arch/arm64/kvm/vgic/vgic-mmio.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * VGIC MMIO handling functions
   4  */
   5
   6 #include <linux/bitops.h>
   7 #include <linux/bsearch.h>
   8 #include <linux/interrupt.h>
   9 #include <linux/irq.h>
  10 #include <linux/kvm.h>
  11 #include <linux/kvm_host.h>
  12 #include <kvm/iodev.h>
  13 #include <kvm/arm_arch_timer.h>
  14 #include <kvm/arm_vgic.h>
  15
  16 #include "vgic.h"
  17 #include "vgic-mmio.h"
  18
  19 unsigned long vgic_mmio_read_raz(struct kvm_vcpu *vcpu,
  20                                  gpa_t addr, unsigned int len)
  21 {
  22         return 0;
  23 }
  24
  25 unsigned long vgic_mmio_read_rao(struct kvm_vcpu *vcpu,
  26                                  gpa_t addr, unsigned int len)
  27 {
  28         return -1UL;
  29 }
  30
  31 void vgic_mmio_write_wi(struct kvm_vcpu *vcpu, gpa_t addr,
  32                         unsigned int len, unsigned long val)
  33 {
  34         /* Ignore */
  35 }
  36
  37 int vgic_mmio_uaccess_write_wi(struct kvm_vcpu *vcpu, gpa_t addr,
  38                                unsigned int len, unsigned long val)
  39 {
  40         /* Ignore */
  41         return 0;
  42 }
  43
  44 unsigned long vgic_mmio_read_group(struct kvm_vcpu *vcpu,
  45                                    gpa_t addr, unsigned int len)
  46 {
  47         u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
  48         u32 value = 0;
  49         int i;
  50
  51         /* Loop over all IRQs affected by this read */
  52         for (i = 0; i < len * 8; i++) {
  53                 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
  54
  55                 if (irq->group)
  56                         value |= BIT(i);
  57
  58                 vgic_put_irq(vcpu->kvm, irq);
  59         }
  60
  61         return value;
  62 }
  63
  64 static void vgic_update_vsgi(struct vgic_irq *irq)
  65 {
  66         WARN_ON(its_prop_update_vsgi(irq->host_irq, irq->priority, irq->group));
  67 }
  68
  69 void vgic_mmio_write_group(struct kvm_vcpu *vcpu, gpa_t addr,
  70                            unsigned int len, unsigned long val)
  71 {
  72         u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
  73         int i;
  74         unsigned long flags;
  75
  76         for (i = 0; i < len * 8; i++) {
  77                 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
  78
  79                 raw_spin_lock_irqsave(&irq->irq_lock, flags);
  80                 irq->group = !!(val & BIT(i));
  81                 if (irq->hw && vgic_irq_is_sgi(irq->intid)) {
  82                         vgic_update_vsgi(irq);
  83                         raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
  84                 } else {
  85                         vgic_queue_irq_unlock(vcpu->kvm, irq, flags);
  86                 }
  87
  88                 vgic_put_irq(vcpu->kvm, irq);
  89         }
  90 }
  91
  92 /*
  93  * Read accesses to both GICD_ICENABLER and GICD_ISENABLER return the value
  94  * of the enabled bit, so there is only one function for both here.
  95  */
  96 unsigned long vgic_mmio_read_enable(struct kvm_vcpu *vcpu,
  97                                     gpa_t addr, unsigned int len)
  98 {
  99         u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
 100         u32 value = 0;
 101         int i;
 102
 103         /* Loop over all IRQs affected by this read */
 104         for (i = 0; i < len * 8; i++) {
 105                 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 106
 107                 if (irq->enabled)
 108                         value |= (1U << i);
 109
 110                 vgic_put_irq(vcpu->kvm, irq);
 111         }
 112
 113         return value;
 114 }
 115
 116 void vgic_mmio_write_senable(struct kvm_vcpu *vcpu,
 117                              gpa_t addr, unsigned int len,
 118                              unsigned long val)
 119 {
 120         u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
 121         int i;
 122         unsigned long flags;
 123
 124         for_each_set_bit(i, &val, len * 8) {
 125                 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 126
 127                 raw_spin_lock_irqsave(&irq->irq_lock, flags);
 128                 if (irq->hw && vgic_irq_is_sgi(irq->intid)) {
 129                         if (!irq->enabled) {
 130                                 struct irq_data *data;
 131
 132                                 irq->enabled = true;
 133                                 data = &irq_to_desc(irq->host_irq)->irq_data;
 134                                 while (irqd_irq_disabled(data))
 135                                         enable_irq(irq->host_irq);
 136                         }
 137
 138                         raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
 139                         vgic_put_irq(vcpu->kvm, irq);
 140
 141                         continue;
 142                 } else if (vgic_irq_is_mapped_level(irq)) {
 143                         bool was_high = irq->line_level;
 144
 145                         /*
 146                          * We need to update the state of the interrupt because
 147                          * the guest might have changed the state of the device
 148                          * while the interrupt was disabled at the VGIC level.
 149                          */
 150                         irq->line_level = vgic_get_phys_line_level(irq);
 151                         /*
 152                          * Deactivate the physical interrupt so the GIC will let
 153                          * us know when it is asserted again.
 154                          */
 155                         if (!irq->active && was_high && !irq->line_level)
 156                                 vgic_irq_set_phys_active(irq, false);
 157                 }
 158                 irq->enabled = true;
 159                 vgic_queue_irq_unlock(vcpu->kvm, irq, flags);
 160
 161                 vgic_put_irq(vcpu->kvm, irq);
 162         }
 163 }
 164
 165 void vgic_mmio_write_cenable(struct kvm_vcpu *vcpu,
 166                              gpa_t addr, unsigned int len,
 167                              unsigned long val)
 168 {
 169         u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
 170         int i;
 171         unsigned long flags;
 172
 173         for_each_set_bit(i, &val, len * 8) {
 174                 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 175
 176                 raw_spin_lock_irqsave(&irq->irq_lock, flags);
 177                 if (irq->hw && vgic_irq_is_sgi(irq->intid) && irq->enabled)
 178                         disable_irq_nosync(irq->host_irq);
 179
 180                 irq->enabled = false;
 181
 182                 raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
 183                 vgic_put_irq(vcpu->kvm, irq);
 184         }
 185 }
 186
 187 int vgic_uaccess_write_senable(struct kvm_vcpu *vcpu,
 188                                gpa_t addr, unsigned int len,
 189                                unsigned long val)
 190 {
 191         u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
 192         int i;
 193         unsigned long flags;
 194
 195         for_each_set_bit(i, &val, len * 8) {
 196                 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 197
 198                 raw_spin_lock_irqsave(&irq->irq_lock, flags);
 199                 irq->enabled = true;
 200                 vgic_queue_irq_unlock(vcpu->kvm, irq, flags);
 201
 202                 vgic_put_irq(vcpu->kvm, irq);
 203         }
 204
 205         return 0;
 206 }
 207
 208 int vgic_uaccess_write_cenable(struct kvm_vcpu *vcpu,
 209                                gpa_t addr, unsigned int len,
 210                                unsigned long val)
 211 {
 212         u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
 213         int i;
 214         unsigned long flags;
 215
 216         for_each_set_bit(i, &val, len * 8) {
 217                 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 218
 219                 raw_spin_lock_irqsave(&irq->irq_lock, flags);
 220                 irq->enabled = false;
 221                 raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
 222
 223                 vgic_put_irq(vcpu->kvm, irq);
 224         }
 225
 226         return 0;
 227 }
 228
 229 unsigned long vgic_mmio_read_pending(struct kvm_vcpu *vcpu,
 230                                      gpa_t addr, unsigned int len)
 231 {
 232         u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
 233         u32 value = 0;
 234         int i;
 235
 236         /* Loop over all IRQs affected by this read */
 237         for (i = 0; i < len * 8; i++) {
 238                 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 239                 unsigned long flags;
 240                 bool val;
 241
 242                 raw_spin_lock_irqsave(&irq->irq_lock, flags);
 243                 if (irq->hw && vgic_irq_is_sgi(irq->intid)) {
 244                         int err;
 245
 246                         val = false;
 247                         err = irq_get_irqchip_state(irq->host_irq,
 248                                                     IRQCHIP_STATE_PENDING,
 249                                                     &val);
 250                         WARN_RATELIMIT(err, "IRQ %d", irq->host_irq);
 251                 } else {
 252                         val = irq_is_pending(irq);
 253                 }
 254
 255                 value |= ((u32)val << i);
 256                 raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
 257
 258                 vgic_put_irq(vcpu->kvm, irq);
 259         }
 260
 261         return value;
 262 }
 263
 264 static bool is_vgic_v2_sgi(struct kvm_vcpu *vcpu, struct vgic_irq *irq)
 265 {
 266         return (vgic_irq_is_sgi(irq->intid) &&
 267                 vcpu->kvm->arch.vgic.vgic_model == KVM_DEV_TYPE_ARM_VGIC_V2);
 268 }
 269
 270 void vgic_mmio_write_spending(struct kvm_vcpu *vcpu,
 271                               gpa_t addr, unsigned int len,
 272                               unsigned long val)
 273 {
 274         u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
 275         int i;
 276         unsigned long flags;
 277
 278         for_each_set_bit(i, &val, len * 8) {
 279                 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 280
 281                 /* GICD_ISPENDR0 SGI bits are WI */
 282                 if (is_vgic_v2_sgi(vcpu, irq)) {
 283                         vgic_put_irq(vcpu->kvm, irq);
 284                         continue;
 285                 }
 286
 287                 raw_spin_lock_irqsave(&irq->irq_lock, flags);
 288
 289                 if (irq->hw && vgic_irq_is_sgi(irq->intid)) {
 290                         /* HW SGI? Ask the GIC to inject it */
 291                         int err;
 292                         err = irq_set_irqchip_state(irq->host_irq,
 293                                                     IRQCHIP_STATE_PENDING,
 294                                                     true);
 295                         WARN_RATELIMIT(err, "IRQ %d", irq->host_irq);
 296
 297                         raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
 298                         vgic_put_irq(vcpu->kvm, irq);
 299
 300                         continue;
 301                 }
 302
 303                 irq->pending_latch = true;
 304                 if (irq->hw)
 305                         vgic_irq_set_phys_active(irq, true);
 306
 307                 vgic_queue_irq_unlock(vcpu->kvm, irq, flags);
 308                 vgic_put_irq(vcpu->kvm, irq);
 309         }
 310 }
 311
 312 int vgic_uaccess_write_spending(struct kvm_vcpu *vcpu,
 313                                 gpa_t addr, unsigned int len,
 314                                 unsigned long val)
 315 {
 316         u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
 317         int i;
 318         unsigned long flags;
 319
 320         for_each_set_bit(i, &val, len * 8) {
 321                 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 322
 323                 raw_spin_lock_irqsave(&irq->irq_lock, flags);
 324                 irq->pending_latch = true;
 325
 326                 /*
 327                  * GICv2 SGIs are terribly broken. We can't restore
 328                  * the source of the interrupt, so just pick the vcpu
 329                  * itself as the source...
 330                  */
 331                 if (is_vgic_v2_sgi(vcpu, irq))
 332                         irq->source |= BIT(vcpu->vcpu_id);
 333
 334                 vgic_queue_irq_unlock(vcpu->kvm, irq, flags);
 335
 336                 vgic_put_irq(vcpu->kvm, irq);
 337         }
 338
 339         return 0;
 340 }
 341
 342 /* Must be called with irq->irq_lock held */
 343 static void vgic_hw_irq_cpending(struct kvm_vcpu *vcpu, struct vgic_irq *irq)
 344 {
 345         irq->pending_latch = false;
 346
 347         /*
 348          * We don't want the guest to effectively mask the physical
 349          * interrupt by doing a write to SPENDR followed by a write to
 350          * CPENDR for HW interrupts, so we clear the active state on
 351          * the physical side if the virtual interrupt is not active.
 352          * This may lead to taking an additional interrupt on the
 353          * host, but that should not be a problem as the worst that
 354          * can happen is an additional vgic injection.  We also clear
 355          * the pending state to maintain proper semantics for edge HW
 356          * interrupts.
 357          */
 358         vgic_irq_set_phys_pending(irq, false);
 359         if (!irq->active)
 360                 vgic_irq_set_phys_active(irq, false);
 361 }
 362
 363 void vgic_mmio_write_cpending(struct kvm_vcpu *vcpu,
 364                               gpa_t addr, unsigned int len,
 365                               unsigned long val)
 366 {
 367         u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
 368         int i;
 369         unsigned long flags;
 370
 371         for_each_set_bit(i, &val, len * 8) {
 372                 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 373
 374                 /* GICD_ICPENDR0 SGI bits are WI */
 375                 if (is_vgic_v2_sgi(vcpu, irq)) {
 376                         vgic_put_irq(vcpu->kvm, irq);
 377                         continue;
 378                 }
 379
 380                 raw_spin_lock_irqsave(&irq->irq_lock, flags);
 381
 382                 if (irq->hw && vgic_irq_is_sgi(irq->intid)) {
 383                         /* HW SGI? Ask the GIC to clear its pending bit */
 384                         int err;
 385                         err = irq_set_irqchip_state(irq->host_irq,
 386                                                     IRQCHIP_STATE_PENDING,
 387                                                     false);
 388                         WARN_RATELIMIT(err, "IRQ %d", irq->host_irq);
 389
 390                         raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
 391                         vgic_put_irq(vcpu->kvm, irq);
 392
 393                         continue;
 394                 }
 395
 396                 if (irq->hw)
 397                         vgic_hw_irq_cpending(vcpu, irq);
 398                 else
 399                         irq->pending_latch = false;
 400
 401                 raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
 402                 vgic_put_irq(vcpu->kvm, irq);
 403         }
 404 }
 405
 406 int vgic_uaccess_write_cpending(struct kvm_vcpu *vcpu,
 407                                 gpa_t addr, unsigned int len,
 408                                 unsigned long val)
 409 {
 410         u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
 411         int i;
 412         unsigned long flags;
 413
 414         for_each_set_bit(i, &val, len * 8) {
 415                 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 416
 417                 raw_spin_lock_irqsave(&irq->irq_lock, flags);
 418                 /*
 419                  * More fun with GICv2 SGIs! If we're clearing one of them
 420                  * from userspace, which source vcpu to clear? Let's not
 421                  * even think of it, and blow the whole set.
 422                  */
 423                 if (is_vgic_v2_sgi(vcpu, irq))
 424                         irq->source = 0;
 425
 426                 irq->pending_latch = false;
 427
 428                 raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
 429
 430                 vgic_put_irq(vcpu->kvm, irq);
 431         }
 432
 433         return 0;
 434 }
 435
 436 /*
 437  * If we are fiddling with an IRQ's active state, we have to make sure the IRQ
 438  * is not queued on some running VCPU's LRs, because then the change to the
 439  * active state can be overwritten when the VCPU's state is synced coming back
 440  * from the guest.
 441  *
 442  * For shared interrupts as well as GICv3 private interrupts, we have to
 443  * stop all the VCPUs because interrupts can be migrated while we don't hold
 444  * the IRQ locks and we don't want to be chasing moving targets.
 445  *
 446  * For GICv2 private interrupts we don't have to do anything because
 447  * userspace accesses to the VGIC state already require all VCPUs to be
 448  * stopped, and only the VCPU itself can modify its private interrupts
 449  * active state, which guarantees that the VCPU is not running.
 450  */
 451 static void vgic_access_active_prepare(struct kvm_vcpu *vcpu, u32 intid)
 452 {
 453         if (vcpu->kvm->arch.vgic.vgic_model == KVM_DEV_TYPE_ARM_VGIC_V3 ||
 454             intid >= VGIC_NR_PRIVATE_IRQS)
 455                 kvm_arm_halt_guest(vcpu->kvm);
 456 }
 457
 458 /* See vgic_access_active_prepare */
 459 static void vgic_access_active_finish(struct kvm_vcpu *vcpu, u32 intid)
 460 {
 461         if (vcpu->kvm->arch.vgic.vgic_model == KVM_DEV_TYPE_ARM_VGIC_V3 ||
 462             intid >= VGIC_NR_PRIVATE_IRQS)
 463                 kvm_arm_resume_guest(vcpu->kvm);
 464 }
 465
 466 static unsigned long __vgic_mmio_read_active(struct kvm_vcpu *vcpu,
 467                                              gpa_t addr, unsigned int len)
 468 {
 469         u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
 470         u32 value = 0;
 471         int i;
 472
 473         /* Loop over all IRQs affected by this read */
 474         for (i = 0; i < len * 8; i++) {
 475                 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 476
 477                 /*
 478                  * Even for HW interrupts, don't evaluate the HW state as
 479                  * all the guest is interested in is the virtual state.
 480                  */
 481                 if (irq->active)
 482                         value |= (1U << i);
 483
 484                 vgic_put_irq(vcpu->kvm, irq);
 485         }
 486
 487         return value;
 488 }
 489
 490 unsigned long vgic_mmio_read_active(struct kvm_vcpu *vcpu,
 491                                     gpa_t addr, unsigned int len)
 492 {
 493         u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
 494         u32 val;
 495
 496         mutex_lock(&vcpu->kvm->lock);
 497         vgic_access_active_prepare(vcpu, intid);
 498
 499         val = __vgic_mmio_read_active(vcpu, addr, len);
 500
 501         vgic_access_active_finish(vcpu, intid);
 502         mutex_unlock(&vcpu->kvm->lock);
 503
 504         return val;
 505 }
 506
 507 unsigned long vgic_uaccess_read_active(struct kvm_vcpu *vcpu,
 508                                     gpa_t addr, unsigned int len)
 509 {
 510         return __vgic_mmio_read_active(vcpu, addr, len);
 511 }
 512
 513 /* Must be called with irq->irq_lock held */
 514 static void vgic_hw_irq_change_active(struct kvm_vcpu *vcpu, struct vgic_irq *irq,
 515                                       bool active, bool is_uaccess)
 516 {
 517         if (is_uaccess)
 518                 return;
 519
 520         irq->active = active;
 521         vgic_irq_set_phys_active(irq, active);
 522 }
 523
 524 static void vgic_mmio_change_active(struct kvm_vcpu *vcpu, struct vgic_irq *irq,
 525                                     bool active)
 526 {
 527         unsigned long flags;
 528         struct kvm_vcpu *requester_vcpu = kvm_get_running_vcpu();
 529
 530         raw_spin_lock_irqsave(&irq->irq_lock, flags);
 531
 532         if (irq->hw && !vgic_irq_is_sgi(irq->intid)) {
 533                 vgic_hw_irq_change_active(vcpu, irq, active, !requester_vcpu);
 534         } else if (irq->hw && vgic_irq_is_sgi(irq->intid)) {
 535                 /*
 536                  * GICv4.1 VSGI feature doesn't track an active state,
 537                  * so let's not kid ourselves, there is nothing we can
 538                  * do here.
 539                  */
 540                 irq->active = false;
 541         } else {
 542                 u32 model = vcpu->kvm->arch.vgic.vgic_model;
 543                 u8 active_source;
 544
 545                 irq->active = active;
 546
 547                 /*
 548                  * The GICv2 architecture indicates that the source CPUID for
 549                  * an SGI should be provided during an EOI which implies that
 550                  * the active state is stored somewhere, but at the same time
 551                  * this state is not architecturally exposed anywhere and we
 552                  * have no way of knowing the right source.
 553                  *
 554                  * This may lead to a VCPU not being able to receive
 555                  * additional instances of a particular SGI after migration
 556                  * for a GICv2 VM on some GIC implementations.  Oh well.
 557                  */
 558                 active_source = (requester_vcpu) ? requester_vcpu->vcpu_id : 0;
 559
 560                 if (model == KVM_DEV_TYPE_ARM_VGIC_V2 &&
 561                     active && vgic_irq_is_sgi(irq->intid))
 562                         irq->active_source = active_source;
 563         }
 564
 565         if (irq->active)
 566                 vgic_queue_irq_unlock(vcpu->kvm, irq, flags);
 567         else
 568                 raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
 569 }
 570
 571 static void __vgic_mmio_write_cactive(struct kvm_vcpu *vcpu,
 572                                       gpa_t addr, unsigned int len,
 573                                       unsigned long val)
 574 {
 575         u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
 576         int i;
 577
 578         for_each_set_bit(i, &val, len * 8) {
 579                 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 580                 vgic_mmio_change_active(vcpu, irq, false);
 581                 vgic_put_irq(vcpu->kvm, irq);
 582         }
 583 }
 584
 585 void vgic_mmio_write_cactive(struct kvm_vcpu *vcpu,
 586                              gpa_t addr, unsigned int len,
 587                              unsigned long val)
 588 {
 589         u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
 590
 591         mutex_lock(&vcpu->kvm->lock);
 592         vgic_access_active_prepare(vcpu, intid);
 593
 594         __vgic_mmio_write_cactive(vcpu, addr, len, val);
 595
 596         vgic_access_active_finish(vcpu, intid);
 597         mutex_unlock(&vcpu->kvm->lock);
 598 }
 599
 600 int vgic_mmio_uaccess_write_cactive(struct kvm_vcpu *vcpu,
 601                                      gpa_t addr, unsigned int len,
 602                                      unsigned long val)
 603 {
 604         __vgic_mmio_write_cactive(vcpu, addr, len, val);
 605         return 0;
 606 }
 607
 608 static void __vgic_mmio_write_sactive(struct kvm_vcpu *vcpu,
 609                                       gpa_t addr, unsigned int len,
 610                                       unsigned long val)
 611 {
 612         u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
 613         int i;
 614
 615         for_each_set_bit(i, &val, len * 8) {
 616                 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 617                 vgic_mmio_change_active(vcpu, irq, true);
 618                 vgic_put_irq(vcpu->kvm, irq);
 619         }
 620 }
 621
 622 void vgic_mmio_write_sactive(struct kvm_vcpu *vcpu,
 623                              gpa_t addr, unsigned int len,
 624                              unsigned long val)
 625 {
 626         u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
 627
 628         mutex_lock(&vcpu->kvm->lock);
 629         vgic_access_active_prepare(vcpu, intid);
 630
 631         __vgic_mmio_write_sactive(vcpu, addr, len, val);
 632
 633         vgic_access_active_finish(vcpu, intid);
 634         mutex_unlock(&vcpu->kvm->lock);
 635 }
 636
 637 int vgic_mmio_uaccess_write_sactive(struct kvm_vcpu *vcpu,
 638                                      gpa_t addr, unsigned int len,
 639                                      unsigned long val)
 640 {
 641         __vgic_mmio_write_sactive(vcpu, addr, len, val);
 642         return 0;
 643 }
 644
 645 unsigned long vgic_mmio_read_priority(struct kvm_vcpu *vcpu,
 646                                       gpa_t addr, unsigned int len)
 647 {
 648         u32 intid = VGIC_ADDR_TO_INTID(addr, 8);
 649         int i;
 650         u64 val = 0;
 651
 652         for (i = 0; i < len; i++) {
 653                 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 654
 655                 val |= (u64)irq->priority << (i * 8);
 656
 657                 vgic_put_irq(vcpu->kvm, irq);
 658         }
 659
 660         return val;
 661 }
 662
 663 /*
 664  * We currently don't handle changing the priority of an interrupt that
 665  * is already pending on a VCPU. If there is a need for this, we would
 666  * need to make this VCPU exit and re-evaluate the priorities, potentially
 667  * leading to this interrupt getting presented now to the guest (if it has
 668  * been masked by the priority mask before).
 669  */
 670 void vgic_mmio_write_priority(struct kvm_vcpu *vcpu,
 671                               gpa_t addr, unsigned int len,
 672                               unsigned long val)
 673 {
 674         u32 intid = VGIC_ADDR_TO_INTID(addr, 8);
 675         int i;
 676         unsigned long flags;
 677
 678         for (i = 0; i < len; i++) {
 679                 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 680
 681                 raw_spin_lock_irqsave(&irq->irq_lock, flags);
 682                 /* Narrow the priority range to what we actually support */
 683                 irq->priority = (val >> (i * 8)) & GENMASK(7, 8 - VGIC_PRI_BITS);
 684                 if (irq->hw && vgic_irq_is_sgi(irq->intid))
 685                         vgic_update_vsgi(irq);
 686                 raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
 687
 688                 vgic_put_irq(vcpu->kvm, irq);
 689         }
 690 }
 691
 692 unsigned long vgic_mmio_read_config(struct kvm_vcpu *vcpu,
 693                                     gpa_t addr, unsigned int len)
 694 {
 695         u32 intid = VGIC_ADDR_TO_INTID(addr, 2);
 696         u32 value = 0;
 697         int i;
 698
 699         for (i = 0; i < len * 4; i++) {
 700                 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 701
 702                 if (irq->config == VGIC_CONFIG_EDGE)
 703                         value |= (2U << (i * 2));
 704
 705                 vgic_put_irq(vcpu->kvm, irq);
 706         }
 707
 708         return value;
 709 }
 710
 711 void vgic_mmio_write_config(struct kvm_vcpu *vcpu,
 712                             gpa_t addr, unsigned int len,
 713                             unsigned long val)
 714 {
 715         u32 intid = VGIC_ADDR_TO_INTID(addr, 2);
 716         int i;
 717         unsigned long flags;
 718
 719         for (i = 0; i < len * 4; i++) {
 720                 struct vgic_irq *irq;
 721
 722                 /*
 723                  * The configuration cannot be changed for SGIs in general,
 724                  * for PPIs this is IMPLEMENTATION DEFINED. The arch timer
 725                  * code relies on PPIs being level triggered, so we also
 726                  * make them read-only here.
 727                  */
 728                 if (intid + i < VGIC_NR_PRIVATE_IRQS)
 729                         continue;
 730
 731                 irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 732                 raw_spin_lock_irqsave(&irq->irq_lock, flags);
 733
 734                 if (test_bit(i * 2 + 1, &val))
 735                         irq->config = VGIC_CONFIG_EDGE;
 736                 else
 737                         irq->config = VGIC_CONFIG_LEVEL;
 738
 739                 raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
 740                 vgic_put_irq(vcpu->kvm, irq);
 741         }
 742 }
 743
 744 u64 vgic_read_irq_line_level_info(struct kvm_vcpu *vcpu, u32 intid)
 745 {
 746         int i;
 747         u64 val = 0;
 748         int nr_irqs = vcpu->kvm->arch.vgic.nr_spis + VGIC_NR_PRIVATE_IRQS;
 749
 750         for (i = 0; i < 32; i++) {
 751                 struct vgic_irq *irq;
 752
 753                 if ((intid + i) < VGIC_NR_SGIS || (intid + i) >= nr_irqs)
 754                         continue;
 755
 756                 irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 757                 if (irq->config == VGIC_CONFIG_LEVEL && irq->line_level)
 758                         val |= (1U << i);
 759
 760                 vgic_put_irq(vcpu->kvm, irq);
 761         }
 762
 763         return val;
 764 }
 765
 766 void vgic_write_irq_line_level_info(struct kvm_vcpu *vcpu, u32 intid,
 767                                     const u64 val)
 768 {
 769         int i;
 770         int nr_irqs = vcpu->kvm->arch.vgic.nr_spis + VGIC_NR_PRIVATE_IRQS;
 771         unsigned long flags;
 772
 773         for (i = 0; i < 32; i++) {
 774                 struct vgic_irq *irq;
 775                 bool new_level;
 776
 777                 if ((intid + i) < VGIC_NR_SGIS || (intid + i) >= nr_irqs)
 778                         continue;
 779
 780                 irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
 781
 782                 /*
 783                  * Line level is set irrespective of irq type
 784                  * (level or edge) to avoid dependency that VM should
 785                  * restore irq config before line level.
 786                  */
 787                 new_level = !!(val & (1U << i));
 788                 raw_spin_lock_irqsave(&irq->irq_lock, flags);
 789                 irq->line_level = new_level;
 790                 if (new_level)
 791                         vgic_queue_irq_unlock(vcpu->kvm, irq, flags);
 792                 else
 793                         raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
 794
 795                 vgic_put_irq(vcpu->kvm, irq);
 796         }
 797 }
 798
 799 static int match_region(const void *key, const void *elt)
 800 {
 801         const unsigned int offset = (unsigned long)key;
 802         const struct vgic_register_region *region = elt;
 803
 804         if (offset < region->reg_offset)
 805                 return -1;
 806
 807         if (offset >= region->reg_offset + region->len)
 808                 return 1;
 809
 810         return 0;
 811 }
 812
 813 const struct vgic_register_region *
 814 vgic_find_mmio_region(const struct vgic_register_region *regions,
 815                       int nr_regions, unsigned int offset)
 816 {
 817         return bsearch((void *)(uintptr_t)offset, regions, nr_regions,
 818                        sizeof(regions[0]), match_region);
 819 }
 820
 821 void vgic_set_vmcr(struct kvm_vcpu *vcpu, struct vgic_vmcr *vmcr)
 822 {
 823         if (kvm_vgic_global_state.type == VGIC_V2)
 824                 vgic_v2_set_vmcr(vcpu, vmcr);
 825         else
 826                 vgic_v3_set_vmcr(vcpu, vmcr);
 827 }
 828
 829 void vgic_get_vmcr(struct kvm_vcpu *vcpu, struct vgic_vmcr *vmcr)
 830 {
 831         if (kvm_vgic_global_state.type == VGIC_V2)
 832                 vgic_v2_get_vmcr(vcpu, vmcr);
 833         else
 834                 vgic_v3_get_vmcr(vcpu, vmcr);
 835 }
 836
 837 /*
 838  * kvm_mmio_read_buf() returns a value in a format where it can be converted
 839  * to a byte array and be directly observed as the guest wanted it to appear
 840  * in memory if it had done the store itself, which is LE for the GIC, as the
 841  * guest knows the GIC is always LE.
 842  *
 843  * We convert this value to the CPUs native format to deal with it as a data
 844  * value.
 845  */
 846 unsigned long vgic_data_mmio_bus_to_host(const void *val, unsigned int len)
 847 {
 848         unsigned long data = kvm_mmio_read_buf(val, len);
 849
 850         switch (len) {
 851         case 1:
 852                 return data;
 853         case 2:
 854                 return le16_to_cpu(data);
 855         case 4:
 856                 return le32_to_cpu(data);
 857         default:
 858                 return le64_to_cpu(data);
 859         }
 860 }
 861
 862 /*
 863  * kvm_mmio_write_buf() expects a value in a format such that if converted to
 864  * a byte array it is observed as the guest would see it if it could perform
 865  * the load directly.  Since the GIC is LE, and the guest knows this, the
 866  * guest expects a value in little endian format.
 867  *
 868  * We convert the data value from the CPUs native format to LE so that the
 869  * value is returned in the proper format.
 870  */
 871 void vgic_data_host_to_mmio_bus(void *buf, unsigned int len,
 872                                 unsigned long data)
 873 {
 874         switch (len) {
 875         case 1:
 876                 break;
 877         case 2:
 878                 data = cpu_to_le16(data);
 879                 break;
 880         case 4:
 881                 data = cpu_to_le32(data);
 882                 break;
 883         default:
 884                 data = cpu_to_le64(data);
 885         }
 886
 887         kvm_mmio_write_buf(buf, len, data);
 888 }
 889
 890 static
 891 struct vgic_io_device *kvm_to_vgic_iodev(const struct kvm_io_device *dev)
 892 {
 893         return container_of(dev, struct vgic_io_device, dev);
 894 }
 895
 896 static bool check_region(const struct kvm *kvm,
 897                          const struct vgic_register_region *region,
 898                          gpa_t addr, int len)
 899 {
 900         int flags, nr_irqs = kvm->arch.vgic.nr_spis + VGIC_NR_PRIVATE_IRQS;
 901
 902         switch (len) {
 903         case sizeof(u8):
 904                 flags = VGIC_ACCESS_8bit;
 905                 break;
 906         case sizeof(u32):
 907                 flags = VGIC_ACCESS_32bit;
 908                 break;
 909         case sizeof(u64):
 910                 flags = VGIC_ACCESS_64bit;
 911                 break;
 912         default:
 913                 return false;
 914         }
 915
 916         if ((region->access_flags & flags) && IS_ALIGNED(addr, len)) {
 917                 if (!region->bits_per_irq)
 918                         return true;
 919
 920                 /* Do we access a non-allocated IRQ? */
 921                 return VGIC_ADDR_TO_INTID(addr, region->bits_per_irq) < nr_irqs;
 922         }
 923
 924         return false;
 925 }
 926
 927 const struct vgic_register_region *
 928 vgic_get_mmio_region(struct kvm_vcpu *vcpu, struct vgic_io_device *iodev,
 929                      gpa_t addr, int len)
 930 {
 931         const struct vgic_register_region *region;
 932
 933         region = vgic_find_mmio_region(iodev->regions, iodev->nr_regions,
 934                                        addr - iodev->base_addr);
 935         if (!region || !check_region(vcpu->kvm, region, addr, len))
 936                 return NULL;
 937
 938         return region;
 939 }
 940
 941 static int vgic_uaccess_read(struct kvm_vcpu *vcpu, struct kvm_io_device *dev,
 942                              gpa_t addr, u32 *val)
 943 {
 944         struct vgic_io_device *iodev = kvm_to_vgic_iodev(dev);
 945         const struct vgic_register_region *region;
 946         struct kvm_vcpu *r_vcpu;
 947
 948         region = vgic_get_mmio_region(vcpu, iodev, addr, sizeof(u32));
 949         if (!region) {
 950                 *val = 0;
 951                 return 0;
 952         }
 953
 954         r_vcpu = iodev->redist_vcpu ? iodev->redist_vcpu : vcpu;
 955         if (region->uaccess_read)
 956                 *val = region->uaccess_read(r_vcpu, addr, sizeof(u32));
 957         else
 958                 *val = region->read(r_vcpu, addr, sizeof(u32));
 959
 960         return 0;
 961 }
 962
 963 static int vgic_uaccess_write(struct kvm_vcpu *vcpu, struct kvm_io_device *dev,
 964                               gpa_t addr, const u32 *val)
 965 {
 966         struct vgic_io_device *iodev = kvm_to_vgic_iodev(dev);
 967         const struct vgic_register_region *region;
 968         struct kvm_vcpu *r_vcpu;
 969
 970         region = vgic_get_mmio_region(vcpu, iodev, addr, sizeof(u32));
 971         if (!region)
 972                 return 0;
 973
 974         r_vcpu = iodev->redist_vcpu ? iodev->redist_vcpu : vcpu;
 975         if (region->uaccess_write)
 976                 return region->uaccess_write(r_vcpu, addr, sizeof(u32), *val);
 977
 978         region->write(r_vcpu, addr, sizeof(u32), *val);
 979         return 0;
 980 }
 981
 982 /*
 983  * Userland access to VGIC registers.
 984  */
 985 int vgic_uaccess(struct kvm_vcpu *vcpu, struct vgic_io_device *dev,
 986                  bool is_write, int offset, u32 *val)
 987 {
 988         if (is_write)
 989                 return vgic_uaccess_write(vcpu, &dev->dev, offset, val);
 990         else
 991                 return vgic_uaccess_read(vcpu, &dev->dev, offset, val);
 992 }
 993
 994 static int dispatch_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *dev,
 995                               gpa_t addr, int len, void *val)
 996 {
 997         struct vgic_io_device *iodev = kvm_to_vgic_iodev(dev);
 998         const struct vgic_register_region *region;
 999         unsigned long data = 0;
1000
1001         region = vgic_get_mmio_region(vcpu, iodev, addr, len);
1002         if (!region) {
1003                 memset(val, 0, len);
1004                 return 0;
1005         }
1006
1007         switch (iodev->iodev_type) {
1008         case IODEV_CPUIF:
1009                 data = region->read(vcpu, addr, len);
1010                 break;
1011         case IODEV_DIST:
1012                 data = region->read(vcpu, addr, len);
1013                 break;
1014         case IODEV_REDIST:
1015                 data = region->read(iodev->redist_vcpu, addr, len);
1016                 break;
1017         case IODEV_ITS:
1018                 data = region->its_read(vcpu->kvm, iodev->its, addr, len);
1019                 break;
1020         }
1021
1022         vgic_data_host_to_mmio_bus(val, len, data);
1023         return 0;
1024 }
1025
1026 static int dispatch_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *dev,
1027                                gpa_t addr, int len, const void *val)
1028 {
1029         struct vgic_io_device *iodev = kvm_to_vgic_iodev(dev);
1030         const struct vgic_register_region *region;
1031         unsigned long data = vgic_data_mmio_bus_to_host(val, len);
1032
1033         region = vgic_get_mmio_region(vcpu, iodev, addr, len);
1034         if (!region)
1035                 return 0;
1036
1037         switch (iodev->iodev_type) {
1038         case IODEV_CPUIF:
1039                 region->write(vcpu, addr, len, data);
1040                 break;
1041         case IODEV_DIST:
1042                 region->write(vcpu, addr, len, data);
1043                 break;
1044         case IODEV_REDIST:
1045                 region->write(iodev->redist_vcpu, addr, len, data);
1046                 break;
1047         case IODEV_ITS:
1048                 region->its_write(vcpu->kvm, iodev->its, addr, len, data);
1049                 break;
1050         }
1051
1052         return 0;
1053 }
1054
1055 struct kvm_io_device_ops kvm_io_gic_ops = {
1056         .read = dispatch_mmio_read,
1057         .write = dispatch_mmio_write,
1058 };
1059
1060 int vgic_register_dist_iodev(struct kvm *kvm, gpa_t dist_base_address,
1061                              enum vgic_type type)
1062 {
1063         struct vgic_io_device *io_device = &kvm->arch.vgic.dist_iodev;
1064         int ret = 0;
1065         unsigned int len;
1066
1067         switch (type) {
1068         case VGIC_V2:
1069                 len = vgic_v2_init_dist_iodev(io_device);
1070                 break;
1071         case VGIC_V3:
1072                 len = vgic_v3_init_dist_iodev(io_device);
1073                 break;
1074         default:
1075                 BUG_ON(1);
1076         }
1077
1078         io_device->base_addr = dist_base_address;
1079         io_device->iodev_type = IODEV_DIST;
1080         io_device->redist_vcpu = NULL;
1081
1082         mutex_lock(&kvm->slots_lock);
1083         ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, dist_base_address,
1084                                       len, &io_device->dev);
1085         mutex_unlock(&kvm->slots_lock);
1086
1087         return ret;
1088 }