arch/s390/kvm/priv.c

   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/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          * Note: running nested under z/VM can result in intercepts for other
 630          * function codes, e.g. PQAP(QCI). We do not support this and bail out.
 631          */
 632         reg0 = vcpu->run->s.regs.gprs[0];
 633         fc = (reg0 >> 24) & 0xff;
 634         if (fc != 0x03)
 635                 return -EOPNOTSUPP;
 636
 637         /* PQAP instruction is allowed for guest kernel only */
 638         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 639                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 640
 641         /* Common PQAP instruction specification exceptions */
 642         /* bits 41-47 must all be zeros */
 643         if (reg0 & 0x007f0000UL)
 644                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 645         /* APFT not install and T bit set */
 646         if (!test_kvm_facility(vcpu->kvm, 15) && (reg0 & 0x00800000UL))
 647                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 648         /* APXA not installed and APID greater 64 or APQI greater 16 */
 649         if (!(vcpu->kvm->arch.crypto.crycbd & 0x02) && (reg0 & 0x0000c0f0UL))
 650                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 651
 652         /* AQIC function code specific exception */
 653         /* facility 65 not present for AQIC function code */
 654         if (!test_kvm_facility(vcpu->kvm, 65))
 655                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 656
 657         /*
 658          * Verify that the hook callback is registered, lock the owner
 659          * and call the hook.
 660          */
 661         if (vcpu->kvm->arch.crypto.pqap_hook) {
 662                 if (!try_module_get(vcpu->kvm->arch.crypto.pqap_hook->owner))
 663                         return -EOPNOTSUPP;
 664                 ret = vcpu->kvm->arch.crypto.pqap_hook->hook(vcpu);
 665                 module_put(vcpu->kvm->arch.crypto.pqap_hook->owner);
 666                 if (!ret && vcpu->run->s.regs.gprs[1] & 0x00ff0000)
 667                         kvm_s390_set_psw_cc(vcpu, 3);
 668                 return ret;
 669         }
 670         /*
 671          * A vfio_driver must register a hook.
 672          * No hook means no driver to enable the SIE CRYCB and no queues.
 673          * We send this response to the guest.
 674          */
 675         status.response_code = 0x01;
 676         memcpy(&vcpu->run->s.regs.gprs[1], &status, sizeof(status));
 677         kvm_s390_set_psw_cc(vcpu, 3);
 678         return 0;
 679 }
 680
 681 static int handle_stfl(struct kvm_vcpu *vcpu)
 682 {
 683         int rc;
 684         unsigned int fac;
 685
 686         vcpu->stat.instruction_stfl++;
 687
 688         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 689                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 690
 691         /*
 692          * We need to shift the lower 32 facility bits (bit 0-31) from a u64
 693          * into a u32 memory representation. They will remain bits 0-31.
 694          */
 695         fac = *vcpu->kvm->arch.model.fac_list >> 32;
 696         rc = write_guest_lc(vcpu, offsetof(struct lowcore, stfl_fac_list),
 697                             &fac, sizeof(fac));
 698         if (rc)
 699                 return rc;
 700         VCPU_EVENT(vcpu, 3, "STFL: store facility list 0x%x", fac);
 701         trace_kvm_s390_handle_stfl(vcpu, fac);
 702         return 0;
 703 }
 704
 705 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
 706 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
 707 #define PSW_ADDR_24 0x0000000000ffffffUL
 708 #define PSW_ADDR_31 0x000000007fffffffUL
 709
 710 int is_valid_psw(psw_t *psw)
 711 {
 712         if (psw->mask & PSW_MASK_UNASSIGNED)
 713                 return 0;
 714         if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
 715                 if (psw->addr & ~PSW_ADDR_31)
 716                         return 0;
 717         }
 718         if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
 719                 return 0;
 720         if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
 721                 return 0;
 722         if (psw->addr & 1)
 723                 return 0;
 724         return 1;
 725 }
 726
 727 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
 728 {
 729         psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
 730         psw_compat_t new_psw;
 731         u64 addr;
 732         int rc;
 733         u8 ar;
 734
 735         vcpu->stat.instruction_lpsw++;
 736
 737         if (gpsw->mask & PSW_MASK_PSTATE)
 738                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 739
 740         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 741         if (addr & 7)
 742                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 743
 744         rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
 745         if (rc)
 746                 return kvm_s390_inject_prog_cond(vcpu, rc);
 747         if (!(new_psw.mask & PSW32_MASK_BASE))
 748                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 749         gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
 750         gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
 751         gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
 752         if (!is_valid_psw(gpsw))
 753                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 754         return 0;
 755 }
 756
 757 static int handle_lpswe(struct kvm_vcpu *vcpu)
 758 {
 759         psw_t new_psw;
 760         u64 addr;
 761         int rc;
 762         u8 ar;
 763
 764         vcpu->stat.instruction_lpswe++;
 765
 766         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 767                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 768
 769         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 770         if (addr & 7)
 771                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 772         rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
 773         if (rc)
 774                 return kvm_s390_inject_prog_cond(vcpu, rc);
 775         vcpu->arch.sie_block->gpsw = new_psw;
 776         if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
 777                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 778         return 0;
 779 }
 780
 781 static int handle_stidp(struct kvm_vcpu *vcpu)
 782 {
 783         u64 stidp_data = vcpu->kvm->arch.model.cpuid;
 784         u64 operand2;
 785         int rc;
 786         u8 ar;
 787
 788         vcpu->stat.instruction_stidp++;
 789
 790         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 791                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 792
 793         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 794
 795         if (operand2 & 7)
 796                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 797
 798         rc = write_guest(vcpu, operand2, ar, &stidp_data, sizeof(stidp_data));
 799         if (rc)
 800                 return kvm_s390_inject_prog_cond(vcpu, rc);
 801
 802         VCPU_EVENT(vcpu, 3, "STIDP: store cpu id 0x%llx", stidp_data);
 803         return 0;
 804 }
 805
 806 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
 807 {
 808         int cpus = 0;
 809         int n;
 810
 811         cpus = atomic_read(&vcpu->kvm->online_vcpus);
 812
 813         /* deal with other level 3 hypervisors */
 814         if (stsi(mem, 3, 2, 2))
 815                 mem->count = 0;
 816         if (mem->count < 8)
 817                 mem->count++;
 818         for (n = mem->count - 1; n > 0 ; n--)
 819                 memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
 820
 821         memset(&mem->vm[0], 0, sizeof(mem->vm[0]));
 822         mem->vm[0].cpus_total = cpus;
 823         mem->vm[0].cpus_configured = cpus;
 824         mem->vm[0].cpus_standby = 0;
 825         mem->vm[0].cpus_reserved = 0;
 826         mem->vm[0].caf = 1000;
 827         memcpy(mem->vm[0].name, "KVMguest", 8);
 828         ASCEBC(mem->vm[0].name, 8);
 829         memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
 830         ASCEBC(mem->vm[0].cpi, 16);
 831 }
 832
 833 static void insert_stsi_usr_data(struct kvm_vcpu *vcpu, u64 addr, u8 ar,
 834                                  u8 fc, u8 sel1, u16 sel2)
 835 {
 836         vcpu->run->exit_reason = KVM_EXIT_S390_STSI;
 837         vcpu->run->s390_stsi.addr = addr;
 838         vcpu->run->s390_stsi.ar = ar;
 839         vcpu->run->s390_stsi.fc = fc;
 840         vcpu->run->s390_stsi.sel1 = sel1;
 841         vcpu->run->s390_stsi.sel2 = sel2;
 842 }
 843
 844 static int handle_stsi(struct kvm_vcpu *vcpu)
 845 {
 846         int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
 847         int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
 848         int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
 849         unsigned long mem = 0;
 850         u64 operand2;
 851         int rc = 0;
 852         u8 ar;
 853
 854         vcpu->stat.instruction_stsi++;
 855         VCPU_EVENT(vcpu, 3, "STSI: fc: %u sel1: %u sel2: %u", fc, sel1, sel2);
 856
 857         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 858                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 859
 860         if (fc > 3) {
 861                 kvm_s390_set_psw_cc(vcpu, 3);
 862                 return 0;
 863         }
 864
 865         if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
 866             || vcpu->run->s.regs.gprs[1] & 0xffff0000)
 867                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 868
 869         if (fc == 0) {
 870                 vcpu->run->s.regs.gprs[0] = 3 << 28;
 871                 kvm_s390_set_psw_cc(vcpu, 0);
 872                 return 0;
 873         }
 874
 875         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 876
 877         if (!kvm_s390_pv_cpu_is_protected(vcpu) && (operand2 & 0xfff))
 878                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 879
 880         switch (fc) {
 881         case 1: /* same handling for 1 and 2 */
 882         case 2:
 883                 mem = get_zeroed_page(GFP_KERNEL);
 884                 if (!mem)
 885                         goto out_no_data;
 886                 if (stsi((void *) mem, fc, sel1, sel2))
 887                         goto out_no_data;
 888                 break;
 889         case 3:
 890                 if (sel1 != 2 || sel2 != 2)
 891                         goto out_no_data;
 892                 mem = get_zeroed_page(GFP_KERNEL);
 893                 if (!mem)
 894                         goto out_no_data;
 895                 handle_stsi_3_2_2(vcpu, (void *) mem);
 896                 break;
 897         }
 898         if (kvm_s390_pv_cpu_is_protected(vcpu)) {
 899                 memcpy((void *)sida_origin(vcpu->arch.sie_block), (void *)mem,
 900                        PAGE_SIZE);
 901                 rc = 0;
 902         } else {
 903                 rc = write_guest(vcpu, operand2, ar, (void *)mem, PAGE_SIZE);
 904         }
 905         if (rc) {
 906                 rc = kvm_s390_inject_prog_cond(vcpu, rc);
 907                 goto out;
 908         }
 909         if (vcpu->kvm->arch.user_stsi) {
 910                 insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
 911                 rc = -EREMOTE;
 912         }
 913         trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
 914         free_page(mem);
 915         kvm_s390_set_psw_cc(vcpu, 0);
 916         vcpu->run->s.regs.gprs[0] = 0;
 917         return rc;
 918 out_no_data:
 919         kvm_s390_set_psw_cc(vcpu, 3);
 920 out:
 921         free_page(mem);
 922         return rc;
 923 }
 924
 925 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
 926 {
 927         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
 928         case 0x02:
 929                 return handle_stidp(vcpu);
 930         case 0x04:
 931                 return handle_set_clock(vcpu);
 932         case 0x10:
 933                 return handle_set_prefix(vcpu);
 934         case 0x11:
 935                 return handle_store_prefix(vcpu);
 936         case 0x12:
 937                 return handle_store_cpu_address(vcpu);
 938         case 0x14:
 939                 return kvm_s390_handle_vsie(vcpu);
 940         case 0x21:
 941         case 0x50:
 942                 return handle_ipte_interlock(vcpu);
 943         case 0x29:
 944                 return handle_iske(vcpu);
 945         case 0x2a:
 946                 return handle_rrbe(vcpu);
 947         case 0x2b:
 948                 return handle_sske(vcpu);
 949         case 0x2c:
 950                 return handle_test_block(vcpu);
 951         case 0x30:
 952         case 0x31:
 953         case 0x32:
 954         case 0x33:
 955         case 0x34:
 956         case 0x35:
 957         case 0x36:
 958         case 0x37:
 959         case 0x38:
 960         case 0x39:
 961         case 0x3a:
 962         case 0x3b:
 963         case 0x3c:
 964         case 0x5f:
 965         case 0x74:
 966         case 0x76:
 967                 return handle_io_inst(vcpu);
 968         case 0x56:
 969                 return handle_sthyi(vcpu);
 970         case 0x7d:
 971                 return handle_stsi(vcpu);
 972         case 0xaf:
 973                 return handle_pqap(vcpu);
 974         case 0xb1:
 975                 return handle_stfl(vcpu);
 976         case 0xb2:
 977                 return handle_lpswe(vcpu);
 978         default:
 979                 return -EOPNOTSUPP;
 980         }
 981 }
 982
 983 static int handle_epsw(struct kvm_vcpu *vcpu)
 984 {
 985         int reg1, reg2;
 986
 987         vcpu->stat.instruction_epsw++;
 988
 989         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 990
 991         /* This basically extracts the mask half of the psw. */
 992         vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
 993         vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
 994         if (reg2) {
 995                 vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
 996                 vcpu->run->s.regs.gprs[reg2] |=
 997                         vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
 998         }
 999         return 0;
1000 }
1001
1002 #define PFMF_RESERVED   0xfffc0101UL
1003 #define PFMF_SK         0x00020000UL
1004 #define PFMF_CF         0x00010000UL
1005 #define PFMF_UI         0x00008000UL
1006 #define PFMF_FSC        0x00007000UL
1007 #define PFMF_NQ         0x00000800UL
1008 #define PFMF_MR         0x00000400UL
1009 #define PFMF_MC         0x00000200UL
1010 #define PFMF_KEY        0x000000feUL
1011
1012 static int handle_pfmf(struct kvm_vcpu *vcpu)
1013 {
1014         bool mr = false, mc = false, nq;
1015         int reg1, reg2;
1016         unsigned long start, end;
1017         unsigned char key;
1018
1019         vcpu->stat.instruction_pfmf++;
1020
1021         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
1022
1023         if (!test_kvm_facility(vcpu->kvm, 8))
1024                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1025
1026         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1027                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1028
1029         if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
1030                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1031
1032         /* Only provide non-quiescing support if enabled for the guest */
1033         if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ &&
1034             !test_kvm_facility(vcpu->kvm, 14))
1035                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1036
1037         /* Only provide conditional-SSKE support if enabled for the guest */
1038         if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK &&
1039             test_kvm_facility(vcpu->kvm, 10)) {
1040                 mr = vcpu->run->s.regs.gprs[reg1] & PFMF_MR;
1041                 mc = vcpu->run->s.regs.gprs[reg1] & PFMF_MC;
1042         }
1043
1044         nq = vcpu->run->s.regs.gprs[reg1] & PFMF_NQ;
1045         key = vcpu->run->s.regs.gprs[reg1] & PFMF_KEY;
1046         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
1047         start = kvm_s390_logical_to_effective(vcpu, start);
1048
1049         if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1050                 if (kvm_s390_check_low_addr_prot_real(vcpu, start))
1051                         return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
1052         }
1053
1054         switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1055         case 0x00000000:
1056                 /* only 4k frames specify a real address */
1057                 start = kvm_s390_real_to_abs(vcpu, start);
1058                 end = (start + PAGE_SIZE) & ~(PAGE_SIZE - 1);
1059                 break;
1060         case 0x00001000:
1061                 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
1062                 break;
1063         case 0x00002000:
1064                 /* only support 2G frame size if EDAT2 is available and we are
1065                    not in 24-bit addressing mode */
1066                 if (!test_kvm_facility(vcpu->kvm, 78) ||
1067                     psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_24BIT)
1068                         return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1069                 end = (start + _REGION3_SIZE) & ~(_REGION3_SIZE - 1);
1070                 break;
1071         default:
1072                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1073         }
1074
1075         while (start != end) {
1076                 unsigned long vmaddr;
1077                 bool unlocked = false;
1078
1079                 /* Translate guest address to host address */
1080                 vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
1081                 if (kvm_is_error_hva(vmaddr))
1082                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1083
1084                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1085                         if (kvm_clear_guest(vcpu->kvm, start, PAGE_SIZE))
1086                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1087                 }
1088
1089                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
1090                         int rc = kvm_s390_skey_check_enable(vcpu);
1091
1092                         if (rc)
1093                                 return rc;
1094                         down_read(&current->mm->mmap_sem);
1095                         rc = cond_set_guest_storage_key(current->mm, vmaddr,
1096                                                         key, NULL, nq, mr, mc);
1097                         if (rc < 0) {
1098                                 rc = fixup_user_fault(current, current->mm, vmaddr,
1099                                                       FAULT_FLAG_WRITE, &unlocked);
1100                                 rc = !rc ? -EAGAIN : rc;
1101                         }
1102                         up_read(&current->mm->mmap_sem);
1103                         if (rc == -EFAULT)
1104                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1105                         if (rc == -EAGAIN)
1106                                 continue;
1107                         if (rc < 0)
1108                                 return rc;
1109                 }
1110                 start += PAGE_SIZE;
1111         }
1112         if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1113                 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT) {
1114                         vcpu->run->s.regs.gprs[reg2] = end;
1115                 } else {
1116                         vcpu->run->s.regs.gprs[reg2] &= ~0xffffffffUL;
1117                         end = kvm_s390_logical_to_effective(vcpu, end);
1118                         vcpu->run->s.regs.gprs[reg2] |= end;
1119                 }
1120         }
1121         return 0;
1122 }
1123
1124 /*
1125  * Must be called with relevant read locks held (kvm->mm->mmap_sem, kvm->srcu)
1126  */
1127 static inline int __do_essa(struct kvm_vcpu *vcpu, const int orc)
1128 {
1129         int r1, r2, nappended, entries;
1130         unsigned long gfn, hva, res, pgstev, ptev;
1131         unsigned long *cbrlo;
1132
1133         /*
1134          * We don't need to set SD.FPF.SK to 1 here, because if we have a
1135          * machine check here we either handle it or crash
1136          */
1137
1138         kvm_s390_get_regs_rre(vcpu, &r1, &r2);
1139         gfn = vcpu->run->s.regs.gprs[r2] >> PAGE_SHIFT;
1140         hva = gfn_to_hva(vcpu->kvm, gfn);
1141         entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1142
1143         if (kvm_is_error_hva(hva))
1144                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1145
1146         nappended = pgste_perform_essa(vcpu->kvm->mm, hva, orc, &ptev, &pgstev);
1147         if (nappended < 0) {
1148                 res = orc ? 0x10 : 0;
1149                 vcpu->run->s.regs.gprs[r1] = res; /* Exception Indication */
1150                 return 0;
1151         }
1152         res = (pgstev & _PGSTE_GPS_USAGE_MASK) >> 22;
1153         /*
1154          * Set the block-content state part of the result. 0 means resident, so
1155          * nothing to do if the page is valid. 2 is for preserved pages
1156          * (non-present and non-zero), and 3 for zero pages (non-present and
1157          * zero).
1158          */
1159         if (ptev & _PAGE_INVALID) {
1160                 res |= 2;
1161                 if (pgstev & _PGSTE_GPS_ZERO)
1162                         res |= 1;
1163         }
1164         if (pgstev & _PGSTE_GPS_NODAT)
1165                 res |= 0x20;
1166         vcpu->run->s.regs.gprs[r1] = res;
1167         /*
1168          * It is possible that all the normal 511 slots were full, in which case
1169          * we will now write in the 512th slot, which is reserved for host use.
1170          * In both cases we let the normal essa handling code process all the
1171          * slots, including the reserved one, if needed.
1172          */
1173         if (nappended > 0) {
1174                 cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo & PAGE_MASK);
1175                 cbrlo[entries] = gfn << PAGE_SHIFT;
1176         }
1177
1178         if (orc) {
1179                 struct kvm_memory_slot *ms = gfn_to_memslot(vcpu->kvm, gfn);
1180
1181                 /* Increment only if we are really flipping the bit */
1182                 if (ms && !test_and_set_bit(gfn - ms->base_gfn, kvm_second_dirty_bitmap(ms)))
1183                         atomic64_inc(&vcpu->kvm->arch.cmma_dirty_pages);
1184         }
1185
1186         return nappended;
1187 }
1188
1189 static int handle_essa(struct kvm_vcpu *vcpu)
1190 {
1191         /* entries expected to be 1FF */
1192         int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1193         unsigned long *cbrlo;
1194         struct gmap *gmap;
1195         int i, orc;
1196
1197         VCPU_EVENT(vcpu, 4, "ESSA: release %d pages", entries);
1198         gmap = vcpu->arch.gmap;
1199         vcpu->stat.instruction_essa++;
1200         if (!vcpu->kvm->arch.use_cmma)
1201                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1202
1203         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1204                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1205         /* Check for invalid operation request code */
1206         orc = (vcpu->arch.sie_block->ipb & 0xf0000000) >> 28;
1207         /* ORCs 0-6 are always valid */
1208         if (orc > (test_kvm_facility(vcpu->kvm, 147) ? ESSA_SET_STABLE_NODAT
1209                                                 : ESSA_SET_STABLE_IF_RESIDENT))
1210                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1211
1212         if (!vcpu->kvm->arch.migration_mode) {
1213                 /*
1214                  * CMMA is enabled in the KVM settings, but is disabled in
1215                  * the SIE block and in the mm_context, and we are not doing
1216                  * a migration. Enable CMMA in the mm_context.
1217                  * Since we need to take a write lock to write to the context
1218                  * to avoid races with storage keys handling, we check if the
1219                  * value really needs to be written to; if the value is
1220                  * already correct, we do nothing and avoid the lock.
1221                  */
1222                 if (vcpu->kvm->mm->context.uses_cmm == 0) {
1223                         down_write(&vcpu->kvm->mm->mmap_sem);
1224                         vcpu->kvm->mm->context.uses_cmm = 1;
1225                         up_write(&vcpu->kvm->mm->mmap_sem);
1226                 }
1227                 /*
1228                  * If we are here, we are supposed to have CMMA enabled in
1229                  * the SIE block. Enabling CMMA works on a per-CPU basis,
1230                  * while the context use_cmma flag is per process.
1231                  * It's possible that the context flag is enabled and the
1232                  * SIE flag is not, so we set the flag always; if it was
1233                  * already set, nothing changes, otherwise we enable it
1234                  * on this CPU too.
1235                  */
1236                 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
1237                 /* Retry the ESSA instruction */
1238                 kvm_s390_retry_instr(vcpu);
1239         } else {
1240                 int srcu_idx;
1241
1242                 down_read(&vcpu->kvm->mm->mmap_sem);
1243                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1244                 i = __do_essa(vcpu, orc);
1245                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1246                 up_read(&vcpu->kvm->mm->mmap_sem);
1247                 if (i < 0)
1248                         return i;
1249                 /* Account for the possible extra cbrl entry */
1250                 entries += i;
1251         }
1252         vcpu->arch.sie_block->cbrlo &= PAGE_MASK;       /* reset nceo */
1253         cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
1254         down_read(&gmap->mm->mmap_sem);
1255         for (i = 0; i < entries; ++i)
1256                 __gmap_zap(gmap, cbrlo[i]);
1257         up_read(&gmap->mm->mmap_sem);
1258         return 0;
1259 }
1260
1261 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
1262 {
1263         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1264         case 0x8a:
1265         case 0x8e:
1266         case 0x8f:
1267                 return handle_ipte_interlock(vcpu);
1268         case 0x8d:
1269                 return handle_epsw(vcpu);
1270         case 0xab:
1271                 return handle_essa(vcpu);
1272         case 0xaf:
1273                 return handle_pfmf(vcpu);
1274         default:
1275                 return -EOPNOTSUPP;
1276         }
1277 }
1278
1279 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
1280 {
1281         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1282         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1283         int reg, rc, nr_regs;
1284         u32 ctl_array[16];
1285         u64 ga;
1286         u8 ar;
1287
1288         vcpu->stat.instruction_lctl++;
1289
1290         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1291                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1292
1293         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1294
1295         if (ga & 3)
1296                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1297
1298         VCPU_EVENT(vcpu, 4, "LCTL: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1299         trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
1300
1301         nr_regs = ((reg3 - reg1) & 0xf) + 1;
1302         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1303         if (rc)
1304                 return kvm_s390_inject_prog_cond(vcpu, rc);
1305         reg = reg1;
1306         nr_regs = 0;
1307         do {
1308                 vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
1309                 vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
1310                 if (reg == reg3)
1311                         break;
1312                 reg = (reg + 1) % 16;
1313         } while (1);
1314         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1315         return 0;
1316 }
1317
1318 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
1319 {
1320         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1321         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1322         int reg, rc, nr_regs;
1323         u32 ctl_array[16];
1324         u64 ga;
1325         u8 ar;
1326
1327         vcpu->stat.instruction_stctl++;
1328
1329         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1330                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1331
1332         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1333
1334         if (ga & 3)
1335                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1336
1337         VCPU_EVENT(vcpu, 4, "STCTL r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1338         trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
1339
1340         reg = reg1;
1341         nr_regs = 0;
1342         do {
1343                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1344                 if (reg == reg3)
1345                         break;
1346                 reg = (reg + 1) % 16;
1347         } while (1);
1348         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1349         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1350 }
1351
1352 static int handle_lctlg(struct kvm_vcpu *vcpu)
1353 {
1354         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1355         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1356         int reg, rc, nr_regs;
1357         u64 ctl_array[16];
1358         u64 ga;
1359         u8 ar;
1360
1361         vcpu->stat.instruction_lctlg++;
1362
1363         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1364                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1365
1366         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1367
1368         if (ga & 7)
1369                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1370
1371         VCPU_EVENT(vcpu, 4, "LCTLG: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1372         trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
1373
1374         nr_regs = ((reg3 - reg1) & 0xf) + 1;
1375         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1376         if (rc)
1377                 return kvm_s390_inject_prog_cond(vcpu, rc);
1378         reg = reg1;
1379         nr_regs = 0;
1380         do {
1381                 vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
1382                 if (reg == reg3)
1383                         break;
1384                 reg = (reg + 1) % 16;
1385         } while (1);
1386         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1387         return 0;
1388 }
1389
1390 static int handle_stctg(struct kvm_vcpu *vcpu)
1391 {
1392         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1393         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1394         int reg, rc, nr_regs;
1395         u64 ctl_array[16];
1396         u64 ga;
1397         u8 ar;
1398
1399         vcpu->stat.instruction_stctg++;
1400
1401         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1402                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1403
1404         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1405
1406         if (ga & 7)
1407                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1408
1409         VCPU_EVENT(vcpu, 4, "STCTG r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1410         trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
1411
1412         reg = reg1;
1413         nr_regs = 0;
1414         do {
1415                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1416                 if (reg == reg3)
1417                         break;
1418                 reg = (reg + 1) % 16;
1419         } while (1);
1420         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1421         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1422 }
1423
1424 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
1425 {
1426         switch (vcpu->arch.sie_block->ipb & 0x000000ff) {
1427         case 0x25:
1428                 return handle_stctg(vcpu);
1429         case 0x2f:
1430                 return handle_lctlg(vcpu);
1431         case 0x60:
1432         case 0x61:
1433         case 0x62:
1434                 return handle_ri(vcpu);
1435         default:
1436                 return -EOPNOTSUPP;
1437         }
1438 }
1439
1440 static int handle_tprot(struct kvm_vcpu *vcpu)
1441 {
1442         u64 address1, address2;
1443         unsigned long hva, gpa;
1444         int ret = 0, cc = 0;
1445         bool writable;
1446         u8 ar;
1447
1448         vcpu->stat.instruction_tprot++;
1449
1450         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1451                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1452
1453         kvm_s390_get_base_disp_sse(vcpu, &address1, &address2, &ar, NULL);
1454
1455         /* we only handle the Linux memory detection case:
1456          * access key == 0
1457          * everything else goes to userspace. */
1458         if (address2 & 0xf0)
1459                 return -EOPNOTSUPP;
1460         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1461                 ipte_lock(vcpu);
1462         ret = guest_translate_address(vcpu, address1, ar, &gpa, GACC_STORE);
1463         if (ret == PGM_PROTECTION) {
1464                 /* Write protected? Try again with read-only... */
1465                 cc = 1;
1466                 ret = guest_translate_address(vcpu, address1, ar, &gpa,
1467                                               GACC_FETCH);
1468         }
1469         if (ret) {
1470                 if (ret == PGM_ADDRESSING || ret == PGM_TRANSLATION_SPEC) {
1471                         ret = kvm_s390_inject_program_int(vcpu, ret);
1472                 } else if (ret > 0) {
1473                         /* Translation not available */
1474                         kvm_s390_set_psw_cc(vcpu, 3);
1475                         ret = 0;
1476                 }
1477                 goto out_unlock;
1478         }
1479
1480         hva = gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
1481         if (kvm_is_error_hva(hva)) {
1482                 ret = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1483         } else {
1484                 if (!writable)
1485                         cc = 1;         /* Write not permitted ==> read-only */
1486                 kvm_s390_set_psw_cc(vcpu, cc);
1487                 /* Note: CC2 only occurs for storage keys (not supported yet) */
1488         }
1489 out_unlock:
1490         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1491                 ipte_unlock(vcpu);
1492         return ret;
1493 }
1494
1495 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
1496 {
1497         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1498         case 0x01:
1499                 return handle_tprot(vcpu);
1500         default:
1501                 return -EOPNOTSUPP;
1502         }
1503 }
1504
1505 static int handle_sckpf(struct kvm_vcpu *vcpu)
1506 {
1507         u32 value;
1508
1509         vcpu->stat.instruction_sckpf++;
1510
1511         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1512                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1513
1514         if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
1515                 return kvm_s390_inject_program_int(vcpu,
1516                                                    PGM_SPECIFICATION);
1517
1518         value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
1519         vcpu->arch.sie_block->todpr = value;
1520
1521         return 0;
1522 }
1523
1524 static int handle_ptff(struct kvm_vcpu *vcpu)
1525 {
1526         vcpu->stat.instruction_ptff++;
1527
1528         /* we don't emulate any control instructions yet */
1529         kvm_s390_set_psw_cc(vcpu, 3);
1530         return 0;
1531 }
1532
1533 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1534 {
1535         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1536         case 0x04:
1537                 return handle_ptff(vcpu);
1538         case 0x07:
1539                 return handle_sckpf(vcpu);
1540         default:
1541                 return -EOPNOTSUPP;
1542         }
1543 }