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_protection(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 cc, rc;
 233         u64 addr;
 234
 235         rc = 0;
 236         addr = kvm_s390_get_base_disp_s(vcpu);
 237         if (addr & 3)
 238                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 239         cc = 0;
 240         inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0);
 241         if (!inti)
 242                 goto no_interrupt;
 243         cc = 1;
 244         tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr;
 245         tpi_data[1] = inti->io.io_int_parm;
 246         tpi_data[2] = inti->io.io_int_word;
 247         if (addr) {
 248                 /*
 249                  * Store the two-word I/O interruption code into the
 250                  * provided area.
 251                  */
 252                 len = sizeof(tpi_data) - 4;
 253                 rc = write_guest(vcpu, addr, &tpi_data, len);
 254                 if (rc)
 255                         return kvm_s390_inject_prog_cond(vcpu, rc);
 256         } else {
 257                 /*
 258                  * Store the three-word I/O interruption code into
 259                  * the appropriate lowcore area.
 260                  */
 261                 len = sizeof(tpi_data);
 262                 if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len))
 263                         rc = -EFAULT;
 264         }
 265         /*
 266          * If we encounter a problem storing the interruption code, the
 267          * instruction is suppressed from the guest's view: reinject the
 268          * interrupt.
 269          */
 270         if (!rc)
 271                 kfree(inti);
 272         else
 273                 kvm_s390_reinject_io_int(vcpu->kvm, inti);
 274 no_interrupt:
 275         /* Set condition code and we're done. */
 276         if (!rc)
 277                 kvm_s390_set_psw_cc(vcpu, cc);
 278         return rc ? -EFAULT : 0;
 279 }
 280
 281 static int handle_tsch(struct kvm_vcpu *vcpu)
 282 {
 283         struct kvm_s390_interrupt_info *inti;
 284
 285         inti = kvm_s390_get_io_int(vcpu->kvm, 0,
 286                                    vcpu->run->s.regs.gprs[1]);
 287
 288         /*
 289          * Prepare exit to userspace.
 290          * We indicate whether we dequeued a pending I/O interrupt
 291          * so that userspace can re-inject it if the instruction gets
 292          * a program check. While this may re-order the pending I/O
 293          * interrupts, this is no problem since the priority is kept
 294          * intact.
 295          */
 296         vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
 297         vcpu->run->s390_tsch.dequeued = !!inti;
 298         if (inti) {
 299                 vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
 300                 vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
 301                 vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
 302                 vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
 303         }
 304         vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
 305         kfree(inti);
 306         return -EREMOTE;
 307 }
 308
 309 static int handle_io_inst(struct kvm_vcpu *vcpu)
 310 {
 311         VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
 312
 313         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 314                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 315
 316         if (vcpu->kvm->arch.css_support) {
 317                 /*
 318                  * Most I/O instructions will be handled by userspace.
 319                  * Exceptions are tpi and the interrupt portion of tsch.
 320                  */
 321                 if (vcpu->arch.sie_block->ipa == 0xb236)
 322                         return handle_tpi(vcpu);
 323                 if (vcpu->arch.sie_block->ipa == 0xb235)
 324                         return handle_tsch(vcpu);
 325                 /* Handle in userspace. */
 326                 return -EOPNOTSUPP;
 327         } else {
 328                 /*
 329                  * Set condition code 3 to stop the guest from issuing channel
 330                  * I/O instructions.
 331                  */
 332                 kvm_s390_set_psw_cc(vcpu, 3);
 333                 return 0;
 334         }
 335 }
 336
 337 static int handle_stfl(struct kvm_vcpu *vcpu)
 338 {
 339         int rc;
 340
 341         vcpu->stat.instruction_stfl++;
 342
 343         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 344                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 345
 346         rc = write_guest_lc(vcpu, offsetof(struct _lowcore, stfl_fac_list),
 347                             vfacilities, 4);
 348         if (rc)
 349                 return rc;
 350         VCPU_EVENT(vcpu, 5, "store facility list value %x",
 351                    *(unsigned int *) vfacilities);
 352         trace_kvm_s390_handle_stfl(vcpu, *(unsigned int *) vfacilities);
 353         return 0;
 354 }
 355
 356 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
 357 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
 358 #define PSW_ADDR_24 0x0000000000ffffffUL
 359 #define PSW_ADDR_31 0x000000007fffffffUL
 360
 361 int is_valid_psw(psw_t *psw)
 362 {
 363         if (psw->mask & PSW_MASK_UNASSIGNED)
 364                 return 0;
 365         if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
 366                 if (psw->addr & ~PSW_ADDR_31)
 367                         return 0;
 368         }
 369         if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
 370                 return 0;
 371         if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
 372                 return 0;
 373         if (psw->addr & 1)
 374                 return 0;
 375         return 1;
 376 }
 377
 378 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
 379 {
 380         psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
 381         psw_compat_t new_psw;
 382         u64 addr;
 383         int rc;
 384
 385         if (gpsw->mask & PSW_MASK_PSTATE)
 386                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 387
 388         addr = kvm_s390_get_base_disp_s(vcpu);
 389         if (addr & 7)
 390                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 391
 392         rc = read_guest(vcpu, addr, &new_psw, sizeof(new_psw));
 393         if (rc)
 394                 return kvm_s390_inject_prog_cond(vcpu, rc);
 395         if (!(new_psw.mask & PSW32_MASK_BASE))
 396                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 397         gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
 398         gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
 399         gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
 400         if (!is_valid_psw(gpsw))
 401                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 402         return 0;
 403 }
 404
 405 static int handle_lpswe(struct kvm_vcpu *vcpu)
 406 {
 407         psw_t new_psw;
 408         u64 addr;
 409         int rc;
 410
 411         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 412                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 413
 414         addr = kvm_s390_get_base_disp_s(vcpu);
 415         if (addr & 7)
 416                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 417         rc = read_guest(vcpu, addr, &new_psw, sizeof(new_psw));
 418         if (rc)
 419                 return kvm_s390_inject_prog_cond(vcpu, rc);
 420         vcpu->arch.sie_block->gpsw = new_psw;
 421         if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
 422                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 423         return 0;
 424 }
 425
 426 static int handle_stidp(struct kvm_vcpu *vcpu)
 427 {
 428         u64 stidp_data = vcpu->arch.stidp_data;
 429         u64 operand2;
 430         int rc;
 431
 432         vcpu->stat.instruction_stidp++;
 433
 434         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 435                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 436
 437         operand2 = kvm_s390_get_base_disp_s(vcpu);
 438
 439         if (operand2 & 7)
 440                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 441
 442         rc = write_guest(vcpu, operand2, &stidp_data, sizeof(stidp_data));
 443         if (rc)
 444                 return kvm_s390_inject_prog_cond(vcpu, rc);
 445
 446         VCPU_EVENT(vcpu, 5, "%s", "store cpu id");
 447         return 0;
 448 }
 449
 450 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
 451 {
 452         int cpus = 0;
 453         int n;
 454
 455         cpus = atomic_read(&vcpu->kvm->online_vcpus);
 456
 457         /* deal with other level 3 hypervisors */
 458         if (stsi(mem, 3, 2, 2))
 459                 mem->count = 0;
 460         if (mem->count < 8)
 461                 mem->count++;
 462         for (n = mem->count - 1; n > 0 ; n--)
 463                 memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
 464
 465         mem->vm[0].cpus_total = cpus;
 466         mem->vm[0].cpus_configured = cpus;
 467         mem->vm[0].cpus_standby = 0;
 468         mem->vm[0].cpus_reserved = 0;
 469         mem->vm[0].caf = 1000;
 470         memcpy(mem->vm[0].name, "KVMguest", 8);
 471         ASCEBC(mem->vm[0].name, 8);
 472         memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
 473         ASCEBC(mem->vm[0].cpi, 16);
 474 }
 475
 476 static int handle_stsi(struct kvm_vcpu *vcpu)
 477 {
 478         int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
 479         int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
 480         int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
 481         unsigned long mem = 0;
 482         u64 operand2;
 483         int rc = 0;
 484
 485         vcpu->stat.instruction_stsi++;
 486         VCPU_EVENT(vcpu, 4, "stsi: fc: %x sel1: %x sel2: %x", fc, sel1, sel2);
 487
 488         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 489                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 490
 491         if (fc > 3) {
 492                 kvm_s390_set_psw_cc(vcpu, 3);
 493                 return 0;
 494         }
 495
 496         if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
 497             || vcpu->run->s.regs.gprs[1] & 0xffff0000)
 498                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 499
 500         if (fc == 0) {
 501                 vcpu->run->s.regs.gprs[0] = 3 << 28;
 502                 kvm_s390_set_psw_cc(vcpu, 0);
 503                 return 0;
 504         }
 505
 506         operand2 = kvm_s390_get_base_disp_s(vcpu);
 507
 508         if (operand2 & 0xfff)
 509                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 510
 511         switch (fc) {
 512         case 1: /* same handling for 1 and 2 */
 513         case 2:
 514                 mem = get_zeroed_page(GFP_KERNEL);
 515                 if (!mem)
 516                         goto out_no_data;
 517                 if (stsi((void *) mem, fc, sel1, sel2))
 518                         goto out_no_data;
 519                 break;
 520         case 3:
 521                 if (sel1 != 2 || sel2 != 2)
 522                         goto out_no_data;
 523                 mem = get_zeroed_page(GFP_KERNEL);
 524                 if (!mem)
 525                         goto out_no_data;
 526                 handle_stsi_3_2_2(vcpu, (void *) mem);
 527                 break;
 528         }
 529
 530         rc = write_guest(vcpu, operand2, (void *)mem, PAGE_SIZE);
 531         if (rc) {
 532                 rc = kvm_s390_inject_prog_cond(vcpu, rc);
 533                 goto out;
 534         }
 535         trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
 536         free_page(mem);
 537         kvm_s390_set_psw_cc(vcpu, 0);
 538         vcpu->run->s.regs.gprs[0] = 0;
 539         return 0;
 540 out_no_data:
 541         kvm_s390_set_psw_cc(vcpu, 3);
 542 out:
 543         free_page(mem);
 544         return rc;
 545 }
 546
 547 static const intercept_handler_t b2_handlers[256] = {
 548         [0x02] = handle_stidp,
 549         [0x04] = handle_set_clock,
 550         [0x10] = handle_set_prefix,
 551         [0x11] = handle_store_prefix,
 552         [0x12] = handle_store_cpu_address,
 553         [0x21] = handle_ipte_interlock,
 554         [0x29] = handle_skey,
 555         [0x2a] = handle_skey,
 556         [0x2b] = handle_skey,
 557         [0x2c] = handle_test_block,
 558         [0x30] = handle_io_inst,
 559         [0x31] = handle_io_inst,
 560         [0x32] = handle_io_inst,
 561         [0x33] = handle_io_inst,
 562         [0x34] = handle_io_inst,
 563         [0x35] = handle_io_inst,
 564         [0x36] = handle_io_inst,
 565         [0x37] = handle_io_inst,
 566         [0x38] = handle_io_inst,
 567         [0x39] = handle_io_inst,
 568         [0x3a] = handle_io_inst,
 569         [0x3b] = handle_io_inst,
 570         [0x3c] = handle_io_inst,
 571         [0x50] = handle_ipte_interlock,
 572         [0x5f] = handle_io_inst,
 573         [0x74] = handle_io_inst,
 574         [0x76] = handle_io_inst,
 575         [0x7d] = handle_stsi,
 576         [0xb1] = handle_stfl,
 577         [0xb2] = handle_lpswe,
 578 };
 579
 580 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
 581 {
 582         intercept_handler_t handler;
 583
 584         /*
 585          * A lot of B2 instructions are priviledged. Here we check for
 586          * the privileged ones, that we can handle in the kernel.
 587          * Anything else goes to userspace.
 588          */
 589         handler = b2_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
 590         if (handler)
 591                 return handler(vcpu);
 592
 593         return -EOPNOTSUPP;
 594 }
 595
 596 static int handle_epsw(struct kvm_vcpu *vcpu)
 597 {
 598         int reg1, reg2;
 599
 600         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 601
 602         /* This basically extracts the mask half of the psw. */
 603         vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
 604         vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
 605         if (reg2) {
 606                 vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
 607                 vcpu->run->s.regs.gprs[reg2] |=
 608                         vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
 609         }
 610         return 0;
 611 }
 612
 613 #define PFMF_RESERVED   0xfffc0101UL
 614 #define PFMF_SK         0x00020000UL
 615 #define PFMF_CF         0x00010000UL
 616 #define PFMF_UI         0x00008000UL
 617 #define PFMF_FSC        0x00007000UL
 618 #define PFMF_NQ         0x00000800UL
 619 #define PFMF_MR         0x00000400UL
 620 #define PFMF_MC         0x00000200UL
 621 #define PFMF_KEY        0x000000feUL
 622
 623 static int handle_pfmf(struct kvm_vcpu *vcpu)
 624 {
 625         int reg1, reg2;
 626         unsigned long start, end;
 627
 628         vcpu->stat.instruction_pfmf++;
 629
 630         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 631
 632         if (!MACHINE_HAS_PFMF)
 633                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
 634
 635         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 636                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 637
 638         if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
 639                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 640
 641         /* Only provide non-quiescing support if the host supports it */
 642         if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ && !test_facility(14))
 643                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 644
 645         /* No support for conditional-SSKE */
 646         if (vcpu->run->s.regs.gprs[reg1] & (PFMF_MR | PFMF_MC))
 647                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 648
 649         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 650         start = kvm_s390_logical_to_effective(vcpu, start);
 651
 652         switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
 653         case 0x00000000:
 654                 end = (start + (1UL << 12)) & ~((1UL << 12) - 1);
 655                 break;
 656         case 0x00001000:
 657                 end = (start + (1UL << 20)) & ~((1UL << 20) - 1);
 658                 break;
 659         /* We dont support EDAT2
 660         case 0x00002000:
 661                 end = (start + (1UL << 31)) & ~((1UL << 31) - 1);
 662                 break;*/
 663         default:
 664                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 665         }
 666
 667         if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
 668                 if (kvm_s390_check_low_addr_protection(vcpu, start))
 669                         return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
 670         }
 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                         int rc = __skey_check_enable(vcpu);
 691
 692                         if (rc)
 693                                 return rc;
 694                         if (set_guest_storage_key(current->mm, useraddr,
 695                                         vcpu->run->s.regs.gprs[reg1] & PFMF_KEY,
 696                                         vcpu->run->s.regs.gprs[reg1] & PFMF_NQ))
 697                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 698                 }
 699
 700                 start += PAGE_SIZE;
 701         }
 702         if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC)
 703                 vcpu->run->s.regs.gprs[reg2] = end;
 704         return 0;
 705 }
 706
 707 static int handle_essa(struct kvm_vcpu *vcpu)
 708 {
 709         /* entries expected to be 1FF */
 710         int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
 711         unsigned long *cbrlo, cbrle;
 712         struct gmap *gmap;
 713         int i;
 714
 715         VCPU_EVENT(vcpu, 5, "cmma release %d pages", entries);
 716         gmap = vcpu->arch.gmap;
 717         vcpu->stat.instruction_essa++;
 718         if (!kvm_s390_cmma_enabled(vcpu->kvm))
 719                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
 720
 721         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 722                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 723
 724         if (((vcpu->arch.sie_block->ipb & 0xf0000000) >> 28) > 6)
 725                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 726
 727         /* Rewind PSW to repeat the ESSA instruction */
 728         kvm_s390_rewind_psw(vcpu, 4);
 729         vcpu->arch.sie_block->cbrlo &= PAGE_MASK;       /* reset nceo */
 730         cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
 731         down_read(&gmap->mm->mmap_sem);
 732         for (i = 0; i < entries; ++i) {
 733                 cbrle = cbrlo[i];
 734                 if (unlikely(cbrle & ~PAGE_MASK || cbrle < 2 * PAGE_SIZE))
 735                         /* invalid entry */
 736                         break;
 737                 /* try to free backing */
 738                 __gmap_zap(gmap, cbrle);
 739         }
 740         up_read(&gmap->mm->mmap_sem);
 741         if (i < entries)
 742                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 743         return 0;
 744 }
 745
 746 static const intercept_handler_t b9_handlers[256] = {
 747         [0x8a] = handle_ipte_interlock,
 748         [0x8d] = handle_epsw,
 749         [0x8e] = handle_ipte_interlock,
 750         [0x8f] = handle_ipte_interlock,
 751         [0xab] = handle_essa,
 752         [0xaf] = handle_pfmf,
 753 };
 754
 755 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
 756 {
 757         intercept_handler_t handler;
 758
 759         /* This is handled just as for the B2 instructions. */
 760         handler = b9_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
 761         if (handler)
 762                 return handler(vcpu);
 763
 764         return -EOPNOTSUPP;
 765 }
 766
 767 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
 768 {
 769         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
 770         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
 771         int reg, rc, nr_regs;
 772         u32 ctl_array[16];
 773         u64 ga;
 774
 775         vcpu->stat.instruction_lctl++;
 776
 777         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 778                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 779
 780         ga = kvm_s390_get_base_disp_rs(vcpu);
 781
 782         if (ga & 3)
 783                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 784
 785         VCPU_EVENT(vcpu, 5, "lctl r1:%x, r3:%x, addr:%llx", reg1, reg3, ga);
 786         trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
 787
 788         nr_regs = ((reg3 - reg1) & 0xf) + 1;
 789         rc = read_guest(vcpu, ga, ctl_array, nr_regs * sizeof(u32));
 790         if (rc)
 791                 return kvm_s390_inject_prog_cond(vcpu, rc);
 792         reg = reg1;
 793         nr_regs = 0;
 794         do {
 795                 vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
 796                 vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
 797                 if (reg == reg3)
 798                         break;
 799                 reg = (reg + 1) % 16;
 800         } while (1);
 801         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 802         return 0;
 803 }
 804
 805 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
 806 {
 807         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
 808         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
 809         int reg, rc, nr_regs;
 810         u32 ctl_array[16];
 811         u64 ga;
 812
 813         vcpu->stat.instruction_stctl++;
 814
 815         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 816                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 817
 818         ga = kvm_s390_get_base_disp_rs(vcpu);
 819
 820         if (ga & 3)
 821                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 822
 823         VCPU_EVENT(vcpu, 5, "stctl r1:%x, r3:%x, addr:%llx", reg1, reg3, ga);
 824         trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
 825
 826         reg = reg1;
 827         nr_regs = 0;
 828         do {
 829                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
 830                 if (reg == reg3)
 831                         break;
 832                 reg = (reg + 1) % 16;
 833         } while (1);
 834         rc = write_guest(vcpu, ga, ctl_array, nr_regs * sizeof(u32));
 835         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 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         int reg, rc, nr_regs;
 843         u64 ctl_array[16];
 844         u64 ga;
 845
 846         vcpu->stat.instruction_lctlg++;
 847
 848         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 849                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 850
 851         ga = kvm_s390_get_base_disp_rsy(vcpu);
 852
 853         if (ga & 7)
 854                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 855
 856         VCPU_EVENT(vcpu, 5, "lctlg r1:%x, r3:%x, addr:%llx", reg1, reg3, ga);
 857         trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
 858
 859         nr_regs = ((reg3 - reg1) & 0xf) + 1;
 860         rc = read_guest(vcpu, ga, ctl_array, nr_regs * sizeof(u64));
 861         if (rc)
 862                 return kvm_s390_inject_prog_cond(vcpu, rc);
 863         reg = reg1;
 864         nr_regs = 0;
 865         do {
 866                 vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
 867                 if (reg == reg3)
 868                         break;
 869                 reg = (reg + 1) % 16;
 870         } while (1);
 871         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 872         return 0;
 873 }
 874
 875 static int handle_stctg(struct kvm_vcpu *vcpu)
 876 {
 877         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
 878         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
 879         int reg, rc, nr_regs;
 880         u64 ctl_array[16];
 881         u64 ga;
 882
 883         vcpu->stat.instruction_stctg++;
 884
 885         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 886                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 887
 888         ga = kvm_s390_get_base_disp_rsy(vcpu);
 889
 890         if (ga & 7)
 891                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 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         reg = reg1;
 897         nr_regs = 0;
 898         do {
 899                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
 900                 if (reg == reg3)
 901                         break;
 902                 reg = (reg + 1) % 16;
 903         } while (1);
 904         rc = write_guest(vcpu, ga, ctl_array, nr_regs * sizeof(u64));
 905         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
 906 }
 907
 908 static const intercept_handler_t eb_handlers[256] = {
 909         [0x2f] = handle_lctlg,
 910         [0x25] = handle_stctg,
 911 };
 912
 913 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
 914 {
 915         intercept_handler_t handler;
 916
 917         handler = eb_handlers[vcpu->arch.sie_block->ipb & 0xff];
 918         if (handler)
 919                 return handler(vcpu);
 920         return -EOPNOTSUPP;
 921 }
 922
 923 static int handle_tprot(struct kvm_vcpu *vcpu)
 924 {
 925         u64 address1, address2;
 926         unsigned long hva, gpa;
 927         int ret = 0, cc = 0;
 928         bool writable;
 929
 930         vcpu->stat.instruction_tprot++;
 931
 932         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 933                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 934
 935         kvm_s390_get_base_disp_sse(vcpu, &address1, &address2);
 936
 937         /* we only handle the Linux memory detection case:
 938          * access key == 0
 939          * everything else goes to userspace. */
 940         if (address2 & 0xf0)
 941                 return -EOPNOTSUPP;
 942         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
 943                 ipte_lock(vcpu);
 944         ret = guest_translate_address(vcpu, address1, &gpa, 1);
 945         if (ret == PGM_PROTECTION) {
 946                 /* Write protected? Try again with read-only... */
 947                 cc = 1;
 948                 ret = guest_translate_address(vcpu, address1, &gpa, 0);
 949         }
 950         if (ret) {
 951                 if (ret == PGM_ADDRESSING || ret == PGM_TRANSLATION_SPEC) {
 952                         ret = kvm_s390_inject_program_int(vcpu, ret);
 953                 } else if (ret > 0) {
 954                         /* Translation not available */
 955                         kvm_s390_set_psw_cc(vcpu, 3);
 956                         ret = 0;
 957                 }
 958                 goto out_unlock;
 959         }
 960
 961         hva = gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
 962         if (kvm_is_error_hva(hva)) {
 963                 ret = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 964         } else {
 965                 if (!writable)
 966                         cc = 1;         /* Write not permitted ==> read-only */
 967                 kvm_s390_set_psw_cc(vcpu, cc);
 968                 /* Note: CC2 only occurs for storage keys (not supported yet) */
 969         }
 970 out_unlock:
 971         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
 972                 ipte_unlock(vcpu);
 973         return ret;
 974 }
 975
 976 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
 977 {
 978         /* For e5xx... instructions we only handle TPROT */
 979         if ((vcpu->arch.sie_block->ipa & 0x00ff) == 0x01)
 980                 return handle_tprot(vcpu);
 981         return -EOPNOTSUPP;
 982 }
 983
 984 static int handle_sckpf(struct kvm_vcpu *vcpu)
 985 {
 986         u32 value;
 987
 988         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 989                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 990
 991         if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
 992                 return kvm_s390_inject_program_int(vcpu,
 993                                                    PGM_SPECIFICATION);
 994
 995         value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
 996         vcpu->arch.sie_block->todpr = value;
 997
 998         return 0;
 999 }
1000
1001 static const intercept_handler_t x01_handlers[256] = {
1002         [0x07] = handle_sckpf,
1003 };
1004
1005 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1006 {
1007         intercept_handler_t handler;
1008
1009         handler = x01_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
1010         if (handler)
1011                 return handler(vcpu);
1012         return -EOPNOTSUPP;
1013 }