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 = kvm_s390_get_prefix(vcpu);
 123
 124         /* get the value */
 125         rc = write_guest(vcpu, operand2, &address, sizeof(address));
 126         if (rc)
 127                 return kvm_s390_inject_prog_cond(vcpu, rc);
 128
 129         VCPU_EVENT(vcpu, 5, "storing prefix to %x", address);
 130         trace_kvm_s390_handle_prefix(vcpu, 0, address);
 131         return 0;
 132 }
 133
 134 static int handle_store_cpu_address(struct kvm_vcpu *vcpu)
 135 {
 136         u16 vcpu_id = vcpu->vcpu_id;
 137         u64 ga;
 138         int rc;
 139
 140         vcpu->stat.instruction_stap++;
 141
 142         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 143                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 144
 145         ga = kvm_s390_get_base_disp_s(vcpu);
 146
 147         if (ga & 1)
 148                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 149
 150         rc = write_guest(vcpu, ga, &vcpu_id, sizeof(vcpu_id));
 151         if (rc)
 152                 return kvm_s390_inject_prog_cond(vcpu, rc);
 153
 154         VCPU_EVENT(vcpu, 5, "storing cpu address to %llx", ga);
 155         trace_kvm_s390_handle_stap(vcpu, ga);
 156         return 0;
 157 }
 158
 159 static int __skey_check_enable(struct kvm_vcpu *vcpu)
 160 {
 161         int rc = 0;
 162         if (!(vcpu->arch.sie_block->ictl & (ICTL_ISKE | ICTL_SSKE | ICTL_RRBE)))
 163                 return rc;
 164
 165         rc = s390_enable_skey();
 166         trace_kvm_s390_skey_related_inst(vcpu);
 167         vcpu->arch.sie_block->ictl &= ~(ICTL_ISKE | ICTL_SSKE | ICTL_RRBE);
 168         return rc;
 169 }
 170
 171
 172 static int handle_skey(struct kvm_vcpu *vcpu)
 173 {
 174         int rc = __skey_check_enable(vcpu);
 175
 176         if (rc)
 177                 return rc;
 178         vcpu->stat.instruction_storage_key++;
 179
 180         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 181                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 182
 183         kvm_s390_rewind_psw(vcpu, 4);
 184         VCPU_EVENT(vcpu, 4, "%s", "retrying storage key operation");
 185         return 0;
 186 }
 187
 188 static int handle_ipte_interlock(struct kvm_vcpu *vcpu)
 189 {
 190         vcpu->stat.instruction_ipte_interlock++;
 191         if (psw_bits(vcpu->arch.sie_block->gpsw).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         kvm_s390_rewind_psw(vcpu, 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_logical_to_effective(vcpu, addr);
 210         if (kvm_s390_check_low_addr_prot_real(vcpu, addr))
 211                 return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
 212         addr = kvm_s390_real_to_abs(vcpu, addr);
 213
 214         if (kvm_is_error_gpa(vcpu->kvm, addr))
 215                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 216         /*
 217          * We don't expect errors on modern systems, and do not care
 218          * about storage keys (yet), so let's just clear the page.
 219          */
 220         if (kvm_clear_guest(vcpu->kvm, addr, PAGE_SIZE))
 221                 return -EFAULT;
 222         kvm_s390_set_psw_cc(vcpu, 0);
 223         vcpu->run->s.regs.gprs[0] = 0;
 224         return 0;
 225 }
 226
 227 static int handle_tpi(struct kvm_vcpu *vcpu)
 228 {
 229         struct kvm_s390_interrupt_info *inti;
 230         unsigned long len;
 231         u32 tpi_data[3];
 232         int rc;
 233         u64 addr;
 234
 235         addr = kvm_s390_get_base_disp_s(vcpu);
 236         if (addr & 3)
 237                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 238
 239         inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0);
 240         if (!inti) {
 241                 kvm_s390_set_psw_cc(vcpu, 0);
 242                 return 0;
 243         }
 244
 245         tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr;
 246         tpi_data[1] = inti->io.io_int_parm;
 247         tpi_data[2] = inti->io.io_int_word;
 248         if (addr) {
 249                 /*
 250                  * Store the two-word I/O interruption code into the
 251                  * provided area.
 252                  */
 253                 len = sizeof(tpi_data) - 4;
 254                 rc = write_guest(vcpu, addr, &tpi_data, len);
 255                 if (rc) {
 256                         rc = kvm_s390_inject_prog_cond(vcpu, rc);
 257                         goto reinject_interrupt;
 258                 }
 259         } else {
 260                 /*
 261                  * Store the three-word I/O interruption code into
 262                  * the appropriate lowcore area.
 263                  */
 264                 len = sizeof(tpi_data);
 265                 if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len)) {
 266                         /* failed writes to the low core are not recoverable */
 267                         rc = -EFAULT;
 268                         goto reinject_interrupt;
 269                 }
 270         }
 271
 272         /* irq was successfully handed to the guest */
 273         kfree(inti);
 274         kvm_s390_set_psw_cc(vcpu, 1);
 275         return 0;
 276 reinject_interrupt:
 277         /*
 278          * If we encounter a problem storing the interruption code, the
 279          * instruction is suppressed from the guest's view: reinject the
 280          * interrupt.
 281          */
 282         if (kvm_s390_reinject_io_int(vcpu->kvm, inti)) {
 283                 kfree(inti);
 284                 rc = -EFAULT;
 285         }
 286         /* don't set the cc, a pgm irq was injected or we drop to user space */
 287         return rc ? -EFAULT : 0;
 288 }
 289
 290 static int handle_tsch(struct kvm_vcpu *vcpu)
 291 {
 292         struct kvm_s390_interrupt_info *inti;
 293
 294         inti = kvm_s390_get_io_int(vcpu->kvm, 0,
 295                                    vcpu->run->s.regs.gprs[1]);
 296
 297         /*
 298          * Prepare exit to userspace.
 299          * We indicate whether we dequeued a pending I/O interrupt
 300          * so that userspace can re-inject it if the instruction gets
 301          * a program check. While this may re-order the pending I/O
 302          * interrupts, this is no problem since the priority is kept
 303          * intact.
 304          */
 305         vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
 306         vcpu->run->s390_tsch.dequeued = !!inti;
 307         if (inti) {
 308                 vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
 309                 vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
 310                 vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
 311                 vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
 312         }
 313         vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
 314         kfree(inti);
 315         return -EREMOTE;
 316 }
 317
 318 static int handle_io_inst(struct kvm_vcpu *vcpu)
 319 {
 320         VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
 321
 322         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 323                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 324
 325         if (vcpu->kvm->arch.css_support) {
 326                 /*
 327                  * Most I/O instructions will be handled by userspace.
 328                  * Exceptions are tpi and the interrupt portion of tsch.
 329                  */
 330                 if (vcpu->arch.sie_block->ipa == 0xb236)
 331                         return handle_tpi(vcpu);
 332                 if (vcpu->arch.sie_block->ipa == 0xb235)
 333                         return handle_tsch(vcpu);
 334                 /* Handle in userspace. */
 335                 return -EOPNOTSUPP;
 336         } else {
 337                 /*
 338                  * Set condition code 3 to stop the guest from issuing channel
 339                  * I/O instructions.
 340                  */
 341                 kvm_s390_set_psw_cc(vcpu, 3);
 342                 return 0;
 343         }
 344 }
 345
 346 static int handle_stfl(struct kvm_vcpu *vcpu)
 347 {
 348         int rc;
 349         unsigned int fac;
 350
 351         vcpu->stat.instruction_stfl++;
 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         /*
 357          * We need to shift the lower 32 facility bits (bit 0-31) from a u64
 358          * into a u32 memory representation. They will remain bits 0-31.
 359          */
 360         fac = *vcpu->kvm->arch.model.fac->list >> 32;
 361         rc = write_guest_lc(vcpu, offsetof(struct _lowcore, stfl_fac_list),
 362                             &fac, sizeof(fac));
 363         if (rc)
 364                 return rc;
 365         VCPU_EVENT(vcpu, 5, "store facility list value %x", fac);
 366         trace_kvm_s390_handle_stfl(vcpu, fac);
 367         return 0;
 368 }
 369
 370 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
 371 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
 372 #define PSW_ADDR_24 0x0000000000ffffffUL
 373 #define PSW_ADDR_31 0x000000007fffffffUL
 374
 375 int is_valid_psw(psw_t *psw)
 376 {
 377         if (psw->mask & PSW_MASK_UNASSIGNED)
 378                 return 0;
 379         if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
 380                 if (psw->addr & ~PSW_ADDR_31)
 381                         return 0;
 382         }
 383         if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
 384                 return 0;
 385         if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
 386                 return 0;
 387         if (psw->addr & 1)
 388                 return 0;
 389         return 1;
 390 }
 391
 392 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
 393 {
 394         psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
 395         psw_compat_t new_psw;
 396         u64 addr;
 397         int rc;
 398
 399         if (gpsw->mask & PSW_MASK_PSTATE)
 400                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 401
 402         addr = kvm_s390_get_base_disp_s(vcpu);
 403         if (addr & 7)
 404                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 405
 406         rc = read_guest(vcpu, addr, &new_psw, sizeof(new_psw));
 407         if (rc)
 408                 return kvm_s390_inject_prog_cond(vcpu, rc);
 409         if (!(new_psw.mask & PSW32_MASK_BASE))
 410                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 411         gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
 412         gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
 413         gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
 414         if (!is_valid_psw(gpsw))
 415                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 416         return 0;
 417 }
 418
 419 static int handle_lpswe(struct kvm_vcpu *vcpu)
 420 {
 421         psw_t new_psw;
 422         u64 addr;
 423         int rc;
 424
 425         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 426                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 427
 428         addr = kvm_s390_get_base_disp_s(vcpu);
 429         if (addr & 7)
 430                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 431         rc = read_guest(vcpu, addr, &new_psw, sizeof(new_psw));
 432         if (rc)
 433                 return kvm_s390_inject_prog_cond(vcpu, rc);
 434         vcpu->arch.sie_block->gpsw = new_psw;
 435         if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
 436                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 437         return 0;
 438 }
 439
 440 static int handle_stidp(struct kvm_vcpu *vcpu)
 441 {
 442         u64 stidp_data = vcpu->arch.stidp_data;
 443         u64 operand2;
 444         int rc;
 445
 446         vcpu->stat.instruction_stidp++;
 447
 448         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 449                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 450
 451         operand2 = kvm_s390_get_base_disp_s(vcpu);
 452
 453         if (operand2 & 7)
 454                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 455
 456         rc = write_guest(vcpu, operand2, &stidp_data, sizeof(stidp_data));
 457         if (rc)
 458                 return kvm_s390_inject_prog_cond(vcpu, rc);
 459
 460         VCPU_EVENT(vcpu, 5, "%s", "store cpu id");
 461         return 0;
 462 }
 463
 464 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
 465 {
 466         int cpus = 0;
 467         int n;
 468
 469         cpus = atomic_read(&vcpu->kvm->online_vcpus);
 470
 471         /* deal with other level 3 hypervisors */
 472         if (stsi(mem, 3, 2, 2))
 473                 mem->count = 0;
 474         if (mem->count < 8)
 475                 mem->count++;
 476         for (n = mem->count - 1; n > 0 ; n--)
 477                 memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
 478
 479         memset(&mem->vm[0], 0, sizeof(mem->vm[0]));
 480         mem->vm[0].cpus_total = cpus;
 481         mem->vm[0].cpus_configured = cpus;
 482         mem->vm[0].cpus_standby = 0;
 483         mem->vm[0].cpus_reserved = 0;
 484         mem->vm[0].caf = 1000;
 485         memcpy(mem->vm[0].name, "KVMguest", 8);
 486         ASCEBC(mem->vm[0].name, 8);
 487         memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
 488         ASCEBC(mem->vm[0].cpi, 16);
 489 }
 490
 491 static int handle_stsi(struct kvm_vcpu *vcpu)
 492 {
 493         int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
 494         int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
 495         int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
 496         unsigned long mem = 0;
 497         u64 operand2;
 498         int rc = 0;
 499
 500         vcpu->stat.instruction_stsi++;
 501         VCPU_EVENT(vcpu, 4, "stsi: fc: %x sel1: %x sel2: %x", fc, sel1, sel2);
 502
 503         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 504                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 505
 506         if (fc > 3) {
 507                 kvm_s390_set_psw_cc(vcpu, 3);
 508                 return 0;
 509         }
 510
 511         if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
 512             || vcpu->run->s.regs.gprs[1] & 0xffff0000)
 513                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 514
 515         if (fc == 0) {
 516                 vcpu->run->s.regs.gprs[0] = 3 << 28;
 517                 kvm_s390_set_psw_cc(vcpu, 0);
 518                 return 0;
 519         }
 520
 521         operand2 = kvm_s390_get_base_disp_s(vcpu);
 522
 523         if (operand2 & 0xfff)
 524                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 525
 526         switch (fc) {
 527         case 1: /* same handling for 1 and 2 */
 528         case 2:
 529                 mem = get_zeroed_page(GFP_KERNEL);
 530                 if (!mem)
 531                         goto out_no_data;
 532                 if (stsi((void *) mem, fc, sel1, sel2))
 533                         goto out_no_data;
 534                 break;
 535         case 3:
 536                 if (sel1 != 2 || sel2 != 2)
 537                         goto out_no_data;
 538                 mem = get_zeroed_page(GFP_KERNEL);
 539                 if (!mem)
 540                         goto out_no_data;
 541                 handle_stsi_3_2_2(vcpu, (void *) mem);
 542                 break;
 543         }
 544
 545         rc = write_guest(vcpu, operand2, (void *)mem, PAGE_SIZE);
 546         if (rc) {
 547                 rc = kvm_s390_inject_prog_cond(vcpu, rc);
 548                 goto out;
 549         }
 550         trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
 551         free_page(mem);
 552         kvm_s390_set_psw_cc(vcpu, 0);
 553         vcpu->run->s.regs.gprs[0] = 0;
 554         return 0;
 555 out_no_data:
 556         kvm_s390_set_psw_cc(vcpu, 3);
 557 out:
 558         free_page(mem);
 559         return rc;
 560 }
 561
 562 static const intercept_handler_t b2_handlers[256] = {
 563         [0x02] = handle_stidp,
 564         [0x04] = handle_set_clock,
 565         [0x10] = handle_set_prefix,
 566         [0x11] = handle_store_prefix,
 567         [0x12] = handle_store_cpu_address,
 568         [0x21] = handle_ipte_interlock,
 569         [0x29] = handle_skey,
 570         [0x2a] = handle_skey,
 571         [0x2b] = handle_skey,
 572         [0x2c] = handle_test_block,
 573         [0x30] = handle_io_inst,
 574         [0x31] = handle_io_inst,
 575         [0x32] = handle_io_inst,
 576         [0x33] = handle_io_inst,
 577         [0x34] = handle_io_inst,
 578         [0x35] = handle_io_inst,
 579         [0x36] = handle_io_inst,
 580         [0x37] = handle_io_inst,
 581         [0x38] = handle_io_inst,
 582         [0x39] = handle_io_inst,
 583         [0x3a] = handle_io_inst,
 584         [0x3b] = handle_io_inst,
 585         [0x3c] = handle_io_inst,
 586         [0x50] = handle_ipte_interlock,
 587         [0x5f] = handle_io_inst,
 588         [0x74] = handle_io_inst,
 589         [0x76] = handle_io_inst,
 590         [0x7d] = handle_stsi,
 591         [0xb1] = handle_stfl,
 592         [0xb2] = handle_lpswe,
 593 };
 594
 595 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
 596 {
 597         intercept_handler_t handler;
 598
 599         /*
 600          * A lot of B2 instructions are priviledged. Here we check for
 601          * the privileged ones, that we can handle in the kernel.
 602          * Anything else goes to userspace.
 603          */
 604         handler = b2_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
 605         if (handler)
 606                 return handler(vcpu);
 607
 608         return -EOPNOTSUPP;
 609 }
 610
 611 static int handle_epsw(struct kvm_vcpu *vcpu)
 612 {
 613         int reg1, reg2;
 614
 615         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 616
 617         /* This basically extracts the mask half of the psw. */
 618         vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
 619         vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
 620         if (reg2) {
 621                 vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
 622                 vcpu->run->s.regs.gprs[reg2] |=
 623                         vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
 624         }
 625         return 0;
 626 }
 627
 628 #define PFMF_RESERVED   0xfffc0101UL
 629 #define PFMF_SK         0x00020000UL
 630 #define PFMF_CF         0x00010000UL
 631 #define PFMF_UI         0x00008000UL
 632 #define PFMF_FSC        0x00007000UL
 633 #define PFMF_NQ         0x00000800UL
 634 #define PFMF_MR         0x00000400UL
 635 #define PFMF_MC         0x00000200UL
 636 #define PFMF_KEY        0x000000feUL
 637
 638 static int handle_pfmf(struct kvm_vcpu *vcpu)
 639 {
 640         int reg1, reg2;
 641         unsigned long start, end;
 642
 643         vcpu->stat.instruction_pfmf++;
 644
 645         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 646
 647         if (!MACHINE_HAS_PFMF)
 648                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
 649
 650         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 651                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 652
 653         if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
 654                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 655
 656         /* Only provide non-quiescing support if the host supports it */
 657         if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ && !test_facility(14))
 658                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 659
 660         /* No support for conditional-SSKE */
 661         if (vcpu->run->s.regs.gprs[reg1] & (PFMF_MR | PFMF_MC))
 662                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 663
 664         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 665         start = kvm_s390_logical_to_effective(vcpu, start);
 666
 667         switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
 668         case 0x00000000:
 669                 end = (start + (1UL << 12)) & ~((1UL << 12) - 1);
 670                 break;
 671         case 0x00001000:
 672                 end = (start + (1UL << 20)) & ~((1UL << 20) - 1);
 673                 break;
 674         /* We dont support EDAT2
 675         case 0x00002000:
 676                 end = (start + (1UL << 31)) & ~((1UL << 31) - 1);
 677                 break;*/
 678         default:
 679                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 680         }
 681
 682         if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
 683                 if (kvm_s390_check_low_addr_prot_real(vcpu, start))
 684                         return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
 685         }
 686
 687         while (start < end) {
 688                 unsigned long useraddr, abs_addr;
 689
 690                 /* Translate guest address to host address */
 691                 if ((vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) == 0)
 692                         abs_addr = kvm_s390_real_to_abs(vcpu, start);
 693                 else
 694                         abs_addr = start;
 695                 useraddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(abs_addr));
 696                 if (kvm_is_error_hva(useraddr))
 697                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 698
 699                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
 700                         if (clear_user((void __user *)useraddr, PAGE_SIZE))
 701                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 702                 }
 703
 704                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
 705                         int rc = __skey_check_enable(vcpu);
 706
 707                         if (rc)
 708                                 return rc;
 709                         if (set_guest_storage_key(current->mm, useraddr,
 710                                         vcpu->run->s.regs.gprs[reg1] & PFMF_KEY,
 711                                         vcpu->run->s.regs.gprs[reg1] & PFMF_NQ))
 712                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 713                 }
 714
 715                 start += PAGE_SIZE;
 716         }
 717         if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC)
 718                 vcpu->run->s.regs.gprs[reg2] = end;
 719         return 0;
 720 }
 721
 722 static int handle_essa(struct kvm_vcpu *vcpu)
 723 {
 724         /* entries expected to be 1FF */
 725         int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
 726         unsigned long *cbrlo, cbrle;
 727         struct gmap *gmap;
 728         int i;
 729
 730         VCPU_EVENT(vcpu, 5, "cmma release %d pages", entries);
 731         gmap = vcpu->arch.gmap;
 732         vcpu->stat.instruction_essa++;
 733         if (!kvm_s390_cmma_enabled(vcpu->kvm))
 734                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
 735
 736         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 737                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 738
 739         if (((vcpu->arch.sie_block->ipb & 0xf0000000) >> 28) > 6)
 740                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 741
 742         /* Rewind PSW to repeat the ESSA instruction */
 743         kvm_s390_rewind_psw(vcpu, 4);
 744         vcpu->arch.sie_block->cbrlo &= PAGE_MASK;       /* reset nceo */
 745         cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
 746         down_read(&gmap->mm->mmap_sem);
 747         for (i = 0; i < entries; ++i) {
 748                 cbrle = cbrlo[i];
 749                 if (unlikely(cbrle & ~PAGE_MASK || cbrle < 2 * PAGE_SIZE))
 750                         /* invalid entry */
 751                         break;
 752                 /* try to free backing */
 753                 __gmap_zap(gmap, cbrle);
 754         }
 755         up_read(&gmap->mm->mmap_sem);
 756         if (i < entries)
 757                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 758         return 0;
 759 }
 760
 761 static const intercept_handler_t b9_handlers[256] = {
 762         [0x8a] = handle_ipte_interlock,
 763         [0x8d] = handle_epsw,
 764         [0x8e] = handle_ipte_interlock,
 765         [0x8f] = handle_ipte_interlock,
 766         [0xab] = handle_essa,
 767         [0xaf] = handle_pfmf,
 768 };
 769
 770 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
 771 {
 772         intercept_handler_t handler;
 773
 774         /* This is handled just as for the B2 instructions. */
 775         handler = b9_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
 776         if (handler)
 777                 return handler(vcpu);
 778
 779         return -EOPNOTSUPP;
 780 }
 781
 782 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
 783 {
 784         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
 785         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
 786         int reg, rc, nr_regs;
 787         u32 ctl_array[16];
 788         u64 ga;
 789
 790         vcpu->stat.instruction_lctl++;
 791
 792         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 793                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 794
 795         ga = kvm_s390_get_base_disp_rs(vcpu);
 796
 797         if (ga & 3)
 798                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 799
 800         VCPU_EVENT(vcpu, 5, "lctl r1:%x, r3:%x, addr:%llx", reg1, reg3, ga);
 801         trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
 802
 803         nr_regs = ((reg3 - reg1) & 0xf) + 1;
 804         rc = read_guest(vcpu, ga, ctl_array, nr_regs * sizeof(u32));
 805         if (rc)
 806                 return kvm_s390_inject_prog_cond(vcpu, rc);
 807         reg = reg1;
 808         nr_regs = 0;
 809         do {
 810                 vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
 811                 vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
 812                 if (reg == reg3)
 813                         break;
 814                 reg = (reg + 1) % 16;
 815         } while (1);
 816         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 817         return 0;
 818 }
 819
 820 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
 821 {
 822         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
 823         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
 824         int reg, rc, nr_regs;
 825         u32 ctl_array[16];
 826         u64 ga;
 827
 828         vcpu->stat.instruction_stctl++;
 829
 830         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 831                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 832
 833         ga = kvm_s390_get_base_disp_rs(vcpu);
 834
 835         if (ga & 3)
 836                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 837
 838         VCPU_EVENT(vcpu, 5, "stctl r1:%x, r3:%x, addr:%llx", reg1, reg3, ga);
 839         trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
 840
 841         reg = reg1;
 842         nr_regs = 0;
 843         do {
 844                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
 845                 if (reg == reg3)
 846                         break;
 847                 reg = (reg + 1) % 16;
 848         } while (1);
 849         rc = write_guest(vcpu, ga, ctl_array, nr_regs * sizeof(u32));
 850         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
 851 }
 852
 853 static int handle_lctlg(struct kvm_vcpu *vcpu)
 854 {
 855         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
 856         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
 857         int reg, rc, nr_regs;
 858         u64 ctl_array[16];
 859         u64 ga;
 860
 861         vcpu->stat.instruction_lctlg++;
 862
 863         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 864                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 865
 866         ga = kvm_s390_get_base_disp_rsy(vcpu);
 867
 868         if (ga & 7)
 869                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 870
 871         VCPU_EVENT(vcpu, 5, "lctlg r1:%x, r3:%x, addr:%llx", reg1, reg3, ga);
 872         trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
 873
 874         nr_regs = ((reg3 - reg1) & 0xf) + 1;
 875         rc = read_guest(vcpu, ga, ctl_array, nr_regs * sizeof(u64));
 876         if (rc)
 877                 return kvm_s390_inject_prog_cond(vcpu, rc);
 878         reg = reg1;
 879         nr_regs = 0;
 880         do {
 881                 vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
 882                 if (reg == reg3)
 883                         break;
 884                 reg = (reg + 1) % 16;
 885         } while (1);
 886         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 887         return 0;
 888 }
 889
 890 static int handle_stctg(struct kvm_vcpu *vcpu)
 891 {
 892         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
 893         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
 894         int reg, rc, nr_regs;
 895         u64 ctl_array[16];
 896         u64 ga;
 897
 898         vcpu->stat.instruction_stctg++;
 899
 900         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 901                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 902
 903         ga = kvm_s390_get_base_disp_rsy(vcpu);
 904
 905         if (ga & 7)
 906                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 907
 908         VCPU_EVENT(vcpu, 5, "stctg r1:%x, r3:%x, addr:%llx", reg1, reg3, ga);
 909         trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
 910
 911         reg = reg1;
 912         nr_regs = 0;
 913         do {
 914                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
 915                 if (reg == reg3)
 916                         break;
 917                 reg = (reg + 1) % 16;
 918         } while (1);
 919         rc = write_guest(vcpu, ga, ctl_array, nr_regs * sizeof(u64));
 920         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
 921 }
 922
 923 static const intercept_handler_t eb_handlers[256] = {
 924         [0x2f] = handle_lctlg,
 925         [0x25] = handle_stctg,
 926 };
 927
 928 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
 929 {
 930         intercept_handler_t handler;
 931
 932         handler = eb_handlers[vcpu->arch.sie_block->ipb & 0xff];
 933         if (handler)
 934                 return handler(vcpu);
 935         return -EOPNOTSUPP;
 936 }
 937
 938 static int handle_tprot(struct kvm_vcpu *vcpu)
 939 {
 940         u64 address1, address2;
 941         unsigned long hva, gpa;
 942         int ret = 0, cc = 0;
 943         bool writable;
 944
 945         vcpu->stat.instruction_tprot++;
 946
 947         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 948                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 949
 950         kvm_s390_get_base_disp_sse(vcpu, &address1, &address2);
 951
 952         /* we only handle the Linux memory detection case:
 953          * access key == 0
 954          * everything else goes to userspace. */
 955         if (address2 & 0xf0)
 956                 return -EOPNOTSUPP;
 957         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
 958                 ipte_lock(vcpu);
 959         ret = guest_translate_address(vcpu, address1, &gpa, 1);
 960         if (ret == PGM_PROTECTION) {
 961                 /* Write protected? Try again with read-only... */
 962                 cc = 1;
 963                 ret = guest_translate_address(vcpu, address1, &gpa, 0);
 964         }
 965         if (ret) {
 966                 if (ret == PGM_ADDRESSING || ret == PGM_TRANSLATION_SPEC) {
 967                         ret = kvm_s390_inject_program_int(vcpu, ret);
 968                 } else if (ret > 0) {
 969                         /* Translation not available */
 970                         kvm_s390_set_psw_cc(vcpu, 3);
 971                         ret = 0;
 972                 }
 973                 goto out_unlock;
 974         }
 975
 976         hva = gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
 977         if (kvm_is_error_hva(hva)) {
 978                 ret = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 979         } else {
 980                 if (!writable)
 981                         cc = 1;         /* Write not permitted ==> read-only */
 982                 kvm_s390_set_psw_cc(vcpu, cc);
 983                 /* Note: CC2 only occurs for storage keys (not supported yet) */
 984         }
 985 out_unlock:
 986         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
 987                 ipte_unlock(vcpu);
 988         return ret;
 989 }
 990
 991 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
 992 {
 993         /* For e5xx... instructions we only handle TPROT */
 994         if ((vcpu->arch.sie_block->ipa & 0x00ff) == 0x01)
 995                 return handle_tprot(vcpu);
 996         return -EOPNOTSUPP;
 997 }
 998
 999 static int handle_sckpf(struct kvm_vcpu *vcpu)
1000 {
1001         u32 value;
1002
1003         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1004                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1005
1006         if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
1007                 return kvm_s390_inject_program_int(vcpu,
1008                                                    PGM_SPECIFICATION);
1009
1010         value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
1011         vcpu->arch.sie_block->todpr = value;
1012
1013         return 0;
1014 }
1015
1016 static const intercept_handler_t x01_handlers[256] = {
1017         [0x07] = handle_sckpf,
1018 };
1019
1020 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1021 {
1022         intercept_handler_t handler;
1023
1024         handler = x01_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
1025         if (handler)
1026                 return handler(vcpu);
1027         return -EOPNOTSUPP;
1028 }