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;
46 static bool xive_has_single_esc;
48 int xive_native_populate_irq_data(u32 hw_irq, struct xive_irq_data *data)
50 __be64 flags, eoi_page, trig_page;
51 __be32 esb_shift, src_chip;
55 memset(data, 0, sizeof(*data));
57 rc = opal_xive_get_irq_info(hw_irq, &flags, &eoi_page, &trig_page,
58 &esb_shift, &src_chip);
60 pr_err("opal_xive_get_irq_info(0x%x) returned %lld\n",
65 opal_flags = be64_to_cpu(flags);
66 if (opal_flags & OPAL_XIVE_IRQ_STORE_EOI)
67 data->flags |= XIVE_IRQ_FLAG_STORE_EOI;
68 if (opal_flags & OPAL_XIVE_IRQ_LSI)
69 data->flags |= XIVE_IRQ_FLAG_LSI;
70 if (opal_flags & OPAL_XIVE_IRQ_SHIFT_BUG)
71 data->flags |= XIVE_IRQ_FLAG_SHIFT_BUG;
72 if (opal_flags & OPAL_XIVE_IRQ_MASK_VIA_FW)
73 data->flags |= XIVE_IRQ_FLAG_MASK_FW;
74 if (opal_flags & OPAL_XIVE_IRQ_EOI_VIA_FW)
75 data->flags |= XIVE_IRQ_FLAG_EOI_FW;
76 data->eoi_page = be64_to_cpu(eoi_page);
77 data->trig_page = be64_to_cpu(trig_page);
78 data->esb_shift = be32_to_cpu(esb_shift);
79 data->src_chip = be32_to_cpu(src_chip);
81 data->eoi_mmio = ioremap(data->eoi_page, 1u << data->esb_shift);
82 if (!data->eoi_mmio) {
83 pr_err("Failed to map EOI page for irq 0x%x\n", hw_irq);
87 data->hw_irq = hw_irq;
91 if (data->trig_page == data->eoi_page) {
92 data->trig_mmio = data->eoi_mmio;
96 data->trig_mmio = ioremap(data->trig_page, 1u << data->esb_shift);
97 if (!data->trig_mmio) {
98 pr_err("Failed to map trigger page for irq 0x%x\n", hw_irq);
103 EXPORT_SYMBOL_GPL(xive_native_populate_irq_data);
105 int xive_native_configure_irq(u32 hw_irq, u32 target, u8 prio, u32 sw_irq)
110 rc = opal_xive_set_irq_config(hw_irq, target, prio, sw_irq);
113 msleep(OPAL_BUSY_DELAY_MS);
115 return rc == 0 ? 0 : -ENXIO;
117 EXPORT_SYMBOL_GPL(xive_native_configure_irq);
120 /* This can be called multiple time to change a queue configuration */
121 int xive_native_configure_queue(u32 vp_id, struct xive_q *q, u8 prio,
122 __be32 *qpage, u32 order, bool can_escalate)
127 u64 flags, qpage_phys;
129 /* If there's an actual queue page, clean it */
133 qpage_phys = __pa(qpage);
137 /* Initialize the rest of the fields */
138 q->msk = order ? ((1u << (order - 2)) - 1) : 0;
142 rc = opal_xive_get_queue_info(vp_id, prio, NULL, NULL,
147 pr_err("Error %lld getting queue info prio %d\n", rc, prio);
151 q->eoi_phys = be64_to_cpu(qeoi_page_be);
154 flags = OPAL_XIVE_EQ_ALWAYS_NOTIFY | OPAL_XIVE_EQ_ENABLED;
156 /* Escalation needed ? */
158 q->esc_irq = be32_to_cpu(esc_irq_be);
159 flags |= OPAL_XIVE_EQ_ESCALATE;
162 /* Configure and enable the queue in HW */
164 rc = opal_xive_set_queue_info(vp_id, prio, qpage_phys, order, flags);
167 msleep(OPAL_BUSY_DELAY_MS);
170 pr_err("Error %lld setting queue for prio %d\n", rc, prio);
174 * KVM code requires all of the above to be visible before
175 * q->qpage is set due to how it manages IPI EOIs
183 EXPORT_SYMBOL_GPL(xive_native_configure_queue);
185 static void __xive_native_disable_queue(u32 vp_id, struct xive_q *q, u8 prio)
189 /* Disable the queue in HW */
191 rc = opal_xive_set_queue_info(vp_id, prio, 0, 0, 0);
194 msleep(OPAL_BUSY_DELAY_MS);
197 pr_err("Error %lld disabling queue for prio %d\n", rc, prio);
200 void xive_native_disable_queue(u32 vp_id, struct xive_q *q, u8 prio)
202 __xive_native_disable_queue(vp_id, q, prio);
204 EXPORT_SYMBOL_GPL(xive_native_disable_queue);
206 static int xive_native_setup_queue(unsigned int cpu, struct xive_cpu *xc, u8 prio)
208 struct xive_q *q = &xc->queue[prio];
211 qpage = xive_queue_page_alloc(cpu, xive_queue_shift);
213 return PTR_ERR(qpage);
215 return xive_native_configure_queue(get_hard_smp_processor_id(cpu),
216 q, prio, qpage, xive_queue_shift, false);
219 static void xive_native_cleanup_queue(unsigned int cpu, struct xive_cpu *xc, u8 prio)
221 struct xive_q *q = &xc->queue[prio];
222 unsigned int alloc_order;
225 * We use the variant with no iounmap as this is called on exec
226 * from an IPI and iounmap isn't safe
228 __xive_native_disable_queue(get_hard_smp_processor_id(cpu), q, prio);
229 alloc_order = xive_alloc_order(xive_queue_shift);
230 free_pages((unsigned long)q->qpage, alloc_order);
234 static bool xive_native_match(struct device_node *node)
236 return of_device_is_compatible(node, "ibm,opal-xive-vc");
239 static s64 opal_xive_allocate_irq(u32 chip_id)
241 s64 irq = opal_xive_allocate_irq_raw(chip_id);
244 * Old versions of skiboot can incorrectly return 0xffffffff to
245 * indicate no space, fix it up here.
247 return irq == 0xffffffff ? OPAL_RESOURCE : irq;
251 static int xive_native_get_ipi(unsigned int cpu, struct xive_cpu *xc)
253 struct device_node *np;
254 unsigned int chip_id;
257 /* Find the chip ID */
258 np = of_get_cpu_node(cpu, NULL);
260 if (of_property_read_u32(np, "ibm,chip-id", &chip_id) < 0)
264 /* Allocate an IPI and populate info about it */
266 irq = opal_xive_allocate_irq(chip_id);
267 if (irq == OPAL_BUSY) {
268 msleep(OPAL_BUSY_DELAY_MS);
272 pr_err("Failed to allocate IPI on CPU %d\n", cpu);
280 #endif /* CONFIG_SMP */
282 u32 xive_native_alloc_irq(void)
287 rc = opal_xive_allocate_irq(OPAL_XIVE_ANY_CHIP);
290 msleep(OPAL_BUSY_DELAY_MS);
296 EXPORT_SYMBOL_GPL(xive_native_alloc_irq);
298 void xive_native_free_irq(u32 irq)
301 s64 rc = opal_xive_free_irq(irq);
304 msleep(OPAL_BUSY_DELAY_MS);
307 EXPORT_SYMBOL_GPL(xive_native_free_irq);
310 static void xive_native_put_ipi(unsigned int cpu, struct xive_cpu *xc)
315 if (xc->hw_ipi == XIVE_BAD_IRQ)
318 rc = opal_xive_free_irq(xc->hw_ipi);
319 if (rc == OPAL_BUSY) {
320 msleep(OPAL_BUSY_DELAY_MS);
323 xc->hw_ipi = XIVE_BAD_IRQ;
327 #endif /* CONFIG_SMP */
329 static void xive_native_shutdown(void)
331 /* Switch the XIVE to emulation mode */
332 opal_xive_reset(OPAL_XIVE_MODE_EMU);
336 * Perform an "ack" cycle on the current thread, thus
337 * grabbing the pending active priorities and updating
338 * the CPPR to the most favored one.
340 static void xive_native_update_pending(struct xive_cpu *xc)
345 /* Perform the acknowledge hypervisor to register cycle */
346 ack = be16_to_cpu(__raw_readw(xive_tima + TM_SPC_ACK_HV_REG));
348 /* Synchronize subsequent queue accesses */
352 * Grab the CPPR and the "HE" field which indicates the source
353 * of the hypervisor interrupt (if any)
356 he = (ack >> 8) >> 6;
358 case TM_QW3_NSR_HE_NONE: /* Nothing to see here */
360 case TM_QW3_NSR_HE_PHYS: /* Physical thread interrupt */
363 /* Mark the priority pending */
364 xc->pending_prio |= 1 << cppr;
367 * A new interrupt should never have a CPPR less favored
368 * than our current one.
370 if (cppr >= xc->cppr)
371 pr_err("CPU %d odd ack CPPR, got %d at %d\n",
372 smp_processor_id(), cppr, xc->cppr);
374 /* Update our idea of what the CPPR is */
377 case TM_QW3_NSR_HE_POOL: /* HV Pool interrupt (unused) */
378 case TM_QW3_NSR_HE_LSI: /* Legacy FW LSI (unused) */
379 pr_err("CPU %d got unexpected interrupt type HE=%d\n",
380 smp_processor_id(), he);
385 static void xive_native_eoi(u32 hw_irq)
388 * Not normally used except if specific interrupts need
389 * a workaround on EOI.
391 opal_int_eoi(hw_irq);
394 static void xive_native_setup_cpu(unsigned int cpu, struct xive_cpu *xc)
401 if (xive_pool_vps == XIVE_INVALID_VP)
404 /* Check if pool VP already active, if it is, pull it */
405 if (in_be32(xive_tima + TM_QW2_HV_POOL + TM_WORD2) & TM_QW2W2_VP)
406 in_be64(xive_tima + TM_SPC_PULL_POOL_CTX);
408 /* Enable the pool VP */
409 vp = xive_pool_vps + cpu;
411 rc = opal_xive_set_vp_info(vp, OPAL_XIVE_VP_ENABLED, 0);
414 msleep(OPAL_BUSY_DELAY_MS);
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 /* Push it on the CPU (set LSMFB to 0xff to skip backlog scan) */
430 out_be32(xive_tima + TM_QW2_HV_POOL + TM_WORD0, 0xff);
431 out_be32(xive_tima + TM_QW2_HV_POOL + TM_WORD2, TM_QW2W2_VP | vp_cam);
434 static void xive_native_teardown_cpu(unsigned int cpu, struct xive_cpu *xc)
439 if (xive_pool_vps == XIVE_INVALID_VP)
442 /* Pull the pool VP from the CPU */
443 in_be64(xive_tima + TM_SPC_PULL_POOL_CTX);
446 vp = xive_pool_vps + cpu;
448 rc = opal_xive_set_vp_info(vp, 0, 0);
451 msleep(OPAL_BUSY_DELAY_MS);
455 void xive_native_sync_source(u32 hw_irq)
457 opal_xive_sync(XIVE_SYNC_EAS, hw_irq);
459 EXPORT_SYMBOL_GPL(xive_native_sync_source);
461 static const struct xive_ops xive_native_ops = {
462 .populate_irq_data = xive_native_populate_irq_data,
463 .configure_irq = xive_native_configure_irq,
464 .setup_queue = xive_native_setup_queue,
465 .cleanup_queue = xive_native_cleanup_queue,
466 .match = xive_native_match,
467 .shutdown = xive_native_shutdown,
468 .update_pending = xive_native_update_pending,
469 .eoi = xive_native_eoi,
470 .setup_cpu = xive_native_setup_cpu,
471 .teardown_cpu = xive_native_teardown_cpu,
472 .sync_source = xive_native_sync_source,
474 .get_ipi = xive_native_get_ipi,
475 .put_ipi = xive_native_put_ipi,
476 #endif /* CONFIG_SMP */
480 static bool xive_parse_provisioning(struct device_node *np)
484 if (of_property_read_u32(np, "ibm,xive-provision-page-size",
485 &xive_provision_size) < 0)
487 rc = of_property_count_elems_of_size(np, "ibm,xive-provision-chips", 4);
489 pr_err("Error %d getting provision chips array\n", rc);
492 xive_provision_chip_count = rc;
496 xive_provision_chips = kcalloc(4, xive_provision_chip_count,
498 if (WARN_ON(!xive_provision_chips))
501 rc = of_property_read_u32_array(np, "ibm,xive-provision-chips",
502 xive_provision_chips,
503 xive_provision_chip_count);
505 pr_err("Error %d reading provision chips array\n", rc);
509 xive_provision_cache = kmem_cache_create("xive-provision",
513 if (!xive_provision_cache) {
514 pr_err("Failed to allocate provision cache\n");
520 static void xive_native_setup_pools(void)
522 /* Allocate a pool big enough */
523 pr_debug("XIVE: Allocating VP block for pool size %u\n", nr_cpu_ids);
525 xive_pool_vps = xive_native_alloc_vp_block(nr_cpu_ids);
526 if (WARN_ON(xive_pool_vps == XIVE_INVALID_VP))
527 pr_err("XIVE: Failed to allocate pool VP, KVM might not function\n");
529 pr_debug("XIVE: Pool VPs allocated at 0x%x for %u max CPUs\n",
530 xive_pool_vps, nr_cpu_ids);
533 u32 xive_native_default_eq_shift(void)
535 return xive_queue_shift;
537 EXPORT_SYMBOL_GPL(xive_native_default_eq_shift);
539 bool __init xive_native_init(void)
541 struct device_node *np;
544 struct property *prop;
550 if (xive_cmdline_disabled)
553 pr_devel("xive_native_init()\n");
554 np = of_find_compatible_node(NULL, NULL, "ibm,opal-xive-pe");
556 pr_devel("not found !\n");
559 pr_devel("Found %pOF\n", np);
561 /* Resource 1 is HV window */
562 if (of_address_to_resource(np, 1, &r)) {
563 pr_err("Failed to get thread mgmnt area resource\n");
566 tima = ioremap(r.start, resource_size(&r));
568 pr_err("Failed to map thread mgmnt area\n");
572 /* Read number of priorities */
573 if (of_property_read_u32(np, "ibm,xive-#priorities", &val) == 0)
576 /* Iterate the EQ sizes and pick one */
577 of_property_for_each_u32(np, "ibm,xive-eq-sizes", prop, p, val) {
578 xive_queue_shift = val;
579 if (val == PAGE_SHIFT)
583 /* Do we support single escalation */
584 if (of_get_property(np, "single-escalation-support", NULL) != NULL)
585 xive_has_single_esc = true;
587 /* Configure Thread Management areas for KVM */
588 for_each_possible_cpu(cpu)
589 kvmppc_set_xive_tima(cpu, r.start, tima);
591 /* Grab size of provisionning pages */
592 xive_parse_provisioning(np);
594 /* Switch the XIVE to exploitation mode */
595 rc = opal_xive_reset(OPAL_XIVE_MODE_EXPL);
597 pr_err("Switch to exploitation mode failed with error %lld\n", rc);
601 /* Setup some dummy HV pool VPs */
602 xive_native_setup_pools();
604 /* Initialize XIVE core with our backend */
605 if (!xive_core_init(&xive_native_ops, tima, TM_QW3_HV_PHYS,
607 opal_xive_reset(OPAL_XIVE_MODE_EMU);
610 pr_info("Using %dkB queues\n", 1 << (xive_queue_shift - 10));
614 static bool xive_native_provision_pages(void)
619 for (i = 0; i < xive_provision_chip_count; i++) {
620 u32 chip = xive_provision_chips[i];
623 * XXX TODO: Try to make the allocation local to the node where
626 p = kmem_cache_alloc(xive_provision_cache, GFP_KERNEL);
628 pr_err("Failed to allocate provisioning page\n");
632 opal_xive_donate_page(chip, __pa(p));
637 u32 xive_native_alloc_vp_block(u32 max_vcpus)
642 order = fls(max_vcpus) - 1;
643 if (max_vcpus > (1 << order))
646 pr_debug("VP block alloc, for max VCPUs %d use order %d\n",
650 rc = opal_xive_alloc_vp_block(order);
653 msleep(OPAL_BUSY_DELAY_MS);
655 case OPAL_XIVE_PROVISIONING:
656 if (!xive_native_provision_pages())
657 return XIVE_INVALID_VP;
661 pr_err("OPAL failed to allocate VCPUs order %d, err %lld\n",
663 return XIVE_INVALID_VP;
669 EXPORT_SYMBOL_GPL(xive_native_alloc_vp_block);
671 void xive_native_free_vp_block(u32 vp_base)
675 if (vp_base == XIVE_INVALID_VP)
678 rc = opal_xive_free_vp_block(vp_base);
680 pr_warn("OPAL error %lld freeing VP block\n", rc);
682 EXPORT_SYMBOL_GPL(xive_native_free_vp_block);
684 int xive_native_enable_vp(u32 vp_id, bool single_escalation)
687 u64 flags = OPAL_XIVE_VP_ENABLED;
689 if (single_escalation)
690 flags |= OPAL_XIVE_VP_SINGLE_ESCALATION;
692 rc = opal_xive_set_vp_info(vp_id, flags, 0);
695 msleep(OPAL_BUSY_DELAY_MS);
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);
709 msleep(OPAL_BUSY_DELAY_MS);
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);
731 bool xive_native_has_single_escalation(void)
733 return xive_has_single_esc;
735 EXPORT_SYMBOL_GPL(xive_native_has_single_escalation);