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
, KVM_MAX_VCPUS
);
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 union kvm_ioapic_redirect_entry
*e
;
114 e
= &ioapic
->redirtbl
[RTC_GSI
];
115 if (!kvm_apic_match_dest(vcpu
, NULL
, 0, e
->fields
.dest_id
,
116 e
->fields
.dest_mode
))
119 new_val
= kvm_apic_pending_eoi(vcpu
, e
->fields
.vector
);
120 old_val
= test_bit(vcpu
->vcpu_id
, ioapic
->rtc_status
.dest_map
);
122 if (new_val
== old_val
)
126 __set_bit(vcpu
->vcpu_id
, ioapic
->rtc_status
.dest_map
);
127 ioapic
->rtc_status
.pending_eoi
++;
129 __clear_bit(vcpu
->vcpu_id
, ioapic
->rtc_status
.dest_map
);
130 ioapic
->rtc_status
.pending_eoi
--;
131 rtc_status_pending_eoi_check_valid(ioapic
);
135 void kvm_rtc_eoi_tracking_restore_one(struct kvm_vcpu
*vcpu
)
137 struct kvm_ioapic
*ioapic
= vcpu
->kvm
->arch
.vioapic
;
139 spin_lock(&ioapic
->lock
);
140 __rtc_irq_eoi_tracking_restore_one(vcpu
);
141 spin_unlock(&ioapic
->lock
);
144 static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic
*ioapic
)
146 struct kvm_vcpu
*vcpu
;
149 if (RTC_GSI
>= IOAPIC_NUM_PINS
)
152 rtc_irq_eoi_tracking_reset(ioapic
);
153 kvm_for_each_vcpu(i
, vcpu
, ioapic
->kvm
)
154 __rtc_irq_eoi_tracking_restore_one(vcpu
);
157 static void rtc_irq_eoi(struct kvm_ioapic
*ioapic
, struct kvm_vcpu
*vcpu
)
159 if (test_and_clear_bit(vcpu
->vcpu_id
, ioapic
->rtc_status
.dest_map
)) {
160 --ioapic
->rtc_status
.pending_eoi
;
161 rtc_status_pending_eoi_check_valid(ioapic
);
165 static bool rtc_irq_check_coalesced(struct kvm_ioapic
*ioapic
)
167 if (ioapic
->rtc_status
.pending_eoi
> 0)
168 return true; /* coalesced */
173 static int ioapic_set_irq(struct kvm_ioapic
*ioapic
, unsigned int irq
,
174 int irq_level
, bool line_status
)
176 union kvm_ioapic_redirect_entry entry
;
181 entry
= ioapic
->redirtbl
[irq
];
182 edge
= (entry
.fields
.trig_mode
== IOAPIC_EDGE_TRIG
);
185 ioapic
->irr
&= ~mask
;
191 * Return 0 for coalesced interrupts; for edge-triggered interrupts,
192 * this only happens if a previous edge has not been delivered due
193 * do masking. For level interrupts, the remote_irr field tells
194 * us if the interrupt is waiting for an EOI.
196 * RTC is special: it is edge-triggered, but userspace likes to know
197 * if it has been already ack-ed via EOI because coalesced RTC
198 * interrupts lead to time drift in Windows guests. So we track
199 * EOI manually for the RTC interrupt.
201 if (irq
== RTC_GSI
&& line_status
&&
202 rtc_irq_check_coalesced(ioapic
)) {
207 old_irr
= ioapic
->irr
;
209 if ((edge
&& old_irr
== ioapic
->irr
) ||
210 (!edge
&& entry
.fields
.remote_irr
)) {
215 ret
= ioapic_service(ioapic
, irq
, line_status
);
218 trace_kvm_ioapic_set_irq(entry
.bits
, irq
, ret
== 0);
222 static void kvm_ioapic_inject_all(struct kvm_ioapic
*ioapic
, unsigned long irr
)
226 rtc_irq_eoi_tracking_reset(ioapic
);
227 for_each_set_bit(idx
, &irr
, IOAPIC_NUM_PINS
)
228 ioapic_set_irq(ioapic
, idx
, 1, true);
230 kvm_rtc_eoi_tracking_restore_all(ioapic
);
234 static void update_handled_vectors(struct kvm_ioapic
*ioapic
)
236 DECLARE_BITMAP(handled_vectors
, 256);
239 memset(handled_vectors
, 0, sizeof(handled_vectors
));
240 for (i
= 0; i
< IOAPIC_NUM_PINS
; ++i
)
241 __set_bit(ioapic
->redirtbl
[i
].fields
.vector
, handled_vectors
);
242 memcpy(ioapic
->handled_vectors
, handled_vectors
,
243 sizeof(handled_vectors
));
247 void kvm_ioapic_scan_entry(struct kvm_vcpu
*vcpu
, u64
*eoi_exit_bitmap
,
250 struct kvm_ioapic
*ioapic
= vcpu
->kvm
->arch
.vioapic
;
251 union kvm_ioapic_redirect_entry
*e
;
254 spin_lock(&ioapic
->lock
);
255 for (index
= 0; index
< IOAPIC_NUM_PINS
; index
++) {
256 e
= &ioapic
->redirtbl
[index
];
257 if (e
->fields
.trig_mode
== IOAPIC_LEVEL_TRIG
||
258 kvm_irq_has_notifier(ioapic
->kvm
, KVM_IRQCHIP_IOAPIC
, index
) ||
260 if (kvm_apic_match_dest(vcpu
, NULL
, 0,
261 e
->fields
.dest_id
, e
->fields
.dest_mode
)) {
262 __set_bit(e
->fields
.vector
,
263 (unsigned long *)eoi_exit_bitmap
);
264 if (e
->fields
.trig_mode
== IOAPIC_LEVEL_TRIG
)
265 __set_bit(e
->fields
.vector
,
266 (unsigned long *)tmr
);
270 spin_unlock(&ioapic
->lock
);
274 void kvm_vcpu_request_scan_ioapic(struct kvm
*kvm
)
276 struct kvm_ioapic
*ioapic
= kvm
->arch
.vioapic
;
280 kvm_make_scan_ioapic_request(kvm
);
283 void kvm_vcpu_request_scan_ioapic(struct kvm
*kvm
)
289 static void ioapic_write_indirect(struct kvm_ioapic
*ioapic
, u32 val
)
292 bool mask_before
, mask_after
;
293 union kvm_ioapic_redirect_entry
*e
;
295 switch (ioapic
->ioregsel
) {
296 case IOAPIC_REG_VERSION
:
297 /* Writes are ignored. */
300 case IOAPIC_REG_APIC_ID
:
301 ioapic
->id
= (val
>> 24) & 0xf;
304 case IOAPIC_REG_ARB_ID
:
308 index
= (ioapic
->ioregsel
- 0x10) >> 1;
310 ioapic_debug("change redir index %x val %x\n", index
, val
);
311 if (index
>= IOAPIC_NUM_PINS
)
313 e
= &ioapic
->redirtbl
[index
];
314 mask_before
= e
->fields
.mask
;
315 if (ioapic
->ioregsel
& 1) {
316 e
->bits
&= 0xffffffff;
317 e
->bits
|= (u64
) val
<< 32;
319 e
->bits
&= ~0xffffffffULL
;
320 e
->bits
|= (u32
) val
;
321 e
->fields
.remote_irr
= 0;
323 update_handled_vectors(ioapic
);
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_vcpu_request_scan_ioapic(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
)
344 ioapic_debug("dest=%x dest_mode=%x delivery_mode=%x "
345 "vector=%x trig_mode=%x\n",
346 entry
->fields
.dest_id
, entry
->fields
.dest_mode
,
347 entry
->fields
.delivery_mode
, entry
->fields
.vector
,
348 entry
->fields
.trig_mode
);
350 irqe
.dest_id
= entry
->fields
.dest_id
;
351 irqe
.vector
= entry
->fields
.vector
;
352 irqe
.dest_mode
= entry
->fields
.dest_mode
;
353 irqe
.trig_mode
= entry
->fields
.trig_mode
;
354 irqe
.delivery_mode
= entry
->fields
.delivery_mode
<< 8;
358 if (irqe
.trig_mode
== IOAPIC_EDGE_TRIG
)
359 ioapic
->irr
&= ~(1 << irq
);
361 if (irq
== RTC_GSI
&& line_status
) {
363 * pending_eoi cannot ever become negative (see
364 * rtc_status_pending_eoi_check_valid) and the caller
365 * ensures that it is only called if it is >= zero, namely
366 * if rtc_irq_check_coalesced returns false).
368 BUG_ON(ioapic
->rtc_status
.pending_eoi
!= 0);
369 ret
= kvm_irq_delivery_to_apic(ioapic
->kvm
, NULL
, &irqe
,
370 ioapic
->rtc_status
.dest_map
);
371 ioapic
->rtc_status
.pending_eoi
= (ret
< 0 ? 0 : ret
);
373 ret
= kvm_irq_delivery_to_apic(ioapic
->kvm
, NULL
, &irqe
, NULL
);
375 if (ret
&& irqe
.trig_mode
== IOAPIC_LEVEL_TRIG
)
376 entry
->fields
.remote_irr
= 1;
381 int kvm_ioapic_set_irq(struct kvm_ioapic
*ioapic
, int irq
, int irq_source_id
,
382 int level
, bool line_status
)
386 BUG_ON(irq
< 0 || irq
>= IOAPIC_NUM_PINS
);
388 spin_lock(&ioapic
->lock
);
389 irq_level
= __kvm_irq_line_state(&ioapic
->irq_states
[irq
],
390 irq_source_id
, level
);
391 ret
= ioapic_set_irq(ioapic
, irq
, irq_level
, line_status
);
393 spin_unlock(&ioapic
->lock
);
398 void kvm_ioapic_clear_all(struct kvm_ioapic
*ioapic
, int irq_source_id
)
402 spin_lock(&ioapic
->lock
);
403 for (i
= 0; i
< KVM_IOAPIC_NUM_PINS
; i
++)
404 __clear_bit(irq_source_id
, &ioapic
->irq_states
[i
]);
405 spin_unlock(&ioapic
->lock
);
408 static void kvm_ioapic_eoi_inject_work(struct work_struct
*work
)
411 struct kvm_ioapic
*ioapic
= container_of(work
, struct kvm_ioapic
,
413 spin_lock(&ioapic
->lock
);
414 for (i
= 0; i
< IOAPIC_NUM_PINS
; i
++) {
415 union kvm_ioapic_redirect_entry
*ent
= &ioapic
->redirtbl
[i
];
417 if (ent
->fields
.trig_mode
!= IOAPIC_LEVEL_TRIG
)
420 if (ioapic
->irr
& (1 << i
) && !ent
->fields
.remote_irr
)
421 ioapic_service(ioapic
, i
, false);
423 spin_unlock(&ioapic
->lock
);
426 #define IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT 10000
428 static void __kvm_ioapic_update_eoi(struct kvm_vcpu
*vcpu
,
429 struct kvm_ioapic
*ioapic
, int vector
, int trigger_mode
)
433 for (i
= 0; i
< IOAPIC_NUM_PINS
; i
++) {
434 union kvm_ioapic_redirect_entry
*ent
= &ioapic
->redirtbl
[i
];
436 if (ent
->fields
.vector
!= vector
)
440 rtc_irq_eoi(ioapic
, vcpu
);
442 * We are dropping lock while calling ack notifiers because ack
443 * notifier callbacks for assigned devices call into IOAPIC
444 * recursively. Since remote_irr is cleared only after call
445 * to notifiers if the same vector will be delivered while lock
446 * is dropped it will be put into irr and will be delivered
447 * after ack notifier returns.
449 spin_unlock(&ioapic
->lock
);
450 kvm_notify_acked_irq(ioapic
->kvm
, KVM_IRQCHIP_IOAPIC
, i
);
451 spin_lock(&ioapic
->lock
);
453 if (trigger_mode
!= IOAPIC_LEVEL_TRIG
)
456 ASSERT(ent
->fields
.trig_mode
== IOAPIC_LEVEL_TRIG
);
457 ent
->fields
.remote_irr
= 0;
458 if (!ent
->fields
.mask
&& (ioapic
->irr
& (1 << i
))) {
459 ++ioapic
->irq_eoi
[i
];
460 if (ioapic
->irq_eoi
[i
] == IOAPIC_SUCCESSIVE_IRQ_MAX_COUNT
) {
462 * Real hardware does not deliver the interrupt
463 * immediately during eoi broadcast, and this
464 * lets a buggy guest make slow progress
465 * even if it does not correctly handle a
466 * level-triggered interrupt. Emulate this
467 * behavior if we detect an interrupt storm.
469 schedule_delayed_work(&ioapic
->eoi_inject
, HZ
/ 100);
470 ioapic
->irq_eoi
[i
] = 0;
471 trace_kvm_ioapic_delayed_eoi_inj(ent
->bits
);
473 ioapic_service(ioapic
, i
, false);
476 ioapic
->irq_eoi
[i
] = 0;
481 bool kvm_ioapic_handles_vector(struct kvm
*kvm
, int vector
)
483 struct kvm_ioapic
*ioapic
= kvm
->arch
.vioapic
;
485 return test_bit(vector
, ioapic
->handled_vectors
);
488 void kvm_ioapic_update_eoi(struct kvm_vcpu
*vcpu
, int vector
, int trigger_mode
)
490 struct kvm_ioapic
*ioapic
= vcpu
->kvm
->arch
.vioapic
;
492 spin_lock(&ioapic
->lock
);
493 __kvm_ioapic_update_eoi(vcpu
, ioapic
, vector
, trigger_mode
);
494 spin_unlock(&ioapic
->lock
);
497 static inline struct kvm_ioapic
*to_ioapic(struct kvm_io_device
*dev
)
499 return container_of(dev
, struct kvm_ioapic
, dev
);
502 static inline int ioapic_in_range(struct kvm_ioapic
*ioapic
, gpa_t addr
)
504 return ((addr
>= ioapic
->base_address
&&
505 (addr
< ioapic
->base_address
+ IOAPIC_MEM_LENGTH
)));
508 static int ioapic_mmio_read(struct kvm_io_device
*this, gpa_t addr
, int len
,
511 struct kvm_ioapic
*ioapic
= to_ioapic(this);
513 if (!ioapic_in_range(ioapic
, addr
))
516 ioapic_debug("addr %lx\n", (unsigned long)addr
);
517 ASSERT(!(addr
& 0xf)); /* check alignment */
520 spin_lock(&ioapic
->lock
);
522 case IOAPIC_REG_SELECT
:
523 result
= ioapic
->ioregsel
;
526 case IOAPIC_REG_WINDOW
:
527 result
= ioapic_read_indirect(ioapic
, addr
, len
);
534 spin_unlock(&ioapic
->lock
);
538 *(u64
*) val
= result
;
543 memcpy(val
, (char *)&result
, len
);
546 printk(KERN_WARNING
"ioapic: wrong length %d\n", len
);
551 static int ioapic_mmio_write(struct kvm_io_device
*this, gpa_t addr
, int len
,
554 struct kvm_ioapic
*ioapic
= to_ioapic(this);
556 if (!ioapic_in_range(ioapic
, addr
))
559 ioapic_debug("ioapic_mmio_write addr=%p len=%d val=%p\n",
560 (void*)addr
, len
, val
);
561 ASSERT(!(addr
& 0xf)); /* check alignment */
575 printk(KERN_WARNING
"ioapic: Unsupported size %d\n", len
);
580 spin_lock(&ioapic
->lock
);
582 case IOAPIC_REG_SELECT
:
583 ioapic
->ioregsel
= data
& 0xFF; /* 8-bit register */
586 case IOAPIC_REG_WINDOW
:
587 ioapic_write_indirect(ioapic
, data
);
591 __kvm_ioapic_update_eoi(NULL
, ioapic
, data
, IOAPIC_LEVEL_TRIG
);
598 spin_unlock(&ioapic
->lock
);
602 static void kvm_ioapic_reset(struct kvm_ioapic
*ioapic
)
606 cancel_delayed_work_sync(&ioapic
->eoi_inject
);
607 for (i
= 0; i
< IOAPIC_NUM_PINS
; i
++)
608 ioapic
->redirtbl
[i
].fields
.mask
= 1;
609 ioapic
->base_address
= IOAPIC_DEFAULT_BASE_ADDRESS
;
610 ioapic
->ioregsel
= 0;
613 memset(ioapic
->irq_eoi
, 0x00, IOAPIC_NUM_PINS
);
614 rtc_irq_eoi_tracking_reset(ioapic
);
615 update_handled_vectors(ioapic
);
618 static const struct kvm_io_device_ops ioapic_mmio_ops
= {
619 .read
= ioapic_mmio_read
,
620 .write
= ioapic_mmio_write
,
623 int kvm_ioapic_init(struct kvm
*kvm
)
625 struct kvm_ioapic
*ioapic
;
628 ioapic
= kzalloc(sizeof(struct kvm_ioapic
), GFP_KERNEL
);
631 spin_lock_init(&ioapic
->lock
);
632 INIT_DELAYED_WORK(&ioapic
->eoi_inject
, kvm_ioapic_eoi_inject_work
);
633 kvm
->arch
.vioapic
= ioapic
;
634 kvm_ioapic_reset(ioapic
);
635 kvm_iodevice_init(&ioapic
->dev
, &ioapic_mmio_ops
);
637 mutex_lock(&kvm
->slots_lock
);
638 ret
= kvm_io_bus_register_dev(kvm
, KVM_MMIO_BUS
, ioapic
->base_address
,
639 IOAPIC_MEM_LENGTH
, &ioapic
->dev
);
640 mutex_unlock(&kvm
->slots_lock
);
642 kvm
->arch
.vioapic
= NULL
;
649 void kvm_ioapic_destroy(struct kvm
*kvm
)
651 struct kvm_ioapic
*ioapic
= kvm
->arch
.vioapic
;
653 cancel_delayed_work_sync(&ioapic
->eoi_inject
);
655 kvm_io_bus_unregister_dev(kvm
, KVM_MMIO_BUS
, &ioapic
->dev
);
656 kvm
->arch
.vioapic
= NULL
;
661 int kvm_get_ioapic(struct kvm
*kvm
, struct kvm_ioapic_state
*state
)
663 struct kvm_ioapic
*ioapic
= ioapic_irqchip(kvm
);
667 spin_lock(&ioapic
->lock
);
668 memcpy(state
, ioapic
, sizeof(struct kvm_ioapic_state
));
669 spin_unlock(&ioapic
->lock
);
673 int kvm_set_ioapic(struct kvm
*kvm
, struct kvm_ioapic_state
*state
)
675 struct kvm_ioapic
*ioapic
= ioapic_irqchip(kvm
);
679 spin_lock(&ioapic
->lock
);
680 memcpy(ioapic
, state
, sizeof(struct kvm_ioapic_state
));
682 update_handled_vectors(ioapic
);
683 kvm_vcpu_request_scan_ioapic(kvm
);
684 kvm_ioapic_inject_all(ioapic
, state
->irr
);
685 spin_unlock(&ioapic
->lock
);