1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * VGICv3 MMIO handling functions 4 */ 5 6 #include <linux/bitfield.h> 7 #include <linux/irqchip/arm-gic-v3.h> 8 #include <linux/kvm.h> 9 #include <linux/kvm_host.h> 10 #include <linux/interrupt.h> 11 #include <kvm/iodev.h> 12 #include <kvm/arm_vgic.h> 13 14 #include <asm/kvm_emulate.h> 15 #include <asm/kvm_arm.h> 16 #include <asm/kvm_mmu.h> 17 18 #include "vgic.h" 19 #include "vgic-mmio.h" 20 21 /* extract @num bytes at @offset bytes offset in data */ 22 unsigned long extract_bytes(u64 data, unsigned int offset, 23 unsigned int num) 24 { 25 return (data >> (offset * 8)) & GENMASK_ULL(num * 8 - 1, 0); 26 } 27 28 /* allows updates of any half of a 64-bit register (or the whole thing) */ 29 u64 update_64bit_reg(u64 reg, unsigned int offset, unsigned int len, 30 unsigned long val) 31 { 32 int lower = (offset & 4) * 8; 33 int upper = lower + 8 * len - 1; 34 35 reg &= ~GENMASK_ULL(upper, lower); 36 val &= GENMASK_ULL(len * 8 - 1, 0); 37 38 return reg | ((u64)val << lower); 39 } 40 41 bool vgic_has_its(struct kvm *kvm) 42 { 43 struct vgic_dist *dist = &kvm->arch.vgic; 44 45 if (dist->vgic_model != KVM_DEV_TYPE_ARM_VGIC_V3) 46 return false; 47 48 return dist->has_its; 49 } 50 51 bool vgic_supports_direct_msis(struct kvm *kvm) 52 { 53 return (kvm_vgic_global_state.has_gicv4_1 || 54 (kvm_vgic_global_state.has_gicv4 && vgic_has_its(kvm))); 55 } 56 57 /* 58 * The Revision field in the IIDR have the following meanings: 59 * 60 * Revision 2: Interrupt groups are guest-configurable and signaled using 61 * their configured groups. 62 */ 63 64 static unsigned long vgic_mmio_read_v3_misc(struct kvm_vcpu *vcpu, 65 gpa_t addr, unsigned int len) 66 { 67 struct vgic_dist *vgic = &vcpu->kvm->arch.vgic; 68 u32 value = 0; 69 70 switch (addr & 0x0c) { 71 case GICD_CTLR: 72 if (vgic->enabled) 73 value |= GICD_CTLR_ENABLE_SS_G1; 74 value |= GICD_CTLR_ARE_NS | GICD_CTLR_DS; 75 if (vgic->nassgireq) 76 value |= GICD_CTLR_nASSGIreq; 77 break; 78 case GICD_TYPER: 79 value = vgic->nr_spis + VGIC_NR_PRIVATE_IRQS; 80 value = (value >> 5) - 1; 81 if (vgic_has_its(vcpu->kvm)) { 82 value |= (INTERRUPT_ID_BITS_ITS - 1) << 19; 83 value |= GICD_TYPER_LPIS; 84 } else { 85 value |= (INTERRUPT_ID_BITS_SPIS - 1) << 19; 86 } 87 break; 88 case GICD_TYPER2: 89 if (kvm_vgic_global_state.has_gicv4_1 && gic_cpuif_has_vsgi()) 90 value = GICD_TYPER2_nASSGIcap; 91 break; 92 case GICD_IIDR: 93 value = (PRODUCT_ID_KVM << GICD_IIDR_PRODUCT_ID_SHIFT) | 94 (vgic->implementation_rev << GICD_IIDR_REVISION_SHIFT) | 95 (IMPLEMENTER_ARM << GICD_IIDR_IMPLEMENTER_SHIFT); 96 break; 97 default: 98 return 0; 99 } 100 101 return value; 102 } 103 104 static void vgic_mmio_write_v3_misc(struct kvm_vcpu *vcpu, 105 gpa_t addr, unsigned int len, 106 unsigned long val) 107 { 108 struct vgic_dist *dist = &vcpu->kvm->arch.vgic; 109 110 switch (addr & 0x0c) { 111 case GICD_CTLR: { 112 bool was_enabled, is_hwsgi; 113 114 mutex_lock(&vcpu->kvm->arch.config_lock); 115 116 was_enabled = dist->enabled; 117 is_hwsgi = dist->nassgireq; 118 119 dist->enabled = val & GICD_CTLR_ENABLE_SS_G1; 120 121 /* Not a GICv4.1? No HW SGIs */ 122 if (!kvm_vgic_global_state.has_gicv4_1 || !gic_cpuif_has_vsgi()) 123 val &= ~GICD_CTLR_nASSGIreq; 124 125 /* Dist stays enabled? nASSGIreq is RO */ 126 if (was_enabled && dist->enabled) { 127 val &= ~GICD_CTLR_nASSGIreq; 128 val |= FIELD_PREP(GICD_CTLR_nASSGIreq, is_hwsgi); 129 } 130 131 /* Switching HW SGIs? */ 132 dist->nassgireq = val & GICD_CTLR_nASSGIreq; 133 if (is_hwsgi != dist->nassgireq) 134 vgic_v4_configure_vsgis(vcpu->kvm); 135 136 if (kvm_vgic_global_state.has_gicv4_1 && 137 was_enabled != dist->enabled) 138 kvm_make_all_cpus_request(vcpu->kvm, KVM_REQ_RELOAD_GICv4); 139 else if (!was_enabled && dist->enabled) 140 vgic_kick_vcpus(vcpu->kvm); 141 142 mutex_unlock(&vcpu->kvm->arch.config_lock); 143 break; 144 } 145 case GICD_TYPER: 146 case GICD_TYPER2: 147 case GICD_IIDR: 148 /* This is at best for documentation purposes... */ 149 return; 150 } 151 } 152 153 static int vgic_mmio_uaccess_write_v3_misc(struct kvm_vcpu *vcpu, 154 gpa_t addr, unsigned int len, 155 unsigned long val) 156 { 157 struct vgic_dist *dist = &vcpu->kvm->arch.vgic; 158 u32 reg; 159 160 switch (addr & 0x0c) { 161 case GICD_TYPER2: 162 if (val != vgic_mmio_read_v3_misc(vcpu, addr, len)) 163 return -EINVAL; 164 return 0; 165 case GICD_IIDR: 166 reg = vgic_mmio_read_v3_misc(vcpu, addr, len); 167 if ((reg ^ val) & ~GICD_IIDR_REVISION_MASK) 168 return -EINVAL; 169 170 reg = FIELD_GET(GICD_IIDR_REVISION_MASK, reg); 171 switch (reg) { 172 case KVM_VGIC_IMP_REV_2: 173 case KVM_VGIC_IMP_REV_3: 174 dist->implementation_rev = reg; 175 return 0; 176 default: 177 return -EINVAL; 178 } 179 case GICD_CTLR: 180 /* Not a GICv4.1? No HW SGIs */ 181 if (!kvm_vgic_global_state.has_gicv4_1) 182 val &= ~GICD_CTLR_nASSGIreq; 183 184 dist->enabled = val & GICD_CTLR_ENABLE_SS_G1; 185 dist->nassgireq = val & GICD_CTLR_nASSGIreq; 186 return 0; 187 } 188 189 vgic_mmio_write_v3_misc(vcpu, addr, len, val); 190 return 0; 191 } 192 193 static unsigned long vgic_mmio_read_irouter(struct kvm_vcpu *vcpu, 194 gpa_t addr, unsigned int len) 195 { 196 int intid = VGIC_ADDR_TO_INTID(addr, 64); 197 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, NULL, intid); 198 unsigned long ret = 0; 199 200 if (!irq) 201 return 0; 202 203 /* The upper word is RAZ for us. */ 204 if (!(addr & 4)) 205 ret = extract_bytes(READ_ONCE(irq->mpidr), addr & 7, len); 206 207 vgic_put_irq(vcpu->kvm, irq); 208 return ret; 209 } 210 211 static void vgic_mmio_write_irouter(struct kvm_vcpu *vcpu, 212 gpa_t addr, unsigned int len, 213 unsigned long val) 214 { 215 int intid = VGIC_ADDR_TO_INTID(addr, 64); 216 struct vgic_irq *irq; 217 unsigned long flags; 218 219 /* The upper word is WI for us since we don't implement Aff3. */ 220 if (addr & 4) 221 return; 222 223 irq = vgic_get_irq(vcpu->kvm, NULL, intid); 224 225 if (!irq) 226 return; 227 228 raw_spin_lock_irqsave(&irq->irq_lock, flags); 229 230 /* We only care about and preserve Aff0, Aff1 and Aff2. */ 231 irq->mpidr = val & GENMASK(23, 0); 232 irq->target_vcpu = kvm_mpidr_to_vcpu(vcpu->kvm, irq->mpidr); 233 234 raw_spin_unlock_irqrestore(&irq->irq_lock, flags); 235 vgic_put_irq(vcpu->kvm, irq); 236 } 237 238 bool vgic_lpis_enabled(struct kvm_vcpu *vcpu) 239 { 240 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu; 241 242 return atomic_read(&vgic_cpu->ctlr) == GICR_CTLR_ENABLE_LPIS; 243 } 244 245 static unsigned long vgic_mmio_read_v3r_ctlr(struct kvm_vcpu *vcpu, 246 gpa_t addr, unsigned int len) 247 { 248 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu; 249 unsigned long val; 250 251 val = atomic_read(&vgic_cpu->ctlr); 252 if (vgic_get_implementation_rev(vcpu) >= KVM_VGIC_IMP_REV_3) 253 val |= GICR_CTLR_IR | GICR_CTLR_CES; 254 255 return val; 256 } 257 258 static void vgic_mmio_write_v3r_ctlr(struct kvm_vcpu *vcpu, 259 gpa_t addr, unsigned int len, 260 unsigned long val) 261 { 262 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu; 263 u32 ctlr; 264 265 if (!vgic_has_its(vcpu->kvm)) 266 return; 267 268 if (!(val & GICR_CTLR_ENABLE_LPIS)) { 269 /* 270 * Don't disable if RWP is set, as there already an 271 * ongoing disable. Funky guest... 272 */ 273 ctlr = atomic_cmpxchg_acquire(&vgic_cpu->ctlr, 274 GICR_CTLR_ENABLE_LPIS, 275 GICR_CTLR_RWP); 276 if (ctlr != GICR_CTLR_ENABLE_LPIS) 277 return; 278 279 vgic_flush_pending_lpis(vcpu); 280 vgic_its_invalidate_cache(vcpu->kvm); 281 atomic_set_release(&vgic_cpu->ctlr, 0); 282 } else { 283 ctlr = atomic_cmpxchg_acquire(&vgic_cpu->ctlr, 0, 284 GICR_CTLR_ENABLE_LPIS); 285 if (ctlr != 0) 286 return; 287 288 vgic_enable_lpis(vcpu); 289 } 290 } 291 292 static bool vgic_mmio_vcpu_rdist_is_last(struct kvm_vcpu *vcpu) 293 { 294 struct vgic_dist *vgic = &vcpu->kvm->arch.vgic; 295 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu; 296 struct vgic_redist_region *iter, *rdreg = vgic_cpu->rdreg; 297 298 if (!rdreg) 299 return false; 300 301 if (vgic_cpu->rdreg_index < rdreg->free_index - 1) { 302 return false; 303 } else if (rdreg->count && vgic_cpu->rdreg_index == (rdreg->count - 1)) { 304 struct list_head *rd_regions = &vgic->rd_regions; 305 gpa_t end = rdreg->base + rdreg->count * KVM_VGIC_V3_REDIST_SIZE; 306 307 /* 308 * the rdist is the last one of the redist region, 309 * check whether there is no other contiguous rdist region 310 */ 311 list_for_each_entry(iter, rd_regions, list) { 312 if (iter->base == end && iter->free_index > 0) 313 return false; 314 } 315 } 316 return true; 317 } 318 319 static unsigned long vgic_mmio_read_v3r_typer(struct kvm_vcpu *vcpu, 320 gpa_t addr, unsigned int len) 321 { 322 unsigned long mpidr = kvm_vcpu_get_mpidr_aff(vcpu); 323 int target_vcpu_id = vcpu->vcpu_id; 324 u64 value; 325 326 value = (u64)(mpidr & GENMASK(23, 0)) << 32; 327 value |= ((target_vcpu_id & 0xffff) << 8); 328 329 if (vgic_has_its(vcpu->kvm)) 330 value |= GICR_TYPER_PLPIS; 331 332 if (vgic_mmio_vcpu_rdist_is_last(vcpu)) 333 value |= GICR_TYPER_LAST; 334 335 return extract_bytes(value, addr & 7, len); 336 } 337 338 static unsigned long vgic_mmio_read_v3r_iidr(struct kvm_vcpu *vcpu, 339 gpa_t addr, unsigned int len) 340 { 341 return (PRODUCT_ID_KVM << 24) | (IMPLEMENTER_ARM << 0); 342 } 343 344 static unsigned long vgic_mmio_read_v3_idregs(struct kvm_vcpu *vcpu, 345 gpa_t addr, unsigned int len) 346 { 347 switch (addr & 0xffff) { 348 case GICD_PIDR2: 349 /* report a GICv3 compliant implementation */ 350 return 0x3b; 351 } 352 353 return 0; 354 } 355 356 static int vgic_v3_uaccess_write_pending(struct kvm_vcpu *vcpu, 357 gpa_t addr, unsigned int len, 358 unsigned long val) 359 { 360 u32 intid = VGIC_ADDR_TO_INTID(addr, 1); 361 int i; 362 unsigned long flags; 363 364 for (i = 0; i < len * 8; i++) { 365 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i); 366 367 raw_spin_lock_irqsave(&irq->irq_lock, flags); 368 if (test_bit(i, &val)) { 369 /* 370 * pending_latch is set irrespective of irq type 371 * (level or edge) to avoid dependency that VM should 372 * restore irq config before pending info. 373 */ 374 irq->pending_latch = true; 375 vgic_queue_irq_unlock(vcpu->kvm, irq, flags); 376 } else { 377 irq->pending_latch = false; 378 raw_spin_unlock_irqrestore(&irq->irq_lock, flags); 379 } 380 381 vgic_put_irq(vcpu->kvm, irq); 382 } 383 384 return 0; 385 } 386 387 /* We want to avoid outer shareable. */ 388 u64 vgic_sanitise_shareability(u64 field) 389 { 390 switch (field) { 391 case GIC_BASER_OuterShareable: 392 return GIC_BASER_InnerShareable; 393 default: 394 return field; 395 } 396 } 397 398 /* Avoid any inner non-cacheable mapping. */ 399 u64 vgic_sanitise_inner_cacheability(u64 field) 400 { 401 switch (field) { 402 case GIC_BASER_CACHE_nCnB: 403 case GIC_BASER_CACHE_nC: 404 return GIC_BASER_CACHE_RaWb; 405 default: 406 return field; 407 } 408 } 409 410 /* Non-cacheable or same-as-inner are OK. */ 411 u64 vgic_sanitise_outer_cacheability(u64 field) 412 { 413 switch (field) { 414 case GIC_BASER_CACHE_SameAsInner: 415 case GIC_BASER_CACHE_nC: 416 return field; 417 default: 418 return GIC_BASER_CACHE_SameAsInner; 419 } 420 } 421 422 u64 vgic_sanitise_field(u64 reg, u64 field_mask, int field_shift, 423 u64 (*sanitise_fn)(u64)) 424 { 425 u64 field = (reg & field_mask) >> field_shift; 426 427 field = sanitise_fn(field) << field_shift; 428 return (reg & ~field_mask) | field; 429 } 430 431 #define PROPBASER_RES0_MASK \ 432 (GENMASK_ULL(63, 59) | GENMASK_ULL(55, 52) | GENMASK_ULL(6, 5)) 433 #define PENDBASER_RES0_MASK \ 434 (BIT_ULL(63) | GENMASK_ULL(61, 59) | GENMASK_ULL(55, 52) | \ 435 GENMASK_ULL(15, 12) | GENMASK_ULL(6, 0)) 436 437 static u64 vgic_sanitise_pendbaser(u64 reg) 438 { 439 reg = vgic_sanitise_field(reg, GICR_PENDBASER_SHAREABILITY_MASK, 440 GICR_PENDBASER_SHAREABILITY_SHIFT, 441 vgic_sanitise_shareability); 442 reg = vgic_sanitise_field(reg, GICR_PENDBASER_INNER_CACHEABILITY_MASK, 443 GICR_PENDBASER_INNER_CACHEABILITY_SHIFT, 444 vgic_sanitise_inner_cacheability); 445 reg = vgic_sanitise_field(reg, GICR_PENDBASER_OUTER_CACHEABILITY_MASK, 446 GICR_PENDBASER_OUTER_CACHEABILITY_SHIFT, 447 vgic_sanitise_outer_cacheability); 448 449 reg &= ~PENDBASER_RES0_MASK; 450 451 return reg; 452 } 453 454 static u64 vgic_sanitise_propbaser(u64 reg) 455 { 456 reg = vgic_sanitise_field(reg, GICR_PROPBASER_SHAREABILITY_MASK, 457 GICR_PROPBASER_SHAREABILITY_SHIFT, 458 vgic_sanitise_shareability); 459 reg = vgic_sanitise_field(reg, GICR_PROPBASER_INNER_CACHEABILITY_MASK, 460 GICR_PROPBASER_INNER_CACHEABILITY_SHIFT, 461 vgic_sanitise_inner_cacheability); 462 reg = vgic_sanitise_field(reg, GICR_PROPBASER_OUTER_CACHEABILITY_MASK, 463 GICR_PROPBASER_OUTER_CACHEABILITY_SHIFT, 464 vgic_sanitise_outer_cacheability); 465 466 reg &= ~PROPBASER_RES0_MASK; 467 return reg; 468 } 469 470 static unsigned long vgic_mmio_read_propbase(struct kvm_vcpu *vcpu, 471 gpa_t addr, unsigned int len) 472 { 473 struct vgic_dist *dist = &vcpu->kvm->arch.vgic; 474 475 return extract_bytes(dist->propbaser, addr & 7, len); 476 } 477 478 static void vgic_mmio_write_propbase(struct kvm_vcpu *vcpu, 479 gpa_t addr, unsigned int len, 480 unsigned long val) 481 { 482 struct vgic_dist *dist = &vcpu->kvm->arch.vgic; 483 u64 old_propbaser, propbaser; 484 485 /* Storing a value with LPIs already enabled is undefined */ 486 if (vgic_lpis_enabled(vcpu)) 487 return; 488 489 do { 490 old_propbaser = READ_ONCE(dist->propbaser); 491 propbaser = old_propbaser; 492 propbaser = update_64bit_reg(propbaser, addr & 4, len, val); 493 propbaser = vgic_sanitise_propbaser(propbaser); 494 } while (cmpxchg64(&dist->propbaser, old_propbaser, 495 propbaser) != old_propbaser); 496 } 497 498 static unsigned long vgic_mmio_read_pendbase(struct kvm_vcpu *vcpu, 499 gpa_t addr, unsigned int len) 500 { 501 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu; 502 u64 value = vgic_cpu->pendbaser; 503 504 value &= ~GICR_PENDBASER_PTZ; 505 506 return extract_bytes(value, addr & 7, len); 507 } 508 509 static void vgic_mmio_write_pendbase(struct kvm_vcpu *vcpu, 510 gpa_t addr, unsigned int len, 511 unsigned long val) 512 { 513 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu; 514 u64 old_pendbaser, pendbaser; 515 516 /* Storing a value with LPIs already enabled is undefined */ 517 if (vgic_lpis_enabled(vcpu)) 518 return; 519 520 do { 521 old_pendbaser = READ_ONCE(vgic_cpu->pendbaser); 522 pendbaser = old_pendbaser; 523 pendbaser = update_64bit_reg(pendbaser, addr & 4, len, val); 524 pendbaser = vgic_sanitise_pendbaser(pendbaser); 525 } while (cmpxchg64(&vgic_cpu->pendbaser, old_pendbaser, 526 pendbaser) != old_pendbaser); 527 } 528 529 static unsigned long vgic_mmio_read_sync(struct kvm_vcpu *vcpu, 530 gpa_t addr, unsigned int len) 531 { 532 return !!atomic_read(&vcpu->arch.vgic_cpu.syncr_busy); 533 } 534 535 static void vgic_set_rdist_busy(struct kvm_vcpu *vcpu, bool busy) 536 { 537 if (busy) { 538 atomic_inc(&vcpu->arch.vgic_cpu.syncr_busy); 539 smp_mb__after_atomic(); 540 } else { 541 smp_mb__before_atomic(); 542 atomic_dec(&vcpu->arch.vgic_cpu.syncr_busy); 543 } 544 } 545 546 static void vgic_mmio_write_invlpi(struct kvm_vcpu *vcpu, 547 gpa_t addr, unsigned int len, 548 unsigned long val) 549 { 550 struct vgic_irq *irq; 551 552 /* 553 * If the guest wrote only to the upper 32bit part of the 554 * register, drop the write on the floor, as it is only for 555 * vPEs (which we don't support for obvious reasons). 556 * 557 * Also discard the access if LPIs are not enabled. 558 */ 559 if ((addr & 4) || !vgic_lpis_enabled(vcpu)) 560 return; 561 562 vgic_set_rdist_busy(vcpu, true); 563 564 irq = vgic_get_irq(vcpu->kvm, NULL, lower_32_bits(val)); 565 if (irq) { 566 vgic_its_inv_lpi(vcpu->kvm, irq); 567 vgic_put_irq(vcpu->kvm, irq); 568 } 569 570 vgic_set_rdist_busy(vcpu, false); 571 } 572 573 static void vgic_mmio_write_invall(struct kvm_vcpu *vcpu, 574 gpa_t addr, unsigned int len, 575 unsigned long val) 576 { 577 /* See vgic_mmio_write_invlpi() for the early return rationale */ 578 if ((addr & 4) || !vgic_lpis_enabled(vcpu)) 579 return; 580 581 vgic_set_rdist_busy(vcpu, true); 582 vgic_its_invall(vcpu); 583 vgic_set_rdist_busy(vcpu, false); 584 } 585 586 /* 587 * The GICv3 per-IRQ registers are split to control PPIs and SGIs in the 588 * redistributors, while SPIs are covered by registers in the distributor 589 * block. Trying to set private IRQs in this block gets ignored. 590 * We take some special care here to fix the calculation of the register 591 * offset. 592 */ 593 #define REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(off, rd, wr, ur, uw, bpi, acc) \ 594 { \ 595 .reg_offset = off, \ 596 .bits_per_irq = bpi, \ 597 .len = (bpi * VGIC_NR_PRIVATE_IRQS) / 8, \ 598 .access_flags = acc, \ 599 .read = vgic_mmio_read_raz, \ 600 .write = vgic_mmio_write_wi, \ 601 }, { \ 602 .reg_offset = off + (bpi * VGIC_NR_PRIVATE_IRQS) / 8, \ 603 .bits_per_irq = bpi, \ 604 .len = (bpi * (1024 - VGIC_NR_PRIVATE_IRQS)) / 8, \ 605 .access_flags = acc, \ 606 .read = rd, \ 607 .write = wr, \ 608 .uaccess_read = ur, \ 609 .uaccess_write = uw, \ 610 } 611 612 static const struct vgic_register_region vgic_v3_dist_registers[] = { 613 REGISTER_DESC_WITH_LENGTH_UACCESS(GICD_CTLR, 614 vgic_mmio_read_v3_misc, vgic_mmio_write_v3_misc, 615 NULL, vgic_mmio_uaccess_write_v3_misc, 616 16, VGIC_ACCESS_32bit), 617 REGISTER_DESC_WITH_LENGTH(GICD_STATUSR, 618 vgic_mmio_read_rao, vgic_mmio_write_wi, 4, 619 VGIC_ACCESS_32bit), 620 REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_IGROUPR, 621 vgic_mmio_read_group, vgic_mmio_write_group, NULL, NULL, 1, 622 VGIC_ACCESS_32bit), 623 REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ISENABLER, 624 vgic_mmio_read_enable, vgic_mmio_write_senable, 625 NULL, vgic_uaccess_write_senable, 1, 626 VGIC_ACCESS_32bit), 627 REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ICENABLER, 628 vgic_mmio_read_enable, vgic_mmio_write_cenable, 629 NULL, vgic_uaccess_write_cenable, 1, 630 VGIC_ACCESS_32bit), 631 REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ISPENDR, 632 vgic_mmio_read_pending, vgic_mmio_write_spending, 633 vgic_uaccess_read_pending, vgic_v3_uaccess_write_pending, 1, 634 VGIC_ACCESS_32bit), 635 REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ICPENDR, 636 vgic_mmio_read_pending, vgic_mmio_write_cpending, 637 vgic_mmio_read_raz, vgic_mmio_uaccess_write_wi, 1, 638 VGIC_ACCESS_32bit), 639 REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ISACTIVER, 640 vgic_mmio_read_active, vgic_mmio_write_sactive, 641 vgic_uaccess_read_active, vgic_mmio_uaccess_write_sactive, 1, 642 VGIC_ACCESS_32bit), 643 REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ICACTIVER, 644 vgic_mmio_read_active, vgic_mmio_write_cactive, 645 vgic_uaccess_read_active, vgic_mmio_uaccess_write_cactive, 646 1, VGIC_ACCESS_32bit), 647 REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_IPRIORITYR, 648 vgic_mmio_read_priority, vgic_mmio_write_priority, NULL, NULL, 649 8, VGIC_ACCESS_32bit | VGIC_ACCESS_8bit), 650 REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ITARGETSR, 651 vgic_mmio_read_raz, vgic_mmio_write_wi, NULL, NULL, 8, 652 VGIC_ACCESS_32bit | VGIC_ACCESS_8bit), 653 REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ICFGR, 654 vgic_mmio_read_config, vgic_mmio_write_config, NULL, NULL, 2, 655 VGIC_ACCESS_32bit), 656 REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_IGRPMODR, 657 vgic_mmio_read_raz, vgic_mmio_write_wi, NULL, NULL, 1, 658 VGIC_ACCESS_32bit), 659 REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_IROUTER, 660 vgic_mmio_read_irouter, vgic_mmio_write_irouter, NULL, NULL, 64, 661 VGIC_ACCESS_64bit | VGIC_ACCESS_32bit), 662 REGISTER_DESC_WITH_LENGTH(GICD_IDREGS, 663 vgic_mmio_read_v3_idregs, vgic_mmio_write_wi, 48, 664 VGIC_ACCESS_32bit), 665 }; 666 667 static const struct vgic_register_region vgic_v3_rd_registers[] = { 668 /* RD_base registers */ 669 REGISTER_DESC_WITH_LENGTH(GICR_CTLR, 670 vgic_mmio_read_v3r_ctlr, vgic_mmio_write_v3r_ctlr, 4, 671 VGIC_ACCESS_32bit), 672 REGISTER_DESC_WITH_LENGTH(GICR_STATUSR, 673 vgic_mmio_read_raz, vgic_mmio_write_wi, 4, 674 VGIC_ACCESS_32bit), 675 REGISTER_DESC_WITH_LENGTH(GICR_IIDR, 676 vgic_mmio_read_v3r_iidr, vgic_mmio_write_wi, 4, 677 VGIC_ACCESS_32bit), 678 REGISTER_DESC_WITH_LENGTH_UACCESS(GICR_TYPER, 679 vgic_mmio_read_v3r_typer, vgic_mmio_write_wi, 680 NULL, vgic_mmio_uaccess_write_wi, 8, 681 VGIC_ACCESS_64bit | VGIC_ACCESS_32bit), 682 REGISTER_DESC_WITH_LENGTH(GICR_WAKER, 683 vgic_mmio_read_raz, vgic_mmio_write_wi, 4, 684 VGIC_ACCESS_32bit), 685 REGISTER_DESC_WITH_LENGTH(GICR_PROPBASER, 686 vgic_mmio_read_propbase, vgic_mmio_write_propbase, 8, 687 VGIC_ACCESS_64bit | VGIC_ACCESS_32bit), 688 REGISTER_DESC_WITH_LENGTH(GICR_PENDBASER, 689 vgic_mmio_read_pendbase, vgic_mmio_write_pendbase, 8, 690 VGIC_ACCESS_64bit | VGIC_ACCESS_32bit), 691 REGISTER_DESC_WITH_LENGTH(GICR_INVLPIR, 692 vgic_mmio_read_raz, vgic_mmio_write_invlpi, 8, 693 VGIC_ACCESS_64bit | VGIC_ACCESS_32bit), 694 REGISTER_DESC_WITH_LENGTH(GICR_INVALLR, 695 vgic_mmio_read_raz, vgic_mmio_write_invall, 8, 696 VGIC_ACCESS_64bit | VGIC_ACCESS_32bit), 697 REGISTER_DESC_WITH_LENGTH(GICR_SYNCR, 698 vgic_mmio_read_sync, vgic_mmio_write_wi, 4, 699 VGIC_ACCESS_32bit), 700 REGISTER_DESC_WITH_LENGTH(GICR_IDREGS, 701 vgic_mmio_read_v3_idregs, vgic_mmio_write_wi, 48, 702 VGIC_ACCESS_32bit), 703 /* SGI_base registers */ 704 REGISTER_DESC_WITH_LENGTH(SZ_64K + GICR_IGROUPR0, 705 vgic_mmio_read_group, vgic_mmio_write_group, 4, 706 VGIC_ACCESS_32bit), 707 REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ISENABLER0, 708 vgic_mmio_read_enable, vgic_mmio_write_senable, 709 NULL, vgic_uaccess_write_senable, 4, 710 VGIC_ACCESS_32bit), 711 REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ICENABLER0, 712 vgic_mmio_read_enable, vgic_mmio_write_cenable, 713 NULL, vgic_uaccess_write_cenable, 4, 714 VGIC_ACCESS_32bit), 715 REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ISPENDR0, 716 vgic_mmio_read_pending, vgic_mmio_write_spending, 717 vgic_uaccess_read_pending, vgic_v3_uaccess_write_pending, 4, 718 VGIC_ACCESS_32bit), 719 REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ICPENDR0, 720 vgic_mmio_read_pending, vgic_mmio_write_cpending, 721 vgic_mmio_read_raz, vgic_mmio_uaccess_write_wi, 4, 722 VGIC_ACCESS_32bit), 723 REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ISACTIVER0, 724 vgic_mmio_read_active, vgic_mmio_write_sactive, 725 vgic_uaccess_read_active, vgic_mmio_uaccess_write_sactive, 4, 726 VGIC_ACCESS_32bit), 727 REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ICACTIVER0, 728 vgic_mmio_read_active, vgic_mmio_write_cactive, 729 vgic_uaccess_read_active, vgic_mmio_uaccess_write_cactive, 4, 730 VGIC_ACCESS_32bit), 731 REGISTER_DESC_WITH_LENGTH(SZ_64K + GICR_IPRIORITYR0, 732 vgic_mmio_read_priority, vgic_mmio_write_priority, 32, 733 VGIC_ACCESS_32bit | VGIC_ACCESS_8bit), 734 REGISTER_DESC_WITH_LENGTH(SZ_64K + GICR_ICFGR0, 735 vgic_mmio_read_config, vgic_mmio_write_config, 8, 736 VGIC_ACCESS_32bit), 737 REGISTER_DESC_WITH_LENGTH(SZ_64K + GICR_IGRPMODR0, 738 vgic_mmio_read_raz, vgic_mmio_write_wi, 4, 739 VGIC_ACCESS_32bit), 740 REGISTER_DESC_WITH_LENGTH(SZ_64K + GICR_NSACR, 741 vgic_mmio_read_raz, vgic_mmio_write_wi, 4, 742 VGIC_ACCESS_32bit), 743 }; 744 745 unsigned int vgic_v3_init_dist_iodev(struct vgic_io_device *dev) 746 { 747 dev->regions = vgic_v3_dist_registers; 748 dev->nr_regions = ARRAY_SIZE(vgic_v3_dist_registers); 749 750 kvm_iodevice_init(&dev->dev, &kvm_io_gic_ops); 751 752 return SZ_64K; 753 } 754 755 /** 756 * vgic_register_redist_iodev - register a single redist iodev 757 * @vcpu: The VCPU to which the redistributor belongs 758 * 759 * Register a KVM iodev for this VCPU's redistributor using the address 760 * provided. 761 * 762 * Return 0 on success, -ERRNO otherwise. 763 */ 764 int vgic_register_redist_iodev(struct kvm_vcpu *vcpu) 765 { 766 struct kvm *kvm = vcpu->kvm; 767 struct vgic_dist *vgic = &kvm->arch.vgic; 768 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu; 769 struct vgic_io_device *rd_dev = &vcpu->arch.vgic_cpu.rd_iodev; 770 struct vgic_redist_region *rdreg; 771 gpa_t rd_base; 772 int ret = 0; 773 774 lockdep_assert_held(&kvm->slots_lock); 775 mutex_lock(&kvm->arch.config_lock); 776 777 if (!IS_VGIC_ADDR_UNDEF(vgic_cpu->rd_iodev.base_addr)) 778 goto out_unlock; 779 780 /* 781 * We may be creating VCPUs before having set the base address for the 782 * redistributor region, in which case we will come back to this 783 * function for all VCPUs when the base address is set. Just return 784 * without doing any work for now. 785 */ 786 rdreg = vgic_v3_rdist_free_slot(&vgic->rd_regions); 787 if (!rdreg) 788 goto out_unlock; 789 790 if (!vgic_v3_check_base(kvm)) { 791 ret = -EINVAL; 792 goto out_unlock; 793 } 794 795 vgic_cpu->rdreg = rdreg; 796 vgic_cpu->rdreg_index = rdreg->free_index; 797 798 rd_base = rdreg->base + rdreg->free_index * KVM_VGIC_V3_REDIST_SIZE; 799 800 kvm_iodevice_init(&rd_dev->dev, &kvm_io_gic_ops); 801 rd_dev->base_addr = rd_base; 802 rd_dev->iodev_type = IODEV_REDIST; 803 rd_dev->regions = vgic_v3_rd_registers; 804 rd_dev->nr_regions = ARRAY_SIZE(vgic_v3_rd_registers); 805 rd_dev->redist_vcpu = vcpu; 806 807 mutex_unlock(&kvm->arch.config_lock); 808 809 ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, rd_base, 810 2 * SZ_64K, &rd_dev->dev); 811 if (ret) 812 return ret; 813 814 /* Protected by slots_lock */ 815 rdreg->free_index++; 816 return 0; 817 818 out_unlock: 819 mutex_unlock(&kvm->arch.config_lock); 820 return ret; 821 } 822 823 static void vgic_unregister_redist_iodev(struct kvm_vcpu *vcpu) 824 { 825 struct vgic_io_device *rd_dev = &vcpu->arch.vgic_cpu.rd_iodev; 826 827 kvm_io_bus_unregister_dev(vcpu->kvm, KVM_MMIO_BUS, &rd_dev->dev); 828 } 829 830 static int vgic_register_all_redist_iodevs(struct kvm *kvm) 831 { 832 struct kvm_vcpu *vcpu; 833 unsigned long c; 834 int ret = 0; 835 836 kvm_for_each_vcpu(c, vcpu, kvm) { 837 ret = vgic_register_redist_iodev(vcpu); 838 if (ret) 839 break; 840 } 841 842 if (ret) { 843 /* The current c failed, so iterate over the previous ones. */ 844 int i; 845 846 for (i = 0; i < c; i++) { 847 vcpu = kvm_get_vcpu(kvm, i); 848 vgic_unregister_redist_iodev(vcpu); 849 } 850 } 851 852 return ret; 853 } 854 855 /** 856 * vgic_v3_alloc_redist_region - Allocate a new redistributor region 857 * 858 * Performs various checks before inserting the rdist region in the list. 859 * Those tests depend on whether the size of the rdist region is known 860 * (ie. count != 0). The list is sorted by rdist region index. 861 * 862 * @kvm: kvm handle 863 * @index: redist region index 864 * @base: base of the new rdist region 865 * @count: number of redistributors the region is made of (0 in the old style 866 * single region, whose size is induced from the number of vcpus) 867 * 868 * Return 0 on success, < 0 otherwise 869 */ 870 static int vgic_v3_alloc_redist_region(struct kvm *kvm, uint32_t index, 871 gpa_t base, uint32_t count) 872 { 873 struct vgic_dist *d = &kvm->arch.vgic; 874 struct vgic_redist_region *rdreg; 875 struct list_head *rd_regions = &d->rd_regions; 876 int nr_vcpus = atomic_read(&kvm->online_vcpus); 877 size_t size = count ? count * KVM_VGIC_V3_REDIST_SIZE 878 : nr_vcpus * KVM_VGIC_V3_REDIST_SIZE; 879 int ret; 880 881 /* cross the end of memory ? */ 882 if (base + size < base) 883 return -EINVAL; 884 885 if (list_empty(rd_regions)) { 886 if (index != 0) 887 return -EINVAL; 888 } else { 889 rdreg = list_last_entry(rd_regions, 890 struct vgic_redist_region, list); 891 892 /* Don't mix single region and discrete redist regions */ 893 if (!count && rdreg->count) 894 return -EINVAL; 895 896 if (!count) 897 return -EEXIST; 898 899 if (index != rdreg->index + 1) 900 return -EINVAL; 901 } 902 903 /* 904 * For legacy single-region redistributor regions (!count), 905 * check that the redistributor region does not overlap with the 906 * distributor's address space. 907 */ 908 if (!count && !IS_VGIC_ADDR_UNDEF(d->vgic_dist_base) && 909 vgic_dist_overlap(kvm, base, size)) 910 return -EINVAL; 911 912 /* collision with any other rdist region? */ 913 if (vgic_v3_rdist_overlap(kvm, base, size)) 914 return -EINVAL; 915 916 rdreg = kzalloc(sizeof(*rdreg), GFP_KERNEL_ACCOUNT); 917 if (!rdreg) 918 return -ENOMEM; 919 920 rdreg->base = VGIC_ADDR_UNDEF; 921 922 ret = vgic_check_iorange(kvm, rdreg->base, base, SZ_64K, size); 923 if (ret) 924 goto free; 925 926 rdreg->base = base; 927 rdreg->count = count; 928 rdreg->free_index = 0; 929 rdreg->index = index; 930 931 list_add_tail(&rdreg->list, rd_regions); 932 return 0; 933 free: 934 kfree(rdreg); 935 return ret; 936 } 937 938 void vgic_v3_free_redist_region(struct vgic_redist_region *rdreg) 939 { 940 list_del(&rdreg->list); 941 kfree(rdreg); 942 } 943 944 int vgic_v3_set_redist_base(struct kvm *kvm, u32 index, u64 addr, u32 count) 945 { 946 int ret; 947 948 mutex_lock(&kvm->arch.config_lock); 949 ret = vgic_v3_alloc_redist_region(kvm, index, addr, count); 950 mutex_unlock(&kvm->arch.config_lock); 951 if (ret) 952 return ret; 953 954 /* 955 * Register iodevs for each existing VCPU. Adding more VCPUs 956 * afterwards will register the iodevs when needed. 957 */ 958 ret = vgic_register_all_redist_iodevs(kvm); 959 if (ret) { 960 struct vgic_redist_region *rdreg; 961 962 mutex_lock(&kvm->arch.config_lock); 963 rdreg = vgic_v3_rdist_region_from_index(kvm, index); 964 vgic_v3_free_redist_region(rdreg); 965 mutex_unlock(&kvm->arch.config_lock); 966 return ret; 967 } 968 969 return 0; 970 } 971 972 int vgic_v3_has_attr_regs(struct kvm_device *dev, struct kvm_device_attr *attr) 973 { 974 const struct vgic_register_region *region; 975 struct vgic_io_device iodev; 976 struct vgic_reg_attr reg_attr; 977 struct kvm_vcpu *vcpu; 978 gpa_t addr; 979 int ret; 980 981 ret = vgic_v3_parse_attr(dev, attr, ®_attr); 982 if (ret) 983 return ret; 984 985 vcpu = reg_attr.vcpu; 986 addr = reg_attr.addr; 987 988 switch (attr->group) { 989 case KVM_DEV_ARM_VGIC_GRP_DIST_REGS: 990 iodev.regions = vgic_v3_dist_registers; 991 iodev.nr_regions = ARRAY_SIZE(vgic_v3_dist_registers); 992 iodev.base_addr = 0; 993 break; 994 case KVM_DEV_ARM_VGIC_GRP_REDIST_REGS:{ 995 iodev.regions = vgic_v3_rd_registers; 996 iodev.nr_regions = ARRAY_SIZE(vgic_v3_rd_registers); 997 iodev.base_addr = 0; 998 break; 999 } 1000 case KVM_DEV_ARM_VGIC_GRP_CPU_SYSREGS: 1001 return vgic_v3_has_cpu_sysregs_attr(vcpu, attr); 1002 default: 1003 return -ENXIO; 1004 } 1005 1006 /* We only support aligned 32-bit accesses. */ 1007 if (addr & 3) 1008 return -ENXIO; 1009 1010 region = vgic_get_mmio_region(vcpu, &iodev, addr, sizeof(u32)); 1011 if (!region) 1012 return -ENXIO; 1013 1014 return 0; 1015 } 1016 1017 /* 1018 * The ICC_SGI* registers encode the affinity differently from the MPIDR, 1019 * so provide a wrapper to use the existing defines to isolate a certain 1020 * affinity level. 1021 */ 1022 #define SGI_AFFINITY_LEVEL(reg, level) \ 1023 ((((reg) & ICC_SGI1R_AFFINITY_## level ##_MASK) \ 1024 >> ICC_SGI1R_AFFINITY_## level ##_SHIFT) << MPIDR_LEVEL_SHIFT(level)) 1025 1026 static void vgic_v3_queue_sgi(struct kvm_vcpu *vcpu, u32 sgi, bool allow_group1) 1027 { 1028 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, sgi); 1029 unsigned long flags; 1030 1031 raw_spin_lock_irqsave(&irq->irq_lock, flags); 1032 1033 /* 1034 * An access targeting Group0 SGIs can only generate 1035 * those, while an access targeting Group1 SGIs can 1036 * generate interrupts of either group. 1037 */ 1038 if (!irq->group || allow_group1) { 1039 if (!irq->hw) { 1040 irq->pending_latch = true; 1041 vgic_queue_irq_unlock(vcpu->kvm, irq, flags); 1042 } else { 1043 /* HW SGI? Ask the GIC to inject it */ 1044 int err; 1045 err = irq_set_irqchip_state(irq->host_irq, 1046 IRQCHIP_STATE_PENDING, 1047 true); 1048 WARN_RATELIMIT(err, "IRQ %d", irq->host_irq); 1049 raw_spin_unlock_irqrestore(&irq->irq_lock, flags); 1050 } 1051 } else { 1052 raw_spin_unlock_irqrestore(&irq->irq_lock, flags); 1053 } 1054 1055 vgic_put_irq(vcpu->kvm, irq); 1056 } 1057 1058 /** 1059 * vgic_v3_dispatch_sgi - handle SGI requests from VCPUs 1060 * @vcpu: The VCPU requesting a SGI 1061 * @reg: The value written into ICC_{ASGI1,SGI0,SGI1}R by that VCPU 1062 * @allow_group1: Does the sysreg access allow generation of G1 SGIs 1063 * 1064 * With GICv3 (and ARE=1) CPUs trigger SGIs by writing to a system register. 1065 * This will trap in sys_regs.c and call this function. 1066 * This ICC_SGI1R_EL1 register contains the upper three affinity levels of the 1067 * target processors as well as a bitmask of 16 Aff0 CPUs. 1068 * 1069 * If the interrupt routing mode bit is not set, we iterate over the Aff0 1070 * bits and signal the VCPUs matching the provided Aff{3,2,1}. 1071 * 1072 * If this bit is set, we signal all, but not the calling VCPU. 1073 */ 1074 void vgic_v3_dispatch_sgi(struct kvm_vcpu *vcpu, u64 reg, bool allow_group1) 1075 { 1076 struct kvm *kvm = vcpu->kvm; 1077 struct kvm_vcpu *c_vcpu; 1078 unsigned long target_cpus; 1079 u64 mpidr; 1080 u32 sgi, aff0; 1081 unsigned long c; 1082 1083 sgi = FIELD_GET(ICC_SGI1R_SGI_ID_MASK, reg); 1084 1085 /* Broadcast */ 1086 if (unlikely(reg & BIT_ULL(ICC_SGI1R_IRQ_ROUTING_MODE_BIT))) { 1087 kvm_for_each_vcpu(c, c_vcpu, kvm) { 1088 /* Don't signal the calling VCPU */ 1089 if (c_vcpu == vcpu) 1090 continue; 1091 1092 vgic_v3_queue_sgi(c_vcpu, sgi, allow_group1); 1093 } 1094 1095 return; 1096 } 1097 1098 /* We iterate over affinities to find the corresponding vcpus */ 1099 mpidr = SGI_AFFINITY_LEVEL(reg, 3); 1100 mpidr |= SGI_AFFINITY_LEVEL(reg, 2); 1101 mpidr |= SGI_AFFINITY_LEVEL(reg, 1); 1102 target_cpus = FIELD_GET(ICC_SGI1R_TARGET_LIST_MASK, reg); 1103 1104 for_each_set_bit(aff0, &target_cpus, hweight_long(ICC_SGI1R_TARGET_LIST_MASK)) { 1105 c_vcpu = kvm_mpidr_to_vcpu(kvm, mpidr | aff0); 1106 if (c_vcpu) 1107 vgic_v3_queue_sgi(c_vcpu, sgi, allow_group1); 1108 } 1109 } 1110 1111 int vgic_v3_dist_uaccess(struct kvm_vcpu *vcpu, bool is_write, 1112 int offset, u32 *val) 1113 { 1114 struct vgic_io_device dev = { 1115 .regions = vgic_v3_dist_registers, 1116 .nr_regions = ARRAY_SIZE(vgic_v3_dist_registers), 1117 }; 1118 1119 return vgic_uaccess(vcpu, &dev, is_write, offset, val); 1120 } 1121 1122 int vgic_v3_redist_uaccess(struct kvm_vcpu *vcpu, bool is_write, 1123 int offset, u32 *val) 1124 { 1125 struct vgic_io_device rd_dev = { 1126 .regions = vgic_v3_rd_registers, 1127 .nr_regions = ARRAY_SIZE(vgic_v3_rd_registers), 1128 }; 1129 1130 return vgic_uaccess(vcpu, &rd_dev, is_write, offset, val); 1131 } 1132 1133 int vgic_v3_line_level_info_uaccess(struct kvm_vcpu *vcpu, bool is_write, 1134 u32 intid, u32 *val) 1135 { 1136 if (intid % 32) 1137 return -EINVAL; 1138 1139 if (is_write) 1140 vgic_write_irq_line_level_info(vcpu, intid, *val); 1141 else 1142 *val = vgic_read_irq_line_level_info(vcpu, intid); 1143 1144 return 0; 1145 } 1146