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 void __skey_check_enable(struct kvm_vcpu *vcpu)
 160 {
 161         if (!(vcpu->arch.sie_block->ictl & (ICTL_ISKE | ICTL_SSKE | ICTL_RRBE)))
 162                 return;
 163
 164         s390_enable_skey();
 165         trace_kvm_s390_skey_related_inst(vcpu);
 166         vcpu->arch.sie_block->ictl &= ~(ICTL_ISKE | ICTL_SSKE | ICTL_RRBE);
 167 }
 168
 169
 170 static int handle_skey(struct kvm_vcpu *vcpu)
 171 {
 172         __skey_check_enable(vcpu);
 173
 174         vcpu->stat.instruction_storage_key++;
 175
 176         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 177                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 178
 179         kvm_s390_rewind_psw(vcpu, 4);
 180         VCPU_EVENT(vcpu, 4, "%s", "retrying storage key operation");
 181         return 0;
 182 }
 183
 184 static int handle_ipte_interlock(struct kvm_vcpu *vcpu)
 185 {
 186         vcpu->stat.instruction_ipte_interlock++;
 187         if (psw_bits(vcpu->arch.sie_block->gpsw).p)
 188                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 189         wait_event(vcpu->kvm->arch.ipte_wq, !ipte_lock_held(vcpu));
 190         kvm_s390_rewind_psw(vcpu, 4);
 191         VCPU_EVENT(vcpu, 4, "%s", "retrying ipte interlock operation");
 192         return 0;
 193 }
 194
 195 static int handle_test_block(struct kvm_vcpu *vcpu)
 196 {
 197         gpa_t addr;
 198         int reg2;
 199
 200         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 201                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 202
 203         kvm_s390_get_regs_rre(vcpu, NULL, &reg2);
 204         addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 205         addr = kvm_s390_logical_to_effective(vcpu, addr);
 206         if (kvm_s390_check_low_addr_protection(vcpu, addr))
 207                 return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
 208         addr = kvm_s390_real_to_abs(vcpu, addr);
 209
 210         if (kvm_is_error_gpa(vcpu->kvm, addr))
 211                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 212         /*
 213          * We don't expect errors on modern systems, and do not care
 214          * about storage keys (yet), so let's just clear the page.
 215          */
 216         if (kvm_clear_guest(vcpu->kvm, addr, PAGE_SIZE))
 217                 return -EFAULT;
 218         kvm_s390_set_psw_cc(vcpu, 0);
 219         vcpu->run->s.regs.gprs[0] = 0;
 220         return 0;
 221 }
 222
 223 static int handle_tpi(struct kvm_vcpu *vcpu)
 224 {
 225         struct kvm_s390_interrupt_info *inti;
 226         unsigned long len;
 227         u32 tpi_data[3];
 228         int cc, rc;
 229         u64 addr;
 230
 231         rc = 0;
 232         addr = kvm_s390_get_base_disp_s(vcpu);
 233         if (addr & 3)
 234                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 235         cc = 0;
 236         inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0);
 237         if (!inti)
 238                 goto no_interrupt;
 239         cc = 1;
 240         tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr;
 241         tpi_data[1] = inti->io.io_int_parm;
 242         tpi_data[2] = inti->io.io_int_word;
 243         if (addr) {
 244                 /*
 245                  * Store the two-word I/O interruption code into the
 246                  * provided area.
 247                  */
 248                 len = sizeof(tpi_data) - 4;
 249                 rc = write_guest(vcpu, addr, &tpi_data, len);
 250                 if (rc)
 251                         return kvm_s390_inject_prog_cond(vcpu, rc);
 252         } else {
 253                 /*
 254                  * Store the three-word I/O interruption code into
 255                  * the appropriate lowcore area.
 256                  */
 257                 len = sizeof(tpi_data);
 258                 if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len))
 259                         rc = -EFAULT;
 260         }
 261         /*
 262          * If we encounter a problem storing the interruption code, the
 263          * instruction is suppressed from the guest's view: reinject the
 264          * interrupt.
 265          */
 266         if (!rc)
 267                 kfree(inti);
 268         else
 269                 kvm_s390_reinject_io_int(vcpu->kvm, inti);
 270 no_interrupt:
 271         /* Set condition code and we're done. */
 272         if (!rc)
 273                 kvm_s390_set_psw_cc(vcpu, cc);
 274         return rc ? -EFAULT : 0;
 275 }
 276
 277 static int handle_tsch(struct kvm_vcpu *vcpu)
 278 {
 279         struct kvm_s390_interrupt_info *inti;
 280
 281         inti = kvm_s390_get_io_int(vcpu->kvm, 0,
 282                                    vcpu->run->s.regs.gprs[1]);
 283
 284         /*
 285          * Prepare exit to userspace.
 286          * We indicate whether we dequeued a pending I/O interrupt
 287          * so that userspace can re-inject it if the instruction gets
 288          * a program check. While this may re-order the pending I/O
 289          * interrupts, this is no problem since the priority is kept
 290          * intact.
 291          */
 292         vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
 293         vcpu->run->s390_tsch.dequeued = !!inti;
 294         if (inti) {
 295                 vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
 296                 vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
 297                 vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
 298                 vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
 299         }
 300         vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
 301         kfree(inti);
 302         return -EREMOTE;
 303 }
 304
 305 static int handle_io_inst(struct kvm_vcpu *vcpu)
 306 {
 307         VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
 308
 309         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 310                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 311
 312         if (vcpu->kvm->arch.css_support) {
 313                 /*
 314                  * Most I/O instructions will be handled by userspace.
 315                  * Exceptions are tpi and the interrupt portion of tsch.
 316                  */
 317                 if (vcpu->arch.sie_block->ipa == 0xb236)
 318                         return handle_tpi(vcpu);
 319                 if (vcpu->arch.sie_block->ipa == 0xb235)
 320                         return handle_tsch(vcpu);
 321                 /* Handle in userspace. */
 322                 return -EOPNOTSUPP;
 323         } else {
 324                 /*
 325                  * Set condition code 3 to stop the guest from issuing channel
 326                  * I/O instructions.
 327                  */
 328                 kvm_s390_set_psw_cc(vcpu, 3);
 329                 return 0;
 330         }
 331 }
 332
 333 static int handle_stfl(struct kvm_vcpu *vcpu)
 334 {
 335         int rc;
 336
 337         vcpu->stat.instruction_stfl++;
 338
 339         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 340                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 341
 342         rc = write_guest_lc(vcpu, offsetof(struct _lowcore, stfl_fac_list),
 343                             vfacilities, 4);
 344         if (rc)
 345                 return rc;
 346         VCPU_EVENT(vcpu, 5, "store facility list value %x",
 347                    *(unsigned int *) vfacilities);
 348         trace_kvm_s390_handle_stfl(vcpu, *(unsigned int *) vfacilities);
 349         return 0;
 350 }
 351
 352 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
 353 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
 354 #define PSW_ADDR_24 0x0000000000ffffffUL
 355 #define PSW_ADDR_31 0x000000007fffffffUL
 356
 357 int is_valid_psw(psw_t *psw)
 358 {
 359         if (psw->mask & PSW_MASK_UNASSIGNED)
 360                 return 0;
 361         if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
 362                 if (psw->addr & ~PSW_ADDR_31)
 363                         return 0;
 364         }
 365         if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
 366                 return 0;
 367         if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
 368                 return 0;
 369         if (psw->addr & 1)
 370                 return 0;
 371         return 1;
 372 }
 373
 374 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
 375 {
 376         psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
 377         psw_compat_t new_psw;
 378         u64 addr;
 379         int rc;
 380
 381         if (gpsw->mask & PSW_MASK_PSTATE)
 382                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 383
 384         addr = kvm_s390_get_base_disp_s(vcpu);
 385         if (addr & 7)
 386                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 387
 388         rc = read_guest(vcpu, addr, &new_psw, sizeof(new_psw));
 389         if (rc)
 390                 return kvm_s390_inject_prog_cond(vcpu, rc);
 391         if (!(new_psw.mask & PSW32_MASK_BASE))
 392                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 393         gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
 394         gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
 395         gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
 396         if (!is_valid_psw(gpsw))
 397                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 398         return 0;
 399 }
 400
 401 static int handle_lpswe(struct kvm_vcpu *vcpu)
 402 {
 403         psw_t new_psw;
 404         u64 addr;
 405         int rc;
 406
 407         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 408                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 409
 410         addr = kvm_s390_get_base_disp_s(vcpu);
 411         if (addr & 7)
 412                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 413         rc = read_guest(vcpu, addr, &new_psw, sizeof(new_psw));
 414         if (rc)
 415                 return kvm_s390_inject_prog_cond(vcpu, rc);
 416         vcpu->arch.sie_block->gpsw = new_psw;
 417         if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
 418                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 419         return 0;
 420 }
 421
 422 static int handle_stidp(struct kvm_vcpu *vcpu)
 423 {
 424         u64 stidp_data = vcpu->arch.stidp_data;
 425         u64 operand2;
 426         int rc;
 427
 428         vcpu->stat.instruction_stidp++;
 429
 430         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 431                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 432
 433         operand2 = kvm_s390_get_base_disp_s(vcpu);
 434
 435         if (operand2 & 7)
 436                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 437
 438         rc = write_guest(vcpu, operand2, &stidp_data, sizeof(stidp_data));
 439         if (rc)
 440                 return kvm_s390_inject_prog_cond(vcpu, rc);
 441
 442         VCPU_EVENT(vcpu, 5, "%s", "store cpu id");
 443         return 0;
 444 }
 445
 446 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
 447 {
 448         int cpus = 0;
 449         int n;
 450
 451         cpus = atomic_read(&vcpu->kvm->online_vcpus);
 452
 453         /* deal with other level 3 hypervisors */
 454         if (stsi(mem, 3, 2, 2))
 455                 mem->count = 0;
 456         if (mem->count < 8)
 457                 mem->count++;
 458         for (n = mem->count - 1; n > 0 ; n--)
 459                 memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
 460
 461         mem->vm[0].cpus_total = cpus;
 462         mem->vm[0].cpus_configured = cpus;
 463         mem->vm[0].cpus_standby = 0;
 464         mem->vm[0].cpus_reserved = 0;
 465         mem->vm[0].caf = 1000;
 466         memcpy(mem->vm[0].name, "KVMguest", 8);
 467         ASCEBC(mem->vm[0].name, 8);
 468         memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
 469         ASCEBC(mem->vm[0].cpi, 16);
 470 }
 471
 472 static int handle_stsi(struct kvm_vcpu *vcpu)
 473 {
 474         int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
 475         int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
 476         int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
 477         unsigned long mem = 0;
 478         u64 operand2;
 479         int rc = 0;
 480
 481         vcpu->stat.instruction_stsi++;
 482         VCPU_EVENT(vcpu, 4, "stsi: fc: %x sel1: %x sel2: %x", fc, sel1, sel2);
 483
 484         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 485                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 486
 487         if (fc > 3) {
 488                 kvm_s390_set_psw_cc(vcpu, 3);
 489                 return 0;
 490         }
 491
 492         if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
 493             || vcpu->run->s.regs.gprs[1] & 0xffff0000)
 494                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 495
 496         if (fc == 0) {
 497                 vcpu->run->s.regs.gprs[0] = 3 << 28;
 498                 kvm_s390_set_psw_cc(vcpu, 0);
 499                 return 0;
 500         }
 501
 502         operand2 = kvm_s390_get_base_disp_s(vcpu);
 503
 504         if (operand2 & 0xfff)
 505                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 506
 507         switch (fc) {
 508         case 1: /* same handling for 1 and 2 */
 509         case 2:
 510                 mem = get_zeroed_page(GFP_KERNEL);
 511                 if (!mem)
 512                         goto out_no_data;
 513                 if (stsi((void *) mem, fc, sel1, sel2))
 514                         goto out_no_data;
 515                 break;
 516         case 3:
 517                 if (sel1 != 2 || sel2 != 2)
 518                         goto out_no_data;
 519                 mem = get_zeroed_page(GFP_KERNEL);
 520                 if (!mem)
 521                         goto out_no_data;
 522                 handle_stsi_3_2_2(vcpu, (void *) mem);
 523                 break;
 524         }
 525
 526         rc = write_guest(vcpu, operand2, (void *)mem, PAGE_SIZE);
 527         if (rc) {
 528                 rc = kvm_s390_inject_prog_cond(vcpu, rc);
 529                 goto out;
 530         }
 531         trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
 532         free_page(mem);
 533         kvm_s390_set_psw_cc(vcpu, 0);
 534         vcpu->run->s.regs.gprs[0] = 0;
 535         return 0;
 536 out_no_data:
 537         kvm_s390_set_psw_cc(vcpu, 3);
 538 out:
 539         free_page(mem);
 540         return rc;
 541 }
 542
 543 static const intercept_handler_t b2_handlers[256] = {
 544         [0x02] = handle_stidp,
 545         [0x04] = handle_set_clock,
 546         [0x10] = handle_set_prefix,
 547         [0x11] = handle_store_prefix,
 548         [0x12] = handle_store_cpu_address,
 549         [0x21] = handle_ipte_interlock,
 550         [0x29] = handle_skey,
 551         [0x2a] = handle_skey,
 552         [0x2b] = handle_skey,
 553         [0x2c] = handle_test_block,
 554         [0x30] = handle_io_inst,
 555         [0x31] = handle_io_inst,
 556         [0x32] = handle_io_inst,
 557         [0x33] = handle_io_inst,
 558         [0x34] = handle_io_inst,
 559         [0x35] = handle_io_inst,
 560         [0x36] = handle_io_inst,
 561         [0x37] = handle_io_inst,
 562         [0x38] = handle_io_inst,
 563         [0x39] = handle_io_inst,
 564         [0x3a] = handle_io_inst,
 565         [0x3b] = handle_io_inst,
 566         [0x3c] = handle_io_inst,
 567         [0x50] = handle_ipte_interlock,
 568         [0x5f] = handle_io_inst,
 569         [0x74] = handle_io_inst,
 570         [0x76] = handle_io_inst,
 571         [0x7d] = handle_stsi,
 572         [0xb1] = handle_stfl,
 573         [0xb2] = handle_lpswe,
 574 };
 575
 576 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
 577 {
 578         intercept_handler_t handler;
 579
 580         /*
 581          * A lot of B2 instructions are priviledged. Here we check for
 582          * the privileged ones, that we can handle in the kernel.
 583          * Anything else goes to userspace.
 584          */
 585         handler = b2_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
 586         if (handler)
 587                 return handler(vcpu);
 588
 589         return -EOPNOTSUPP;
 590 }
 591
 592 static int handle_epsw(struct kvm_vcpu *vcpu)
 593 {
 594         int reg1, reg2;
 595
 596         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 597
 598         /* This basically extracts the mask half of the psw. */
 599         vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
 600         vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
 601         if (reg2) {
 602                 vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
 603                 vcpu->run->s.regs.gprs[reg2] |=
 604                         vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
 605         }
 606         return 0;
 607 }
 608
 609 #define PFMF_RESERVED   0xfffc0101UL
 610 #define PFMF_SK         0x00020000UL
 611 #define PFMF_CF         0x00010000UL
 612 #define PFMF_UI         0x00008000UL
 613 #define PFMF_FSC        0x00007000UL
 614 #define PFMF_NQ         0x00000800UL
 615 #define PFMF_MR         0x00000400UL
 616 #define PFMF_MC         0x00000200UL
 617 #define PFMF_KEY        0x000000feUL
 618
 619 static int handle_pfmf(struct kvm_vcpu *vcpu)
 620 {
 621         int reg1, reg2;
 622         unsigned long start, end;
 623
 624         vcpu->stat.instruction_pfmf++;
 625
 626         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 627
 628         if (!MACHINE_HAS_PFMF)
 629                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
 630
 631         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 632                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 633
 634         if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
 635                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 636
 637         /* Only provide non-quiescing support if the host supports it */
 638         if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ && !test_facility(14))
 639                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 640
 641         /* No support for conditional-SSKE */
 642         if (vcpu->run->s.regs.gprs[reg1] & (PFMF_MR | PFMF_MC))
 643                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 644
 645         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 646         start = kvm_s390_logical_to_effective(vcpu, start);
 647
 648         switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
 649         case 0x00000000:
 650                 end = (start + (1UL << 12)) & ~((1UL << 12) - 1);
 651                 break;
 652         case 0x00001000:
 653                 end = (start + (1UL << 20)) & ~((1UL << 20) - 1);
 654                 break;
 655         /* We dont support EDAT2
 656         case 0x00002000:
 657                 end = (start + (1UL << 31)) & ~((1UL << 31) - 1);
 658                 break;*/
 659         default:
 660                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 661         }
 662
 663         if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
 664                 if (kvm_s390_check_low_addr_protection(vcpu, start))
 665                         return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
 666         }
 667
 668         while (start < end) {
 669                 unsigned long useraddr, abs_addr;
 670
 671                 /* Translate guest address to host address */
 672                 if ((vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) == 0)
 673                         abs_addr = kvm_s390_real_to_abs(vcpu, start);
 674                 else
 675                         abs_addr = start;
 676                 useraddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(abs_addr));
 677                 if (kvm_is_error_hva(useraddr))
 678                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 679
 680                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
 681                         if (clear_user((void __user *)useraddr, PAGE_SIZE))
 682                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 683                 }
 684
 685                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
 686                         __skey_check_enable(vcpu);
 687                         if (set_guest_storage_key(current->mm, useraddr,
 688                                         vcpu->run->s.regs.gprs[reg1] & PFMF_KEY,
 689                                         vcpu->run->s.regs.gprs[reg1] & PFMF_NQ))
 690                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 691                 }
 692
 693                 start += PAGE_SIZE;
 694         }
 695         if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC)
 696                 vcpu->run->s.regs.gprs[reg2] = end;
 697         return 0;
 698 }
 699
 700 static int handle_essa(struct kvm_vcpu *vcpu)
 701 {
 702         /* entries expected to be 1FF */
 703         int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
 704         unsigned long *cbrlo, cbrle;
 705         struct gmap *gmap;
 706         int i;
 707
 708         VCPU_EVENT(vcpu, 5, "cmma release %d pages", entries);
 709         gmap = vcpu->arch.gmap;
 710         vcpu->stat.instruction_essa++;
 711         if (!kvm_s390_cmma_enabled(vcpu->kvm))
 712                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
 713
 714         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 715                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 716
 717         if (((vcpu->arch.sie_block->ipb & 0xf0000000) >> 28) > 6)
 718                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 719
 720         /* Rewind PSW to repeat the ESSA instruction */
 721         kvm_s390_rewind_psw(vcpu, 4);
 722         vcpu->arch.sie_block->cbrlo &= PAGE_MASK;       /* reset nceo */
 723         cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
 724         down_read(&gmap->mm->mmap_sem);
 725         for (i = 0; i < entries; ++i) {
 726                 cbrle = cbrlo[i];
 727                 if (unlikely(cbrle & ~PAGE_MASK || cbrle < 2 * PAGE_SIZE))
 728                         /* invalid entry */
 729                         break;
 730                 /* try to free backing */
 731                 __gmap_zap(gmap, cbrle);
 732         }
 733         up_read(&gmap->mm->mmap_sem);
 734         if (i < entries)
 735                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 736         return 0;
 737 }
 738
 739 static const intercept_handler_t b9_handlers[256] = {
 740         [0x8a] = handle_ipte_interlock,
 741         [0x8d] = handle_epsw,
 742         [0x8e] = handle_ipte_interlock,
 743         [0x8f] = handle_ipte_interlock,
 744         [0xab] = handle_essa,
 745         [0xaf] = handle_pfmf,
 746 };
 747
 748 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
 749 {
 750         intercept_handler_t handler;
 751
 752         /* This is handled just as for the B2 instructions. */
 753         handler = b9_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
 754         if (handler)
 755                 return handler(vcpu);
 756
 757         return -EOPNOTSUPP;
 758 }
 759
 760 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
 761 {
 762         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
 763         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
 764         int reg, rc, nr_regs;
 765         u32 ctl_array[16];
 766         u64 ga;
 767
 768         vcpu->stat.instruction_lctl++;
 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         ga = kvm_s390_get_base_disp_rs(vcpu);
 774
 775         if (ga & 3)
 776                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 777
 778         VCPU_EVENT(vcpu, 5, "lctl r1:%x, r3:%x, addr:%llx", reg1, reg3, ga);
 779         trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
 780
 781         nr_regs = ((reg3 - reg1) & 0xf) + 1;
 782         rc = read_guest(vcpu, ga, ctl_array, nr_regs * sizeof(u32));
 783         if (rc)
 784                 return kvm_s390_inject_prog_cond(vcpu, rc);
 785         reg = reg1;
 786         nr_regs = 0;
 787         do {
 788                 vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
 789                 vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
 790                 if (reg == reg3)
 791                         break;
 792                 reg = (reg + 1) % 16;
 793         } while (1);
 794         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 795         return 0;
 796 }
 797
 798 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
 799 {
 800         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
 801         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
 802         int reg, rc, nr_regs;
 803         u32 ctl_array[16];
 804         u64 ga;
 805
 806         vcpu->stat.instruction_stctl++;
 807
 808         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 809                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 810
 811         ga = kvm_s390_get_base_disp_rs(vcpu);
 812
 813         if (ga & 3)
 814                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 815
 816         VCPU_EVENT(vcpu, 5, "stctl r1:%x, r3:%x, addr:%llx", reg1, reg3, ga);
 817         trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
 818
 819         reg = reg1;
 820         nr_regs = 0;
 821         do {
 822                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
 823                 if (reg == reg3)
 824                         break;
 825                 reg = (reg + 1) % 16;
 826         } while (1);
 827         rc = write_guest(vcpu, ga, ctl_array, nr_regs * sizeof(u32));
 828         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
 829 }
 830
 831 static int handle_lctlg(struct kvm_vcpu *vcpu)
 832 {
 833         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
 834         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
 835         int reg, rc, nr_regs;
 836         u64 ctl_array[16];
 837         u64 ga;
 838
 839         vcpu->stat.instruction_lctlg++;
 840
 841         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 842                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 843
 844         ga = kvm_s390_get_base_disp_rsy(vcpu);
 845
 846         if (ga & 7)
 847                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 848
 849         VCPU_EVENT(vcpu, 5, "lctlg r1:%x, r3:%x, addr:%llx", reg1, reg3, ga);
 850         trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
 851
 852         nr_regs = ((reg3 - reg1) & 0xf) + 1;
 853         rc = read_guest(vcpu, ga, ctl_array, nr_regs * sizeof(u64));
 854         if (rc)
 855                 return kvm_s390_inject_prog_cond(vcpu, rc);
 856         reg = reg1;
 857         nr_regs = 0;
 858         do {
 859                 vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
 860                 if (reg == reg3)
 861                         break;
 862                 reg = (reg + 1) % 16;
 863         } while (1);
 864         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 865         return 0;
 866 }
 867
 868 static int handle_stctg(struct kvm_vcpu *vcpu)
 869 {
 870         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
 871         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
 872         int reg, rc, nr_regs;
 873         u64 ctl_array[16];
 874         u64 ga;
 875
 876         vcpu->stat.instruction_stctg++;
 877
 878         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 879                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 880
 881         ga = kvm_s390_get_base_disp_rsy(vcpu);
 882
 883         if (ga & 7)
 884                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 885
 886         VCPU_EVENT(vcpu, 5, "stctg r1:%x, r3:%x, addr:%llx", reg1, reg3, ga);
 887         trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
 888
 889         reg = reg1;
 890         nr_regs = 0;
 891         do {
 892                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
 893                 if (reg == reg3)
 894                         break;
 895                 reg = (reg + 1) % 16;
 896         } while (1);
 897         rc = write_guest(vcpu, ga, ctl_array, nr_regs * sizeof(u64));
 898         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
 899 }
 900
 901 static const intercept_handler_t eb_handlers[256] = {
 902         [0x2f] = handle_lctlg,
 903         [0x25] = handle_stctg,
 904 };
 905
 906 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
 907 {
 908         intercept_handler_t handler;
 909
 910         handler = eb_handlers[vcpu->arch.sie_block->ipb & 0xff];
 911         if (handler)
 912                 return handler(vcpu);
 913         return -EOPNOTSUPP;
 914 }
 915
 916 static int handle_tprot(struct kvm_vcpu *vcpu)
 917 {
 918         u64 address1, address2;
 919         unsigned long hva, gpa;
 920         int ret = 0, cc = 0;
 921         bool writable;
 922
 923         vcpu->stat.instruction_tprot++;
 924
 925         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 926                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 927
 928         kvm_s390_get_base_disp_sse(vcpu, &address1, &address2);
 929
 930         /* we only handle the Linux memory detection case:
 931          * access key == 0
 932          * everything else goes to userspace. */
 933         if (address2 & 0xf0)
 934                 return -EOPNOTSUPP;
 935         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
 936                 ipte_lock(vcpu);
 937         ret = guest_translate_address(vcpu, address1, &gpa, 1);
 938         if (ret == PGM_PROTECTION) {
 939                 /* Write protected? Try again with read-only... */
 940                 cc = 1;
 941                 ret = guest_translate_address(vcpu, address1, &gpa, 0);
 942         }
 943         if (ret) {
 944                 if (ret == PGM_ADDRESSING || ret == PGM_TRANSLATION_SPEC) {
 945                         ret = kvm_s390_inject_program_int(vcpu, ret);
 946                 } else if (ret > 0) {
 947                         /* Translation not available */
 948                         kvm_s390_set_psw_cc(vcpu, 3);
 949                         ret = 0;
 950                 }
 951                 goto out_unlock;
 952         }
 953
 954         hva = gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
 955         if (kvm_is_error_hva(hva)) {
 956                 ret = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 957         } else {
 958                 if (!writable)
 959                         cc = 1;         /* Write not permitted ==> read-only */
 960                 kvm_s390_set_psw_cc(vcpu, cc);
 961                 /* Note: CC2 only occurs for storage keys (not supported yet) */
 962         }
 963 out_unlock:
 964         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
 965                 ipte_unlock(vcpu);
 966         return ret;
 967 }
 968
 969 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
 970 {
 971         /* For e5xx... instructions we only handle TPROT */
 972         if ((vcpu->arch.sie_block->ipa & 0x00ff) == 0x01)
 973                 return handle_tprot(vcpu);
 974         return -EOPNOTSUPP;
 975 }
 976
 977 static int handle_sckpf(struct kvm_vcpu *vcpu)
 978 {
 979         u32 value;
 980
 981         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 982                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 983
 984         if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
 985                 return kvm_s390_inject_program_int(vcpu,
 986                                                    PGM_SPECIFICATION);
 987
 988         value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
 989         vcpu->arch.sie_block->todpr = value;
 990
 991         return 0;
 992 }
 993
 994 static const intercept_handler_t x01_handlers[256] = {
 995         [0x07] = handle_sckpf,
 996 };
 997
 998 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
 999 {
1000         intercept_handler_t handler;
1001
1002         handler = x01_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
1003         if (handler)
1004                 return handler(vcpu);
1005         return -EOPNOTSUPP;
1006 }