u32 vp_id, u32 *out_cam_id, u32 *out_chip_id)
754 {
755 	__be64 vp_cam_be;
756 	__be32 vp_chip_id_be;
757 	s64 rc;
758 
759 	rc = opal_xive_get_vp_info(vp_id, NULL, &vp_cam_be, NULL, &vp_chip_id_be);
760 	if (rc) {
761 		vp_err(vp_id, "Failed to get VP info : %lld\n", rc);
762 		return -EIO;
763 	}
764 	*out_cam_id = be64_to_cpu(vp_cam_be) & 0xffffffffu;
765 	*out_chip_id = be32_to_cpu(vp_chip_id_be);
766 
767 	return 0;
768 }
769 EXPORT_SYMBOL_GPL(xive_native_get_vp_info);
770 
771 bool xive_native_has_single_escalation(void)
772 {
773 	return xive_has_single_esc;
774 }
775 EXPORT_SYMBOL_GPL(xive_native_has_single_escalation);
776 
777 bool xive_native_has_save_restore(void)
778 {
779 	return xive_has_save_restore;
780 }
781 EXPORT_SYMBOL_GPL(xive_native_has_save_restore);
782 
783 int xive_native_get_queue_info(u32 vp_id, u32 prio,
784 			       u64 *out_qpage,
785 			       u64 *out_qsize,
786 			       u64 *out_qeoi_page,
787 			       u32 *out_escalate_irq,
788 			       u64 *out_qflags)
789 {
790 	__be64 qpage;
791 	__be64 qsize;
792 	__be64 qeoi_page;
793 	__be32 escalate_irq;
794 	__be64 qflags;
795 	s64 rc;
796 
797 	rc = opal_xive_get_queue_info(vp_id, prio, &qpage, &qsize,
798 				      &qeoi_page, &escalate_irq, &qflags);
799 	if (rc) {
800 		vp_err(vp_id, "failed to get queue %d info : %lld\n", prio, rc);
801 		return -EIO;
802 	}
803 
804 	if (out_qpage)
805 		*out_qpage = be64_to_cpu(qpage);
806 	if (out_qsize)
807 		*out_qsize = be32_to_cpu(qsize);
808 	if (out_qeoi_page)
809 		*out_qeoi_page = be64_to_cpu(qeoi_page);
810 	if (out_escalate_irq)
811 		*out_escalate_irq = be32_to_cpu(escalate_irq);
812 	if (out_qflags)
813 		*out_qflags = be64_to_cpu(qflags);
814 
815 	return 0;
816 }
817 EXPORT_SYMBOL_GPL(xive_native_get_queue_info);
818 
819 int xive_native_get_queue_state(u32 vp_id, u32 prio, u32 *qtoggle, u32 *qindex)
820 {
821 	__be32 opal_qtoggle;
822 	__be32 opal_qindex;
823 	s64 rc;
824 
825 	rc = opal_xive_get_queue_state(vp_id, prio, &opal_qtoggle,
826 				       &opal_qindex);
827 	if (rc) {
828 		vp_err(vp_id, "failed to get queue %d state : %lld\n", prio, rc);
829 		return -EIO;
830 	}
831 
832 	if (qtoggle)
833 		*qtoggle = be32_to_cpu(opal_qtoggle);
834 	if (qindex)
835 		*qindex = be32_to_cpu(opal_qindex);
836 
837 	return 0;
838 }
839 EXPORT_SYMBOL_GPL(xive_native_get_queue_state);
840 
841 int xive_native_set_queue_state(u32 vp_id, u32 prio, u32 qtoggle, u32 qindex)
842 {
843 	s64 rc;
844 
845 	rc = opal_xive_set_queue_state(vp_id, prio, qtoggle, qindex);
846 	if (rc) {
847 		vp_err(vp_id, "failed to set queue %d state : %lld\n", prio, rc);
848 		return -EIO;
849 	}
850 
851 	return 0;
852 }
853 EXPORT_SYMBOL_GPL(xive_native_set_queue_state);
854 
855 bool xive_native_has_queue_state_support(void)
856 {
857 	return opal_check_token(OPAL_XIVE_GET_QUEUE_STATE) &&
858 		opal_check_token(OPAL_XIVE_SET_QUEUE_STATE);
859 }
860 EXPORT_SYMBOL_GPL(xive_native_has_queue_state_support);
861 
862 int xive_native_get_vp_state(u32 vp_id, u64 *out_state)
863 {
864 	__be64 state;
865 	s64 rc;
866 
867 	rc = opal_xive_get_vp_state(vp_id, &state);
868 	if (rc) {
869 		vp_err(vp_id, "failed to get vp state : %lld\n", rc);
870 		return -EIO;
871 	}
872 
873 	if (out_state)
874 		*out_state = be64_to_cpu(state);
875 	return 0;
876 }
877 EXPORT_SYMBOL_GPL(xive_native_get_vp_state);
878 
879 machine_arch_initcall(powernv, xive_core_debug_init);
880