| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * in-kernel handling for sie intercepts |
| 4 | * |
| 5 | * Copyright IBM Corp. 2008, 2020 |
| 6 | * |
| 7 | * Author(s): Carsten Otte <cotte@de.ibm.com> |
| 8 | * Christian Borntraeger <borntraeger@de.ibm.com> |
| 9 | */ |
| 10 | |
| 11 | #include <linux/kvm_host.h> |
| 12 | #include <linux/errno.h> |
| 13 | #include <linux/pagemap.h> |
| 14 | |
| 15 | #include <asm/asm-offsets.h> |
| 16 | #include <asm/irq.h> |
| 17 | #include <asm/sysinfo.h> |
| 18 | #include <asm/uv.h> |
| 19 | |
| 20 | #include "kvm-s390.h" |
| 21 | #include "gaccess.h" |
| 22 | #include "trace.h" |
| 23 | #include "trace-s390.h" |
| 24 | |
| 25 | u8 kvm_s390_get_ilen(struct kvm_vcpu *vcpu) |
| 26 | { |
| 27 | struct kvm_s390_sie_block *sie_block = vcpu->arch.sie_block; |
| 28 | u8 ilen = 0; |
| 29 | |
| 30 | switch (vcpu->arch.sie_block->icptcode) { |
| 31 | case ICPT_INST: |
| 32 | case ICPT_INSTPROGI: |
| 33 | case ICPT_OPEREXC: |
| 34 | case ICPT_PARTEXEC: |
| 35 | case ICPT_IOINST: |
| 36 | /* instruction only stored for these icptcodes */ |
| 37 | ilen = insn_length(vcpu->arch.sie_block->ipa >> 8); |
| 38 | /* Use the length of the EXECUTE instruction if necessary */ |
| 39 | if (sie_block->icptstatus & 1) { |
| 40 | ilen = (sie_block->icptstatus >> 4) & 0x6; |
| 41 | if (!ilen) |
| 42 | ilen = 4; |
| 43 | } |
| 44 | break; |
| 45 | case ICPT_PROGI: |
| 46 | /* bit 1+2 of pgmilc are the ilc, so we directly get ilen */ |
| 47 | ilen = vcpu->arch.sie_block->pgmilc & 0x6; |
| 48 | break; |
| 49 | } |
| 50 | return ilen; |
| 51 | } |
| 52 | |
| 53 | static int handle_stop(struct kvm_vcpu *vcpu) |
| 54 | { |
| 55 | struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int; |
| 56 | int rc = 0; |
| 57 | uint8_t flags, stop_pending; |
| 58 | |
| 59 | vcpu->stat.exit_stop_request++; |
| 60 | |
| 61 | /* delay the stop if any non-stop irq is pending */ |
| 62 | if (kvm_s390_vcpu_has_irq(vcpu, 1)) |
| 63 | return 0; |
| 64 | |
| 65 | /* avoid races with the injection/SIGP STOP code */ |
| 66 | spin_lock(&li->lock); |
| 67 | flags = li->irq.stop.flags; |
| 68 | stop_pending = kvm_s390_is_stop_irq_pending(vcpu); |
| 69 | spin_unlock(&li->lock); |
| 70 | |
| 71 | trace_kvm_s390_stop_request(stop_pending, flags); |
| 72 | if (!stop_pending) |
| 73 | return 0; |
| 74 | |
| 75 | if (flags & KVM_S390_STOP_FLAG_STORE_STATUS) { |
| 76 | rc = kvm_s390_vcpu_store_status(vcpu, |
| 77 | KVM_S390_STORE_STATUS_NOADDR); |
| 78 | if (rc) |
| 79 | return rc; |
| 80 | } |
| 81 | |
| 82 | /* |
| 83 | * no need to check the return value of vcpu_stop as it can only have |
| 84 | * an error for protvirt, but protvirt means user cpu state |
| 85 | */ |
| 86 | if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) |
| 87 | kvm_s390_vcpu_stop(vcpu); |
| 88 | return -EOPNOTSUPP; |
| 89 | } |
| 90 | |
| 91 | static int handle_validity(struct kvm_vcpu *vcpu) |
| 92 | { |
| 93 | int viwhy = vcpu->arch.sie_block->ipb >> 16; |
| 94 | |
| 95 | vcpu->stat.exit_validity++; |
| 96 | trace_kvm_s390_intercept_validity(vcpu, viwhy); |
| 97 | KVM_EVENT(3, "validity intercept 0x%x for pid %u (kvm 0x%p)", viwhy, |
| 98 | current->pid, vcpu->kvm); |
| 99 | |
| 100 | /* do not warn on invalid runtime instrumentation mode */ |
| 101 | WARN_ONCE(viwhy != 0x44, "kvm: unhandled validity intercept 0x%x\n", |
| 102 | viwhy); |
| 103 | return -EINVAL; |
| 104 | } |
| 105 | |
| 106 | static int handle_instruction(struct kvm_vcpu *vcpu) |
| 107 | { |
| 108 | vcpu->stat.exit_instruction++; |
| 109 | trace_kvm_s390_intercept_instruction(vcpu, |
| 110 | vcpu->arch.sie_block->ipa, |
| 111 | vcpu->arch.sie_block->ipb); |
| 112 | |
| 113 | switch (vcpu->arch.sie_block->ipa >> 8) { |
| 114 | case 0x01: |
| 115 | return kvm_s390_handle_01(vcpu); |
| 116 | case 0x82: |
| 117 | return kvm_s390_handle_lpsw(vcpu); |
| 118 | case 0x83: |
| 119 | return kvm_s390_handle_diag(vcpu); |
| 120 | case 0xaa: |
| 121 | return kvm_s390_handle_aa(vcpu); |
| 122 | case 0xae: |
| 123 | return kvm_s390_handle_sigp(vcpu); |
| 124 | case 0xb2: |
| 125 | return kvm_s390_handle_b2(vcpu); |
| 126 | case 0xb6: |
| 127 | return kvm_s390_handle_stctl(vcpu); |
| 128 | case 0xb7: |
| 129 | return kvm_s390_handle_lctl(vcpu); |
| 130 | case 0xb9: |
| 131 | return kvm_s390_handle_b9(vcpu); |
| 132 | case 0xe3: |
| 133 | return kvm_s390_handle_e3(vcpu); |
| 134 | case 0xe5: |
| 135 | return kvm_s390_handle_e5(vcpu); |
| 136 | case 0xeb: |
| 137 | return kvm_s390_handle_eb(vcpu); |
| 138 | default: |
| 139 | return -EOPNOTSUPP; |
| 140 | } |
| 141 | } |
| 142 | |
| 143 | static int inject_prog_on_prog_intercept(struct kvm_vcpu *vcpu) |
| 144 | { |
| 145 | struct kvm_s390_pgm_info pgm_info = { |
| 146 | .code = vcpu->arch.sie_block->iprcc, |
| 147 | /* the PSW has already been rewound */ |
| 148 | .flags = KVM_S390_PGM_FLAGS_NO_REWIND, |
| 149 | }; |
| 150 | |
| 151 | switch (vcpu->arch.sie_block->iprcc & ~PGM_PER) { |
| 152 | case PGM_AFX_TRANSLATION: |
| 153 | case PGM_ASX_TRANSLATION: |
| 154 | case PGM_EX_TRANSLATION: |
| 155 | case PGM_LFX_TRANSLATION: |
| 156 | case PGM_LSTE_SEQUENCE: |
| 157 | case PGM_LSX_TRANSLATION: |
| 158 | case PGM_LX_TRANSLATION: |
| 159 | case PGM_PRIMARY_AUTHORITY: |
| 160 | case PGM_SECONDARY_AUTHORITY: |
| 161 | case PGM_SPACE_SWITCH: |
| 162 | pgm_info.trans_exc_code = vcpu->arch.sie_block->tecmc; |
| 163 | break; |
| 164 | case PGM_ALEN_TRANSLATION: |
| 165 | case PGM_ALE_SEQUENCE: |
| 166 | case PGM_ASTE_INSTANCE: |
| 167 | case PGM_ASTE_SEQUENCE: |
| 168 | case PGM_ASTE_VALIDITY: |
| 169 | case PGM_EXTENDED_AUTHORITY: |
| 170 | pgm_info.exc_access_id = vcpu->arch.sie_block->eai; |
| 171 | break; |
| 172 | case PGM_ASCE_TYPE: |
| 173 | case PGM_PAGE_TRANSLATION: |
| 174 | case PGM_REGION_FIRST_TRANS: |
| 175 | case PGM_REGION_SECOND_TRANS: |
| 176 | case PGM_REGION_THIRD_TRANS: |
| 177 | case PGM_SEGMENT_TRANSLATION: |
| 178 | pgm_info.trans_exc_code = vcpu->arch.sie_block->tecmc; |
| 179 | pgm_info.exc_access_id = vcpu->arch.sie_block->eai; |
| 180 | pgm_info.op_access_id = vcpu->arch.sie_block->oai; |
| 181 | break; |
| 182 | case PGM_MONITOR: |
| 183 | pgm_info.mon_class_nr = vcpu->arch.sie_block->mcn; |
| 184 | pgm_info.mon_code = vcpu->arch.sie_block->tecmc; |
| 185 | break; |
| 186 | case PGM_VECTOR_PROCESSING: |
| 187 | case PGM_DATA: |
| 188 | pgm_info.data_exc_code = vcpu->arch.sie_block->dxc; |
| 189 | break; |
| 190 | case PGM_PROTECTION: |
| 191 | pgm_info.trans_exc_code = vcpu->arch.sie_block->tecmc; |
| 192 | pgm_info.exc_access_id = vcpu->arch.sie_block->eai; |
| 193 | break; |
| 194 | default: |
| 195 | break; |
| 196 | } |
| 197 | |
| 198 | if (vcpu->arch.sie_block->iprcc & PGM_PER) { |
| 199 | pgm_info.per_code = vcpu->arch.sie_block->perc; |
| 200 | pgm_info.per_atmid = vcpu->arch.sie_block->peratmid; |
| 201 | pgm_info.per_address = vcpu->arch.sie_block->peraddr; |
| 202 | pgm_info.per_access_id = vcpu->arch.sie_block->peraid; |
| 203 | } |
| 204 | return kvm_s390_inject_prog_irq(vcpu, &pgm_info); |
| 205 | } |
| 206 | |
| 207 | /* |
| 208 | * restore ITDB to program-interruption TDB in guest lowcore |
| 209 | * and set TX abort indication if required |
| 210 | */ |
| 211 | static int handle_itdb(struct kvm_vcpu *vcpu) |
| 212 | { |
| 213 | struct kvm_s390_itdb *itdb; |
| 214 | int rc; |
| 215 | |
| 216 | if (!IS_TE_ENABLED(vcpu) || !IS_ITDB_VALID(vcpu)) |
| 217 | return 0; |
| 218 | if (current->thread.per_flags & PER_FLAG_NO_TE) |
| 219 | return 0; |
| 220 | itdb = phys_to_virt(vcpu->arch.sie_block->itdba); |
| 221 | rc = write_guest_lc(vcpu, __LC_PGM_TDB, itdb, sizeof(*itdb)); |
| 222 | if (rc) |
| 223 | return rc; |
| 224 | memset(itdb, 0, sizeof(*itdb)); |
| 225 | |
| 226 | return 0; |
| 227 | } |
| 228 | |
| 229 | #define per_event(vcpu) (vcpu->arch.sie_block->iprcc & PGM_PER) |
| 230 | |
| 231 | static bool should_handle_per_event(const struct kvm_vcpu *vcpu) |
| 232 | { |
| 233 | if (!guestdbg_enabled(vcpu) || !per_event(vcpu)) |
| 234 | return false; |
| 235 | if (guestdbg_sstep_enabled(vcpu) && |
| 236 | vcpu->arch.sie_block->iprcc != PGM_PER) { |
| 237 | /* |
| 238 | * __vcpu_run() will exit after delivering the concurrently |
| 239 | * indicated condition. |
| 240 | */ |
| 241 | return false; |
| 242 | } |
| 243 | return true; |
| 244 | } |
| 245 | |
| 246 | static int handle_prog(struct kvm_vcpu *vcpu) |
| 247 | { |
| 248 | psw_t psw; |
| 249 | int rc; |
| 250 | |
| 251 | vcpu->stat.exit_program_interruption++; |
| 252 | |
| 253 | /* |
| 254 | * Intercept 8 indicates a loop of specification exceptions |
| 255 | * for protected guests. |
| 256 | */ |
| 257 | if (kvm_s390_pv_cpu_is_protected(vcpu)) |
| 258 | return -EOPNOTSUPP; |
| 259 | |
| 260 | if (should_handle_per_event(vcpu)) { |
| 261 | rc = kvm_s390_handle_per_event(vcpu); |
| 262 | if (rc) |
| 263 | return rc; |
| 264 | /* the interrupt might have been filtered out completely */ |
| 265 | if (vcpu->arch.sie_block->iprcc == 0) |
| 266 | return 0; |
| 267 | } |
| 268 | |
| 269 | trace_kvm_s390_intercept_prog(vcpu, vcpu->arch.sie_block->iprcc); |
| 270 | if (vcpu->arch.sie_block->iprcc == PGM_SPECIFICATION) { |
| 271 | rc = read_guest_lc(vcpu, __LC_PGM_NEW_PSW, &psw, sizeof(psw_t)); |
| 272 | if (rc) |
| 273 | return rc; |
| 274 | /* Avoid endless loops of specification exceptions */ |
| 275 | if (!is_valid_psw(&psw)) |
| 276 | return -EOPNOTSUPP; |
| 277 | } |
| 278 | rc = handle_itdb(vcpu); |
| 279 | if (rc) |
| 280 | return rc; |
| 281 | |
| 282 | return inject_prog_on_prog_intercept(vcpu); |
| 283 | } |
| 284 | |
| 285 | /** |
| 286 | * handle_external_interrupt - used for external interruption interceptions |
| 287 | * @vcpu: virtual cpu |
| 288 | * |
| 289 | * This interception occurs if: |
| 290 | * - the CPUSTAT_EXT_INT bit was already set when the external interrupt |
| 291 | * occurred. In this case, the interrupt needs to be injected manually to |
| 292 | * preserve interrupt priority. |
| 293 | * - the external new PSW has external interrupts enabled, which will cause an |
| 294 | * interruption loop. We drop to userspace in this case. |
| 295 | * |
| 296 | * The latter case can be detected by inspecting the external mask bit in the |
| 297 | * external new psw. |
| 298 | * |
| 299 | * Under PV, only the latter case can occur, since interrupt priorities are |
| 300 | * handled in the ultravisor. |
| 301 | */ |
| 302 | static int handle_external_interrupt(struct kvm_vcpu *vcpu) |
| 303 | { |
| 304 | u16 eic = vcpu->arch.sie_block->eic; |
| 305 | struct kvm_s390_irq irq; |
| 306 | psw_t newpsw; |
| 307 | int rc; |
| 308 | |
| 309 | vcpu->stat.exit_external_interrupt++; |
| 310 | |
| 311 | if (kvm_s390_pv_cpu_is_protected(vcpu)) { |
| 312 | newpsw = vcpu->arch.sie_block->gpsw; |
| 313 | } else { |
| 314 | rc = read_guest_lc(vcpu, __LC_EXT_NEW_PSW, &newpsw, sizeof(psw_t)); |
| 315 | if (rc) |
| 316 | return rc; |
| 317 | } |
| 318 | |
| 319 | /* |
| 320 | * Clock comparator or timer interrupt with external interrupt enabled |
| 321 | * will cause interrupt loop. Drop to userspace. |
| 322 | */ |
| 323 | if ((eic == EXT_IRQ_CLK_COMP || eic == EXT_IRQ_CPU_TIMER) && |
| 324 | (newpsw.mask & PSW_MASK_EXT)) |
| 325 | return -EOPNOTSUPP; |
| 326 | |
| 327 | switch (eic) { |
| 328 | case EXT_IRQ_CLK_COMP: |
| 329 | irq.type = KVM_S390_INT_CLOCK_COMP; |
| 330 | break; |
| 331 | case EXT_IRQ_CPU_TIMER: |
| 332 | irq.type = KVM_S390_INT_CPU_TIMER; |
| 333 | break; |
| 334 | case EXT_IRQ_EXTERNAL_CALL: |
| 335 | irq.type = KVM_S390_INT_EXTERNAL_CALL; |
| 336 | irq.u.extcall.code = vcpu->arch.sie_block->extcpuaddr; |
| 337 | rc = kvm_s390_inject_vcpu(vcpu, &irq); |
| 338 | /* ignore if another external call is already pending */ |
| 339 | if (rc == -EBUSY) |
| 340 | return 0; |
| 341 | return rc; |
| 342 | default: |
| 343 | return -EOPNOTSUPP; |
| 344 | } |
| 345 | |
| 346 | return kvm_s390_inject_vcpu(vcpu, &irq); |
| 347 | } |
| 348 | |
| 349 | /** |
| 350 | * handle_mvpg_pei - Handle MOVE PAGE partial execution interception. |
| 351 | * @vcpu: virtual cpu |
| 352 | * |
| 353 | * This interception can only happen for guests with DAT disabled and |
| 354 | * addresses that are currently not mapped in the host. Thus we try to |
| 355 | * set up the mappings for the corresponding user pages here (or throw |
| 356 | * addressing exceptions in case of illegal guest addresses). |
| 357 | */ |
| 358 | static int handle_mvpg_pei(struct kvm_vcpu *vcpu) |
| 359 | { |
| 360 | unsigned long srcaddr, dstaddr; |
| 361 | int reg1, reg2, rc; |
| 362 | |
| 363 | kvm_s390_get_regs_rre(vcpu, ®1, ®2); |
| 364 | |
| 365 | /* Ensure that the source is paged-in, no actual access -> no key checking */ |
| 366 | rc = guest_translate_address_with_key(vcpu, vcpu->run->s.regs.gprs[reg2], |
| 367 | reg2, &srcaddr, GACC_FETCH, 0); |
| 368 | if (rc) |
| 369 | return kvm_s390_inject_prog_cond(vcpu, rc); |
| 370 | rc = kvm_s390_handle_dat_fault(vcpu, srcaddr, 0); |
| 371 | if (rc != 0) |
| 372 | return rc; |
| 373 | |
| 374 | /* Ensure that the source is paged-in, no actual access -> no key checking */ |
| 375 | rc = guest_translate_address_with_key(vcpu, vcpu->run->s.regs.gprs[reg1], |
| 376 | reg1, &dstaddr, GACC_STORE, 0); |
| 377 | if (rc) |
| 378 | return kvm_s390_inject_prog_cond(vcpu, rc); |
| 379 | rc = kvm_s390_handle_dat_fault(vcpu, dstaddr, FOLL_WRITE); |
| 380 | if (rc != 0) |
| 381 | return rc; |
| 382 | |
| 383 | kvm_s390_retry_instr(vcpu); |
| 384 | |
| 385 | return 0; |
| 386 | } |
| 387 | |
| 388 | static int handle_partial_execution(struct kvm_vcpu *vcpu) |
| 389 | { |
| 390 | vcpu->stat.exit_pei++; |
| 391 | |
| 392 | if (vcpu->arch.sie_block->ipa == 0xb254) /* MVPG */ |
| 393 | return handle_mvpg_pei(vcpu); |
| 394 | if (vcpu->arch.sie_block->ipa >> 8 == 0xae) /* SIGP */ |
| 395 | return kvm_s390_handle_sigp_pei(vcpu); |
| 396 | |
| 397 | return -EOPNOTSUPP; |
| 398 | } |
| 399 | |
| 400 | /* |
| 401 | * Handle the sthyi instruction that provides the guest with system |
| 402 | * information, like current CPU resources available at each level of |
| 403 | * the machine. |
| 404 | */ |
| 405 | int handle_sthyi(struct kvm_vcpu *vcpu) |
| 406 | { |
| 407 | int reg1, reg2, cc = 0, r = 0; |
| 408 | u64 code, addr, rc = 0; |
| 409 | struct sthyi_sctns *sctns = NULL; |
| 410 | |
| 411 | if (!test_kvm_facility(vcpu->kvm, 74)) |
| 412 | return kvm_s390_inject_program_int(vcpu, PGM_OPERATION); |
| 413 | |
| 414 | kvm_s390_get_regs_rre(vcpu, ®1, ®2); |
| 415 | code = vcpu->run->s.regs.gprs[reg1]; |
| 416 | addr = vcpu->run->s.regs.gprs[reg2]; |
| 417 | |
| 418 | vcpu->stat.instruction_sthyi++; |
| 419 | VCPU_EVENT(vcpu, 3, "STHYI: fc: %llu addr: 0x%016llx", code, addr); |
| 420 | trace_kvm_s390_handle_sthyi(vcpu, code, addr); |
| 421 | |
| 422 | if (reg1 == reg2 || reg1 & 1 || reg2 & 1) |
| 423 | return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); |
| 424 | |
| 425 | if (code & 0xffff) { |
| 426 | cc = 3; |
| 427 | rc = 4; |
| 428 | goto out; |
| 429 | } |
| 430 | |
| 431 | if (!kvm_s390_pv_cpu_is_protected(vcpu) && (addr & ~PAGE_MASK)) |
| 432 | return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); |
| 433 | |
| 434 | sctns = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT); |
| 435 | if (!sctns) |
| 436 | return -ENOMEM; |
| 437 | |
| 438 | cc = sthyi_fill(sctns, &rc); |
| 439 | if (cc < 0) { |
| 440 | free_page((unsigned long)sctns); |
| 441 | return cc; |
| 442 | } |
| 443 | out: |
| 444 | if (!cc) { |
| 445 | if (kvm_s390_pv_cpu_is_protected(vcpu)) { |
| 446 | memcpy(sida_addr(vcpu->arch.sie_block), sctns, PAGE_SIZE); |
| 447 | } else { |
| 448 | r = write_guest(vcpu, addr, reg2, sctns, PAGE_SIZE); |
| 449 | if (r) { |
| 450 | free_page((unsigned long)sctns); |
| 451 | return kvm_s390_inject_prog_cond(vcpu, r); |
| 452 | } |
| 453 | } |
| 454 | } |
| 455 | |
| 456 | free_page((unsigned long)sctns); |
| 457 | vcpu->run->s.regs.gprs[reg2 + 1] = rc; |
| 458 | kvm_s390_set_psw_cc(vcpu, cc); |
| 459 | return r; |
| 460 | } |
| 461 | |
| 462 | static int handle_operexc(struct kvm_vcpu *vcpu) |
| 463 | { |
| 464 | psw_t oldpsw, newpsw; |
| 465 | int rc; |
| 466 | |
| 467 | vcpu->stat.exit_operation_exception++; |
| 468 | trace_kvm_s390_handle_operexc(vcpu, vcpu->arch.sie_block->ipa, |
| 469 | vcpu->arch.sie_block->ipb); |
| 470 | |
| 471 | if (vcpu->arch.sie_block->ipa == 0xb256) |
| 472 | return handle_sthyi(vcpu); |
| 473 | |
| 474 | if (vcpu->arch.sie_block->ipa == 0 && vcpu->kvm->arch.user_instr0) |
| 475 | return -EOPNOTSUPP; |
| 476 | rc = read_guest_lc(vcpu, __LC_PGM_NEW_PSW, &newpsw, sizeof(psw_t)); |
| 477 | if (rc) |
| 478 | return rc; |
| 479 | /* |
| 480 | * Avoid endless loops of operation exceptions, if the pgm new |
| 481 | * PSW will cause a new operation exception. |
| 482 | * The heuristic checks if the pgm new psw is within 6 bytes before |
| 483 | * the faulting psw address (with same DAT, AS settings) and the |
| 484 | * new psw is not a wait psw and the fault was not triggered by |
| 485 | * problem state. |
| 486 | */ |
| 487 | oldpsw = vcpu->arch.sie_block->gpsw; |
| 488 | if (oldpsw.addr - newpsw.addr <= 6 && |
| 489 | !(newpsw.mask & PSW_MASK_WAIT) && |
| 490 | !(oldpsw.mask & PSW_MASK_PSTATE) && |
| 491 | (newpsw.mask & PSW_MASK_ASC) == (oldpsw.mask & PSW_MASK_ASC) && |
| 492 | (newpsw.mask & PSW_MASK_DAT) == (oldpsw.mask & PSW_MASK_DAT)) |
| 493 | return -EOPNOTSUPP; |
| 494 | |
| 495 | return kvm_s390_inject_program_int(vcpu, PGM_OPERATION); |
| 496 | } |
| 497 | |
| 498 | static int handle_pv_spx(struct kvm_vcpu *vcpu) |
| 499 | { |
| 500 | u32 pref = *(u32 *)sida_addr(vcpu->arch.sie_block); |
| 501 | |
| 502 | kvm_s390_set_prefix(vcpu, pref); |
| 503 | trace_kvm_s390_handle_prefix(vcpu, 1, pref); |
| 504 | return 0; |
| 505 | } |
| 506 | |
| 507 | static int handle_pv_sclp(struct kvm_vcpu *vcpu) |
| 508 | { |
| 509 | struct kvm_s390_float_interrupt *fi = &vcpu->kvm->arch.float_int; |
| 510 | |
| 511 | spin_lock(&fi->lock); |
| 512 | /* |
| 513 | * 2 cases: |
| 514 | * a: an sccb answering interrupt was already pending or in flight. |
| 515 | * As the sccb value is not known we can simply set some value to |
| 516 | * trigger delivery of a saved SCCB. UV will then use its saved |
| 517 | * copy of the SCCB value. |
| 518 | * b: an error SCCB interrupt needs to be injected so we also inject |
| 519 | * a fake SCCB address. Firmware will use the proper one. |
| 520 | * This makes sure, that both errors and real sccb returns will only |
| 521 | * be delivered after a notification intercept (instruction has |
| 522 | * finished) but not after others. |
| 523 | */ |
| 524 | fi->srv_signal.ext_params |= 0x43000; |
| 525 | set_bit(IRQ_PEND_EXT_SERVICE, &fi->pending_irqs); |
| 526 | clear_bit(IRQ_PEND_EXT_SERVICE, &fi->masked_irqs); |
| 527 | spin_unlock(&fi->lock); |
| 528 | return 0; |
| 529 | } |
| 530 | |
| 531 | static int handle_pv_uvc(struct kvm_vcpu *vcpu) |
| 532 | { |
| 533 | struct uv_cb_share *guest_uvcb = sida_addr(vcpu->arch.sie_block); |
| 534 | struct uv_cb_cts uvcb = { |
| 535 | .header.cmd = UVC_CMD_UNPIN_PAGE_SHARED, |
| 536 | .header.len = sizeof(uvcb), |
| 537 | .guest_handle = kvm_s390_pv_get_handle(vcpu->kvm), |
| 538 | .gaddr = guest_uvcb->paddr, |
| 539 | }; |
| 540 | int rc; |
| 541 | |
| 542 | if (guest_uvcb->header.cmd != UVC_CMD_REMOVE_SHARED_ACCESS) { |
| 543 | WARN_ONCE(1, "Unexpected notification intercept for UVC 0x%x\n", |
| 544 | guest_uvcb->header.cmd); |
| 545 | return 0; |
| 546 | } |
| 547 | rc = kvm_s390_pv_make_secure(vcpu->kvm, uvcb.gaddr, &uvcb); |
| 548 | /* |
| 549 | * If the unpin did not succeed, the guest will exit again for the UVC |
| 550 | * and we will retry the unpin. |
| 551 | */ |
| 552 | if (rc == -EINVAL || rc == -ENXIO) |
| 553 | return 0; |
| 554 | /* |
| 555 | * If we got -EAGAIN here, we simply return it. It will eventually |
| 556 | * get propagated all the way to userspace, which should then try |
| 557 | * again. |
| 558 | */ |
| 559 | return rc; |
| 560 | } |
| 561 | |
| 562 | static int handle_pv_notification(struct kvm_vcpu *vcpu) |
| 563 | { |
| 564 | int ret; |
| 565 | |
| 566 | if (vcpu->arch.sie_block->ipa == 0xb210) |
| 567 | return handle_pv_spx(vcpu); |
| 568 | if (vcpu->arch.sie_block->ipa == 0xb220) |
| 569 | return handle_pv_sclp(vcpu); |
| 570 | if (vcpu->arch.sie_block->ipa == 0xb9a4) |
| 571 | return handle_pv_uvc(vcpu); |
| 572 | if (vcpu->arch.sie_block->ipa >> 8 == 0xae) { |
| 573 | /* |
| 574 | * Besides external call, other SIGP orders also cause a |
| 575 | * 108 (pv notify) intercept. In contrast to external call, |
| 576 | * these orders need to be emulated and hence the appropriate |
| 577 | * place to handle them is in handle_instruction(). |
| 578 | * So first try kvm_s390_handle_sigp_pei() and if that isn't |
| 579 | * successful, go on with handle_instruction(). |
| 580 | */ |
| 581 | ret = kvm_s390_handle_sigp_pei(vcpu); |
| 582 | if (!ret) |
| 583 | return ret; |
| 584 | } |
| 585 | |
| 586 | return handle_instruction(vcpu); |
| 587 | } |
| 588 | |
| 589 | static bool should_handle_per_ifetch(const struct kvm_vcpu *vcpu, int rc) |
| 590 | { |
| 591 | /* Process PER, also if the instruction is processed in user space. */ |
| 592 | if (!(vcpu->arch.sie_block->icptstatus & 0x02)) |
| 593 | return false; |
| 594 | if (rc != 0 && rc != -EOPNOTSUPP) |
| 595 | return false; |
| 596 | if (guestdbg_sstep_enabled(vcpu) && vcpu->arch.local_int.pending_irqs) |
| 597 | /* __vcpu_run() will exit after delivering the interrupt. */ |
| 598 | return false; |
| 599 | return true; |
| 600 | } |
| 601 | |
| 602 | int kvm_handle_sie_intercept(struct kvm_vcpu *vcpu) |
| 603 | { |
| 604 | int rc, per_rc = 0; |
| 605 | |
| 606 | if (kvm_is_ucontrol(vcpu->kvm)) |
| 607 | return -EOPNOTSUPP; |
| 608 | |
| 609 | switch (vcpu->arch.sie_block->icptcode) { |
| 610 | case ICPT_EXTREQ: |
| 611 | vcpu->stat.exit_external_request++; |
| 612 | return 0; |
| 613 | case ICPT_IOREQ: |
| 614 | vcpu->stat.exit_io_request++; |
| 615 | return 0; |
| 616 | case ICPT_INST: |
| 617 | rc = handle_instruction(vcpu); |
| 618 | break; |
| 619 | case ICPT_PROGI: |
| 620 | return handle_prog(vcpu); |
| 621 | case ICPT_EXTINT: |
| 622 | return handle_external_interrupt(vcpu); |
| 623 | case ICPT_WAIT: |
| 624 | return kvm_s390_handle_wait(vcpu); |
| 625 | case ICPT_VALIDITY: |
| 626 | return handle_validity(vcpu); |
| 627 | case ICPT_STOP: |
| 628 | return handle_stop(vcpu); |
| 629 | case ICPT_OPEREXC: |
| 630 | rc = handle_operexc(vcpu); |
| 631 | break; |
| 632 | case ICPT_PARTEXEC: |
| 633 | rc = handle_partial_execution(vcpu); |
| 634 | break; |
| 635 | case ICPT_KSS: |
| 636 | /* Instruction will be redriven, skip the PER check. */ |
| 637 | return kvm_s390_skey_check_enable(vcpu); |
| 638 | case ICPT_MCHKREQ: |
| 639 | case ICPT_INT_ENABLE: |
| 640 | /* |
| 641 | * PSW bit 13 or a CR (0, 6, 14) changed and we might |
| 642 | * now be able to deliver interrupts. The pre-run code |
| 643 | * will take care of this. |
| 644 | */ |
| 645 | rc = 0; |
| 646 | break; |
| 647 | case ICPT_PV_INSTR: |
| 648 | rc = handle_instruction(vcpu); |
| 649 | break; |
| 650 | case ICPT_PV_NOTIFY: |
| 651 | rc = handle_pv_notification(vcpu); |
| 652 | break; |
| 653 | case ICPT_PV_PREF: |
| 654 | rc = 0; |
| 655 | kvm_s390_pv_convert_to_secure(vcpu->kvm, kvm_s390_get_prefix(vcpu)); |
| 656 | kvm_s390_pv_convert_to_secure(vcpu->kvm, kvm_s390_get_prefix(vcpu) + PAGE_SIZE); |
| 657 | break; |
| 658 | default: |
| 659 | return -EOPNOTSUPP; |
| 660 | } |
| 661 | |
| 662 | if (should_handle_per_ifetch(vcpu, rc)) |
| 663 | per_rc = kvm_s390_handle_per_ifetch_icpt(vcpu); |
| 664 | return per_rc ? per_rc : rc; |
| 665 | } |