KVM: Add resampling irqfds for level triggered interrupts
[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 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e,
106                 struct kvm *kvm, int irq_source_id, int level)
107 {
108         struct kvm_lapic_irq irq;
109
110         if (!level)
111                 return -1;
112
113         trace_kvm_msi_set_irq(e->msi.address_lo, e->msi.data);
114
115         irq.dest_id = (e->msi.address_lo &
116                         MSI_ADDR_DEST_ID_MASK) >> MSI_ADDR_DEST_ID_SHIFT;
117         irq.vector = (e->msi.data &
118                         MSI_DATA_VECTOR_MASK) >> MSI_DATA_VECTOR_SHIFT;
119         irq.dest_mode = (1 << MSI_ADDR_DEST_MODE_SHIFT) & e->msi.address_lo;
120         irq.trig_mode = (1 << MSI_DATA_TRIGGER_SHIFT) & e->msi.data;
121         irq.delivery_mode = e->msi.data & 0x700;
122         irq.level = 1;
123         irq.shorthand = 0;
124
125         /* TODO Deal with RH bit of MSI message address */
126         return kvm_irq_delivery_to_apic(kvm, NULL, &irq);
127 }
128
129 int kvm_send_userspace_msi(struct kvm *kvm, struct kvm_msi *msi)
130 {
131         struct kvm_kernel_irq_routing_entry route;
132
133         if (!irqchip_in_kernel(kvm) || msi->flags != 0)
134                 return -EINVAL;
135
136         route.msi.address_lo = msi->address_lo;
137         route.msi.address_hi = msi->address_hi;
138         route.msi.data = msi->data;
139
140         return kvm_set_msi(&route, kvm, KVM_USERSPACE_IRQ_SOURCE_ID, 1);
141 }
142
143 /*
144  * Return value:
145  *  < 0   Interrupt was ignored (masked or not delivered for other reasons)
146  *  = 0   Interrupt was coalesced (previous irq is still pending)
147  *  > 0   Number of CPUs interrupt was delivered to
148  */
149 int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level)
150 {
151         struct kvm_kernel_irq_routing_entry *e, irq_set[KVM_NR_IRQCHIPS];
152         int ret = -1, i = 0;
153         struct kvm_irq_routing_table *irq_rt;
154         struct hlist_node *n;
155
156         trace_kvm_set_irq(irq, level, irq_source_id);
157
158         /* Not possible to detect if the guest uses the PIC or the
159          * IOAPIC.  So set the bit in both. The guest will ignore
160          * writes to the unused one.
161          */
162         rcu_read_lock();
163         irq_rt = rcu_dereference(kvm->irq_routing);
164         if (irq < irq_rt->nr_rt_entries)
165                 hlist_for_each_entry(e, n, &irq_rt->map[irq], link)
166                         irq_set[i++] = *e;
167         rcu_read_unlock();
168
169         while(i--) {
170                 int r;
171                 r = irq_set[i].set(&irq_set[i], kvm, irq_source_id, level);
172                 if (r < 0)
173                         continue;
174
175                 ret = r + ((ret < 0) ? 0 : ret);
176         }
177
178         return ret;
179 }
180
181 void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin)
182 {
183         struct kvm_irq_ack_notifier *kian;
184         struct hlist_node *n;
185         int gsi;
186
187         trace_kvm_ack_irq(irqchip, pin);
188
189         rcu_read_lock();
190         gsi = rcu_dereference(kvm->irq_routing)->chip[irqchip][pin];
191         if (gsi != -1)
192                 hlist_for_each_entry_rcu(kian, n, &kvm->irq_ack_notifier_list,
193                                          link)
194                         if (kian->gsi == gsi)
195                                 kian->irq_acked(kian);
196         rcu_read_unlock();
197 }
198
199 void kvm_register_irq_ack_notifier(struct kvm *kvm,
200                                    struct kvm_irq_ack_notifier *kian)
201 {
202         mutex_lock(&kvm->irq_lock);
203         hlist_add_head_rcu(&kian->link, &kvm->irq_ack_notifier_list);
204         mutex_unlock(&kvm->irq_lock);
205 }
206
207 void kvm_unregister_irq_ack_notifier(struct kvm *kvm,
208                                     struct kvm_irq_ack_notifier *kian)
209 {
210         mutex_lock(&kvm->irq_lock);
211         hlist_del_init_rcu(&kian->link);
212         mutex_unlock(&kvm->irq_lock);
213         synchronize_rcu();
214 }
215
216 int kvm_request_irq_source_id(struct kvm *kvm)
217 {
218         unsigned long *bitmap = &kvm->arch.irq_sources_bitmap;
219         int irq_source_id;
220
221         mutex_lock(&kvm->irq_lock);
222         irq_source_id = find_first_zero_bit(bitmap, BITS_PER_LONG);
223
224         if (irq_source_id >= BITS_PER_LONG) {
225                 printk(KERN_WARNING "kvm: exhaust allocatable IRQ sources!\n");
226                 irq_source_id = -EFAULT;
227                 goto unlock;
228         }
229
230         ASSERT(irq_source_id != KVM_USERSPACE_IRQ_SOURCE_ID);
231 #ifdef CONFIG_X86
232         ASSERT(irq_source_id != KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID);
233 #endif
234         set_bit(irq_source_id, bitmap);
235 unlock:
236         mutex_unlock(&kvm->irq_lock);
237
238         return irq_source_id;
239 }
240
241 void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id)
242 {
243         ASSERT(irq_source_id != KVM_USERSPACE_IRQ_SOURCE_ID);
244 #ifdef CONFIG_X86
245         ASSERT(irq_source_id != KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID);
246 #endif
247
248         mutex_lock(&kvm->irq_lock);
249         if (irq_source_id < 0 ||
250             irq_source_id >= BITS_PER_LONG) {
251                 printk(KERN_ERR "kvm: IRQ source ID out of range!\n");
252                 goto unlock;
253         }
254         clear_bit(irq_source_id, &kvm->arch.irq_sources_bitmap);
255         if (!irqchip_in_kernel(kvm))
256                 goto unlock;
257
258         kvm_ioapic_clear_all(kvm->arch.vioapic, irq_source_id);
259 #ifdef CONFIG_X86
260         kvm_pic_clear_all(pic_irqchip(kvm), irq_source_id);
261 #endif
262 unlock:
263         mutex_unlock(&kvm->irq_lock);
264 }
265
266 void kvm_register_irq_mask_notifier(struct kvm *kvm, int irq,
267                                     struct kvm_irq_mask_notifier *kimn)
268 {
269         mutex_lock(&kvm->irq_lock);
270         kimn->irq = irq;
271         hlist_add_head_rcu(&kimn->link, &kvm->mask_notifier_list);
272         mutex_unlock(&kvm->irq_lock);
273 }
274
275 void kvm_unregister_irq_mask_notifier(struct kvm *kvm, int irq,
276                                       struct kvm_irq_mask_notifier *kimn)
277 {
278         mutex_lock(&kvm->irq_lock);
279         hlist_del_rcu(&kimn->link);
280         mutex_unlock(&kvm->irq_lock);
281         synchronize_rcu();
282 }
283
284 void kvm_fire_mask_notifiers(struct kvm *kvm, unsigned irqchip, unsigned pin,
285                              bool mask)
286 {
287         struct kvm_irq_mask_notifier *kimn;
288         struct hlist_node *n;
289         int gsi;
290
291         rcu_read_lock();
292         gsi = rcu_dereference(kvm->irq_routing)->chip[irqchip][pin];
293         if (gsi != -1)
294                 hlist_for_each_entry_rcu(kimn, n, &kvm->mask_notifier_list, link)
295                         if (kimn->irq == gsi)
296                                 kimn->func(kimn, mask);
297         rcu_read_unlock();
298 }
299
300 void kvm_free_irq_routing(struct kvm *kvm)
301 {
302         /* Called only during vm destruction. Nobody can use the pointer
303            at this stage */
304         kfree(kvm->irq_routing);
305 }
306
307 static int setup_routing_entry(struct kvm_irq_routing_table *rt,
308                                struct kvm_kernel_irq_routing_entry *e,
309                                const struct kvm_irq_routing_entry *ue)
310 {
311         int r = -EINVAL;
312         int delta;
313         unsigned max_pin;
314         struct kvm_kernel_irq_routing_entry *ei;
315         struct hlist_node *n;
316
317         /*
318          * Do not allow GSI to be mapped to the same irqchip more than once.
319          * Allow only one to one mapping between GSI and MSI.
320          */
321         hlist_for_each_entry(ei, n, &rt->map[ue->gsi], link)
322                 if (ei->type == KVM_IRQ_ROUTING_MSI ||
323                     ue->type == KVM_IRQ_ROUTING_MSI ||
324                     ue->u.irqchip.irqchip == ei->irqchip.irqchip)
325                         return r;
326
327         e->gsi = ue->gsi;
328         e->type = ue->type;
329         switch (ue->type) {
330         case KVM_IRQ_ROUTING_IRQCHIP:
331                 delta = 0;
332                 switch (ue->u.irqchip.irqchip) {
333                 case KVM_IRQCHIP_PIC_MASTER:
334                         e->set = kvm_set_pic_irq;
335                         max_pin = PIC_NUM_PINS;
336                         break;
337                 case KVM_IRQCHIP_PIC_SLAVE:
338                         e->set = kvm_set_pic_irq;
339                         max_pin = PIC_NUM_PINS;
340                         delta = 8;
341                         break;
342                 case KVM_IRQCHIP_IOAPIC:
343                         max_pin = KVM_IOAPIC_NUM_PINS;
344                         e->set = kvm_set_ioapic_irq;
345                         break;
346                 default:
347                         goto out;
348                 }
349                 e->irqchip.irqchip = ue->u.irqchip.irqchip;
350                 e->irqchip.pin = ue->u.irqchip.pin + delta;
351                 if (e->irqchip.pin >= max_pin)
352                         goto out;
353                 rt->chip[ue->u.irqchip.irqchip][e->irqchip.pin] = ue->gsi;
354                 break;
355         case KVM_IRQ_ROUTING_MSI:
356                 e->set = kvm_set_msi;
357                 e->msi.address_lo = ue->u.msi.address_lo;
358                 e->msi.address_hi = ue->u.msi.address_hi;
359                 e->msi.data = ue->u.msi.data;
360                 break;
361         default:
362                 goto out;
363         }
364
365         hlist_add_head(&e->link, &rt->map[e->gsi]);
366         r = 0;
367 out:
368         return r;
369 }
370
371
372 int kvm_set_irq_routing(struct kvm *kvm,
373                         const struct kvm_irq_routing_entry *ue,
374                         unsigned nr,
375                         unsigned flags)
376 {
377         struct kvm_irq_routing_table *new, *old;
378         u32 i, j, nr_rt_entries = 0;
379         int r;
380
381         for (i = 0; i < nr; ++i) {
382                 if (ue[i].gsi >= KVM_MAX_IRQ_ROUTES)
383                         return -EINVAL;
384                 nr_rt_entries = max(nr_rt_entries, ue[i].gsi);
385         }
386
387         nr_rt_entries += 1;
388
389         new = kzalloc(sizeof(*new) + (nr_rt_entries * sizeof(struct hlist_head))
390                       + (nr * sizeof(struct kvm_kernel_irq_routing_entry)),
391                       GFP_KERNEL);
392
393         if (!new)
394                 return -ENOMEM;
395
396         new->rt_entries = (void *)&new->map[nr_rt_entries];
397
398         new->nr_rt_entries = nr_rt_entries;
399         for (i = 0; i < 3; i++)
400                 for (j = 0; j < KVM_IOAPIC_NUM_PINS; j++)
401                         new->chip[i][j] = -1;
402
403         for (i = 0; i < nr; ++i) {
404                 r = -EINVAL;
405                 if (ue->flags)
406                         goto out;
407                 r = setup_routing_entry(new, &new->rt_entries[i], ue);
408                 if (r)
409                         goto out;
410                 ++ue;
411         }
412
413         mutex_lock(&kvm->irq_lock);
414         old = kvm->irq_routing;
415         kvm_irq_routing_update(kvm, new);
416         mutex_unlock(&kvm->irq_lock);
417
418         synchronize_rcu();
419
420         new = old;
421         r = 0;
422
423 out:
424         kfree(new);
425         return r;
426 }
427
428 #define IOAPIC_ROUTING_ENTRY(irq) \
429         { .gsi = irq, .type = KVM_IRQ_ROUTING_IRQCHIP,  \
430           .u.irqchip.irqchip = KVM_IRQCHIP_IOAPIC, .u.irqchip.pin = (irq) }
431 #define ROUTING_ENTRY1(irq) IOAPIC_ROUTING_ENTRY(irq)
432
433 #ifdef CONFIG_X86
434 #  define PIC_ROUTING_ENTRY(irq) \
435         { .gsi = irq, .type = KVM_IRQ_ROUTING_IRQCHIP,  \
436           .u.irqchip.irqchip = SELECT_PIC(irq), .u.irqchip.pin = (irq) % 8 }
437 #  define ROUTING_ENTRY2(irq) \
438         IOAPIC_ROUTING_ENTRY(irq), PIC_ROUTING_ENTRY(irq)
439 #else
440 #  define ROUTING_ENTRY2(irq) \
441         IOAPIC_ROUTING_ENTRY(irq)
442 #endif
443
444 static const struct kvm_irq_routing_entry default_routing[] = {
445         ROUTING_ENTRY2(0), ROUTING_ENTRY2(1),
446         ROUTING_ENTRY2(2), ROUTING_ENTRY2(3),
447         ROUTING_ENTRY2(4), ROUTING_ENTRY2(5),
448         ROUTING_ENTRY2(6), ROUTING_ENTRY2(7),
449         ROUTING_ENTRY2(8), ROUTING_ENTRY2(9),
450         ROUTING_ENTRY2(10), ROUTING_ENTRY2(11),
451         ROUTING_ENTRY2(12), ROUTING_ENTRY2(13),
452         ROUTING_ENTRY2(14), ROUTING_ENTRY2(15),
453         ROUTING_ENTRY1(16), ROUTING_ENTRY1(17),
454         ROUTING_ENTRY1(18), ROUTING_ENTRY1(19),
455         ROUTING_ENTRY1(20), ROUTING_ENTRY1(21),
456         ROUTING_ENTRY1(22), ROUTING_ENTRY1(23),
457 #ifdef CONFIG_IA64
458         ROUTING_ENTRY1(24), ROUTING_ENTRY1(25),
459         ROUTING_ENTRY1(26), ROUTING_ENTRY1(27),
460         ROUTING_ENTRY1(28), ROUTING_ENTRY1(29),
461         ROUTING_ENTRY1(30), ROUTING_ENTRY1(31),
462         ROUTING_ENTRY1(32), ROUTING_ENTRY1(33),
463         ROUTING_ENTRY1(34), ROUTING_ENTRY1(35),
464         ROUTING_ENTRY1(36), ROUTING_ENTRY1(37),
465         ROUTING_ENTRY1(38), ROUTING_ENTRY1(39),
466         ROUTING_ENTRY1(40), ROUTING_ENTRY1(41),
467         ROUTING_ENTRY1(42), ROUTING_ENTRY1(43),
468         ROUTING_ENTRY1(44), ROUTING_ENTRY1(45),
469         ROUTING_ENTRY1(46), ROUTING_ENTRY1(47),
470 #endif
471 };
472
473 int kvm_setup_default_irq_routing(struct kvm *kvm)
474 {
475         return kvm_set_irq_routing(kvm, default_routing,
476                                    ARRAY_SIZE(default_routing), 0);
477 }