Loading...
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
37static u32 xive_provision_size;
38static u32 *xive_provision_chips;
39static u32 xive_provision_chip_count;
40static u32 xive_queue_shift;
41static u32 xive_pool_vps = XIVE_INVALID_VP;
42static struct kmem_cache *xive_provision_cache;
43static bool xive_has_single_esc;
44bool xive_has_save_restore;
45
46int 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}
97EXPORT_SYMBOL_GPL(xive_native_populate_irq_data);
98
99int 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}
111EXPORT_SYMBOL_GPL(xive_native_configure_irq);
112
113static 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 */
131int 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 }
190fail:
191 return rc;
192}
193EXPORT_SYMBOL_GPL(xive_native_configure_queue);
194
195static 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
210void xive_native_disable_queue(u32 vp_id, struct xive_q *q, u8 prio)
211{
212 __xive_native_disable_queue(vp_id, q, prio);
213}
214EXPORT_SYMBOL_GPL(xive_native_disable_queue);
215
216static 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
229static 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
244static bool xive_native_match(struct device_node *node)
245{
246 return of_device_is_compatible(node, "ibm,opal-xive-vc");
247}
248
249static 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
261static 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
283u32 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}
297EXPORT_SYMBOL_GPL(xive_native_alloc_irq_on_chip);
298
299void 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}
308EXPORT_SYMBOL_GPL(xive_native_free_irq);
309
310#ifdef CONFIG_SMP
311static 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
330static 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 */
341static 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
386static void xive_native_prepare_cpu(unsigned int cpu, struct xive_cpu *xc)
387{
388 xc->chip_id = cpu_to_chip_id(cpu);
389}
390
391static 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
431static 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
452void xive_native_sync_source(u32 hw_irq)
453{
454 opal_xive_sync(XIVE_SYNC_EAS, hw_irq);
455}
456EXPORT_SYMBOL_GPL(xive_native_sync_source);
457
458void xive_native_sync_queue(u32 hw_irq)
459{
460 opal_xive_sync(XIVE_SYNC_QUEUE, hw_irq);
461}
462EXPORT_SYMBOL_GPL(xive_native_sync_queue);
463
464#ifdef CONFIG_DEBUG_FS
465static 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
472static 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
495static 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
535static void __init xive_native_setup_pools(void)
536{
537 /* Allocate a pool big enough */
538 pr_debug("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("Failed to allocate pool VP, KVM might not function\n");
543
544 pr_debug("Pool VPs allocated at 0x%x for %u max CPUs\n",
545 xive_pool_vps, nr_cpu_ids);
546}
547
548u32 xive_native_default_eq_shift(void)
549{
550 return xive_queue_shift;
551}
552EXPORT_SYMBOL_GPL(xive_native_default_eq_shift);
553
554unsigned long xive_tima_os;
555EXPORT_SYMBOL_GPL(xive_tima_os);
556
557bool __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 xive_has_single_esc = of_property_read_bool(np, "single-escalation-support");
603
604 xive_has_save_restore = of_property_read_bool(np, "vp-save-restore");
605
606 /* Configure Thread Management areas for KVM */
607 for_each_possible_cpu(cpu)
608 kvmppc_set_xive_tima(cpu, r.start, tima);
609
610 /* Resource 2 is OS window */
611 if (of_address_to_resource(np, 2, &r)) {
612 pr_err("Failed to get thread mgmnt area resource\n");
613 goto err_put;
614 }
615
616 xive_tima_os = r.start;
617
618 /* Grab size of provisioning pages */
619 xive_parse_provisioning(np);
620
621 /* Switch the XIVE to exploitation mode */
622 rc = opal_xive_reset(OPAL_XIVE_MODE_EXPL);
623 if (rc) {
624 pr_err("Switch to exploitation mode failed with error %lld\n", rc);
625 goto err_put;
626 }
627
628 /* Setup some dummy HV pool VPs */
629 xive_native_setup_pools();
630
631 /* Initialize XIVE core with our backend */
632 if (!xive_core_init(np, &xive_native_ops, tima, TM_QW3_HV_PHYS,
633 max_prio)) {
634 opal_xive_reset(OPAL_XIVE_MODE_EMU);
635 goto err_put;
636 }
637 of_node_put(np);
638 pr_info("Using %dkB queues\n", 1 << (xive_queue_shift - 10));
639 return true;
640
641err_put:
642 of_node_put(np);
643 return false;
644}
645
646static bool xive_native_provision_pages(void)
647{
648 u32 i;
649 void *p;
650
651 for (i = 0; i < xive_provision_chip_count; i++) {
652 u32 chip = xive_provision_chips[i];
653
654 /*
655 * XXX TODO: Try to make the allocation local to the node where
656 * the chip resides.
657 */
658 p = kmem_cache_alloc(xive_provision_cache, GFP_KERNEL);
659 if (!p) {
660 pr_err("Failed to allocate provisioning page\n");
661 return false;
662 }
663 kmemleak_ignore(p);
664 opal_xive_donate_page(chip, __pa(p));
665 }
666 return true;
667}
668
669u32 xive_native_alloc_vp_block(u32 max_vcpus)
670{
671 s64 rc;
672 u32 order;
673
674 order = fls(max_vcpus) - 1;
675 if (max_vcpus > (1 << order))
676 order++;
677
678 pr_debug("VP block alloc, for max VCPUs %d use order %d\n",
679 max_vcpus, order);
680
681 for (;;) {
682 rc = opal_xive_alloc_vp_block(order);
683 switch (rc) {
684 case OPAL_BUSY:
685 msleep(OPAL_BUSY_DELAY_MS);
686 break;
687 case OPAL_XIVE_PROVISIONING:
688 if (!xive_native_provision_pages())
689 return XIVE_INVALID_VP;
690 break;
691 default:
692 if (rc < 0) {
693 pr_err("OPAL failed to allocate VCPUs order %d, err %lld\n",
694 order, rc);
695 return XIVE_INVALID_VP;
696 }
697 return rc;
698 }
699 }
700}
701EXPORT_SYMBOL_GPL(xive_native_alloc_vp_block);
702
703void xive_native_free_vp_block(u32 vp_base)
704{
705 s64 rc;
706
707 if (vp_base == XIVE_INVALID_VP)
708 return;
709
710 rc = opal_xive_free_vp_block(vp_base);
711 if (rc < 0)
712 pr_warn("OPAL error %lld freeing VP block\n", rc);
713}
714EXPORT_SYMBOL_GPL(xive_native_free_vp_block);
715
716int xive_native_enable_vp(u32 vp_id, bool single_escalation)
717{
718 s64 rc;
719 u64 flags = OPAL_XIVE_VP_ENABLED;
720
721 if (single_escalation)
722 flags |= OPAL_XIVE_VP_SINGLE_ESCALATION;
723 for (;;) {
724 rc = opal_xive_set_vp_info(vp_id, flags, 0);
725 if (rc != OPAL_BUSY)
726 break;
727 msleep(OPAL_BUSY_DELAY_MS);
728 }
729 if (rc)
730 vp_err(vp_id, "Failed to enable VP : %lld\n", rc);
731 return rc ? -EIO : 0;
732}
733EXPORT_SYMBOL_GPL(xive_native_enable_vp);
734
735int xive_native_disable_vp(u32 vp_id)
736{
737 s64 rc;
738
739 for (;;) {
740 rc = opal_xive_set_vp_info(vp_id, 0, 0);
741 if (rc != OPAL_BUSY)
742 break;
743 msleep(OPAL_BUSY_DELAY_MS);
744 }
745 if (rc)
746 vp_err(vp_id, "Failed to disable VP : %lld\n", rc);
747 return rc ? -EIO : 0;
748}
749EXPORT_SYMBOL_GPL(xive_native_disable_vp);
750
751int xive_native_get_vp_info(u32 vp_id, u32 *out_cam_id, u32 *out_chip_id)
752{
753 __be64 vp_cam_be;
754 __be32 vp_chip_id_be;
755 s64 rc;
756
757 rc = opal_xive_get_vp_info(vp_id, NULL, &vp_cam_be, NULL, &vp_chip_id_be);
758 if (rc) {
759 vp_err(vp_id, "Failed to get VP info : %lld\n", rc);
760 return -EIO;
761 }
762 *out_cam_id = be64_to_cpu(vp_cam_be) & 0xffffffffu;
763 *out_chip_id = be32_to_cpu(vp_chip_id_be);
764
765 return 0;
766}
767EXPORT_SYMBOL_GPL(xive_native_get_vp_info);
768
769bool xive_native_has_single_escalation(void)
770{
771 return xive_has_single_esc;
772}
773EXPORT_SYMBOL_GPL(xive_native_has_single_escalation);
774
775bool xive_native_has_save_restore(void)
776{
777 return xive_has_save_restore;
778}
779EXPORT_SYMBOL_GPL(xive_native_has_save_restore);
780
781int xive_native_get_queue_info(u32 vp_id, u32 prio,
782 u64 *out_qpage,
783 u64 *out_qsize,
784 u64 *out_qeoi_page,
785 u32 *out_escalate_irq,
786 u64 *out_qflags)
787{
788 __be64 qpage;
789 __be64 qsize;
790 __be64 qeoi_page;
791 __be32 escalate_irq;
792 __be64 qflags;
793 s64 rc;
794
795 rc = opal_xive_get_queue_info(vp_id, prio, &qpage, &qsize,
796 &qeoi_page, &escalate_irq, &qflags);
797 if (rc) {
798 vp_err(vp_id, "failed to get queue %d info : %lld\n", prio, rc);
799 return -EIO;
800 }
801
802 if (out_qpage)
803 *out_qpage = be64_to_cpu(qpage);
804 if (out_qsize)
805 *out_qsize = be64_to_cpu(qsize);
806 if (out_qeoi_page)
807 *out_qeoi_page = be64_to_cpu(qeoi_page);
808 if (out_escalate_irq)
809 *out_escalate_irq = be32_to_cpu(escalate_irq);
810 if (out_qflags)
811 *out_qflags = be64_to_cpu(qflags);
812
813 return 0;
814}
815EXPORT_SYMBOL_GPL(xive_native_get_queue_info);
816
817int xive_native_get_queue_state(u32 vp_id, u32 prio, u32 *qtoggle, u32 *qindex)
818{
819 __be32 opal_qtoggle;
820 __be32 opal_qindex;
821 s64 rc;
822
823 rc = opal_xive_get_queue_state(vp_id, prio, &opal_qtoggle,
824 &opal_qindex);
825 if (rc) {
826 vp_err(vp_id, "failed to get queue %d state : %lld\n", prio, rc);
827 return -EIO;
828 }
829
830 if (qtoggle)
831 *qtoggle = be32_to_cpu(opal_qtoggle);
832 if (qindex)
833 *qindex = be32_to_cpu(opal_qindex);
834
835 return 0;
836}
837EXPORT_SYMBOL_GPL(xive_native_get_queue_state);
838
839int xive_native_set_queue_state(u32 vp_id, u32 prio, u32 qtoggle, u32 qindex)
840{
841 s64 rc;
842
843 rc = opal_xive_set_queue_state(vp_id, prio, qtoggle, qindex);
844 if (rc) {
845 vp_err(vp_id, "failed to set queue %d state : %lld\n", prio, rc);
846 return -EIO;
847 }
848
849 return 0;
850}
851EXPORT_SYMBOL_GPL(xive_native_set_queue_state);
852
853bool xive_native_has_queue_state_support(void)
854{
855 return opal_check_token(OPAL_XIVE_GET_QUEUE_STATE) &&
856 opal_check_token(OPAL_XIVE_SET_QUEUE_STATE);
857}
858EXPORT_SYMBOL_GPL(xive_native_has_queue_state_support);
859
860int xive_native_get_vp_state(u32 vp_id, u64 *out_state)
861{
862 __be64 state;
863 s64 rc;
864
865 rc = opal_xive_get_vp_state(vp_id, &state);
866 if (rc) {
867 vp_err(vp_id, "failed to get vp state : %lld\n", rc);
868 return -EIO;
869 }
870
871 if (out_state)
872 *out_state = be64_to_cpu(state);
873 return 0;
874}
875EXPORT_SYMBOL_GPL(xive_native_get_vp_state);
876
877machine_arch_initcall(powernv, xive_core_debug_init);
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#include <linux/kmemleak.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/opal.h>
32#include <asm/kvm_ppc.h>
33
34#include "xive-internal.h"
35
36
37static u32 xive_provision_size;
38static u32 *xive_provision_chips;
39static u32 xive_provision_chip_count;
40static u32 xive_queue_shift;
41static u32 xive_pool_vps = XIVE_INVALID_VP;
42static struct kmem_cache *xive_provision_cache;
43static bool xive_has_single_esc;
44
45int xive_native_populate_irq_data(u32 hw_irq, struct xive_irq_data *data)
46{
47 __be64 flags, eoi_page, trig_page;
48 __be32 esb_shift, src_chip;
49 u64 opal_flags;
50 s64 rc;
51
52 memset(data, 0, sizeof(*data));
53
54 rc = opal_xive_get_irq_info(hw_irq, &flags, &eoi_page, &trig_page,
55 &esb_shift, &src_chip);
56 if (rc) {
57 pr_err("opal_xive_get_irq_info(0x%x) returned %lld\n",
58 hw_irq, rc);
59 return -EINVAL;
60 }
61
62 opal_flags = be64_to_cpu(flags);
63 if (opal_flags & OPAL_XIVE_IRQ_STORE_EOI)
64 data->flags |= XIVE_IRQ_FLAG_STORE_EOI;
65 if (opal_flags & OPAL_XIVE_IRQ_LSI)
66 data->flags |= XIVE_IRQ_FLAG_LSI;
67 data->eoi_page = be64_to_cpu(eoi_page);
68 data->trig_page = be64_to_cpu(trig_page);
69 data->esb_shift = be32_to_cpu(esb_shift);
70 data->src_chip = be32_to_cpu(src_chip);
71
72 data->eoi_mmio = ioremap(data->eoi_page, 1u << data->esb_shift);
73 if (!data->eoi_mmio) {
74 pr_err("Failed to map EOI page for irq 0x%x\n", hw_irq);
75 return -ENOMEM;
76 }
77
78 data->hw_irq = hw_irq;
79
80 if (!data->trig_page)
81 return 0;
82 if (data->trig_page == data->eoi_page) {
83 data->trig_mmio = data->eoi_mmio;
84 return 0;
85 }
86
87 data->trig_mmio = ioremap(data->trig_page, 1u << data->esb_shift);
88 if (!data->trig_mmio) {
89 pr_err("Failed to map trigger page for irq 0x%x\n", hw_irq);
90 return -ENOMEM;
91 }
92 return 0;
93}
94EXPORT_SYMBOL_GPL(xive_native_populate_irq_data);
95
96int xive_native_configure_irq(u32 hw_irq, u32 target, u8 prio, u32 sw_irq)
97{
98 s64 rc;
99
100 for (;;) {
101 rc = opal_xive_set_irq_config(hw_irq, target, prio, sw_irq);
102 if (rc != OPAL_BUSY)
103 break;
104 msleep(OPAL_BUSY_DELAY_MS);
105 }
106 return rc == 0 ? 0 : -ENXIO;
107}
108EXPORT_SYMBOL_GPL(xive_native_configure_irq);
109
110static int xive_native_get_irq_config(u32 hw_irq, u32 *target, u8 *prio,
111 u32 *sw_irq)
112{
113 s64 rc;
114 __be64 vp;
115 __be32 lirq;
116
117 rc = opal_xive_get_irq_config(hw_irq, &vp, prio, &lirq);
118
119 *target = be64_to_cpu(vp);
120 *sw_irq = be32_to_cpu(lirq);
121
122 return rc == 0 ? 0 : -ENXIO;
123}
124
125#define vp_err(vp, fmt, ...) pr_err("VP[0x%x]: " fmt, vp, ##__VA_ARGS__)
126
127/* This can be called multiple time to change a queue configuration */
128int xive_native_configure_queue(u32 vp_id, struct xive_q *q, u8 prio,
129 __be32 *qpage, u32 order, bool can_escalate)
130{
131 s64 rc = 0;
132 __be64 qeoi_page_be;
133 __be32 esc_irq_be;
134 u64 flags, qpage_phys;
135
136 /* If there's an actual queue page, clean it */
137 if (order) {
138 if (WARN_ON(!qpage))
139 return -EINVAL;
140 qpage_phys = __pa(qpage);
141 } else
142 qpage_phys = 0;
143
144 /* Initialize the rest of the fields */
145 q->msk = order ? ((1u << (order - 2)) - 1) : 0;
146 q->idx = 0;
147 q->toggle = 0;
148
149 rc = opal_xive_get_queue_info(vp_id, prio, NULL, NULL,
150 &qeoi_page_be,
151 &esc_irq_be,
152 NULL);
153 if (rc) {
154 vp_err(vp_id, "Failed to get queue %d info : %lld\n", prio, rc);
155 rc = -EIO;
156 goto fail;
157 }
158 q->eoi_phys = be64_to_cpu(qeoi_page_be);
159
160 /* Default flags */
161 flags = OPAL_XIVE_EQ_ALWAYS_NOTIFY | OPAL_XIVE_EQ_ENABLED;
162
163 /* Escalation needed ? */
164 if (can_escalate) {
165 q->esc_irq = be32_to_cpu(esc_irq_be);
166 flags |= OPAL_XIVE_EQ_ESCALATE;
167 }
168
169 /* Configure and enable the queue in HW */
170 for (;;) {
171 rc = opal_xive_set_queue_info(vp_id, prio, qpage_phys, order, flags);
172 if (rc != OPAL_BUSY)
173 break;
174 msleep(OPAL_BUSY_DELAY_MS);
175 }
176 if (rc) {
177 vp_err(vp_id, "Failed to set queue %d info: %lld\n", prio, rc);
178 rc = -EIO;
179 } else {
180 /*
181 * KVM code requires all of the above to be visible before
182 * q->qpage is set due to how it manages IPI EOIs
183 */
184 wmb();
185 q->qpage = qpage;
186 }
187fail:
188 return rc;
189}
190EXPORT_SYMBOL_GPL(xive_native_configure_queue);
191
192static void __xive_native_disable_queue(u32 vp_id, struct xive_q *q, u8 prio)
193{
194 s64 rc;
195
196 /* Disable the queue in HW */
197 for (;;) {
198 rc = opal_xive_set_queue_info(vp_id, prio, 0, 0, 0);
199 if (rc != OPAL_BUSY)
200 break;
201 msleep(OPAL_BUSY_DELAY_MS);
202 }
203 if (rc)
204 vp_err(vp_id, "Failed to disable queue %d : %lld\n", prio, rc);
205}
206
207void xive_native_disable_queue(u32 vp_id, struct xive_q *q, u8 prio)
208{
209 __xive_native_disable_queue(vp_id, q, prio);
210}
211EXPORT_SYMBOL_GPL(xive_native_disable_queue);
212
213static int xive_native_setup_queue(unsigned int cpu, struct xive_cpu *xc, u8 prio)
214{
215 struct xive_q *q = &xc->queue[prio];
216 __be32 *qpage;
217
218 qpage = xive_queue_page_alloc(cpu, xive_queue_shift);
219 if (IS_ERR(qpage))
220 return PTR_ERR(qpage);
221
222 return xive_native_configure_queue(get_hard_smp_processor_id(cpu),
223 q, prio, qpage, xive_queue_shift, false);
224}
225
226static void xive_native_cleanup_queue(unsigned int cpu, struct xive_cpu *xc, u8 prio)
227{
228 struct xive_q *q = &xc->queue[prio];
229 unsigned int alloc_order;
230
231 /*
232 * We use the variant with no iounmap as this is called on exec
233 * from an IPI and iounmap isn't safe
234 */
235 __xive_native_disable_queue(get_hard_smp_processor_id(cpu), q, prio);
236 alloc_order = xive_alloc_order(xive_queue_shift);
237 free_pages((unsigned long)q->qpage, alloc_order);
238 q->qpage = NULL;
239}
240
241static bool xive_native_match(struct device_node *node)
242{
243 return of_device_is_compatible(node, "ibm,opal-xive-vc");
244}
245
246static s64 opal_xive_allocate_irq(u32 chip_id)
247{
248 s64 irq = opal_xive_allocate_irq_raw(chip_id);
249
250 /*
251 * Old versions of skiboot can incorrectly return 0xffffffff to
252 * indicate no space, fix it up here.
253 */
254 return irq == 0xffffffff ? OPAL_RESOURCE : irq;
255}
256
257#ifdef CONFIG_SMP
258static int xive_native_get_ipi(unsigned int cpu, struct xive_cpu *xc)
259{
260 s64 irq;
261
262 /* Allocate an IPI and populate info about it */
263 for (;;) {
264 irq = opal_xive_allocate_irq(xc->chip_id);
265 if (irq == OPAL_BUSY) {
266 msleep(OPAL_BUSY_DELAY_MS);
267 continue;
268 }
269 if (irq < 0) {
270 pr_err("Failed to allocate IPI on CPU %d\n", cpu);
271 return -ENXIO;
272 }
273 xc->hw_ipi = irq;
274 break;
275 }
276 return 0;
277}
278#endif /* CONFIG_SMP */
279
280u32 xive_native_alloc_irq_on_chip(u32 chip_id)
281{
282 s64 rc;
283
284 for (;;) {
285 rc = opal_xive_allocate_irq(chip_id);
286 if (rc != OPAL_BUSY)
287 break;
288 msleep(OPAL_BUSY_DELAY_MS);
289 }
290 if (rc < 0)
291 return 0;
292 return rc;
293}
294EXPORT_SYMBOL_GPL(xive_native_alloc_irq_on_chip);
295
296void xive_native_free_irq(u32 irq)
297{
298 for (;;) {
299 s64 rc = opal_xive_free_irq(irq);
300 if (rc != OPAL_BUSY)
301 break;
302 msleep(OPAL_BUSY_DELAY_MS);
303 }
304}
305EXPORT_SYMBOL_GPL(xive_native_free_irq);
306
307#ifdef CONFIG_SMP
308static void xive_native_put_ipi(unsigned int cpu, struct xive_cpu *xc)
309{
310 s64 rc;
311
312 /* Free the IPI */
313 if (xc->hw_ipi == XIVE_BAD_IRQ)
314 return;
315 for (;;) {
316 rc = opal_xive_free_irq(xc->hw_ipi);
317 if (rc == OPAL_BUSY) {
318 msleep(OPAL_BUSY_DELAY_MS);
319 continue;
320 }
321 xc->hw_ipi = XIVE_BAD_IRQ;
322 break;
323 }
324}
325#endif /* CONFIG_SMP */
326
327static void xive_native_shutdown(void)
328{
329 /* Switch the XIVE to emulation mode */
330 opal_xive_reset(OPAL_XIVE_MODE_EMU);
331}
332
333/*
334 * Perform an "ack" cycle on the current thread, thus
335 * grabbing the pending active priorities and updating
336 * the CPPR to the most favored one.
337 */
338static void xive_native_update_pending(struct xive_cpu *xc)
339{
340 u8 he, cppr;
341 u16 ack;
342
343 /* Perform the acknowledge hypervisor to register cycle */
344 ack = be16_to_cpu(__raw_readw(xive_tima + TM_SPC_ACK_HV_REG));
345
346 /* Synchronize subsequent queue accesses */
347 mb();
348
349 /*
350 * Grab the CPPR and the "HE" field which indicates the source
351 * of the hypervisor interrupt (if any)
352 */
353 cppr = ack & 0xff;
354 he = (ack >> 8) >> 6;
355 switch(he) {
356 case TM_QW3_NSR_HE_NONE: /* Nothing to see here */
357 break;
358 case TM_QW3_NSR_HE_PHYS: /* Physical thread interrupt */
359 if (cppr == 0xff)
360 return;
361 /* Mark the priority pending */
362 xc->pending_prio |= 1 << cppr;
363
364 /*
365 * A new interrupt should never have a CPPR less favored
366 * than our current one.
367 */
368 if (cppr >= xc->cppr)
369 pr_err("CPU %d odd ack CPPR, got %d at %d\n",
370 smp_processor_id(), cppr, xc->cppr);
371
372 /* Update our idea of what the CPPR is */
373 xc->cppr = cppr;
374 break;
375 case TM_QW3_NSR_HE_POOL: /* HV Pool interrupt (unused) */
376 case TM_QW3_NSR_HE_LSI: /* Legacy FW LSI (unused) */
377 pr_err("CPU %d got unexpected interrupt type HE=%d\n",
378 smp_processor_id(), he);
379 return;
380 }
381}
382
383static void xive_native_prepare_cpu(unsigned int cpu, struct xive_cpu *xc)
384{
385 xc->chip_id = cpu_to_chip_id(cpu);
386}
387
388static void xive_native_setup_cpu(unsigned int cpu, struct xive_cpu *xc)
389{
390 s64 rc;
391 u32 vp;
392 __be64 vp_cam_be;
393 u64 vp_cam;
394
395 if (xive_pool_vps == XIVE_INVALID_VP)
396 return;
397
398 /* Check if pool VP already active, if it is, pull it */
399 if (in_be32(xive_tima + TM_QW2_HV_POOL + TM_WORD2) & TM_QW2W2_VP)
400 in_be64(xive_tima + TM_SPC_PULL_POOL_CTX);
401
402 /* Enable the pool VP */
403 vp = xive_pool_vps + cpu;
404 for (;;) {
405 rc = opal_xive_set_vp_info(vp, OPAL_XIVE_VP_ENABLED, 0);
406 if (rc != OPAL_BUSY)
407 break;
408 msleep(OPAL_BUSY_DELAY_MS);
409 }
410 if (rc) {
411 pr_err("Failed to enable pool VP on CPU %d\n", cpu);
412 return;
413 }
414
415 /* Grab it's CAM value */
416 rc = opal_xive_get_vp_info(vp, NULL, &vp_cam_be, NULL, NULL);
417 if (rc) {
418 pr_err("Failed to get pool VP info CPU %d\n", cpu);
419 return;
420 }
421 vp_cam = be64_to_cpu(vp_cam_be);
422
423 /* Push it on the CPU (set LSMFB to 0xff to skip backlog scan) */
424 out_be32(xive_tima + TM_QW2_HV_POOL + TM_WORD0, 0xff);
425 out_be32(xive_tima + TM_QW2_HV_POOL + TM_WORD2, TM_QW2W2_VP | vp_cam);
426}
427
428static void xive_native_teardown_cpu(unsigned int cpu, struct xive_cpu *xc)
429{
430 s64 rc;
431 u32 vp;
432
433 if (xive_pool_vps == XIVE_INVALID_VP)
434 return;
435
436 /* Pull the pool VP from the CPU */
437 in_be64(xive_tima + TM_SPC_PULL_POOL_CTX);
438
439 /* Disable it */
440 vp = xive_pool_vps + cpu;
441 for (;;) {
442 rc = opal_xive_set_vp_info(vp, 0, 0);
443 if (rc != OPAL_BUSY)
444 break;
445 msleep(OPAL_BUSY_DELAY_MS);
446 }
447}
448
449void xive_native_sync_source(u32 hw_irq)
450{
451 opal_xive_sync(XIVE_SYNC_EAS, hw_irq);
452}
453EXPORT_SYMBOL_GPL(xive_native_sync_source);
454
455void xive_native_sync_queue(u32 hw_irq)
456{
457 opal_xive_sync(XIVE_SYNC_QUEUE, hw_irq);
458}
459EXPORT_SYMBOL_GPL(xive_native_sync_queue);
460
461static const struct xive_ops xive_native_ops = {
462 .populate_irq_data = xive_native_populate_irq_data,
463 .configure_irq = xive_native_configure_irq,
464 .get_irq_config = xive_native_get_irq_config,
465 .setup_queue = xive_native_setup_queue,
466 .cleanup_queue = xive_native_cleanup_queue,
467 .match = xive_native_match,
468 .shutdown = xive_native_shutdown,
469 .update_pending = xive_native_update_pending,
470 .prepare_cpu = xive_native_prepare_cpu,
471 .setup_cpu = xive_native_setup_cpu,
472 .teardown_cpu = xive_native_teardown_cpu,
473 .sync_source = xive_native_sync_source,
474#ifdef CONFIG_SMP
475 .get_ipi = xive_native_get_ipi,
476 .put_ipi = xive_native_put_ipi,
477#endif /* CONFIG_SMP */
478 .name = "native",
479};
480
481static bool xive_parse_provisioning(struct device_node *np)
482{
483 int rc;
484
485 if (of_property_read_u32(np, "ibm,xive-provision-page-size",
486 &xive_provision_size) < 0)
487 return true;
488 rc = of_property_count_elems_of_size(np, "ibm,xive-provision-chips", 4);
489 if (rc < 0) {
490 pr_err("Error %d getting provision chips array\n", rc);
491 return false;
492 }
493 xive_provision_chip_count = rc;
494 if (rc == 0)
495 return true;
496
497 xive_provision_chips = kcalloc(4, xive_provision_chip_count,
498 GFP_KERNEL);
499 if (WARN_ON(!xive_provision_chips))
500 return false;
501
502 rc = of_property_read_u32_array(np, "ibm,xive-provision-chips",
503 xive_provision_chips,
504 xive_provision_chip_count);
505 if (rc < 0) {
506 pr_err("Error %d reading provision chips array\n", rc);
507 return false;
508 }
509
510 xive_provision_cache = kmem_cache_create("xive-provision",
511 xive_provision_size,
512 xive_provision_size,
513 0, NULL);
514 if (!xive_provision_cache) {
515 pr_err("Failed to allocate provision cache\n");
516 return false;
517 }
518 return true;
519}
520
521static void xive_native_setup_pools(void)
522{
523 /* Allocate a pool big enough */
524 pr_debug("XIVE: Allocating VP block for pool size %u\n", nr_cpu_ids);
525
526 xive_pool_vps = xive_native_alloc_vp_block(nr_cpu_ids);
527 if (WARN_ON(xive_pool_vps == XIVE_INVALID_VP))
528 pr_err("XIVE: Failed to allocate pool VP, KVM might not function\n");
529
530 pr_debug("XIVE: Pool VPs allocated at 0x%x for %u max CPUs\n",
531 xive_pool_vps, nr_cpu_ids);
532}
533
534u32 xive_native_default_eq_shift(void)
535{
536 return xive_queue_shift;
537}
538EXPORT_SYMBOL_GPL(xive_native_default_eq_shift);
539
540unsigned long xive_tima_os;
541EXPORT_SYMBOL_GPL(xive_tima_os);
542
543bool __init xive_native_init(void)
544{
545 struct device_node *np;
546 struct resource r;
547 void __iomem *tima;
548 struct property *prop;
549 u8 max_prio = 7;
550 const __be32 *p;
551 u32 val, cpu;
552 s64 rc;
553
554 if (xive_cmdline_disabled)
555 return false;
556
557 pr_devel("xive_native_init()\n");
558 np = of_find_compatible_node(NULL, NULL, "ibm,opal-xive-pe");
559 if (!np) {
560 pr_devel("not found !\n");
561 return false;
562 }
563 pr_devel("Found %pOF\n", np);
564
565 /* Resource 1 is HV window */
566 if (of_address_to_resource(np, 1, &r)) {
567 pr_err("Failed to get thread mgmnt area resource\n");
568 return false;
569 }
570 tima = ioremap(r.start, resource_size(&r));
571 if (!tima) {
572 pr_err("Failed to map thread mgmnt area\n");
573 return false;
574 }
575
576 /* Read number of priorities */
577 if (of_property_read_u32(np, "ibm,xive-#priorities", &val) == 0)
578 max_prio = val - 1;
579
580 /* Iterate the EQ sizes and pick one */
581 of_property_for_each_u32(np, "ibm,xive-eq-sizes", prop, p, val) {
582 xive_queue_shift = val;
583 if (val == PAGE_SHIFT)
584 break;
585 }
586
587 /* Do we support single escalation */
588 if (of_get_property(np, "single-escalation-support", NULL) != NULL)
589 xive_has_single_esc = true;
590
591 /* Configure Thread Management areas for KVM */
592 for_each_possible_cpu(cpu)
593 kvmppc_set_xive_tima(cpu, r.start, tima);
594
595 /* Resource 2 is OS window */
596 if (of_address_to_resource(np, 2, &r)) {
597 pr_err("Failed to get thread mgmnt area resource\n");
598 return false;
599 }
600
601 xive_tima_os = r.start;
602
603 /* Grab size of provisionning pages */
604 xive_parse_provisioning(np);
605
606 /* Switch the XIVE to exploitation mode */
607 rc = opal_xive_reset(OPAL_XIVE_MODE_EXPL);
608 if (rc) {
609 pr_err("Switch to exploitation mode failed with error %lld\n", rc);
610 return false;
611 }
612
613 /* Setup some dummy HV pool VPs */
614 xive_native_setup_pools();
615
616 /* Initialize XIVE core with our backend */
617 if (!xive_core_init(np, &xive_native_ops, tima, TM_QW3_HV_PHYS,
618 max_prio)) {
619 opal_xive_reset(OPAL_XIVE_MODE_EMU);
620 return false;
621 }
622 pr_info("Using %dkB queues\n", 1 << (xive_queue_shift - 10));
623 return true;
624}
625
626static bool xive_native_provision_pages(void)
627{
628 u32 i;
629 void *p;
630
631 for (i = 0; i < xive_provision_chip_count; i++) {
632 u32 chip = xive_provision_chips[i];
633
634 /*
635 * XXX TODO: Try to make the allocation local to the node where
636 * the chip resides.
637 */
638 p = kmem_cache_alloc(xive_provision_cache, GFP_KERNEL);
639 if (!p) {
640 pr_err("Failed to allocate provisioning page\n");
641 return false;
642 }
643 kmemleak_ignore(p);
644 opal_xive_donate_page(chip, __pa(p));
645 }
646 return true;
647}
648
649u32 xive_native_alloc_vp_block(u32 max_vcpus)
650{
651 s64 rc;
652 u32 order;
653
654 order = fls(max_vcpus) - 1;
655 if (max_vcpus > (1 << order))
656 order++;
657
658 pr_debug("VP block alloc, for max VCPUs %d use order %d\n",
659 max_vcpus, order);
660
661 for (;;) {
662 rc = opal_xive_alloc_vp_block(order);
663 switch (rc) {
664 case OPAL_BUSY:
665 msleep(OPAL_BUSY_DELAY_MS);
666 break;
667 case OPAL_XIVE_PROVISIONING:
668 if (!xive_native_provision_pages())
669 return XIVE_INVALID_VP;
670 break;
671 default:
672 if (rc < 0) {
673 pr_err("OPAL failed to allocate VCPUs order %d, err %lld\n",
674 order, rc);
675 return XIVE_INVALID_VP;
676 }
677 return rc;
678 }
679 }
680}
681EXPORT_SYMBOL_GPL(xive_native_alloc_vp_block);
682
683void xive_native_free_vp_block(u32 vp_base)
684{
685 s64 rc;
686
687 if (vp_base == XIVE_INVALID_VP)
688 return;
689
690 rc = opal_xive_free_vp_block(vp_base);
691 if (rc < 0)
692 pr_warn("OPAL error %lld freeing VP block\n", rc);
693}
694EXPORT_SYMBOL_GPL(xive_native_free_vp_block);
695
696int xive_native_enable_vp(u32 vp_id, bool single_escalation)
697{
698 s64 rc;
699 u64 flags = OPAL_XIVE_VP_ENABLED;
700
701 if (single_escalation)
702 flags |= OPAL_XIVE_VP_SINGLE_ESCALATION;
703 for (;;) {
704 rc = opal_xive_set_vp_info(vp_id, flags, 0);
705 if (rc != OPAL_BUSY)
706 break;
707 msleep(OPAL_BUSY_DELAY_MS);
708 }
709 if (rc)
710 vp_err(vp_id, "Failed to enable VP : %lld\n", rc);
711 return rc ? -EIO : 0;
712}
713EXPORT_SYMBOL_GPL(xive_native_enable_vp);
714
715int xive_native_disable_vp(u32 vp_id)
716{
717 s64 rc;
718
719 for (;;) {
720 rc = opal_xive_set_vp_info(vp_id, 0, 0);
721 if (rc != OPAL_BUSY)
722 break;
723 msleep(OPAL_BUSY_DELAY_MS);
724 }
725 if (rc)
726 vp_err(vp_id, "Failed to disable VP : %lld\n", rc);
727 return rc ? -EIO : 0;
728}
729EXPORT_SYMBOL_GPL(xive_native_disable_vp);
730
731int xive_native_get_vp_info(u32 vp_id, u32 *out_cam_id, u32 *out_chip_id)
732{
733 __be64 vp_cam_be;
734 __be32 vp_chip_id_be;
735 s64 rc;
736
737 rc = opal_xive_get_vp_info(vp_id, NULL, &vp_cam_be, NULL, &vp_chip_id_be);
738 if (rc) {
739 vp_err(vp_id, "Failed to get VP info : %lld\n", rc);
740 return -EIO;
741 }
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}
747EXPORT_SYMBOL_GPL(xive_native_get_vp_info);
748
749bool xive_native_has_single_escalation(void)
750{
751 return xive_has_single_esc;
752}
753EXPORT_SYMBOL_GPL(xive_native_has_single_escalation);
754
755int 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 vp_err(vp_id, "failed to get queue %d info : %lld\n", 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}
789EXPORT_SYMBOL_GPL(xive_native_get_queue_info);
790
791int 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 vp_err(vp_id, "failed to get queue %d state : %lld\n", prio, rc);
801 return -EIO;
802 }
803
804 if (qtoggle)
805 *qtoggle = be32_to_cpu(opal_qtoggle);
806 if (qindex)
807 *qindex = be32_to_cpu(opal_qindex);
808
809 return 0;
810}
811EXPORT_SYMBOL_GPL(xive_native_get_queue_state);
812
813int xive_native_set_queue_state(u32 vp_id, u32 prio, u32 qtoggle, u32 qindex)
814{
815 s64 rc;
816
817 rc = opal_xive_set_queue_state(vp_id, prio, qtoggle, qindex);
818 if (rc) {
819 vp_err(vp_id, "failed to set queue %d state : %lld\n", prio, rc);
820 return -EIO;
821 }
822
823 return 0;
824}
825EXPORT_SYMBOL_GPL(xive_native_set_queue_state);
826
827bool xive_native_has_queue_state_support(void)
828{
829 return opal_check_token(OPAL_XIVE_GET_QUEUE_STATE) &&
830 opal_check_token(OPAL_XIVE_SET_QUEUE_STATE);
831}
832EXPORT_SYMBOL_GPL(xive_native_has_queue_state_support);
833
834int xive_native_get_vp_state(u32 vp_id, u64 *out_state)
835{
836 __be64 state;
837 s64 rc;
838
839 rc = opal_xive_get_vp_state(vp_id, &state);
840 if (rc) {
841 vp_err(vp_id, "failed to get vp state : %lld\n", rc);
842 return -EIO;
843 }
844
845 if (out_state)
846 *out_state = be64_to_cpu(state);
847 return 0;
848}
849EXPORT_SYMBOL_GPL(xive_native_get_vp_state);
850
851machine_arch_initcall(powernv, xive_core_debug_init);