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