656fa455e154abedceba3883cf865adfa73f873b
[linux-3.10.git] / virt / kvm / irq_comm.c
1 /*
2  * irq_comm.c: Common API for in kernel interrupt controller
3  * Copyright (c) 2007, Intel Corporation.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms and conditions of the GNU General Public License,
7  * version 2, as published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope it will be useful, but WITHOUT
10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
12  * more details.
13  *
14  * You should have received a copy of the GNU General Public License along with
15  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
16  * Place - Suite 330, Boston, MA 02111-1307 USA.
17  * Authors:
18  *   Yaozu (Eddie) Dong <Eddie.dong@intel.com>
19  *
20  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
21  */
22
23 #include <linux/kvm_host.h>
24 #include <linux/slab.h>
25 #include <trace/events/kvm.h>
26
27 #include <asm/msidef.h>
28 #ifdef CONFIG_IA64
29 #include <asm/iosapic.h>
30 #endif
31
32 #include "irq.h"
33
34 #include "ioapic.h"
35
36 static int kvm_set_pic_irq(struct kvm_kernel_irq_routing_entry *e,
37                            struct kvm *kvm, int irq_source_id, int level)
38 {
39 #ifdef CONFIG_X86
40         struct kvm_pic *pic = pic_irqchip(kvm);
41         return kvm_pic_set_irq(pic, e->irqchip.pin, irq_source_id, level);
42 #else
43         return -1;
44 #endif
45 }
46
47 static int kvm_set_ioapic_irq(struct kvm_kernel_irq_routing_entry *e,
48                               struct kvm *kvm, int irq_source_id, int level)
49 {
50         struct kvm_ioapic *ioapic = kvm->arch.vioapic;
51         return kvm_ioapic_set_irq(ioapic, e->irqchip.pin, irq_source_id, level);
52 }
53
54 inline static bool kvm_is_dm_lowest_prio(struct kvm_lapic_irq *irq)
55 {
56 #ifdef CONFIG_IA64
57         return irq->delivery_mode ==
58                 (IOSAPIC_LOWEST_PRIORITY << IOSAPIC_DELIVERY_SHIFT);
59 #else
60         return irq->delivery_mode == APIC_DM_LOWEST;
61 #endif
62 }
63
64 int kvm_irq_delivery_to_apic(struct kvm *kvm, struct kvm_lapic *src,
65                 struct kvm_lapic_irq *irq)
66 {
67         int i, r = -1;
68         struct kvm_vcpu *vcpu, *lowest = NULL;
69
70         if (irq->dest_mode == 0 && irq->dest_id == 0xff &&
71                         kvm_is_dm_lowest_prio(irq)) {
72                 printk(KERN_INFO "kvm: apic: phys broadcast and lowest prio\n");
73                 irq->delivery_mode = APIC_DM_FIXED;
74         }
75
76         if (kvm_irq_delivery_to_apic_fast(kvm, src, irq, &r))
77                 return r;
78
79         kvm_for_each_vcpu(i, vcpu, kvm) {
80                 if (!kvm_apic_present(vcpu))
81                         continue;
82
83                 if (!kvm_apic_match_dest(vcpu, src, irq->shorthand,
84                                         irq->dest_id, irq->dest_mode))
85                         continue;
86
87                 if (!kvm_is_dm_lowest_prio(irq)) {
88                         if (r < 0)
89                                 r = 0;
90                         r += kvm_apic_set_irq(vcpu, irq);
91                 } else if (kvm_lapic_enabled(vcpu)) {
92                         if (!lowest)
93                                 lowest = vcpu;
94                         else if (kvm_apic_compare_prio(vcpu, lowest) < 0)
95                                 lowest = vcpu;
96                 }
97         }
98
99         if (lowest)
100                 r = kvm_apic_set_irq(lowest, irq);
101
102         return r;
103 }
104
105 static inline void kvm_set_msi_irq(struct kvm_kernel_irq_routing_entry *e,
106                                    struct kvm_lapic_irq *irq)
107 {
108         trace_kvm_msi_set_irq(e->msi.address_lo, e->msi.data);
109
110         irq->dest_id = (e->msi.address_lo &
111                         MSI_ADDR_DEST_ID_MASK) >> MSI_ADDR_DEST_ID_SHIFT;
112         irq->vector = (e->msi.data &
113                         MSI_DATA_VECTOR_MASK) >> MSI_DATA_VECTOR_SHIFT;
114         irq->dest_mode = (1 << MSI_ADDR_DEST_MODE_SHIFT) & e->msi.address_lo;
115         irq->trig_mode = (1 << MSI_DATA_TRIGGER_SHIFT) & e->msi.data;
116         irq->delivery_mode = e->msi.data & 0x700;
117         irq->level = 1;
118         irq->shorthand = 0;
119         /* TODO Deal with RH bit of MSI message address */
120 }
121
122 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e,
123                 struct kvm *kvm, int irq_source_id, int level)
124 {
125         struct kvm_lapic_irq irq;
126
127         if (!level)
128                 return -1;
129
130         kvm_set_msi_irq(e, &irq);
131
132         return kvm_irq_delivery_to_apic(kvm, NULL, &irq);
133 }
134
135
136 static int kvm_set_msi_inatomic(struct kvm_kernel_irq_routing_entry *e,
137                          struct kvm *kvm)
138 {
139         struct kvm_lapic_irq irq;
140         int r;
141
142         kvm_set_msi_irq(e, &irq);
143
144         if (kvm_irq_delivery_to_apic_fast(kvm, NULL, &irq, &r))
145                 return r;
146         else
147                 return -EWOULDBLOCK;
148 }
149
150 int kvm_send_userspace_msi(struct kvm *kvm, struct kvm_msi *msi)
151 {
152         struct kvm_kernel_irq_routing_entry route;
153
154         if (!irqchip_in_kernel(kvm) || msi->flags != 0)
155                 return -EINVAL;
156
157         route.msi.address_lo = msi->address_lo;
158         route.msi.address_hi = msi->address_hi;
159         route.msi.data = msi->data;
160
161         return kvm_set_msi(&route, kvm, KVM_USERSPACE_IRQ_SOURCE_ID, 1);
162 }
163
164 /*
165  * Return value:
166  *  < 0   Interrupt was ignored (masked or not delivered for other reasons)
167  *  = 0   Interrupt was coalesced (previous irq is still pending)
168  *  > 0   Number of CPUs interrupt was delivered to
169  */
170 int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level)
171 {
172         struct kvm_kernel_irq_routing_entry *e, irq_set[KVM_NR_IRQCHIPS];
173         int ret = -1, i = 0;
174         struct kvm_irq_routing_table *irq_rt;
175         struct hlist_node *n;
176
177         trace_kvm_set_irq(irq, level, irq_source_id);
178
179         /* Not possible to detect if the guest uses the PIC or the
180          * IOAPIC.  So set the bit in both. The guest will ignore
181          * writes to the unused one.
182          */
183         rcu_read_lock();
184         irq_rt = rcu_dereference(kvm->irq_routing);
185         if (irq < irq_rt->nr_rt_entries)
186                 hlist_for_each_entry(e, n, &irq_rt->map[irq], link)
187                         irq_set[i++] = *e;
188         rcu_read_unlock();
189
190         while(i--) {
191                 int r;
192                 r = irq_set[i].set(&irq_set[i], kvm, irq_source_id, level);
193                 if (r < 0)
194                         continue;
195
196                 ret = r + ((ret < 0) ? 0 : ret);
197         }
198
199         return ret;
200 }
201
202 /*
203  * Deliver an IRQ in an atomic context if we can, or return a failure,
204  * user can retry in a process context.
205  * Return value:
206  *  -EWOULDBLOCK - Can't deliver in atomic context: retry in a process context.
207  *  Other values - No need to retry.
208  */
209 int kvm_set_irq_inatomic(struct kvm *kvm, int irq_source_id, u32 irq, int level)
210 {
211         struct kvm_kernel_irq_routing_entry *e;
212         int ret = -EINVAL;
213         struct kvm_irq_routing_table *irq_rt;
214         struct hlist_node *n;
215
216         trace_kvm_set_irq(irq, level, irq_source_id);
217
218         /*
219          * Injection into either PIC or IOAPIC might need to scan all CPUs,
220          * which would need to be retried from thread context;  when same GSI
221          * is connected to both PIC and IOAPIC, we'd have to report a
222          * partial failure here.
223          * Since there's no easy way to do this, we only support injecting MSI
224          * which is limited to 1:1 GSI mapping.
225          */
226         rcu_read_lock();
227         irq_rt = rcu_dereference(kvm->irq_routing);
228         if (irq < irq_rt->nr_rt_entries)
229                 hlist_for_each_entry(e, n, &irq_rt->map[irq], link) {
230                         if (likely(e->type == KVM_IRQ_ROUTING_MSI))
231                                 ret = kvm_set_msi_inatomic(e, kvm);
232                         else
233                                 ret = -EWOULDBLOCK;
234                         break;
235                 }
236         rcu_read_unlock();
237         return ret;
238 }
239
240 void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin)
241 {
242         struct kvm_irq_ack_notifier *kian;
243         struct hlist_node *n;
244         int gsi;
245
246         trace_kvm_ack_irq(irqchip, pin);
247
248         rcu_read_lock();
249         gsi = rcu_dereference(kvm->irq_routing)->chip[irqchip][pin];
250         if (gsi != -1)
251                 hlist_for_each_entry_rcu(kian, n, &kvm->irq_ack_notifier_list,
252                                          link)
253                         if (kian->gsi == gsi)
254                                 kian->irq_acked(kian);
255         rcu_read_unlock();
256 }
257
258 void kvm_register_irq_ack_notifier(struct kvm *kvm,
259                                    struct kvm_irq_ack_notifier *kian)
260 {
261         mutex_lock(&kvm->irq_lock);
262         hlist_add_head_rcu(&kian->link, &kvm->irq_ack_notifier_list);
263         mutex_unlock(&kvm->irq_lock);
264 }
265
266 void kvm_unregister_irq_ack_notifier(struct kvm *kvm,
267                                     struct kvm_irq_ack_notifier *kian)
268 {
269         mutex_lock(&kvm->irq_lock);
270         hlist_del_init_rcu(&kian->link);
271         mutex_unlock(&kvm->irq_lock);
272         synchronize_rcu();
273 }
274
275 int kvm_request_irq_source_id(struct kvm *kvm)
276 {
277         unsigned long *bitmap = &kvm->arch.irq_sources_bitmap;
278         int irq_source_id;
279
280         mutex_lock(&kvm->irq_lock);
281         irq_source_id = find_first_zero_bit(bitmap, BITS_PER_LONG);
282
283         if (irq_source_id >= BITS_PER_LONG) {
284                 printk(KERN_WARNING "kvm: exhaust allocatable IRQ sources!\n");
285                 irq_source_id = -EFAULT;
286                 goto unlock;
287         }
288
289         ASSERT(irq_source_id != KVM_USERSPACE_IRQ_SOURCE_ID);
290 #ifdef CONFIG_X86
291         ASSERT(irq_source_id != KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID);
292 #endif
293         set_bit(irq_source_id, bitmap);
294 unlock:
295         mutex_unlock(&kvm->irq_lock);
296
297         return irq_source_id;
298 }
299
300 void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id)
301 {
302         ASSERT(irq_source_id != KVM_USERSPACE_IRQ_SOURCE_ID);
303 #ifdef CONFIG_X86
304         ASSERT(irq_source_id != KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID);
305 #endif
306
307         mutex_lock(&kvm->irq_lock);
308         if (irq_source_id < 0 ||
309             irq_source_id >= BITS_PER_LONG) {
310                 printk(KERN_ERR "kvm: IRQ source ID out of range!\n");
311                 goto unlock;
312         }
313         clear_bit(irq_source_id, &kvm->arch.irq_sources_bitmap);
314         if (!irqchip_in_kernel(kvm))
315                 goto unlock;
316
317         kvm_ioapic_clear_all(kvm->arch.vioapic, irq_source_id);
318 #ifdef CONFIG_X86
319         kvm_pic_clear_all(pic_irqchip(kvm), irq_source_id);
320 #endif
321 unlock:
322         mutex_unlock(&kvm->irq_lock);
323 }
324
325 void kvm_register_irq_mask_notifier(struct kvm *kvm, int irq,
326                                     struct kvm_irq_mask_notifier *kimn)
327 {
328         mutex_lock(&kvm->irq_lock);
329         kimn->irq = irq;
330         hlist_add_head_rcu(&kimn->link, &kvm->mask_notifier_list);
331         mutex_unlock(&kvm->irq_lock);
332 }
333
334 void kvm_unregister_irq_mask_notifier(struct kvm *kvm, int irq,
335                                       struct kvm_irq_mask_notifier *kimn)
336 {
337         mutex_lock(&kvm->irq_lock);
338         hlist_del_rcu(&kimn->link);
339         mutex_unlock(&kvm->irq_lock);
340         synchronize_rcu();
341 }
342
343 void kvm_fire_mask_notifiers(struct kvm *kvm, unsigned irqchip, unsigned pin,
344                              bool mask)
345 {
346         struct kvm_irq_mask_notifier *kimn;
347         struct hlist_node *n;
348         int gsi;
349
350         rcu_read_lock();
351         gsi = rcu_dereference(kvm->irq_routing)->chip[irqchip][pin];
352         if (gsi != -1)
353                 hlist_for_each_entry_rcu(kimn, n, &kvm->mask_notifier_list, link)
354                         if (kimn->irq == gsi)
355                                 kimn->func(kimn, mask);
356         rcu_read_unlock();
357 }
358
359 void kvm_free_irq_routing(struct kvm *kvm)
360 {
361         /* Called only during vm destruction. Nobody can use the pointer
362            at this stage */
363         kfree(kvm->irq_routing);
364 }
365
366 static int setup_routing_entry(struct kvm_irq_routing_table *rt,
367                                struct kvm_kernel_irq_routing_entry *e,
368                                const struct kvm_irq_routing_entry *ue)
369 {
370         int r = -EINVAL;
371         int delta;
372         unsigned max_pin;
373         struct kvm_kernel_irq_routing_entry *ei;
374         struct hlist_node *n;
375
376         /*
377          * Do not allow GSI to be mapped to the same irqchip more than once.
378          * Allow only one to one mapping between GSI and MSI.
379          */
380         hlist_for_each_entry(ei, n, &rt->map[ue->gsi], link)
381                 if (ei->type == KVM_IRQ_ROUTING_MSI ||
382                     ue->type == KVM_IRQ_ROUTING_MSI ||
383                     ue->u.irqchip.irqchip == ei->irqchip.irqchip)
384                         return r;
385
386         e->gsi = ue->gsi;
387         e->type = ue->type;
388         switch (ue->type) {
389         case KVM_IRQ_ROUTING_IRQCHIP:
390                 delta = 0;
391                 switch (ue->u.irqchip.irqchip) {
392                 case KVM_IRQCHIP_PIC_MASTER:
393                         e->set = kvm_set_pic_irq;
394                         max_pin = PIC_NUM_PINS;
395                         break;
396                 case KVM_IRQCHIP_PIC_SLAVE:
397                         e->set = kvm_set_pic_irq;
398                         max_pin = PIC_NUM_PINS;
399                         delta = 8;
400                         break;
401                 case KVM_IRQCHIP_IOAPIC:
402                         max_pin = KVM_IOAPIC_NUM_PINS;
403                         e->set = kvm_set_ioapic_irq;
404                         break;
405                 default:
406                         goto out;
407                 }
408                 e->irqchip.irqchip = ue->u.irqchip.irqchip;
409                 e->irqchip.pin = ue->u.irqchip.pin + delta;
410                 if (e->irqchip.pin >= max_pin)
411                         goto out;
412                 rt->chip[ue->u.irqchip.irqchip][e->irqchip.pin] = ue->gsi;
413                 break;
414         case KVM_IRQ_ROUTING_MSI:
415                 e->set = kvm_set_msi;
416                 e->msi.address_lo = ue->u.msi.address_lo;
417                 e->msi.address_hi = ue->u.msi.address_hi;
418                 e->msi.data = ue->u.msi.data;
419                 break;
420         default:
421                 goto out;
422         }
423
424         hlist_add_head(&e->link, &rt->map[e->gsi]);
425         r = 0;
426 out:
427         return r;
428 }
429
430
431 int kvm_set_irq_routing(struct kvm *kvm,
432                         const struct kvm_irq_routing_entry *ue,
433                         unsigned nr,
434                         unsigned flags)
435 {
436         struct kvm_irq_routing_table *new, *old;
437         u32 i, j, nr_rt_entries = 0;
438         int r;
439
440         for (i = 0; i < nr; ++i) {
441                 if (ue[i].gsi >= KVM_MAX_IRQ_ROUTES)
442                         return -EINVAL;
443                 nr_rt_entries = max(nr_rt_entries, ue[i].gsi);
444         }
445
446         nr_rt_entries += 1;
447
448         new = kzalloc(sizeof(*new) + (nr_rt_entries * sizeof(struct hlist_head))
449                       + (nr * sizeof(struct kvm_kernel_irq_routing_entry)),
450                       GFP_KERNEL);
451
452         if (!new)
453                 return -ENOMEM;
454
455         new->rt_entries = (void *)&new->map[nr_rt_entries];
456
457         new->nr_rt_entries = nr_rt_entries;
458         for (i = 0; i < 3; i++)
459                 for (j = 0; j < KVM_IOAPIC_NUM_PINS; j++)
460                         new->chip[i][j] = -1;
461
462         for (i = 0; i < nr; ++i) {
463                 r = -EINVAL;
464                 if (ue->flags)
465                         goto out;
466                 r = setup_routing_entry(new, &new->rt_entries[i], ue);
467                 if (r)
468                         goto out;
469                 ++ue;
470         }
471
472         mutex_lock(&kvm->irq_lock);
473         old = kvm->irq_routing;
474         kvm_irq_routing_update(kvm, new);
475         mutex_unlock(&kvm->irq_lock);
476
477         synchronize_rcu();
478
479         new = old;
480         r = 0;
481
482 out:
483         kfree(new);
484         return r;
485 }
486
487 #define IOAPIC_ROUTING_ENTRY(irq) \
488         { .gsi = irq, .type = KVM_IRQ_ROUTING_IRQCHIP,  \
489           .u.irqchip.irqchip = KVM_IRQCHIP_IOAPIC, .u.irqchip.pin = (irq) }
490 #define ROUTING_ENTRY1(irq) IOAPIC_ROUTING_ENTRY(irq)
491
492 #ifdef CONFIG_X86
493 #  define PIC_ROUTING_ENTRY(irq) \
494         { .gsi = irq, .type = KVM_IRQ_ROUTING_IRQCHIP,  \
495           .u.irqchip.irqchip = SELECT_PIC(irq), .u.irqchip.pin = (irq) % 8 }
496 #  define ROUTING_ENTRY2(irq) \
497         IOAPIC_ROUTING_ENTRY(irq), PIC_ROUTING_ENTRY(irq)
498 #else
499 #  define ROUTING_ENTRY2(irq) \
500         IOAPIC_ROUTING_ENTRY(irq)
501 #endif
502
503 static const struct kvm_irq_routing_entry default_routing[] = {
504         ROUTING_ENTRY2(0), ROUTING_ENTRY2(1),
505         ROUTING_ENTRY2(2), ROUTING_ENTRY2(3),
506         ROUTING_ENTRY2(4), ROUTING_ENTRY2(5),
507         ROUTING_ENTRY2(6), ROUTING_ENTRY2(7),
508         ROUTING_ENTRY2(8), ROUTING_ENTRY2(9),
509         ROUTING_ENTRY2(10), ROUTING_ENTRY2(11),
510         ROUTING_ENTRY2(12), ROUTING_ENTRY2(13),
511         ROUTING_ENTRY2(14), ROUTING_ENTRY2(15),
512         ROUTING_ENTRY1(16), ROUTING_ENTRY1(17),
513         ROUTING_ENTRY1(18), ROUTING_ENTRY1(19),
514         ROUTING_ENTRY1(20), ROUTING_ENTRY1(21),
515         ROUTING_ENTRY1(22), ROUTING_ENTRY1(23),
516 #ifdef CONFIG_IA64
517         ROUTING_ENTRY1(24), ROUTING_ENTRY1(25),
518         ROUTING_ENTRY1(26), ROUTING_ENTRY1(27),
519         ROUTING_ENTRY1(28), ROUTING_ENTRY1(29),
520         ROUTING_ENTRY1(30), ROUTING_ENTRY1(31),
521         ROUTING_ENTRY1(32), ROUTING_ENTRY1(33),
522         ROUTING_ENTRY1(34), ROUTING_ENTRY1(35),
523         ROUTING_ENTRY1(36), ROUTING_ENTRY1(37),
524         ROUTING_ENTRY1(38), ROUTING_ENTRY1(39),
525         ROUTING_ENTRY1(40), ROUTING_ENTRY1(41),
526         ROUTING_ENTRY1(42), ROUTING_ENTRY1(43),
527         ROUTING_ENTRY1(44), ROUTING_ENTRY1(45),
528         ROUTING_ENTRY1(46), ROUTING_ENTRY1(47),
529 #endif
530 };
531
532 int kvm_setup_default_irq_routing(struct kvm *kvm)
533 {
534         return kvm_set_irq_routing(kvm, default_routing,
535                                    ARRAY_SIZE(default_routing), 0);
536 }