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