1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * handling privileged instructions 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.h> 12 #include <linux/gfp.h> 13 #include <linux/errno.h> 14 #include <linux/mm_types.h> 15 #include <linux/pgtable.h> 16 17 #include <asm/asm-offsets.h> 18 #include <asm/facility.h> 19 #include <asm/current.h> 20 #include <asm/debug.h> 21 #include <asm/ebcdic.h> 22 #include <asm/sysinfo.h> 23 #include <asm/page-states.h> 24 #include <asm/gmap.h> 25 #include <asm/io.h> 26 #include <asm/ptrace.h> 27 #include <asm/sclp.h> 28 #include <asm/ap.h> 29 #include "gaccess.h" 30 #include "kvm-s390.h" 31 #include "trace.h" 32 33 static int handle_ri(struct kvm_vcpu *vcpu) 34 { 35 vcpu->stat.instruction_ri++; 36 37 if (test_kvm_facility(vcpu->kvm, 64)) { 38 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (lazy)"); 39 vcpu->arch.sie_block->ecb3 |= ECB3_RI; 40 kvm_s390_retry_instr(vcpu); 41 return 0; 42 } else 43 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION); 44 } 45 46 int kvm_s390_handle_aa(struct kvm_vcpu *vcpu) 47 { 48 if ((vcpu->arch.sie_block->ipa & 0xf) <= 4) 49 return handle_ri(vcpu); 50 else 51 return -EOPNOTSUPP; 52 } 53 54 static int handle_gs(struct kvm_vcpu *vcpu) 55 { 56 vcpu->stat.instruction_gs++; 57 58 if (test_kvm_facility(vcpu->kvm, 133)) { 59 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (lazy)"); 60 preempt_disable(); 61 __ctl_set_bit(2, 4); 62 current->thread.gs_cb = (struct gs_cb *)&vcpu->run->s.regs.gscb; 63 restore_gs_cb(current->thread.gs_cb); 64 preempt_enable(); 65 vcpu->arch.sie_block->ecb |= ECB_GS; 66 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT; 67 vcpu->arch.gs_enabled = 1; 68 kvm_s390_retry_instr(vcpu); 69 return 0; 70 } else 71 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION); 72 } 73 74 int kvm_s390_handle_e3(struct kvm_vcpu *vcpu) 75 { 76 int code = vcpu->arch.sie_block->ipb & 0xff; 77 78 if (code == 0x49 || code == 0x4d) 79 return handle_gs(vcpu); 80 else 81 return -EOPNOTSUPP; 82 } 83 /* Handle SCK (SET CLOCK) interception */ 84 static int handle_set_clock(struct kvm_vcpu *vcpu) 85 { 86 struct kvm_s390_vm_tod_clock gtod = { 0 }; 87 int rc; 88 u8 ar; 89 u64 op2; 90 91 vcpu->stat.instruction_sck++; 92 93 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 94 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 95 96 op2 = kvm_s390_get_base_disp_s(vcpu, &ar); 97 if (op2 & 7) /* Operand must be on a doubleword boundary */ 98 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 99 rc = read_guest(vcpu, op2, ar, >od.tod, sizeof(gtod.tod)); 100 if (rc) 101 return kvm_s390_inject_prog_cond(vcpu, rc); 102 103 VCPU_EVENT(vcpu, 3, "SCK: setting guest TOD to 0x%llx", gtod.tod); 104 /* 105 * To set the TOD clock the kvm lock must be taken, but the vcpu lock 106 * is already held in handle_set_clock. The usual lock order is the 107 * opposite. As SCK is deprecated and should not be used in several 108 * cases, for example when the multiple epoch facility or TOD clock 109 * steering facility is installed (see Principles of Operation), a 110 * slow path can be used. If the lock can not be taken via try_lock, 111 * the instruction will be retried via -EAGAIN at a later point in 112 * time. 113 */ 114 if (!kvm_s390_try_set_tod_clock(vcpu->kvm, >od)) { 115 kvm_s390_retry_instr(vcpu); 116 return -EAGAIN; 117 } 118 119 kvm_s390_set_psw_cc(vcpu, 0); 120 return 0; 121 } 122 123 static int handle_set_prefix(struct kvm_vcpu *vcpu) 124 { 125 u64 operand2; 126 u32 address; 127 int rc; 128 u8 ar; 129 130 vcpu->stat.instruction_spx++; 131 132 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 133 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 134 135 operand2 = kvm_s390_get_base_disp_s(vcpu, &ar); 136 137 /* must be word boundary */ 138 if (operand2 & 3) 139 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 140 141 /* get the value */ 142 rc = read_guest(vcpu, operand2, ar, &address, sizeof(address)); 143 if (rc) 144 return kvm_s390_inject_prog_cond(vcpu, rc); 145 146 address &= 0x7fffe000u; 147 148 /* 149 * Make sure the new value is valid memory. We only need to check the 150 * first page, since address is 8k aligned and memory pieces are always 151 * at least 1MB aligned and have at least a size of 1MB. 152 */ 153 if (kvm_is_error_gpa(vcpu->kvm, address)) 154 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING); 155 156 kvm_s390_set_prefix(vcpu, address); 157 trace_kvm_s390_handle_prefix(vcpu, 1, address); 158 return 0; 159 } 160 161 static int handle_store_prefix(struct kvm_vcpu *vcpu) 162 { 163 u64 operand2; 164 u32 address; 165 int rc; 166 u8 ar; 167 168 vcpu->stat.instruction_stpx++; 169 170 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 171 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 172 173 operand2 = kvm_s390_get_base_disp_s(vcpu, &ar); 174 175 /* must be word boundary */ 176 if (operand2 & 3) 177 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 178 179 address = kvm_s390_get_prefix(vcpu); 180 181 /* get the value */ 182 rc = write_guest(vcpu, operand2, ar, &address, sizeof(address)); 183 if (rc) 184 return kvm_s390_inject_prog_cond(vcpu, rc); 185 186 VCPU_EVENT(vcpu, 3, "STPX: storing prefix 0x%x into 0x%llx", address, operand2); 187 trace_kvm_s390_handle_prefix(vcpu, 0, address); 188 return 0; 189 } 190 191 static int handle_store_cpu_address(struct kvm_vcpu *vcpu) 192 { 193 u16 vcpu_id = vcpu->vcpu_id; 194 u64 ga; 195 int rc; 196 u8 ar; 197 198 vcpu->stat.instruction_stap++; 199 200 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 201 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 202 203 ga = kvm_s390_get_base_disp_s(vcpu, &ar); 204 205 if (ga & 1) 206 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 207 208 rc = write_guest(vcpu, ga, ar, &vcpu_id, sizeof(vcpu_id)); 209 if (rc) 210 return kvm_s390_inject_prog_cond(vcpu, rc); 211 212 VCPU_EVENT(vcpu, 3, "STAP: storing cpu address (%u) to 0x%llx", vcpu_id, ga); 213 trace_kvm_s390_handle_stap(vcpu, ga); 214 return 0; 215 } 216 217 int kvm_s390_skey_check_enable(struct kvm_vcpu *vcpu) 218 { 219 int rc; 220 221 trace_kvm_s390_skey_related_inst(vcpu); 222 /* Already enabled? */ 223 if (vcpu->arch.skey_enabled) 224 return 0; 225 226 rc = s390_enable_skey(); 227 VCPU_EVENT(vcpu, 3, "enabling storage keys for guest: %d", rc); 228 if (rc) 229 return rc; 230 231 if (kvm_s390_test_cpuflags(vcpu, CPUSTAT_KSS)) 232 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_KSS); 233 if (!vcpu->kvm->arch.use_skf) 234 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE; 235 else 236 vcpu->arch.sie_block->ictl &= ~(ICTL_ISKE | ICTL_SSKE | ICTL_RRBE); 237 vcpu->arch.skey_enabled = true; 238 return 0; 239 } 240 241 static int try_handle_skey(struct kvm_vcpu *vcpu) 242 { 243 int rc; 244 245 rc = kvm_s390_skey_check_enable(vcpu); 246 if (rc) 247 return rc; 248 if (vcpu->kvm->arch.use_skf) { 249 /* with storage-key facility, SIE interprets it for us */ 250 kvm_s390_retry_instr(vcpu); 251 VCPU_EVENT(vcpu, 4, "%s", "retrying storage key operation"); 252 return -EAGAIN; 253 } 254 return 0; 255 } 256 257 static int handle_iske(struct kvm_vcpu *vcpu) 258 { 259 unsigned long gaddr, vmaddr; 260 unsigned char key; 261 int reg1, reg2; 262 bool unlocked; 263 int rc; 264 265 vcpu->stat.instruction_iske++; 266 267 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 268 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 269 270 rc = try_handle_skey(vcpu); 271 if (rc) 272 return rc != -EAGAIN ? rc : 0; 273 274 kvm_s390_get_regs_rre(vcpu, ®1, ®2); 275 276 gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK; 277 gaddr = kvm_s390_logical_to_effective(vcpu, gaddr); 278 gaddr = kvm_s390_real_to_abs(vcpu, gaddr); 279 vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr)); 280 if (kvm_is_error_hva(vmaddr)) 281 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING); 282 retry: 283 unlocked = false; 284 mmap_read_lock(current->mm); 285 rc = get_guest_storage_key(current->mm, vmaddr, &key); 286 287 if (rc) { 288 rc = fixup_user_fault(current->mm, vmaddr, 289 FAULT_FLAG_WRITE, &unlocked); 290 if (!rc) { 291 mmap_read_unlock(current->mm); 292 goto retry; 293 } 294 } 295 mmap_read_unlock(current->mm); 296 if (rc == -EFAULT) 297 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING); 298 if (rc < 0) 299 return rc; 300 vcpu->run->s.regs.gprs[reg1] &= ~0xff; 301 vcpu->run->s.regs.gprs[reg1] |= key; 302 return 0; 303 } 304 305 static int handle_rrbe(struct kvm_vcpu *vcpu) 306 { 307 unsigned long vmaddr, gaddr; 308 int reg1, reg2; 309 bool unlocked; 310 int rc; 311 312 vcpu->stat.instruction_rrbe++; 313 314 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 315 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 316 317 rc = try_handle_skey(vcpu); 318 if (rc) 319 return rc != -EAGAIN ? rc : 0; 320 321 kvm_s390_get_regs_rre(vcpu, ®1, ®2); 322 323 gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK; 324 gaddr = kvm_s390_logical_to_effective(vcpu, gaddr); 325 gaddr = kvm_s390_real_to_abs(vcpu, gaddr); 326 vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr)); 327 if (kvm_is_error_hva(vmaddr)) 328 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING); 329 retry: 330 unlocked = false; 331 mmap_read_lock(current->mm); 332 rc = reset_guest_reference_bit(current->mm, vmaddr); 333 if (rc < 0) { 334 rc = fixup_user_fault(current->mm, vmaddr, 335 FAULT_FLAG_WRITE, &unlocked); 336 if (!rc) { 337 mmap_read_unlock(current->mm); 338 goto retry; 339 } 340 } 341 mmap_read_unlock(current->mm); 342 if (rc == -EFAULT) 343 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING); 344 if (rc < 0) 345 return rc; 346 kvm_s390_set_psw_cc(vcpu, rc); 347 return 0; 348 } 349 350 #define SSKE_NQ 0x8 351 #define SSKE_MR 0x4 352 #define SSKE_MC 0x2 353 #define SSKE_MB 0x1 354 static int handle_sske(struct kvm_vcpu *vcpu) 355 { 356 unsigned char m3 = vcpu->arch.sie_block->ipb >> 28; 357 unsigned long start, end; 358 unsigned char key, oldkey; 359 int reg1, reg2; 360 bool unlocked; 361 int rc; 362 363 vcpu->stat.instruction_sske++; 364 365 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 366 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 367 368 rc = try_handle_skey(vcpu); 369 if (rc) 370 return rc != -EAGAIN ? rc : 0; 371 372 if (!test_kvm_facility(vcpu->kvm, 8)) 373 m3 &= ~SSKE_MB; 374 if (!test_kvm_facility(vcpu->kvm, 10)) 375 m3 &= ~(SSKE_MC | SSKE_MR); 376 if (!test_kvm_facility(vcpu->kvm, 14)) 377 m3 &= ~SSKE_NQ; 378 379 kvm_s390_get_regs_rre(vcpu, ®1, ®2); 380 381 key = vcpu->run->s.regs.gprs[reg1] & 0xfe; 382 start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK; 383 start = kvm_s390_logical_to_effective(vcpu, start); 384 if (m3 & SSKE_MB) { 385 /* start already designates an absolute address */ 386 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1); 387 } else { 388 start = kvm_s390_real_to_abs(vcpu, start); 389 end = start + PAGE_SIZE; 390 } 391 392 while (start != end) { 393 unsigned long vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start)); 394 unlocked = false; 395 396 if (kvm_is_error_hva(vmaddr)) 397 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING); 398 399 mmap_read_lock(current->mm); 400 rc = cond_set_guest_storage_key(current->mm, vmaddr, key, &oldkey, 401 m3 & SSKE_NQ, m3 & SSKE_MR, 402 m3 & SSKE_MC); 403 404 if (rc < 0) { 405 rc = fixup_user_fault(current->mm, vmaddr, 406 FAULT_FLAG_WRITE, &unlocked); 407 rc = !rc ? -EAGAIN : rc; 408 } 409 mmap_read_unlock(current->mm); 410 if (rc == -EFAULT) 411 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING); 412 if (rc == -EAGAIN) 413 continue; 414 if (rc < 0) 415 return rc; 416 start += PAGE_SIZE; 417 } 418 419 if (m3 & (SSKE_MC | SSKE_MR)) { 420 if (m3 & SSKE_MB) { 421 /* skey in reg1 is unpredictable */ 422 kvm_s390_set_psw_cc(vcpu, 3); 423 } else { 424 kvm_s390_set_psw_cc(vcpu, rc); 425 vcpu->run->s.regs.gprs[reg1] &= ~0xff00UL; 426 vcpu->run->s.regs.gprs[reg1] |= (u64) oldkey << 8; 427 } 428 } 429 if (m3 & SSKE_MB) { 430 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT) 431 vcpu->run->s.regs.gprs[reg2] &= ~PAGE_MASK; 432 else 433 vcpu->run->s.regs.gprs[reg2] &= ~0xfffff000UL; 434 end = kvm_s390_logical_to_effective(vcpu, end); 435 vcpu->run->s.regs.gprs[reg2] |= end; 436 } 437 return 0; 438 } 439 440 static int handle_ipte_interlock(struct kvm_vcpu *vcpu) 441 { 442 vcpu->stat.instruction_ipte_interlock++; 443 if (psw_bits(vcpu->arch.sie_block->gpsw).pstate) 444 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 445 wait_event(vcpu->kvm->arch.ipte_wq, !ipte_lock_held(vcpu)); 446 kvm_s390_retry_instr(vcpu); 447 VCPU_EVENT(vcpu, 4, "%s", "retrying ipte interlock operation"); 448 return 0; 449 } 450 451 static int handle_test_block(struct kvm_vcpu *vcpu) 452 { 453 gpa_t addr; 454 int reg2; 455 456 vcpu->stat.instruction_tb++; 457 458 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 459 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 460 461 kvm_s390_get_regs_rre(vcpu, NULL, ®2); 462 addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK; 463 addr = kvm_s390_logical_to_effective(vcpu, addr); 464 if (kvm_s390_check_low_addr_prot_real(vcpu, addr)) 465 return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm); 466 addr = kvm_s390_real_to_abs(vcpu, addr); 467 468 if (kvm_is_error_gpa(vcpu->kvm, addr)) 469 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING); 470 /* 471 * We don't expect errors on modern systems, and do not care 472 * about storage keys (yet), so let's just clear the page. 473 */ 474 if (kvm_clear_guest(vcpu->kvm, addr, PAGE_SIZE)) 475 return -EFAULT; 476 kvm_s390_set_psw_cc(vcpu, 0); 477 vcpu->run->s.regs.gprs[0] = 0; 478 return 0; 479 } 480 481 static int handle_tpi(struct kvm_vcpu *vcpu) 482 { 483 struct kvm_s390_interrupt_info *inti; 484 unsigned long len; 485 u32 tpi_data[3]; 486 int rc; 487 u64 addr; 488 u8 ar; 489 490 vcpu->stat.instruction_tpi++; 491 492 addr = kvm_s390_get_base_disp_s(vcpu, &ar); 493 if (addr & 3) 494 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 495 496 inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0); 497 if (!inti) { 498 kvm_s390_set_psw_cc(vcpu, 0); 499 return 0; 500 } 501 502 tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr; 503 tpi_data[1] = inti->io.io_int_parm; 504 tpi_data[2] = inti->io.io_int_word; 505 if (addr) { 506 /* 507 * Store the two-word I/O interruption code into the 508 * provided area. 509 */ 510 len = sizeof(tpi_data) - 4; 511 rc = write_guest(vcpu, addr, ar, &tpi_data, len); 512 if (rc) { 513 rc = kvm_s390_inject_prog_cond(vcpu, rc); 514 goto reinject_interrupt; 515 } 516 } else { 517 /* 518 * Store the three-word I/O interruption code into 519 * the appropriate lowcore area. 520 */ 521 len = sizeof(tpi_data); 522 if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len)) { 523 /* failed writes to the low core are not recoverable */ 524 rc = -EFAULT; 525 goto reinject_interrupt; 526 } 527 } 528 529 /* irq was successfully handed to the guest */ 530 kfree(inti); 531 kvm_s390_set_psw_cc(vcpu, 1); 532 return 0; 533 reinject_interrupt: 534 /* 535 * If we encounter a problem storing the interruption code, the 536 * instruction is suppressed from the guest's view: reinject the 537 * interrupt. 538 */ 539 if (kvm_s390_reinject_io_int(vcpu->kvm, inti)) { 540 kfree(inti); 541 rc = -EFAULT; 542 } 543 /* don't set the cc, a pgm irq was injected or we drop to user space */ 544 return rc ? -EFAULT : 0; 545 } 546 547 static int handle_tsch(struct kvm_vcpu *vcpu) 548 { 549 struct kvm_s390_interrupt_info *inti = NULL; 550 const u64 isc_mask = 0xffUL << 24; /* all iscs set */ 551 552 vcpu->stat.instruction_tsch++; 553 554 /* a valid schid has at least one bit set */ 555 if (vcpu->run->s.regs.gprs[1]) 556 inti = kvm_s390_get_io_int(vcpu->kvm, isc_mask, 557 vcpu->run->s.regs.gprs[1]); 558 559 /* 560 * Prepare exit to userspace. 561 * We indicate whether we dequeued a pending I/O interrupt 562 * so that userspace can re-inject it if the instruction gets 563 * a program check. While this may re-order the pending I/O 564 * interrupts, this is no problem since the priority is kept 565 * intact. 566 */ 567 vcpu->run->exit_reason = KVM_EXIT_S390_TSCH; 568 vcpu->run->s390_tsch.dequeued = !!inti; 569 if (inti) { 570 vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id; 571 vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr; 572 vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm; 573 vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word; 574 } 575 vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb; 576 kfree(inti); 577 return -EREMOTE; 578 } 579 580 static int handle_io_inst(struct kvm_vcpu *vcpu) 581 { 582 VCPU_EVENT(vcpu, 4, "%s", "I/O instruction"); 583 584 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 585 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 586 587 if (vcpu->kvm->arch.css_support) { 588 /* 589 * Most I/O instructions will be handled by userspace. 590 * Exceptions are tpi and the interrupt portion of tsch. 591 */ 592 if (vcpu->arch.sie_block->ipa == 0xb236) 593 return handle_tpi(vcpu); 594 if (vcpu->arch.sie_block->ipa == 0xb235) 595 return handle_tsch(vcpu); 596 /* Handle in userspace. */ 597 vcpu->stat.instruction_io_other++; 598 return -EOPNOTSUPP; 599 } else { 600 /* 601 * Set condition code 3 to stop the guest from issuing channel 602 * I/O instructions. 603 */ 604 kvm_s390_set_psw_cc(vcpu, 3); 605 return 0; 606 } 607 } 608 609 /* 610 * handle_pqap: Handling pqap interception 611 * @vcpu: the vcpu having issue the pqap instruction 612 * 613 * We now support PQAP/AQIC instructions and we need to correctly 614 * answer the guest even if no dedicated driver's hook is available. 615 * 616 * The intercepting code calls a dedicated callback for this instruction 617 * if a driver did register one in the CRYPTO satellite of the 618 * SIE block. 619 * 620 * If no callback is available, the queues are not available, return this 621 * response code to the caller and set CC to 3. 622 * Else return the response code returned by the callback. 623 */ 624 static int handle_pqap(struct kvm_vcpu *vcpu) 625 { 626 struct ap_queue_status status = {}; 627 crypto_hook pqap_hook; 628 unsigned long reg0; 629 int ret; 630 uint8_t fc; 631 632 /* Verify that the AP instruction are available */ 633 if (!ap_instructions_available()) 634 return -EOPNOTSUPP; 635 /* Verify that the guest is allowed to use AP instructions */ 636 if (!(vcpu->arch.sie_block->eca & ECA_APIE)) 637 return -EOPNOTSUPP; 638 /* 639 * The only possibly intercepted functions when AP instructions are 640 * available for the guest are AQIC and TAPQ with the t bit set 641 * since we do not set IC.3 (FIII) we currently will only intercept 642 * the AQIC function code. 643 * Note: running nested under z/VM can result in intercepts for other 644 * function codes, e.g. PQAP(QCI). We do not support this and bail out. 645 */ 646 reg0 = vcpu->run->s.regs.gprs[0]; 647 fc = (reg0 >> 24) & 0xff; 648 if (fc != 0x03) 649 return -EOPNOTSUPP; 650 651 /* PQAP instruction is allowed for guest kernel only */ 652 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 653 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 654 655 /* Common PQAP instruction specification exceptions */ 656 /* bits 41-47 must all be zeros */ 657 if (reg0 & 0x007f0000UL) 658 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 659 /* APFT not install and T bit set */ 660 if (!test_kvm_facility(vcpu->kvm, 15) && (reg0 & 0x00800000UL)) 661 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 662 /* APXA not installed and APID greater 64 or APQI greater 16 */ 663 if (!(vcpu->kvm->arch.crypto.crycbd & 0x02) && (reg0 & 0x0000c0f0UL)) 664 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 665 666 /* AQIC function code specific exception */ 667 /* facility 65 not present for AQIC function code */ 668 if (!test_kvm_facility(vcpu->kvm, 65)) 669 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 670 671 /* 672 * If the hook callback is registered, there will be a pointer to the 673 * hook function pointer in the kvm_s390_crypto structure. Lock the 674 * owner, retrieve the hook function pointer and call the hook. 675 */ 676 down_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem); 677 if (vcpu->kvm->arch.crypto.pqap_hook) { 678 pqap_hook = *vcpu->kvm->arch.crypto.pqap_hook; 679 ret = pqap_hook(vcpu); 680 if (!ret && vcpu->run->s.regs.gprs[1] & 0x00ff0000) 681 kvm_s390_set_psw_cc(vcpu, 3); 682 up_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem); 683 return ret; 684 } 685 up_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem); 686 /* 687 * A vfio_driver must register a hook. 688 * No hook means no driver to enable the SIE CRYCB and no queues. 689 * We send this response to the guest. 690 */ 691 status.response_code = 0x01; 692 memcpy(&vcpu->run->s.regs.gprs[1], &status, sizeof(status)); 693 kvm_s390_set_psw_cc(vcpu, 3); 694 return 0; 695 } 696 697 static int handle_stfl(struct kvm_vcpu *vcpu) 698 { 699 int rc; 700 unsigned int fac; 701 702 vcpu->stat.instruction_stfl++; 703 704 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 705 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 706 707 /* 708 * We need to shift the lower 32 facility bits (bit 0-31) from a u64 709 * into a u32 memory representation. They will remain bits 0-31. 710 */ 711 fac = *vcpu->kvm->arch.model.fac_list >> 32; 712 rc = write_guest_lc(vcpu, offsetof(struct lowcore, stfl_fac_list), 713 &fac, sizeof(fac)); 714 if (rc) 715 return rc; 716 VCPU_EVENT(vcpu, 3, "STFL: store facility list 0x%x", fac); 717 trace_kvm_s390_handle_stfl(vcpu, fac); 718 return 0; 719 } 720 721 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA) 722 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL 723 #define PSW_ADDR_24 0x0000000000ffffffUL 724 #define PSW_ADDR_31 0x000000007fffffffUL 725 726 int is_valid_psw(psw_t *psw) 727 { 728 if (psw->mask & PSW_MASK_UNASSIGNED) 729 return 0; 730 if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) { 731 if (psw->addr & ~PSW_ADDR_31) 732 return 0; 733 } 734 if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24)) 735 return 0; 736 if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_EA) 737 return 0; 738 if (psw->addr & 1) 739 return 0; 740 return 1; 741 } 742 743 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu) 744 { 745 psw_t *gpsw = &vcpu->arch.sie_block->gpsw; 746 psw_compat_t new_psw; 747 u64 addr; 748 int rc; 749 u8 ar; 750 751 vcpu->stat.instruction_lpsw++; 752 753 if (gpsw->mask & PSW_MASK_PSTATE) 754 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 755 756 addr = kvm_s390_get_base_disp_s(vcpu, &ar); 757 if (addr & 7) 758 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 759 760 rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw)); 761 if (rc) 762 return kvm_s390_inject_prog_cond(vcpu, rc); 763 if (!(new_psw.mask & PSW32_MASK_BASE)) 764 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 765 gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32; 766 gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE; 767 gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE; 768 if (!is_valid_psw(gpsw)) 769 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 770 return 0; 771 } 772 773 static int handle_lpswe(struct kvm_vcpu *vcpu) 774 { 775 psw_t new_psw; 776 u64 addr; 777 int rc; 778 u8 ar; 779 780 vcpu->stat.instruction_lpswe++; 781 782 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 783 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 784 785 addr = kvm_s390_get_base_disp_s(vcpu, &ar); 786 if (addr & 7) 787 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 788 rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw)); 789 if (rc) 790 return kvm_s390_inject_prog_cond(vcpu, rc); 791 vcpu->arch.sie_block->gpsw = new_psw; 792 if (!is_valid_psw(&vcpu->arch.sie_block->gpsw)) 793 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 794 return 0; 795 } 796 797 static int handle_stidp(struct kvm_vcpu *vcpu) 798 { 799 u64 stidp_data = vcpu->kvm->arch.model.cpuid; 800 u64 operand2; 801 int rc; 802 u8 ar; 803 804 vcpu->stat.instruction_stidp++; 805 806 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 807 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 808 809 operand2 = kvm_s390_get_base_disp_s(vcpu, &ar); 810 811 if (operand2 & 7) 812 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 813 814 rc = write_guest(vcpu, operand2, ar, &stidp_data, sizeof(stidp_data)); 815 if (rc) 816 return kvm_s390_inject_prog_cond(vcpu, rc); 817 818 VCPU_EVENT(vcpu, 3, "STIDP: store cpu id 0x%llx", stidp_data); 819 return 0; 820 } 821 822 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem) 823 { 824 int cpus = 0; 825 int n; 826 827 cpus = atomic_read(&vcpu->kvm->online_vcpus); 828 829 /* deal with other level 3 hypervisors */ 830 if (stsi(mem, 3, 2, 2)) 831 mem->count = 0; 832 if (mem->count < 8) 833 mem->count++; 834 for (n = mem->count - 1; n > 0 ; n--) 835 memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0])); 836 837 memset(&mem->vm[0], 0, sizeof(mem->vm[0])); 838 mem->vm[0].cpus_total = cpus; 839 mem->vm[0].cpus_configured = cpus; 840 mem->vm[0].cpus_standby = 0; 841 mem->vm[0].cpus_reserved = 0; 842 mem->vm[0].caf = 1000; 843 memcpy(mem->vm[0].name, "KVMguest", 8); 844 ASCEBC(mem->vm[0].name, 8); 845 memcpy(mem->vm[0].cpi, "KVM/Linux ", 16); 846 ASCEBC(mem->vm[0].cpi, 16); 847 } 848 849 static void insert_stsi_usr_data(struct kvm_vcpu *vcpu, u64 addr, u8 ar, 850 u8 fc, u8 sel1, u16 sel2) 851 { 852 vcpu->run->exit_reason = KVM_EXIT_S390_STSI; 853 vcpu->run->s390_stsi.addr = addr; 854 vcpu->run->s390_stsi.ar = ar; 855 vcpu->run->s390_stsi.fc = fc; 856 vcpu->run->s390_stsi.sel1 = sel1; 857 vcpu->run->s390_stsi.sel2 = sel2; 858 } 859 860 static int handle_stsi(struct kvm_vcpu *vcpu) 861 { 862 int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28; 863 int sel1 = vcpu->run->s.regs.gprs[0] & 0xff; 864 int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff; 865 unsigned long mem = 0; 866 u64 operand2; 867 int rc = 0; 868 u8 ar; 869 870 vcpu->stat.instruction_stsi++; 871 VCPU_EVENT(vcpu, 3, "STSI: fc: %u sel1: %u sel2: %u", fc, sel1, sel2); 872 873 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 874 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 875 876 if (fc > 3) { 877 kvm_s390_set_psw_cc(vcpu, 3); 878 return 0; 879 } 880 881 if (vcpu->run->s.regs.gprs[0] & 0x0fffff00 882 || vcpu->run->s.regs.gprs[1] & 0xffff0000) 883 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 884 885 if (fc == 0) { 886 vcpu->run->s.regs.gprs[0] = 3 << 28; 887 kvm_s390_set_psw_cc(vcpu, 0); 888 return 0; 889 } 890 891 operand2 = kvm_s390_get_base_disp_s(vcpu, &ar); 892 893 if (!kvm_s390_pv_cpu_is_protected(vcpu) && (operand2 & 0xfff)) 894 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 895 896 switch (fc) { 897 case 1: /* same handling for 1 and 2 */ 898 case 2: 899 mem = get_zeroed_page(GFP_KERNEL_ACCOUNT); 900 if (!mem) 901 goto out_no_data; 902 if (stsi((void *) mem, fc, sel1, sel2)) 903 goto out_no_data; 904 break; 905 case 3: 906 if (sel1 != 2 || sel2 != 2) 907 goto out_no_data; 908 mem = get_zeroed_page(GFP_KERNEL_ACCOUNT); 909 if (!mem) 910 goto out_no_data; 911 handle_stsi_3_2_2(vcpu, (void *) mem); 912 break; 913 } 914 if (kvm_s390_pv_cpu_is_protected(vcpu)) { 915 memcpy((void *)sida_origin(vcpu->arch.sie_block), (void *)mem, 916 PAGE_SIZE); 917 rc = 0; 918 } else { 919 rc = write_guest(vcpu, operand2, ar, (void *)mem, PAGE_SIZE); 920 } 921 if (rc) { 922 rc = kvm_s390_inject_prog_cond(vcpu, rc); 923 goto out; 924 } 925 if (vcpu->kvm->arch.user_stsi) { 926 insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2); 927 rc = -EREMOTE; 928 } 929 trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2); 930 free_page(mem); 931 kvm_s390_set_psw_cc(vcpu, 0); 932 vcpu->run->s.regs.gprs[0] = 0; 933 return rc; 934 out_no_data: 935 kvm_s390_set_psw_cc(vcpu, 3); 936 out: 937 free_page(mem); 938 return rc; 939 } 940 941 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu) 942 { 943 switch (vcpu->arch.sie_block->ipa & 0x00ff) { 944 case 0x02: 945 return handle_stidp(vcpu); 946 case 0x04: 947 return handle_set_clock(vcpu); 948 case 0x10: 949 return handle_set_prefix(vcpu); 950 case 0x11: 951 return handle_store_prefix(vcpu); 952 case 0x12: 953 return handle_store_cpu_address(vcpu); 954 case 0x14: 955 return kvm_s390_handle_vsie(vcpu); 956 case 0x21: 957 case 0x50: 958 return handle_ipte_interlock(vcpu); 959 case 0x29: 960 return handle_iske(vcpu); 961 case 0x2a: 962 return handle_rrbe(vcpu); 963 case 0x2b: 964 return handle_sske(vcpu); 965 case 0x2c: 966 return handle_test_block(vcpu); 967 case 0x30: 968 case 0x31: 969 case 0x32: 970 case 0x33: 971 case 0x34: 972 case 0x35: 973 case 0x36: 974 case 0x37: 975 case 0x38: 976 case 0x39: 977 case 0x3a: 978 case 0x3b: 979 case 0x3c: 980 case 0x5f: 981 case 0x74: 982 case 0x76: 983 return handle_io_inst(vcpu); 984 case 0x56: 985 return handle_sthyi(vcpu); 986 case 0x7d: 987 return handle_stsi(vcpu); 988 case 0xaf: 989 return handle_pqap(vcpu); 990 case 0xb1: 991 return handle_stfl(vcpu); 992 case 0xb2: 993 return handle_lpswe(vcpu); 994 default: 995 return -EOPNOTSUPP; 996 } 997 } 998 999 static int handle_epsw(struct kvm_vcpu *vcpu) 1000 { 1001 int reg1, reg2; 1002 1003 vcpu->stat.instruction_epsw++; 1004 1005 kvm_s390_get_regs_rre(vcpu, ®1, ®2); 1006 1007 /* This basically extracts the mask half of the psw. */ 1008 vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL; 1009 vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32; 1010 if (reg2) { 1011 vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL; 1012 vcpu->run->s.regs.gprs[reg2] |= 1013 vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL; 1014 } 1015 return 0; 1016 } 1017 1018 #define PFMF_RESERVED 0xfffc0101UL 1019 #define PFMF_SK 0x00020000UL 1020 #define PFMF_CF 0x00010000UL 1021 #define PFMF_UI 0x00008000UL 1022 #define PFMF_FSC 0x00007000UL 1023 #define PFMF_NQ 0x00000800UL 1024 #define PFMF_MR 0x00000400UL 1025 #define PFMF_MC 0x00000200UL 1026 #define PFMF_KEY 0x000000feUL 1027 1028 static int handle_pfmf(struct kvm_vcpu *vcpu) 1029 { 1030 bool mr = false, mc = false, nq; 1031 int reg1, reg2; 1032 unsigned long start, end; 1033 unsigned char key; 1034 1035 vcpu->stat.instruction_pfmf++; 1036 1037 kvm_s390_get_regs_rre(vcpu, ®1, ®2); 1038 1039 if (!test_kvm_facility(vcpu->kvm, 8)) 1040 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION); 1041 1042 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 1043 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 1044 1045 if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED) 1046 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 1047 1048 /* Only provide non-quiescing support if enabled for the guest */ 1049 if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ && 1050 !test_kvm_facility(vcpu->kvm, 14)) 1051 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 1052 1053 /* Only provide conditional-SSKE support if enabled for the guest */ 1054 if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK && 1055 test_kvm_facility(vcpu->kvm, 10)) { 1056 mr = vcpu->run->s.regs.gprs[reg1] & PFMF_MR; 1057 mc = vcpu->run->s.regs.gprs[reg1] & PFMF_MC; 1058 } 1059 1060 nq = vcpu->run->s.regs.gprs[reg1] & PFMF_NQ; 1061 key = vcpu->run->s.regs.gprs[reg1] & PFMF_KEY; 1062 start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK; 1063 start = kvm_s390_logical_to_effective(vcpu, start); 1064 1065 if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) { 1066 if (kvm_s390_check_low_addr_prot_real(vcpu, start)) 1067 return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm); 1068 } 1069 1070 switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) { 1071 case 0x00000000: 1072 /* only 4k frames specify a real address */ 1073 start = kvm_s390_real_to_abs(vcpu, start); 1074 end = (start + PAGE_SIZE) & ~(PAGE_SIZE - 1); 1075 break; 1076 case 0x00001000: 1077 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1); 1078 break; 1079 case 0x00002000: 1080 /* only support 2G frame size if EDAT2 is available and we are 1081 not in 24-bit addressing mode */ 1082 if (!test_kvm_facility(vcpu->kvm, 78) || 1083 psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_24BIT) 1084 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 1085 end = (start + _REGION3_SIZE) & ~(_REGION3_SIZE - 1); 1086 break; 1087 default: 1088 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 1089 } 1090 1091 while (start != end) { 1092 unsigned long vmaddr; 1093 bool unlocked = false; 1094 1095 /* Translate guest address to host address */ 1096 vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start)); 1097 if (kvm_is_error_hva(vmaddr)) 1098 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING); 1099 1100 if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) { 1101 if (kvm_clear_guest(vcpu->kvm, start, PAGE_SIZE)) 1102 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING); 1103 } 1104 1105 if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) { 1106 int rc = kvm_s390_skey_check_enable(vcpu); 1107 1108 if (rc) 1109 return rc; 1110 mmap_read_lock(current->mm); 1111 rc = cond_set_guest_storage_key(current->mm, vmaddr, 1112 key, NULL, nq, mr, mc); 1113 if (rc < 0) { 1114 rc = fixup_user_fault(current->mm, vmaddr, 1115 FAULT_FLAG_WRITE, &unlocked); 1116 rc = !rc ? -EAGAIN : rc; 1117 } 1118 mmap_read_unlock(current->mm); 1119 if (rc == -EFAULT) 1120 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING); 1121 if (rc == -EAGAIN) 1122 continue; 1123 if (rc < 0) 1124 return rc; 1125 } 1126 start += PAGE_SIZE; 1127 } 1128 if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) { 1129 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT) { 1130 vcpu->run->s.regs.gprs[reg2] = end; 1131 } else { 1132 vcpu->run->s.regs.gprs[reg2] &= ~0xffffffffUL; 1133 end = kvm_s390_logical_to_effective(vcpu, end); 1134 vcpu->run->s.regs.gprs[reg2] |= end; 1135 } 1136 } 1137 return 0; 1138 } 1139 1140 /* 1141 * Must be called with relevant read locks held (kvm->mm->mmap_lock, kvm->srcu) 1142 */ 1143 static inline int __do_essa(struct kvm_vcpu *vcpu, const int orc) 1144 { 1145 int r1, r2, nappended, entries; 1146 unsigned long gfn, hva, res, pgstev, ptev; 1147 unsigned long *cbrlo; 1148 1149 /* 1150 * We don't need to set SD.FPF.SK to 1 here, because if we have a 1151 * machine check here we either handle it or crash 1152 */ 1153 1154 kvm_s390_get_regs_rre(vcpu, &r1, &r2); 1155 gfn = vcpu->run->s.regs.gprs[r2] >> PAGE_SHIFT; 1156 hva = gfn_to_hva(vcpu->kvm, gfn); 1157 entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3; 1158 1159 if (kvm_is_error_hva(hva)) 1160 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING); 1161 1162 nappended = pgste_perform_essa(vcpu->kvm->mm, hva, orc, &ptev, &pgstev); 1163 if (nappended < 0) { 1164 res = orc ? 0x10 : 0; 1165 vcpu->run->s.regs.gprs[r1] = res; /* Exception Indication */ 1166 return 0; 1167 } 1168 res = (pgstev & _PGSTE_GPS_USAGE_MASK) >> 22; 1169 /* 1170 * Set the block-content state part of the result. 0 means resident, so 1171 * nothing to do if the page is valid. 2 is for preserved pages 1172 * (non-present and non-zero), and 3 for zero pages (non-present and 1173 * zero). 1174 */ 1175 if (ptev & _PAGE_INVALID) { 1176 res |= 2; 1177 if (pgstev & _PGSTE_GPS_ZERO) 1178 res |= 1; 1179 } 1180 if (pgstev & _PGSTE_GPS_NODAT) 1181 res |= 0x20; 1182 vcpu->run->s.regs.gprs[r1] = res; 1183 /* 1184 * It is possible that all the normal 511 slots were full, in which case 1185 * we will now write in the 512th slot, which is reserved for host use. 1186 * In both cases we let the normal essa handling code process all the 1187 * slots, including the reserved one, if needed. 1188 */ 1189 if (nappended > 0) { 1190 cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo & PAGE_MASK); 1191 cbrlo[entries] = gfn << PAGE_SHIFT; 1192 } 1193 1194 if (orc) { 1195 struct kvm_memory_slot *ms = gfn_to_memslot(vcpu->kvm, gfn); 1196 1197 /* Increment only if we are really flipping the bit */ 1198 if (ms && !test_and_set_bit(gfn - ms->base_gfn, kvm_second_dirty_bitmap(ms))) 1199 atomic64_inc(&vcpu->kvm->arch.cmma_dirty_pages); 1200 } 1201 1202 return nappended; 1203 } 1204 1205 static int handle_essa(struct kvm_vcpu *vcpu) 1206 { 1207 /* entries expected to be 1FF */ 1208 int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3; 1209 unsigned long *cbrlo; 1210 struct gmap *gmap; 1211 int i, orc; 1212 1213 VCPU_EVENT(vcpu, 4, "ESSA: release %d pages", entries); 1214 gmap = vcpu->arch.gmap; 1215 vcpu->stat.instruction_essa++; 1216 if (!vcpu->kvm->arch.use_cmma) 1217 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION); 1218 1219 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 1220 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 1221 /* Check for invalid operation request code */ 1222 orc = (vcpu->arch.sie_block->ipb & 0xf0000000) >> 28; 1223 /* ORCs 0-6 are always valid */ 1224 if (orc > (test_kvm_facility(vcpu->kvm, 147) ? ESSA_SET_STABLE_NODAT 1225 : ESSA_SET_STABLE_IF_RESIDENT)) 1226 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 1227 1228 if (!vcpu->kvm->arch.migration_mode) { 1229 /* 1230 * CMMA is enabled in the KVM settings, but is disabled in 1231 * the SIE block and in the mm_context, and we are not doing 1232 * a migration. Enable CMMA in the mm_context. 1233 * Since we need to take a write lock to write to the context 1234 * to avoid races with storage keys handling, we check if the 1235 * value really needs to be written to; if the value is 1236 * already correct, we do nothing and avoid the lock. 1237 */ 1238 if (vcpu->kvm->mm->context.uses_cmm == 0) { 1239 mmap_write_lock(vcpu->kvm->mm); 1240 vcpu->kvm->mm->context.uses_cmm = 1; 1241 mmap_write_unlock(vcpu->kvm->mm); 1242 } 1243 /* 1244 * If we are here, we are supposed to have CMMA enabled in 1245 * the SIE block. Enabling CMMA works on a per-CPU basis, 1246 * while the context use_cmma flag is per process. 1247 * It's possible that the context flag is enabled and the 1248 * SIE flag is not, so we set the flag always; if it was 1249 * already set, nothing changes, otherwise we enable it 1250 * on this CPU too. 1251 */ 1252 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA; 1253 /* Retry the ESSA instruction */ 1254 kvm_s390_retry_instr(vcpu); 1255 } else { 1256 int srcu_idx; 1257 1258 mmap_read_lock(vcpu->kvm->mm); 1259 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); 1260 i = __do_essa(vcpu, orc); 1261 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx); 1262 mmap_read_unlock(vcpu->kvm->mm); 1263 if (i < 0) 1264 return i; 1265 /* Account for the possible extra cbrl entry */ 1266 entries += i; 1267 } 1268 vcpu->arch.sie_block->cbrlo &= PAGE_MASK; /* reset nceo */ 1269 cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo); 1270 mmap_read_lock(gmap->mm); 1271 for (i = 0; i < entries; ++i) 1272 __gmap_zap(gmap, cbrlo[i]); 1273 mmap_read_unlock(gmap->mm); 1274 return 0; 1275 } 1276 1277 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu) 1278 { 1279 switch (vcpu->arch.sie_block->ipa & 0x00ff) { 1280 case 0x8a: 1281 case 0x8e: 1282 case 0x8f: 1283 return handle_ipte_interlock(vcpu); 1284 case 0x8d: 1285 return handle_epsw(vcpu); 1286 case 0xab: 1287 return handle_essa(vcpu); 1288 case 0xaf: 1289 return handle_pfmf(vcpu); 1290 default: 1291 return -EOPNOTSUPP; 1292 } 1293 } 1294 1295 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu) 1296 { 1297 int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4; 1298 int reg3 = vcpu->arch.sie_block->ipa & 0x000f; 1299 int reg, rc, nr_regs; 1300 u32 ctl_array[16]; 1301 u64 ga; 1302 u8 ar; 1303 1304 vcpu->stat.instruction_lctl++; 1305 1306 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 1307 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 1308 1309 ga = kvm_s390_get_base_disp_rs(vcpu, &ar); 1310 1311 if (ga & 3) 1312 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 1313 1314 VCPU_EVENT(vcpu, 4, "LCTL: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga); 1315 trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga); 1316 1317 nr_regs = ((reg3 - reg1) & 0xf) + 1; 1318 rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32)); 1319 if (rc) 1320 return kvm_s390_inject_prog_cond(vcpu, rc); 1321 reg = reg1; 1322 nr_regs = 0; 1323 do { 1324 vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul; 1325 vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++]; 1326 if (reg == reg3) 1327 break; 1328 reg = (reg + 1) % 16; 1329 } while (1); 1330 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); 1331 return 0; 1332 } 1333 1334 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu) 1335 { 1336 int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4; 1337 int reg3 = vcpu->arch.sie_block->ipa & 0x000f; 1338 int reg, rc, nr_regs; 1339 u32 ctl_array[16]; 1340 u64 ga; 1341 u8 ar; 1342 1343 vcpu->stat.instruction_stctl++; 1344 1345 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 1346 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 1347 1348 ga = kvm_s390_get_base_disp_rs(vcpu, &ar); 1349 1350 if (ga & 3) 1351 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 1352 1353 VCPU_EVENT(vcpu, 4, "STCTL r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga); 1354 trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga); 1355 1356 reg = reg1; 1357 nr_regs = 0; 1358 do { 1359 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg]; 1360 if (reg == reg3) 1361 break; 1362 reg = (reg + 1) % 16; 1363 } while (1); 1364 rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32)); 1365 return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0; 1366 } 1367 1368 static int handle_lctlg(struct kvm_vcpu *vcpu) 1369 { 1370 int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4; 1371 int reg3 = vcpu->arch.sie_block->ipa & 0x000f; 1372 int reg, rc, nr_regs; 1373 u64 ctl_array[16]; 1374 u64 ga; 1375 u8 ar; 1376 1377 vcpu->stat.instruction_lctlg++; 1378 1379 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 1380 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 1381 1382 ga = kvm_s390_get_base_disp_rsy(vcpu, &ar); 1383 1384 if (ga & 7) 1385 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 1386 1387 VCPU_EVENT(vcpu, 4, "LCTLG: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga); 1388 trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga); 1389 1390 nr_regs = ((reg3 - reg1) & 0xf) + 1; 1391 rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64)); 1392 if (rc) 1393 return kvm_s390_inject_prog_cond(vcpu, rc); 1394 reg = reg1; 1395 nr_regs = 0; 1396 do { 1397 vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++]; 1398 if (reg == reg3) 1399 break; 1400 reg = (reg + 1) % 16; 1401 } while (1); 1402 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); 1403 return 0; 1404 } 1405 1406 static int handle_stctg(struct kvm_vcpu *vcpu) 1407 { 1408 int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4; 1409 int reg3 = vcpu->arch.sie_block->ipa & 0x000f; 1410 int reg, rc, nr_regs; 1411 u64 ctl_array[16]; 1412 u64 ga; 1413 u8 ar; 1414 1415 vcpu->stat.instruction_stctg++; 1416 1417 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 1418 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 1419 1420 ga = kvm_s390_get_base_disp_rsy(vcpu, &ar); 1421 1422 if (ga & 7) 1423 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); 1424 1425 VCPU_EVENT(vcpu, 4, "STCTG r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga); 1426 trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga); 1427 1428 reg = reg1; 1429 nr_regs = 0; 1430 do { 1431 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg]; 1432 if (reg == reg3) 1433 break; 1434 reg = (reg + 1) % 16; 1435 } while (1); 1436 rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64)); 1437 return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0; 1438 } 1439 1440 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu) 1441 { 1442 switch (vcpu->arch.sie_block->ipb & 0x000000ff) { 1443 case 0x25: 1444 return handle_stctg(vcpu); 1445 case 0x2f: 1446 return handle_lctlg(vcpu); 1447 case 0x60: 1448 case 0x61: 1449 case 0x62: 1450 return handle_ri(vcpu); 1451 default: 1452 return -EOPNOTSUPP; 1453 } 1454 } 1455 1456 static int handle_tprot(struct kvm_vcpu *vcpu) 1457 { 1458 u64 address, operand2; 1459 unsigned long gpa; 1460 u8 access_key; 1461 bool writable; 1462 int ret, cc; 1463 u8 ar; 1464 1465 vcpu->stat.instruction_tprot++; 1466 1467 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 1468 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 1469 1470 kvm_s390_get_base_disp_sse(vcpu, &address, &operand2, &ar, NULL); 1471 access_key = (operand2 & 0xf0) >> 4; 1472 1473 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT) 1474 ipte_lock(vcpu); 1475 1476 ret = guest_translate_address_with_key(vcpu, address, ar, &gpa, 1477 GACC_STORE, access_key); 1478 if (ret == 0) { 1479 gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable); 1480 } else if (ret == PGM_PROTECTION) { 1481 writable = false; 1482 /* Write protected? Try again with read-only... */ 1483 ret = guest_translate_address_with_key(vcpu, address, ar, &gpa, 1484 GACC_FETCH, access_key); 1485 } 1486 if (ret >= 0) { 1487 cc = -1; 1488 1489 /* Fetching permitted; storing permitted */ 1490 if (ret == 0 && writable) 1491 cc = 0; 1492 /* Fetching permitted; storing not permitted */ 1493 else if (ret == 0 && !writable) 1494 cc = 1; 1495 /* Fetching not permitted; storing not permitted */ 1496 else if (ret == PGM_PROTECTION) 1497 cc = 2; 1498 /* Translation not available */ 1499 else if (ret != PGM_ADDRESSING && ret != PGM_TRANSLATION_SPEC) 1500 cc = 3; 1501 1502 if (cc != -1) { 1503 kvm_s390_set_psw_cc(vcpu, cc); 1504 ret = 0; 1505 } else { 1506 ret = kvm_s390_inject_program_int(vcpu, ret); 1507 } 1508 } 1509 1510 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT) 1511 ipte_unlock(vcpu); 1512 return ret; 1513 } 1514 1515 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu) 1516 { 1517 switch (vcpu->arch.sie_block->ipa & 0x00ff) { 1518 case 0x01: 1519 return handle_tprot(vcpu); 1520 default: 1521 return -EOPNOTSUPP; 1522 } 1523 } 1524 1525 static int handle_sckpf(struct kvm_vcpu *vcpu) 1526 { 1527 u32 value; 1528 1529 vcpu->stat.instruction_sckpf++; 1530 1531 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) 1532 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); 1533 1534 if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000) 1535 return kvm_s390_inject_program_int(vcpu, 1536 PGM_SPECIFICATION); 1537 1538 value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff; 1539 vcpu->arch.sie_block->todpr = value; 1540 1541 return 0; 1542 } 1543 1544 static int handle_ptff(struct kvm_vcpu *vcpu) 1545 { 1546 vcpu->stat.instruction_ptff++; 1547 1548 /* we don't emulate any control instructions yet */ 1549 kvm_s390_set_psw_cc(vcpu, 3); 1550 return 0; 1551 } 1552 1553 int kvm_s390_handle_01(struct kvm_vcpu *vcpu) 1554 { 1555 switch (vcpu->arch.sie_block->ipa & 0x00ff) { 1556 case 0x04: 1557 return handle_ptff(vcpu); 1558 case 0x07: 1559 return handle_sckpf(vcpu); 1560 default: 1561 return -EOPNOTSUPP; 1562 } 1563 } 1564