arch/s390/kvm/priv.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * handling privileged instructions
   4  *
   5  * Copyright IBM Corp. 2008, 2018
   6  *
   7  *    Author(s): Carsten Otte <cotte@de.ibm.com>
   8  *               Christian Borntraeger <borntraeger@de.ibm.com>
   9  */
  10
  11 #include <linux/kvm.h>
  12 #include <linux/gfp.h>
  13 #include <linux/errno.h>
  14 #include <linux/compat.h>
  15 #include <linux/mm_types.h>
  16
  17 #include <asm/asm-offsets.h>
  18 #include <asm/facility.h>
  19 #include <asm/current.h>
  20 #include <asm/debug.h>
  21 #include <asm/ebcdic.h>
  22 #include <asm/sysinfo.h>
  23 #include <asm/pgtable.h>
  24 #include <asm/page-states.h>
  25 #include <asm/pgalloc.h>
  26 #include <asm/gmap.h>
  27 #include <asm/io.h>
  28 #include <asm/ptrace.h>
  29 #include <asm/sclp.h>
  30 #include "gaccess.h"
  31 #include "kvm-s390.h"
  32 #include "trace.h"
  33
  34 static int handle_ri(struct kvm_vcpu *vcpu)
  35 {
  36         vcpu->stat.instruction_ri++;
  37
  38         if (test_kvm_facility(vcpu->kvm, 64)) {
  39                 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (lazy)");
  40                 vcpu->arch.sie_block->ecb3 |= ECB3_RI;
  41                 kvm_s390_retry_instr(vcpu);
  42                 return 0;
  43         } else
  44                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
  45 }
  46
  47 int kvm_s390_handle_aa(struct kvm_vcpu *vcpu)
  48 {
  49         if ((vcpu->arch.sie_block->ipa & 0xf) <= 4)
  50                 return handle_ri(vcpu);
  51         else
  52                 return -EOPNOTSUPP;
  53 }
  54
  55 static int handle_gs(struct kvm_vcpu *vcpu)
  56 {
  57         vcpu->stat.instruction_gs++;
  58
  59         if (test_kvm_facility(vcpu->kvm, 133)) {
  60                 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (lazy)");
  61                 preempt_disable();
  62                 __ctl_set_bit(2, 4);
  63                 current->thread.gs_cb = (struct gs_cb *)&vcpu->run->s.regs.gscb;
  64                 restore_gs_cb(current->thread.gs_cb);
  65                 preempt_enable();
  66                 vcpu->arch.sie_block->ecb |= ECB_GS;
  67                 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
  68                 vcpu->arch.gs_enabled = 1;
  69                 kvm_s390_retry_instr(vcpu);
  70                 return 0;
  71         } else
  72                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
  73 }
  74
  75 int kvm_s390_handle_e3(struct kvm_vcpu *vcpu)
  76 {
  77         int code = vcpu->arch.sie_block->ipb & 0xff;
  78
  79         if (code == 0x49 || code == 0x4d)
  80                 return handle_gs(vcpu);
  81         else
  82                 return -EOPNOTSUPP;
  83 }
  84 /* Handle SCK (SET CLOCK) interception */
  85 static int handle_set_clock(struct kvm_vcpu *vcpu)
  86 {
  87         struct kvm_s390_vm_tod_clock gtod = { 0 };
  88         int rc;
  89         u8 ar;
  90         u64 op2;
  91
  92         vcpu->stat.instruction_sck++;
  93
  94         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
  95                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
  96
  97         op2 = kvm_s390_get_base_disp_s(vcpu, &ar);
  98         if (op2 & 7)    /* Operand must be on a doubleword boundary */
  99                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 100         rc = read_guest(vcpu, op2, ar, &gtod.tod, sizeof(gtod.tod));
 101         if (rc)
 102                 return kvm_s390_inject_prog_cond(vcpu, rc);
 103
 104         VCPU_EVENT(vcpu, 3, "SCK: setting guest TOD to 0x%llx", gtod.tod);
 105         kvm_s390_set_tod_clock(vcpu->kvm, &gtod);
 106
 107         kvm_s390_set_psw_cc(vcpu, 0);
 108         return 0;
 109 }
 110
 111 static int handle_set_prefix(struct kvm_vcpu *vcpu)
 112 {
 113         u64 operand2;
 114         u32 address;
 115         int rc;
 116         u8 ar;
 117
 118         vcpu->stat.instruction_spx++;
 119
 120         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 121                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 122
 123         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 124
 125         /* must be word boundary */
 126         if (operand2 & 3)
 127                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 128
 129         /* get the value */
 130         rc = read_guest(vcpu, operand2, ar, &address, sizeof(address));
 131         if (rc)
 132                 return kvm_s390_inject_prog_cond(vcpu, rc);
 133
 134         address &= 0x7fffe000u;
 135
 136         /*
 137          * Make sure the new value is valid memory. We only need to check the
 138          * first page, since address is 8k aligned and memory pieces are always
 139          * at least 1MB aligned and have at least a size of 1MB.
 140          */
 141         if (kvm_is_error_gpa(vcpu->kvm, address))
 142                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 143
 144         kvm_s390_set_prefix(vcpu, address);
 145         trace_kvm_s390_handle_prefix(vcpu, 1, address);
 146         return 0;
 147 }
 148
 149 static int handle_store_prefix(struct kvm_vcpu *vcpu)
 150 {
 151         u64 operand2;
 152         u32 address;
 153         int rc;
 154         u8 ar;
 155
 156         vcpu->stat.instruction_stpx++;
 157
 158         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 159                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 160
 161         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 162
 163         /* must be word boundary */
 164         if (operand2 & 3)
 165                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 166
 167         address = kvm_s390_get_prefix(vcpu);
 168
 169         /* get the value */
 170         rc = write_guest(vcpu, operand2, ar, &address, sizeof(address));
 171         if (rc)
 172                 return kvm_s390_inject_prog_cond(vcpu, rc);
 173
 174         VCPU_EVENT(vcpu, 3, "STPX: storing prefix 0x%x into 0x%llx", address, operand2);
 175         trace_kvm_s390_handle_prefix(vcpu, 0, address);
 176         return 0;
 177 }
 178
 179 static int handle_store_cpu_address(struct kvm_vcpu *vcpu)
 180 {
 181         u16 vcpu_id = vcpu->vcpu_id;
 182         u64 ga;
 183         int rc;
 184         u8 ar;
 185
 186         vcpu->stat.instruction_stap++;
 187
 188         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 189                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 190
 191         ga = kvm_s390_get_base_disp_s(vcpu, &ar);
 192
 193         if (ga & 1)
 194                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 195
 196         rc = write_guest(vcpu, ga, ar, &vcpu_id, sizeof(vcpu_id));
 197         if (rc)
 198                 return kvm_s390_inject_prog_cond(vcpu, rc);
 199
 200         VCPU_EVENT(vcpu, 3, "STAP: storing cpu address (%u) to 0x%llx", vcpu_id, ga);
 201         trace_kvm_s390_handle_stap(vcpu, ga);
 202         return 0;
 203 }
 204
 205 int kvm_s390_skey_check_enable(struct kvm_vcpu *vcpu)
 206 {
 207         int rc;
 208
 209         trace_kvm_s390_skey_related_inst(vcpu);
 210         /* Already enabled? */
 211         if (vcpu->arch.skey_enabled)
 212                 return 0;
 213
 214         rc = s390_enable_skey();
 215         VCPU_EVENT(vcpu, 3, "enabling storage keys for guest: %d", rc);
 216         if (rc)
 217                 return rc;
 218
 219         if (kvm_s390_test_cpuflags(vcpu, CPUSTAT_KSS))
 220                 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_KSS);
 221         if (!vcpu->kvm->arch.use_skf)
 222                 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
 223         else
 224                 vcpu->arch.sie_block->ictl &= ~(ICTL_ISKE | ICTL_SSKE | ICTL_RRBE);
 225         vcpu->arch.skey_enabled = true;
 226         return 0;
 227 }
 228
 229 static int try_handle_skey(struct kvm_vcpu *vcpu)
 230 {
 231         int rc;
 232
 233         rc = kvm_s390_skey_check_enable(vcpu);
 234         if (rc)
 235                 return rc;
 236         if (vcpu->kvm->arch.use_skf) {
 237                 /* with storage-key facility, SIE interprets it for us */
 238                 kvm_s390_retry_instr(vcpu);
 239                 VCPU_EVENT(vcpu, 4, "%s", "retrying storage key operation");
 240                 return -EAGAIN;
 241         }
 242         return 0;
 243 }
 244
 245 static int handle_iske(struct kvm_vcpu *vcpu)
 246 {
 247         unsigned long gaddr, vmaddr;
 248         unsigned char key;
 249         int reg1, reg2;
 250         bool unlocked;
 251         int rc;
 252
 253         vcpu->stat.instruction_iske++;
 254
 255         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 256                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 257
 258         rc = try_handle_skey(vcpu);
 259         if (rc)
 260                 return rc != -EAGAIN ? rc : 0;
 261
 262         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 263
 264         gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 265         gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
 266         gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
 267         vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr));
 268         if (kvm_is_error_hva(vmaddr))
 269                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 270 retry:
 271         unlocked = false;
 272         down_read(&current->mm->mmap_sem);
 273         rc = get_guest_storage_key(current->mm, vmaddr, &key);
 274
 275         if (rc) {
 276                 rc = fixup_user_fault(current, current->mm, vmaddr,
 277                                       FAULT_FLAG_WRITE, &unlocked);
 278                 if (!rc) {
 279                         up_read(&current->mm->mmap_sem);
 280                         goto retry;
 281                 }
 282         }
 283         up_read(&current->mm->mmap_sem);
 284         if (rc == -EFAULT)
 285                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 286         if (rc < 0)
 287                 return rc;
 288         vcpu->run->s.regs.gprs[reg1] &= ~0xff;
 289         vcpu->run->s.regs.gprs[reg1] |= key;
 290         return 0;
 291 }
 292
 293 static int handle_rrbe(struct kvm_vcpu *vcpu)
 294 {
 295         unsigned long vmaddr, gaddr;
 296         int reg1, reg2;
 297         bool unlocked;
 298         int rc;
 299
 300         vcpu->stat.instruction_rrbe++;
 301
 302         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 303                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 304
 305         rc = try_handle_skey(vcpu);
 306         if (rc)
 307                 return rc != -EAGAIN ? rc : 0;
 308
 309         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 310
 311         gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 312         gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
 313         gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
 314         vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr));
 315         if (kvm_is_error_hva(vmaddr))
 316                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 317 retry:
 318         unlocked = false;
 319         down_read(&current->mm->mmap_sem);
 320         rc = reset_guest_reference_bit(current->mm, vmaddr);
 321         if (rc < 0) {
 322                 rc = fixup_user_fault(current, current->mm, vmaddr,
 323                                       FAULT_FLAG_WRITE, &unlocked);
 324                 if (!rc) {
 325                         up_read(&current->mm->mmap_sem);
 326                         goto retry;
 327                 }
 328         }
 329         up_read(&current->mm->mmap_sem);
 330         if (rc == -EFAULT)
 331                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 332         if (rc < 0)
 333                 return rc;
 334         kvm_s390_set_psw_cc(vcpu, rc);
 335         return 0;
 336 }
 337
 338 #define SSKE_NQ 0x8
 339 #define SSKE_MR 0x4
 340 #define SSKE_MC 0x2
 341 #define SSKE_MB 0x1
 342 static int handle_sske(struct kvm_vcpu *vcpu)
 343 {
 344         unsigned char m3 = vcpu->arch.sie_block->ipb >> 28;
 345         unsigned long start, end;
 346         unsigned char key, oldkey;
 347         int reg1, reg2;
 348         bool unlocked;
 349         int rc;
 350
 351         vcpu->stat.instruction_sske++;
 352
 353         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 354                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 355
 356         rc = try_handle_skey(vcpu);
 357         if (rc)
 358                 return rc != -EAGAIN ? rc : 0;
 359
 360         if (!test_kvm_facility(vcpu->kvm, 8))
 361                 m3 &= ~SSKE_MB;
 362         if (!test_kvm_facility(vcpu->kvm, 10))
 363                 m3 &= ~(SSKE_MC | SSKE_MR);
 364         if (!test_kvm_facility(vcpu->kvm, 14))
 365                 m3 &= ~SSKE_NQ;
 366
 367         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 368
 369         key = vcpu->run->s.regs.gprs[reg1] & 0xfe;
 370         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 371         start = kvm_s390_logical_to_effective(vcpu, start);
 372         if (m3 & SSKE_MB) {
 373                 /* start already designates an absolute address */
 374                 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
 375         } else {
 376                 start = kvm_s390_real_to_abs(vcpu, start);
 377                 end = start + PAGE_SIZE;
 378         }
 379
 380         while (start != end) {
 381                 unsigned long vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
 382                 unlocked = false;
 383
 384                 if (kvm_is_error_hva(vmaddr))
 385                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 386
 387                 down_read(&current->mm->mmap_sem);
 388                 rc = cond_set_guest_storage_key(current->mm, vmaddr, key, &oldkey,
 389                                                 m3 & SSKE_NQ, m3 & SSKE_MR,
 390                                                 m3 & SSKE_MC);
 391
 392                 if (rc < 0) {
 393                         rc = fixup_user_fault(current, current->mm, vmaddr,
 394                                               FAULT_FLAG_WRITE, &unlocked);
 395                         rc = !rc ? -EAGAIN : rc;
 396                 }
 397                 up_read(&current->mm->mmap_sem);
 398                 if (rc == -EFAULT)
 399                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 400                 if (rc == -EAGAIN)
 401                         continue;
 402                 if (rc < 0)
 403                         return rc;
 404                 start += PAGE_SIZE;
 405         }
 406
 407         if (m3 & (SSKE_MC | SSKE_MR)) {
 408                 if (m3 & SSKE_MB) {
 409                         /* skey in reg1 is unpredictable */
 410                         kvm_s390_set_psw_cc(vcpu, 3);
 411                 } else {
 412                         kvm_s390_set_psw_cc(vcpu, rc);
 413                         vcpu->run->s.regs.gprs[reg1] &= ~0xff00UL;
 414                         vcpu->run->s.regs.gprs[reg1] |= (u64) oldkey << 8;
 415                 }
 416         }
 417         if (m3 & SSKE_MB) {
 418                 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT)
 419                         vcpu->run->s.regs.gprs[reg2] &= ~PAGE_MASK;
 420                 else
 421                         vcpu->run->s.regs.gprs[reg2] &= ~0xfffff000UL;
 422                 end = kvm_s390_logical_to_effective(vcpu, end);
 423                 vcpu->run->s.regs.gprs[reg2] |= end;
 424         }
 425         return 0;
 426 }
 427
 428 static int handle_ipte_interlock(struct kvm_vcpu *vcpu)
 429 {
 430         vcpu->stat.instruction_ipte_interlock++;
 431         if (psw_bits(vcpu->arch.sie_block->gpsw).pstate)
 432                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 433         wait_event(vcpu->kvm->arch.ipte_wq, !ipte_lock_held(vcpu));
 434         kvm_s390_retry_instr(vcpu);
 435         VCPU_EVENT(vcpu, 4, "%s", "retrying ipte interlock operation");
 436         return 0;
 437 }
 438
 439 static int handle_test_block(struct kvm_vcpu *vcpu)
 440 {
 441         gpa_t addr;
 442         int reg2;
 443
 444         vcpu->stat.instruction_tb++;
 445
 446         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 447                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 448
 449         kvm_s390_get_regs_rre(vcpu, NULL, &reg2);
 450         addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 451         addr = kvm_s390_logical_to_effective(vcpu, addr);
 452         if (kvm_s390_check_low_addr_prot_real(vcpu, addr))
 453                 return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
 454         addr = kvm_s390_real_to_abs(vcpu, addr);
 455
 456         if (kvm_is_error_gpa(vcpu->kvm, addr))
 457                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 458         /*
 459          * We don't expect errors on modern systems, and do not care
 460          * about storage keys (yet), so let's just clear the page.
 461          */
 462         if (kvm_clear_guest(vcpu->kvm, addr, PAGE_SIZE))
 463                 return -EFAULT;
 464         kvm_s390_set_psw_cc(vcpu, 0);
 465         vcpu->run->s.regs.gprs[0] = 0;
 466         return 0;
 467 }
 468
 469 static int handle_tpi(struct kvm_vcpu *vcpu)
 470 {
 471         struct kvm_s390_interrupt_info *inti;
 472         unsigned long len;
 473         u32 tpi_data[3];
 474         int rc;
 475         u64 addr;
 476         u8 ar;
 477
 478         vcpu->stat.instruction_tpi++;
 479
 480         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 481         if (addr & 3)
 482                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 483
 484         inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0);
 485         if (!inti) {
 486                 kvm_s390_set_psw_cc(vcpu, 0);
 487                 return 0;
 488         }
 489
 490         tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr;
 491         tpi_data[1] = inti->io.io_int_parm;
 492         tpi_data[2] = inti->io.io_int_word;
 493         if (addr) {
 494                 /*
 495                  * Store the two-word I/O interruption code into the
 496                  * provided area.
 497                  */
 498                 len = sizeof(tpi_data) - 4;
 499                 rc = write_guest(vcpu, addr, ar, &tpi_data, len);
 500                 if (rc) {
 501                         rc = kvm_s390_inject_prog_cond(vcpu, rc);
 502                         goto reinject_interrupt;
 503                 }
 504         } else {
 505                 /*
 506                  * Store the three-word I/O interruption code into
 507                  * the appropriate lowcore area.
 508                  */
 509                 len = sizeof(tpi_data);
 510                 if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len)) {
 511                         /* failed writes to the low core are not recoverable */
 512                         rc = -EFAULT;
 513                         goto reinject_interrupt;
 514                 }
 515         }
 516
 517         /* irq was successfully handed to the guest */
 518         kfree(inti);
 519         kvm_s390_set_psw_cc(vcpu, 1);
 520         return 0;
 521 reinject_interrupt:
 522         /*
 523          * If we encounter a problem storing the interruption code, the
 524          * instruction is suppressed from the guest's view: reinject the
 525          * interrupt.
 526          */
 527         if (kvm_s390_reinject_io_int(vcpu->kvm, inti)) {
 528                 kfree(inti);
 529                 rc = -EFAULT;
 530         }
 531         /* don't set the cc, a pgm irq was injected or we drop to user space */
 532         return rc ? -EFAULT : 0;
 533 }
 534
 535 static int handle_tsch(struct kvm_vcpu *vcpu)
 536 {
 537         struct kvm_s390_interrupt_info *inti = NULL;
 538         const u64 isc_mask = 0xffUL << 24; /* all iscs set */
 539
 540         vcpu->stat.instruction_tsch++;
 541
 542         /* a valid schid has at least one bit set */
 543         if (vcpu->run->s.regs.gprs[1])
 544                 inti = kvm_s390_get_io_int(vcpu->kvm, isc_mask,
 545                                            vcpu->run->s.regs.gprs[1]);
 546
 547         /*
 548          * Prepare exit to userspace.
 549          * We indicate whether we dequeued a pending I/O interrupt
 550          * so that userspace can re-inject it if the instruction gets
 551          * a program check. While this may re-order the pending I/O
 552          * interrupts, this is no problem since the priority is kept
 553          * intact.
 554          */
 555         vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
 556         vcpu->run->s390_tsch.dequeued = !!inti;
 557         if (inti) {
 558                 vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
 559                 vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
 560                 vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
 561                 vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
 562         }
 563         vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
 564         kfree(inti);
 565         return -EREMOTE;
 566 }
 567
 568 static int handle_io_inst(struct kvm_vcpu *vcpu)
 569 {
 570         VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
 571
 572         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 573                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 574
 575         if (vcpu->kvm->arch.css_support) {
 576                 /*
 577                  * Most I/O instructions will be handled by userspace.
 578                  * Exceptions are tpi and the interrupt portion of tsch.
 579                  */
 580                 if (vcpu->arch.sie_block->ipa == 0xb236)
 581                         return handle_tpi(vcpu);
 582                 if (vcpu->arch.sie_block->ipa == 0xb235)
 583                         return handle_tsch(vcpu);
 584                 /* Handle in userspace. */
 585                 vcpu->stat.instruction_io_other++;
 586                 return -EOPNOTSUPP;
 587         } else {
 588                 /*
 589                  * Set condition code 3 to stop the guest from issuing channel
 590                  * I/O instructions.
 591                  */
 592                 kvm_s390_set_psw_cc(vcpu, 3);
 593                 return 0;
 594         }
 595 }
 596
 597 static int handle_stfl(struct kvm_vcpu *vcpu)
 598 {
 599         int rc;
 600         unsigned int fac;
 601
 602         vcpu->stat.instruction_stfl++;
 603
 604         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 605                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 606
 607         /*
 608          * We need to shift the lower 32 facility bits (bit 0-31) from a u64
 609          * into a u32 memory representation. They will remain bits 0-31.
 610          */
 611         fac = *vcpu->kvm->arch.model.fac_list >> 32;
 612         rc = write_guest_lc(vcpu, offsetof(struct lowcore, stfl_fac_list),
 613                             &fac, sizeof(fac));
 614         if (rc)
 615                 return rc;
 616         VCPU_EVENT(vcpu, 3, "STFL: store facility list 0x%x", fac);
 617         trace_kvm_s390_handle_stfl(vcpu, fac);
 618         return 0;
 619 }
 620
 621 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
 622 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
 623 #define PSW_ADDR_24 0x0000000000ffffffUL
 624 #define PSW_ADDR_31 0x000000007fffffffUL
 625
 626 int is_valid_psw(psw_t *psw)
 627 {
 628         if (psw->mask & PSW_MASK_UNASSIGNED)
 629                 return 0;
 630         if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
 631                 if (psw->addr & ~PSW_ADDR_31)
 632                         return 0;
 633         }
 634         if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
 635                 return 0;
 636         if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
 637                 return 0;
 638         if (psw->addr & 1)
 639                 return 0;
 640         return 1;
 641 }
 642
 643 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
 644 {
 645         psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
 646         psw_compat_t new_psw;
 647         u64 addr;
 648         int rc;
 649         u8 ar;
 650
 651         vcpu->stat.instruction_lpsw++;
 652
 653         if (gpsw->mask & PSW_MASK_PSTATE)
 654                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 655
 656         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 657         if (addr & 7)
 658                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 659
 660         rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
 661         if (rc)
 662                 return kvm_s390_inject_prog_cond(vcpu, rc);
 663         if (!(new_psw.mask & PSW32_MASK_BASE))
 664                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 665         gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
 666         gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
 667         gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
 668         if (!is_valid_psw(gpsw))
 669                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 670         return 0;
 671 }
 672
 673 static int handle_lpswe(struct kvm_vcpu *vcpu)
 674 {
 675         psw_t new_psw;
 676         u64 addr;
 677         int rc;
 678         u8 ar;
 679
 680         vcpu->stat.instruction_lpswe++;
 681
 682         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 683                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 684
 685         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 686         if (addr & 7)
 687                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 688         rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
 689         if (rc)
 690                 return kvm_s390_inject_prog_cond(vcpu, rc);
 691         vcpu->arch.sie_block->gpsw = new_psw;
 692         if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
 693                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 694         return 0;
 695 }
 696
 697 static int handle_stidp(struct kvm_vcpu *vcpu)
 698 {
 699         u64 stidp_data = vcpu->kvm->arch.model.cpuid;
 700         u64 operand2;
 701         int rc;
 702         u8 ar;
 703
 704         vcpu->stat.instruction_stidp++;
 705
 706         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 707                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 708
 709         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 710
 711         if (operand2 & 7)
 712                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 713
 714         rc = write_guest(vcpu, operand2, ar, &stidp_data, sizeof(stidp_data));
 715         if (rc)
 716                 return kvm_s390_inject_prog_cond(vcpu, rc);
 717
 718         VCPU_EVENT(vcpu, 3, "STIDP: store cpu id 0x%llx", stidp_data);
 719         return 0;
 720 }
 721
 722 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
 723 {
 724         int cpus = 0;
 725         int n;
 726
 727         cpus = atomic_read(&vcpu->kvm->online_vcpus);
 728
 729         /* deal with other level 3 hypervisors */
 730         if (stsi(mem, 3, 2, 2))
 731                 mem->count = 0;
 732         if (mem->count < 8)
 733                 mem->count++;
 734         for (n = mem->count - 1; n > 0 ; n--)
 735                 memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
 736
 737         memset(&mem->vm[0], 0, sizeof(mem->vm[0]));
 738         mem->vm[0].cpus_total = cpus;
 739         mem->vm[0].cpus_configured = cpus;
 740         mem->vm[0].cpus_standby = 0;
 741         mem->vm[0].cpus_reserved = 0;
 742         mem->vm[0].caf = 1000;
 743         memcpy(mem->vm[0].name, "KVMguest", 8);
 744         ASCEBC(mem->vm[0].name, 8);
 745         memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
 746         ASCEBC(mem->vm[0].cpi, 16);
 747 }
 748
 749 static void insert_stsi_usr_data(struct kvm_vcpu *vcpu, u64 addr, u8 ar,
 750                                  u8 fc, u8 sel1, u16 sel2)
 751 {
 752         vcpu->run->exit_reason = KVM_EXIT_S390_STSI;
 753         vcpu->run->s390_stsi.addr = addr;
 754         vcpu->run->s390_stsi.ar = ar;
 755         vcpu->run->s390_stsi.fc = fc;
 756         vcpu->run->s390_stsi.sel1 = sel1;
 757         vcpu->run->s390_stsi.sel2 = sel2;
 758 }
 759
 760 static int handle_stsi(struct kvm_vcpu *vcpu)
 761 {
 762         int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
 763         int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
 764         int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
 765         unsigned long mem = 0;
 766         u64 operand2;
 767         int rc = 0;
 768         u8 ar;
 769
 770         vcpu->stat.instruction_stsi++;
 771         VCPU_EVENT(vcpu, 3, "STSI: fc: %u sel1: %u sel2: %u", fc, sel1, sel2);
 772
 773         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 774                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 775
 776         if (fc > 3) {
 777                 kvm_s390_set_psw_cc(vcpu, 3);
 778                 return 0;
 779         }
 780
 781         if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
 782             || vcpu->run->s.regs.gprs[1] & 0xffff0000)
 783                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 784
 785         if (fc == 0) {
 786                 vcpu->run->s.regs.gprs[0] = 3 << 28;
 787                 kvm_s390_set_psw_cc(vcpu, 0);
 788                 return 0;
 789         }
 790
 791         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 792
 793         if (operand2 & 0xfff)
 794                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 795
 796         switch (fc) {
 797         case 1: /* same handling for 1 and 2 */
 798         case 2:
 799                 mem = get_zeroed_page(GFP_KERNEL);
 800                 if (!mem)
 801                         goto out_no_data;
 802                 if (stsi((void *) mem, fc, sel1, sel2))
 803                         goto out_no_data;
 804                 break;
 805         case 3:
 806                 if (sel1 != 2 || sel2 != 2)
 807                         goto out_no_data;
 808                 mem = get_zeroed_page(GFP_KERNEL);
 809                 if (!mem)
 810                         goto out_no_data;
 811                 handle_stsi_3_2_2(vcpu, (void *) mem);
 812                 break;
 813         }
 814
 815         rc = write_guest(vcpu, operand2, ar, (void *)mem, PAGE_SIZE);
 816         if (rc) {
 817                 rc = kvm_s390_inject_prog_cond(vcpu, rc);
 818                 goto out;
 819         }
 820         if (vcpu->kvm->arch.user_stsi) {
 821                 insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
 822                 rc = -EREMOTE;
 823         }
 824         trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
 825         free_page(mem);
 826         kvm_s390_set_psw_cc(vcpu, 0);
 827         vcpu->run->s.regs.gprs[0] = 0;
 828         return rc;
 829 out_no_data:
 830         kvm_s390_set_psw_cc(vcpu, 3);
 831 out:
 832         free_page(mem);
 833         return rc;
 834 }
 835
 836 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
 837 {
 838         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
 839         case 0x02:
 840                 return handle_stidp(vcpu);
 841         case 0x04:
 842                 return handle_set_clock(vcpu);
 843         case 0x10:
 844                 return handle_set_prefix(vcpu);
 845         case 0x11:
 846                 return handle_store_prefix(vcpu);
 847         case 0x12:
 848                 return handle_store_cpu_address(vcpu);
 849         case 0x14:
 850                 return kvm_s390_handle_vsie(vcpu);
 851         case 0x21:
 852         case 0x50:
 853                 return handle_ipte_interlock(vcpu);
 854         case 0x29:
 855                 return handle_iske(vcpu);
 856         case 0x2a:
 857                 return handle_rrbe(vcpu);
 858         case 0x2b:
 859                 return handle_sske(vcpu);
 860         case 0x2c:
 861                 return handle_test_block(vcpu);
 862         case 0x30:
 863         case 0x31:
 864         case 0x32:
 865         case 0x33:
 866         case 0x34:
 867         case 0x35:
 868         case 0x36:
 869         case 0x37:
 870         case 0x38:
 871         case 0x39:
 872         case 0x3a:
 873         case 0x3b:
 874         case 0x3c:
 875         case 0x5f:
 876         case 0x74:
 877         case 0x76:
 878                 return handle_io_inst(vcpu);
 879         case 0x56:
 880                 return handle_sthyi(vcpu);
 881         case 0x7d:
 882                 return handle_stsi(vcpu);
 883         case 0xb1:
 884                 return handle_stfl(vcpu);
 885         case 0xb2:
 886                 return handle_lpswe(vcpu);
 887         default:
 888                 return -EOPNOTSUPP;
 889         }
 890 }
 891
 892 static int handle_epsw(struct kvm_vcpu *vcpu)
 893 {
 894         int reg1, reg2;
 895
 896         vcpu->stat.instruction_epsw++;
 897
 898         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 899
 900         /* This basically extracts the mask half of the psw. */
 901         vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
 902         vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
 903         if (reg2) {
 904                 vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
 905                 vcpu->run->s.regs.gprs[reg2] |=
 906                         vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
 907         }
 908         return 0;
 909 }
 910
 911 #define PFMF_RESERVED   0xfffc0101UL
 912 #define PFMF_SK         0x00020000UL
 913 #define PFMF_CF         0x00010000UL
 914 #define PFMF_UI         0x00008000UL
 915 #define PFMF_FSC        0x00007000UL
 916 #define PFMF_NQ         0x00000800UL
 917 #define PFMF_MR         0x00000400UL
 918 #define PFMF_MC         0x00000200UL
 919 #define PFMF_KEY        0x000000feUL
 920
 921 static int handle_pfmf(struct kvm_vcpu *vcpu)
 922 {
 923         bool mr = false, mc = false, nq;
 924         int reg1, reg2;
 925         unsigned long start, end;
 926         unsigned char key;
 927
 928         vcpu->stat.instruction_pfmf++;
 929
 930         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 931
 932         if (!test_kvm_facility(vcpu->kvm, 8))
 933                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
 934
 935         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 936                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 937
 938         if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
 939                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 940
 941         /* Only provide non-quiescing support if enabled for the guest */
 942         if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ &&
 943             !test_kvm_facility(vcpu->kvm, 14))
 944                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 945
 946         /* Only provide conditional-SSKE support if enabled for the guest */
 947         if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK &&
 948             test_kvm_facility(vcpu->kvm, 10)) {
 949                 mr = vcpu->run->s.regs.gprs[reg1] & PFMF_MR;
 950                 mc = vcpu->run->s.regs.gprs[reg1] & PFMF_MC;
 951         }
 952
 953         nq = vcpu->run->s.regs.gprs[reg1] & PFMF_NQ;
 954         key = vcpu->run->s.regs.gprs[reg1] & PFMF_KEY;
 955         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 956         start = kvm_s390_logical_to_effective(vcpu, start);
 957
 958         if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
 959                 if (kvm_s390_check_low_addr_prot_real(vcpu, start))
 960                         return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
 961         }
 962
 963         switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
 964         case 0x00000000:
 965                 /* only 4k frames specify a real address */
 966                 start = kvm_s390_real_to_abs(vcpu, start);
 967                 end = (start + PAGE_SIZE) & ~(PAGE_SIZE - 1);
 968                 break;
 969         case 0x00001000:
 970                 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
 971                 break;
 972         case 0x00002000:
 973                 /* only support 2G frame size if EDAT2 is available and we are
 974                    not in 24-bit addressing mode */
 975                 if (!test_kvm_facility(vcpu->kvm, 78) ||
 976                     psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_24BIT)
 977                         return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 978                 end = (start + _REGION3_SIZE) & ~(_REGION3_SIZE - 1);
 979                 break;
 980         default:
 981                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 982         }
 983
 984         while (start != end) {
 985                 unsigned long vmaddr;
 986                 bool unlocked = false;
 987
 988                 /* Translate guest address to host address */
 989                 vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
 990                 if (kvm_is_error_hva(vmaddr))
 991                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 992
 993                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
 994                         if (kvm_clear_guest(vcpu->kvm, start, PAGE_SIZE))
 995                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 996                 }
 997
 998                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
 999                         int rc = kvm_s390_skey_check_enable(vcpu);
1000
1001                         if (rc)
1002                                 return rc;
1003                         down_read(&current->mm->mmap_sem);
1004                         rc = cond_set_guest_storage_key(current->mm, vmaddr,
1005                                                         key, NULL, nq, mr, mc);
1006                         if (rc < 0) {
1007                                 rc = fixup_user_fault(current, current->mm, vmaddr,
1008                                                       FAULT_FLAG_WRITE, &unlocked);
1009                                 rc = !rc ? -EAGAIN : rc;
1010                         }
1011                         up_read(&current->mm->mmap_sem);
1012                         if (rc == -EFAULT)
1013                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1014                         if (rc == -EAGAIN)
1015                                 continue;
1016                         if (rc < 0)
1017                                 return rc;
1018                 }
1019                 start += PAGE_SIZE;
1020         }
1021         if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1022                 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT) {
1023                         vcpu->run->s.regs.gprs[reg2] = end;
1024                 } else {
1025                         vcpu->run->s.regs.gprs[reg2] &= ~0xffffffffUL;
1026                         end = kvm_s390_logical_to_effective(vcpu, end);
1027                         vcpu->run->s.regs.gprs[reg2] |= end;
1028                 }
1029         }
1030         return 0;
1031 }
1032
1033 /*
1034  * Must be called with relevant read locks held (kvm->mm->mmap_sem, kvm->srcu)
1035  */
1036 static inline int __do_essa(struct kvm_vcpu *vcpu, const int orc)
1037 {
1038         int r1, r2, nappended, entries;
1039         unsigned long gfn, hva, res, pgstev, ptev;
1040         unsigned long *cbrlo;
1041
1042         /*
1043          * We don't need to set SD.FPF.SK to 1 here, because if we have a
1044          * machine check here we either handle it or crash
1045          */
1046
1047         kvm_s390_get_regs_rre(vcpu, &r1, &r2);
1048         gfn = vcpu->run->s.regs.gprs[r2] >> PAGE_SHIFT;
1049         hva = gfn_to_hva(vcpu->kvm, gfn);
1050         entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1051
1052         if (kvm_is_error_hva(hva))
1053                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1054
1055         nappended = pgste_perform_essa(vcpu->kvm->mm, hva, orc, &ptev, &pgstev);
1056         if (nappended < 0) {
1057                 res = orc ? 0x10 : 0;
1058                 vcpu->run->s.regs.gprs[r1] = res; /* Exception Indication */
1059                 return 0;
1060         }
1061         res = (pgstev & _PGSTE_GPS_USAGE_MASK) >> 22;
1062         /*
1063          * Set the block-content state part of the result. 0 means resident, so
1064          * nothing to do if the page is valid. 2 is for preserved pages
1065          * (non-present and non-zero), and 3 for zero pages (non-present and
1066          * zero).
1067          */
1068         if (ptev & _PAGE_INVALID) {
1069                 res |= 2;
1070                 if (pgstev & _PGSTE_GPS_ZERO)
1071                         res |= 1;
1072         }
1073         if (pgstev & _PGSTE_GPS_NODAT)
1074                 res |= 0x20;
1075         vcpu->run->s.regs.gprs[r1] = res;
1076         /*
1077          * It is possible that all the normal 511 slots were full, in which case
1078          * we will now write in the 512th slot, which is reserved for host use.
1079          * In both cases we let the normal essa handling code process all the
1080          * slots, including the reserved one, if needed.
1081          */
1082         if (nappended > 0) {
1083                 cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo & PAGE_MASK);
1084                 cbrlo[entries] = gfn << PAGE_SHIFT;
1085         }
1086
1087         if (orc) {
1088                 struct kvm_memory_slot *ms = gfn_to_memslot(vcpu->kvm, gfn);
1089
1090                 /* Increment only if we are really flipping the bit */
1091                 if (ms && !test_and_set_bit(gfn - ms->base_gfn, kvm_second_dirty_bitmap(ms)))
1092                         atomic64_inc(&vcpu->kvm->arch.cmma_dirty_pages);
1093         }
1094
1095         return nappended;
1096 }
1097
1098 static int handle_essa(struct kvm_vcpu *vcpu)
1099 {
1100         /* entries expected to be 1FF */
1101         int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1102         unsigned long *cbrlo;
1103         struct gmap *gmap;
1104         int i, orc;
1105
1106         VCPU_EVENT(vcpu, 4, "ESSA: release %d pages", entries);
1107         gmap = vcpu->arch.gmap;
1108         vcpu->stat.instruction_essa++;
1109         if (!vcpu->kvm->arch.use_cmma)
1110                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1111
1112         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1113                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1114         /* Check for invalid operation request code */
1115         orc = (vcpu->arch.sie_block->ipb & 0xf0000000) >> 28;
1116         /* ORCs 0-6 are always valid */
1117         if (orc > (test_kvm_facility(vcpu->kvm, 147) ? ESSA_SET_STABLE_NODAT
1118                                                 : ESSA_SET_STABLE_IF_RESIDENT))
1119                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1120
1121         if (!vcpu->kvm->arch.migration_mode) {
1122                 /*
1123                  * CMMA is enabled in the KVM settings, but is disabled in
1124                  * the SIE block and in the mm_context, and we are not doing
1125                  * a migration. Enable CMMA in the mm_context.
1126                  * Since we need to take a write lock to write to the context
1127                  * to avoid races with storage keys handling, we check if the
1128                  * value really needs to be written to; if the value is
1129                  * already correct, we do nothing and avoid the lock.
1130                  */
1131                 if (vcpu->kvm->mm->context.uses_cmm == 0) {
1132                         down_write(&vcpu->kvm->mm->mmap_sem);
1133                         vcpu->kvm->mm->context.uses_cmm = 1;
1134                         up_write(&vcpu->kvm->mm->mmap_sem);
1135                 }
1136                 /*
1137                  * If we are here, we are supposed to have CMMA enabled in
1138                  * the SIE block. Enabling CMMA works on a per-CPU basis,
1139                  * while the context use_cmma flag is per process.
1140                  * It's possible that the context flag is enabled and the
1141                  * SIE flag is not, so we set the flag always; if it was
1142                  * already set, nothing changes, otherwise we enable it
1143                  * on this CPU too.
1144                  */
1145                 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
1146                 /* Retry the ESSA instruction */
1147                 kvm_s390_retry_instr(vcpu);
1148         } else {
1149                 int srcu_idx;
1150
1151                 down_read(&vcpu->kvm->mm->mmap_sem);
1152                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1153                 i = __do_essa(vcpu, orc);
1154                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1155                 up_read(&vcpu->kvm->mm->mmap_sem);
1156                 if (i < 0)
1157                         return i;
1158                 /* Account for the possible extra cbrl entry */
1159                 entries += i;
1160         }
1161         vcpu->arch.sie_block->cbrlo &= PAGE_MASK;       /* reset nceo */
1162         cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
1163         down_read(&gmap->mm->mmap_sem);
1164         for (i = 0; i < entries; ++i)
1165                 __gmap_zap(gmap, cbrlo[i]);
1166         up_read(&gmap->mm->mmap_sem);
1167         return 0;
1168 }
1169
1170 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
1171 {
1172         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1173         case 0x8a:
1174         case 0x8e:
1175         case 0x8f:
1176                 return handle_ipte_interlock(vcpu);
1177         case 0x8d:
1178                 return handle_epsw(vcpu);
1179         case 0xab:
1180                 return handle_essa(vcpu);
1181         case 0xaf:
1182                 return handle_pfmf(vcpu);
1183         default:
1184                 return -EOPNOTSUPP;
1185         }
1186 }
1187
1188 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
1189 {
1190         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1191         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1192         int reg, rc, nr_regs;
1193         u32 ctl_array[16];
1194         u64 ga;
1195         u8 ar;
1196
1197         vcpu->stat.instruction_lctl++;
1198
1199         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1200                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1201
1202         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1203
1204         if (ga & 3)
1205                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1206
1207         VCPU_EVENT(vcpu, 4, "LCTL: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1208         trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
1209
1210         nr_regs = ((reg3 - reg1) & 0xf) + 1;
1211         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1212         if (rc)
1213                 return kvm_s390_inject_prog_cond(vcpu, rc);
1214         reg = reg1;
1215         nr_regs = 0;
1216         do {
1217                 vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
1218                 vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
1219                 if (reg == reg3)
1220                         break;
1221                 reg = (reg + 1) % 16;
1222         } while (1);
1223         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1224         return 0;
1225 }
1226
1227 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
1228 {
1229         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1230         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1231         int reg, rc, nr_regs;
1232         u32 ctl_array[16];
1233         u64 ga;
1234         u8 ar;
1235
1236         vcpu->stat.instruction_stctl++;
1237
1238         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1239                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1240
1241         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1242
1243         if (ga & 3)
1244                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1245
1246         VCPU_EVENT(vcpu, 4, "STCTL r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1247         trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
1248
1249         reg = reg1;
1250         nr_regs = 0;
1251         do {
1252                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1253                 if (reg == reg3)
1254                         break;
1255                 reg = (reg + 1) % 16;
1256         } while (1);
1257         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1258         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1259 }
1260
1261 static int handle_lctlg(struct kvm_vcpu *vcpu)
1262 {
1263         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1264         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1265         int reg, rc, nr_regs;
1266         u64 ctl_array[16];
1267         u64 ga;
1268         u8 ar;
1269
1270         vcpu->stat.instruction_lctlg++;
1271
1272         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1273                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1274
1275         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1276
1277         if (ga & 7)
1278                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1279
1280         VCPU_EVENT(vcpu, 4, "LCTLG: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1281         trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
1282
1283         nr_regs = ((reg3 - reg1) & 0xf) + 1;
1284         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1285         if (rc)
1286                 return kvm_s390_inject_prog_cond(vcpu, rc);
1287         reg = reg1;
1288         nr_regs = 0;
1289         do {
1290                 vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
1291                 if (reg == reg3)
1292                         break;
1293                 reg = (reg + 1) % 16;
1294         } while (1);
1295         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1296         return 0;
1297 }
1298
1299 static int handle_stctg(struct kvm_vcpu *vcpu)
1300 {
1301         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1302         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1303         int reg, rc, nr_regs;
1304         u64 ctl_array[16];
1305         u64 ga;
1306         u8 ar;
1307
1308         vcpu->stat.instruction_stctg++;
1309
1310         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1311                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1312
1313         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1314
1315         if (ga & 7)
1316                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1317
1318         VCPU_EVENT(vcpu, 4, "STCTG r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1319         trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
1320
1321         reg = reg1;
1322         nr_regs = 0;
1323         do {
1324                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1325                 if (reg == reg3)
1326                         break;
1327                 reg = (reg + 1) % 16;
1328         } while (1);
1329         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1330         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1331 }
1332
1333 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
1334 {
1335         switch (vcpu->arch.sie_block->ipb & 0x000000ff) {
1336         case 0x25:
1337                 return handle_stctg(vcpu);
1338         case 0x2f:
1339                 return handle_lctlg(vcpu);
1340         case 0x60:
1341         case 0x61:
1342         case 0x62:
1343                 return handle_ri(vcpu);
1344         default:
1345                 return -EOPNOTSUPP;
1346         }
1347 }
1348
1349 static int handle_tprot(struct kvm_vcpu *vcpu)
1350 {
1351         u64 address1, address2;
1352         unsigned long hva, gpa;
1353         int ret = 0, cc = 0;
1354         bool writable;
1355         u8 ar;
1356
1357         vcpu->stat.instruction_tprot++;
1358
1359         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1360                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1361
1362         kvm_s390_get_base_disp_sse(vcpu, &address1, &address2, &ar, NULL);
1363
1364         /* we only handle the Linux memory detection case:
1365          * access key == 0
1366          * everything else goes to userspace. */
1367         if (address2 & 0xf0)
1368                 return -EOPNOTSUPP;
1369         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1370                 ipte_lock(vcpu);
1371         ret = guest_translate_address(vcpu, address1, ar, &gpa, GACC_STORE);
1372         if (ret == PGM_PROTECTION) {
1373                 /* Write protected? Try again with read-only... */
1374                 cc = 1;
1375                 ret = guest_translate_address(vcpu, address1, ar, &gpa,
1376                                               GACC_FETCH);
1377         }
1378         if (ret) {
1379                 if (ret == PGM_ADDRESSING || ret == PGM_TRANSLATION_SPEC) {
1380                         ret = kvm_s390_inject_program_int(vcpu, ret);
1381                 } else if (ret > 0) {
1382                         /* Translation not available */
1383                         kvm_s390_set_psw_cc(vcpu, 3);
1384                         ret = 0;
1385                 }
1386                 goto out_unlock;
1387         }
1388
1389         hva = gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
1390         if (kvm_is_error_hva(hva)) {
1391                 ret = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1392         } else {
1393                 if (!writable)
1394                         cc = 1;         /* Write not permitted ==> read-only */
1395                 kvm_s390_set_psw_cc(vcpu, cc);
1396                 /* Note: CC2 only occurs for storage keys (not supported yet) */
1397         }
1398 out_unlock:
1399         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1400                 ipte_unlock(vcpu);
1401         return ret;
1402 }
1403
1404 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
1405 {
1406         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1407         case 0x01:
1408                 return handle_tprot(vcpu);
1409         default:
1410                 return -EOPNOTSUPP;
1411         }
1412 }
1413
1414 static int handle_sckpf(struct kvm_vcpu *vcpu)
1415 {
1416         u32 value;
1417
1418         vcpu->stat.instruction_sckpf++;
1419
1420         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1421                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1422
1423         if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
1424                 return kvm_s390_inject_program_int(vcpu,
1425                                                    PGM_SPECIFICATION);
1426
1427         value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
1428         vcpu->arch.sie_block->todpr = value;
1429
1430         return 0;
1431 }
1432
1433 static int handle_ptff(struct kvm_vcpu *vcpu)
1434 {
1435         vcpu->stat.instruction_ptff++;
1436
1437         /* we don't emulate any control instructions yet */
1438         kvm_s390_set_psw_cc(vcpu, 3);
1439         return 0;
1440 }
1441
1442 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1443 {
1444         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1445         case 0x04:
1446                 return handle_ptff(vcpu);
1447         case 0x07:
1448                 return handle_sckpf(vcpu);
1449         default:
1450                 return -EOPNOTSUPP;
1451         }
1452 }