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 <asm/ap.h>
  31 #include "gaccess.h"
  32 #include "kvm-s390.h"
  33 #include "trace.h"
  34
  35 static int handle_ri(struct kvm_vcpu *vcpu)
  36 {
  37         vcpu->stat.instruction_ri++;
  38
  39         if (test_kvm_facility(vcpu->kvm, 64)) {
  40                 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (lazy)");
  41                 vcpu->arch.sie_block->ecb3 |= ECB3_RI;
  42                 kvm_s390_retry_instr(vcpu);
  43                 return 0;
  44         } else
  45                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
  46 }
  47
  48 int kvm_s390_handle_aa(struct kvm_vcpu *vcpu)
  49 {
  50         if ((vcpu->arch.sie_block->ipa & 0xf) <= 4)
  51                 return handle_ri(vcpu);
  52         else
  53                 return -EOPNOTSUPP;
  54 }
  55
  56 static int handle_gs(struct kvm_vcpu *vcpu)
  57 {
  58         vcpu->stat.instruction_gs++;
  59
  60         if (test_kvm_facility(vcpu->kvm, 133)) {
  61                 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (lazy)");
  62                 preempt_disable();
  63                 __ctl_set_bit(2, 4);
  64                 current->thread.gs_cb = (struct gs_cb *)&vcpu->run->s.regs.gscb;
  65                 restore_gs_cb(current->thread.gs_cb);
  66                 preempt_enable();
  67                 vcpu->arch.sie_block->ecb |= ECB_GS;
  68                 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
  69                 vcpu->arch.gs_enabled = 1;
  70                 kvm_s390_retry_instr(vcpu);
  71                 return 0;
  72         } else
  73                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
  74 }
  75
  76 int kvm_s390_handle_e3(struct kvm_vcpu *vcpu)
  77 {
  78         int code = vcpu->arch.sie_block->ipb & 0xff;
  79
  80         if (code == 0x49 || code == 0x4d)
  81                 return handle_gs(vcpu);
  82         else
  83                 return -EOPNOTSUPP;
  84 }
  85 /* Handle SCK (SET CLOCK) interception */
  86 static int handle_set_clock(struct kvm_vcpu *vcpu)
  87 {
  88         struct kvm_s390_vm_tod_clock gtod = { 0 };
  89         int rc;
  90         u8 ar;
  91         u64 op2;
  92
  93         vcpu->stat.instruction_sck++;
  94
  95         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
  96                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
  97
  98         op2 = kvm_s390_get_base_disp_s(vcpu, &ar);
  99         if (op2 & 7)    /* Operand must be on a doubleword boundary */
 100                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 101         rc = read_guest(vcpu, op2, ar, &gtod.tod, sizeof(gtod.tod));
 102         if (rc)
 103                 return kvm_s390_inject_prog_cond(vcpu, rc);
 104
 105         VCPU_EVENT(vcpu, 3, "SCK: setting guest TOD to 0x%llx", gtod.tod);
 106         kvm_s390_set_tod_clock(vcpu->kvm, &gtod);
 107
 108         kvm_s390_set_psw_cc(vcpu, 0);
 109         return 0;
 110 }
 111
 112 static int handle_set_prefix(struct kvm_vcpu *vcpu)
 113 {
 114         u64 operand2;
 115         u32 address;
 116         int rc;
 117         u8 ar;
 118
 119         vcpu->stat.instruction_spx++;
 120
 121         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 122                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 123
 124         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 125
 126         /* must be word boundary */
 127         if (operand2 & 3)
 128                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 129
 130         /* get the value */
 131         rc = read_guest(vcpu, operand2, ar, &address, sizeof(address));
 132         if (rc)
 133                 return kvm_s390_inject_prog_cond(vcpu, rc);
 134
 135         address &= 0x7fffe000u;
 136
 137         /*
 138          * Make sure the new value is valid memory. We only need to check the
 139          * first page, since address is 8k aligned and memory pieces are always
 140          * at least 1MB aligned and have at least a size of 1MB.
 141          */
 142         if (kvm_is_error_gpa(vcpu->kvm, address))
 143                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 144
 145         kvm_s390_set_prefix(vcpu, address);
 146         trace_kvm_s390_handle_prefix(vcpu, 1, address);
 147         return 0;
 148 }
 149
 150 static int handle_store_prefix(struct kvm_vcpu *vcpu)
 151 {
 152         u64 operand2;
 153         u32 address;
 154         int rc;
 155         u8 ar;
 156
 157         vcpu->stat.instruction_stpx++;
 158
 159         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 160                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 161
 162         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 163
 164         /* must be word boundary */
 165         if (operand2 & 3)
 166                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 167
 168         address = kvm_s390_get_prefix(vcpu);
 169
 170         /* get the value */
 171         rc = write_guest(vcpu, operand2, ar, &address, sizeof(address));
 172         if (rc)
 173                 return kvm_s390_inject_prog_cond(vcpu, rc);
 174
 175         VCPU_EVENT(vcpu, 3, "STPX: storing prefix 0x%x into 0x%llx", address, operand2);
 176         trace_kvm_s390_handle_prefix(vcpu, 0, address);
 177         return 0;
 178 }
 179
 180 static int handle_store_cpu_address(struct kvm_vcpu *vcpu)
 181 {
 182         u16 vcpu_id = vcpu->vcpu_id;
 183         u64 ga;
 184         int rc;
 185         u8 ar;
 186
 187         vcpu->stat.instruction_stap++;
 188
 189         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 190                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 191
 192         ga = kvm_s390_get_base_disp_s(vcpu, &ar);
 193
 194         if (ga & 1)
 195                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 196
 197         rc = write_guest(vcpu, ga, ar, &vcpu_id, sizeof(vcpu_id));
 198         if (rc)
 199                 return kvm_s390_inject_prog_cond(vcpu, rc);
 200
 201         VCPU_EVENT(vcpu, 3, "STAP: storing cpu address (%u) to 0x%llx", vcpu_id, ga);
 202         trace_kvm_s390_handle_stap(vcpu, ga);
 203         return 0;
 204 }
 205
 206 int kvm_s390_skey_check_enable(struct kvm_vcpu *vcpu)
 207 {
 208         int rc;
 209
 210         trace_kvm_s390_skey_related_inst(vcpu);
 211         /* Already enabled? */
 212         if (vcpu->arch.skey_enabled)
 213                 return 0;
 214
 215         rc = s390_enable_skey();
 216         VCPU_EVENT(vcpu, 3, "enabling storage keys for guest: %d", rc);
 217         if (rc)
 218                 return rc;
 219
 220         if (kvm_s390_test_cpuflags(vcpu, CPUSTAT_KSS))
 221                 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_KSS);
 222         if (!vcpu->kvm->arch.use_skf)
 223                 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
 224         else
 225                 vcpu->arch.sie_block->ictl &= ~(ICTL_ISKE | ICTL_SSKE | ICTL_RRBE);
 226         vcpu->arch.skey_enabled = true;
 227         return 0;
 228 }
 229
 230 static int try_handle_skey(struct kvm_vcpu *vcpu)
 231 {
 232         int rc;
 233
 234         rc = kvm_s390_skey_check_enable(vcpu);
 235         if (rc)
 236                 return rc;
 237         if (vcpu->kvm->arch.use_skf) {
 238                 /* with storage-key facility, SIE interprets it for us */
 239                 kvm_s390_retry_instr(vcpu);
 240                 VCPU_EVENT(vcpu, 4, "%s", "retrying storage key operation");
 241                 return -EAGAIN;
 242         }
 243         return 0;
 244 }
 245
 246 static int handle_iske(struct kvm_vcpu *vcpu)
 247 {
 248         unsigned long gaddr, vmaddr;
 249         unsigned char key;
 250         int reg1, reg2;
 251         bool unlocked;
 252         int rc;
 253
 254         vcpu->stat.instruction_iske++;
 255
 256         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 257                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 258
 259         rc = try_handle_skey(vcpu);
 260         if (rc)
 261                 return rc != -EAGAIN ? rc : 0;
 262
 263         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 264
 265         gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 266         gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
 267         gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
 268         vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr));
 269         if (kvm_is_error_hva(vmaddr))
 270                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 271 retry:
 272         unlocked = false;
 273         down_read(&current->mm->mmap_sem);
 274         rc = get_guest_storage_key(current->mm, vmaddr, &key);
 275
 276         if (rc) {
 277                 rc = fixup_user_fault(current, current->mm, vmaddr,
 278                                       FAULT_FLAG_WRITE, &unlocked);
 279                 if (!rc) {
 280                         up_read(&current->mm->mmap_sem);
 281                         goto retry;
 282                 }
 283         }
 284         up_read(&current->mm->mmap_sem);
 285         if (rc == -EFAULT)
 286                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 287         if (rc < 0)
 288                 return rc;
 289         vcpu->run->s.regs.gprs[reg1] &= ~0xff;
 290         vcpu->run->s.regs.gprs[reg1] |= key;
 291         return 0;
 292 }
 293
 294 static int handle_rrbe(struct kvm_vcpu *vcpu)
 295 {
 296         unsigned long vmaddr, gaddr;
 297         int reg1, reg2;
 298         bool unlocked;
 299         int rc;
 300
 301         vcpu->stat.instruction_rrbe++;
 302
 303         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 304                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 305
 306         rc = try_handle_skey(vcpu);
 307         if (rc)
 308                 return rc != -EAGAIN ? rc : 0;
 309
 310         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 311
 312         gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 313         gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
 314         gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
 315         vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr));
 316         if (kvm_is_error_hva(vmaddr))
 317                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 318 retry:
 319         unlocked = false;
 320         down_read(&current->mm->mmap_sem);
 321         rc = reset_guest_reference_bit(current->mm, vmaddr);
 322         if (rc < 0) {
 323                 rc = fixup_user_fault(current, current->mm, vmaddr,
 324                                       FAULT_FLAG_WRITE, &unlocked);
 325                 if (!rc) {
 326                         up_read(&current->mm->mmap_sem);
 327                         goto retry;
 328                 }
 329         }
 330         up_read(&current->mm->mmap_sem);
 331         if (rc == -EFAULT)
 332                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 333         if (rc < 0)
 334                 return rc;
 335         kvm_s390_set_psw_cc(vcpu, rc);
 336         return 0;
 337 }
 338
 339 #define SSKE_NQ 0x8
 340 #define SSKE_MR 0x4
 341 #define SSKE_MC 0x2
 342 #define SSKE_MB 0x1
 343 static int handle_sske(struct kvm_vcpu *vcpu)
 344 {
 345         unsigned char m3 = vcpu->arch.sie_block->ipb >> 28;
 346         unsigned long start, end;
 347         unsigned char key, oldkey;
 348         int reg1, reg2;
 349         bool unlocked;
 350         int rc;
 351
 352         vcpu->stat.instruction_sske++;
 353
 354         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 355                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 356
 357         rc = try_handle_skey(vcpu);
 358         if (rc)
 359                 return rc != -EAGAIN ? rc : 0;
 360
 361         if (!test_kvm_facility(vcpu->kvm, 8))
 362                 m3 &= ~SSKE_MB;
 363         if (!test_kvm_facility(vcpu->kvm, 10))
 364                 m3 &= ~(SSKE_MC | SSKE_MR);
 365         if (!test_kvm_facility(vcpu->kvm, 14))
 366                 m3 &= ~SSKE_NQ;
 367
 368         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 369
 370         key = vcpu->run->s.regs.gprs[reg1] & 0xfe;
 371         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 372         start = kvm_s390_logical_to_effective(vcpu, start);
 373         if (m3 & SSKE_MB) {
 374                 /* start already designates an absolute address */
 375                 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
 376         } else {
 377                 start = kvm_s390_real_to_abs(vcpu, start);
 378                 end = start + PAGE_SIZE;
 379         }
 380
 381         while (start != end) {
 382                 unsigned long vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
 383                 unlocked = false;
 384
 385                 if (kvm_is_error_hva(vmaddr))
 386                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 387
 388                 down_read(&current->mm->mmap_sem);
 389                 rc = cond_set_guest_storage_key(current->mm, vmaddr, key, &oldkey,
 390                                                 m3 & SSKE_NQ, m3 & SSKE_MR,
 391                                                 m3 & SSKE_MC);
 392
 393                 if (rc < 0) {
 394                         rc = fixup_user_fault(current, current->mm, vmaddr,
 395                                               FAULT_FLAG_WRITE, &unlocked);
 396                         rc = !rc ? -EAGAIN : rc;
 397                 }
 398                 up_read(&current->mm->mmap_sem);
 399                 if (rc == -EFAULT)
 400                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 401                 if (rc == -EAGAIN)
 402                         continue;
 403                 if (rc < 0)
 404                         return rc;
 405                 start += PAGE_SIZE;
 406         }
 407
 408         if (m3 & (SSKE_MC | SSKE_MR)) {
 409                 if (m3 & SSKE_MB) {
 410                         /* skey in reg1 is unpredictable */
 411                         kvm_s390_set_psw_cc(vcpu, 3);
 412                 } else {
 413                         kvm_s390_set_psw_cc(vcpu, rc);
 414                         vcpu->run->s.regs.gprs[reg1] &= ~0xff00UL;
 415                         vcpu->run->s.regs.gprs[reg1] |= (u64) oldkey << 8;
 416                 }
 417         }
 418         if (m3 & SSKE_MB) {
 419                 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT)
 420                         vcpu->run->s.regs.gprs[reg2] &= ~PAGE_MASK;
 421                 else
 422                         vcpu->run->s.regs.gprs[reg2] &= ~0xfffff000UL;
 423                 end = kvm_s390_logical_to_effective(vcpu, end);
 424                 vcpu->run->s.regs.gprs[reg2] |= end;
 425         }
 426         return 0;
 427 }
 428
 429 static int handle_ipte_interlock(struct kvm_vcpu *vcpu)
 430 {
 431         vcpu->stat.instruction_ipte_interlock++;
 432         if (psw_bits(vcpu->arch.sie_block->gpsw).pstate)
 433                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 434         wait_event(vcpu->kvm->arch.ipte_wq, !ipte_lock_held(vcpu));
 435         kvm_s390_retry_instr(vcpu);
 436         VCPU_EVENT(vcpu, 4, "%s", "retrying ipte interlock operation");
 437         return 0;
 438 }
 439
 440 static int handle_test_block(struct kvm_vcpu *vcpu)
 441 {
 442         gpa_t addr;
 443         int reg2;
 444
 445         vcpu->stat.instruction_tb++;
 446
 447         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 448                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 449
 450         kvm_s390_get_regs_rre(vcpu, NULL, &reg2);
 451         addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 452         addr = kvm_s390_logical_to_effective(vcpu, addr);
 453         if (kvm_s390_check_low_addr_prot_real(vcpu, addr))
 454                 return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
 455         addr = kvm_s390_real_to_abs(vcpu, addr);
 456
 457         if (kvm_is_error_gpa(vcpu->kvm, addr))
 458                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 459         /*
 460          * We don't expect errors on modern systems, and do not care
 461          * about storage keys (yet), so let's just clear the page.
 462          */
 463         if (kvm_clear_guest(vcpu->kvm, addr, PAGE_SIZE))
 464                 return -EFAULT;
 465         kvm_s390_set_psw_cc(vcpu, 0);
 466         vcpu->run->s.regs.gprs[0] = 0;
 467         return 0;
 468 }
 469
 470 static int handle_tpi(struct kvm_vcpu *vcpu)
 471 {
 472         struct kvm_s390_interrupt_info *inti;
 473         unsigned long len;
 474         u32 tpi_data[3];
 475         int rc;
 476         u64 addr;
 477         u8 ar;
 478
 479         vcpu->stat.instruction_tpi++;
 480
 481         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 482         if (addr & 3)
 483                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 484
 485         inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0);
 486         if (!inti) {
 487                 kvm_s390_set_psw_cc(vcpu, 0);
 488                 return 0;
 489         }
 490
 491         tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr;
 492         tpi_data[1] = inti->io.io_int_parm;
 493         tpi_data[2] = inti->io.io_int_word;
 494         if (addr) {
 495                 /*
 496                  * Store the two-word I/O interruption code into the
 497                  * provided area.
 498                  */
 499                 len = sizeof(tpi_data) - 4;
 500                 rc = write_guest(vcpu, addr, ar, &tpi_data, len);
 501                 if (rc) {
 502                         rc = kvm_s390_inject_prog_cond(vcpu, rc);
 503                         goto reinject_interrupt;
 504                 }
 505         } else {
 506                 /*
 507                  * Store the three-word I/O interruption code into
 508                  * the appropriate lowcore area.
 509                  */
 510                 len = sizeof(tpi_data);
 511                 if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len)) {
 512                         /* failed writes to the low core are not recoverable */
 513                         rc = -EFAULT;
 514                         goto reinject_interrupt;
 515                 }
 516         }
 517
 518         /* irq was successfully handed to the guest */
 519         kfree(inti);
 520         kvm_s390_set_psw_cc(vcpu, 1);
 521         return 0;
 522 reinject_interrupt:
 523         /*
 524          * If we encounter a problem storing the interruption code, the
 525          * instruction is suppressed from the guest's view: reinject the
 526          * interrupt.
 527          */
 528         if (kvm_s390_reinject_io_int(vcpu->kvm, inti)) {
 529                 kfree(inti);
 530                 rc = -EFAULT;
 531         }
 532         /* don't set the cc, a pgm irq was injected or we drop to user space */
 533         return rc ? -EFAULT : 0;
 534 }
 535
 536 static int handle_tsch(struct kvm_vcpu *vcpu)
 537 {
 538         struct kvm_s390_interrupt_info *inti = NULL;
 539         const u64 isc_mask = 0xffUL << 24; /* all iscs set */
 540
 541         vcpu->stat.instruction_tsch++;
 542
 543         /* a valid schid has at least one bit set */
 544         if (vcpu->run->s.regs.gprs[1])
 545                 inti = kvm_s390_get_io_int(vcpu->kvm, isc_mask,
 546                                            vcpu->run->s.regs.gprs[1]);
 547
 548         /*
 549          * Prepare exit to userspace.
 550          * We indicate whether we dequeued a pending I/O interrupt
 551          * so that userspace can re-inject it if the instruction gets
 552          * a program check. While this may re-order the pending I/O
 553          * interrupts, this is no problem since the priority is kept
 554          * intact.
 555          */
 556         vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
 557         vcpu->run->s390_tsch.dequeued = !!inti;
 558         if (inti) {
 559                 vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
 560                 vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
 561                 vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
 562                 vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
 563         }
 564         vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
 565         kfree(inti);
 566         return -EREMOTE;
 567 }
 568
 569 static int handle_io_inst(struct kvm_vcpu *vcpu)
 570 {
 571         VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
 572
 573         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 574                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 575
 576         if (vcpu->kvm->arch.css_support) {
 577                 /*
 578                  * Most I/O instructions will be handled by userspace.
 579                  * Exceptions are tpi and the interrupt portion of tsch.
 580                  */
 581                 if (vcpu->arch.sie_block->ipa == 0xb236)
 582                         return handle_tpi(vcpu);
 583                 if (vcpu->arch.sie_block->ipa == 0xb235)
 584                         return handle_tsch(vcpu);
 585                 /* Handle in userspace. */
 586                 vcpu->stat.instruction_io_other++;
 587                 return -EOPNOTSUPP;
 588         } else {
 589                 /*
 590                  * Set condition code 3 to stop the guest from issuing channel
 591                  * I/O instructions.
 592                  */
 593                 kvm_s390_set_psw_cc(vcpu, 3);
 594                 return 0;
 595         }
 596 }
 597
 598 /*
 599  * handle_pqap: Handling pqap interception
 600  * @vcpu: the vcpu having issue the pqap instruction
 601  *
 602  * We now support PQAP/AQIC instructions and we need to correctly
 603  * answer the guest even if no dedicated driver's hook is available.
 604  *
 605  * The intercepting code calls a dedicated callback for this instruction
 606  * if a driver did register one in the CRYPTO satellite of the
 607  * SIE block.
 608  *
 609  * If no callback is available, the queues are not available, return this
 610  * response code to the caller and set CC to 3.
 611  * Else return the response code returned by the callback.
 612  */
 613 static int handle_pqap(struct kvm_vcpu *vcpu)
 614 {
 615         struct ap_queue_status status = {};
 616         unsigned long reg0;
 617         int ret;
 618         uint8_t fc;
 619
 620         /* Verify that the AP instruction are available */
 621         if (!ap_instructions_available())
 622                 return -EOPNOTSUPP;
 623         /* Verify that the guest is allowed to use AP instructions */
 624         if (!(vcpu->arch.sie_block->eca & ECA_APIE))
 625                 return -EOPNOTSUPP;
 626         /*
 627          * The only possibly intercepted functions when AP instructions are
 628          * available for the guest are AQIC and TAPQ with the t bit set
 629          * since we do not set IC.3 (FIII) we currently will only intercept
 630          * the AQIC function code.
 631          * Note: running nested under z/VM can result in intercepts for other
 632          * function codes, e.g. PQAP(QCI). We do not support this and bail out.
 633          */
 634         reg0 = vcpu->run->s.regs.gprs[0];
 635         fc = (reg0 >> 24) & 0xff;
 636         if (fc != 0x03)
 637                 return -EOPNOTSUPP;
 638
 639         /* PQAP instruction is allowed for guest kernel only */
 640         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 641                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 642
 643         /* Common PQAP instruction specification exceptions */
 644         /* bits 41-47 must all be zeros */
 645         if (reg0 & 0x007f0000UL)
 646                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 647         /* APFT not install and T bit set */
 648         if (!test_kvm_facility(vcpu->kvm, 15) && (reg0 & 0x00800000UL))
 649                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 650         /* APXA not installed and APID greater 64 or APQI greater 16 */
 651         if (!(vcpu->kvm->arch.crypto.crycbd & 0x02) && (reg0 & 0x0000c0f0UL))
 652                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 653
 654         /* AQIC function code specific exception */
 655         /* facility 65 not present for AQIC function code */
 656         if (!test_kvm_facility(vcpu->kvm, 65))
 657                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 658
 659         /*
 660          * Verify that the hook callback is registered, lock the owner
 661          * and call the hook.
 662          */
 663         if (vcpu->kvm->arch.crypto.pqap_hook) {
 664                 if (!try_module_get(vcpu->kvm->arch.crypto.pqap_hook->owner))
 665                         return -EOPNOTSUPP;
 666                 ret = vcpu->kvm->arch.crypto.pqap_hook->hook(vcpu);
 667                 module_put(vcpu->kvm->arch.crypto.pqap_hook->owner);
 668                 if (!ret && vcpu->run->s.regs.gprs[1] & 0x00ff0000)
 669                         kvm_s390_set_psw_cc(vcpu, 3);
 670                 return ret;
 671         }
 672         /*
 673          * A vfio_driver must register a hook.
 674          * No hook means no driver to enable the SIE CRYCB and no queues.
 675          * We send this response to the guest.
 676          */
 677         status.response_code = 0x01;
 678         memcpy(&vcpu->run->s.regs.gprs[1], &status, sizeof(status));
 679         kvm_s390_set_psw_cc(vcpu, 3);
 680         return 0;
 681 }
 682
 683 static int handle_stfl(struct kvm_vcpu *vcpu)
 684 {
 685         int rc;
 686         unsigned int fac;
 687
 688         vcpu->stat.instruction_stfl++;
 689
 690         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 691                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 692
 693         /*
 694          * We need to shift the lower 32 facility bits (bit 0-31) from a u64
 695          * into a u32 memory representation. They will remain bits 0-31.
 696          */
 697         fac = *vcpu->kvm->arch.model.fac_list >> 32;
 698         rc = write_guest_lc(vcpu, offsetof(struct lowcore, stfl_fac_list),
 699                             &fac, sizeof(fac));
 700         if (rc)
 701                 return rc;
 702         VCPU_EVENT(vcpu, 3, "STFL: store facility list 0x%x", fac);
 703         trace_kvm_s390_handle_stfl(vcpu, fac);
 704         return 0;
 705 }
 706
 707 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
 708 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
 709 #define PSW_ADDR_24 0x0000000000ffffffUL
 710 #define PSW_ADDR_31 0x000000007fffffffUL
 711
 712 int is_valid_psw(psw_t *psw)
 713 {
 714         if (psw->mask & PSW_MASK_UNASSIGNED)
 715                 return 0;
 716         if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
 717                 if (psw->addr & ~PSW_ADDR_31)
 718                         return 0;
 719         }
 720         if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
 721                 return 0;
 722         if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
 723                 return 0;
 724         if (psw->addr & 1)
 725                 return 0;
 726         return 1;
 727 }
 728
 729 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
 730 {
 731         psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
 732         psw_compat_t new_psw;
 733         u64 addr;
 734         int rc;
 735         u8 ar;
 736
 737         vcpu->stat.instruction_lpsw++;
 738
 739         if (gpsw->mask & PSW_MASK_PSTATE)
 740                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 741
 742         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 743         if (addr & 7)
 744                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 745
 746         rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
 747         if (rc)
 748                 return kvm_s390_inject_prog_cond(vcpu, rc);
 749         if (!(new_psw.mask & PSW32_MASK_BASE))
 750                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 751         gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
 752         gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
 753         gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
 754         if (!is_valid_psw(gpsw))
 755                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 756         return 0;
 757 }
 758
 759 static int handle_lpswe(struct kvm_vcpu *vcpu)
 760 {
 761         psw_t new_psw;
 762         u64 addr;
 763         int rc;
 764         u8 ar;
 765
 766         vcpu->stat.instruction_lpswe++;
 767
 768         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 769                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 770
 771         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 772         if (addr & 7)
 773                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 774         rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
 775         if (rc)
 776                 return kvm_s390_inject_prog_cond(vcpu, rc);
 777         vcpu->arch.sie_block->gpsw = new_psw;
 778         if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
 779                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 780         return 0;
 781 }
 782
 783 static int handle_stidp(struct kvm_vcpu *vcpu)
 784 {
 785         u64 stidp_data = vcpu->kvm->arch.model.cpuid;
 786         u64 operand2;
 787         int rc;
 788         u8 ar;
 789
 790         vcpu->stat.instruction_stidp++;
 791
 792         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 793                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 794
 795         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 796
 797         if (operand2 & 7)
 798                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 799
 800         rc = write_guest(vcpu, operand2, ar, &stidp_data, sizeof(stidp_data));
 801         if (rc)
 802                 return kvm_s390_inject_prog_cond(vcpu, rc);
 803
 804         VCPU_EVENT(vcpu, 3, "STIDP: store cpu id 0x%llx", stidp_data);
 805         return 0;
 806 }
 807
 808 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
 809 {
 810         int cpus = 0;
 811         int n;
 812
 813         cpus = atomic_read(&vcpu->kvm->online_vcpus);
 814
 815         /* deal with other level 3 hypervisors */
 816         if (stsi(mem, 3, 2, 2))
 817                 mem->count = 0;
 818         if (mem->count < 8)
 819                 mem->count++;
 820         for (n = mem->count - 1; n > 0 ; n--)
 821                 memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
 822
 823         memset(&mem->vm[0], 0, sizeof(mem->vm[0]));
 824         mem->vm[0].cpus_total = cpus;
 825         mem->vm[0].cpus_configured = cpus;
 826         mem->vm[0].cpus_standby = 0;
 827         mem->vm[0].cpus_reserved = 0;
 828         mem->vm[0].caf = 1000;
 829         memcpy(mem->vm[0].name, "KVMguest", 8);
 830         ASCEBC(mem->vm[0].name, 8);
 831         memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
 832         ASCEBC(mem->vm[0].cpi, 16);
 833 }
 834
 835 static void insert_stsi_usr_data(struct kvm_vcpu *vcpu, u64 addr, u8 ar,
 836                                  u8 fc, u8 sel1, u16 sel2)
 837 {
 838         vcpu->run->exit_reason = KVM_EXIT_S390_STSI;
 839         vcpu->run->s390_stsi.addr = addr;
 840         vcpu->run->s390_stsi.ar = ar;
 841         vcpu->run->s390_stsi.fc = fc;
 842         vcpu->run->s390_stsi.sel1 = sel1;
 843         vcpu->run->s390_stsi.sel2 = sel2;
 844 }
 845
 846 static int handle_stsi(struct kvm_vcpu *vcpu)
 847 {
 848         int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
 849         int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
 850         int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
 851         unsigned long mem = 0;
 852         u64 operand2;
 853         int rc = 0;
 854         u8 ar;
 855
 856         vcpu->stat.instruction_stsi++;
 857         VCPU_EVENT(vcpu, 3, "STSI: fc: %u sel1: %u sel2: %u", fc, sel1, sel2);
 858
 859         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 860                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 861
 862         if (fc > 3) {
 863                 kvm_s390_set_psw_cc(vcpu, 3);
 864                 return 0;
 865         }
 866
 867         if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
 868             || vcpu->run->s.regs.gprs[1] & 0xffff0000)
 869                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 870
 871         if (fc == 0) {
 872                 vcpu->run->s.regs.gprs[0] = 3 << 28;
 873                 kvm_s390_set_psw_cc(vcpu, 0);
 874                 return 0;
 875         }
 876
 877         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 878
 879         if (operand2 & 0xfff)
 880                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 881
 882         switch (fc) {
 883         case 1: /* same handling for 1 and 2 */
 884         case 2:
 885                 mem = get_zeroed_page(GFP_KERNEL);
 886                 if (!mem)
 887                         goto out_no_data;
 888                 if (stsi((void *) mem, fc, sel1, sel2))
 889                         goto out_no_data;
 890                 break;
 891         case 3:
 892                 if (sel1 != 2 || sel2 != 2)
 893                         goto out_no_data;
 894                 mem = get_zeroed_page(GFP_KERNEL);
 895                 if (!mem)
 896                         goto out_no_data;
 897                 handle_stsi_3_2_2(vcpu, (void *) mem);
 898                 break;
 899         }
 900
 901         rc = write_guest(vcpu, operand2, ar, (void *)mem, PAGE_SIZE);
 902         if (rc) {
 903                 rc = kvm_s390_inject_prog_cond(vcpu, rc);
 904                 goto out;
 905         }
 906         if (vcpu->kvm->arch.user_stsi) {
 907                 insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
 908                 rc = -EREMOTE;
 909         }
 910         trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
 911         free_page(mem);
 912         kvm_s390_set_psw_cc(vcpu, 0);
 913         vcpu->run->s.regs.gprs[0] = 0;
 914         return rc;
 915 out_no_data:
 916         kvm_s390_set_psw_cc(vcpu, 3);
 917 out:
 918         free_page(mem);
 919         return rc;
 920 }
 921
 922 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
 923 {
 924         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
 925         case 0x02:
 926                 return handle_stidp(vcpu);
 927         case 0x04:
 928                 return handle_set_clock(vcpu);
 929         case 0x10:
 930                 return handle_set_prefix(vcpu);
 931         case 0x11:
 932                 return handle_store_prefix(vcpu);
 933         case 0x12:
 934                 return handle_store_cpu_address(vcpu);
 935         case 0x14:
 936                 return kvm_s390_handle_vsie(vcpu);
 937         case 0x21:
 938         case 0x50:
 939                 return handle_ipte_interlock(vcpu);
 940         case 0x29:
 941                 return handle_iske(vcpu);
 942         case 0x2a:
 943                 return handle_rrbe(vcpu);
 944         case 0x2b:
 945                 return handle_sske(vcpu);
 946         case 0x2c:
 947                 return handle_test_block(vcpu);
 948         case 0x30:
 949         case 0x31:
 950         case 0x32:
 951         case 0x33:
 952         case 0x34:
 953         case 0x35:
 954         case 0x36:
 955         case 0x37:
 956         case 0x38:
 957         case 0x39:
 958         case 0x3a:
 959         case 0x3b:
 960         case 0x3c:
 961         case 0x5f:
 962         case 0x74:
 963         case 0x76:
 964                 return handle_io_inst(vcpu);
 965         case 0x56:
 966                 return handle_sthyi(vcpu);
 967         case 0x7d:
 968                 return handle_stsi(vcpu);
 969         case 0xaf:
 970                 return handle_pqap(vcpu);
 971         case 0xb1:
 972                 return handle_stfl(vcpu);
 973         case 0xb2:
 974                 return handle_lpswe(vcpu);
 975         default:
 976                 return -EOPNOTSUPP;
 977         }
 978 }
 979
 980 static int handle_epsw(struct kvm_vcpu *vcpu)
 981 {
 982         int reg1, reg2;
 983
 984         vcpu->stat.instruction_epsw++;
 985
 986         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 987
 988         /* This basically extracts the mask half of the psw. */
 989         vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
 990         vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
 991         if (reg2) {
 992                 vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
 993                 vcpu->run->s.regs.gprs[reg2] |=
 994                         vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
 995         }
 996         return 0;
 997 }
 998
 999 #define PFMF_RESERVED   0xfffc0101UL
1000 #define PFMF_SK         0x00020000UL
1001 #define PFMF_CF         0x00010000UL
1002 #define PFMF_UI         0x00008000UL
1003 #define PFMF_FSC        0x00007000UL
1004 #define PFMF_NQ         0x00000800UL
1005 #define PFMF_MR         0x00000400UL
1006 #define PFMF_MC         0x00000200UL
1007 #define PFMF_KEY        0x000000feUL
1008
1009 static int handle_pfmf(struct kvm_vcpu *vcpu)
1010 {
1011         bool mr = false, mc = false, nq;
1012         int reg1, reg2;
1013         unsigned long start, end;
1014         unsigned char key;
1015
1016         vcpu->stat.instruction_pfmf++;
1017
1018         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
1019
1020         if (!test_kvm_facility(vcpu->kvm, 8))
1021                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1022
1023         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1024                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1025
1026         if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
1027                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1028
1029         /* Only provide non-quiescing support if enabled for the guest */
1030         if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ &&
1031             !test_kvm_facility(vcpu->kvm, 14))
1032                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1033
1034         /* Only provide conditional-SSKE support if enabled for the guest */
1035         if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK &&
1036             test_kvm_facility(vcpu->kvm, 10)) {
1037                 mr = vcpu->run->s.regs.gprs[reg1] & PFMF_MR;
1038                 mc = vcpu->run->s.regs.gprs[reg1] & PFMF_MC;
1039         }
1040
1041         nq = vcpu->run->s.regs.gprs[reg1] & PFMF_NQ;
1042         key = vcpu->run->s.regs.gprs[reg1] & PFMF_KEY;
1043         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
1044         start = kvm_s390_logical_to_effective(vcpu, start);
1045
1046         if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1047                 if (kvm_s390_check_low_addr_prot_real(vcpu, start))
1048                         return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
1049         }
1050
1051         switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1052         case 0x00000000:
1053                 /* only 4k frames specify a real address */
1054                 start = kvm_s390_real_to_abs(vcpu, start);
1055                 end = (start + PAGE_SIZE) & ~(PAGE_SIZE - 1);
1056                 break;
1057         case 0x00001000:
1058                 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
1059                 break;
1060         case 0x00002000:
1061                 /* only support 2G frame size if EDAT2 is available and we are
1062                    not in 24-bit addressing mode */
1063                 if (!test_kvm_facility(vcpu->kvm, 78) ||
1064                     psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_24BIT)
1065                         return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1066                 end = (start + _REGION3_SIZE) & ~(_REGION3_SIZE - 1);
1067                 break;
1068         default:
1069                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1070         }
1071
1072         while (start != end) {
1073                 unsigned long vmaddr;
1074                 bool unlocked = false;
1075
1076                 /* Translate guest address to host address */
1077                 vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
1078                 if (kvm_is_error_hva(vmaddr))
1079                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1080
1081                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1082                         if (kvm_clear_guest(vcpu->kvm, start, PAGE_SIZE))
1083                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1084                 }
1085
1086                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
1087                         int rc = kvm_s390_skey_check_enable(vcpu);
1088
1089                         if (rc)
1090                                 return rc;
1091                         down_read(&current->mm->mmap_sem);
1092                         rc = cond_set_guest_storage_key(current->mm, vmaddr,
1093                                                         key, NULL, nq, mr, mc);
1094                         if (rc < 0) {
1095                                 rc = fixup_user_fault(current, current->mm, vmaddr,
1096                                                       FAULT_FLAG_WRITE, &unlocked);
1097                                 rc = !rc ? -EAGAIN : rc;
1098                         }
1099                         up_read(&current->mm->mmap_sem);
1100                         if (rc == -EFAULT)
1101                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1102                         if (rc == -EAGAIN)
1103                                 continue;
1104                         if (rc < 0)
1105                                 return rc;
1106                 }
1107                 start += PAGE_SIZE;
1108         }
1109         if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1110                 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT) {
1111                         vcpu->run->s.regs.gprs[reg2] = end;
1112                 } else {
1113                         vcpu->run->s.regs.gprs[reg2] &= ~0xffffffffUL;
1114                         end = kvm_s390_logical_to_effective(vcpu, end);
1115                         vcpu->run->s.regs.gprs[reg2] |= end;
1116                 }
1117         }
1118         return 0;
1119 }
1120
1121 /*
1122  * Must be called with relevant read locks held (kvm->mm->mmap_sem, kvm->srcu)
1123  */
1124 static inline int __do_essa(struct kvm_vcpu *vcpu, const int orc)
1125 {
1126         int r1, r2, nappended, entries;
1127         unsigned long gfn, hva, res, pgstev, ptev;
1128         unsigned long *cbrlo;
1129
1130         /*
1131          * We don't need to set SD.FPF.SK to 1 here, because if we have a
1132          * machine check here we either handle it or crash
1133          */
1134
1135         kvm_s390_get_regs_rre(vcpu, &r1, &r2);
1136         gfn = vcpu->run->s.regs.gprs[r2] >> PAGE_SHIFT;
1137         hva = gfn_to_hva(vcpu->kvm, gfn);
1138         entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1139
1140         if (kvm_is_error_hva(hva))
1141                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1142
1143         nappended = pgste_perform_essa(vcpu->kvm->mm, hva, orc, &ptev, &pgstev);
1144         if (nappended < 0) {
1145                 res = orc ? 0x10 : 0;
1146                 vcpu->run->s.regs.gprs[r1] = res; /* Exception Indication */
1147                 return 0;
1148         }
1149         res = (pgstev & _PGSTE_GPS_USAGE_MASK) >> 22;
1150         /*
1151          * Set the block-content state part of the result. 0 means resident, so
1152          * nothing to do if the page is valid. 2 is for preserved pages
1153          * (non-present and non-zero), and 3 for zero pages (non-present and
1154          * zero).
1155          */
1156         if (ptev & _PAGE_INVALID) {
1157                 res |= 2;
1158                 if (pgstev & _PGSTE_GPS_ZERO)
1159                         res |= 1;
1160         }
1161         if (pgstev & _PGSTE_GPS_NODAT)
1162                 res |= 0x20;
1163         vcpu->run->s.regs.gprs[r1] = res;
1164         /*
1165          * It is possible that all the normal 511 slots were full, in which case
1166          * we will now write in the 512th slot, which is reserved for host use.
1167          * In both cases we let the normal essa handling code process all the
1168          * slots, including the reserved one, if needed.
1169          */
1170         if (nappended > 0) {
1171                 cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo & PAGE_MASK);
1172                 cbrlo[entries] = gfn << PAGE_SHIFT;
1173         }
1174
1175         if (orc) {
1176                 struct kvm_memory_slot *ms = gfn_to_memslot(vcpu->kvm, gfn);
1177
1178                 /* Increment only if we are really flipping the bit */
1179                 if (ms && !test_and_set_bit(gfn - ms->base_gfn, kvm_second_dirty_bitmap(ms)))
1180                         atomic64_inc(&vcpu->kvm->arch.cmma_dirty_pages);
1181         }
1182
1183         return nappended;
1184 }
1185
1186 static int handle_essa(struct kvm_vcpu *vcpu)
1187 {
1188         /* entries expected to be 1FF */
1189         int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1190         unsigned long *cbrlo;
1191         struct gmap *gmap;
1192         int i, orc;
1193
1194         VCPU_EVENT(vcpu, 4, "ESSA: release %d pages", entries);
1195         gmap = vcpu->arch.gmap;
1196         vcpu->stat.instruction_essa++;
1197         if (!vcpu->kvm->arch.use_cmma)
1198                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1199
1200         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1201                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1202         /* Check for invalid operation request code */
1203         orc = (vcpu->arch.sie_block->ipb & 0xf0000000) >> 28;
1204         /* ORCs 0-6 are always valid */
1205         if (orc > (test_kvm_facility(vcpu->kvm, 147) ? ESSA_SET_STABLE_NODAT
1206                                                 : ESSA_SET_STABLE_IF_RESIDENT))
1207                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1208
1209         if (!vcpu->kvm->arch.migration_mode) {
1210                 /*
1211                  * CMMA is enabled in the KVM settings, but is disabled in
1212                  * the SIE block and in the mm_context, and we are not doing
1213                  * a migration. Enable CMMA in the mm_context.
1214                  * Since we need to take a write lock to write to the context
1215                  * to avoid races with storage keys handling, we check if the
1216                  * value really needs to be written to; if the value is
1217                  * already correct, we do nothing and avoid the lock.
1218                  */
1219                 if (vcpu->kvm->mm->context.uses_cmm == 0) {
1220                         down_write(&vcpu->kvm->mm->mmap_sem);
1221                         vcpu->kvm->mm->context.uses_cmm = 1;
1222                         up_write(&vcpu->kvm->mm->mmap_sem);
1223                 }
1224                 /*
1225                  * If we are here, we are supposed to have CMMA enabled in
1226                  * the SIE block. Enabling CMMA works on a per-CPU basis,
1227                  * while the context use_cmma flag is per process.
1228                  * It's possible that the context flag is enabled and the
1229                  * SIE flag is not, so we set the flag always; if it was
1230                  * already set, nothing changes, otherwise we enable it
1231                  * on this CPU too.
1232                  */
1233                 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
1234                 /* Retry the ESSA instruction */
1235                 kvm_s390_retry_instr(vcpu);
1236         } else {
1237                 int srcu_idx;
1238
1239                 down_read(&vcpu->kvm->mm->mmap_sem);
1240                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1241                 i = __do_essa(vcpu, orc);
1242                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1243                 up_read(&vcpu->kvm->mm->mmap_sem);
1244                 if (i < 0)
1245                         return i;
1246                 /* Account for the possible extra cbrl entry */
1247                 entries += i;
1248         }
1249         vcpu->arch.sie_block->cbrlo &= PAGE_MASK;       /* reset nceo */
1250         cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
1251         down_read(&gmap->mm->mmap_sem);
1252         for (i = 0; i < entries; ++i)
1253                 __gmap_zap(gmap, cbrlo[i]);
1254         up_read(&gmap->mm->mmap_sem);
1255         return 0;
1256 }
1257
1258 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
1259 {
1260         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1261         case 0x8a:
1262         case 0x8e:
1263         case 0x8f:
1264                 return handle_ipte_interlock(vcpu);
1265         case 0x8d:
1266                 return handle_epsw(vcpu);
1267         case 0xab:
1268                 return handle_essa(vcpu);
1269         case 0xaf:
1270                 return handle_pfmf(vcpu);
1271         default:
1272                 return -EOPNOTSUPP;
1273         }
1274 }
1275
1276 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
1277 {
1278         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1279         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1280         int reg, rc, nr_regs;
1281         u32 ctl_array[16];
1282         u64 ga;
1283         u8 ar;
1284
1285         vcpu->stat.instruction_lctl++;
1286
1287         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1288                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1289
1290         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1291
1292         if (ga & 3)
1293                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1294
1295         VCPU_EVENT(vcpu, 4, "LCTL: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1296         trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
1297
1298         nr_regs = ((reg3 - reg1) & 0xf) + 1;
1299         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1300         if (rc)
1301                 return kvm_s390_inject_prog_cond(vcpu, rc);
1302         reg = reg1;
1303         nr_regs = 0;
1304         do {
1305                 vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
1306                 vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
1307                 if (reg == reg3)
1308                         break;
1309                 reg = (reg + 1) % 16;
1310         } while (1);
1311         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1312         return 0;
1313 }
1314
1315 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
1316 {
1317         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1318         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1319         int reg, rc, nr_regs;
1320         u32 ctl_array[16];
1321         u64 ga;
1322         u8 ar;
1323
1324         vcpu->stat.instruction_stctl++;
1325
1326         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1327                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1328
1329         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1330
1331         if (ga & 3)
1332                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1333
1334         VCPU_EVENT(vcpu, 4, "STCTL r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1335         trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
1336
1337         reg = reg1;
1338         nr_regs = 0;
1339         do {
1340                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1341                 if (reg == reg3)
1342                         break;
1343                 reg = (reg + 1) % 16;
1344         } while (1);
1345         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1346         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1347 }
1348
1349 static int handle_lctlg(struct kvm_vcpu *vcpu)
1350 {
1351         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1352         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1353         int reg, rc, nr_regs;
1354         u64 ctl_array[16];
1355         u64 ga;
1356         u8 ar;
1357
1358         vcpu->stat.instruction_lctlg++;
1359
1360         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1361                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1362
1363         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1364
1365         if (ga & 7)
1366                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1367
1368         VCPU_EVENT(vcpu, 4, "LCTLG: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1369         trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
1370
1371         nr_regs = ((reg3 - reg1) & 0xf) + 1;
1372         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1373         if (rc)
1374                 return kvm_s390_inject_prog_cond(vcpu, rc);
1375         reg = reg1;
1376         nr_regs = 0;
1377         do {
1378                 vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
1379                 if (reg == reg3)
1380                         break;
1381                 reg = (reg + 1) % 16;
1382         } while (1);
1383         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1384         return 0;
1385 }
1386
1387 static int handle_stctg(struct kvm_vcpu *vcpu)
1388 {
1389         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1390         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1391         int reg, rc, nr_regs;
1392         u64 ctl_array[16];
1393         u64 ga;
1394         u8 ar;
1395
1396         vcpu->stat.instruction_stctg++;
1397
1398         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1399                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1400
1401         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1402
1403         if (ga & 7)
1404                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1405
1406         VCPU_EVENT(vcpu, 4, "STCTG r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1407         trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
1408
1409         reg = reg1;
1410         nr_regs = 0;
1411         do {
1412                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1413                 if (reg == reg3)
1414                         break;
1415                 reg = (reg + 1) % 16;
1416         } while (1);
1417         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1418         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1419 }
1420
1421 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
1422 {
1423         switch (vcpu->arch.sie_block->ipb & 0x000000ff) {
1424         case 0x25:
1425                 return handle_stctg(vcpu);
1426         case 0x2f:
1427                 return handle_lctlg(vcpu);
1428         case 0x60:
1429         case 0x61:
1430         case 0x62:
1431                 return handle_ri(vcpu);
1432         default:
1433                 return -EOPNOTSUPP;
1434         }
1435 }
1436
1437 static int handle_tprot(struct kvm_vcpu *vcpu)
1438 {
1439         u64 address1, address2;
1440         unsigned long hva, gpa;
1441         int ret = 0, cc = 0;
1442         bool writable;
1443         u8 ar;
1444
1445         vcpu->stat.instruction_tprot++;
1446
1447         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1448                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1449
1450         kvm_s390_get_base_disp_sse(vcpu, &address1, &address2, &ar, NULL);
1451
1452         /* we only handle the Linux memory detection case:
1453          * access key == 0
1454          * everything else goes to userspace. */
1455         if (address2 & 0xf0)
1456                 return -EOPNOTSUPP;
1457         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1458                 ipte_lock(vcpu);
1459         ret = guest_translate_address(vcpu, address1, ar, &gpa, GACC_STORE);
1460         if (ret == PGM_PROTECTION) {
1461                 /* Write protected? Try again with read-only... */
1462                 cc = 1;
1463                 ret = guest_translate_address(vcpu, address1, ar, &gpa,
1464                                               GACC_FETCH);
1465         }
1466         if (ret) {
1467                 if (ret == PGM_ADDRESSING || ret == PGM_TRANSLATION_SPEC) {
1468                         ret = kvm_s390_inject_program_int(vcpu, ret);
1469                 } else if (ret > 0) {
1470                         /* Translation not available */
1471                         kvm_s390_set_psw_cc(vcpu, 3);
1472                         ret = 0;
1473                 }
1474                 goto out_unlock;
1475         }
1476
1477         hva = gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
1478         if (kvm_is_error_hva(hva)) {
1479                 ret = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1480         } else {
1481                 if (!writable)
1482                         cc = 1;         /* Write not permitted ==> read-only */
1483                 kvm_s390_set_psw_cc(vcpu, cc);
1484                 /* Note: CC2 only occurs for storage keys (not supported yet) */
1485         }
1486 out_unlock:
1487         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1488                 ipte_unlock(vcpu);
1489         return ret;
1490 }
1491
1492 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
1493 {
1494         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1495         case 0x01:
1496                 return handle_tprot(vcpu);
1497         default:
1498                 return -EOPNOTSUPP;
1499         }
1500 }
1501
1502 static int handle_sckpf(struct kvm_vcpu *vcpu)
1503 {
1504         u32 value;
1505
1506         vcpu->stat.instruction_sckpf++;
1507
1508         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1509                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1510
1511         if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
1512                 return kvm_s390_inject_program_int(vcpu,
1513                                                    PGM_SPECIFICATION);
1514
1515         value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
1516         vcpu->arch.sie_block->todpr = value;
1517
1518         return 0;
1519 }
1520
1521 static int handle_ptff(struct kvm_vcpu *vcpu)
1522 {
1523         vcpu->stat.instruction_ptff++;
1524
1525         /* we don't emulate any control instructions yet */
1526         kvm_s390_set_psw_cc(vcpu, 3);
1527         return 0;
1528 }
1529
1530 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1531 {
1532         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1533         case 0x04:
1534                 return handle_ptff(vcpu);
1535         case 0x07:
1536                 return handle_sckpf(vcpu);
1537         default:
1538                 return -EOPNOTSUPP;
1539         }
1540 }