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         ar_t ar;
  40         u64 op2;
  41
  42         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
  43                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
  44
  45         op2 = kvm_s390_get_base_disp_s(vcpu, &ar);
  46         if (op2 & 7)    /* Operand must be on a doubleword boundary */
  47                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
  48         rc = read_guest(vcpu, op2, ar, &val, sizeof(val));
  49         if (rc)
  50                 return kvm_s390_inject_prog_cond(vcpu, rc);
  51
  52         if (store_tod_clock(&hostclk)) {
  53                 kvm_s390_set_psw_cc(vcpu, 3);
  54                 return 0;
  55         }
  56         val = (val - hostclk) & ~0x3fUL;
  57
  58         mutex_lock(&vcpu->kvm->lock);
  59         kvm_for_each_vcpu(i, cpup, vcpu->kvm)
  60                 cpup->arch.sie_block->epoch = val;
  61         mutex_unlock(&vcpu->kvm->lock);
  62
  63         kvm_s390_set_psw_cc(vcpu, 0);
  64         return 0;
  65 }
  66
  67 static int handle_set_prefix(struct kvm_vcpu *vcpu)
  68 {
  69         u64 operand2;
  70         u32 address;
  71         int rc;
  72         ar_t ar;
  73
  74         vcpu->stat.instruction_spx++;
  75
  76         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
  77                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
  78
  79         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
  80
  81         /* must be word boundary */
  82         if (operand2 & 3)
  83                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
  84
  85         /* get the value */
  86         rc = read_guest(vcpu, operand2, ar, &address, sizeof(address));
  87         if (rc)
  88                 return kvm_s390_inject_prog_cond(vcpu, rc);
  89
  90         address &= 0x7fffe000u;
  91
  92         /*
  93          * Make sure the new value is valid memory. We only need to check the
  94          * first page, since address is 8k aligned and memory pieces are always
  95          * at least 1MB aligned and have at least a size of 1MB.
  96          */
  97         if (kvm_is_error_gpa(vcpu->kvm, address))
  98                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
  99
 100         kvm_s390_set_prefix(vcpu, address);
 101
 102         VCPU_EVENT(vcpu, 5, "setting prefix to %x", address);
 103         trace_kvm_s390_handle_prefix(vcpu, 1, address);
 104         return 0;
 105 }
 106
 107 static int handle_store_prefix(struct kvm_vcpu *vcpu)
 108 {
 109         u64 operand2;
 110         u32 address;
 111         int rc;
 112         ar_t ar;
 113
 114         vcpu->stat.instruction_stpx++;
 115
 116         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 117                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 118
 119         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 120
 121         /* must be word boundary */
 122         if (operand2 & 3)
 123                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 124
 125         address = kvm_s390_get_prefix(vcpu);
 126
 127         /* get the value */
 128         rc = write_guest(vcpu, operand2, ar, &address, sizeof(address));
 129         if (rc)
 130                 return kvm_s390_inject_prog_cond(vcpu, rc);
 131
 132         VCPU_EVENT(vcpu, 5, "storing prefix to %x", address);
 133         trace_kvm_s390_handle_prefix(vcpu, 0, address);
 134         return 0;
 135 }
 136
 137 static int handle_store_cpu_address(struct kvm_vcpu *vcpu)
 138 {
 139         u16 vcpu_id = vcpu->vcpu_id;
 140         u64 ga;
 141         int rc;
 142         ar_t ar;
 143
 144         vcpu->stat.instruction_stap++;
 145
 146         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 147                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 148
 149         ga = kvm_s390_get_base_disp_s(vcpu, &ar);
 150
 151         if (ga & 1)
 152                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 153
 154         rc = write_guest(vcpu, ga, ar, &vcpu_id, sizeof(vcpu_id));
 155         if (rc)
 156                 return kvm_s390_inject_prog_cond(vcpu, rc);
 157
 158         VCPU_EVENT(vcpu, 5, "storing cpu address to %llx", ga);
 159         trace_kvm_s390_handle_stap(vcpu, ga);
 160         return 0;
 161 }
 162
 163 static int __skey_check_enable(struct kvm_vcpu *vcpu)
 164 {
 165         int rc = 0;
 166         if (!(vcpu->arch.sie_block->ictl & (ICTL_ISKE | ICTL_SSKE | ICTL_RRBE)))
 167                 return rc;
 168
 169         rc = s390_enable_skey();
 170         trace_kvm_s390_skey_related_inst(vcpu);
 171         vcpu->arch.sie_block->ictl &= ~(ICTL_ISKE | ICTL_SSKE | ICTL_RRBE);
 172         return rc;
 173 }
 174
 175
 176 static int handle_skey(struct kvm_vcpu *vcpu)
 177 {
 178         int rc = __skey_check_enable(vcpu);
 179
 180         if (rc)
 181                 return rc;
 182         vcpu->stat.instruction_storage_key++;
 183
 184         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 185                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 186
 187         kvm_s390_rewind_psw(vcpu, 4);
 188         VCPU_EVENT(vcpu, 4, "%s", "retrying storage key operation");
 189         return 0;
 190 }
 191
 192 static int handle_ipte_interlock(struct kvm_vcpu *vcpu)
 193 {
 194         vcpu->stat.instruction_ipte_interlock++;
 195         if (psw_bits(vcpu->arch.sie_block->gpsw).p)
 196                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 197         wait_event(vcpu->kvm->arch.ipte_wq, !ipte_lock_held(vcpu));
 198         kvm_s390_rewind_psw(vcpu, 4);
 199         VCPU_EVENT(vcpu, 4, "%s", "retrying ipte interlock operation");
 200         return 0;
 201 }
 202
 203 static int handle_test_block(struct kvm_vcpu *vcpu)
 204 {
 205         gpa_t addr;
 206         int reg2;
 207
 208         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 209                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 210
 211         kvm_s390_get_regs_rre(vcpu, NULL, &reg2);
 212         addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 213         addr = kvm_s390_logical_to_effective(vcpu, addr);
 214         if (kvm_s390_check_low_addr_prot_real(vcpu, addr))
 215                 return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
 216         addr = kvm_s390_real_to_abs(vcpu, addr);
 217
 218         if (kvm_is_error_gpa(vcpu->kvm, addr))
 219                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 220         /*
 221          * We don't expect errors on modern systems, and do not care
 222          * about storage keys (yet), so let's just clear the page.
 223          */
 224         if (kvm_clear_guest(vcpu->kvm, addr, PAGE_SIZE))
 225                 return -EFAULT;
 226         kvm_s390_set_psw_cc(vcpu, 0);
 227         vcpu->run->s.regs.gprs[0] = 0;
 228         return 0;
 229 }
 230
 231 static int handle_tpi(struct kvm_vcpu *vcpu)
 232 {
 233         struct kvm_s390_interrupt_info *inti;
 234         unsigned long len;
 235         u32 tpi_data[3];
 236         int rc;
 237         u64 addr;
 238         ar_t ar;
 239
 240         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 241         if (addr & 3)
 242                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 243
 244         inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0);
 245         if (!inti) {
 246                 kvm_s390_set_psw_cc(vcpu, 0);
 247                 return 0;
 248         }
 249
 250         tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr;
 251         tpi_data[1] = inti->io.io_int_parm;
 252         tpi_data[2] = inti->io.io_int_word;
 253         if (addr) {
 254                 /*
 255                  * Store the two-word I/O interruption code into the
 256                  * provided area.
 257                  */
 258                 len = sizeof(tpi_data) - 4;
 259                 rc = write_guest(vcpu, addr, ar, &tpi_data, len);
 260                 if (rc) {
 261                         rc = kvm_s390_inject_prog_cond(vcpu, rc);
 262                         goto reinject_interrupt;
 263                 }
 264         } else {
 265                 /*
 266                  * Store the three-word I/O interruption code into
 267                  * the appropriate lowcore area.
 268                  */
 269                 len = sizeof(tpi_data);
 270                 if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len)) {
 271                         /* failed writes to the low core are not recoverable */
 272                         rc = -EFAULT;
 273                         goto reinject_interrupt;
 274                 }
 275         }
 276
 277         /* irq was successfully handed to the guest */
 278         kfree(inti);
 279         kvm_s390_set_psw_cc(vcpu, 1);
 280         return 0;
 281 reinject_interrupt:
 282         /*
 283          * If we encounter a problem storing the interruption code, the
 284          * instruction is suppressed from the guest's view: reinject the
 285          * interrupt.
 286          */
 287         if (kvm_s390_reinject_io_int(vcpu->kvm, inti)) {
 288                 kfree(inti);
 289                 rc = -EFAULT;
 290         }
 291         /* don't set the cc, a pgm irq was injected or we drop to user space */
 292         return rc ? -EFAULT : 0;
 293 }
 294
 295 static int handle_tsch(struct kvm_vcpu *vcpu)
 296 {
 297         struct kvm_s390_interrupt_info *inti;
 298
 299         inti = kvm_s390_get_io_int(vcpu->kvm, 0,
 300                                    vcpu->run->s.regs.gprs[1]);
 301
 302         /*
 303          * Prepare exit to userspace.
 304          * We indicate whether we dequeued a pending I/O interrupt
 305          * so that userspace can re-inject it if the instruction gets
 306          * a program check. While this may re-order the pending I/O
 307          * interrupts, this is no problem since the priority is kept
 308          * intact.
 309          */
 310         vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
 311         vcpu->run->s390_tsch.dequeued = !!inti;
 312         if (inti) {
 313                 vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
 314                 vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
 315                 vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
 316                 vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
 317         }
 318         vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
 319         kfree(inti);
 320         return -EREMOTE;
 321 }
 322
 323 static int handle_io_inst(struct kvm_vcpu *vcpu)
 324 {
 325         VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
 326
 327         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 328                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 329
 330         if (vcpu->kvm->arch.css_support) {
 331                 /*
 332                  * Most I/O instructions will be handled by userspace.
 333                  * Exceptions are tpi and the interrupt portion of tsch.
 334                  */
 335                 if (vcpu->arch.sie_block->ipa == 0xb236)
 336                         return handle_tpi(vcpu);
 337                 if (vcpu->arch.sie_block->ipa == 0xb235)
 338                         return handle_tsch(vcpu);
 339                 /* Handle in userspace. */
 340                 return -EOPNOTSUPP;
 341         } else {
 342                 /*
 343                  * Set condition code 3 to stop the guest from issuing channel
 344                  * I/O instructions.
 345                  */
 346                 kvm_s390_set_psw_cc(vcpu, 3);
 347                 return 0;
 348         }
 349 }
 350
 351 static int handle_stfl(struct kvm_vcpu *vcpu)
 352 {
 353         int rc;
 354         unsigned int fac;
 355
 356         vcpu->stat.instruction_stfl++;
 357
 358         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 359                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 360
 361         /*
 362          * We need to shift the lower 32 facility bits (bit 0-31) from a u64
 363          * into a u32 memory representation. They will remain bits 0-31.
 364          */
 365         fac = *vcpu->kvm->arch.model.fac->list >> 32;
 366         rc = write_guest_lc(vcpu, offsetof(struct _lowcore, stfl_fac_list),
 367                             &fac, sizeof(fac));
 368         if (rc)
 369                 return rc;
 370         VCPU_EVENT(vcpu, 5, "store facility list value %x", fac);
 371         trace_kvm_s390_handle_stfl(vcpu, fac);
 372         return 0;
 373 }
 374
 375 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
 376 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
 377 #define PSW_ADDR_24 0x0000000000ffffffUL
 378 #define PSW_ADDR_31 0x000000007fffffffUL
 379
 380 int is_valid_psw(psw_t *psw)
 381 {
 382         if (psw->mask & PSW_MASK_UNASSIGNED)
 383                 return 0;
 384         if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
 385                 if (psw->addr & ~PSW_ADDR_31)
 386                         return 0;
 387         }
 388         if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
 389                 return 0;
 390         if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
 391                 return 0;
 392         if (psw->addr & 1)
 393                 return 0;
 394         return 1;
 395 }
 396
 397 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
 398 {
 399         psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
 400         psw_compat_t new_psw;
 401         u64 addr;
 402         int rc;
 403         ar_t ar;
 404
 405         if (gpsw->mask & PSW_MASK_PSTATE)
 406                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 407
 408         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 409         if (addr & 7)
 410                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 411
 412         rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
 413         if (rc)
 414                 return kvm_s390_inject_prog_cond(vcpu, rc);
 415         if (!(new_psw.mask & PSW32_MASK_BASE))
 416                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 417         gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
 418         gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
 419         gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
 420         if (!is_valid_psw(gpsw))
 421                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 422         return 0;
 423 }
 424
 425 static int handle_lpswe(struct kvm_vcpu *vcpu)
 426 {
 427         psw_t new_psw;
 428         u64 addr;
 429         int rc;
 430         ar_t ar;
 431
 432         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 433                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 434
 435         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 436         if (addr & 7)
 437                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 438         rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
 439         if (rc)
 440                 return kvm_s390_inject_prog_cond(vcpu, rc);
 441         vcpu->arch.sie_block->gpsw = new_psw;
 442         if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
 443                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 444         return 0;
 445 }
 446
 447 static int handle_stidp(struct kvm_vcpu *vcpu)
 448 {
 449         u64 stidp_data = vcpu->arch.stidp_data;
 450         u64 operand2;
 451         int rc;
 452         ar_t ar;
 453
 454         vcpu->stat.instruction_stidp++;
 455
 456         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 457                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 458
 459         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 460
 461         if (operand2 & 7)
 462                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 463
 464         rc = write_guest(vcpu, operand2, ar, &stidp_data, sizeof(stidp_data));
 465         if (rc)
 466                 return kvm_s390_inject_prog_cond(vcpu, rc);
 467
 468         VCPU_EVENT(vcpu, 5, "%s", "store cpu id");
 469         return 0;
 470 }
 471
 472 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
 473 {
 474         int cpus = 0;
 475         int n;
 476
 477         cpus = atomic_read(&vcpu->kvm->online_vcpus);
 478
 479         /* deal with other level 3 hypervisors */
 480         if (stsi(mem, 3, 2, 2))
 481                 mem->count = 0;
 482         if (mem->count < 8)
 483                 mem->count++;
 484         for (n = mem->count - 1; n > 0 ; n--)
 485                 memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
 486
 487         memset(&mem->vm[0], 0, sizeof(mem->vm[0]));
 488         mem->vm[0].cpus_total = cpus;
 489         mem->vm[0].cpus_configured = cpus;
 490         mem->vm[0].cpus_standby = 0;
 491         mem->vm[0].cpus_reserved = 0;
 492         mem->vm[0].caf = 1000;
 493         memcpy(mem->vm[0].name, "KVMguest", 8);
 494         ASCEBC(mem->vm[0].name, 8);
 495         memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
 496         ASCEBC(mem->vm[0].cpi, 16);
 497 }
 498
 499 static void insert_stsi_usr_data(struct kvm_vcpu *vcpu, u64 addr, ar_t ar,
 500                                  u8 fc, u8 sel1, u16 sel2)
 501 {
 502         vcpu->run->exit_reason = KVM_EXIT_S390_STSI;
 503         vcpu->run->s390_stsi.addr = addr;
 504         vcpu->run->s390_stsi.ar = ar;
 505         vcpu->run->s390_stsi.fc = fc;
 506         vcpu->run->s390_stsi.sel1 = sel1;
 507         vcpu->run->s390_stsi.sel2 = sel2;
 508 }
 509
 510 static int handle_stsi(struct kvm_vcpu *vcpu)
 511 {
 512         int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
 513         int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
 514         int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
 515         unsigned long mem = 0;
 516         u64 operand2;
 517         int rc = 0;
 518         ar_t ar;
 519
 520         vcpu->stat.instruction_stsi++;
 521         VCPU_EVENT(vcpu, 4, "stsi: fc: %x sel1: %x sel2: %x", fc, sel1, sel2);
 522
 523         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 524                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 525
 526         if (fc > 3) {
 527                 kvm_s390_set_psw_cc(vcpu, 3);
 528                 return 0;
 529         }
 530
 531         if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
 532             || vcpu->run->s.regs.gprs[1] & 0xffff0000)
 533                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 534
 535         if (fc == 0) {
 536                 vcpu->run->s.regs.gprs[0] = 3 << 28;
 537                 kvm_s390_set_psw_cc(vcpu, 0);
 538                 return 0;
 539         }
 540
 541         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 542
 543         if (operand2 & 0xfff)
 544                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 545
 546         switch (fc) {
 547         case 1: /* same handling for 1 and 2 */
 548         case 2:
 549                 mem = get_zeroed_page(GFP_KERNEL);
 550                 if (!mem)
 551                         goto out_no_data;
 552                 if (stsi((void *) mem, fc, sel1, sel2))
 553                         goto out_no_data;
 554                 break;
 555         case 3:
 556                 if (sel1 != 2 || sel2 != 2)
 557                         goto out_no_data;
 558                 mem = get_zeroed_page(GFP_KERNEL);
 559                 if (!mem)
 560                         goto out_no_data;
 561                 handle_stsi_3_2_2(vcpu, (void *) mem);
 562                 break;
 563         }
 564
 565         rc = write_guest(vcpu, operand2, ar, (void *)mem, PAGE_SIZE);
 566         if (rc) {
 567                 rc = kvm_s390_inject_prog_cond(vcpu, rc);
 568                 goto out;
 569         }
 570         if (vcpu->kvm->arch.user_stsi) {
 571                 insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
 572                 rc = -EREMOTE;
 573         }
 574         trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
 575         free_page(mem);
 576         kvm_s390_set_psw_cc(vcpu, 0);
 577         vcpu->run->s.regs.gprs[0] = 0;
 578         return rc;
 579 out_no_data:
 580         kvm_s390_set_psw_cc(vcpu, 3);
 581 out:
 582         free_page(mem);
 583         return rc;
 584 }
 585
 586 static const intercept_handler_t b2_handlers[256] = {
 587         [0x02] = handle_stidp,
 588         [0x04] = handle_set_clock,
 589         [0x10] = handle_set_prefix,
 590         [0x11] = handle_store_prefix,
 591         [0x12] = handle_store_cpu_address,
 592         [0x21] = handle_ipte_interlock,
 593         [0x29] = handle_skey,
 594         [0x2a] = handle_skey,
 595         [0x2b] = handle_skey,
 596         [0x2c] = handle_test_block,
 597         [0x30] = handle_io_inst,
 598         [0x31] = handle_io_inst,
 599         [0x32] = handle_io_inst,
 600         [0x33] = handle_io_inst,
 601         [0x34] = handle_io_inst,
 602         [0x35] = handle_io_inst,
 603         [0x36] = handle_io_inst,
 604         [0x37] = handle_io_inst,
 605         [0x38] = handle_io_inst,
 606         [0x39] = handle_io_inst,
 607         [0x3a] = handle_io_inst,
 608         [0x3b] = handle_io_inst,
 609         [0x3c] = handle_io_inst,
 610         [0x50] = handle_ipte_interlock,
 611         [0x5f] = handle_io_inst,
 612         [0x74] = handle_io_inst,
 613         [0x76] = handle_io_inst,
 614         [0x7d] = handle_stsi,
 615         [0xb1] = handle_stfl,
 616         [0xb2] = handle_lpswe,
 617 };
 618
 619 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
 620 {
 621         intercept_handler_t handler;
 622
 623         /*
 624          * A lot of B2 instructions are priviledged. Here we check for
 625          * the privileged ones, that we can handle in the kernel.
 626          * Anything else goes to userspace.
 627          */
 628         handler = b2_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
 629         if (handler)
 630                 return handler(vcpu);
 631
 632         return -EOPNOTSUPP;
 633 }
 634
 635 static int handle_epsw(struct kvm_vcpu *vcpu)
 636 {
 637         int reg1, reg2;
 638
 639         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 640
 641         /* This basically extracts the mask half of the psw. */
 642         vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
 643         vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
 644         if (reg2) {
 645                 vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
 646                 vcpu->run->s.regs.gprs[reg2] |=
 647                         vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
 648         }
 649         return 0;
 650 }
 651
 652 #define PFMF_RESERVED   0xfffc0101UL
 653 #define PFMF_SK         0x00020000UL
 654 #define PFMF_CF         0x00010000UL
 655 #define PFMF_UI         0x00008000UL
 656 #define PFMF_FSC        0x00007000UL
 657 #define PFMF_NQ         0x00000800UL
 658 #define PFMF_MR         0x00000400UL
 659 #define PFMF_MC         0x00000200UL
 660 #define PFMF_KEY        0x000000feUL
 661
 662 static int handle_pfmf(struct kvm_vcpu *vcpu)
 663 {
 664         int reg1, reg2;
 665         unsigned long start, end;
 666
 667         vcpu->stat.instruction_pfmf++;
 668
 669         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 670
 671         if (!MACHINE_HAS_PFMF)
 672                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
 673
 674         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 675                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 676
 677         if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
 678                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 679
 680         /* Only provide non-quiescing support if the host supports it */
 681         if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ && !test_facility(14))
 682                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 683
 684         /* No support for conditional-SSKE */
 685         if (vcpu->run->s.regs.gprs[reg1] & (PFMF_MR | PFMF_MC))
 686                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 687
 688         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 689         start = kvm_s390_logical_to_effective(vcpu, start);
 690
 691         switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
 692         case 0x00000000:
 693                 end = (start + (1UL << 12)) & ~((1UL << 12) - 1);
 694                 break;
 695         case 0x00001000:
 696                 end = (start + (1UL << 20)) & ~((1UL << 20) - 1);
 697                 break;
 698         /* We dont support EDAT2
 699         case 0x00002000:
 700                 end = (start + (1UL << 31)) & ~((1UL << 31) - 1);
 701                 break;*/
 702         default:
 703                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 704         }
 705
 706         if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
 707                 if (kvm_s390_check_low_addr_prot_real(vcpu, start))
 708                         return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
 709         }
 710
 711         while (start < end) {
 712                 unsigned long useraddr, abs_addr;
 713
 714                 /* Translate guest address to host address */
 715                 if ((vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) == 0)
 716                         abs_addr = kvm_s390_real_to_abs(vcpu, start);
 717                 else
 718                         abs_addr = start;
 719                 useraddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(abs_addr));
 720                 if (kvm_is_error_hva(useraddr))
 721                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 722
 723                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
 724                         if (clear_user((void __user *)useraddr, PAGE_SIZE))
 725                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 726                 }
 727
 728                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
 729                         int rc = __skey_check_enable(vcpu);
 730
 731                         if (rc)
 732                                 return rc;
 733                         if (set_guest_storage_key(current->mm, useraddr,
 734                                         vcpu->run->s.regs.gprs[reg1] & PFMF_KEY,
 735                                         vcpu->run->s.regs.gprs[reg1] & PFMF_NQ))
 736                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 737                 }
 738
 739                 start += PAGE_SIZE;
 740         }
 741         if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC)
 742                 vcpu->run->s.regs.gprs[reg2] = end;
 743         return 0;
 744 }
 745
 746 static int handle_essa(struct kvm_vcpu *vcpu)
 747 {
 748         /* entries expected to be 1FF */
 749         int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
 750         unsigned long *cbrlo, cbrle;
 751         struct gmap *gmap;
 752         int i;
 753
 754         VCPU_EVENT(vcpu, 5, "cmma release %d pages", entries);
 755         gmap = vcpu->arch.gmap;
 756         vcpu->stat.instruction_essa++;
 757         if (!kvm_s390_cmma_enabled(vcpu->kvm))
 758                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
 759
 760         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 761                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 762
 763         if (((vcpu->arch.sie_block->ipb & 0xf0000000) >> 28) > 6)
 764                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 765
 766         /* Rewind PSW to repeat the ESSA instruction */
 767         kvm_s390_rewind_psw(vcpu, 4);
 768         vcpu->arch.sie_block->cbrlo &= PAGE_MASK;       /* reset nceo */
 769         cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
 770         down_read(&gmap->mm->mmap_sem);
 771         for (i = 0; i < entries; ++i) {
 772                 cbrle = cbrlo[i];
 773                 if (unlikely(cbrle & ~PAGE_MASK || cbrle < 2 * PAGE_SIZE))
 774                         /* invalid entry */
 775                         break;
 776                 /* try to free backing */
 777                 __gmap_zap(gmap, cbrle);
 778         }
 779         up_read(&gmap->mm->mmap_sem);
 780         if (i < entries)
 781                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 782         return 0;
 783 }
 784
 785 static const intercept_handler_t b9_handlers[256] = {
 786         [0x8a] = handle_ipte_interlock,
 787         [0x8d] = handle_epsw,
 788         [0x8e] = handle_ipte_interlock,
 789         [0x8f] = handle_ipte_interlock,
 790         [0xab] = handle_essa,
 791         [0xaf] = handle_pfmf,
 792 };
 793
 794 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
 795 {
 796         intercept_handler_t handler;
 797
 798         /* This is handled just as for the B2 instructions. */
 799         handler = b9_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
 800         if (handler)
 801                 return handler(vcpu);
 802
 803         return -EOPNOTSUPP;
 804 }
 805
 806 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
 807 {
 808         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
 809         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
 810         int reg, rc, nr_regs;
 811         u32 ctl_array[16];
 812         u64 ga;
 813         ar_t ar;
 814
 815         vcpu->stat.instruction_lctl++;
 816
 817         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 818                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 819
 820         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
 821
 822         if (ga & 3)
 823                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 824
 825         VCPU_EVENT(vcpu, 5, "lctl r1:%x, r3:%x, addr:%llx", reg1, reg3, ga);
 826         trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
 827
 828         nr_regs = ((reg3 - reg1) & 0xf) + 1;
 829         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
 830         if (rc)
 831                 return kvm_s390_inject_prog_cond(vcpu, rc);
 832         reg = reg1;
 833         nr_regs = 0;
 834         do {
 835                 vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
 836                 vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
 837                 if (reg == reg3)
 838                         break;
 839                 reg = (reg + 1) % 16;
 840         } while (1);
 841         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 842         return 0;
 843 }
 844
 845 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
 846 {
 847         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
 848         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
 849         int reg, rc, nr_regs;
 850         u32 ctl_array[16];
 851         u64 ga;
 852         ar_t ar;
 853
 854         vcpu->stat.instruction_stctl++;
 855
 856         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 857                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 858
 859         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
 860
 861         if (ga & 3)
 862                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 863
 864         VCPU_EVENT(vcpu, 5, "stctl r1:%x, r3:%x, addr:%llx", reg1, reg3, ga);
 865         trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
 866
 867         reg = reg1;
 868         nr_regs = 0;
 869         do {
 870                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
 871                 if (reg == reg3)
 872                         break;
 873                 reg = (reg + 1) % 16;
 874         } while (1);
 875         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
 876         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
 877 }
 878
 879 static int handle_lctlg(struct kvm_vcpu *vcpu)
 880 {
 881         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
 882         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
 883         int reg, rc, nr_regs;
 884         u64 ctl_array[16];
 885         u64 ga;
 886         ar_t ar;
 887
 888         vcpu->stat.instruction_lctlg++;
 889
 890         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 891                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 892
 893         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
 894
 895         if (ga & 7)
 896                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 897
 898         VCPU_EVENT(vcpu, 5, "lctlg r1:%x, r3:%x, addr:%llx", reg1, reg3, ga);
 899         trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
 900
 901         nr_regs = ((reg3 - reg1) & 0xf) + 1;
 902         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
 903         if (rc)
 904                 return kvm_s390_inject_prog_cond(vcpu, rc);
 905         reg = reg1;
 906         nr_regs = 0;
 907         do {
 908                 vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
 909                 if (reg == reg3)
 910                         break;
 911                 reg = (reg + 1) % 16;
 912         } while (1);
 913         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 914         return 0;
 915 }
 916
 917 static int handle_stctg(struct kvm_vcpu *vcpu)
 918 {
 919         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
 920         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
 921         int reg, rc, nr_regs;
 922         u64 ctl_array[16];
 923         u64 ga;
 924         ar_t ar;
 925
 926         vcpu->stat.instruction_stctg++;
 927
 928         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 929                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 930
 931         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
 932
 933         if (ga & 7)
 934                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 935
 936         VCPU_EVENT(vcpu, 5, "stctg r1:%x, r3:%x, addr:%llx", reg1, reg3, ga);
 937         trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
 938
 939         reg = reg1;
 940         nr_regs = 0;
 941         do {
 942                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
 943                 if (reg == reg3)
 944                         break;
 945                 reg = (reg + 1) % 16;
 946         } while (1);
 947         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
 948         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
 949 }
 950
 951 static const intercept_handler_t eb_handlers[256] = {
 952         [0x2f] = handle_lctlg,
 953         [0x25] = handle_stctg,
 954 };
 955
 956 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
 957 {
 958         intercept_handler_t handler;
 959
 960         handler = eb_handlers[vcpu->arch.sie_block->ipb & 0xff];
 961         if (handler)
 962                 return handler(vcpu);
 963         return -EOPNOTSUPP;
 964 }
 965
 966 static int handle_tprot(struct kvm_vcpu *vcpu)
 967 {
 968         u64 address1, address2;
 969         unsigned long hva, gpa;
 970         int ret = 0, cc = 0;
 971         bool writable;
 972         ar_t ar;
 973
 974         vcpu->stat.instruction_tprot++;
 975
 976         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 977                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 978
 979         kvm_s390_get_base_disp_sse(vcpu, &address1, &address2, &ar, NULL);
 980
 981         /* we only handle the Linux memory detection case:
 982          * access key == 0
 983          * everything else goes to userspace. */
 984         if (address2 & 0xf0)
 985                 return -EOPNOTSUPP;
 986         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
 987                 ipte_lock(vcpu);
 988         ret = guest_translate_address(vcpu, address1, ar, &gpa, 1);
 989         if (ret == PGM_PROTECTION) {
 990                 /* Write protected? Try again with read-only... */
 991                 cc = 1;
 992                 ret = guest_translate_address(vcpu, address1, ar, &gpa, 0);
 993         }
 994         if (ret) {
 995                 if (ret == PGM_ADDRESSING || ret == PGM_TRANSLATION_SPEC) {
 996                         ret = kvm_s390_inject_program_int(vcpu, ret);
 997                 } else if (ret > 0) {
 998                         /* Translation not available */
 999                         kvm_s390_set_psw_cc(vcpu, 3);
1000                         ret = 0;
1001                 }
1002                 goto out_unlock;
1003         }
1004
1005         hva = gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
1006         if (kvm_is_error_hva(hva)) {
1007                 ret = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1008         } else {
1009                 if (!writable)
1010                         cc = 1;         /* Write not permitted ==> read-only */
1011                 kvm_s390_set_psw_cc(vcpu, cc);
1012                 /* Note: CC2 only occurs for storage keys (not supported yet) */
1013         }
1014 out_unlock:
1015         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1016                 ipte_unlock(vcpu);
1017         return ret;
1018 }
1019
1020 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
1021 {
1022         /* For e5xx... instructions we only handle TPROT */
1023         if ((vcpu->arch.sie_block->ipa & 0x00ff) == 0x01)
1024                 return handle_tprot(vcpu);
1025         return -EOPNOTSUPP;
1026 }
1027
1028 static int handle_sckpf(struct kvm_vcpu *vcpu)
1029 {
1030         u32 value;
1031
1032         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1033                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1034
1035         if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
1036                 return kvm_s390_inject_program_int(vcpu,
1037                                                    PGM_SPECIFICATION);
1038
1039         value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
1040         vcpu->arch.sie_block->todpr = value;
1041
1042         return 0;
1043 }
1044
1045 static const intercept_handler_t x01_handlers[256] = {
1046         [0x07] = handle_sckpf,
1047 };
1048
1049 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1050 {
1051         intercept_handler_t handler;
1052
1053         handler = x01_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
1054         if (handler)
1055                 return handler(vcpu);
1056         return -EOPNOTSUPP;
1057 }