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 <[email protected]>
   8  *               Christian Borntraeger <[email protected]>
   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 #include <linux/pgtable.h>
  17
  18 #include <asm/asm-offsets.h>
  19 #include <asm/facility.h>
  20 #include <asm/current.h>
  21 #include <asm/debug.h>
  22 #include <asm/ebcdic.h>
  23 #include <asm/sysinfo.h>
  24 #include <asm/page-states.h>
  25 #include <asm/gmap.h>
  26 #include <asm/io.h>
  27 #include <asm/ptrace.h>
  28 #include <asm/sclp.h>
  29 #include <asm/ap.h>
  30 #include "gaccess.h"
  31 #include "kvm-s390.h"
  32 #include "trace.h"
  33
  34 static int handle_ri(struct kvm_vcpu *vcpu)
  35 {
  36         vcpu->stat.instruction_ri++;
  37
  38         if (test_kvm_facility(vcpu->kvm, 64)) {
  39                 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (lazy)");
  40                 vcpu->arch.sie_block->ecb3 |= ECB3_RI;
  41                 kvm_s390_retry_instr(vcpu);
  42                 return 0;
  43         } else
  44                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
  45 }
  46
  47 int kvm_s390_handle_aa(struct kvm_vcpu *vcpu)
  48 {
  49         if ((vcpu->arch.sie_block->ipa & 0xf) <= 4)
  50                 return handle_ri(vcpu);
  51         else
  52                 return -EOPNOTSUPP;
  53 }
  54
  55 static int handle_gs(struct kvm_vcpu *vcpu)
  56 {
  57         vcpu->stat.instruction_gs++;
  58
  59         if (test_kvm_facility(vcpu->kvm, 133)) {
  60                 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (lazy)");
  61                 preempt_disable();
  62                 __ctl_set_bit(2, 4);
  63                 current->thread.gs_cb = (struct gs_cb *)&vcpu->run->s.regs.gscb;
  64                 restore_gs_cb(current->thread.gs_cb);
  65                 preempt_enable();
  66                 vcpu->arch.sie_block->ecb |= ECB_GS;
  67                 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
  68                 vcpu->arch.gs_enabled = 1;
  69                 kvm_s390_retry_instr(vcpu);
  70                 return 0;
  71         } else
  72                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
  73 }
  74
  75 int kvm_s390_handle_e3(struct kvm_vcpu *vcpu)
  76 {
  77         int code = vcpu->arch.sie_block->ipb & 0xff;
  78
  79         if (code == 0x49 || code == 0x4d)
  80                 return handle_gs(vcpu);
  81         else
  82                 return -EOPNOTSUPP;
  83 }
  84 /* Handle SCK (SET CLOCK) interception */
  85 static int handle_set_clock(struct kvm_vcpu *vcpu)
  86 {
  87         struct kvm_s390_vm_tod_clock gtod = { 0 };
  88         int rc;
  89         u8 ar;
  90         u64 op2;
  91
  92         vcpu->stat.instruction_sck++;
  93
  94         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
  95                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
  96
  97         op2 = kvm_s390_get_base_disp_s(vcpu, &ar);
  98         if (op2 & 7)    /* Operand must be on a doubleword boundary */
  99                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 100         rc = read_guest(vcpu, op2, ar, &gtod.tod, sizeof(gtod.tod));
 101         if (rc)
 102                 return kvm_s390_inject_prog_cond(vcpu, rc);
 103
 104         VCPU_EVENT(vcpu, 3, "SCK: setting guest TOD to 0x%llx", gtod.tod);
 105         kvm_s390_set_tod_clock(vcpu->kvm, &gtod);
 106
 107         kvm_s390_set_psw_cc(vcpu, 0);
 108         return 0;
 109 }
 110
 111 static int handle_set_prefix(struct kvm_vcpu *vcpu)
 112 {
 113         u64 operand2;
 114         u32 address;
 115         int rc;
 116         u8 ar;
 117
 118         vcpu->stat.instruction_spx++;
 119
 120         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 121                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 122
 123         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 124
 125         /* must be word boundary */
 126         if (operand2 & 3)
 127                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 128
 129         /* get the value */
 130         rc = read_guest(vcpu, operand2, ar, &address, sizeof(address));
 131         if (rc)
 132                 return kvm_s390_inject_prog_cond(vcpu, rc);
 133
 134         address &= 0x7fffe000u;
 135
 136         /*
 137          * Make sure the new value is valid memory. We only need to check the
 138          * first page, since address is 8k aligned and memory pieces are always
 139          * at least 1MB aligned and have at least a size of 1MB.
 140          */
 141         if (kvm_is_error_gpa(vcpu->kvm, address))
 142                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 143
 144         kvm_s390_set_prefix(vcpu, address);
 145         trace_kvm_s390_handle_prefix(vcpu, 1, address);
 146         return 0;
 147 }
 148
 149 static int handle_store_prefix(struct kvm_vcpu *vcpu)
 150 {
 151         u64 operand2;
 152         u32 address;
 153         int rc;
 154         u8 ar;
 155
 156         vcpu->stat.instruction_stpx++;
 157
 158         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 159                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 160
 161         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 162
 163         /* must be word boundary */
 164         if (operand2 & 3)
 165                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 166
 167         address = kvm_s390_get_prefix(vcpu);
 168
 169         /* get the value */
 170         rc = write_guest(vcpu, operand2, ar, &address, sizeof(address));
 171         if (rc)
 172                 return kvm_s390_inject_prog_cond(vcpu, rc);
 173
 174         VCPU_EVENT(vcpu, 3, "STPX: storing prefix 0x%x into 0x%llx", address, operand2);
 175         trace_kvm_s390_handle_prefix(vcpu, 0, address);
 176         return 0;
 177 }
 178
 179 static int handle_store_cpu_address(struct kvm_vcpu *vcpu)
 180 {
 181         u16 vcpu_id = vcpu->vcpu_id;
 182         u64 ga;
 183         int rc;
 184         u8 ar;
 185
 186         vcpu->stat.instruction_stap++;
 187
 188         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 189                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 190
 191         ga = kvm_s390_get_base_disp_s(vcpu, &ar);
 192
 193         if (ga & 1)
 194                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 195
 196         rc = write_guest(vcpu, ga, ar, &vcpu_id, sizeof(vcpu_id));
 197         if (rc)
 198                 return kvm_s390_inject_prog_cond(vcpu, rc);
 199
 200         VCPU_EVENT(vcpu, 3, "STAP: storing cpu address (%u) to 0x%llx", vcpu_id, ga);
 201         trace_kvm_s390_handle_stap(vcpu, ga);
 202         return 0;
 203 }
 204
 205 int kvm_s390_skey_check_enable(struct kvm_vcpu *vcpu)
 206 {
 207         int rc;
 208
 209         trace_kvm_s390_skey_related_inst(vcpu);
 210         /* Already enabled? */
 211         if (vcpu->arch.skey_enabled)
 212                 return 0;
 213
 214         rc = s390_enable_skey();
 215         VCPU_EVENT(vcpu, 3, "enabling storage keys for guest: %d", rc);
 216         if (rc)
 217                 return rc;
 218
 219         if (kvm_s390_test_cpuflags(vcpu, CPUSTAT_KSS))
 220                 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_KSS);
 221         if (!vcpu->kvm->arch.use_skf)
 222                 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
 223         else
 224                 vcpu->arch.sie_block->ictl &= ~(ICTL_ISKE | ICTL_SSKE | ICTL_RRBE);
 225         vcpu->arch.skey_enabled = true;
 226         return 0;
 227 }
 228
 229 static int try_handle_skey(struct kvm_vcpu *vcpu)
 230 {
 231         int rc;
 232
 233         rc = kvm_s390_skey_check_enable(vcpu);
 234         if (rc)
 235                 return rc;
 236         if (vcpu->kvm->arch.use_skf) {
 237                 /* with storage-key facility, SIE interprets it for us */
 238                 kvm_s390_retry_instr(vcpu);
 239                 VCPU_EVENT(vcpu, 4, "%s", "retrying storage key operation");
 240                 return -EAGAIN;
 241         }
 242         return 0;
 243 }
 244
 245 static int handle_iske(struct kvm_vcpu *vcpu)
 246 {
 247         unsigned long gaddr, vmaddr;
 248         unsigned char key;
 249         int reg1, reg2;
 250         bool unlocked;
 251         int rc;
 252
 253         vcpu->stat.instruction_iske++;
 254
 255         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 256                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 257
 258         rc = try_handle_skey(vcpu);
 259         if (rc)
 260                 return rc != -EAGAIN ? rc : 0;
 261
 262         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 263
 264         gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 265         gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
 266         gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
 267         vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr));
 268         if (kvm_is_error_hva(vmaddr))
 269                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 270 retry:
 271         unlocked = false;
 272         mmap_read_lock(current->mm);
 273         rc = get_guest_storage_key(current->mm, vmaddr, &key);
 274
 275         if (rc) {
 276                 rc = fixup_user_fault(current->mm, vmaddr,
 277                                       FAULT_FLAG_WRITE, &unlocked);
 278                 if (!rc) {
 279                         mmap_read_unlock(current->mm);
 280                         goto retry;
 281                 }
 282         }
 283         mmap_read_unlock(current->mm);
 284         if (rc == -EFAULT)
 285                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 286         if (rc < 0)
 287                 return rc;
 288         vcpu->run->s.regs.gprs[reg1] &= ~0xff;
 289         vcpu->run->s.regs.gprs[reg1] |= key;
 290         return 0;
 291 }
 292
 293 static int handle_rrbe(struct kvm_vcpu *vcpu)
 294 {
 295         unsigned long vmaddr, gaddr;
 296         int reg1, reg2;
 297         bool unlocked;
 298         int rc;
 299
 300         vcpu->stat.instruction_rrbe++;
 301
 302         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 303                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 304
 305         rc = try_handle_skey(vcpu);
 306         if (rc)
 307                 return rc != -EAGAIN ? rc : 0;
 308
 309         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 310
 311         gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 312         gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
 313         gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
 314         vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr));
 315         if (kvm_is_error_hva(vmaddr))
 316                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 317 retry:
 318         unlocked = false;
 319         mmap_read_lock(current->mm);
 320         rc = reset_guest_reference_bit(current->mm, vmaddr);
 321         if (rc < 0) {
 322                 rc = fixup_user_fault(current->mm, vmaddr,
 323                                       FAULT_FLAG_WRITE, &unlocked);
 324                 if (!rc) {
 325                         mmap_read_unlock(current->mm);
 326                         goto retry;
 327                 }
 328         }
 329         mmap_read_unlock(current->mm);
 330         if (rc == -EFAULT)
 331                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 332         if (rc < 0)
 333                 return rc;
 334         kvm_s390_set_psw_cc(vcpu, rc);
 335         return 0;
 336 }
 337
 338 #define SSKE_NQ 0x8
 339 #define SSKE_MR 0x4
 340 #define SSKE_MC 0x2
 341 #define SSKE_MB 0x1
 342 static int handle_sske(struct kvm_vcpu *vcpu)
 343 {
 344         unsigned char m3 = vcpu->arch.sie_block->ipb >> 28;
 345         unsigned long start, end;
 346         unsigned char key, oldkey;
 347         int reg1, reg2;
 348         bool unlocked;
 349         int rc;
 350
 351         vcpu->stat.instruction_sske++;
 352
 353         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 354                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 355
 356         rc = try_handle_skey(vcpu);
 357         if (rc)
 358                 return rc != -EAGAIN ? rc : 0;
 359
 360         if (!test_kvm_facility(vcpu->kvm, 8))
 361                 m3 &= ~SSKE_MB;
 362         if (!test_kvm_facility(vcpu->kvm, 10))
 363                 m3 &= ~(SSKE_MC | SSKE_MR);
 364         if (!test_kvm_facility(vcpu->kvm, 14))
 365                 m3 &= ~SSKE_NQ;
 366
 367         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 368
 369         key = vcpu->run->s.regs.gprs[reg1] & 0xfe;
 370         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 371         start = kvm_s390_logical_to_effective(vcpu, start);
 372         if (m3 & SSKE_MB) {
 373                 /* start already designates an absolute address */
 374                 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
 375         } else {
 376                 start = kvm_s390_real_to_abs(vcpu, start);
 377                 end = start + PAGE_SIZE;
 378         }
 379
 380         while (start != end) {
 381                 unsigned long vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
 382                 unlocked = false;
 383
 384                 if (kvm_is_error_hva(vmaddr))
 385                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 386
 387                 mmap_read_lock(current->mm);
 388                 rc = cond_set_guest_storage_key(current->mm, vmaddr, key, &oldkey,
 389                                                 m3 & SSKE_NQ, m3 & SSKE_MR,
 390                                                 m3 & SSKE_MC);
 391
 392                 if (rc < 0) {
 393                         rc = fixup_user_fault(current->mm, vmaddr,
 394                                               FAULT_FLAG_WRITE, &unlocked);
 395                         rc = !rc ? -EAGAIN : rc;
 396                 }
 397                 mmap_read_unlock(current->mm);
 398                 if (rc == -EFAULT)
 399                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 400                 if (rc == -EAGAIN)
 401                         continue;
 402                 if (rc < 0)
 403                         return rc;
 404                 start += PAGE_SIZE;
 405         }
 406
 407         if (m3 & (SSKE_MC | SSKE_MR)) {
 408                 if (m3 & SSKE_MB) {
 409                         /* skey in reg1 is unpredictable */
 410                         kvm_s390_set_psw_cc(vcpu, 3);
 411                 } else {
 412                         kvm_s390_set_psw_cc(vcpu, rc);
 413                         vcpu->run->s.regs.gprs[reg1] &= ~0xff00UL;
 414                         vcpu->run->s.regs.gprs[reg1] |= (u64) oldkey << 8;
 415                 }
 416         }
 417         if (m3 & SSKE_MB) {
 418                 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT)
 419                         vcpu->run->s.regs.gprs[reg2] &= ~PAGE_MASK;
 420                 else
 421                         vcpu->run->s.regs.gprs[reg2] &= ~0xfffff000UL;
 422                 end = kvm_s390_logical_to_effective(vcpu, end);
 423                 vcpu->run->s.regs.gprs[reg2] |= end;
 424         }
 425         return 0;
 426 }
 427
 428 static int handle_ipte_interlock(struct kvm_vcpu *vcpu)
 429 {
 430         vcpu->stat.instruction_ipte_interlock++;
 431         if (psw_bits(vcpu->arch.sie_block->gpsw).pstate)
 432                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 433         wait_event(vcpu->kvm->arch.ipte_wq, !ipte_lock_held(vcpu));
 434         kvm_s390_retry_instr(vcpu);
 435         VCPU_EVENT(vcpu, 4, "%s", "retrying ipte interlock operation");
 436         return 0;
 437 }
 438
 439 static int handle_test_block(struct kvm_vcpu *vcpu)
 440 {
 441         gpa_t addr;
 442         int reg2;
 443
 444         vcpu->stat.instruction_tb++;
 445
 446         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 447                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 448
 449         kvm_s390_get_regs_rre(vcpu, NULL, &reg2);
 450         addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 451         addr = kvm_s390_logical_to_effective(vcpu, addr);
 452         if (kvm_s390_check_low_addr_prot_real(vcpu, addr))
 453                 return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
 454         addr = kvm_s390_real_to_abs(vcpu, addr);
 455
 456         if (kvm_is_error_gpa(vcpu->kvm, addr))
 457                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 458         /*
 459          * We don't expect errors on modern systems, and do not care
 460          * about storage keys (yet), so let's just clear the page.
 461          */
 462         if (kvm_clear_guest(vcpu->kvm, addr, PAGE_SIZE))
 463                 return -EFAULT;
 464         kvm_s390_set_psw_cc(vcpu, 0);
 465         vcpu->run->s.regs.gprs[0] = 0;
 466         return 0;
 467 }
 468
 469 static int handle_tpi(struct kvm_vcpu *vcpu)
 470 {
 471         struct kvm_s390_interrupt_info *inti;
 472         unsigned long len;
 473         u32 tpi_data[3];
 474         int rc;
 475         u64 addr;
 476         u8 ar;
 477
 478         vcpu->stat.instruction_tpi++;
 479
 480         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 481         if (addr & 3)
 482                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 483
 484         inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0);
 485         if (!inti) {
 486                 kvm_s390_set_psw_cc(vcpu, 0);
 487                 return 0;
 488         }
 489
 490         tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr;
 491         tpi_data[1] = inti->io.io_int_parm;
 492         tpi_data[2] = inti->io.io_int_word;
 493         if (addr) {
 494                 /*
 495                  * Store the two-word I/O interruption code into the
 496                  * provided area.
 497                  */
 498                 len = sizeof(tpi_data) - 4;
 499                 rc = write_guest(vcpu, addr, ar, &tpi_data, len);
 500                 if (rc) {
 501                         rc = kvm_s390_inject_prog_cond(vcpu, rc);
 502                         goto reinject_interrupt;
 503                 }
 504         } else {
 505                 /*
 506                  * Store the three-word I/O interruption code into
 507                  * the appropriate lowcore area.
 508                  */
 509                 len = sizeof(tpi_data);
 510                 if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len)) {
 511                         /* failed writes to the low core are not recoverable */
 512                         rc = -EFAULT;
 513                         goto reinject_interrupt;
 514                 }
 515         }
 516
 517         /* irq was successfully handed to the guest */
 518         kfree(inti);
 519         kvm_s390_set_psw_cc(vcpu, 1);
 520         return 0;
 521 reinject_interrupt:
 522         /*
 523          * If we encounter a problem storing the interruption code, the
 524          * instruction is suppressed from the guest's view: reinject the
 525          * interrupt.
 526          */
 527         if (kvm_s390_reinject_io_int(vcpu->kvm, inti)) {
 528                 kfree(inti);
 529                 rc = -EFAULT;
 530         }
 531         /* don't set the cc, a pgm irq was injected or we drop to user space */
 532         return rc ? -EFAULT : 0;
 533 }
 534
 535 static int handle_tsch(struct kvm_vcpu *vcpu)
 536 {
 537         struct kvm_s390_interrupt_info *inti = NULL;
 538         const u64 isc_mask = 0xffUL << 24; /* all iscs set */
 539
 540         vcpu->stat.instruction_tsch++;
 541
 542         /* a valid schid has at least one bit set */
 543         if (vcpu->run->s.regs.gprs[1])
 544                 inti = kvm_s390_get_io_int(vcpu->kvm, isc_mask,
 545                                            vcpu->run->s.regs.gprs[1]);
 546
 547         /*
 548          * Prepare exit to userspace.
 549          * We indicate whether we dequeued a pending I/O interrupt
 550          * so that userspace can re-inject it if the instruction gets
 551          * a program check. While this may re-order the pending I/O
 552          * interrupts, this is no problem since the priority is kept
 553          * intact.
 554          */
 555         vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
 556         vcpu->run->s390_tsch.dequeued = !!inti;
 557         if (inti) {
 558                 vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
 559                 vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
 560                 vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
 561                 vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
 562         }
 563         vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
 564         kfree(inti);
 565         return -EREMOTE;
 566 }
 567
 568 static int handle_io_inst(struct kvm_vcpu *vcpu)
 569 {
 570         VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
 571
 572         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 573                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 574
 575         if (vcpu->kvm->arch.css_support) {
 576                 /*
 577                  * Most I/O instructions will be handled by userspace.
 578                  * Exceptions are tpi and the interrupt portion of tsch.
 579                  */
 580                 if (vcpu->arch.sie_block->ipa == 0xb236)
 581                         return handle_tpi(vcpu);
 582                 if (vcpu->arch.sie_block->ipa == 0xb235)
 583                         return handle_tsch(vcpu);
 584                 /* Handle in userspace. */
 585                 vcpu->stat.instruction_io_other++;
 586                 return -EOPNOTSUPP;
 587         } else {
 588                 /*
 589                  * Set condition code 3 to stop the guest from issuing channel
 590                  * I/O instructions.
 591                  */
 592                 kvm_s390_set_psw_cc(vcpu, 3);
 593                 return 0;
 594         }
 595 }
 596
 597 /*
 598  * handle_pqap: Handling pqap interception
 599  * @vcpu: the vcpu having issue the pqap instruction
 600  *
 601  * We now support PQAP/AQIC instructions and we need to correctly
 602  * answer the guest even if no dedicated driver's hook is available.
 603  *
 604  * The intercepting code calls a dedicated callback for this instruction
 605  * if a driver did register one in the CRYPTO satellite of the
 606  * SIE block.
 607  *
 608  * If no callback is available, the queues are not available, return this
 609  * response code to the caller and set CC to 3.
 610  * Else return the response code returned by the callback.
 611  */
 612 static int handle_pqap(struct kvm_vcpu *vcpu)
 613 {
 614         struct ap_queue_status status = {};
 615         crypto_hook pqap_hook;
 616         unsigned long reg0;
 617         int ret;
 618         uint8_t fc;
 619
 620         /* Verify that the AP instruction are available */
 621         if (!ap_instructions_available())
 622                 return -EOPNOTSUPP;
 623         /* Verify that the guest is allowed to use AP instructions */
 624         if (!(vcpu->arch.sie_block->eca & ECA_APIE))
 625                 return -EOPNOTSUPP;
 626         /*
 627          * The only possibly intercepted functions when AP instructions are
 628          * available for the guest are AQIC and TAPQ with the t bit set
 629          * since we do not set IC.3 (FIII) we currently will only intercept
 630          * the AQIC function code.
 631          * Note: running nested under z/VM can result in intercepts for other
 632          * function codes, e.g. PQAP(QCI). We do not support this and bail out.
 633          */
 634         reg0 = vcpu->run->s.regs.gprs[0];
 635         fc = (reg0 >> 24) & 0xff;
 636         if (fc != 0x03)
 637                 return -EOPNOTSUPP;
 638
 639         /* PQAP instruction is allowed for guest kernel only */
 640         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 641                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 642
 643         /* Common PQAP instruction specification exceptions */
 644         /* bits 41-47 must all be zeros */
 645         if (reg0 & 0x007f0000UL)
 646                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 647         /* APFT not install and T bit set */
 648         if (!test_kvm_facility(vcpu->kvm, 15) && (reg0 & 0x00800000UL))
 649                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 650         /* APXA not installed and APID greater 64 or APQI greater 16 */
 651         if (!(vcpu->kvm->arch.crypto.crycbd & 0x02) && (reg0 & 0x0000c0f0UL))
 652                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 653
 654         /* AQIC function code specific exception */
 655         /* facility 65 not present for AQIC function code */
 656         if (!test_kvm_facility(vcpu->kvm, 65))
 657                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 658
 659         /*
 660          * If the hook callback is registered, there will be a pointer to the
 661          * hook function pointer in the kvm_s390_crypto structure. Lock the
 662          * owner, retrieve the hook function pointer and call the hook.
 663          */
 664         down_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
 665         if (vcpu->kvm->arch.crypto.pqap_hook) {
 666                 pqap_hook = *vcpu->kvm->arch.crypto.pqap_hook;
 667                 ret = pqap_hook(vcpu);
 668                 if (!ret && vcpu->run->s.regs.gprs[1] & 0x00ff0000)
 669                         kvm_s390_set_psw_cc(vcpu, 3);
 670                 up_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
 671                 return ret;
 672         }
 673         up_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
 674         /*
 675          * A vfio_driver must register a hook.
 676          * No hook means no driver to enable the SIE CRYCB and no queues.
 677          * We send this response to the guest.
 678          */
 679         status.response_code = 0x01;
 680         memcpy(&vcpu->run->s.regs.gprs[1], &status, sizeof(status));
 681         kvm_s390_set_psw_cc(vcpu, 3);
 682         return 0;
 683 }
 684
 685 static int handle_stfl(struct kvm_vcpu *vcpu)
 686 {
 687         int rc;
 688         unsigned int fac;
 689
 690         vcpu->stat.instruction_stfl++;
 691
 692         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 693                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 694
 695         /*
 696          * We need to shift the lower 32 facility bits (bit 0-31) from a u64
 697          * into a u32 memory representation. They will remain bits 0-31.
 698          */
 699         fac = *vcpu->kvm->arch.model.fac_list >> 32;
 700         rc = write_guest_lc(vcpu, offsetof(struct lowcore, stfl_fac_list),
 701                             &fac, sizeof(fac));
 702         if (rc)
 703                 return rc;
 704         VCPU_EVENT(vcpu, 3, "STFL: store facility list 0x%x", fac);
 705         trace_kvm_s390_handle_stfl(vcpu, fac);
 706         return 0;
 707 }
 708
 709 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
 710 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
 711 #define PSW_ADDR_24 0x0000000000ffffffUL
 712 #define PSW_ADDR_31 0x000000007fffffffUL
 713
 714 int is_valid_psw(psw_t *psw)
 715 {
 716         if (psw->mask & PSW_MASK_UNASSIGNED)
 717                 return 0;
 718         if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
 719                 if (psw->addr & ~PSW_ADDR_31)
 720                         return 0;
 721         }
 722         if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
 723                 return 0;
 724         if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
 725                 return 0;
 726         if (psw->addr & 1)
 727                 return 0;
 728         return 1;
 729 }
 730
 731 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
 732 {
 733         psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
 734         psw_compat_t new_psw;
 735         u64 addr;
 736         int rc;
 737         u8 ar;
 738
 739         vcpu->stat.instruction_lpsw++;
 740
 741         if (gpsw->mask & PSW_MASK_PSTATE)
 742                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 743
 744         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 745         if (addr & 7)
 746                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 747
 748         rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
 749         if (rc)
 750                 return kvm_s390_inject_prog_cond(vcpu, rc);
 751         if (!(new_psw.mask & PSW32_MASK_BASE))
 752                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 753         gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
 754         gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
 755         gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
 756         if (!is_valid_psw(gpsw))
 757                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 758         return 0;
 759 }
 760
 761 static int handle_lpswe(struct kvm_vcpu *vcpu)
 762 {
 763         psw_t new_psw;
 764         u64 addr;
 765         int rc;
 766         u8 ar;
 767
 768         vcpu->stat.instruction_lpswe++;
 769
 770         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 771                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 772
 773         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 774         if (addr & 7)
 775                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 776         rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
 777         if (rc)
 778                 return kvm_s390_inject_prog_cond(vcpu, rc);
 779         vcpu->arch.sie_block->gpsw = new_psw;
 780         if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
 781                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 782         return 0;
 783 }
 784
 785 static int handle_stidp(struct kvm_vcpu *vcpu)
 786 {
 787         u64 stidp_data = vcpu->kvm->arch.model.cpuid;
 788         u64 operand2;
 789         int rc;
 790         u8 ar;
 791
 792         vcpu->stat.instruction_stidp++;
 793
 794         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 795                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 796
 797         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 798
 799         if (operand2 & 7)
 800                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 801
 802         rc = write_guest(vcpu, operand2, ar, &stidp_data, sizeof(stidp_data));
 803         if (rc)
 804                 return kvm_s390_inject_prog_cond(vcpu, rc);
 805
 806         VCPU_EVENT(vcpu, 3, "STIDP: store cpu id 0x%llx", stidp_data);
 807         return 0;
 808 }
 809
 810 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
 811 {
 812         int cpus = 0;
 813         int n;
 814
 815         cpus = atomic_read(&vcpu->kvm->online_vcpus);
 816
 817         /* deal with other level 3 hypervisors */
 818         if (stsi(mem, 3, 2, 2))
 819                 mem->count = 0;
 820         if (mem->count < 8)
 821                 mem->count++;
 822         for (n = mem->count - 1; n > 0 ; n--)
 823                 memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
 824
 825         memset(&mem->vm[0], 0, sizeof(mem->vm[0]));
 826         mem->vm[0].cpus_total = cpus;
 827         mem->vm[0].cpus_configured = cpus;
 828         mem->vm[0].cpus_standby = 0;
 829         mem->vm[0].cpus_reserved = 0;
 830         mem->vm[0].caf = 1000;
 831         memcpy(mem->vm[0].name, "KVMguest", 8);
 832         ASCEBC(mem->vm[0].name, 8);
 833         memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
 834         ASCEBC(mem->vm[0].cpi, 16);
 835 }
 836
 837 static void insert_stsi_usr_data(struct kvm_vcpu *vcpu, u64 addr, u8 ar,
 838                                  u8 fc, u8 sel1, u16 sel2)
 839 {
 840         vcpu->run->exit_reason = KVM_EXIT_S390_STSI;
 841         vcpu->run->s390_stsi.addr = addr;
 842         vcpu->run->s390_stsi.ar = ar;
 843         vcpu->run->s390_stsi.fc = fc;
 844         vcpu->run->s390_stsi.sel1 = sel1;
 845         vcpu->run->s390_stsi.sel2 = sel2;
 846 }
 847
 848 static int handle_stsi(struct kvm_vcpu *vcpu)
 849 {
 850         int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
 851         int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
 852         int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
 853         unsigned long mem = 0;
 854         u64 operand2;
 855         int rc = 0;
 856         u8 ar;
 857
 858         vcpu->stat.instruction_stsi++;
 859         VCPU_EVENT(vcpu, 3, "STSI: fc: %u sel1: %u sel2: %u", fc, sel1, sel2);
 860
 861         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 862                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 863
 864         if (fc > 3) {
 865                 kvm_s390_set_psw_cc(vcpu, 3);
 866                 return 0;
 867         }
 868
 869         if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
 870             || vcpu->run->s.regs.gprs[1] & 0xffff0000)
 871                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 872
 873         if (fc == 0) {
 874                 vcpu->run->s.regs.gprs[0] = 3 << 28;
 875                 kvm_s390_set_psw_cc(vcpu, 0);
 876                 return 0;
 877         }
 878
 879         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 880
 881         if (!kvm_s390_pv_cpu_is_protected(vcpu) && (operand2 & 0xfff))
 882                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 883
 884         switch (fc) {
 885         case 1: /* same handling for 1 and 2 */
 886         case 2:
 887                 mem = get_zeroed_page(GFP_KERNEL_ACCOUNT);
 888                 if (!mem)
 889                         goto out_no_data;
 890                 if (stsi((void *) mem, fc, sel1, sel2))
 891                         goto out_no_data;
 892                 break;
 893         case 3:
 894                 if (sel1 != 2 || sel2 != 2)
 895                         goto out_no_data;
 896                 mem = get_zeroed_page(GFP_KERNEL_ACCOUNT);
 897                 if (!mem)
 898                         goto out_no_data;
 899                 handle_stsi_3_2_2(vcpu, (void *) mem);
 900                 break;
 901         }
 902         if (kvm_s390_pv_cpu_is_protected(vcpu)) {
 903                 memcpy((void *)sida_origin(vcpu->arch.sie_block), (void *)mem,
 904                        PAGE_SIZE);
 905                 rc = 0;
 906         } else {
 907                 rc = write_guest(vcpu, operand2, ar, (void *)mem, PAGE_SIZE);
 908         }
 909         if (rc) {
 910                 rc = kvm_s390_inject_prog_cond(vcpu, rc);
 911                 goto out;
 912         }
 913         if (vcpu->kvm->arch.user_stsi) {
 914                 insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
 915                 rc = -EREMOTE;
 916         }
 917         trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
 918         free_page(mem);
 919         kvm_s390_set_psw_cc(vcpu, 0);
 920         vcpu->run->s.regs.gprs[0] = 0;
 921         return rc;
 922 out_no_data:
 923         kvm_s390_set_psw_cc(vcpu, 3);
 924 out:
 925         free_page(mem);
 926         return rc;
 927 }
 928
 929 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
 930 {
 931         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
 932         case 0x02:
 933                 return handle_stidp(vcpu);
 934         case 0x04:
 935                 return handle_set_clock(vcpu);
 936         case 0x10:
 937                 return handle_set_prefix(vcpu);
 938         case 0x11:
 939                 return handle_store_prefix(vcpu);
 940         case 0x12:
 941                 return handle_store_cpu_address(vcpu);
 942         case 0x14:
 943                 return kvm_s390_handle_vsie(vcpu);
 944         case 0x21:
 945         case 0x50:
 946                 return handle_ipte_interlock(vcpu);
 947         case 0x29:
 948                 return handle_iske(vcpu);
 949         case 0x2a:
 950                 return handle_rrbe(vcpu);
 951         case 0x2b:
 952                 return handle_sske(vcpu);
 953         case 0x2c:
 954                 return handle_test_block(vcpu);
 955         case 0x30:
 956         case 0x31:
 957         case 0x32:
 958         case 0x33:
 959         case 0x34:
 960         case 0x35:
 961         case 0x36:
 962         case 0x37:
 963         case 0x38:
 964         case 0x39:
 965         case 0x3a:
 966         case 0x3b:
 967         case 0x3c:
 968         case 0x5f:
 969         case 0x74:
 970         case 0x76:
 971                 return handle_io_inst(vcpu);
 972         case 0x56:
 973                 return handle_sthyi(vcpu);
 974         case 0x7d:
 975                 return handle_stsi(vcpu);
 976         case 0xaf:
 977                 return handle_pqap(vcpu);
 978         case 0xb1:
 979                 return handle_stfl(vcpu);
 980         case 0xb2:
 981                 return handle_lpswe(vcpu);
 982         default:
 983                 return -EOPNOTSUPP;
 984         }
 985 }
 986
 987 static int handle_epsw(struct kvm_vcpu *vcpu)
 988 {
 989         int reg1, reg2;
 990
 991         vcpu->stat.instruction_epsw++;
 992
 993         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 994
 995         /* This basically extracts the mask half of the psw. */
 996         vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
 997         vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
 998         if (reg2) {
 999                 vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
1000                 vcpu->run->s.regs.gprs[reg2] |=
1001                         vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
1002         }
1003         return 0;
1004 }
1005
1006 #define PFMF_RESERVED   0xfffc0101UL
1007 #define PFMF_SK         0x00020000UL
1008 #define PFMF_CF         0x00010000UL
1009 #define PFMF_UI         0x00008000UL
1010 #define PFMF_FSC        0x00007000UL
1011 #define PFMF_NQ         0x00000800UL
1012 #define PFMF_MR         0x00000400UL
1013 #define PFMF_MC         0x00000200UL
1014 #define PFMF_KEY        0x000000feUL
1015
1016 static int handle_pfmf(struct kvm_vcpu *vcpu)
1017 {
1018         bool mr = false, mc = false, nq;
1019         int reg1, reg2;
1020         unsigned long start, end;
1021         unsigned char key;
1022
1023         vcpu->stat.instruction_pfmf++;
1024
1025         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
1026
1027         if (!test_kvm_facility(vcpu->kvm, 8))
1028                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1029
1030         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1031                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1032
1033         if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
1034                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1035
1036         /* Only provide non-quiescing support if enabled for the guest */
1037         if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ &&
1038             !test_kvm_facility(vcpu->kvm, 14))
1039                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1040
1041         /* Only provide conditional-SSKE support if enabled for the guest */
1042         if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK &&
1043             test_kvm_facility(vcpu->kvm, 10)) {
1044                 mr = vcpu->run->s.regs.gprs[reg1] & PFMF_MR;
1045                 mc = vcpu->run->s.regs.gprs[reg1] & PFMF_MC;
1046         }
1047
1048         nq = vcpu->run->s.regs.gprs[reg1] & PFMF_NQ;
1049         key = vcpu->run->s.regs.gprs[reg1] & PFMF_KEY;
1050         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
1051         start = kvm_s390_logical_to_effective(vcpu, start);
1052
1053         if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1054                 if (kvm_s390_check_low_addr_prot_real(vcpu, start))
1055                         return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
1056         }
1057
1058         switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1059         case 0x00000000:
1060                 /* only 4k frames specify a real address */
1061                 start = kvm_s390_real_to_abs(vcpu, start);
1062                 end = (start + PAGE_SIZE) & ~(PAGE_SIZE - 1);
1063                 break;
1064         case 0x00001000:
1065                 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
1066                 break;
1067         case 0x00002000:
1068                 /* only support 2G frame size if EDAT2 is available and we are
1069                    not in 24-bit addressing mode */
1070                 if (!test_kvm_facility(vcpu->kvm, 78) ||
1071                     psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_24BIT)
1072                         return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1073                 end = (start + _REGION3_SIZE) & ~(_REGION3_SIZE - 1);
1074                 break;
1075         default:
1076                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1077         }
1078
1079         while (start != end) {
1080                 unsigned long vmaddr;
1081                 bool unlocked = false;
1082
1083                 /* Translate guest address to host address */
1084                 vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
1085                 if (kvm_is_error_hva(vmaddr))
1086                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1087
1088                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1089                         if (kvm_clear_guest(vcpu->kvm, start, PAGE_SIZE))
1090                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1091                 }
1092
1093                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
1094                         int rc = kvm_s390_skey_check_enable(vcpu);
1095
1096                         if (rc)
1097                                 return rc;
1098                         mmap_read_lock(current->mm);
1099                         rc = cond_set_guest_storage_key(current->mm, vmaddr,
1100                                                         key, NULL, nq, mr, mc);
1101                         if (rc < 0) {
1102                                 rc = fixup_user_fault(current->mm, vmaddr,
1103                                                       FAULT_FLAG_WRITE, &unlocked);
1104                                 rc = !rc ? -EAGAIN : rc;
1105                         }
1106                         mmap_read_unlock(current->mm);
1107                         if (rc == -EFAULT)
1108                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1109                         if (rc == -EAGAIN)
1110                                 continue;
1111                         if (rc < 0)
1112                                 return rc;
1113                 }
1114                 start += PAGE_SIZE;
1115         }
1116         if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1117                 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT) {
1118                         vcpu->run->s.regs.gprs[reg2] = end;
1119                 } else {
1120                         vcpu->run->s.regs.gprs[reg2] &= ~0xffffffffUL;
1121                         end = kvm_s390_logical_to_effective(vcpu, end);
1122                         vcpu->run->s.regs.gprs[reg2] |= end;
1123                 }
1124         }
1125         return 0;
1126 }
1127
1128 /*
1129  * Must be called with relevant read locks held (kvm->mm->mmap_lock, kvm->srcu)
1130  */
1131 static inline int __do_essa(struct kvm_vcpu *vcpu, const int orc)
1132 {
1133         int r1, r2, nappended, entries;
1134         unsigned long gfn, hva, res, pgstev, ptev;
1135         unsigned long *cbrlo;
1136
1137         /*
1138          * We don't need to set SD.FPF.SK to 1 here, because if we have a
1139          * machine check here we either handle it or crash
1140          */
1141
1142         kvm_s390_get_regs_rre(vcpu, &r1, &r2);
1143         gfn = vcpu->run->s.regs.gprs[r2] >> PAGE_SHIFT;
1144         hva = gfn_to_hva(vcpu->kvm, gfn);
1145         entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1146
1147         if (kvm_is_error_hva(hva))
1148                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1149
1150         nappended = pgste_perform_essa(vcpu->kvm->mm, hva, orc, &ptev, &pgstev);
1151         if (nappended < 0) {
1152                 res = orc ? 0x10 : 0;
1153                 vcpu->run->s.regs.gprs[r1] = res; /* Exception Indication */
1154                 return 0;
1155         }
1156         res = (pgstev & _PGSTE_GPS_USAGE_MASK) >> 22;
1157         /*
1158          * Set the block-content state part of the result. 0 means resident, so
1159          * nothing to do if the page is valid. 2 is for preserved pages
1160          * (non-present and non-zero), and 3 for zero pages (non-present and
1161          * zero).
1162          */
1163         if (ptev & _PAGE_INVALID) {
1164                 res |= 2;
1165                 if (pgstev & _PGSTE_GPS_ZERO)
1166                         res |= 1;
1167         }
1168         if (pgstev & _PGSTE_GPS_NODAT)
1169                 res |= 0x20;
1170         vcpu->run->s.regs.gprs[r1] = res;
1171         /*
1172          * It is possible that all the normal 511 slots were full, in which case
1173          * we will now write in the 512th slot, which is reserved for host use.
1174          * In both cases we let the normal essa handling code process all the
1175          * slots, including the reserved one, if needed.
1176          */
1177         if (nappended > 0) {
1178                 cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo & PAGE_MASK);
1179                 cbrlo[entries] = gfn << PAGE_SHIFT;
1180         }
1181
1182         if (orc) {
1183                 struct kvm_memory_slot *ms = gfn_to_memslot(vcpu->kvm, gfn);
1184
1185                 /* Increment only if we are really flipping the bit */
1186                 if (ms && !test_and_set_bit(gfn - ms->base_gfn, kvm_second_dirty_bitmap(ms)))
1187                         atomic64_inc(&vcpu->kvm->arch.cmma_dirty_pages);
1188         }
1189
1190         return nappended;
1191 }
1192
1193 static int handle_essa(struct kvm_vcpu *vcpu)
1194 {
1195         /* entries expected to be 1FF */
1196         int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1197         unsigned long *cbrlo;
1198         struct gmap *gmap;
1199         int i, orc;
1200
1201         VCPU_EVENT(vcpu, 4, "ESSA: release %d pages", entries);
1202         gmap = vcpu->arch.gmap;
1203         vcpu->stat.instruction_essa++;
1204         if (!vcpu->kvm->arch.use_cmma)
1205                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1206
1207         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1208                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1209         /* Check for invalid operation request code */
1210         orc = (vcpu->arch.sie_block->ipb & 0xf0000000) >> 28;
1211         /* ORCs 0-6 are always valid */
1212         if (orc > (test_kvm_facility(vcpu->kvm, 147) ? ESSA_SET_STABLE_NODAT
1213                                                 : ESSA_SET_STABLE_IF_RESIDENT))
1214                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1215
1216         if (!vcpu->kvm->arch.migration_mode) {
1217                 /*
1218                  * CMMA is enabled in the KVM settings, but is disabled in
1219                  * the SIE block and in the mm_context, and we are not doing
1220                  * a migration. Enable CMMA in the mm_context.
1221                  * Since we need to take a write lock to write to the context
1222                  * to avoid races with storage keys handling, we check if the
1223                  * value really needs to be written to; if the value is
1224                  * already correct, we do nothing and avoid the lock.
1225                  */
1226                 if (vcpu->kvm->mm->context.uses_cmm == 0) {
1227                         mmap_write_lock(vcpu->kvm->mm);
1228                         vcpu->kvm->mm->context.uses_cmm = 1;
1229                         mmap_write_unlock(vcpu->kvm->mm);
1230                 }
1231                 /*
1232                  * If we are here, we are supposed to have CMMA enabled in
1233                  * the SIE block. Enabling CMMA works on a per-CPU basis,
1234                  * while the context use_cmma flag is per process.
1235                  * It's possible that the context flag is enabled and the
1236                  * SIE flag is not, so we set the flag always; if it was
1237                  * already set, nothing changes, otherwise we enable it
1238                  * on this CPU too.
1239                  */
1240                 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
1241                 /* Retry the ESSA instruction */
1242                 kvm_s390_retry_instr(vcpu);
1243         } else {
1244                 int srcu_idx;
1245
1246                 mmap_read_lock(vcpu->kvm->mm);
1247                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1248                 i = __do_essa(vcpu, orc);
1249                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1250                 mmap_read_unlock(vcpu->kvm->mm);
1251                 if (i < 0)
1252                         return i;
1253                 /* Account for the possible extra cbrl entry */
1254                 entries += i;
1255         }
1256         vcpu->arch.sie_block->cbrlo &= PAGE_MASK;       /* reset nceo */
1257         cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
1258         mmap_read_lock(gmap->mm);
1259         for (i = 0; i < entries; ++i)
1260                 __gmap_zap(gmap, cbrlo[i]);
1261         mmap_read_unlock(gmap->mm);
1262         return 0;
1263 }
1264
1265 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
1266 {
1267         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1268         case 0x8a:
1269         case 0x8e:
1270         case 0x8f:
1271                 return handle_ipte_interlock(vcpu);
1272         case 0x8d:
1273                 return handle_epsw(vcpu);
1274         case 0xab:
1275                 return handle_essa(vcpu);
1276         case 0xaf:
1277                 return handle_pfmf(vcpu);
1278         default:
1279                 return -EOPNOTSUPP;
1280         }
1281 }
1282
1283 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
1284 {
1285         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1286         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1287         int reg, rc, nr_regs;
1288         u32 ctl_array[16];
1289         u64 ga;
1290         u8 ar;
1291
1292         vcpu->stat.instruction_lctl++;
1293
1294         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1295                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1296
1297         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1298
1299         if (ga & 3)
1300                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1301
1302         VCPU_EVENT(vcpu, 4, "LCTL: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1303         trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
1304
1305         nr_regs = ((reg3 - reg1) & 0xf) + 1;
1306         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1307         if (rc)
1308                 return kvm_s390_inject_prog_cond(vcpu, rc);
1309         reg = reg1;
1310         nr_regs = 0;
1311         do {
1312                 vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
1313                 vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
1314                 if (reg == reg3)
1315                         break;
1316                 reg = (reg + 1) % 16;
1317         } while (1);
1318         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1319         return 0;
1320 }
1321
1322 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
1323 {
1324         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1325         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1326         int reg, rc, nr_regs;
1327         u32 ctl_array[16];
1328         u64 ga;
1329         u8 ar;
1330
1331         vcpu->stat.instruction_stctl++;
1332
1333         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1334                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1335
1336         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1337
1338         if (ga & 3)
1339                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1340
1341         VCPU_EVENT(vcpu, 4, "STCTL r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1342         trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
1343
1344         reg = reg1;
1345         nr_regs = 0;
1346         do {
1347                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1348                 if (reg == reg3)
1349                         break;
1350                 reg = (reg + 1) % 16;
1351         } while (1);
1352         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1353         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1354 }
1355
1356 static int handle_lctlg(struct kvm_vcpu *vcpu)
1357 {
1358         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1359         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1360         int reg, rc, nr_regs;
1361         u64 ctl_array[16];
1362         u64 ga;
1363         u8 ar;
1364
1365         vcpu->stat.instruction_lctlg++;
1366
1367         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1368                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1369
1370         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1371
1372         if (ga & 7)
1373                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1374
1375         VCPU_EVENT(vcpu, 4, "LCTLG: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1376         trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
1377
1378         nr_regs = ((reg3 - reg1) & 0xf) + 1;
1379         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1380         if (rc)
1381                 return kvm_s390_inject_prog_cond(vcpu, rc);
1382         reg = reg1;
1383         nr_regs = 0;
1384         do {
1385                 vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
1386                 if (reg == reg3)
1387                         break;
1388                 reg = (reg + 1) % 16;
1389         } while (1);
1390         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1391         return 0;
1392 }
1393
1394 static int handle_stctg(struct kvm_vcpu *vcpu)
1395 {
1396         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1397         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1398         int reg, rc, nr_regs;
1399         u64 ctl_array[16];
1400         u64 ga;
1401         u8 ar;
1402
1403         vcpu->stat.instruction_stctg++;
1404
1405         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1406                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1407
1408         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1409
1410         if (ga & 7)
1411                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1412
1413         VCPU_EVENT(vcpu, 4, "STCTG r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1414         trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
1415
1416         reg = reg1;
1417         nr_regs = 0;
1418         do {
1419                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1420                 if (reg == reg3)
1421                         break;
1422                 reg = (reg + 1) % 16;
1423         } while (1);
1424         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1425         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1426 }
1427
1428 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
1429 {
1430         switch (vcpu->arch.sie_block->ipb & 0x000000ff) {
1431         case 0x25:
1432                 return handle_stctg(vcpu);
1433         case 0x2f:
1434                 return handle_lctlg(vcpu);
1435         case 0x60:
1436         case 0x61:
1437         case 0x62:
1438                 return handle_ri(vcpu);
1439         default:
1440                 return -EOPNOTSUPP;
1441         }
1442 }
1443
1444 static int handle_tprot(struct kvm_vcpu *vcpu)
1445 {
1446         u64 address1, address2;
1447         unsigned long hva, gpa;
1448         int ret = 0, cc = 0;
1449         bool writable;
1450         u8 ar;
1451
1452         vcpu->stat.instruction_tprot++;
1453
1454         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1455                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1456
1457         kvm_s390_get_base_disp_sse(vcpu, &address1, &address2, &ar, NULL);
1458
1459         /* we only handle the Linux memory detection case:
1460          * access key == 0
1461          * everything else goes to userspace. */
1462         if (address2 & 0xf0)
1463                 return -EOPNOTSUPP;
1464         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1465                 ipte_lock(vcpu);
1466         ret = guest_translate_address(vcpu, address1, ar, &gpa, GACC_STORE);
1467         if (ret == PGM_PROTECTION) {
1468                 /* Write protected? Try again with read-only... */
1469                 cc = 1;
1470                 ret = guest_translate_address(vcpu, address1, ar, &gpa,
1471                                               GACC_FETCH);
1472         }
1473         if (ret) {
1474                 if (ret == PGM_ADDRESSING || ret == PGM_TRANSLATION_SPEC) {
1475                         ret = kvm_s390_inject_program_int(vcpu, ret);
1476                 } else if (ret > 0) {
1477                         /* Translation not available */
1478                         kvm_s390_set_psw_cc(vcpu, 3);
1479                         ret = 0;
1480                 }
1481                 goto out_unlock;
1482         }
1483
1484         hva = gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
1485         if (kvm_is_error_hva(hva)) {
1486                 ret = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1487         } else {
1488                 if (!writable)
1489                         cc = 1;         /* Write not permitted ==> read-only */
1490                 kvm_s390_set_psw_cc(vcpu, cc);
1491                 /* Note: CC2 only occurs for storage keys (not supported yet) */
1492         }
1493 out_unlock:
1494         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1495                 ipte_unlock(vcpu);
1496         return ret;
1497 }
1498
1499 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
1500 {
1501         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1502         case 0x01:
1503                 return handle_tprot(vcpu);
1504         default:
1505                 return -EOPNOTSUPP;
1506         }
1507 }
1508
1509 static int handle_sckpf(struct kvm_vcpu *vcpu)
1510 {
1511         u32 value;
1512
1513         vcpu->stat.instruction_sckpf++;
1514
1515         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1516                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1517
1518         if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
1519                 return kvm_s390_inject_program_int(vcpu,
1520                                                    PGM_SPECIFICATION);
1521
1522         value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
1523         vcpu->arch.sie_block->todpr = value;
1524
1525         return 0;
1526 }
1527
1528 static int handle_ptff(struct kvm_vcpu *vcpu)
1529 {
1530         vcpu->stat.instruction_ptff++;
1531
1532         /* we don't emulate any control instructions yet */
1533         kvm_s390_set_psw_cc(vcpu, 3);
1534         return 0;
1535 }
1536
1537 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1538 {
1539         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1540         case 0x04:
1541                 return handle_ptff(vcpu);
1542         case 0x07:
1543                 return handle_sckpf(vcpu);
1544         default:
1545                 return -EOPNOTSUPP;
1546         }
1547 }