2 * Copyright (C) 2001 MandrakeSoft S.A.
3 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
8 * http://www.linux-mandrake.com/
9 * http://www.mandrakesoft.com/
11 * This library is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU Lesser General Public
13 * License as published by the Free Software Foundation; either
14 * version 2 of the License, or (at your option) any later version.
16 * This library is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * Lesser General Public License for more details.
21 * You should have received a copy of the GNU Lesser General Public
22 * License along with this library; if not, write to the Free Software
23 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
25 * Yunhong Jiang <yunhong.jiang@intel.com>
26 * Yaozu (Eddie) Dong <eddie.dong@intel.com>
27 * Based on Xen 3.1 code.
30 #include <linux/kvm_host.h>
31 #include <linux/kvm.h>
33 #include <linux/highmem.h>
34 #include <linux/smp.h>
35 #include <linux/hrtimer.h>
37 #include <linux/slab.h>
38 #include <linux/export.h>
39 #include <asm/processor.h>
41 #include <asm/current.h>
42 #include <trace/events/kvm.h>
49 #define ioapic_debug(fmt,arg...) printk(KERN_WARNING fmt,##arg)
51 #define ioapic_debug(fmt, arg...)
53 static int ioapic_service(struct kvm_ioapic
*vioapic
, int irq
,
56 static unsigned long ioapic_read_indirect(struct kvm_ioapic
*ioapic
,
60 unsigned long result
= 0;
62 switch (ioapic
->ioregsel
) {
63 case IOAPIC_REG_VERSION
:
64 result
= ((((IOAPIC_NUM_PINS
- 1) & 0xff) << 16)
65 | (IOAPIC_VERSION_ID
& 0xff));
68 case IOAPIC_REG_APIC_ID
:
69 case IOAPIC_REG_ARB_ID
:
70 result
= ((ioapic
->id
& 0xf) << 24);
75 u32 redir_index
= (ioapic
->ioregsel
- 0x10) >> 1;
78 if (redir_index
< IOAPIC_NUM_PINS
)
80 ioapic
->redirtbl
[redir_index
].bits
;
82 redir_content
= ~0ULL;
84 result
= (ioapic
->ioregsel
& 0x1) ?
85 (redir_content
>> 32) & 0xffffffff :
86 redir_content
& 0xffffffff;
94 static void rtc_irq_eoi_tracking_reset(struct kvm_ioapic
*ioapic
)
96 ioapic
->rtc_status
.pending_eoi
= 0;
97 bitmap_zero(ioapic
->rtc_status
.dest_map
.map
, KVM_MAX_VCPU_ID
);
100 static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic
*ioapic
);
102 static void rtc_status_pending_eoi_check_valid(struct kvm_ioapic
*ioapic
)
104 if (WARN_ON(ioapic
->rtc_status
.pending_eoi
< 0))
105 kvm_rtc_eoi_tracking_restore_all(ioapic
);
108 static void __rtc_irq_eoi_tracking_restore_one(struct kvm_vcpu
*vcpu
)
110 bool new_val
, old_val
;
111 struct kvm_ioapic
*ioapic
= vcpu
->kvm
->arch
.vioapic
;
112 struct dest_map
*dest_map
= &ioapic
->rtc_status
.dest_map
;
113 union kvm_ioapic_redirect_entry
*e
;
115 e
= &ioapic
->redirtbl
[RTC_GSI
];
116 if (!kvm_apic_match_dest(vcpu
, NULL
, 0, e
->fields
.dest_id
,
117 e
->fields
.dest_mode
))
120 new_val
= kvm_apic_pending_eoi(vcpu
, e
->fields
.vector
);
121 old_val
= test_bit(vcpu
->vcpu_id
, dest_map
->map
);
123 if (new_val
== old_val
)
127 __set_bit(vcpu
->vcpu_id
, dest_map
->map
);
128 dest_map
->vectors
[vcpu
->vcpu_id
] = e
->fields
.vector
;
129 ioapic
->rtc_status
.pending_eoi
++;
131 __clear_bit(vcpu
->vcpu_id
, dest_map
->map
);
132 ioapic
->rtc_status
.pending_eoi
--;
133 rtc_status_pending_eoi_check_valid(ioapic
);
137 void kvm_rtc_eoi_tracking_restore_one(struct kvm_vcpu
*vcpu
)
139 struct kvm_ioapic
*ioapic
= vcpu
->kvm
->arch
.vioapic
;
141 spin_lock(&ioapic
->lock
);
142 __rtc_irq_eoi_tracking_restore_one(vcpu
);
143 spin_unlock(&ioapic
->lock
);
146 static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic
*ioapic
)
148 struct kvm_vcpu
*vcpu
;
151 if (RTC_GSI
>= IOAPIC_NUM_PINS
)
154 rtc_irq_eoi_tracking_reset(ioapic
);
155 kvm_for_each_vcpu(i
, vcpu
, ioapic
->kvm
)
156 __rtc_irq_eoi_tracking_restore_one(vcpu
);
159 static void rtc_irq_eoi(struct kvm_ioapic
*ioapic
, struct kvm_vcpu
*vcpu
)
161 if (test_and_clear_bit(vcpu
->vcpu_id
,
162 ioapic
->rtc_status
.dest_map
.map
)) {
163 --ioapic
->rtc_status
.pending_eoi
;
164 rtc_status_pending_eoi_check_valid(ioapic
);
168 static bool rtc_irq_check_coalesced(struct kvm_ioapic
*ioapic
)
170 if (ioapic
->rtc_status
.pending_eoi
> 0)
171 return true; /* coalesced */
176 static int ioapic_set_irq(struct kvm_ioapic
*ioapic
, unsigned int irq
,
177 int irq_level
, bool line_status
)
179 union kvm_ioapic_redirect_entry entry
;
184 entry
= ioapic
->redirtbl
[irq
];
185 edge
= (entry
.fields
.trig_mode
== IOAPIC_EDGE_TRIG
);
188 ioapic
->irr
&= ~mask
;
194 * Return 0 for coalesced interrupts; for edge-triggered interrupts,
195 * this only happens if a previous edge has not been delivered due
196 * do masking. For level interrupts, the remote_irr field tells
197 * us if the interrupt is waiting for an EOI.
199 * RTC is special: it is edge-triggered, but userspace likes to know
200 * if it has been already ack-ed via EOI because coalesced RTC
201 * interrupts lead to time drift in Windows guests. So we track
202 * EOI manually for the RTC interrupt.
204 if (irq
== RTC_GSI
&& line_status
&&
205 rtc_irq_check_coalesced(ioapic
)) {
210 old_irr
= ioapic
->irr
;
213 ioapic
->irr_delivered
&= ~mask
;
214 if (old_irr
== ioapic
->irr
) {
220 ret
= ioapic_service(ioapic
, irq
, line_status
);
223 trace_kvm_ioapic_set_irq(entry
.bits
, irq
, ret
== 0);
227 static void kvm_ioapic_inject_all(struct kvm_ioapic
*ioapic
, unsigned long irr
)
231 rtc_irq_eoi_tracking_reset(ioapic
);
232 for_each_set_bit(idx
, &irr
, IOAPIC_NUM_PINS
)
233 ioapic_set_irq(ioapic
, idx
, 1, true);
235 kvm_rtc_eoi_tracking_restore_all(ioapic
);
239 void kvm_ioapic_scan_entry(struct kvm_vcpu
*vcpu
, ulong
*ioapic_handled_vectors
)
241 struct kvm_ioapic
*ioapic
= vcpu
->kvm
->arch
.vioapic
;
242 struct dest_map
*dest_map
= &ioapic
->rtc_status
.dest_map
;
243 union kvm_ioapic_redirect_entry
*e
;
246 spin_lock(&ioapic
->lock
);
248 /* Make sure we see any missing RTC EOI */
249 if (test_bit(vcpu
->vcpu_id
, dest_map
->map
))
250 __set_bit(dest_map
->vectors
[vcpu
->vcpu_id
],
251 ioapic_handled_vectors
);
253 for (index
= 0; index
< IOAPIC_NUM_PINS
; index
++) {
254 e
= &ioapic
->redirtbl
[index
];
255 if (e
->fields
.trig_mode
== IOAPIC_LEVEL_TRIG
||
256 kvm_irq_has_notifier(ioapic
->kvm
, KVM_IRQCHIP_IOAPIC
, index
) ||
258 if (kvm_apic_match_dest(vcpu
, NULL
, 0,
259 e
->fields
.dest_id
, e
->fields
.dest_mode
) ||
260 kvm_apic_pending_eoi(vcpu
, e
->fields
.vector
))
261 __set_bit(e
->fields
.vector
,
262 ioapic_handled_vectors
);
265 spin_unlock(&ioapic
->lock
);
268 void kvm_arch_post_irq_ack_notifier_list_update(struct kvm
*kvm
)
270 if (!ioapic_in_kernel(kvm
))
272 kvm_make_scan_ioapic_request(kvm
);
275 static void ioapic_write_indirect(struct kvm_ioapic
*ioapic
, u32 val
)
278 bool mask_before
, mask_after
;
279 int old_remote_irr
, old_delivery_status
;
280 union kvm_ioapic_redirect_entry
*e
;
282 switch (ioapic
->ioregsel
) {
283 case IOAPIC_REG_VERSION
:
284 /* Writes are ignored. */
287 case IOAPIC_REG_APIC_ID
:
288 ioapic
->id
= (val
>> 24) & 0xf;
291 case IOAPIC_REG_ARB_ID
:
295 index
= (ioapic
->ioregsel
- 0x10) >> 1;
297 ioapic_debug("change redir index %x val %x\n", index
, val
);
298 if (index
>= IOAPIC_NUM_PINS
)
300 e
= &ioapic
->redirtbl
[index
];
301 mask_before
= e
->fields
.mask
;
302 /* Preserve read-only fields */
303 old_remote_irr
= e
->fields
.remote_irr
;
304 old_delivery_status
= e
->fields
.delivery_status
;
305 if (ioapic
->ioregsel
& 1) {
306 e
->bits
&= 0xffffffff;
307 e
->bits
|= (u64
) val
<< 32;
309 e
->bits
&= ~0xffffffffULL
;
310 e
->bits
|= (u32
) val
;
312 e
->fields
.remote_irr
= old_remote_irr
;
313 e
->fields
.delivery_status
= old_delivery_status
;
316 * Some OSes (Linux, Xen) assume that Remote IRR bit will
317 * be cleared by IOAPIC hardware when the entry is configured
318 * as edge-triggered. This behavior is used to simulate an
319 * explicit EOI on IOAPICs that don't have the EOI register.
321 if (e
->fields
.trig_mode
== IOAPIC_EDGE_TRIG
)
322 e
->fields
.remote_irr
= 0;
324 mask_after
= e
->fields
.mask
;
325 if (mask_before
!= mask_after
)
326 kvm_fire_mask_notifiers(ioapic
->kvm
, KVM_IRQCHIP_IOAPIC
, index
, mask_after
);
327 if (e
->fields
.trig_mode
== IOAPIC_LEVEL_TRIG
328 && ioapic
->irr
& (1 << index
))
329 ioapic_service(ioapic
, index
, false);
330 kvm_make_scan_ioapic_request(ioapic
->kvm
);
335 static int ioapic_service(struct kvm_ioapic
*ioapic
, int irq
, bool line_status
)
337 union kvm_ioapic_redirect_entry
*entry
= &ioapic
->redirtbl
[irq
];
338 struct kvm_lapic_irq irqe
;
341 if (entry
->fields
.mask
||
342 (entry
->fields
.trig_mode
== IOAPIC_LEVEL_TRIG
&&
343 entry
->fields
.remote_irr
))
346 ioapic_debug("dest=%x dest_mode=%x delivery_mode=%x "
347 "vector=%x trig_mode=%x\n",
348 entry
->fields
.dest_id
, entry
->fields
.dest_mode
,
349 entry
->fields
.delivery_mode
, entry
->fields
.vector
,
350 entry
->fields
.trig_mode
);
352 irqe
.dest_id
= entry
->fields
.dest_id
;
353 irqe
.vector
= entry
->fields
.vector
;
354 irqe
.dest_mode
= entry
->fields
.dest_mode
;
355 irqe
.trig_mode
= entry
->fields
.trig_mode
;
356 irqe
.delivery_mode
= entry
->fields
.delivery_mode
<< 8;
359 irqe
.msi_redir_hint
= false;
361 if (irqe
.trig_mode
== IOAPIC_EDGE_TRIG
)
362 ioapic
->irr_delivered
|= 1 << irq
;
364 if (irq
== RTC_GSI
&& line_status
) {
366 * pending_eoi cannot ever become negative (see
367 * rtc_status_pending_eoi_check_valid) and the caller
368 * ensures that it is only called if it is >= zero, namely
369 * if rtc_irq_check_coalesced returns false).
371 BUG_ON(ioapic
->rtc_status
.pending_eoi
!= 0);
372 ret
= kvm_irq_delivery_to_apic(ioapic
->kvm
, NULL
, &irqe
,
373 &ioapic
->rtc_status
.dest_map
);
374 ioapic
->rtc_status
.pending_eoi
= (ret
< 0 ? 0 : ret
);
376 ret
= kvm_irq_delivery_to_apic(ioapic
->kvm
, NULL
, &irqe
, NULL
);
378 if (ret
&& irqe
.trig_mode
== IOAPIC_LEVEL_TRIG
)
379 entry
->fields
.remote_irr
= 1;
384 int kvm_ioapic_set_irq(struct kvm_ioapic
*ioapic
, int irq
, int irq_source_id
,
385 int level
, bool line_status
)
389 BUG_ON(irq
< 0 || irq
>= IOAPIC_NUM_PINS
);
391 spin_lock(&ioapic
->lock
);
392 irq_level
= __kvm_irq_line_state(&ioapic
->irq_states
[irq
],
393 irq_source_id
, level
);
394 ret
= ioapic_set_irq(ioapic
, irq
, irq_level
, line_status
);
396 spin_unlock(&ioapic
->lock
);
401 void kvm_ioapic_clear_all(struct kvm_ioapic
*ioapic
, int irq_source_id
)
405 spin_lock(&ioapic
->lock
);
406 for (i
= 0; i
< KVM_IOAPIC_NUM_PINS
; i
++)
407 __clear_bit(irq_source_id
, &ioapic
->irq_states
[i
]);
408 spin_unlock(&ioapic
->lock
);
411 static void kvm_ioapic_eoi_inject_work(struct work_struct
*work
)
414 struct kvm_ioapic
*ioapic
= container_of(work
, struct kvm_ioapic
,
416 spin_lock(&ioapic
->lock
);
417 for (i
= 0; i
< IOAPIC_NUM_PINS
; i
++) {
418 union kvm_ioapic_redirect_entry
*ent
= &ioapic
->redirtbl
[i
];
420 if (ent
->fields
.trig_mode
!= IOAPIC_LEVEL_TRIG
)
423 if (ioapic
->irr
& (1 << i
) && !ent
->fields
.remote_irr
)
424 ioapic_service(ioapic
, i
, false);
426 spin_unlock(&ioapic
->lock
);
429 #define IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT 10000
431 static void __kvm_ioapic_update_eoi(struct kvm_vcpu
*vcpu
,
432 struct kvm_ioapic
*ioapic
, int vector
, int trigger_mode
)
434 struct dest_map
*dest_map
= &ioapic
->rtc_status
.dest_map
;
435 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
438 /* RTC special handling */
439 if (test_bit(vcpu
->vcpu_id
, dest_map
->map
) &&
440 vector
== dest_map
->vectors
[vcpu
->vcpu_id
])
441 rtc_irq_eoi(ioapic
, vcpu
);
443 for (i
= 0; i
< IOAPIC_NUM_PINS
; i
++) {
444 union kvm_ioapic_redirect_entry
*ent
= &ioapic
->redirtbl
[i
];
446 if (ent
->fields
.vector
!= vector
)
450 * We are dropping lock while calling ack notifiers because ack
451 * notifier callbacks for assigned devices call into IOAPIC
452 * recursively. Since remote_irr is cleared only after call
453 * to notifiers if the same vector will be delivered while lock
454 * is dropped it will be put into irr and will be delivered
455 * after ack notifier returns.
457 spin_unlock(&ioapic
->lock
);
458 kvm_notify_acked_irq(ioapic
->kvm
, KVM_IRQCHIP_IOAPIC
, i
);
459 spin_lock(&ioapic
->lock
);
461 if (trigger_mode
!= IOAPIC_LEVEL_TRIG
||
462 kvm_lapic_get_reg(apic
, APIC_SPIV
) & APIC_SPIV_DIRECTED_EOI
)
465 ASSERT(ent
->fields
.trig_mode
== IOAPIC_LEVEL_TRIG
);
466 ent
->fields
.remote_irr
= 0;
467 if (!ent
->fields
.mask
&& (ioapic
->irr
& (1 << i
))) {
468 ++ioapic
->irq_eoi
[i
];
469 if (ioapic
->irq_eoi
[i
] == IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT
) {
471 * Real hardware does not deliver the interrupt
472 * immediately during eoi broadcast, and this
473 * lets a buggy guest make slow progress
474 * even if it does not correctly handle a
475 * level-triggered interrupt. Emulate this
476 * behavior if we detect an interrupt storm.
478 schedule_delayed_work(&ioapic
->eoi_inject
, HZ
/ 100);
479 ioapic
->irq_eoi
[i
] = 0;
480 trace_kvm_ioapic_delayed_eoi_inj(ent
->bits
);
482 ioapic_service(ioapic
, i
, false);
485 ioapic
->irq_eoi
[i
] = 0;
490 void kvm_ioapic_update_eoi(struct kvm_vcpu
*vcpu
, int vector
, int trigger_mode
)
492 struct kvm_ioapic
*ioapic
= vcpu
->kvm
->arch
.vioapic
;
494 spin_lock(&ioapic
->lock
);
495 __kvm_ioapic_update_eoi(vcpu
, ioapic
, vector
, trigger_mode
);
496 spin_unlock(&ioapic
->lock
);
499 static inline struct kvm_ioapic
*to_ioapic(struct kvm_io_device
*dev
)
501 return container_of(dev
, struct kvm_ioapic
, dev
);
504 static inline int ioapic_in_range(struct kvm_ioapic
*ioapic
, gpa_t addr
)
506 return ((addr
>= ioapic
->base_address
&&
507 (addr
< ioapic
->base_address
+ IOAPIC_MEM_LENGTH
)));
510 static int ioapic_mmio_read(struct kvm_vcpu
*vcpu
, struct kvm_io_device
*this,
511 gpa_t addr
, int len
, void *val
)
513 struct kvm_ioapic
*ioapic
= to_ioapic(this);
515 if (!ioapic_in_range(ioapic
, addr
))
518 ioapic_debug("addr %lx\n", (unsigned long)addr
);
519 ASSERT(!(addr
& 0xf)); /* check alignment */
522 spin_lock(&ioapic
->lock
);
524 case IOAPIC_REG_SELECT
:
525 result
= ioapic
->ioregsel
;
528 case IOAPIC_REG_WINDOW
:
529 result
= ioapic_read_indirect(ioapic
, addr
, len
);
536 spin_unlock(&ioapic
->lock
);
540 *(u64
*) val
= result
;
545 memcpy(val
, (char *)&result
, len
);
548 printk(KERN_WARNING
"ioapic: wrong length %d\n", len
);
553 static int ioapic_mmio_write(struct kvm_vcpu
*vcpu
, struct kvm_io_device
*this,
554 gpa_t addr
, int len
, const void *val
)
556 struct kvm_ioapic
*ioapic
= to_ioapic(this);
558 if (!ioapic_in_range(ioapic
, addr
))
561 ioapic_debug("ioapic_mmio_write addr=%p len=%d val=%p\n",
562 (void*)addr
, len
, val
);
563 ASSERT(!(addr
& 0xf)); /* check alignment */
577 printk(KERN_WARNING
"ioapic: Unsupported size %d\n", len
);
582 spin_lock(&ioapic
->lock
);
584 case IOAPIC_REG_SELECT
:
585 ioapic
->ioregsel
= data
& 0xFF; /* 8-bit register */
588 case IOAPIC_REG_WINDOW
:
589 ioapic_write_indirect(ioapic
, data
);
595 spin_unlock(&ioapic
->lock
);
599 static void kvm_ioapic_reset(struct kvm_ioapic
*ioapic
)
603 cancel_delayed_work_sync(&ioapic
->eoi_inject
);
604 for (i
= 0; i
< IOAPIC_NUM_PINS
; i
++)
605 ioapic
->redirtbl
[i
].fields
.mask
= 1;
606 ioapic
->base_address
= IOAPIC_DEFAULT_BASE_ADDRESS
;
607 ioapic
->ioregsel
= 0;
609 ioapic
->irr_delivered
= 0;
611 memset(ioapic
->irq_eoi
, 0x00, sizeof(ioapic
->irq_eoi
));
612 rtc_irq_eoi_tracking_reset(ioapic
);
615 static const struct kvm_io_device_ops ioapic_mmio_ops
= {
616 .read
= ioapic_mmio_read
,
617 .write
= ioapic_mmio_write
,
620 int kvm_ioapic_init(struct kvm
*kvm
)
622 struct kvm_ioapic
*ioapic
;
625 ioapic
= kzalloc(sizeof(struct kvm_ioapic
), GFP_KERNEL
);
628 spin_lock_init(&ioapic
->lock
);
629 INIT_DELAYED_WORK(&ioapic
->eoi_inject
, kvm_ioapic_eoi_inject_work
);
630 kvm
->arch
.vioapic
= ioapic
;
631 kvm_ioapic_reset(ioapic
);
632 kvm_iodevice_init(&ioapic
->dev
, &ioapic_mmio_ops
);
634 mutex_lock(&kvm
->slots_lock
);
635 ret
= kvm_io_bus_register_dev(kvm
, KVM_MMIO_BUS
, ioapic
->base_address
,
636 IOAPIC_MEM_LENGTH
, &ioapic
->dev
);
637 mutex_unlock(&kvm
->slots_lock
);
639 kvm
->arch
.vioapic
= NULL
;
646 void kvm_ioapic_destroy(struct kvm
*kvm
)
648 struct kvm_ioapic
*ioapic
= kvm
->arch
.vioapic
;
653 cancel_delayed_work_sync(&ioapic
->eoi_inject
);
654 mutex_lock(&kvm
->slots_lock
);
655 kvm_io_bus_unregister_dev(kvm
, KVM_MMIO_BUS
, &ioapic
->dev
);
656 mutex_unlock(&kvm
->slots_lock
);
657 kvm
->arch
.vioapic
= NULL
;
661 void kvm_get_ioapic(struct kvm
*kvm
, struct kvm_ioapic_state
*state
)
663 struct kvm_ioapic
*ioapic
= kvm
->arch
.vioapic
;
665 spin_lock(&ioapic
->lock
);
666 memcpy(state
, ioapic
, sizeof(struct kvm_ioapic_state
));
667 state
->irr
&= ~ioapic
->irr_delivered
;
668 spin_unlock(&ioapic
->lock
);
671 void kvm_set_ioapic(struct kvm
*kvm
, struct kvm_ioapic_state
*state
)
673 struct kvm_ioapic
*ioapic
= kvm
->arch
.vioapic
;
675 spin_lock(&ioapic
->lock
);
676 memcpy(ioapic
, state
, sizeof(struct kvm_ioapic_state
));
678 ioapic
->irr_delivered
= 0;
679 kvm_make_scan_ioapic_request(kvm
);
680 kvm_ioapic_inject_all(ioapic
, state
->irr
);
681 spin_unlock(&ioapic
->lock
);