arch/s390/kvm/priv.c

   1 /*
   2  * handling privileged instructions
   3  *
   4  * Copyright IBM Corp. 2008, 2013
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License (version 2 only)
   8  * as published by the Free Software Foundation.
   9  *
  10  *    Author(s): Carsten Otte <cotte@de.ibm.com>
  11  *               Christian Borntraeger <borntraeger@de.ibm.com>
  12  */
  13
  14 #include <linux/kvm.h>
  15 #include <linux/gfp.h>
  16 #include <linux/errno.h>
  17 #include <linux/compat.h>
  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/pgtable.h>
  25 #include <asm/pgalloc.h>
  26 #include <asm/io.h>
  27 #include <asm/ptrace.h>
  28 #include <asm/compat.h>
  29 #include "gaccess.h"
  30 #include "kvm-s390.h"
  31 #include "trace.h"
  32
  33 /* Handle SCK (SET CLOCK) interception */
  34 static int handle_set_clock(struct kvm_vcpu *vcpu)
  35 {
  36         struct kvm_vcpu *cpup;
  37         s64 hostclk, val;
  38         int i, rc;
  39         u64 op2;
  40
  41         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
  42                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
  43
  44         op2 = kvm_s390_get_base_disp_s(vcpu);
  45         if (op2 & 7)    /* Operand must be on a doubleword boundary */
  46                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
  47         rc = read_guest(vcpu, op2, &val, sizeof(val));
  48         if (rc)
  49                 return kvm_s390_inject_prog_cond(vcpu, rc);
  50
  51         if (store_tod_clock(&hostclk)) {
  52                 kvm_s390_set_psw_cc(vcpu, 3);
  53                 return 0;
  54         }
  55         val = (val - hostclk) & ~0x3fUL;
  56
  57         mutex_lock(&vcpu->kvm->lock);
  58         kvm_for_each_vcpu(i, cpup, vcpu->kvm)
  59                 cpup->arch.sie_block->epoch = val;
  60         mutex_unlock(&vcpu->kvm->lock);
  61
  62         kvm_s390_set_psw_cc(vcpu, 0);
  63         return 0;
  64 }
  65
  66 static int handle_set_prefix(struct kvm_vcpu *vcpu)
  67 {
  68         u64 operand2;
  69         u32 address;
  70         int rc;
  71
  72         vcpu->stat.instruction_spx++;
  73
  74         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
  75                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
  76
  77         operand2 = kvm_s390_get_base_disp_s(vcpu);
  78
  79         /* must be word boundary */
  80         if (operand2 & 3)
  81                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
  82
  83         /* get the value */
  84         rc = read_guest(vcpu, operand2, &address, sizeof(address));
  85         if (rc)
  86                 return kvm_s390_inject_prog_cond(vcpu, rc);
  87
  88         address &= 0x7fffe000u;
  89
  90         /*
  91          * Make sure the new value is valid memory. We only need to check the
  92          * first page, since address is 8k aligned and memory pieces are always
  93          * at least 1MB aligned and have at least a size of 1MB.
  94          */
  95         if (kvm_is_error_gpa(vcpu->kvm, address))
  96                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
  97
  98         kvm_s390_set_prefix(vcpu, address);
  99
 100         VCPU_EVENT(vcpu, 5, "setting prefix to %x", address);
 101         trace_kvm_s390_handle_prefix(vcpu, 1, address);
 102         return 0;
 103 }
 104
 105 static int handle_store_prefix(struct kvm_vcpu *vcpu)
 106 {
 107         u64 operand2;
 108         u32 address;
 109         int rc;
 110
 111         vcpu->stat.instruction_stpx++;
 112
 113         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 114                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 115
 116         operand2 = kvm_s390_get_base_disp_s(vcpu);
 117
 118         /* must be word boundary */
 119         if (operand2 & 3)
 120                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 121
 122         address = vcpu->arch.sie_block->prefix;
 123         address = address & 0x7fffe000u;
 124
 125         /* get the value */
 126         rc = write_guest(vcpu, operand2, &address, sizeof(address));
 127         if (rc)
 128                 return kvm_s390_inject_prog_cond(vcpu, rc);
 129
 130         VCPU_EVENT(vcpu, 5, "storing prefix to %x", address);
 131         trace_kvm_s390_handle_prefix(vcpu, 0, address);
 132         return 0;
 133 }
 134
 135 static int handle_store_cpu_address(struct kvm_vcpu *vcpu)
 136 {
 137         u16 vcpu_id = vcpu->vcpu_id;
 138         u64 ga;
 139         int rc;
 140
 141         vcpu->stat.instruction_stap++;
 142
 143         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 144                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 145
 146         ga = kvm_s390_get_base_disp_s(vcpu);
 147
 148         if (ga & 1)
 149                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 150
 151         rc = write_guest(vcpu, ga, &vcpu_id, sizeof(vcpu_id));
 152         if (rc)
 153                 return kvm_s390_inject_prog_cond(vcpu, rc);
 154
 155         VCPU_EVENT(vcpu, 5, "storing cpu address to %llx", ga);
 156         trace_kvm_s390_handle_stap(vcpu, ga);
 157         return 0;
 158 }
 159
 160 static void __skey_check_enable(struct kvm_vcpu *vcpu)
 161 {
 162         if (!(vcpu->arch.sie_block->ictl & (ICTL_ISKE | ICTL_SSKE | ICTL_RRBE)))
 163                 return;
 164
 165         s390_enable_skey();
 166         trace_kvm_s390_skey_related_inst(vcpu);
 167         vcpu->arch.sie_block->ictl &= ~(ICTL_ISKE | ICTL_SSKE | ICTL_RRBE);
 168 }
 169
 170
 171 static int handle_skey(struct kvm_vcpu *vcpu)
 172 {
 173         __skey_check_enable(vcpu);
 174
 175         vcpu->stat.instruction_storage_key++;
 176
 177         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 178                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 179
 180         vcpu->arch.sie_block->gpsw.addr =
 181                 __rewind_psw(vcpu->arch.sie_block->gpsw, 4);
 182         VCPU_EVENT(vcpu, 4, "%s", "retrying storage key operation");
 183         return 0;
 184 }
 185
 186 static int handle_ipte_interlock(struct kvm_vcpu *vcpu)
 187 {
 188         psw_t *psw = &vcpu->arch.sie_block->gpsw;
 189
 190         vcpu->stat.instruction_ipte_interlock++;
 191         if (psw_bits(*psw).p)
 192                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 193         wait_event(vcpu->kvm->arch.ipte_wq, !ipte_lock_held(vcpu));
 194         psw->addr = __rewind_psw(*psw, 4);
 195         VCPU_EVENT(vcpu, 4, "%s", "retrying ipte interlock operation");
 196         return 0;
 197 }
 198
 199 static int handle_test_block(struct kvm_vcpu *vcpu)
 200 {
 201         gpa_t addr;
 202         int reg2;
 203
 204         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 205                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 206
 207         kvm_s390_get_regs_rre(vcpu, NULL, &reg2);
 208         addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 209         addr = kvm_s390_real_to_abs(vcpu, addr);
 210
 211         if (kvm_is_error_gpa(vcpu->kvm, addr))
 212                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 213         /*
 214          * We don't expect errors on modern systems, and do not care
 215          * about storage keys (yet), so let's just clear the page.
 216          */
 217         if (kvm_clear_guest(vcpu->kvm, addr, PAGE_SIZE))
 218                 return -EFAULT;
 219         kvm_s390_set_psw_cc(vcpu, 0);
 220         vcpu->run->s.regs.gprs[0] = 0;
 221         return 0;
 222 }
 223
 224 static int handle_tpi(struct kvm_vcpu *vcpu)
 225 {
 226         struct kvm_s390_interrupt_info *inti;
 227         unsigned long len;
 228         u32 tpi_data[3];
 229         int cc, rc;
 230         u64 addr;
 231
 232         rc = 0;
 233         addr = kvm_s390_get_base_disp_s(vcpu);
 234         if (addr & 3)
 235                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 236         cc = 0;
 237         inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0);
 238         if (!inti)
 239                 goto no_interrupt;
 240         cc = 1;
 241         tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr;
 242         tpi_data[1] = inti->io.io_int_parm;
 243         tpi_data[2] = inti->io.io_int_word;
 244         if (addr) {
 245                 /*
 246                  * Store the two-word I/O interruption code into the
 247                  * provided area.
 248                  */
 249                 len = sizeof(tpi_data) - 4;
 250                 rc = write_guest(vcpu, addr, &tpi_data, len);
 251                 if (rc)
 252                         return kvm_s390_inject_prog_cond(vcpu, rc);
 253         } else {
 254                 /*
 255                  * Store the three-word I/O interruption code into
 256                  * the appropriate lowcore area.
 257                  */
 258                 len = sizeof(tpi_data);
 259                 if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len))
 260                         rc = -EFAULT;
 261         }
 262         /*
 263          * If we encounter a problem storing the interruption code, the
 264          * instruction is suppressed from the guest's view: reinject the
 265          * interrupt.
 266          */
 267         if (!rc)
 268                 kfree(inti);
 269         else
 270                 kvm_s390_reinject_io_int(vcpu->kvm, inti);
 271 no_interrupt:
 272         /* Set condition code and we're done. */
 273         if (!rc)
 274                 kvm_s390_set_psw_cc(vcpu, cc);
 275         return rc ? -EFAULT : 0;
 276 }
 277
 278 static int handle_tsch(struct kvm_vcpu *vcpu)
 279 {
 280         struct kvm_s390_interrupt_info *inti;
 281
 282         inti = kvm_s390_get_io_int(vcpu->kvm, 0,
 283                                    vcpu->run->s.regs.gprs[1]);
 284
 285         /*
 286          * Prepare exit to userspace.
 287          * We indicate whether we dequeued a pending I/O interrupt
 288          * so that userspace can re-inject it if the instruction gets
 289          * a program check. While this may re-order the pending I/O
 290          * interrupts, this is no problem since the priority is kept
 291          * intact.
 292          */
 293         vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
 294         vcpu->run->s390_tsch.dequeued = !!inti;
 295         if (inti) {
 296                 vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
 297                 vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
 298                 vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
 299                 vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
 300         }
 301         vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
 302         kfree(inti);
 303         return -EREMOTE;
 304 }
 305
 306 static int handle_io_inst(struct kvm_vcpu *vcpu)
 307 {
 308         VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
 309
 310         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 311                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 312
 313         if (vcpu->kvm->arch.css_support) {
 314                 /*
 315                  * Most I/O instructions will be handled by userspace.
 316                  * Exceptions are tpi and the interrupt portion of tsch.
 317                  */
 318                 if (vcpu->arch.sie_block->ipa == 0xb236)
 319                         return handle_tpi(vcpu);
 320                 if (vcpu->arch.sie_block->ipa == 0xb235)
 321                         return handle_tsch(vcpu);
 322                 /* Handle in userspace. */
 323                 return -EOPNOTSUPP;
 324         } else {
 325                 /*
 326                  * Set condition code 3 to stop the guest from issuing channel
 327                  * I/O instructions.
 328                  */
 329                 kvm_s390_set_psw_cc(vcpu, 3);
 330                 return 0;
 331         }
 332 }
 333
 334 static int handle_stfl(struct kvm_vcpu *vcpu)
 335 {
 336         int rc;
 337
 338         vcpu->stat.instruction_stfl++;
 339
 340         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 341                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 342
 343         rc = write_guest_lc(vcpu, offsetof(struct _lowcore, stfl_fac_list),
 344                             vfacilities, 4);
 345         if (rc)
 346                 return rc;
 347         VCPU_EVENT(vcpu, 5, "store facility list value %x",
 348                    *(unsigned int *) vfacilities);
 349         trace_kvm_s390_handle_stfl(vcpu, *(unsigned int *) vfacilities);
 350         return 0;
 351 }
 352
 353 static void handle_new_psw(struct kvm_vcpu *vcpu)
 354 {
 355         /* Check whether the new psw is enabled for machine checks. */
 356         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_MCHECK)
 357                 kvm_s390_deliver_pending_machine_checks(vcpu);
 358 }
 359
 360 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
 361 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
 362 #define PSW_ADDR_24 0x0000000000ffffffUL
 363 #define PSW_ADDR_31 0x000000007fffffffUL
 364
 365 static int is_valid_psw(psw_t *psw) {
 366         if (psw->mask & PSW_MASK_UNASSIGNED)
 367                 return 0;
 368         if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
 369                 if (psw->addr & ~PSW_ADDR_31)
 370                         return 0;
 371         }
 372         if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
 373                 return 0;
 374         if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
 375                 return 0;
 376         return 1;
 377 }
 378
 379 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
 380 {
 381         psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
 382         psw_compat_t new_psw;
 383         u64 addr;
 384         int rc;
 385
 386         if (gpsw->mask & PSW_MASK_PSTATE)
 387                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 388
 389         addr = kvm_s390_get_base_disp_s(vcpu);
 390         if (addr & 7)
 391                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 392
 393         rc = read_guest(vcpu, addr, &new_psw, sizeof(new_psw));
 394         if (rc)
 395                 return kvm_s390_inject_prog_cond(vcpu, rc);
 396         if (!(new_psw.mask & PSW32_MASK_BASE))
 397                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 398         gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
 399         gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
 400         gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
 401         if (!is_valid_psw(gpsw))
 402                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 403         handle_new_psw(vcpu);
 404         return 0;
 405 }
 406
 407 static int handle_lpswe(struct kvm_vcpu *vcpu)
 408 {
 409         psw_t new_psw;
 410         u64 addr;
 411         int rc;
 412
 413         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 414                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 415
 416         addr = kvm_s390_get_base_disp_s(vcpu);
 417         if (addr & 7)
 418                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 419         rc = read_guest(vcpu, addr, &new_psw, sizeof(new_psw));
 420         if (rc)
 421                 return kvm_s390_inject_prog_cond(vcpu, rc);
 422         vcpu->arch.sie_block->gpsw = new_psw;
 423         if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
 424                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 425         handle_new_psw(vcpu);
 426         return 0;
 427 }
 428
 429 static int handle_stidp(struct kvm_vcpu *vcpu)
 430 {
 431         u64 stidp_data = vcpu->arch.stidp_data;
 432         u64 operand2;
 433         int rc;
 434
 435         vcpu->stat.instruction_stidp++;
 436
 437         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 438                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 439
 440         operand2 = kvm_s390_get_base_disp_s(vcpu);
 441
 442         if (operand2 & 7)
 443                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 444
 445         rc = write_guest(vcpu, operand2, &stidp_data, sizeof(stidp_data));
 446         if (rc)
 447                 return kvm_s390_inject_prog_cond(vcpu, rc);
 448
 449         VCPU_EVENT(vcpu, 5, "%s", "store cpu id");
 450         return 0;
 451 }
 452
 453 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
 454 {
 455         int cpus = 0;
 456         int n;
 457
 458         cpus = atomic_read(&vcpu->kvm->online_vcpus);
 459
 460         /* deal with other level 3 hypervisors */
 461         if (stsi(mem, 3, 2, 2))
 462                 mem->count = 0;
 463         if (mem->count < 8)
 464                 mem->count++;
 465         for (n = mem->count - 1; n > 0 ; n--)
 466                 memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
 467
 468         mem->vm[0].cpus_total = cpus;
 469         mem->vm[0].cpus_configured = cpus;
 470         mem->vm[0].cpus_standby = 0;
 471         mem->vm[0].cpus_reserved = 0;
 472         mem->vm[0].caf = 1000;
 473         memcpy(mem->vm[0].name, "KVMguest", 8);
 474         ASCEBC(mem->vm[0].name, 8);
 475         memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
 476         ASCEBC(mem->vm[0].cpi, 16);
 477 }
 478
 479 static int handle_stsi(struct kvm_vcpu *vcpu)
 480 {
 481         int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
 482         int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
 483         int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
 484         unsigned long mem = 0;
 485         u64 operand2;
 486         int rc = 0;
 487
 488         vcpu->stat.instruction_stsi++;
 489         VCPU_EVENT(vcpu, 4, "stsi: fc: %x sel1: %x sel2: %x", fc, sel1, sel2);
 490
 491         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 492                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 493
 494         if (fc > 3) {
 495                 kvm_s390_set_psw_cc(vcpu, 3);
 496                 return 0;
 497         }
 498
 499         if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
 500             || vcpu->run->s.regs.gprs[1] & 0xffff0000)
 501                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 502
 503         if (fc == 0) {
 504                 vcpu->run->s.regs.gprs[0] = 3 << 28;
 505                 kvm_s390_set_psw_cc(vcpu, 0);
 506                 return 0;
 507         }
 508
 509         operand2 = kvm_s390_get_base_disp_s(vcpu);
 510
 511         if (operand2 & 0xfff)
 512                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 513
 514         switch (fc) {
 515         case 1: /* same handling for 1 and 2 */
 516         case 2:
 517                 mem = get_zeroed_page(GFP_KERNEL);
 518                 if (!mem)
 519                         goto out_no_data;
 520                 if (stsi((void *) mem, fc, sel1, sel2))
 521                         goto out_no_data;
 522                 break;
 523         case 3:
 524                 if (sel1 != 2 || sel2 != 2)
 525                         goto out_no_data;
 526                 mem = get_zeroed_page(GFP_KERNEL);
 527                 if (!mem)
 528                         goto out_no_data;
 529                 handle_stsi_3_2_2(vcpu, (void *) mem);
 530                 break;
 531         }
 532
 533         rc = write_guest(vcpu, operand2, (void *)mem, PAGE_SIZE);
 534         if (rc) {
 535                 rc = kvm_s390_inject_prog_cond(vcpu, rc);
 536                 goto out;
 537         }
 538         trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
 539         free_page(mem);
 540         kvm_s390_set_psw_cc(vcpu, 0);
 541         vcpu->run->s.regs.gprs[0] = 0;
 542         return 0;
 543 out_no_data:
 544         kvm_s390_set_psw_cc(vcpu, 3);
 545 out:
 546         free_page(mem);
 547         return rc;
 548 }
 549
 550 static const intercept_handler_t b2_handlers[256] = {
 551         [0x02] = handle_stidp,
 552         [0x04] = handle_set_clock,
 553         [0x10] = handle_set_prefix,
 554         [0x11] = handle_store_prefix,
 555         [0x12] = handle_store_cpu_address,
 556         [0x21] = handle_ipte_interlock,
 557         [0x29] = handle_skey,
 558         [0x2a] = handle_skey,
 559         [0x2b] = handle_skey,
 560         [0x2c] = handle_test_block,
 561         [0x30] = handle_io_inst,
 562         [0x31] = handle_io_inst,
 563         [0x32] = handle_io_inst,
 564         [0x33] = handle_io_inst,
 565         [0x34] = handle_io_inst,
 566         [0x35] = handle_io_inst,
 567         [0x36] = handle_io_inst,
 568         [0x37] = handle_io_inst,
 569         [0x38] = handle_io_inst,
 570         [0x39] = handle_io_inst,
 571         [0x3a] = handle_io_inst,
 572         [0x3b] = handle_io_inst,
 573         [0x3c] = handle_io_inst,
 574         [0x50] = handle_ipte_interlock,
 575         [0x5f] = handle_io_inst,
 576         [0x74] = handle_io_inst,
 577         [0x76] = handle_io_inst,
 578         [0x7d] = handle_stsi,
 579         [0xb1] = handle_stfl,
 580         [0xb2] = handle_lpswe,
 581 };
 582
 583 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
 584 {
 585         intercept_handler_t handler;
 586
 587         /*
 588          * A lot of B2 instructions are priviledged. Here we check for
 589          * the privileged ones, that we can handle in the kernel.
 590          * Anything else goes to userspace.
 591          */
 592         handler = b2_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
 593         if (handler)
 594                 return handler(vcpu);
 595
 596         return -EOPNOTSUPP;
 597 }
 598
 599 static int handle_epsw(struct kvm_vcpu *vcpu)
 600 {
 601         int reg1, reg2;
 602
 603         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 604
 605         /* This basically extracts the mask half of the psw. */
 606         vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
 607         vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
 608         if (reg2) {
 609                 vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
 610                 vcpu->run->s.regs.gprs[reg2] |=
 611                         vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
 612         }
 613         return 0;
 614 }
 615
 616 #define PFMF_RESERVED   0xfffc0101UL
 617 #define PFMF_SK         0x00020000UL
 618 #define PFMF_CF         0x00010000UL
 619 #define PFMF_UI         0x00008000UL
 620 #define PFMF_FSC        0x00007000UL
 621 #define PFMF_NQ         0x00000800UL
 622 #define PFMF_MR         0x00000400UL
 623 #define PFMF_MC         0x00000200UL
 624 #define PFMF_KEY        0x000000feUL
 625
 626 static int handle_pfmf(struct kvm_vcpu *vcpu)
 627 {
 628         int reg1, reg2;
 629         unsigned long start, end;
 630
 631         vcpu->stat.instruction_pfmf++;
 632
 633         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 634
 635         if (!MACHINE_HAS_PFMF)
 636                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
 637
 638         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 639                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 640
 641         if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
 642                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 643
 644         /* Only provide non-quiescing support if the host supports it */
 645         if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ && !test_facility(14))
 646                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 647
 648         /* No support for conditional-SSKE */
 649         if (vcpu->run->s.regs.gprs[reg1] & (PFMF_MR | PFMF_MC))
 650                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 651
 652         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 653         if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
 654                 if (kvm_s390_check_low_addr_protection(vcpu, start))
 655                         return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
 656         }
 657
 658         switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
 659         case 0x00000000:
 660                 end = (start + (1UL << 12)) & ~((1UL << 12) - 1);
 661                 break;
 662         case 0x00001000:
 663                 end = (start + (1UL << 20)) & ~((1UL << 20) - 1);
 664                 break;
 665         /* We dont support EDAT2
 666         case 0x00002000:
 667                 end = (start + (1UL << 31)) & ~((1UL << 31) - 1);
 668                 break;*/
 669         default:
 670                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 671         }
 672         while (start < end) {
 673                 unsigned long useraddr, abs_addr;
 674
 675                 /* Translate guest address to host address */
 676                 if ((vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) == 0)
 677                         abs_addr = kvm_s390_real_to_abs(vcpu, start);
 678                 else
 679                         abs_addr = start;
 680                 useraddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(abs_addr));
 681                 if (kvm_is_error_hva(useraddr))
 682                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 683
 684                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
 685                         if (clear_user((void __user *)useraddr, PAGE_SIZE))
 686                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 687                 }
 688
 689                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
 690                         __skey_check_enable(vcpu);
 691                         if (set_guest_storage_key(current->mm, useraddr,
 692                                         vcpu->run->s.regs.gprs[reg1] & PFMF_KEY,
 693                                         vcpu->run->s.regs.gprs[reg1] & PFMF_NQ))
 694                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 695                 }
 696
 697                 start += PAGE_SIZE;
 698         }
 699         if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC)
 700                 vcpu->run->s.regs.gprs[reg2] = end;
 701         return 0;
 702 }
 703
 704 static int handle_essa(struct kvm_vcpu *vcpu)
 705 {
 706         /* entries expected to be 1FF */
 707         int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
 708         unsigned long *cbrlo, cbrle;
 709         struct gmap *gmap;
 710         int i;
 711
 712         VCPU_EVENT(vcpu, 5, "cmma release %d pages", entries);
 713         gmap = vcpu->arch.gmap;
 714         vcpu->stat.instruction_essa++;
 715         if (!kvm_s390_cmma_enabled(vcpu->kvm))
 716                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
 717
 718         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 719                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 720
 721         if (((vcpu->arch.sie_block->ipb & 0xf0000000) >> 28) > 6)
 722                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 723
 724         /* Rewind PSW to repeat the ESSA instruction */
 725         vcpu->arch.sie_block->gpsw.addr =
 726                 __rewind_psw(vcpu->arch.sie_block->gpsw, 4);
 727         vcpu->arch.sie_block->cbrlo &= PAGE_MASK;       /* reset nceo */
 728         cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
 729         down_read(&gmap->mm->mmap_sem);
 730         for (i = 0; i < entries; ++i) {
 731                 cbrle = cbrlo[i];
 732                 if (unlikely(cbrle & ~PAGE_MASK || cbrle < 2 * PAGE_SIZE))
 733                         /* invalid entry */
 734                         break;
 735                 /* try to free backing */
 736                 __gmap_zap(cbrle, gmap);
 737         }
 738         up_read(&gmap->mm->mmap_sem);
 739         if (i < entries)
 740                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 741         return 0;
 742 }
 743
 744 static const intercept_handler_t b9_handlers[256] = {
 745         [0x8a] = handle_ipte_interlock,
 746         [0x8d] = handle_epsw,
 747         [0x8e] = handle_ipte_interlock,
 748         [0x8f] = handle_ipte_interlock,
 749         [0xab] = handle_essa,
 750         [0xaf] = handle_pfmf,
 751 };
 752
 753 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
 754 {
 755         intercept_handler_t handler;
 756
 757         /* This is handled just as for the B2 instructions. */
 758         handler = b9_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
 759         if (handler)
 760                 return handler(vcpu);
 761
 762         return -EOPNOTSUPP;
 763 }
 764
 765 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
 766 {
 767         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
 768         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
 769         u32 val = 0;
 770         int reg, rc;
 771         u64 ga;
 772
 773         vcpu->stat.instruction_lctl++;
 774
 775         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 776                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 777
 778         ga = kvm_s390_get_base_disp_rs(vcpu);
 779
 780         if (ga & 3)
 781                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 782
 783         VCPU_EVENT(vcpu, 5, "lctl r1:%x, r3:%x, addr:%llx", reg1, reg3, ga);
 784         trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
 785
 786         reg = reg1;
 787         do {
 788                 rc = read_guest(vcpu, ga, &val, sizeof(val));
 789                 if (rc)
 790                         return kvm_s390_inject_prog_cond(vcpu, rc);
 791                 vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
 792                 vcpu->arch.sie_block->gcr[reg] |= val;
 793                 ga += 4;
 794                 if (reg == reg3)
 795                         break;
 796                 reg = (reg + 1) % 16;
 797         } while (1);
 798
 799         return 0;
 800 }
 801
 802 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
 803 {
 804         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
 805         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
 806         u64 ga;
 807         u32 val;
 808         int reg, rc;
 809
 810         vcpu->stat.instruction_stctl++;
 811
 812         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 813                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 814
 815         ga = kvm_s390_get_base_disp_rs(vcpu);
 816
 817         if (ga & 3)
 818                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 819
 820         VCPU_EVENT(vcpu, 5, "stctl r1:%x, r3:%x, addr:%llx", reg1, reg3, ga);
 821         trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
 822
 823         reg = reg1;
 824         do {
 825                 val = vcpu->arch.sie_block->gcr[reg] &  0x00000000fffffffful;
 826                 rc = write_guest(vcpu, ga, &val, sizeof(val));
 827                 if (rc)
 828                         return kvm_s390_inject_prog_cond(vcpu, rc);
 829                 ga += 4;
 830                 if (reg == reg3)
 831                         break;
 832                 reg = (reg + 1) % 16;
 833         } while (1);
 834
 835         return 0;
 836 }
 837
 838 static int handle_lctlg(struct kvm_vcpu *vcpu)
 839 {
 840         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
 841         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
 842         u64 ga, val;
 843         int reg, rc;
 844
 845         vcpu->stat.instruction_lctlg++;
 846
 847         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 848                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 849
 850         ga = kvm_s390_get_base_disp_rsy(vcpu);
 851
 852         if (ga & 7)
 853                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 854
 855         reg = reg1;
 856
 857         VCPU_EVENT(vcpu, 5, "lctlg r1:%x, r3:%x, addr:%llx", reg1, reg3, ga);
 858         trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
 859
 860         do {
 861                 rc = read_guest(vcpu, ga, &val, sizeof(val));
 862                 if (rc)
 863                         return kvm_s390_inject_prog_cond(vcpu, rc);
 864                 vcpu->arch.sie_block->gcr[reg] = val;
 865                 ga += 8;
 866                 if (reg == reg3)
 867                         break;
 868                 reg = (reg + 1) % 16;
 869         } while (1);
 870
 871         return 0;
 872 }
 873
 874 static int handle_stctg(struct kvm_vcpu *vcpu)
 875 {
 876         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
 877         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
 878         u64 ga, val;
 879         int reg, rc;
 880
 881         vcpu->stat.instruction_stctg++;
 882
 883         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 884                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 885
 886         ga = kvm_s390_get_base_disp_rsy(vcpu);
 887
 888         if (ga & 7)
 889                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 890
 891         reg = reg1;
 892
 893         VCPU_EVENT(vcpu, 5, "stctg r1:%x, r3:%x, addr:%llx", reg1, reg3, ga);
 894         trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
 895
 896         do {
 897                 val = vcpu->arch.sie_block->gcr[reg];
 898                 rc = write_guest(vcpu, ga, &val, sizeof(val));
 899                 if (rc)
 900                         return kvm_s390_inject_prog_cond(vcpu, rc);
 901                 ga += 8;
 902                 if (reg == reg3)
 903                         break;
 904                 reg = (reg + 1) % 16;
 905         } while (1);
 906
 907         return 0;
 908 }
 909
 910 static const intercept_handler_t eb_handlers[256] = {
 911         [0x2f] = handle_lctlg,
 912         [0x25] = handle_stctg,
 913 };
 914
 915 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
 916 {
 917         intercept_handler_t handler;
 918
 919         handler = eb_handlers[vcpu->arch.sie_block->ipb & 0xff];
 920         if (handler)
 921                 return handler(vcpu);
 922         return -EOPNOTSUPP;
 923 }
 924
 925 static int handle_tprot(struct kvm_vcpu *vcpu)
 926 {
 927         u64 address1, address2;
 928         struct vm_area_struct *vma;
 929         unsigned long user_address;
 930
 931         vcpu->stat.instruction_tprot++;
 932
 933         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 934                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 935
 936         kvm_s390_get_base_disp_sse(vcpu, &address1, &address2);
 937
 938         /* we only handle the Linux memory detection case:
 939          * access key == 0
 940          * guest DAT == off
 941          * everything else goes to userspace. */
 942         if (address2 & 0xf0)
 943                 return -EOPNOTSUPP;
 944         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
 945                 return -EOPNOTSUPP;
 946
 947         down_read(&current->mm->mmap_sem);
 948         user_address = __gmap_translate(address1, vcpu->arch.gmap);
 949         if (IS_ERR_VALUE(user_address))
 950                 goto out_inject;
 951         vma = find_vma(current->mm, user_address);
 952         if (!vma)
 953                 goto out_inject;
 954         vcpu->arch.sie_block->gpsw.mask &= ~(3ul << 44);
 955         if (!(vma->vm_flags & VM_WRITE) && (vma->vm_flags & VM_READ))
 956                 vcpu->arch.sie_block->gpsw.mask |= (1ul << 44);
 957         if (!(vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_READ))
 958                 vcpu->arch.sie_block->gpsw.mask |= (2ul << 44);
 959
 960         up_read(&current->mm->mmap_sem);
 961         return 0;
 962
 963 out_inject:
 964         up_read(&current->mm->mmap_sem);
 965         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 966 }
 967
 968 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
 969 {
 970         /* For e5xx... instructions we only handle TPROT */
 971         if ((vcpu->arch.sie_block->ipa & 0x00ff) == 0x01)
 972                 return handle_tprot(vcpu);
 973         return -EOPNOTSUPP;
 974 }
 975
 976 static int handle_sckpf(struct kvm_vcpu *vcpu)
 977 {
 978         u32 value;
 979
 980         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 981                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 982
 983         if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
 984                 return kvm_s390_inject_program_int(vcpu,
 985                                                    PGM_SPECIFICATION);
 986
 987         value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
 988         vcpu->arch.sie_block->todpr = value;
 989
 990         return 0;
 991 }
 992
 993 static const intercept_handler_t x01_handlers[256] = {
 994         [0x07] = handle_sckpf,
 995 };
 996
 997 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
 998 {
 999         intercept_handler_t handler;
1000
1001         handler = x01_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
1002         if (handler)
1003                 return handler(vcpu);
1004         return -EOPNOTSUPP;
1005 }