1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) 2019 Western Digital Corporation or its affiliates. 4 * 5 * Authors: 6 * Anup Patel <anup.patel@wdc.com> 7 */ 8 9 #include <linux/bitops.h> 10 #include <linux/errno.h> 11 #include <linux/err.h> 12 #include <linux/kdebug.h> 13 #include <linux/module.h> 14 #include <linux/percpu.h> 15 #include <linux/vmalloc.h> 16 #include <linux/sched/signal.h> 17 #include <linux/fs.h> 18 #include <linux/kvm_host.h> 19 #include <asm/cacheflush.h> 20 #include <asm/kvm_mmu.h> 21 #include <asm/kvm_nacl.h> 22 #include <asm/kvm_vcpu_vector.h> 23 24 #define CREATE_TRACE_POINTS 25 #include "trace.h" 26 27 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = { 28 KVM_GENERIC_VCPU_STATS(), 29 STATS_DESC_COUNTER(VCPU, ecall_exit_stat), 30 STATS_DESC_COUNTER(VCPU, wfi_exit_stat), 31 STATS_DESC_COUNTER(VCPU, wrs_exit_stat), 32 STATS_DESC_COUNTER(VCPU, mmio_exit_user), 33 STATS_DESC_COUNTER(VCPU, mmio_exit_kernel), 34 STATS_DESC_COUNTER(VCPU, csr_exit_user), 35 STATS_DESC_COUNTER(VCPU, csr_exit_kernel), 36 STATS_DESC_COUNTER(VCPU, signal_exits), 37 STATS_DESC_COUNTER(VCPU, exits), 38 STATS_DESC_COUNTER(VCPU, instr_illegal_exits), 39 STATS_DESC_COUNTER(VCPU, load_misaligned_exits), 40 STATS_DESC_COUNTER(VCPU, store_misaligned_exits), 41 STATS_DESC_COUNTER(VCPU, load_access_exits), 42 STATS_DESC_COUNTER(VCPU, store_access_exits), 43 }; 44 45 const struct kvm_stats_header kvm_vcpu_stats_header = { 46 .name_size = KVM_STATS_NAME_SIZE, 47 .num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc), 48 .id_offset = sizeof(struct kvm_stats_header), 49 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE, 50 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE + 51 sizeof(kvm_vcpu_stats_desc), 52 }; 53 54 static void kvm_riscv_vcpu_context_reset(struct kvm_vcpu *vcpu, 55 bool kvm_sbi_reset) 56 { 57 struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr; 58 struct kvm_cpu_context *cntx = &vcpu->arch.guest_context; 59 void *vector_datap = cntx->vector.datap; 60 61 memset(cntx, 0, sizeof(*cntx)); 62 memset(csr, 0, sizeof(*csr)); 63 memset(&vcpu->arch.smstateen_csr, 0, sizeof(vcpu->arch.smstateen_csr)); 64 65 /* Restore datap as it's not a part of the guest context. */ 66 cntx->vector.datap = vector_datap; 67 68 if (kvm_sbi_reset) 69 kvm_riscv_vcpu_sbi_load_reset_state(vcpu); 70 71 /* Setup reset state of shadow SSTATUS and HSTATUS CSRs */ 72 cntx->sstatus = SR_SPP | SR_SPIE; 73 74 cntx->hstatus |= HSTATUS_VTW; 75 cntx->hstatus |= HSTATUS_SPVP; 76 cntx->hstatus |= HSTATUS_SPV; 77 } 78 79 static void kvm_riscv_reset_vcpu(struct kvm_vcpu *vcpu, bool kvm_sbi_reset) 80 { 81 bool loaded; 82 83 /** 84 * The preemption should be disabled here because it races with 85 * kvm_sched_out/kvm_sched_in(called from preempt notifiers) which 86 * also calls vcpu_load/put. 87 */ 88 get_cpu(); 89 loaded = (vcpu->cpu != -1); 90 if (loaded) 91 kvm_arch_vcpu_put(vcpu); 92 93 vcpu->arch.last_exit_cpu = -1; 94 95 kvm_riscv_vcpu_context_reset(vcpu, kvm_sbi_reset); 96 97 kvm_riscv_vcpu_fp_reset(vcpu); 98 99 kvm_riscv_vcpu_vector_reset(vcpu); 100 101 kvm_riscv_vcpu_timer_reset(vcpu); 102 103 kvm_riscv_vcpu_aia_reset(vcpu); 104 105 bitmap_zero(vcpu->arch.irqs_pending, KVM_RISCV_VCPU_NR_IRQS); 106 bitmap_zero(vcpu->arch.irqs_pending_mask, KVM_RISCV_VCPU_NR_IRQS); 107 108 kvm_riscv_vcpu_pmu_reset(vcpu); 109 110 vcpu->arch.hfence_head = 0; 111 vcpu->arch.hfence_tail = 0; 112 memset(vcpu->arch.hfence_queue, 0, sizeof(vcpu->arch.hfence_queue)); 113 114 kvm_riscv_vcpu_sbi_reset(vcpu); 115 116 /* Reset the guest CSRs for hotplug usecase */ 117 if (loaded) 118 kvm_arch_vcpu_load(vcpu, smp_processor_id()); 119 put_cpu(); 120 } 121 122 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id) 123 { 124 return 0; 125 } 126 127 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu) 128 { 129 int rc; 130 131 spin_lock_init(&vcpu->arch.mp_state_lock); 132 133 /* Mark this VCPU never ran */ 134 vcpu->arch.ran_atleast_once = false; 135 136 vcpu->arch.cfg.hedeleg = KVM_HEDELEG_DEFAULT; 137 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO; 138 bitmap_zero(vcpu->arch.isa, RISCV_ISA_EXT_MAX); 139 140 /* Setup ISA features available to VCPU */ 141 kvm_riscv_vcpu_setup_isa(vcpu); 142 143 /* Setup vendor, arch, and implementation details */ 144 vcpu->arch.mvendorid = sbi_get_mvendorid(); 145 vcpu->arch.marchid = sbi_get_marchid(); 146 vcpu->arch.mimpid = sbi_get_mimpid(); 147 148 /* Setup VCPU hfence queue */ 149 spin_lock_init(&vcpu->arch.hfence_lock); 150 151 spin_lock_init(&vcpu->arch.reset_state.lock); 152 153 rc = kvm_riscv_vcpu_alloc_vector_context(vcpu); 154 if (rc) 155 return rc; 156 157 /* Setup VCPU timer */ 158 kvm_riscv_vcpu_timer_init(vcpu); 159 160 /* setup performance monitoring */ 161 kvm_riscv_vcpu_pmu_init(vcpu); 162 163 /* Setup VCPU AIA */ 164 kvm_riscv_vcpu_aia_init(vcpu); 165 166 /* 167 * Setup SBI extensions 168 * NOTE: This must be the last thing to be initialized. 169 */ 170 kvm_riscv_vcpu_sbi_init(vcpu); 171 172 /* Reset VCPU */ 173 kvm_riscv_reset_vcpu(vcpu, false); 174 175 return 0; 176 } 177 178 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) 179 { 180 /** 181 * vcpu with id 0 is the designated boot cpu. 182 * Keep all vcpus with non-zero id in power-off state so that 183 * they can be brought up using SBI HSM extension. 184 */ 185 if (vcpu->vcpu_idx != 0) 186 kvm_riscv_vcpu_power_off(vcpu); 187 } 188 189 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) 190 { 191 kvm_riscv_vcpu_sbi_deinit(vcpu); 192 193 /* Cleanup VCPU AIA context */ 194 kvm_riscv_vcpu_aia_deinit(vcpu); 195 196 /* Cleanup VCPU timer */ 197 kvm_riscv_vcpu_timer_deinit(vcpu); 198 199 kvm_riscv_vcpu_pmu_deinit(vcpu); 200 201 /* Free unused pages pre-allocated for G-stage page table mappings */ 202 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache); 203 204 /* Free vector context space for host and guest kernel */ 205 kvm_riscv_vcpu_free_vector_context(vcpu); 206 } 207 208 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) 209 { 210 return kvm_riscv_vcpu_timer_pending(vcpu); 211 } 212 213 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu) 214 { 215 return (kvm_riscv_vcpu_has_interrupts(vcpu, -1UL) && 216 !kvm_riscv_vcpu_stopped(vcpu) && !vcpu->arch.pause); 217 } 218 219 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) 220 { 221 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; 222 } 223 224 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu) 225 { 226 return (vcpu->arch.guest_context.sstatus & SR_SPP) ? true : false; 227 } 228 229 #ifdef CONFIG_GUEST_PERF_EVENTS 230 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu) 231 { 232 return vcpu->arch.guest_context.sepc; 233 } 234 #endif 235 236 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) 237 { 238 return VM_FAULT_SIGBUS; 239 } 240 241 long kvm_arch_vcpu_async_ioctl(struct file *filp, 242 unsigned int ioctl, unsigned long arg) 243 { 244 struct kvm_vcpu *vcpu = filp->private_data; 245 void __user *argp = (void __user *)arg; 246 247 if (ioctl == KVM_INTERRUPT) { 248 struct kvm_interrupt irq; 249 250 if (copy_from_user(&irq, argp, sizeof(irq))) 251 return -EFAULT; 252 253 if (irq.irq == KVM_INTERRUPT_SET) 254 return kvm_riscv_vcpu_set_interrupt(vcpu, IRQ_VS_EXT); 255 else 256 return kvm_riscv_vcpu_unset_interrupt(vcpu, IRQ_VS_EXT); 257 } 258 259 return -ENOIOCTLCMD; 260 } 261 262 long kvm_arch_vcpu_ioctl(struct file *filp, 263 unsigned int ioctl, unsigned long arg) 264 { 265 struct kvm_vcpu *vcpu = filp->private_data; 266 void __user *argp = (void __user *)arg; 267 long r = -EINVAL; 268 269 switch (ioctl) { 270 case KVM_SET_ONE_REG: 271 case KVM_GET_ONE_REG: { 272 struct kvm_one_reg reg; 273 274 r = -EFAULT; 275 if (copy_from_user(®, argp, sizeof(reg))) 276 break; 277 278 if (ioctl == KVM_SET_ONE_REG) 279 r = kvm_riscv_vcpu_set_reg(vcpu, ®); 280 else 281 r = kvm_riscv_vcpu_get_reg(vcpu, ®); 282 break; 283 } 284 case KVM_GET_REG_LIST: { 285 struct kvm_reg_list __user *user_list = argp; 286 struct kvm_reg_list reg_list; 287 unsigned int n; 288 289 r = -EFAULT; 290 if (copy_from_user(®_list, user_list, sizeof(reg_list))) 291 break; 292 n = reg_list.n; 293 reg_list.n = kvm_riscv_vcpu_num_regs(vcpu); 294 if (copy_to_user(user_list, ®_list, sizeof(reg_list))) 295 break; 296 r = -E2BIG; 297 if (n < reg_list.n) 298 break; 299 r = kvm_riscv_vcpu_copy_reg_indices(vcpu, user_list->reg); 300 break; 301 } 302 default: 303 break; 304 } 305 306 return r; 307 } 308 309 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu, 310 struct kvm_sregs *sregs) 311 { 312 return -EINVAL; 313 } 314 315 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu, 316 struct kvm_sregs *sregs) 317 { 318 return -EINVAL; 319 } 320 321 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) 322 { 323 return -EINVAL; 324 } 325 326 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) 327 { 328 return -EINVAL; 329 } 330 331 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu, 332 struct kvm_translation *tr) 333 { 334 return -EINVAL; 335 } 336 337 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) 338 { 339 return -EINVAL; 340 } 341 342 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) 343 { 344 return -EINVAL; 345 } 346 347 void kvm_riscv_vcpu_flush_interrupts(struct kvm_vcpu *vcpu) 348 { 349 struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr; 350 unsigned long mask, val; 351 352 if (READ_ONCE(vcpu->arch.irqs_pending_mask[0])) { 353 mask = xchg_acquire(&vcpu->arch.irqs_pending_mask[0], 0); 354 val = READ_ONCE(vcpu->arch.irqs_pending[0]) & mask; 355 356 csr->hvip &= ~mask; 357 csr->hvip |= val; 358 } 359 360 /* Flush AIA high interrupts */ 361 kvm_riscv_vcpu_aia_flush_interrupts(vcpu); 362 } 363 364 void kvm_riscv_vcpu_sync_interrupts(struct kvm_vcpu *vcpu) 365 { 366 unsigned long hvip; 367 struct kvm_vcpu_arch *v = &vcpu->arch; 368 struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr; 369 370 /* Read current HVIP and VSIE CSRs */ 371 csr->vsie = ncsr_read(CSR_VSIE); 372 373 /* Sync-up HVIP.VSSIP bit changes does by Guest */ 374 hvip = ncsr_read(CSR_HVIP); 375 if ((csr->hvip ^ hvip) & (1UL << IRQ_VS_SOFT)) { 376 if (hvip & (1UL << IRQ_VS_SOFT)) { 377 if (!test_and_set_bit(IRQ_VS_SOFT, 378 v->irqs_pending_mask)) 379 set_bit(IRQ_VS_SOFT, v->irqs_pending); 380 } else { 381 if (!test_and_set_bit(IRQ_VS_SOFT, 382 v->irqs_pending_mask)) 383 clear_bit(IRQ_VS_SOFT, v->irqs_pending); 384 } 385 } 386 387 /* Sync up the HVIP.LCOFIP bit changes (only clear) by the guest */ 388 if ((csr->hvip ^ hvip) & (1UL << IRQ_PMU_OVF)) { 389 if (!(hvip & (1UL << IRQ_PMU_OVF)) && 390 !test_and_set_bit(IRQ_PMU_OVF, v->irqs_pending_mask)) 391 clear_bit(IRQ_PMU_OVF, v->irqs_pending); 392 } 393 394 /* Sync-up AIA high interrupts */ 395 kvm_riscv_vcpu_aia_sync_interrupts(vcpu); 396 397 /* Sync-up timer CSRs */ 398 kvm_riscv_vcpu_timer_sync(vcpu); 399 } 400 401 int kvm_riscv_vcpu_set_interrupt(struct kvm_vcpu *vcpu, unsigned int irq) 402 { 403 /* 404 * We only allow VS-mode software, timer, and external 405 * interrupts when irq is one of the local interrupts 406 * defined by RISC-V privilege specification. 407 */ 408 if (irq < IRQ_LOCAL_MAX && 409 irq != IRQ_VS_SOFT && 410 irq != IRQ_VS_TIMER && 411 irq != IRQ_VS_EXT && 412 irq != IRQ_PMU_OVF) 413 return -EINVAL; 414 415 set_bit(irq, vcpu->arch.irqs_pending); 416 smp_mb__before_atomic(); 417 set_bit(irq, vcpu->arch.irqs_pending_mask); 418 419 kvm_vcpu_kick(vcpu); 420 421 return 0; 422 } 423 424 int kvm_riscv_vcpu_unset_interrupt(struct kvm_vcpu *vcpu, unsigned int irq) 425 { 426 /* 427 * We only allow VS-mode software, timer, counter overflow and external 428 * interrupts when irq is one of the local interrupts 429 * defined by RISC-V privilege specification. 430 */ 431 if (irq < IRQ_LOCAL_MAX && 432 irq != IRQ_VS_SOFT && 433 irq != IRQ_VS_TIMER && 434 irq != IRQ_VS_EXT && 435 irq != IRQ_PMU_OVF) 436 return -EINVAL; 437 438 clear_bit(irq, vcpu->arch.irqs_pending); 439 smp_mb__before_atomic(); 440 set_bit(irq, vcpu->arch.irqs_pending_mask); 441 442 return 0; 443 } 444 445 bool kvm_riscv_vcpu_has_interrupts(struct kvm_vcpu *vcpu, u64 mask) 446 { 447 unsigned long ie; 448 449 ie = ((vcpu->arch.guest_csr.vsie & VSIP_VALID_MASK) 450 << VSIP_TO_HVIP_SHIFT) & (unsigned long)mask; 451 ie |= vcpu->arch.guest_csr.vsie & ~IRQ_LOCAL_MASK & 452 (unsigned long)mask; 453 if (READ_ONCE(vcpu->arch.irqs_pending[0]) & ie) 454 return true; 455 456 /* Check AIA high interrupts */ 457 return kvm_riscv_vcpu_aia_has_interrupts(vcpu, mask); 458 } 459 460 void __kvm_riscv_vcpu_power_off(struct kvm_vcpu *vcpu) 461 { 462 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_STOPPED); 463 kvm_make_request(KVM_REQ_SLEEP, vcpu); 464 kvm_vcpu_kick(vcpu); 465 } 466 467 void kvm_riscv_vcpu_power_off(struct kvm_vcpu *vcpu) 468 { 469 spin_lock(&vcpu->arch.mp_state_lock); 470 __kvm_riscv_vcpu_power_off(vcpu); 471 spin_unlock(&vcpu->arch.mp_state_lock); 472 } 473 474 void __kvm_riscv_vcpu_power_on(struct kvm_vcpu *vcpu) 475 { 476 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_RUNNABLE); 477 kvm_vcpu_wake_up(vcpu); 478 } 479 480 void kvm_riscv_vcpu_power_on(struct kvm_vcpu *vcpu) 481 { 482 spin_lock(&vcpu->arch.mp_state_lock); 483 __kvm_riscv_vcpu_power_on(vcpu); 484 spin_unlock(&vcpu->arch.mp_state_lock); 485 } 486 487 bool kvm_riscv_vcpu_stopped(struct kvm_vcpu *vcpu) 488 { 489 return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_STOPPED; 490 } 491 492 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, 493 struct kvm_mp_state *mp_state) 494 { 495 *mp_state = READ_ONCE(vcpu->arch.mp_state); 496 497 return 0; 498 } 499 500 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, 501 struct kvm_mp_state *mp_state) 502 { 503 int ret = 0; 504 505 spin_lock(&vcpu->arch.mp_state_lock); 506 507 switch (mp_state->mp_state) { 508 case KVM_MP_STATE_RUNNABLE: 509 WRITE_ONCE(vcpu->arch.mp_state, *mp_state); 510 break; 511 case KVM_MP_STATE_STOPPED: 512 __kvm_riscv_vcpu_power_off(vcpu); 513 break; 514 case KVM_MP_STATE_INIT_RECEIVED: 515 if (vcpu->kvm->arch.mp_state_reset) 516 kvm_riscv_reset_vcpu(vcpu, false); 517 else 518 ret = -EINVAL; 519 break; 520 default: 521 ret = -EINVAL; 522 } 523 524 spin_unlock(&vcpu->arch.mp_state_lock); 525 526 return ret; 527 } 528 529 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu, 530 struct kvm_guest_debug *dbg) 531 { 532 if (dbg->control & KVM_GUESTDBG_ENABLE) { 533 vcpu->guest_debug = dbg->control; 534 vcpu->arch.cfg.hedeleg &= ~BIT(EXC_BREAKPOINT); 535 } else { 536 vcpu->guest_debug = 0; 537 vcpu->arch.cfg.hedeleg |= BIT(EXC_BREAKPOINT); 538 } 539 540 return 0; 541 } 542 543 static void kvm_riscv_vcpu_setup_config(struct kvm_vcpu *vcpu) 544 { 545 const unsigned long *isa = vcpu->arch.isa; 546 struct kvm_vcpu_config *cfg = &vcpu->arch.cfg; 547 548 if (riscv_isa_extension_available(isa, SVPBMT)) 549 cfg->henvcfg |= ENVCFG_PBMTE; 550 551 if (riscv_isa_extension_available(isa, SSTC)) 552 cfg->henvcfg |= ENVCFG_STCE; 553 554 if (riscv_isa_extension_available(isa, ZICBOM)) 555 cfg->henvcfg |= (ENVCFG_CBIE | ENVCFG_CBCFE); 556 557 if (riscv_isa_extension_available(isa, ZICBOZ)) 558 cfg->henvcfg |= ENVCFG_CBZE; 559 560 if (riscv_isa_extension_available(isa, SVADU) && 561 !riscv_isa_extension_available(isa, SVADE)) 562 cfg->henvcfg |= ENVCFG_ADUE; 563 564 if (riscv_has_extension_unlikely(RISCV_ISA_EXT_SMSTATEEN)) { 565 cfg->hstateen0 |= SMSTATEEN0_HSENVCFG; 566 if (riscv_isa_extension_available(isa, SSAIA)) 567 cfg->hstateen0 |= SMSTATEEN0_AIA_IMSIC | 568 SMSTATEEN0_AIA | 569 SMSTATEEN0_AIA_ISEL; 570 if (riscv_isa_extension_available(isa, SMSTATEEN)) 571 cfg->hstateen0 |= SMSTATEEN0_SSTATEEN0; 572 } 573 574 if (vcpu->guest_debug) 575 cfg->hedeleg &= ~BIT(EXC_BREAKPOINT); 576 } 577 578 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) 579 { 580 void *nsh; 581 struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr; 582 struct kvm_vcpu_config *cfg = &vcpu->arch.cfg; 583 584 if (kvm_riscv_nacl_sync_csr_available()) { 585 nsh = nacl_shmem(); 586 nacl_csr_write(nsh, CSR_VSSTATUS, csr->vsstatus); 587 nacl_csr_write(nsh, CSR_VSIE, csr->vsie); 588 nacl_csr_write(nsh, CSR_VSTVEC, csr->vstvec); 589 nacl_csr_write(nsh, CSR_VSSCRATCH, csr->vsscratch); 590 nacl_csr_write(nsh, CSR_VSEPC, csr->vsepc); 591 nacl_csr_write(nsh, CSR_VSCAUSE, csr->vscause); 592 nacl_csr_write(nsh, CSR_VSTVAL, csr->vstval); 593 nacl_csr_write(nsh, CSR_HEDELEG, cfg->hedeleg); 594 nacl_csr_write(nsh, CSR_HVIP, csr->hvip); 595 nacl_csr_write(nsh, CSR_VSATP, csr->vsatp); 596 nacl_csr_write(nsh, CSR_HENVCFG, cfg->henvcfg); 597 if (IS_ENABLED(CONFIG_32BIT)) 598 nacl_csr_write(nsh, CSR_HENVCFGH, cfg->henvcfg >> 32); 599 if (riscv_has_extension_unlikely(RISCV_ISA_EXT_SMSTATEEN)) { 600 nacl_csr_write(nsh, CSR_HSTATEEN0, cfg->hstateen0); 601 if (IS_ENABLED(CONFIG_32BIT)) 602 nacl_csr_write(nsh, CSR_HSTATEEN0H, cfg->hstateen0 >> 32); 603 } 604 } else { 605 csr_write(CSR_VSSTATUS, csr->vsstatus); 606 csr_write(CSR_VSIE, csr->vsie); 607 csr_write(CSR_VSTVEC, csr->vstvec); 608 csr_write(CSR_VSSCRATCH, csr->vsscratch); 609 csr_write(CSR_VSEPC, csr->vsepc); 610 csr_write(CSR_VSCAUSE, csr->vscause); 611 csr_write(CSR_VSTVAL, csr->vstval); 612 csr_write(CSR_HEDELEG, cfg->hedeleg); 613 csr_write(CSR_HVIP, csr->hvip); 614 csr_write(CSR_VSATP, csr->vsatp); 615 csr_write(CSR_HENVCFG, cfg->henvcfg); 616 if (IS_ENABLED(CONFIG_32BIT)) 617 csr_write(CSR_HENVCFGH, cfg->henvcfg >> 32); 618 if (riscv_has_extension_unlikely(RISCV_ISA_EXT_SMSTATEEN)) { 619 csr_write(CSR_HSTATEEN0, cfg->hstateen0); 620 if (IS_ENABLED(CONFIG_32BIT)) 621 csr_write(CSR_HSTATEEN0H, cfg->hstateen0 >> 32); 622 } 623 } 624 625 kvm_riscv_mmu_update_hgatp(vcpu); 626 627 kvm_riscv_vcpu_timer_restore(vcpu); 628 629 kvm_riscv_vcpu_host_fp_save(&vcpu->arch.host_context); 630 kvm_riscv_vcpu_guest_fp_restore(&vcpu->arch.guest_context, 631 vcpu->arch.isa); 632 kvm_riscv_vcpu_host_vector_save(&vcpu->arch.host_context); 633 kvm_riscv_vcpu_guest_vector_restore(&vcpu->arch.guest_context, 634 vcpu->arch.isa); 635 636 kvm_riscv_vcpu_aia_load(vcpu, cpu); 637 638 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu); 639 640 vcpu->cpu = cpu; 641 } 642 643 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) 644 { 645 void *nsh; 646 struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr; 647 648 vcpu->cpu = -1; 649 650 kvm_riscv_vcpu_aia_put(vcpu); 651 652 kvm_riscv_vcpu_guest_fp_save(&vcpu->arch.guest_context, 653 vcpu->arch.isa); 654 kvm_riscv_vcpu_host_fp_restore(&vcpu->arch.host_context); 655 656 kvm_riscv_vcpu_timer_save(vcpu); 657 kvm_riscv_vcpu_guest_vector_save(&vcpu->arch.guest_context, 658 vcpu->arch.isa); 659 kvm_riscv_vcpu_host_vector_restore(&vcpu->arch.host_context); 660 661 if (kvm_riscv_nacl_available()) { 662 nsh = nacl_shmem(); 663 csr->vsstatus = nacl_csr_read(nsh, CSR_VSSTATUS); 664 csr->vsie = nacl_csr_read(nsh, CSR_VSIE); 665 csr->vstvec = nacl_csr_read(nsh, CSR_VSTVEC); 666 csr->vsscratch = nacl_csr_read(nsh, CSR_VSSCRATCH); 667 csr->vsepc = nacl_csr_read(nsh, CSR_VSEPC); 668 csr->vscause = nacl_csr_read(nsh, CSR_VSCAUSE); 669 csr->vstval = nacl_csr_read(nsh, CSR_VSTVAL); 670 csr->hvip = nacl_csr_read(nsh, CSR_HVIP); 671 csr->vsatp = nacl_csr_read(nsh, CSR_VSATP); 672 } else { 673 csr->vsstatus = csr_read(CSR_VSSTATUS); 674 csr->vsie = csr_read(CSR_VSIE); 675 csr->vstvec = csr_read(CSR_VSTVEC); 676 csr->vsscratch = csr_read(CSR_VSSCRATCH); 677 csr->vsepc = csr_read(CSR_VSEPC); 678 csr->vscause = csr_read(CSR_VSCAUSE); 679 csr->vstval = csr_read(CSR_VSTVAL); 680 csr->hvip = csr_read(CSR_HVIP); 681 csr->vsatp = csr_read(CSR_VSATP); 682 } 683 } 684 685 /** 686 * kvm_riscv_check_vcpu_requests - check and handle pending vCPU requests 687 * @vcpu: the VCPU pointer 688 * 689 * Return: 1 if we should enter the guest 690 * 0 if we should exit to userspace 691 */ 692 static int kvm_riscv_check_vcpu_requests(struct kvm_vcpu *vcpu) 693 { 694 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu); 695 696 if (kvm_request_pending(vcpu)) { 697 if (kvm_check_request(KVM_REQ_SLEEP, vcpu)) { 698 kvm_vcpu_srcu_read_unlock(vcpu); 699 rcuwait_wait_event(wait, 700 (!kvm_riscv_vcpu_stopped(vcpu)) && (!vcpu->arch.pause), 701 TASK_INTERRUPTIBLE); 702 kvm_vcpu_srcu_read_lock(vcpu); 703 704 if (kvm_riscv_vcpu_stopped(vcpu) || vcpu->arch.pause) { 705 /* 706 * Awaken to handle a signal, request to 707 * sleep again later. 708 */ 709 kvm_make_request(KVM_REQ_SLEEP, vcpu); 710 } 711 } 712 713 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu)) 714 kvm_riscv_reset_vcpu(vcpu, true); 715 716 if (kvm_check_request(KVM_REQ_UPDATE_HGATP, vcpu)) 717 kvm_riscv_mmu_update_hgatp(vcpu); 718 719 if (kvm_check_request(KVM_REQ_FENCE_I, vcpu)) 720 kvm_riscv_fence_i_process(vcpu); 721 722 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) 723 kvm_riscv_tlb_flush_process(vcpu); 724 725 if (kvm_check_request(KVM_REQ_HFENCE_VVMA_ALL, vcpu)) 726 kvm_riscv_hfence_vvma_all_process(vcpu); 727 728 if (kvm_check_request(KVM_REQ_HFENCE, vcpu)) 729 kvm_riscv_hfence_process(vcpu); 730 731 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu)) 732 kvm_riscv_vcpu_record_steal_time(vcpu); 733 734 if (kvm_dirty_ring_check_request(vcpu)) 735 return 0; 736 } 737 738 return 1; 739 } 740 741 static void kvm_riscv_update_hvip(struct kvm_vcpu *vcpu) 742 { 743 struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr; 744 745 ncsr_write(CSR_HVIP, csr->hvip); 746 kvm_riscv_vcpu_aia_update_hvip(vcpu); 747 } 748 749 static __always_inline void kvm_riscv_vcpu_swap_in_guest_state(struct kvm_vcpu *vcpu) 750 { 751 struct kvm_vcpu_smstateen_csr *smcsr = &vcpu->arch.smstateen_csr; 752 struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr; 753 struct kvm_vcpu_config *cfg = &vcpu->arch.cfg; 754 755 vcpu->arch.host_scounteren = csr_swap(CSR_SCOUNTEREN, csr->scounteren); 756 vcpu->arch.host_senvcfg = csr_swap(CSR_SENVCFG, csr->senvcfg); 757 if (riscv_has_extension_unlikely(RISCV_ISA_EXT_SMSTATEEN) && 758 (cfg->hstateen0 & SMSTATEEN0_SSTATEEN0)) 759 vcpu->arch.host_sstateen0 = csr_swap(CSR_SSTATEEN0, 760 smcsr->sstateen0); 761 } 762 763 static __always_inline void kvm_riscv_vcpu_swap_in_host_state(struct kvm_vcpu *vcpu) 764 { 765 struct kvm_vcpu_smstateen_csr *smcsr = &vcpu->arch.smstateen_csr; 766 struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr; 767 struct kvm_vcpu_config *cfg = &vcpu->arch.cfg; 768 769 csr->scounteren = csr_swap(CSR_SCOUNTEREN, vcpu->arch.host_scounteren); 770 csr->senvcfg = csr_swap(CSR_SENVCFG, vcpu->arch.host_senvcfg); 771 if (riscv_has_extension_unlikely(RISCV_ISA_EXT_SMSTATEEN) && 772 (cfg->hstateen0 & SMSTATEEN0_SSTATEEN0)) 773 smcsr->sstateen0 = csr_swap(CSR_SSTATEEN0, 774 vcpu->arch.host_sstateen0); 775 } 776 777 /* 778 * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while 779 * the vCPU is running. 780 * 781 * This must be noinstr as instrumentation may make use of RCU, and this is not 782 * safe during the EQS. 783 */ 784 static void noinstr kvm_riscv_vcpu_enter_exit(struct kvm_vcpu *vcpu, 785 struct kvm_cpu_trap *trap) 786 { 787 void *nsh; 788 struct kvm_cpu_context *gcntx = &vcpu->arch.guest_context; 789 struct kvm_cpu_context *hcntx = &vcpu->arch.host_context; 790 791 /* 792 * We save trap CSRs (such as SEPC, SCAUSE, STVAL, HTVAL, and 793 * HTINST) here because we do local_irq_enable() after this 794 * function in kvm_arch_vcpu_ioctl_run() which can result in 795 * an interrupt immediately after local_irq_enable() and can 796 * potentially change trap CSRs. 797 */ 798 799 kvm_riscv_vcpu_swap_in_guest_state(vcpu); 800 guest_state_enter_irqoff(); 801 802 if (kvm_riscv_nacl_sync_sret_available()) { 803 nsh = nacl_shmem(); 804 805 if (kvm_riscv_nacl_autoswap_csr_available()) { 806 hcntx->hstatus = 807 nacl_csr_read(nsh, CSR_HSTATUS); 808 nacl_scratch_write_long(nsh, 809 SBI_NACL_SHMEM_AUTOSWAP_OFFSET + 810 SBI_NACL_SHMEM_AUTOSWAP_HSTATUS, 811 gcntx->hstatus); 812 nacl_scratch_write_long(nsh, 813 SBI_NACL_SHMEM_AUTOSWAP_OFFSET, 814 SBI_NACL_SHMEM_AUTOSWAP_FLAG_HSTATUS); 815 } else if (kvm_riscv_nacl_sync_csr_available()) { 816 hcntx->hstatus = nacl_csr_swap(nsh, 817 CSR_HSTATUS, gcntx->hstatus); 818 } else { 819 hcntx->hstatus = csr_swap(CSR_HSTATUS, gcntx->hstatus); 820 } 821 822 nacl_scratch_write_longs(nsh, 823 SBI_NACL_SHMEM_SRET_OFFSET + 824 SBI_NACL_SHMEM_SRET_X(1), 825 &gcntx->ra, 826 SBI_NACL_SHMEM_SRET_X_LAST); 827 828 __kvm_riscv_nacl_switch_to(&vcpu->arch, SBI_EXT_NACL, 829 SBI_EXT_NACL_SYNC_SRET); 830 831 if (kvm_riscv_nacl_autoswap_csr_available()) { 832 nacl_scratch_write_long(nsh, 833 SBI_NACL_SHMEM_AUTOSWAP_OFFSET, 834 0); 835 gcntx->hstatus = nacl_scratch_read_long(nsh, 836 SBI_NACL_SHMEM_AUTOSWAP_OFFSET + 837 SBI_NACL_SHMEM_AUTOSWAP_HSTATUS); 838 } else { 839 gcntx->hstatus = csr_swap(CSR_HSTATUS, hcntx->hstatus); 840 } 841 842 trap->htval = nacl_csr_read(nsh, CSR_HTVAL); 843 trap->htinst = nacl_csr_read(nsh, CSR_HTINST); 844 } else { 845 hcntx->hstatus = csr_swap(CSR_HSTATUS, gcntx->hstatus); 846 847 __kvm_riscv_switch_to(&vcpu->arch); 848 849 gcntx->hstatus = csr_swap(CSR_HSTATUS, hcntx->hstatus); 850 851 trap->htval = csr_read(CSR_HTVAL); 852 trap->htinst = csr_read(CSR_HTINST); 853 } 854 855 trap->sepc = gcntx->sepc; 856 trap->scause = csr_read(CSR_SCAUSE); 857 trap->stval = csr_read(CSR_STVAL); 858 859 vcpu->arch.last_exit_cpu = vcpu->cpu; 860 guest_state_exit_irqoff(); 861 kvm_riscv_vcpu_swap_in_host_state(vcpu); 862 } 863 864 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu) 865 { 866 int ret; 867 struct kvm_cpu_trap trap; 868 struct kvm_run *run = vcpu->run; 869 870 if (!vcpu->arch.ran_atleast_once) 871 kvm_riscv_vcpu_setup_config(vcpu); 872 873 /* Mark this VCPU ran at least once */ 874 vcpu->arch.ran_atleast_once = true; 875 876 kvm_vcpu_srcu_read_lock(vcpu); 877 878 switch (run->exit_reason) { 879 case KVM_EXIT_MMIO: 880 /* Process MMIO value returned from user-space */ 881 ret = kvm_riscv_vcpu_mmio_return(vcpu, vcpu->run); 882 break; 883 case KVM_EXIT_RISCV_SBI: 884 /* Process SBI value returned from user-space */ 885 ret = kvm_riscv_vcpu_sbi_return(vcpu, vcpu->run); 886 break; 887 case KVM_EXIT_RISCV_CSR: 888 /* Process CSR value returned from user-space */ 889 ret = kvm_riscv_vcpu_csr_return(vcpu, vcpu->run); 890 break; 891 default: 892 ret = 0; 893 break; 894 } 895 if (ret) { 896 kvm_vcpu_srcu_read_unlock(vcpu); 897 return ret; 898 } 899 900 if (!vcpu->wants_to_run) { 901 kvm_vcpu_srcu_read_unlock(vcpu); 902 return -EINTR; 903 } 904 905 vcpu_load(vcpu); 906 907 kvm_sigset_activate(vcpu); 908 909 ret = 1; 910 run->exit_reason = KVM_EXIT_UNKNOWN; 911 while (ret > 0) { 912 /* Check conditions before entering the guest */ 913 ret = kvm_xfer_to_guest_mode_handle_work(vcpu); 914 if (ret) 915 continue; 916 ret = 1; 917 918 kvm_riscv_gstage_vmid_update(vcpu); 919 920 ret = kvm_riscv_check_vcpu_requests(vcpu); 921 if (ret <= 0) 922 continue; 923 924 preempt_disable(); 925 926 /* Update AIA HW state before entering guest */ 927 ret = kvm_riscv_vcpu_aia_update(vcpu); 928 if (ret <= 0) { 929 preempt_enable(); 930 continue; 931 } 932 933 local_irq_disable(); 934 935 /* 936 * Ensure we set mode to IN_GUEST_MODE after we disable 937 * interrupts and before the final VCPU requests check. 938 * See the comment in kvm_vcpu_exiting_guest_mode() and 939 * Documentation/virt/kvm/vcpu-requests.rst 940 */ 941 vcpu->mode = IN_GUEST_MODE; 942 943 kvm_vcpu_srcu_read_unlock(vcpu); 944 smp_mb__after_srcu_read_unlock(); 945 946 /* 947 * We might have got VCPU interrupts updated asynchronously 948 * so update it in HW. 949 */ 950 kvm_riscv_vcpu_flush_interrupts(vcpu); 951 952 /* Update HVIP CSR for current CPU */ 953 kvm_riscv_update_hvip(vcpu); 954 955 if (kvm_riscv_gstage_vmid_ver_changed(&vcpu->kvm->arch.vmid) || 956 kvm_request_pending(vcpu) || 957 xfer_to_guest_mode_work_pending()) { 958 vcpu->mode = OUTSIDE_GUEST_MODE; 959 local_irq_enable(); 960 preempt_enable(); 961 kvm_vcpu_srcu_read_lock(vcpu); 962 continue; 963 } 964 965 /* 966 * Sanitize VMID mappings cached (TLB) on current CPU 967 * 968 * Note: This should be done after G-stage VMID has been 969 * updated using kvm_riscv_gstage_vmid_ver_changed() 970 */ 971 kvm_riscv_gstage_vmid_sanitize(vcpu); 972 973 trace_kvm_entry(vcpu); 974 975 guest_timing_enter_irqoff(); 976 977 kvm_riscv_vcpu_enter_exit(vcpu, &trap); 978 979 vcpu->mode = OUTSIDE_GUEST_MODE; 980 vcpu->stat.exits++; 981 982 /* Syncup interrupts state with HW */ 983 kvm_riscv_vcpu_sync_interrupts(vcpu); 984 985 /* 986 * We must ensure that any pending interrupts are taken before 987 * we exit guest timing so that timer ticks are accounted as 988 * guest time. Transiently unmask interrupts so that any 989 * pending interrupts are taken. 990 * 991 * There's no barrier which ensures that pending interrupts are 992 * recognised, so we just hope that the CPU takes any pending 993 * interrupts between the enable and disable. 994 */ 995 local_irq_enable(); 996 local_irq_disable(); 997 998 guest_timing_exit_irqoff(); 999 1000 local_irq_enable(); 1001 1002 trace_kvm_exit(&trap); 1003 1004 preempt_enable(); 1005 1006 kvm_vcpu_srcu_read_lock(vcpu); 1007 1008 ret = kvm_riscv_vcpu_exit(vcpu, run, &trap); 1009 } 1010 1011 kvm_sigset_deactivate(vcpu); 1012 1013 vcpu_put(vcpu); 1014 1015 kvm_vcpu_srcu_read_unlock(vcpu); 1016 1017 return ret; 1018 } 1019