arch/s390/kvm/priv.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * handling privileged instructions
   4  *
   5  * Copyright IBM Corp. 2008, 2018
   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/compat.h>
  15 #include <linux/mm_types.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/pgtable.h>
  24 #include <asm/page-states.h>
  25 #include <asm/pgalloc.h>
  26 #include <asm/gmap.h>
  27 #include <asm/io.h>
  28 #include <asm/ptrace.h>
  29 #include <asm/sclp.h>
  30 #include <asm/ap.h>
  31 #include "gaccess.h"
  32 #include "kvm-s390.h"
  33 #include "trace.h"
  34
  35 static int handle_ri(struct kvm_vcpu *vcpu)
  36 {
  37         vcpu->stat.instruction_ri++;
  38
  39         if (test_kvm_facility(vcpu->kvm, 64)) {
  40                 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (lazy)");
  41                 vcpu->arch.sie_block->ecb3 |= ECB3_RI;
  42                 kvm_s390_retry_instr(vcpu);
  43                 return 0;
  44         } else
  45                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
  46 }
  47
  48 int kvm_s390_handle_aa(struct kvm_vcpu *vcpu)
  49 {
  50         if ((vcpu->arch.sie_block->ipa & 0xf) <= 4)
  51                 return handle_ri(vcpu);
  52         else
  53                 return -EOPNOTSUPP;
  54 }
  55
  56 static int handle_gs(struct kvm_vcpu *vcpu)
  57 {
  58         vcpu->stat.instruction_gs++;
  59
  60         if (test_kvm_facility(vcpu->kvm, 133)) {
  61                 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (lazy)");
  62                 preempt_disable();
  63                 __ctl_set_bit(2, 4);
  64                 current->thread.gs_cb = (struct gs_cb *)&vcpu->run->s.regs.gscb;
  65                 restore_gs_cb(current->thread.gs_cb);
  66                 preempt_enable();
  67                 vcpu->arch.sie_block->ecb |= ECB_GS;
  68                 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
  69                 vcpu->arch.gs_enabled = 1;
  70                 kvm_s390_retry_instr(vcpu);
  71                 return 0;
  72         } else
  73                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
  74 }
  75
  76 int kvm_s390_handle_e3(struct kvm_vcpu *vcpu)
  77 {
  78         int code = vcpu->arch.sie_block->ipb & 0xff;
  79
  80         if (code == 0x49 || code == 0x4d)
  81                 return handle_gs(vcpu);
  82         else
  83                 return -EOPNOTSUPP;
  84 }
  85 /* Handle SCK (SET CLOCK) interception */
  86 static int handle_set_clock(struct kvm_vcpu *vcpu)
  87 {
  88         struct kvm_s390_vm_tod_clock gtod = { 0 };
  89         int rc;
  90         u8 ar;
  91         u64 op2;
  92
  93         vcpu->stat.instruction_sck++;
  94
  95         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
  96                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
  97
  98         op2 = kvm_s390_get_base_disp_s(vcpu, &ar);
  99         if (op2 & 7)    /* Operand must be on a doubleword boundary */
 100                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 101         rc = read_guest(vcpu, op2, ar, &gtod.tod, sizeof(gtod.tod));
 102         if (rc)
 103                 return kvm_s390_inject_prog_cond(vcpu, rc);
 104
 105         VCPU_EVENT(vcpu, 3, "SCK: setting guest TOD to 0x%llx", gtod.tod);
 106         kvm_s390_set_tod_clock(vcpu->kvm, &gtod);
 107
 108         kvm_s390_set_psw_cc(vcpu, 0);
 109         return 0;
 110 }
 111
 112 static int handle_set_prefix(struct kvm_vcpu *vcpu)
 113 {
 114         u64 operand2;
 115         u32 address;
 116         int rc;
 117         u8 ar;
 118
 119         vcpu->stat.instruction_spx++;
 120
 121         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 122                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 123
 124         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 125
 126         /* must be word boundary */
 127         if (operand2 & 3)
 128                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 129
 130         /* get the value */
 131         rc = read_guest(vcpu, operand2, ar, &address, sizeof(address));
 132         if (rc)
 133                 return kvm_s390_inject_prog_cond(vcpu, rc);
 134
 135         address &= 0x7fffe000u;
 136
 137         /*
 138          * Make sure the new value is valid memory. We only need to check the
 139          * first page, since address is 8k aligned and memory pieces are always
 140          * at least 1MB aligned and have at least a size of 1MB.
 141          */
 142         if (kvm_is_error_gpa(vcpu->kvm, address))
 143                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 144
 145         kvm_s390_set_prefix(vcpu, address);
 146         trace_kvm_s390_handle_prefix(vcpu, 1, address);
 147         return 0;
 148 }
 149
 150 static int handle_store_prefix(struct kvm_vcpu *vcpu)
 151 {
 152         u64 operand2;
 153         u32 address;
 154         int rc;
 155         u8 ar;
 156
 157         vcpu->stat.instruction_stpx++;
 158
 159         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 160                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 161
 162         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 163
 164         /* must be word boundary */
 165         if (operand2 & 3)
 166                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 167
 168         address = kvm_s390_get_prefix(vcpu);
 169
 170         /* get the value */
 171         rc = write_guest(vcpu, operand2, ar, &address, sizeof(address));
 172         if (rc)
 173                 return kvm_s390_inject_prog_cond(vcpu, rc);
 174
 175         VCPU_EVENT(vcpu, 3, "STPX: storing prefix 0x%x into 0x%llx", address, operand2);
 176         trace_kvm_s390_handle_prefix(vcpu, 0, address);
 177         return 0;
 178 }
 179
 180 static int handle_store_cpu_address(struct kvm_vcpu *vcpu)
 181 {
 182         u16 vcpu_id = vcpu->vcpu_id;
 183         u64 ga;
 184         int rc;
 185         u8 ar;
 186
 187         vcpu->stat.instruction_stap++;
 188
 189         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 190                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 191
 192         ga = kvm_s390_get_base_disp_s(vcpu, &ar);
 193
 194         if (ga & 1)
 195                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 196
 197         rc = write_guest(vcpu, ga, ar, &vcpu_id, sizeof(vcpu_id));
 198         if (rc)
 199                 return kvm_s390_inject_prog_cond(vcpu, rc);
 200
 201         VCPU_EVENT(vcpu, 3, "STAP: storing cpu address (%u) to 0x%llx", vcpu_id, ga);
 202         trace_kvm_s390_handle_stap(vcpu, ga);
 203         return 0;
 204 }
 205
 206 int kvm_s390_skey_check_enable(struct kvm_vcpu *vcpu)
 207 {
 208         int rc;
 209
 210         trace_kvm_s390_skey_related_inst(vcpu);
 211         /* Already enabled? */
 212         if (vcpu->arch.skey_enabled)
 213                 return 0;
 214
 215         rc = s390_enable_skey();
 216         VCPU_EVENT(vcpu, 3, "enabling storage keys for guest: %d", rc);
 217         if (rc)
 218                 return rc;
 219
 220         if (kvm_s390_test_cpuflags(vcpu, CPUSTAT_KSS))
 221                 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_KSS);
 222         if (!vcpu->kvm->arch.use_skf)
 223                 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
 224         else
 225                 vcpu->arch.sie_block->ictl &= ~(ICTL_ISKE | ICTL_SSKE | ICTL_RRBE);
 226         vcpu->arch.skey_enabled = true;
 227         return 0;
 228 }
 229
 230 static int try_handle_skey(struct kvm_vcpu *vcpu)
 231 {
 232         int rc;
 233
 234         rc = kvm_s390_skey_check_enable(vcpu);
 235         if (rc)
 236                 return rc;
 237         if (vcpu->kvm->arch.use_skf) {
 238                 /* with storage-key facility, SIE interprets it for us */
 239                 kvm_s390_retry_instr(vcpu);
 240                 VCPU_EVENT(vcpu, 4, "%s", "retrying storage key operation");
 241                 return -EAGAIN;
 242         }
 243         return 0;
 244 }
 245
 246 static int handle_iske(struct kvm_vcpu *vcpu)
 247 {
 248         unsigned long gaddr, vmaddr;
 249         unsigned char key;
 250         int reg1, reg2;
 251         bool unlocked;
 252         int rc;
 253
 254         vcpu->stat.instruction_iske++;
 255
 256         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 257                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 258
 259         rc = try_handle_skey(vcpu);
 260         if (rc)
 261                 return rc != -EAGAIN ? rc : 0;
 262
 263         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 264
 265         gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 266         gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
 267         gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
 268         vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr));
 269         if (kvm_is_error_hva(vmaddr))
 270                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 271 retry:
 272         unlocked = false;
 273         down_read(&current->mm->mmap_sem);
 274         rc = get_guest_storage_key(current->mm, vmaddr, &key);
 275
 276         if (rc) {
 277                 rc = fixup_user_fault(current, current->mm, vmaddr,
 278                                       FAULT_FLAG_WRITE, &unlocked);
 279                 if (!rc) {
 280                         up_read(&current->mm->mmap_sem);
 281                         goto retry;
 282                 }
 283         }
 284         up_read(&current->mm->mmap_sem);
 285         if (rc == -EFAULT)
 286                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 287         if (rc < 0)
 288                 return rc;
 289         vcpu->run->s.regs.gprs[reg1] &= ~0xff;
 290         vcpu->run->s.regs.gprs[reg1] |= key;
 291         return 0;
 292 }
 293
 294 static int handle_rrbe(struct kvm_vcpu *vcpu)
 295 {
 296         unsigned long vmaddr, gaddr;
 297         int reg1, reg2;
 298         bool unlocked;
 299         int rc;
 300
 301         vcpu->stat.instruction_rrbe++;
 302
 303         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 304                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 305
 306         rc = try_handle_skey(vcpu);
 307         if (rc)
 308                 return rc != -EAGAIN ? rc : 0;
 309
 310         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 311
 312         gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 313         gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
 314         gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
 315         vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr));
 316         if (kvm_is_error_hva(vmaddr))
 317                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 318 retry:
 319         unlocked = false;
 320         down_read(&current->mm->mmap_sem);
 321         rc = reset_guest_reference_bit(current->mm, vmaddr);
 322         if (rc < 0) {
 323                 rc = fixup_user_fault(current, current->mm, vmaddr,
 324                                       FAULT_FLAG_WRITE, &unlocked);
 325                 if (!rc) {
 326                         up_read(&current->mm->mmap_sem);
 327                         goto retry;
 328                 }
 329         }
 330         up_read(&current->mm->mmap_sem);
 331         if (rc == -EFAULT)
 332                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 333         if (rc < 0)
 334                 return rc;
 335         kvm_s390_set_psw_cc(vcpu, rc);
 336         return 0;
 337 }
 338
 339 #define SSKE_NQ 0x8
 340 #define SSKE_MR 0x4
 341 #define SSKE_MC 0x2
 342 #define SSKE_MB 0x1
 343 static int handle_sske(struct kvm_vcpu *vcpu)
 344 {
 345         unsigned char m3 = vcpu->arch.sie_block->ipb >> 28;
 346         unsigned long start, end;
 347         unsigned char key, oldkey;
 348         int reg1, reg2;
 349         bool unlocked;
 350         int rc;
 351
 352         vcpu->stat.instruction_sske++;
 353
 354         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 355                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 356
 357         rc = try_handle_skey(vcpu);
 358         if (rc)
 359                 return rc != -EAGAIN ? rc : 0;
 360
 361         if (!test_kvm_facility(vcpu->kvm, 8))
 362                 m3 &= ~SSKE_MB;
 363         if (!test_kvm_facility(vcpu->kvm, 10))
 364                 m3 &= ~(SSKE_MC | SSKE_MR);
 365         if (!test_kvm_facility(vcpu->kvm, 14))
 366                 m3 &= ~SSKE_NQ;
 367
 368         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 369
 370         key = vcpu->run->s.regs.gprs[reg1] & 0xfe;
 371         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 372         start = kvm_s390_logical_to_effective(vcpu, start);
 373         if (m3 & SSKE_MB) {
 374                 /* start already designates an absolute address */
 375                 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
 376         } else {
 377                 start = kvm_s390_real_to_abs(vcpu, start);
 378                 end = start + PAGE_SIZE;
 379         }
 380
 381         while (start != end) {
 382                 unsigned long vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
 383                 unlocked = false;
 384
 385                 if (kvm_is_error_hva(vmaddr))
 386                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 387
 388                 down_read(&current->mm->mmap_sem);
 389                 rc = cond_set_guest_storage_key(current->mm, vmaddr, key, &oldkey,
 390                                                 m3 & SSKE_NQ, m3 & SSKE_MR,
 391                                                 m3 & SSKE_MC);
 392
 393                 if (rc < 0) {
 394                         rc = fixup_user_fault(current, current->mm, vmaddr,
 395                                               FAULT_FLAG_WRITE, &unlocked);
 396                         rc = !rc ? -EAGAIN : rc;
 397                 }
 398                 up_read(&current->mm->mmap_sem);
 399                 if (rc == -EFAULT)
 400                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 401                 if (rc < 0)
 402                         return rc;
 403                 start += PAGE_SIZE;
 404         }
 405
 406         if (m3 & (SSKE_MC | SSKE_MR)) {
 407                 if (m3 & SSKE_MB) {
 408                         /* skey in reg1 is unpredictable */
 409                         kvm_s390_set_psw_cc(vcpu, 3);
 410                 } else {
 411                         kvm_s390_set_psw_cc(vcpu, rc);
 412                         vcpu->run->s.regs.gprs[reg1] &= ~0xff00UL;
 413                         vcpu->run->s.regs.gprs[reg1] |= (u64) oldkey << 8;
 414                 }
 415         }
 416         if (m3 & SSKE_MB) {
 417                 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT)
 418                         vcpu->run->s.regs.gprs[reg2] &= ~PAGE_MASK;
 419                 else
 420                         vcpu->run->s.regs.gprs[reg2] &= ~0xfffff000UL;
 421                 end = kvm_s390_logical_to_effective(vcpu, end);
 422                 vcpu->run->s.regs.gprs[reg2] |= end;
 423         }
 424         return 0;
 425 }
 426
 427 static int handle_ipte_interlock(struct kvm_vcpu *vcpu)
 428 {
 429         vcpu->stat.instruction_ipte_interlock++;
 430         if (psw_bits(vcpu->arch.sie_block->gpsw).pstate)
 431                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 432         wait_event(vcpu->kvm->arch.ipte_wq, !ipte_lock_held(vcpu));
 433         kvm_s390_retry_instr(vcpu);
 434         VCPU_EVENT(vcpu, 4, "%s", "retrying ipte interlock operation");
 435         return 0;
 436 }
 437
 438 static int handle_test_block(struct kvm_vcpu *vcpu)
 439 {
 440         gpa_t addr;
 441         int reg2;
 442
 443         vcpu->stat.instruction_tb++;
 444
 445         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 446                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 447
 448         kvm_s390_get_regs_rre(vcpu, NULL, &reg2);
 449         addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 450         addr = kvm_s390_logical_to_effective(vcpu, addr);
 451         if (kvm_s390_check_low_addr_prot_real(vcpu, addr))
 452                 return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
 453         addr = kvm_s390_real_to_abs(vcpu, addr);
 454
 455         if (kvm_is_error_gpa(vcpu->kvm, addr))
 456                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 457         /*
 458          * We don't expect errors on modern systems, and do not care
 459          * about storage keys (yet), so let's just clear the page.
 460          */
 461         if (kvm_clear_guest(vcpu->kvm, addr, PAGE_SIZE))
 462                 return -EFAULT;
 463         kvm_s390_set_psw_cc(vcpu, 0);
 464         vcpu->run->s.regs.gprs[0] = 0;
 465         return 0;
 466 }
 467
 468 static int handle_tpi(struct kvm_vcpu *vcpu)
 469 {
 470         struct kvm_s390_interrupt_info *inti;
 471         unsigned long len;
 472         u32 tpi_data[3];
 473         int rc;
 474         u64 addr;
 475         u8 ar;
 476
 477         vcpu->stat.instruction_tpi++;
 478
 479         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 480         if (addr & 3)
 481                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 482
 483         inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0);
 484         if (!inti) {
 485                 kvm_s390_set_psw_cc(vcpu, 0);
 486                 return 0;
 487         }
 488
 489         tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr;
 490         tpi_data[1] = inti->io.io_int_parm;
 491         tpi_data[2] = inti->io.io_int_word;
 492         if (addr) {
 493                 /*
 494                  * Store the two-word I/O interruption code into the
 495                  * provided area.
 496                  */
 497                 len = sizeof(tpi_data) - 4;
 498                 rc = write_guest(vcpu, addr, ar, &tpi_data, len);
 499                 if (rc) {
 500                         rc = kvm_s390_inject_prog_cond(vcpu, rc);
 501                         goto reinject_interrupt;
 502                 }
 503         } else {
 504                 /*
 505                  * Store the three-word I/O interruption code into
 506                  * the appropriate lowcore area.
 507                  */
 508                 len = sizeof(tpi_data);
 509                 if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len)) {
 510                         /* failed writes to the low core are not recoverable */
 511                         rc = -EFAULT;
 512                         goto reinject_interrupt;
 513                 }
 514         }
 515
 516         /* irq was successfully handed to the guest */
 517         kfree(inti);
 518         kvm_s390_set_psw_cc(vcpu, 1);
 519         return 0;
 520 reinject_interrupt:
 521         /*
 522          * If we encounter a problem storing the interruption code, the
 523          * instruction is suppressed from the guest's view: reinject the
 524          * interrupt.
 525          */
 526         if (kvm_s390_reinject_io_int(vcpu->kvm, inti)) {
 527                 kfree(inti);
 528                 rc = -EFAULT;
 529         }
 530         /* don't set the cc, a pgm irq was injected or we drop to user space */
 531         return rc ? -EFAULT : 0;
 532 }
 533
 534 static int handle_tsch(struct kvm_vcpu *vcpu)
 535 {
 536         struct kvm_s390_interrupt_info *inti = NULL;
 537         const u64 isc_mask = 0xffUL << 24; /* all iscs set */
 538
 539         vcpu->stat.instruction_tsch++;
 540
 541         /* a valid schid has at least one bit set */
 542         if (vcpu->run->s.regs.gprs[1])
 543                 inti = kvm_s390_get_io_int(vcpu->kvm, isc_mask,
 544                                            vcpu->run->s.regs.gprs[1]);
 545
 546         /*
 547          * Prepare exit to userspace.
 548          * We indicate whether we dequeued a pending I/O interrupt
 549          * so that userspace can re-inject it if the instruction gets
 550          * a program check. While this may re-order the pending I/O
 551          * interrupts, this is no problem since the priority is kept
 552          * intact.
 553          */
 554         vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
 555         vcpu->run->s390_tsch.dequeued = !!inti;
 556         if (inti) {
 557                 vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
 558                 vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
 559                 vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
 560                 vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
 561         }
 562         vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
 563         kfree(inti);
 564         return -EREMOTE;
 565 }
 566
 567 static int handle_io_inst(struct kvm_vcpu *vcpu)
 568 {
 569         VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
 570
 571         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 572                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 573
 574         if (vcpu->kvm->arch.css_support) {
 575                 /*
 576                  * Most I/O instructions will be handled by userspace.
 577                  * Exceptions are tpi and the interrupt portion of tsch.
 578                  */
 579                 if (vcpu->arch.sie_block->ipa == 0xb236)
 580                         return handle_tpi(vcpu);
 581                 if (vcpu->arch.sie_block->ipa == 0xb235)
 582                         return handle_tsch(vcpu);
 583                 /* Handle in userspace. */
 584                 vcpu->stat.instruction_io_other++;
 585                 return -EOPNOTSUPP;
 586         } else {
 587                 /*
 588                  * Set condition code 3 to stop the guest from issuing channel
 589                  * I/O instructions.
 590                  */
 591                 kvm_s390_set_psw_cc(vcpu, 3);
 592                 return 0;
 593         }
 594 }
 595
 596 /*
 597  * handle_pqap: Handling pqap interception
 598  * @vcpu: the vcpu having issue the pqap instruction
 599  *
 600  * We now support PQAP/AQIC instructions and we need to correctly
 601  * answer the guest even if no dedicated driver's hook is available.
 602  *
 603  * The intercepting code calls a dedicated callback for this instruction
 604  * if a driver did register one in the CRYPTO satellite of the
 605  * SIE block.
 606  *
 607  * If no callback is available, the queues are not available, return this
 608  * response code to the caller and set CC to 3.
 609  * Else return the response code returned by the callback.
 610  */
 611 static int handle_pqap(struct kvm_vcpu *vcpu)
 612 {
 613         struct ap_queue_status status = {};
 614         unsigned long reg0;
 615         int ret;
 616         uint8_t fc;
 617
 618         /* Verify that the AP instruction are available */
 619         if (!ap_instructions_available())
 620                 return -EOPNOTSUPP;
 621         /* Verify that the guest is allowed to use AP instructions */
 622         if (!(vcpu->arch.sie_block->eca & ECA_APIE))
 623                 return -EOPNOTSUPP;
 624         /*
 625          * The only possibly intercepted functions when AP instructions are
 626          * available for the guest are AQIC and TAPQ with the t bit set
 627          * since we do not set IC.3 (FIII) we currently will only intercept
 628          * the AQIC function code.
 629          */
 630         reg0 = vcpu->run->s.regs.gprs[0];
 631         fc = (reg0 >> 24) & 0xff;
 632         if (WARN_ON_ONCE(fc != 0x03))
 633                 return -EOPNOTSUPP;
 634
 635         /* PQAP instruction is allowed for guest kernel only */
 636         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 637                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 638
 639         /* Common PQAP instruction specification exceptions */
 640         /* bits 41-47 must all be zeros */
 641         if (reg0 & 0x007f0000UL)
 642                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 643         /* APFT not install and T bit set */
 644         if (!test_kvm_facility(vcpu->kvm, 15) && (reg0 & 0x00800000UL))
 645                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 646         /* APXA not installed and APID greater 64 or APQI greater 16 */
 647         if (!(vcpu->kvm->arch.crypto.crycbd & 0x02) && (reg0 & 0x0000c0f0UL))
 648                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 649
 650         /* AQIC function code specific exception */
 651         /* facility 65 not present for AQIC function code */
 652         if (!test_kvm_facility(vcpu->kvm, 65))
 653                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 654
 655         /*
 656          * Verify that the hook callback is registered, lock the owner
 657          * and call the hook.
 658          */
 659         if (vcpu->kvm->arch.crypto.pqap_hook) {
 660                 if (!try_module_get(vcpu->kvm->arch.crypto.pqap_hook->owner))
 661                         return -EOPNOTSUPP;
 662                 ret = vcpu->kvm->arch.crypto.pqap_hook->hook(vcpu);
 663                 module_put(vcpu->kvm->arch.crypto.pqap_hook->owner);
 664                 if (!ret && vcpu->run->s.regs.gprs[1] & 0x00ff0000)
 665                         kvm_s390_set_psw_cc(vcpu, 3);
 666                 return ret;
 667         }
 668         /*
 669          * A vfio_driver must register a hook.
 670          * No hook means no driver to enable the SIE CRYCB and no queues.
 671          * We send this response to the guest.
 672          */
 673         status.response_code = 0x01;
 674         memcpy(&vcpu->run->s.regs.gprs[1], &status, sizeof(status));
 675         kvm_s390_set_psw_cc(vcpu, 3);
 676         return 0;
 677 }
 678
 679 static int handle_stfl(struct kvm_vcpu *vcpu)
 680 {
 681         int rc;
 682         unsigned int fac;
 683
 684         vcpu->stat.instruction_stfl++;
 685
 686         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 687                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 688
 689         /*
 690          * We need to shift the lower 32 facility bits (bit 0-31) from a u64
 691          * into a u32 memory representation. They will remain bits 0-31.
 692          */
 693         fac = *vcpu->kvm->arch.model.fac_list >> 32;
 694         rc = write_guest_lc(vcpu, offsetof(struct lowcore, stfl_fac_list),
 695                             &fac, sizeof(fac));
 696         if (rc)
 697                 return rc;
 698         VCPU_EVENT(vcpu, 3, "STFL: store facility list 0x%x", fac);
 699         trace_kvm_s390_handle_stfl(vcpu, fac);
 700         return 0;
 701 }
 702
 703 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
 704 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
 705 #define PSW_ADDR_24 0x0000000000ffffffUL
 706 #define PSW_ADDR_31 0x000000007fffffffUL
 707
 708 int is_valid_psw(psw_t *psw)
 709 {
 710         if (psw->mask & PSW_MASK_UNASSIGNED)
 711                 return 0;
 712         if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
 713                 if (psw->addr & ~PSW_ADDR_31)
 714                         return 0;
 715         }
 716         if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
 717                 return 0;
 718         if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
 719                 return 0;
 720         if (psw->addr & 1)
 721                 return 0;
 722         return 1;
 723 }
 724
 725 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
 726 {
 727         psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
 728         psw_compat_t new_psw;
 729         u64 addr;
 730         int rc;
 731         u8 ar;
 732
 733         vcpu->stat.instruction_lpsw++;
 734
 735         if (gpsw->mask & PSW_MASK_PSTATE)
 736                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 737
 738         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 739         if (addr & 7)
 740                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 741
 742         rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
 743         if (rc)
 744                 return kvm_s390_inject_prog_cond(vcpu, rc);
 745         if (!(new_psw.mask & PSW32_MASK_BASE))
 746                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 747         gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
 748         gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
 749         gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
 750         if (!is_valid_psw(gpsw))
 751                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 752         return 0;
 753 }
 754
 755 static int handle_lpswe(struct kvm_vcpu *vcpu)
 756 {
 757         psw_t new_psw;
 758         u64 addr;
 759         int rc;
 760         u8 ar;
 761
 762         vcpu->stat.instruction_lpswe++;
 763
 764         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 765                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 766
 767         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 768         if (addr & 7)
 769                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 770         rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
 771         if (rc)
 772                 return kvm_s390_inject_prog_cond(vcpu, rc);
 773         vcpu->arch.sie_block->gpsw = new_psw;
 774         if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
 775                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 776         return 0;
 777 }
 778
 779 static int handle_stidp(struct kvm_vcpu *vcpu)
 780 {
 781         u64 stidp_data = vcpu->kvm->arch.model.cpuid;
 782         u64 operand2;
 783         int rc;
 784         u8 ar;
 785
 786         vcpu->stat.instruction_stidp++;
 787
 788         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 789                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 790
 791         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 792
 793         if (operand2 & 7)
 794                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 795
 796         rc = write_guest(vcpu, operand2, ar, &stidp_data, sizeof(stidp_data));
 797         if (rc)
 798                 return kvm_s390_inject_prog_cond(vcpu, rc);
 799
 800         VCPU_EVENT(vcpu, 3, "STIDP: store cpu id 0x%llx", stidp_data);
 801         return 0;
 802 }
 803
 804 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
 805 {
 806         int cpus = 0;
 807         int n;
 808
 809         cpus = atomic_read(&vcpu->kvm->online_vcpus);
 810
 811         /* deal with other level 3 hypervisors */
 812         if (stsi(mem, 3, 2, 2))
 813                 mem->count = 0;
 814         if (mem->count < 8)
 815                 mem->count++;
 816         for (n = mem->count - 1; n > 0 ; n--)
 817                 memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
 818
 819         memset(&mem->vm[0], 0, sizeof(mem->vm[0]));
 820         mem->vm[0].cpus_total = cpus;
 821         mem->vm[0].cpus_configured = cpus;
 822         mem->vm[0].cpus_standby = 0;
 823         mem->vm[0].cpus_reserved = 0;
 824         mem->vm[0].caf = 1000;
 825         memcpy(mem->vm[0].name, "KVMguest", 8);
 826         ASCEBC(mem->vm[0].name, 8);
 827         memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
 828         ASCEBC(mem->vm[0].cpi, 16);
 829 }
 830
 831 static void insert_stsi_usr_data(struct kvm_vcpu *vcpu, u64 addr, u8 ar,
 832                                  u8 fc, u8 sel1, u16 sel2)
 833 {
 834         vcpu->run->exit_reason = KVM_EXIT_S390_STSI;
 835         vcpu->run->s390_stsi.addr = addr;
 836         vcpu->run->s390_stsi.ar = ar;
 837         vcpu->run->s390_stsi.fc = fc;
 838         vcpu->run->s390_stsi.sel1 = sel1;
 839         vcpu->run->s390_stsi.sel2 = sel2;
 840 }
 841
 842 static int handle_stsi(struct kvm_vcpu *vcpu)
 843 {
 844         int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
 845         int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
 846         int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
 847         unsigned long mem = 0;
 848         u64 operand2;
 849         int rc = 0;
 850         u8 ar;
 851
 852         vcpu->stat.instruction_stsi++;
 853         VCPU_EVENT(vcpu, 3, "STSI: fc: %u sel1: %u sel2: %u", fc, sel1, sel2);
 854
 855         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 856                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 857
 858         if (fc > 3) {
 859                 kvm_s390_set_psw_cc(vcpu, 3);
 860                 return 0;
 861         }
 862
 863         if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
 864             || vcpu->run->s.regs.gprs[1] & 0xffff0000)
 865                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 866
 867         if (fc == 0) {
 868                 vcpu->run->s.regs.gprs[0] = 3 << 28;
 869                 kvm_s390_set_psw_cc(vcpu, 0);
 870                 return 0;
 871         }
 872
 873         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 874
 875         if (operand2 & 0xfff)
 876                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 877
 878         switch (fc) {
 879         case 1: /* same handling for 1 and 2 */
 880         case 2:
 881                 mem = get_zeroed_page(GFP_KERNEL);
 882                 if (!mem)
 883                         goto out_no_data;
 884                 if (stsi((void *) mem, fc, sel1, sel2))
 885                         goto out_no_data;
 886                 break;
 887         case 3:
 888                 if (sel1 != 2 || sel2 != 2)
 889                         goto out_no_data;
 890                 mem = get_zeroed_page(GFP_KERNEL);
 891                 if (!mem)
 892                         goto out_no_data;
 893                 handle_stsi_3_2_2(vcpu, (void *) mem);
 894                 break;
 895         }
 896
 897         rc = write_guest(vcpu, operand2, ar, (void *)mem, PAGE_SIZE);
 898         if (rc) {
 899                 rc = kvm_s390_inject_prog_cond(vcpu, rc);
 900                 goto out;
 901         }
 902         if (vcpu->kvm->arch.user_stsi) {
 903                 insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
 904                 rc = -EREMOTE;
 905         }
 906         trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
 907         free_page(mem);
 908         kvm_s390_set_psw_cc(vcpu, 0);
 909         vcpu->run->s.regs.gprs[0] = 0;
 910         return rc;
 911 out_no_data:
 912         kvm_s390_set_psw_cc(vcpu, 3);
 913 out:
 914         free_page(mem);
 915         return rc;
 916 }
 917
 918 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
 919 {
 920         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
 921         case 0x02:
 922                 return handle_stidp(vcpu);
 923         case 0x04:
 924                 return handle_set_clock(vcpu);
 925         case 0x10:
 926                 return handle_set_prefix(vcpu);
 927         case 0x11:
 928                 return handle_store_prefix(vcpu);
 929         case 0x12:
 930                 return handle_store_cpu_address(vcpu);
 931         case 0x14:
 932                 return kvm_s390_handle_vsie(vcpu);
 933         case 0x21:
 934         case 0x50:
 935                 return handle_ipte_interlock(vcpu);
 936         case 0x29:
 937                 return handle_iske(vcpu);
 938         case 0x2a:
 939                 return handle_rrbe(vcpu);
 940         case 0x2b:
 941                 return handle_sske(vcpu);
 942         case 0x2c:
 943                 return handle_test_block(vcpu);
 944         case 0x30:
 945         case 0x31:
 946         case 0x32:
 947         case 0x33:
 948         case 0x34:
 949         case 0x35:
 950         case 0x36:
 951         case 0x37:
 952         case 0x38:
 953         case 0x39:
 954         case 0x3a:
 955         case 0x3b:
 956         case 0x3c:
 957         case 0x5f:
 958         case 0x74:
 959         case 0x76:
 960                 return handle_io_inst(vcpu);
 961         case 0x56:
 962                 return handle_sthyi(vcpu);
 963         case 0x7d:
 964                 return handle_stsi(vcpu);
 965         case 0xaf:
 966                 return handle_pqap(vcpu);
 967         case 0xb1:
 968                 return handle_stfl(vcpu);
 969         case 0xb2:
 970                 return handle_lpswe(vcpu);
 971         default:
 972                 return -EOPNOTSUPP;
 973         }
 974 }
 975
 976 static int handle_epsw(struct kvm_vcpu *vcpu)
 977 {
 978         int reg1, reg2;
 979
 980         vcpu->stat.instruction_epsw++;
 981
 982         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 983
 984         /* This basically extracts the mask half of the psw. */
 985         vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
 986         vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
 987         if (reg2) {
 988                 vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
 989                 vcpu->run->s.regs.gprs[reg2] |=
 990                         vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
 991         }
 992         return 0;
 993 }
 994
 995 #define PFMF_RESERVED   0xfffc0101UL
 996 #define PFMF_SK         0x00020000UL
 997 #define PFMF_CF         0x00010000UL
 998 #define PFMF_UI         0x00008000UL
 999 #define PFMF_FSC        0x00007000UL
1000 #define PFMF_NQ         0x00000800UL
1001 #define PFMF_MR         0x00000400UL
1002 #define PFMF_MC         0x00000200UL
1003 #define PFMF_KEY        0x000000feUL
1004
1005 static int handle_pfmf(struct kvm_vcpu *vcpu)
1006 {
1007         bool mr = false, mc = false, nq;
1008         int reg1, reg2;
1009         unsigned long start, end;
1010         unsigned char key;
1011
1012         vcpu->stat.instruction_pfmf++;
1013
1014         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
1015
1016         if (!test_kvm_facility(vcpu->kvm, 8))
1017                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1018
1019         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1020                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1021
1022         if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
1023                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1024
1025         /* Only provide non-quiescing support if enabled for the guest */
1026         if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ &&
1027             !test_kvm_facility(vcpu->kvm, 14))
1028                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1029
1030         /* Only provide conditional-SSKE support if enabled for the guest */
1031         if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK &&
1032             test_kvm_facility(vcpu->kvm, 10)) {
1033                 mr = vcpu->run->s.regs.gprs[reg1] & PFMF_MR;
1034                 mc = vcpu->run->s.regs.gprs[reg1] & PFMF_MC;
1035         }
1036
1037         nq = vcpu->run->s.regs.gprs[reg1] & PFMF_NQ;
1038         key = vcpu->run->s.regs.gprs[reg1] & PFMF_KEY;
1039         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
1040         start = kvm_s390_logical_to_effective(vcpu, start);
1041
1042         if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1043                 if (kvm_s390_check_low_addr_prot_real(vcpu, start))
1044                         return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
1045         }
1046
1047         switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1048         case 0x00000000:
1049                 /* only 4k frames specify a real address */
1050                 start = kvm_s390_real_to_abs(vcpu, start);
1051                 end = (start + PAGE_SIZE) & ~(PAGE_SIZE - 1);
1052                 break;
1053         case 0x00001000:
1054                 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
1055                 break;
1056         case 0x00002000:
1057                 /* only support 2G frame size if EDAT2 is available and we are
1058                    not in 24-bit addressing mode */
1059                 if (!test_kvm_facility(vcpu->kvm, 78) ||
1060                     psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_24BIT)
1061                         return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1062                 end = (start + _REGION3_SIZE) & ~(_REGION3_SIZE - 1);
1063                 break;
1064         default:
1065                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1066         }
1067
1068         while (start != end) {
1069                 unsigned long vmaddr;
1070                 bool unlocked = false;
1071
1072                 /* Translate guest address to host address */
1073                 vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
1074                 if (kvm_is_error_hva(vmaddr))
1075                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1076
1077                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1078                         if (kvm_clear_guest(vcpu->kvm, start, PAGE_SIZE))
1079                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1080                 }
1081
1082                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
1083                         int rc = kvm_s390_skey_check_enable(vcpu);
1084
1085                         if (rc)
1086                                 return rc;
1087                         down_read(&current->mm->mmap_sem);
1088                         rc = cond_set_guest_storage_key(current->mm, vmaddr,
1089                                                         key, NULL, nq, mr, mc);
1090                         if (rc < 0) {
1091                                 rc = fixup_user_fault(current, current->mm, vmaddr,
1092                                                       FAULT_FLAG_WRITE, &unlocked);
1093                                 rc = !rc ? -EAGAIN : rc;
1094                         }
1095                         up_read(&current->mm->mmap_sem);
1096                         if (rc == -EFAULT)
1097                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1098                         if (rc == -EAGAIN)
1099                                 continue;
1100                         if (rc < 0)
1101                                 return rc;
1102                 }
1103                 start += PAGE_SIZE;
1104         }
1105         if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1106                 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT) {
1107                         vcpu->run->s.regs.gprs[reg2] = end;
1108                 } else {
1109                         vcpu->run->s.regs.gprs[reg2] &= ~0xffffffffUL;
1110                         end = kvm_s390_logical_to_effective(vcpu, end);
1111                         vcpu->run->s.regs.gprs[reg2] |= end;
1112                 }
1113         }
1114         return 0;
1115 }
1116
1117 /*
1118  * Must be called with relevant read locks held (kvm->mm->mmap_sem, kvm->srcu)
1119  */
1120 static inline int __do_essa(struct kvm_vcpu *vcpu, const int orc)
1121 {
1122         int r1, r2, nappended, entries;
1123         unsigned long gfn, hva, res, pgstev, ptev;
1124         unsigned long *cbrlo;
1125
1126         /*
1127          * We don't need to set SD.FPF.SK to 1 here, because if we have a
1128          * machine check here we either handle it or crash
1129          */
1130
1131         kvm_s390_get_regs_rre(vcpu, &r1, &r2);
1132         gfn = vcpu->run->s.regs.gprs[r2] >> PAGE_SHIFT;
1133         hva = gfn_to_hva(vcpu->kvm, gfn);
1134         entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1135
1136         if (kvm_is_error_hva(hva))
1137                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1138
1139         nappended = pgste_perform_essa(vcpu->kvm->mm, hva, orc, &ptev, &pgstev);
1140         if (nappended < 0) {
1141                 res = orc ? 0x10 : 0;
1142                 vcpu->run->s.regs.gprs[r1] = res; /* Exception Indication */
1143                 return 0;
1144         }
1145         res = (pgstev & _PGSTE_GPS_USAGE_MASK) >> 22;
1146         /*
1147          * Set the block-content state part of the result. 0 means resident, so
1148          * nothing to do if the page is valid. 2 is for preserved pages
1149          * (non-present and non-zero), and 3 for zero pages (non-present and
1150          * zero).
1151          */
1152         if (ptev & _PAGE_INVALID) {
1153                 res |= 2;
1154                 if (pgstev & _PGSTE_GPS_ZERO)
1155                         res |= 1;
1156         }
1157         if (pgstev & _PGSTE_GPS_NODAT)
1158                 res |= 0x20;
1159         vcpu->run->s.regs.gprs[r1] = res;
1160         /*
1161          * It is possible that all the normal 511 slots were full, in which case
1162          * we will now write in the 512th slot, which is reserved for host use.
1163          * In both cases we let the normal essa handling code process all the
1164          * slots, including the reserved one, if needed.
1165          */
1166         if (nappended > 0) {
1167                 cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo & PAGE_MASK);
1168                 cbrlo[entries] = gfn << PAGE_SHIFT;
1169         }
1170
1171         if (orc) {
1172                 struct kvm_memory_slot *ms = gfn_to_memslot(vcpu->kvm, gfn);
1173
1174                 /* Increment only if we are really flipping the bit */
1175                 if (ms && !test_and_set_bit(gfn - ms->base_gfn, kvm_second_dirty_bitmap(ms)))
1176                         atomic64_inc(&vcpu->kvm->arch.cmma_dirty_pages);
1177         }
1178
1179         return nappended;
1180 }
1181
1182 static int handle_essa(struct kvm_vcpu *vcpu)
1183 {
1184         /* entries expected to be 1FF */
1185         int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1186         unsigned long *cbrlo;
1187         struct gmap *gmap;
1188         int i, orc;
1189
1190         VCPU_EVENT(vcpu, 4, "ESSA: release %d pages", entries);
1191         gmap = vcpu->arch.gmap;
1192         vcpu->stat.instruction_essa++;
1193         if (!vcpu->kvm->arch.use_cmma)
1194                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1195
1196         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1197                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1198         /* Check for invalid operation request code */
1199         orc = (vcpu->arch.sie_block->ipb & 0xf0000000) >> 28;
1200         /* ORCs 0-6 are always valid */
1201         if (orc > (test_kvm_facility(vcpu->kvm, 147) ? ESSA_SET_STABLE_NODAT
1202                                                 : ESSA_SET_STABLE_IF_RESIDENT))
1203                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1204
1205         if (!vcpu->kvm->arch.migration_mode) {
1206                 /*
1207                  * CMMA is enabled in the KVM settings, but is disabled in
1208                  * the SIE block and in the mm_context, and we are not doing
1209                  * a migration. Enable CMMA in the mm_context.
1210                  * Since we need to take a write lock to write to the context
1211                  * to avoid races with storage keys handling, we check if the
1212                  * value really needs to be written to; if the value is
1213                  * already correct, we do nothing and avoid the lock.
1214                  */
1215                 if (vcpu->kvm->mm->context.uses_cmm == 0) {
1216                         down_write(&vcpu->kvm->mm->mmap_sem);
1217                         vcpu->kvm->mm->context.uses_cmm = 1;
1218                         up_write(&vcpu->kvm->mm->mmap_sem);
1219                 }
1220                 /*
1221                  * If we are here, we are supposed to have CMMA enabled in
1222                  * the SIE block. Enabling CMMA works on a per-CPU basis,
1223                  * while the context use_cmma flag is per process.
1224                  * It's possible that the context flag is enabled and the
1225                  * SIE flag is not, so we set the flag always; if it was
1226                  * already set, nothing changes, otherwise we enable it
1227                  * on this CPU too.
1228                  */
1229                 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
1230                 /* Retry the ESSA instruction */
1231                 kvm_s390_retry_instr(vcpu);
1232         } else {
1233                 int srcu_idx;
1234
1235                 down_read(&vcpu->kvm->mm->mmap_sem);
1236                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1237                 i = __do_essa(vcpu, orc);
1238                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1239                 up_read(&vcpu->kvm->mm->mmap_sem);
1240                 if (i < 0)
1241                         return i;
1242                 /* Account for the possible extra cbrl entry */
1243                 entries += i;
1244         }
1245         vcpu->arch.sie_block->cbrlo &= PAGE_MASK;       /* reset nceo */
1246         cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
1247         down_read(&gmap->mm->mmap_sem);
1248         for (i = 0; i < entries; ++i)
1249                 __gmap_zap(gmap, cbrlo[i]);
1250         up_read(&gmap->mm->mmap_sem);
1251         return 0;
1252 }
1253
1254 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
1255 {
1256         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1257         case 0x8a:
1258         case 0x8e:
1259         case 0x8f:
1260                 return handle_ipte_interlock(vcpu);
1261         case 0x8d:
1262                 return handle_epsw(vcpu);
1263         case 0xab:
1264                 return handle_essa(vcpu);
1265         case 0xaf:
1266                 return handle_pfmf(vcpu);
1267         default:
1268                 return -EOPNOTSUPP;
1269         }
1270 }
1271
1272 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
1273 {
1274         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1275         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1276         int reg, rc, nr_regs;
1277         u32 ctl_array[16];
1278         u64 ga;
1279         u8 ar;
1280
1281         vcpu->stat.instruction_lctl++;
1282
1283         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1284                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1285
1286         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1287
1288         if (ga & 3)
1289                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1290
1291         VCPU_EVENT(vcpu, 4, "LCTL: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1292         trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
1293
1294         nr_regs = ((reg3 - reg1) & 0xf) + 1;
1295         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1296         if (rc)
1297                 return kvm_s390_inject_prog_cond(vcpu, rc);
1298         reg = reg1;
1299         nr_regs = 0;
1300         do {
1301                 vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
1302                 vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
1303                 if (reg == reg3)
1304                         break;
1305                 reg = (reg + 1) % 16;
1306         } while (1);
1307         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1308         return 0;
1309 }
1310
1311 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
1312 {
1313         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1314         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1315         int reg, rc, nr_regs;
1316         u32 ctl_array[16];
1317         u64 ga;
1318         u8 ar;
1319
1320         vcpu->stat.instruction_stctl++;
1321
1322         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1323                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1324
1325         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1326
1327         if (ga & 3)
1328                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1329
1330         VCPU_EVENT(vcpu, 4, "STCTL r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1331         trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
1332
1333         reg = reg1;
1334         nr_regs = 0;
1335         do {
1336                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1337                 if (reg == reg3)
1338                         break;
1339                 reg = (reg + 1) % 16;
1340         } while (1);
1341         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1342         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1343 }
1344
1345 static int handle_lctlg(struct kvm_vcpu *vcpu)
1346 {
1347         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1348         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1349         int reg, rc, nr_regs;
1350         u64 ctl_array[16];
1351         u64 ga;
1352         u8 ar;
1353
1354         vcpu->stat.instruction_lctlg++;
1355
1356         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1357                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1358
1359         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1360
1361         if (ga & 7)
1362                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1363
1364         VCPU_EVENT(vcpu, 4, "LCTLG: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1365         trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
1366
1367         nr_regs = ((reg3 - reg1) & 0xf) + 1;
1368         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1369         if (rc)
1370                 return kvm_s390_inject_prog_cond(vcpu, rc);
1371         reg = reg1;
1372         nr_regs = 0;
1373         do {
1374                 vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
1375                 if (reg == reg3)
1376                         break;
1377                 reg = (reg + 1) % 16;
1378         } while (1);
1379         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1380         return 0;
1381 }
1382
1383 static int handle_stctg(struct kvm_vcpu *vcpu)
1384 {
1385         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1386         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1387         int reg, rc, nr_regs;
1388         u64 ctl_array[16];
1389         u64 ga;
1390         u8 ar;
1391
1392         vcpu->stat.instruction_stctg++;
1393
1394         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1395                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1396
1397         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1398
1399         if (ga & 7)
1400                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1401
1402         VCPU_EVENT(vcpu, 4, "STCTG r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1403         trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
1404
1405         reg = reg1;
1406         nr_regs = 0;
1407         do {
1408                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1409                 if (reg == reg3)
1410                         break;
1411                 reg = (reg + 1) % 16;
1412         } while (1);
1413         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1414         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1415 }
1416
1417 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
1418 {
1419         switch (vcpu->arch.sie_block->ipb & 0x000000ff) {
1420         case 0x25:
1421                 return handle_stctg(vcpu);
1422         case 0x2f:
1423                 return handle_lctlg(vcpu);
1424         case 0x60:
1425         case 0x61:
1426         case 0x62:
1427                 return handle_ri(vcpu);
1428         default:
1429                 return -EOPNOTSUPP;
1430         }
1431 }
1432
1433 static int handle_tprot(struct kvm_vcpu *vcpu)
1434 {
1435         u64 address1, address2;
1436         unsigned long hva, gpa;
1437         int ret = 0, cc = 0;
1438         bool writable;
1439         u8 ar;
1440
1441         vcpu->stat.instruction_tprot++;
1442
1443         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1444                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1445
1446         kvm_s390_get_base_disp_sse(vcpu, &address1, &address2, &ar, NULL);
1447
1448         /* we only handle the Linux memory detection case:
1449          * access key == 0
1450          * everything else goes to userspace. */
1451         if (address2 & 0xf0)
1452                 return -EOPNOTSUPP;
1453         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1454                 ipte_lock(vcpu);
1455         ret = guest_translate_address(vcpu, address1, ar, &gpa, GACC_STORE);
1456         if (ret == PGM_PROTECTION) {
1457                 /* Write protected? Try again with read-only... */
1458                 cc = 1;
1459                 ret = guest_translate_address(vcpu, address1, ar, &gpa,
1460                                               GACC_FETCH);
1461         }
1462         if (ret) {
1463                 if (ret == PGM_ADDRESSING || ret == PGM_TRANSLATION_SPEC) {
1464                         ret = kvm_s390_inject_program_int(vcpu, ret);
1465                 } else if (ret > 0) {
1466                         /* Translation not available */
1467                         kvm_s390_set_psw_cc(vcpu, 3);
1468                         ret = 0;
1469                 }
1470                 goto out_unlock;
1471         }
1472
1473         hva = gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
1474         if (kvm_is_error_hva(hva)) {
1475                 ret = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1476         } else {
1477                 if (!writable)
1478                         cc = 1;         /* Write not permitted ==> read-only */
1479                 kvm_s390_set_psw_cc(vcpu, cc);
1480                 /* Note: CC2 only occurs for storage keys (not supported yet) */
1481         }
1482 out_unlock:
1483         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1484                 ipte_unlock(vcpu);
1485         return ret;
1486 }
1487
1488 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
1489 {
1490         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1491         case 0x01:
1492                 return handle_tprot(vcpu);
1493         default:
1494                 return -EOPNOTSUPP;
1495         }
1496 }
1497
1498 static int handle_sckpf(struct kvm_vcpu *vcpu)
1499 {
1500         u32 value;
1501
1502         vcpu->stat.instruction_sckpf++;
1503
1504         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1505                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1506
1507         if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
1508                 return kvm_s390_inject_program_int(vcpu,
1509                                                    PGM_SPECIFICATION);
1510
1511         value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
1512         vcpu->arch.sie_block->todpr = value;
1513
1514         return 0;
1515 }
1516
1517 static int handle_ptff(struct kvm_vcpu *vcpu)
1518 {
1519         vcpu->stat.instruction_ptff++;
1520
1521         /* we don't emulate any control instructions yet */
1522         kvm_s390_set_psw_cc(vcpu, 3);
1523         return 0;
1524 }
1525
1526 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1527 {
1528         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1529         case 0x04:
1530                 return handle_ptff(vcpu);
1531         case 0x07:
1532                 return handle_sckpf(vcpu);
1533         default:
1534                 return -EOPNOTSUPP;
1535         }
1536 }