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 affilates.
  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 inline int kvm_irq_line_state(unsigned long *irq_state,
  37                                      int irq_source_id, int level)
  38 {
  39         /* Logical OR for level trig interrupt */
  40         if (level)
  41                 set_bit(irq_source_id, irq_state);
  42         else
  43                 clear_bit(irq_source_id, irq_state);
  44
  45         return !!(*irq_state);
  46 }
  47
  48 static int kvm_set_pic_irq(struct kvm_kernel_irq_routing_entry *e,
  49                            struct kvm *kvm, int irq_source_id, int level)
  50 {
  51 #ifdef CONFIG_X86
  52         struct kvm_pic *pic = pic_irqchip(kvm);
  53         level = kvm_irq_line_state(&pic->irq_states[e->irqchip.pin],
  54                                    irq_source_id, level);
  55         return kvm_pic_set_irq(pic, e->irqchip.pin, level);
  56 #else
  57         return -1;
  58 #endif
  59 }
  60
  61 static int kvm_set_ioapic_irq(struct kvm_kernel_irq_routing_entry *e,
  62                               struct kvm *kvm, int irq_source_id, int level)
  63 {
  64         struct kvm_ioapic *ioapic = kvm->arch.vioapic;
  65         level = kvm_irq_line_state(&ioapic->irq_states[e->irqchip.pin],
  66                                    irq_source_id, level);
  67
  68         return kvm_ioapic_set_irq(ioapic, e->irqchip.pin, level);
  69 }
  70
  71 inline static bool kvm_is_dm_lowest_prio(struct kvm_lapic_irq *irq)
  72 {
  73 #ifdef CONFIG_IA64
  74         return irq->delivery_mode ==
  75                 (IOSAPIC_LOWEST_PRIORITY << IOSAPIC_DELIVERY_SHIFT);
  76 #else
  77         return irq->delivery_mode == APIC_DM_LOWEST;
  78 #endif
  79 }
  80
  81 int kvm_irq_delivery_to_apic(struct kvm *kvm, struct kvm_lapic *src,
  82                 struct kvm_lapic_irq *irq)
  83 {
  84         int i, r = -1;
  85         struct kvm_vcpu *vcpu, *lowest = NULL;
  86
  87         if (irq->dest_mode == 0 && irq->dest_id == 0xff &&
  88                         kvm_is_dm_lowest_prio(irq))
  89                 printk(KERN_INFO "kvm: apic: phys broadcast and lowest prio\n");
  90
  91         kvm_for_each_vcpu(i, vcpu, kvm) {
  92                 if (!kvm_apic_present(vcpu))
  93                         continue;
  94
  95                 if (!kvm_apic_match_dest(vcpu, src, irq->shorthand,
  96                                         irq->dest_id, irq->dest_mode))
  97                         continue;
  98
  99                 if (!kvm_is_dm_lowest_prio(irq)) {
 100                         if (r < 0)
 101                                 r = 0;
 102                         r += kvm_apic_set_irq(vcpu, irq);
 103                 } else if (kvm_lapic_enabled(vcpu)) {
 104                         if (!lowest)
 105                                 lowest = vcpu;
 106                         else if (kvm_apic_compare_prio(vcpu, lowest) < 0)
 107                                 lowest = vcpu;
 108                 }
 109         }
 110
 111         if (lowest)
 112                 r = kvm_apic_set_irq(lowest, irq);
 113
 114         return r;
 115 }
 116
 117 static int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e,
 118                        struct kvm *kvm, int irq_source_id, int level)
 119 {
 120         struct kvm_lapic_irq irq;
 121
 122         if (!level)
 123                 return -1;
 124
 125         trace_kvm_msi_set_irq(e->msi.address_lo, e->msi.data);
 126
 127         irq.dest_id = (e->msi.address_lo &
 128                         MSI_ADDR_DEST_ID_MASK) >> MSI_ADDR_DEST_ID_SHIFT;
 129         irq.vector = (e->msi.data &
 130                         MSI_DATA_VECTOR_MASK) >> MSI_DATA_VECTOR_SHIFT;
 131         irq.dest_mode = (1 << MSI_ADDR_DEST_MODE_SHIFT) & e->msi.address_lo;
 132         irq.trig_mode = (1 << MSI_DATA_TRIGGER_SHIFT) & e->msi.data;
 133         irq.delivery_mode = e->msi.data & 0x700;
 134         irq.level = 1;
 135         irq.shorthand = 0;
 136
 137         /* TODO Deal with RH bit of MSI message address */
 138         return kvm_irq_delivery_to_apic(kvm, NULL, &irq);
 139 }
 140
 141 /*
 142  * Return value:
 143  *  < 0   Interrupt was ignored (masked or not delivered for other reasons)
 144  *  = 0   Interrupt was coalesced (previous irq is still pending)
 145  *  > 0   Number of CPUs interrupt was delivered to
 146  */
 147 int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level)
 148 {
 149         struct kvm_kernel_irq_routing_entry *e, irq_set[KVM_NR_IRQCHIPS];
 150         int ret = -1, i = 0;
 151         struct kvm_irq_routing_table *irq_rt;
 152         struct hlist_node *n;
 153
 154         trace_kvm_set_irq(irq, level, irq_source_id);
 155
 156         /* Not possible to detect if the guest uses the PIC or the
 157          * IOAPIC.  So set the bit in both. The guest will ignore
 158          * writes to the unused one.
 159          */
 160         rcu_read_lock();
 161         irq_rt = rcu_dereference(kvm->irq_routing);
 162         if (irq < irq_rt->nr_rt_entries)
 163                 hlist_for_each_entry(e, n, &irq_rt->map[irq], link)
 164                         irq_set[i++] = *e;
 165         rcu_read_unlock();
 166
 167         while(i--) {
 168                 int r;
 169                 r = irq_set[i].set(&irq_set[i], kvm, irq_source_id, level);
 170                 if (r < 0)
 171                         continue;
 172
 173                 ret = r + ((ret < 0) ? 0 : ret);
 174         }
 175
 176         return ret;
 177 }
 178
 179 void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin)
 180 {
 181         struct kvm_irq_ack_notifier *kian;
 182         struct hlist_node *n;
 183         int gsi;
 184
 185         trace_kvm_ack_irq(irqchip, pin);
 186
 187         rcu_read_lock();
 188         gsi = rcu_dereference(kvm->irq_routing)->chip[irqchip][pin];
 189         if (gsi != -1)
 190                 hlist_for_each_entry_rcu(kian, n, &kvm->irq_ack_notifier_list,
 191                                          link)
 192                         if (kian->gsi == gsi)
 193                                 kian->irq_acked(kian);
 194         rcu_read_unlock();
 195 }
 196
 197 void kvm_register_irq_ack_notifier(struct kvm *kvm,
 198                                    struct kvm_irq_ack_notifier *kian)
 199 {
 200         mutex_lock(&kvm->irq_lock);
 201         hlist_add_head_rcu(&kian->link, &kvm->irq_ack_notifier_list);
 202         mutex_unlock(&kvm->irq_lock);
 203 }
 204
 205 void kvm_unregister_irq_ack_notifier(struct kvm *kvm,
 206                                     struct kvm_irq_ack_notifier *kian)
 207 {
 208         mutex_lock(&kvm->irq_lock);
 209         hlist_del_init_rcu(&kian->link);
 210         mutex_unlock(&kvm->irq_lock);
 211         synchronize_rcu();
 212 }
 213
 214 int kvm_request_irq_source_id(struct kvm *kvm)
 215 {
 216         unsigned long *bitmap = &kvm->arch.irq_sources_bitmap;
 217         int irq_source_id;
 218
 219         mutex_lock(&kvm->irq_lock);
 220         irq_source_id = find_first_zero_bit(bitmap, BITS_PER_LONG);
 221
 222         if (irq_source_id >= BITS_PER_LONG) {
 223                 printk(KERN_WARNING "kvm: exhaust allocatable IRQ sources!\n");
 224                 irq_source_id = -EFAULT;
 225                 goto unlock;
 226         }
 227
 228         ASSERT(irq_source_id != KVM_USERSPACE_IRQ_SOURCE_ID);
 229         set_bit(irq_source_id, bitmap);
 230 unlock:
 231         mutex_unlock(&kvm->irq_lock);
 232
 233         return irq_source_id;
 234 }
 235
 236 void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id)
 237 {
 238         int i;
 239
 240         ASSERT(irq_source_id != KVM_USERSPACE_IRQ_SOURCE_ID);
 241
 242         mutex_lock(&kvm->irq_lock);
 243         if (irq_source_id < 0 ||
 244             irq_source_id >= BITS_PER_LONG) {
 245                 printk(KERN_ERR "kvm: IRQ source ID out of range!\n");
 246                 goto unlock;
 247         }
 248         clear_bit(irq_source_id, &kvm->arch.irq_sources_bitmap);
 249         if (!irqchip_in_kernel(kvm))
 250                 goto unlock;
 251
 252         for (i = 0; i < KVM_IOAPIC_NUM_PINS; i++) {
 253                 clear_bit(irq_source_id, &kvm->arch.vioapic->irq_states[i]);
 254                 if (i >= 16)
 255                         continue;
 256 #ifdef CONFIG_X86
 257                 clear_bit(irq_source_id, &pic_irqchip(kvm)->irq_states[i]);
 258 #endif
 259         }
 260 unlock:
 261         mutex_unlock(&kvm->irq_lock);
 262 }
 263
 264 void kvm_register_irq_mask_notifier(struct kvm *kvm, int irq,
 265                                     struct kvm_irq_mask_notifier *kimn)
 266 {
 267         mutex_lock(&kvm->irq_lock);
 268         kimn->irq = irq;
 269         hlist_add_head_rcu(&kimn->link, &kvm->mask_notifier_list);
 270         mutex_unlock(&kvm->irq_lock);
 271 }
 272
 273 void kvm_unregister_irq_mask_notifier(struct kvm *kvm, int irq,
 274                                       struct kvm_irq_mask_notifier *kimn)
 275 {
 276         mutex_lock(&kvm->irq_lock);
 277         hlist_del_rcu(&kimn->link);
 278         mutex_unlock(&kvm->irq_lock);
 279         synchronize_rcu();
 280 }
 281
 282 void kvm_fire_mask_notifiers(struct kvm *kvm, unsigned irqchip, unsigned pin,
 283                              bool mask)
 284 {
 285         struct kvm_irq_mask_notifier *kimn;
 286         struct hlist_node *n;
 287         int gsi;
 288
 289         rcu_read_lock();
 290         gsi = rcu_dereference(kvm->irq_routing)->chip[irqchip][pin];
 291         if (gsi != -1)
 292                 hlist_for_each_entry_rcu(kimn, n, &kvm->mask_notifier_list, link)
 293                         if (kimn->irq == gsi)
 294                                 kimn->func(kimn, mask);
 295         rcu_read_unlock();
 296 }
 297
 298 void kvm_free_irq_routing(struct kvm *kvm)
 299 {
 300         /* Called only during vm destruction. Nobody can use the pointer
 301            at this stage */
 302         kfree(kvm->irq_routing);
 303 }
 304
 305 static int setup_routing_entry(struct kvm_irq_routing_table *rt,
 306                                struct kvm_kernel_irq_routing_entry *e,
 307                                const struct kvm_irq_routing_entry *ue)
 308 {
 309         int r = -EINVAL;
 310         int delta;
 311         unsigned max_pin;
 312         struct kvm_kernel_irq_routing_entry *ei;
 313         struct hlist_node *n;
 314
 315         /*
 316          * Do not allow GSI to be mapped to the same irqchip more than once.
 317          * Allow only one to one mapping between GSI and MSI.
 318          */
 319         hlist_for_each_entry(ei, n, &rt->map[ue->gsi], link)
 320                 if (ei->type == KVM_IRQ_ROUTING_MSI ||
 321                     ue->u.irqchip.irqchip == ei->irqchip.irqchip)
 322                         return r;
 323
 324         e->gsi = ue->gsi;
 325         e->type = ue->type;
 326         switch (ue->type) {
 327         case KVM_IRQ_ROUTING_IRQCHIP:
 328                 delta = 0;
 329                 switch (ue->u.irqchip.irqchip) {
 330                 case KVM_IRQCHIP_PIC_MASTER:
 331                         e->set = kvm_set_pic_irq;
 332                         max_pin = 16;
 333                         break;
 334                 case KVM_IRQCHIP_PIC_SLAVE:
 335                         e->set = kvm_set_pic_irq;
 336                         max_pin = 16;
 337                         delta = 8;
 338                         break;
 339                 case KVM_IRQCHIP_IOAPIC:
 340                         max_pin = KVM_IOAPIC_NUM_PINS;
 341                         e->set = kvm_set_ioapic_irq;
 342                         break;
 343                 default:
 344                         goto out;
 345                 }
 346                 e->irqchip.irqchip = ue->u.irqchip.irqchip;
 347                 e->irqchip.pin = ue->u.irqchip.pin + delta;
 348                 if (e->irqchip.pin >= max_pin)
 349                         goto out;
 350                 rt->chip[ue->u.irqchip.irqchip][e->irqchip.pin] = ue->gsi;
 351                 break;
 352         case KVM_IRQ_ROUTING_MSI:
 353                 e->set = kvm_set_msi;
 354                 e->msi.address_lo = ue->u.msi.address_lo;
 355                 e->msi.address_hi = ue->u.msi.address_hi;
 356                 e->msi.data = ue->u.msi.data;
 357                 break;
 358         default:
 359                 goto out;
 360         }
 361
 362         hlist_add_head(&e->link, &rt->map[e->gsi]);
 363         r = 0;
 364 out:
 365         return r;
 366 }
 367
 368
 369 int kvm_set_irq_routing(struct kvm *kvm,
 370                         const struct kvm_irq_routing_entry *ue,
 371                         unsigned nr,
 372                         unsigned flags)
 373 {
 374         struct kvm_irq_routing_table *new, *old;
 375         u32 i, j, nr_rt_entries = 0;
 376         int r;
 377
 378         for (i = 0; i < nr; ++i) {
 379                 if (ue[i].gsi >= KVM_MAX_IRQ_ROUTES)
 380                         return -EINVAL;
 381                 nr_rt_entries = max(nr_rt_entries, ue[i].gsi);
 382         }
 383
 384         nr_rt_entries += 1;
 385
 386         new = kzalloc(sizeof(*new) + (nr_rt_entries * sizeof(struct hlist_head))
 387                       + (nr * sizeof(struct kvm_kernel_irq_routing_entry)),
 388                       GFP_KERNEL);
 389
 390         if (!new)
 391                 return -ENOMEM;
 392
 393         new->rt_entries = (void *)&new->map[nr_rt_entries];
 394
 395         new->nr_rt_entries = nr_rt_entries;
 396         for (i = 0; i < 3; i++)
 397                 for (j = 0; j < KVM_IOAPIC_NUM_PINS; j++)
 398                         new->chip[i][j] = -1;
 399
 400         for (i = 0; i < nr; ++i) {
 401                 r = -EINVAL;
 402                 if (ue->flags)
 403                         goto out;
 404                 r = setup_routing_entry(new, &new->rt_entries[i], ue);
 405                 if (r)
 406                         goto out;
 407                 ++ue;
 408         }
 409
 410         mutex_lock(&kvm->irq_lock);
 411         old = kvm->irq_routing;
 412         rcu_assign_pointer(kvm->irq_routing, new);
 413         mutex_unlock(&kvm->irq_lock);
 414         synchronize_rcu();
 415
 416         new = old;
 417         r = 0;
 418
 419 out:
 420         kfree(new);
 421         return r;
 422 }
 423
 424 #define IOAPIC_ROUTING_ENTRY(irq) \
 425         { .gsi = irq, .type = KVM_IRQ_ROUTING_IRQCHIP,  \
 426           .u.irqchip.irqchip = KVM_IRQCHIP_IOAPIC, .u.irqchip.pin = (irq) }
 427 #define ROUTING_ENTRY1(irq) IOAPIC_ROUTING_ENTRY(irq)
 428
 429 #ifdef CONFIG_X86
 430 #  define PIC_ROUTING_ENTRY(irq) \
 431         { .gsi = irq, .type = KVM_IRQ_ROUTING_IRQCHIP,  \
 432           .u.irqchip.irqchip = SELECT_PIC(irq), .u.irqchip.pin = (irq) % 8 }
 433 #  define ROUTING_ENTRY2(irq) \
 434         IOAPIC_ROUTING_ENTRY(irq), PIC_ROUTING_ENTRY(irq)
 435 #else
 436 #  define ROUTING_ENTRY2(irq) \
 437         IOAPIC_ROUTING_ENTRY(irq)
 438 #endif
 439
 440 static const struct kvm_irq_routing_entry default_routing[] = {
 441         ROUTING_ENTRY2(0), ROUTING_ENTRY2(1),
 442         ROUTING_ENTRY2(2), ROUTING_ENTRY2(3),
 443         ROUTING_ENTRY2(4), ROUTING_ENTRY2(5),
 444         ROUTING_ENTRY2(6), ROUTING_ENTRY2(7),
 445         ROUTING_ENTRY2(8), ROUTING_ENTRY2(9),
 446         ROUTING_ENTRY2(10), ROUTING_ENTRY2(11),
 447         ROUTING_ENTRY2(12), ROUTING_ENTRY2(13),
 448         ROUTING_ENTRY2(14), ROUTING_ENTRY2(15),
 449         ROUTING_ENTRY1(16), ROUTING_ENTRY1(17),
 450         ROUTING_ENTRY1(18), ROUTING_ENTRY1(19),
 451         ROUTING_ENTRY1(20), ROUTING_ENTRY1(21),
 452         ROUTING_ENTRY1(22), ROUTING_ENTRY1(23),
 453 #ifdef CONFIG_IA64
 454         ROUTING_ENTRY1(24), ROUTING_ENTRY1(25),
 455         ROUTING_ENTRY1(26), ROUTING_ENTRY1(27),
 456         ROUTING_ENTRY1(28), ROUTING_ENTRY1(29),
 457         ROUTING_ENTRY1(30), ROUTING_ENTRY1(31),
 458         ROUTING_ENTRY1(32), ROUTING_ENTRY1(33),
 459         ROUTING_ENTRY1(34), ROUTING_ENTRY1(35),
 460         ROUTING_ENTRY1(36), ROUTING_ENTRY1(37),
 461         ROUTING_ENTRY1(38), ROUTING_ENTRY1(39),
 462         ROUTING_ENTRY1(40), ROUTING_ENTRY1(41),
 463         ROUTING_ENTRY1(42), ROUTING_ENTRY1(43),
 464         ROUTING_ENTRY1(44), ROUTING_ENTRY1(45),
 465         ROUTING_ENTRY1(46), ROUTING_ENTRY1(47),
 466 #endif
 467 };
 468
 469 int kvm_setup_default_irq_routing(struct kvm *kvm)
 470 {
 471         return kvm_set_irq_routing(kvm, default_routing,
 472                                    ARRAY_SIZE(default_routing), 0);
 473 }