2 * Copyright 2016,2017 IBM Corporation.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
10 #define pr_fmt(fmt) "xive: " fmt
12 #include <linux/types.h>
13 #include <linux/irq.h>
14 #include <linux/debugfs.h>
15 #include <linux/smp.h>
16 #include <linux/interrupt.h>
17 #include <linux/seq_file.h>
18 #include <linux/init.h>
20 #include <linux/slab.h>
21 #include <linux/spinlock.h>
22 #include <linux/delay.h>
23 #include <linux/cpumask.h>
25 #include <linux/kmemleak.h>
31 #include <asm/errno.h>
33 #include <asm/xive-regs.h>
35 #include <asm/kvm_ppc.h>
37 #include "xive-internal.h"
40 static u32 xive_provision_size;
41 static u32 *xive_provision_chips;
42 static u32 xive_provision_chip_count;
43 static u32 xive_queue_shift;
44 static u32 xive_pool_vps = XIVE_INVALID_VP;
45 static struct kmem_cache *xive_provision_cache;
47 int xive_native_populate_irq_data(u32 hw_irq, struct xive_irq_data *data)
49 __be64 flags, eoi_page, trig_page;
50 __be32 esb_shift, src_chip;
54 memset(data, 0, sizeof(*data));
56 rc = opal_xive_get_irq_info(hw_irq, &flags, &eoi_page, &trig_page,
57 &esb_shift, &src_chip);
59 pr_err("opal_xive_get_irq_info(0x%x) returned %lld\n",
64 opal_flags = be64_to_cpu(flags);
65 if (opal_flags & OPAL_XIVE_IRQ_STORE_EOI)
66 data->flags |= XIVE_IRQ_FLAG_STORE_EOI;
67 if (opal_flags & OPAL_XIVE_IRQ_LSI)
68 data->flags |= XIVE_IRQ_FLAG_LSI;
69 if (opal_flags & OPAL_XIVE_IRQ_SHIFT_BUG)
70 data->flags |= XIVE_IRQ_FLAG_SHIFT_BUG;
71 if (opal_flags & OPAL_XIVE_IRQ_MASK_VIA_FW)
72 data->flags |= XIVE_IRQ_FLAG_MASK_FW;
73 if (opal_flags & OPAL_XIVE_IRQ_EOI_VIA_FW)
74 data->flags |= XIVE_IRQ_FLAG_EOI_FW;
75 data->eoi_page = be64_to_cpu(eoi_page);
76 data->trig_page = be64_to_cpu(trig_page);
77 data->esb_shift = be32_to_cpu(esb_shift);
78 data->src_chip = be32_to_cpu(src_chip);
80 data->eoi_mmio = ioremap(data->eoi_page, 1u << data->esb_shift);
81 if (!data->eoi_mmio) {
82 pr_err("Failed to map EOI page for irq 0x%x\n", hw_irq);
86 data->hw_irq = hw_irq;
90 if (data->trig_page == data->eoi_page) {
91 data->trig_mmio = data->eoi_mmio;
95 data->trig_mmio = ioremap(data->trig_page, 1u << data->esb_shift);
96 if (!data->trig_mmio) {
97 pr_err("Failed to map trigger page for irq 0x%x\n", hw_irq);
102 EXPORT_SYMBOL_GPL(xive_native_populate_irq_data);
104 int xive_native_configure_irq(u32 hw_irq, u32 target, u8 prio, u32 sw_irq)
109 rc = opal_xive_set_irq_config(hw_irq, target, prio, sw_irq);
114 return rc == 0 ? 0 : -ENXIO;
116 EXPORT_SYMBOL_GPL(xive_native_configure_irq);
119 /* This can be called multiple time to change a queue configuration */
120 int xive_native_configure_queue(u32 vp_id, struct xive_q *q, u8 prio,
121 __be32 *qpage, u32 order, bool can_escalate)
126 u64 flags, qpage_phys;
128 /* If there's an actual queue page, clean it */
132 qpage_phys = __pa(qpage);
136 /* Initialize the rest of the fields */
137 q->msk = order ? ((1u << (order - 2)) - 1) : 0;
141 rc = opal_xive_get_queue_info(vp_id, prio, NULL, NULL,
146 pr_err("Error %lld getting queue info prio %d\n", rc, prio);
150 q->eoi_phys = be64_to_cpu(qeoi_page_be);
153 flags = OPAL_XIVE_EQ_ALWAYS_NOTIFY | OPAL_XIVE_EQ_ENABLED;
155 /* Escalation needed ? */
157 q->esc_irq = be32_to_cpu(esc_irq_be);
158 flags |= OPAL_XIVE_EQ_ESCALATE;
161 /* Configure and enable the queue in HW */
163 rc = opal_xive_set_queue_info(vp_id, prio, qpage_phys, order, flags);
169 pr_err("Error %lld setting queue for prio %d\n", rc, prio);
173 * KVM code requires all of the above to be visible before
174 * q->qpage is set due to how it manages IPI EOIs
182 EXPORT_SYMBOL_GPL(xive_native_configure_queue);
184 static void __xive_native_disable_queue(u32 vp_id, struct xive_q *q, u8 prio)
188 /* Disable the queue in HW */
190 rc = opal_xive_set_queue_info(vp_id, prio, 0, 0, 0);
196 pr_err("Error %lld disabling queue for prio %d\n", rc, prio);
199 void xive_native_disable_queue(u32 vp_id, struct xive_q *q, u8 prio)
201 __xive_native_disable_queue(vp_id, q, prio);
203 EXPORT_SYMBOL_GPL(xive_native_disable_queue);
205 static int xive_native_setup_queue(unsigned int cpu, struct xive_cpu *xc, u8 prio)
207 struct xive_q *q = &xc->queue[prio];
210 qpage = xive_queue_page_alloc(cpu, xive_queue_shift);
212 return PTR_ERR(qpage);
214 return xive_native_configure_queue(get_hard_smp_processor_id(cpu),
215 q, prio, qpage, xive_queue_shift, false);
218 static void xive_native_cleanup_queue(unsigned int cpu, struct xive_cpu *xc, u8 prio)
220 struct xive_q *q = &xc->queue[prio];
221 unsigned int alloc_order;
224 * We use the variant with no iounmap as this is called on exec
225 * from an IPI and iounmap isn't safe
227 __xive_native_disable_queue(get_hard_smp_processor_id(cpu), q, prio);
228 alloc_order = xive_alloc_order(xive_queue_shift);
229 free_pages((unsigned long)q->qpage, alloc_order);
233 static bool xive_native_match(struct device_node *node)
235 return of_device_is_compatible(node, "ibm,opal-xive-vc");
238 static s64 opal_xive_allocate_irq(u32 chip_id)
240 s64 irq = opal_xive_allocate_irq_raw(chip_id);
243 * Old versions of skiboot can incorrectly return 0xffffffff to
244 * indicate no space, fix it up here.
246 return irq == 0xffffffff ? OPAL_RESOURCE : irq;
250 static int xive_native_get_ipi(unsigned int cpu, struct xive_cpu *xc)
252 struct device_node *np;
253 unsigned int chip_id;
256 /* Find the chip ID */
257 np = of_get_cpu_node(cpu, NULL);
259 if (of_property_read_u32(np, "ibm,chip-id", &chip_id) < 0)
263 /* Allocate an IPI and populate info about it */
265 irq = opal_xive_allocate_irq(chip_id);
266 if (irq == OPAL_BUSY) {
271 pr_err("Failed to allocate IPI on CPU %d\n", cpu);
279 #endif /* CONFIG_SMP */
281 u32 xive_native_alloc_irq(void)
286 rc = opal_xive_allocate_irq(OPAL_XIVE_ANY_CHIP);
295 EXPORT_SYMBOL_GPL(xive_native_alloc_irq);
297 void xive_native_free_irq(u32 irq)
300 s64 rc = opal_xive_free_irq(irq);
306 EXPORT_SYMBOL_GPL(xive_native_free_irq);
309 static void xive_native_put_ipi(unsigned int cpu, struct xive_cpu *xc)
314 if (xc->hw_ipi == XIVE_BAD_IRQ)
317 rc = opal_xive_free_irq(xc->hw_ipi);
318 if (rc == OPAL_BUSY) {
322 xc->hw_ipi = XIVE_BAD_IRQ;
326 #endif /* CONFIG_SMP */
328 static void xive_native_shutdown(void)
330 /* Switch the XIVE to emulation mode */
331 opal_xive_reset(OPAL_XIVE_MODE_EMU);
335 * Perform an "ack" cycle on the current thread, thus
336 * grabbing the pending active priorities and updating
337 * the CPPR to the most favored one.
339 static void xive_native_update_pending(struct xive_cpu *xc)
344 /* Perform the acknowledge hypervisor to register cycle */
345 ack = be16_to_cpu(__raw_readw(xive_tima + TM_SPC_ACK_HV_REG));
347 /* Synchronize subsequent queue accesses */
351 * Grab the CPPR and the "HE" field which indicates the source
352 * of the hypervisor interrupt (if any)
355 he = GETFIELD(TM_QW3_NSR_HE, (ack >> 8));
357 case TM_QW3_NSR_HE_NONE: /* Nothing to see here */
359 case TM_QW3_NSR_HE_PHYS: /* Physical thread interrupt */
362 /* Mark the priority pending */
363 xc->pending_prio |= 1 << cppr;
366 * A new interrupt should never have a CPPR less favored
367 * than our current one.
369 if (cppr >= xc->cppr)
370 pr_err("CPU %d odd ack CPPR, got %d at %d\n",
371 smp_processor_id(), cppr, xc->cppr);
373 /* Update our idea of what the CPPR is */
376 case TM_QW3_NSR_HE_POOL: /* HV Pool interrupt (unused) */
377 case TM_QW3_NSR_HE_LSI: /* Legacy FW LSI (unused) */
378 pr_err("CPU %d got unexpected interrupt type HE=%d\n",
379 smp_processor_id(), he);
384 static void xive_native_eoi(u32 hw_irq)
387 * Not normally used except if specific interrupts need
388 * a workaround on EOI.
390 opal_int_eoi(hw_irq);
393 static void xive_native_setup_cpu(unsigned int cpu, struct xive_cpu *xc)
400 if (xive_pool_vps == XIVE_INVALID_VP)
403 /* Check if pool VP already active, if it is, pull it */
404 if (in_be32(xive_tima + TM_QW2_HV_POOL + TM_WORD2) & TM_QW2W2_VP)
405 in_be64(xive_tima + TM_SPC_PULL_POOL_CTX);
407 /* Enable the pool VP */
408 vp = xive_pool_vps + cpu;
409 pr_debug("CPU %d setting up pool VP 0x%x\n", cpu, vp);
411 rc = opal_xive_set_vp_info(vp, OPAL_XIVE_VP_ENABLED, 0);
417 pr_err("Failed to enable pool VP on CPU %d\n", cpu);
421 /* Grab it's CAM value */
422 rc = opal_xive_get_vp_info(vp, NULL, &vp_cam_be, NULL, NULL);
424 pr_err("Failed to get pool VP info CPU %d\n", cpu);
427 vp_cam = be64_to_cpu(vp_cam_be);
429 pr_debug("VP CAM = %llx\n", vp_cam);
431 /* Push it on the CPU (set LSMFB to 0xff to skip backlog scan) */
432 pr_debug("(Old HW value: %08x)\n",
433 in_be32(xive_tima + TM_QW2_HV_POOL + TM_WORD2));
434 out_be32(xive_tima + TM_QW2_HV_POOL + TM_WORD0, 0xff);
435 out_be32(xive_tima + TM_QW2_HV_POOL + TM_WORD2,
436 TM_QW2W2_VP | vp_cam);
437 pr_debug("(New HW value: %08x)\n",
438 in_be32(xive_tima + TM_QW2_HV_POOL + TM_WORD2));
441 static void xive_native_teardown_cpu(unsigned int cpu, struct xive_cpu *xc)
446 if (xive_pool_vps == XIVE_INVALID_VP)
449 /* Pull the pool VP from the CPU */
450 in_be64(xive_tima + TM_SPC_PULL_POOL_CTX);
453 vp = xive_pool_vps + cpu;
455 rc = opal_xive_set_vp_info(vp, 0, 0);
462 void xive_native_sync_source(u32 hw_irq)
464 opal_xive_sync(XIVE_SYNC_EAS, hw_irq);
466 EXPORT_SYMBOL_GPL(xive_native_sync_source);
468 static const struct xive_ops xive_native_ops = {
469 .populate_irq_data = xive_native_populate_irq_data,
470 .configure_irq = xive_native_configure_irq,
471 .setup_queue = xive_native_setup_queue,
472 .cleanup_queue = xive_native_cleanup_queue,
473 .match = xive_native_match,
474 .shutdown = xive_native_shutdown,
475 .update_pending = xive_native_update_pending,
476 .eoi = xive_native_eoi,
477 .setup_cpu = xive_native_setup_cpu,
478 .teardown_cpu = xive_native_teardown_cpu,
479 .sync_source = xive_native_sync_source,
481 .get_ipi = xive_native_get_ipi,
482 .put_ipi = xive_native_put_ipi,
483 #endif /* CONFIG_SMP */
487 static bool xive_parse_provisioning(struct device_node *np)
491 if (of_property_read_u32(np, "ibm,xive-provision-page-size",
492 &xive_provision_size) < 0)
494 rc = of_property_count_elems_of_size(np, "ibm,xive-provision-chips", 4);
496 pr_err("Error %d getting provision chips array\n", rc);
499 xive_provision_chip_count = rc;
503 xive_provision_chips = kzalloc(4 * xive_provision_chip_count,
505 if (WARN_ON(!xive_provision_chips))
508 rc = of_property_read_u32_array(np, "ibm,xive-provision-chips",
509 xive_provision_chips,
510 xive_provision_chip_count);
512 pr_err("Error %d reading provision chips array\n", rc);
516 xive_provision_cache = kmem_cache_create("xive-provision",
520 if (!xive_provision_cache) {
521 pr_err("Failed to allocate provision cache\n");
527 static void xive_native_setup_pools(void)
529 /* Allocate a pool big enough */
530 pr_debug("XIVE: Allocating VP block for pool size %u\n", nr_cpu_ids);
532 xive_pool_vps = xive_native_alloc_vp_block(nr_cpu_ids);
533 if (WARN_ON(xive_pool_vps == XIVE_INVALID_VP))
534 pr_err("XIVE: Failed to allocate pool VP, KVM might not function\n");
536 pr_debug("XIVE: Pool VPs allocated at 0x%x for %u max CPUs\n",
537 xive_pool_vps, nr_cpu_ids);
540 u32 xive_native_default_eq_shift(void)
542 return xive_queue_shift;
544 EXPORT_SYMBOL_GPL(xive_native_default_eq_shift);
546 bool __init xive_native_init(void)
548 struct device_node *np;
551 struct property *prop;
557 if (xive_cmdline_disabled)
560 pr_devel("xive_native_init()\n");
561 np = of_find_compatible_node(NULL, NULL, "ibm,opal-xive-pe");
563 pr_devel("not found !\n");
566 pr_devel("Found %pOF\n", np);
568 /* Resource 1 is HV window */
569 if (of_address_to_resource(np, 1, &r)) {
570 pr_err("Failed to get thread mgmnt area resource\n");
573 tima = ioremap(r.start, resource_size(&r));
575 pr_err("Failed to map thread mgmnt area\n");
579 /* Read number of priorities */
580 if (of_property_read_u32(np, "ibm,xive-#priorities", &val) == 0)
583 /* Iterate the EQ sizes and pick one */
584 of_property_for_each_u32(np, "ibm,xive-eq-sizes", prop, p, val) {
585 xive_queue_shift = val;
586 if (val == PAGE_SHIFT)
590 /* Configure Thread Management areas for KVM */
591 for_each_possible_cpu(cpu)
592 kvmppc_set_xive_tima(cpu, r.start, tima);
594 /* Grab size of provisionning pages */
595 xive_parse_provisioning(np);
597 /* Switch the XIVE to exploitation mode */
598 rc = opal_xive_reset(OPAL_XIVE_MODE_EXPL);
600 pr_err("Switch to exploitation mode failed with error %lld\n", rc);
604 /* Setup some dummy HV pool VPs */
605 xive_native_setup_pools();
607 /* Initialize XIVE core with our backend */
608 if (!xive_core_init(&xive_native_ops, tima, TM_QW3_HV_PHYS,
610 opal_xive_reset(OPAL_XIVE_MODE_EMU);
613 pr_info("Using %dkB queues\n", 1 << (xive_queue_shift - 10));
617 static bool xive_native_provision_pages(void)
622 for (i = 0; i < xive_provision_chip_count; i++) {
623 u32 chip = xive_provision_chips[i];
626 * XXX TODO: Try to make the allocation local to the node where
629 p = kmem_cache_alloc(xive_provision_cache, GFP_KERNEL);
631 pr_err("Failed to allocate provisioning page\n");
635 opal_xive_donate_page(chip, __pa(p));
640 u32 xive_native_alloc_vp_block(u32 max_vcpus)
645 order = fls(max_vcpus) - 1;
646 if (max_vcpus > (1 << order))
649 pr_debug("VP block alloc, for max VCPUs %d use order %d\n",
653 rc = opal_xive_alloc_vp_block(order);
658 case OPAL_XIVE_PROVISIONING:
659 if (!xive_native_provision_pages())
660 return XIVE_INVALID_VP;
664 pr_err("OPAL failed to allocate VCPUs order %d, err %lld\n",
666 return XIVE_INVALID_VP;
672 EXPORT_SYMBOL_GPL(xive_native_alloc_vp_block);
674 void xive_native_free_vp_block(u32 vp_base)
678 if (vp_base == XIVE_INVALID_VP)
681 rc = opal_xive_free_vp_block(vp_base);
683 pr_warn("OPAL error %lld freeing VP block\n", rc);
685 EXPORT_SYMBOL_GPL(xive_native_free_vp_block);
687 int xive_native_enable_vp(u32 vp_id)
692 rc = opal_xive_set_vp_info(vp_id, OPAL_XIVE_VP_ENABLED, 0);
697 return rc ? -EIO : 0;
699 EXPORT_SYMBOL_GPL(xive_native_enable_vp);
701 int xive_native_disable_vp(u32 vp_id)
706 rc = opal_xive_set_vp_info(vp_id, 0, 0);
711 return rc ? -EIO : 0;
713 EXPORT_SYMBOL_GPL(xive_native_disable_vp);
715 int xive_native_get_vp_info(u32 vp_id, u32 *out_cam_id, u32 *out_chip_id)
718 __be32 vp_chip_id_be;
721 rc = opal_xive_get_vp_info(vp_id, NULL, &vp_cam_be, NULL, &vp_chip_id_be);
724 *out_cam_id = be64_to_cpu(vp_cam_be) & 0xffffffffu;
725 *out_chip_id = be32_to_cpu(vp_chip_id_be);
729 EXPORT_SYMBOL_GPL(xive_native_get_vp_info);