1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Copyright 2016,2017 IBM Corporation.
6 #define pr_fmt(fmt) "xive: " fmt
8 #include <linux/types.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>
16 #include <linux/slab.h>
17 #include <linux/spinlock.h>
18 #include <linux/delay.h>
19 #include <linux/cpumask.h>
21 #include <linux/kmemleak.h>
23 #include <asm/machdep.h>
28 #include <asm/errno.h>
30 #include <asm/xive-regs.h>
32 #include <asm/kvm_ppc.h>
34 #include "xive-internal.h"
37 static u32 xive_provision_size
;
38 static u32
*xive_provision_chips
;
39 static u32 xive_provision_chip_count
;
40 static u32 xive_queue_shift
;
41 static u32 xive_pool_vps
= XIVE_INVALID_VP
;
42 static struct kmem_cache
*xive_provision_cache
;
43 static bool xive_has_single_esc
;
45 int xive_native_populate_irq_data(u32 hw_irq
, struct xive_irq_data
*data
)
47 __be64 flags
, eoi_page
, trig_page
;
48 __be32 esb_shift
, src_chip
;
52 memset(data
, 0, sizeof(*data
));
54 rc
= opal_xive_get_irq_info(hw_irq
, &flags
, &eoi_page
, &trig_page
,
55 &esb_shift
, &src_chip
);
57 pr_err("opal_xive_get_irq_info(0x%x) returned %lld\n",
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
);
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
);
78 data
->hw_irq
= hw_irq
;
82 if (data
->trig_page
== data
->eoi_page
) {
83 data
->trig_mmio
= data
->eoi_mmio
;
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
);
94 EXPORT_SYMBOL_GPL(xive_native_populate_irq_data
);
96 int xive_native_configure_irq(u32 hw_irq
, u32 target
, u8 prio
, u32 sw_irq
)
101 rc
= opal_xive_set_irq_config(hw_irq
, target
, prio
, sw_irq
);
104 msleep(OPAL_BUSY_DELAY_MS
);
106 return rc
== 0 ? 0 : -ENXIO
;
108 EXPORT_SYMBOL_GPL(xive_native_configure_irq
);
110 static int xive_native_get_irq_config(u32 hw_irq
, u32
*target
, u8
*prio
,
117 rc
= opal_xive_get_irq_config(hw_irq
, &vp
, prio
, &lirq
);
119 *target
= be64_to_cpu(vp
);
120 *sw_irq
= be32_to_cpu(lirq
);
122 return rc
== 0 ? 0 : -ENXIO
;
125 #define vp_err(vp, fmt, ...) pr_err("VP[0x%x]: " fmt, vp, ##__VA_ARGS__)
127 /* This can be called multiple time to change a queue configuration */
128 int xive_native_configure_queue(u32 vp_id
, struct xive_q
*q
, u8 prio
,
129 __be32
*qpage
, u32 order
, bool can_escalate
)
134 u64 flags
, qpage_phys
;
136 /* If there's an actual queue page, clean it */
140 qpage_phys
= __pa(qpage
);
144 /* Initialize the rest of the fields */
145 q
->msk
= order
? ((1u << (order
- 2)) - 1) : 0;
149 rc
= opal_xive_get_queue_info(vp_id
, prio
, NULL
, NULL
,
154 vp_err(vp_id
, "Failed to get queue %d info : %lld\n", prio
, rc
);
158 q
->eoi_phys
= be64_to_cpu(qeoi_page_be
);
161 flags
= OPAL_XIVE_EQ_ALWAYS_NOTIFY
| OPAL_XIVE_EQ_ENABLED
;
163 /* Escalation needed ? */
165 q
->esc_irq
= be32_to_cpu(esc_irq_be
);
166 flags
|= OPAL_XIVE_EQ_ESCALATE
;
169 /* Configure and enable the queue in HW */
171 rc
= opal_xive_set_queue_info(vp_id
, prio
, qpage_phys
, order
, flags
);
174 msleep(OPAL_BUSY_DELAY_MS
);
177 vp_err(vp_id
, "Failed to set queue %d info: %lld\n", prio
, rc
);
181 * KVM code requires all of the above to be visible before
182 * q->qpage is set due to how it manages IPI EOIs
190 EXPORT_SYMBOL_GPL(xive_native_configure_queue
);
192 static void __xive_native_disable_queue(u32 vp_id
, struct xive_q
*q
, u8 prio
)
196 /* Disable the queue in HW */
198 rc
= opal_xive_set_queue_info(vp_id
, prio
, 0, 0, 0);
201 msleep(OPAL_BUSY_DELAY_MS
);
204 vp_err(vp_id
, "Failed to disable queue %d : %lld\n", prio
, rc
);
207 void xive_native_disable_queue(u32 vp_id
, struct xive_q
*q
, u8 prio
)
209 __xive_native_disable_queue(vp_id
, q
, prio
);
211 EXPORT_SYMBOL_GPL(xive_native_disable_queue
);
213 static int xive_native_setup_queue(unsigned int cpu
, struct xive_cpu
*xc
, u8 prio
)
215 struct xive_q
*q
= &xc
->queue
[prio
];
218 qpage
= xive_queue_page_alloc(cpu
, xive_queue_shift
);
220 return PTR_ERR(qpage
);
222 return xive_native_configure_queue(get_hard_smp_processor_id(cpu
),
223 q
, prio
, qpage
, xive_queue_shift
, false);
226 static void xive_native_cleanup_queue(unsigned int cpu
, struct xive_cpu
*xc
, u8 prio
)
228 struct xive_q
*q
= &xc
->queue
[prio
];
229 unsigned int alloc_order
;
232 * We use the variant with no iounmap as this is called on exec
233 * from an IPI and iounmap isn't safe
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
);
241 static bool xive_native_match(struct device_node
*node
)
243 return of_device_is_compatible(node
, "ibm,opal-xive-vc");
246 static s64
opal_xive_allocate_irq(u32 chip_id
)
248 s64 irq
= opal_xive_allocate_irq_raw(chip_id
);
251 * Old versions of skiboot can incorrectly return 0xffffffff to
252 * indicate no space, fix it up here.
254 return irq
== 0xffffffff ? OPAL_RESOURCE
: irq
;
258 static int xive_native_get_ipi(unsigned int cpu
, struct xive_cpu
*xc
)
262 /* Allocate an IPI and populate info about it */
264 irq
= opal_xive_allocate_irq(xc
->chip_id
);
265 if (irq
== OPAL_BUSY
) {
266 msleep(OPAL_BUSY_DELAY_MS
);
270 pr_err("Failed to allocate IPI on CPU %d\n", cpu
);
278 #endif /* CONFIG_SMP */
280 u32
xive_native_alloc_irq_on_chip(u32 chip_id
)
285 rc
= opal_xive_allocate_irq(chip_id
);
288 msleep(OPAL_BUSY_DELAY_MS
);
294 EXPORT_SYMBOL_GPL(xive_native_alloc_irq_on_chip
);
296 void xive_native_free_irq(u32 irq
)
299 s64 rc
= opal_xive_free_irq(irq
);
302 msleep(OPAL_BUSY_DELAY_MS
);
305 EXPORT_SYMBOL_GPL(xive_native_free_irq
);
308 static void xive_native_put_ipi(unsigned int cpu
, struct xive_cpu
*xc
)
313 if (xc
->hw_ipi
== XIVE_BAD_IRQ
)
316 rc
= opal_xive_free_irq(xc
->hw_ipi
);
317 if (rc
== OPAL_BUSY
) {
318 msleep(OPAL_BUSY_DELAY_MS
);
321 xc
->hw_ipi
= XIVE_BAD_IRQ
;
325 #endif /* CONFIG_SMP */
327 static void xive_native_shutdown(void)
329 /* Switch the XIVE to emulation mode */
330 opal_xive_reset(OPAL_XIVE_MODE_EMU
);
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.
338 static void xive_native_update_pending(struct xive_cpu
*xc
)
343 /* Perform the acknowledge hypervisor to register cycle */
344 ack
= be16_to_cpu(__raw_readw(xive_tima
+ TM_SPC_ACK_HV_REG
));
346 /* Synchronize subsequent queue accesses */
350 * Grab the CPPR and the "HE" field which indicates the source
351 * of the hypervisor interrupt (if any)
354 he
= (ack
>> 8) >> 6;
356 case TM_QW3_NSR_HE_NONE
: /* Nothing to see here */
358 case TM_QW3_NSR_HE_PHYS
: /* Physical thread interrupt */
361 /* Mark the priority pending */
362 xc
->pending_prio
|= 1 << cppr
;
365 * A new interrupt should never have a CPPR less favored
366 * than our current one.
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
);
372 /* Update our idea of what the CPPR is */
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
);
383 static void xive_native_setup_cpu(unsigned int cpu
, struct xive_cpu
*xc
)
390 if (xive_pool_vps
== XIVE_INVALID_VP
)
393 /* Check if pool VP already active, if it is, pull it */
394 if (in_be32(xive_tima
+ TM_QW2_HV_POOL
+ TM_WORD2
) & TM_QW2W2_VP
)
395 in_be64(xive_tima
+ TM_SPC_PULL_POOL_CTX
);
397 /* Enable the pool VP */
398 vp
= xive_pool_vps
+ cpu
;
400 rc
= opal_xive_set_vp_info(vp
, OPAL_XIVE_VP_ENABLED
, 0);
403 msleep(OPAL_BUSY_DELAY_MS
);
406 pr_err("Failed to enable pool VP on CPU %d\n", cpu
);
410 /* Grab it's CAM value */
411 rc
= opal_xive_get_vp_info(vp
, NULL
, &vp_cam_be
, NULL
, NULL
);
413 pr_err("Failed to get pool VP info CPU %d\n", cpu
);
416 vp_cam
= be64_to_cpu(vp_cam_be
);
418 /* Push it on the CPU (set LSMFB to 0xff to skip backlog scan) */
419 out_be32(xive_tima
+ TM_QW2_HV_POOL
+ TM_WORD0
, 0xff);
420 out_be32(xive_tima
+ TM_QW2_HV_POOL
+ TM_WORD2
, TM_QW2W2_VP
| vp_cam
);
423 static void xive_native_teardown_cpu(unsigned int cpu
, struct xive_cpu
*xc
)
428 if (xive_pool_vps
== XIVE_INVALID_VP
)
431 /* Pull the pool VP from the CPU */
432 in_be64(xive_tima
+ TM_SPC_PULL_POOL_CTX
);
435 vp
= xive_pool_vps
+ cpu
;
437 rc
= opal_xive_set_vp_info(vp
, 0, 0);
440 msleep(OPAL_BUSY_DELAY_MS
);
444 void xive_native_sync_source(u32 hw_irq
)
446 opal_xive_sync(XIVE_SYNC_EAS
, hw_irq
);
448 EXPORT_SYMBOL_GPL(xive_native_sync_source
);
450 void xive_native_sync_queue(u32 hw_irq
)
452 opal_xive_sync(XIVE_SYNC_QUEUE
, hw_irq
);
454 EXPORT_SYMBOL_GPL(xive_native_sync_queue
);
456 static const struct xive_ops xive_native_ops
= {
457 .populate_irq_data
= xive_native_populate_irq_data
,
458 .configure_irq
= xive_native_configure_irq
,
459 .get_irq_config
= xive_native_get_irq_config
,
460 .setup_queue
= xive_native_setup_queue
,
461 .cleanup_queue
= xive_native_cleanup_queue
,
462 .match
= xive_native_match
,
463 .shutdown
= xive_native_shutdown
,
464 .update_pending
= xive_native_update_pending
,
465 .setup_cpu
= xive_native_setup_cpu
,
466 .teardown_cpu
= xive_native_teardown_cpu
,
467 .sync_source
= xive_native_sync_source
,
469 .get_ipi
= xive_native_get_ipi
,
470 .put_ipi
= xive_native_put_ipi
,
471 #endif /* CONFIG_SMP */
475 static bool xive_parse_provisioning(struct device_node
*np
)
479 if (of_property_read_u32(np
, "ibm,xive-provision-page-size",
480 &xive_provision_size
) < 0)
482 rc
= of_property_count_elems_of_size(np
, "ibm,xive-provision-chips", 4);
484 pr_err("Error %d getting provision chips array\n", rc
);
487 xive_provision_chip_count
= rc
;
491 xive_provision_chips
= kcalloc(4, xive_provision_chip_count
,
493 if (WARN_ON(!xive_provision_chips
))
496 rc
= of_property_read_u32_array(np
, "ibm,xive-provision-chips",
497 xive_provision_chips
,
498 xive_provision_chip_count
);
500 pr_err("Error %d reading provision chips array\n", rc
);
504 xive_provision_cache
= kmem_cache_create("xive-provision",
508 if (!xive_provision_cache
) {
509 pr_err("Failed to allocate provision cache\n");
515 static void xive_native_setup_pools(void)
517 /* Allocate a pool big enough */
518 pr_debug("XIVE: Allocating VP block for pool size %u\n", nr_cpu_ids
);
520 xive_pool_vps
= xive_native_alloc_vp_block(nr_cpu_ids
);
521 if (WARN_ON(xive_pool_vps
== XIVE_INVALID_VP
))
522 pr_err("XIVE: Failed to allocate pool VP, KVM might not function\n");
524 pr_debug("XIVE: Pool VPs allocated at 0x%x for %u max CPUs\n",
525 xive_pool_vps
, nr_cpu_ids
);
528 u32
xive_native_default_eq_shift(void)
530 return xive_queue_shift
;
532 EXPORT_SYMBOL_GPL(xive_native_default_eq_shift
);
534 unsigned long xive_tima_os
;
535 EXPORT_SYMBOL_GPL(xive_tima_os
);
537 bool __init
xive_native_init(void)
539 struct device_node
*np
;
542 struct property
*prop
;
548 if (xive_cmdline_disabled
)
551 pr_devel("xive_native_init()\n");
552 np
= of_find_compatible_node(NULL
, NULL
, "ibm,opal-xive-pe");
554 pr_devel("not found !\n");
557 pr_devel("Found %pOF\n", np
);
559 /* Resource 1 is HV window */
560 if (of_address_to_resource(np
, 1, &r
)) {
561 pr_err("Failed to get thread mgmnt area resource\n");
564 tima
= ioremap(r
.start
, resource_size(&r
));
566 pr_err("Failed to map thread mgmnt area\n");
570 /* Read number of priorities */
571 if (of_property_read_u32(np
, "ibm,xive-#priorities", &val
) == 0)
574 /* Iterate the EQ sizes and pick one */
575 of_property_for_each_u32(np
, "ibm,xive-eq-sizes", prop
, p
, val
) {
576 xive_queue_shift
= val
;
577 if (val
== PAGE_SHIFT
)
581 /* Do we support single escalation */
582 if (of_get_property(np
, "single-escalation-support", NULL
) != NULL
)
583 xive_has_single_esc
= true;
585 /* Configure Thread Management areas for KVM */
586 for_each_possible_cpu(cpu
)
587 kvmppc_set_xive_tima(cpu
, r
.start
, tima
);
589 /* Resource 2 is OS window */
590 if (of_address_to_resource(np
, 2, &r
)) {
591 pr_err("Failed to get thread mgmnt area resource\n");
595 xive_tima_os
= r
.start
;
597 /* Grab size of provisionning pages */
598 xive_parse_provisioning(np
);
600 /* Switch the XIVE to exploitation mode */
601 rc
= opal_xive_reset(OPAL_XIVE_MODE_EXPL
);
603 pr_err("Switch to exploitation mode failed with error %lld\n", rc
);
607 /* Setup some dummy HV pool VPs */
608 xive_native_setup_pools();
610 /* Initialize XIVE core with our backend */
611 if (!xive_core_init(np
, &xive_native_ops
, tima
, TM_QW3_HV_PHYS
,
613 opal_xive_reset(OPAL_XIVE_MODE_EMU
);
616 pr_info("Using %dkB queues\n", 1 << (xive_queue_shift
- 10));
620 static bool xive_native_provision_pages(void)
625 for (i
= 0; i
< xive_provision_chip_count
; i
++) {
626 u32 chip
= xive_provision_chips
[i
];
629 * XXX TODO: Try to make the allocation local to the node where
632 p
= kmem_cache_alloc(xive_provision_cache
, GFP_KERNEL
);
634 pr_err("Failed to allocate provisioning page\n");
638 opal_xive_donate_page(chip
, __pa(p
));
643 u32
xive_native_alloc_vp_block(u32 max_vcpus
)
648 order
= fls(max_vcpus
) - 1;
649 if (max_vcpus
> (1 << order
))
652 pr_debug("VP block alloc, for max VCPUs %d use order %d\n",
656 rc
= opal_xive_alloc_vp_block(order
);
659 msleep(OPAL_BUSY_DELAY_MS
);
661 case OPAL_XIVE_PROVISIONING
:
662 if (!xive_native_provision_pages())
663 return XIVE_INVALID_VP
;
667 pr_err("OPAL failed to allocate VCPUs order %d, err %lld\n",
669 return XIVE_INVALID_VP
;
675 EXPORT_SYMBOL_GPL(xive_native_alloc_vp_block
);
677 void xive_native_free_vp_block(u32 vp_base
)
681 if (vp_base
== XIVE_INVALID_VP
)
684 rc
= opal_xive_free_vp_block(vp_base
);
686 pr_warn("OPAL error %lld freeing VP block\n", rc
);
688 EXPORT_SYMBOL_GPL(xive_native_free_vp_block
);
690 int xive_native_enable_vp(u32 vp_id
, bool single_escalation
)
693 u64 flags
= OPAL_XIVE_VP_ENABLED
;
695 if (single_escalation
)
696 flags
|= OPAL_XIVE_VP_SINGLE_ESCALATION
;
698 rc
= opal_xive_set_vp_info(vp_id
, flags
, 0);
701 msleep(OPAL_BUSY_DELAY_MS
);
704 vp_err(vp_id
, "Failed to enable VP : %lld\n", rc
);
705 return rc
? -EIO
: 0;
707 EXPORT_SYMBOL_GPL(xive_native_enable_vp
);
709 int xive_native_disable_vp(u32 vp_id
)
714 rc
= opal_xive_set_vp_info(vp_id
, 0, 0);
717 msleep(OPAL_BUSY_DELAY_MS
);
720 vp_err(vp_id
, "Failed to disable VP : %lld\n", rc
);
721 return rc
? -EIO
: 0;
723 EXPORT_SYMBOL_GPL(xive_native_disable_vp
);
725 int xive_native_get_vp_info(u32 vp_id
, u32
*out_cam_id
, u32
*out_chip_id
)
728 __be32 vp_chip_id_be
;
731 rc
= opal_xive_get_vp_info(vp_id
, NULL
, &vp_cam_be
, NULL
, &vp_chip_id_be
);
733 vp_err(vp_id
, "Failed to get VP info : %lld\n", rc
);
736 *out_cam_id
= be64_to_cpu(vp_cam_be
) & 0xffffffffu
;
737 *out_chip_id
= be32_to_cpu(vp_chip_id_be
);
741 EXPORT_SYMBOL_GPL(xive_native_get_vp_info
);
743 bool xive_native_has_single_escalation(void)
745 return xive_has_single_esc
;
747 EXPORT_SYMBOL_GPL(xive_native_has_single_escalation
);
749 int xive_native_get_queue_info(u32 vp_id
, u32 prio
,
753 u32
*out_escalate_irq
,
763 rc
= opal_xive_get_queue_info(vp_id
, prio
, &qpage
, &qsize
,
764 &qeoi_page
, &escalate_irq
, &qflags
);
766 vp_err(vp_id
, "failed to get queue %d info : %lld\n", prio
, rc
);
771 *out_qpage
= be64_to_cpu(qpage
);
773 *out_qsize
= be32_to_cpu(qsize
);
775 *out_qeoi_page
= be64_to_cpu(qeoi_page
);
776 if (out_escalate_irq
)
777 *out_escalate_irq
= be32_to_cpu(escalate_irq
);
779 *out_qflags
= be64_to_cpu(qflags
);
783 EXPORT_SYMBOL_GPL(xive_native_get_queue_info
);
785 int xive_native_get_queue_state(u32 vp_id
, u32 prio
, u32
*qtoggle
, u32
*qindex
)
791 rc
= opal_xive_get_queue_state(vp_id
, prio
, &opal_qtoggle
,
794 vp_err(vp_id
, "failed to get queue %d state : %lld\n", prio
, rc
);
799 *qtoggle
= be32_to_cpu(opal_qtoggle
);
801 *qindex
= be32_to_cpu(opal_qindex
);
805 EXPORT_SYMBOL_GPL(xive_native_get_queue_state
);
807 int xive_native_set_queue_state(u32 vp_id
, u32 prio
, u32 qtoggle
, u32 qindex
)
811 rc
= opal_xive_set_queue_state(vp_id
, prio
, qtoggle
, qindex
);
813 vp_err(vp_id
, "failed to set queue %d state : %lld\n", prio
, rc
);
819 EXPORT_SYMBOL_GPL(xive_native_set_queue_state
);
821 bool xive_native_has_queue_state_support(void)
823 return opal_check_token(OPAL_XIVE_GET_QUEUE_STATE
) &&
824 opal_check_token(OPAL_XIVE_SET_QUEUE_STATE
);
826 EXPORT_SYMBOL_GPL(xive_native_has_queue_state_support
);
828 int xive_native_get_vp_state(u32 vp_id
, u64
*out_state
)
833 rc
= opal_xive_get_vp_state(vp_id
, &state
);
835 vp_err(vp_id
, "failed to get vp state : %lld\n", rc
);
840 *out_state
= be64_to_cpu(state
);
843 EXPORT_SYMBOL_GPL(xive_native_get_vp_state
);
845 machine_arch_initcall(powernv
, xive_core_debug_init
);