arch/s390/kvm/priv.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * handling privileged instructions
   4  *
   5  * Copyright IBM Corp. 2008, 2020
   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 #include <linux/pgtable.h>
  17
  18 #include <asm/asm-offsets.h>
  19 #include <asm/facility.h>
  20 #include <asm/current.h>
  21 #include <asm/debug.h>
  22 #include <asm/ebcdic.h>
  23 #include <asm/sysinfo.h>
  24 #include <asm/page-states.h>
  25 #include <asm/gmap.h>
  26 #include <asm/io.h>
  27 #include <asm/ptrace.h>
  28 #include <asm/sclp.h>
  29 #include <asm/ap.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         mmap_read_lock(current->mm);
 273         rc = get_guest_storage_key(current->mm, vmaddr, &key);
 274
 275         if (rc) {
 276                 rc = fixup_user_fault(current->mm, vmaddr,
 277                                       FAULT_FLAG_WRITE, &unlocked);
 278                 if (!rc) {
 279                         mmap_read_unlock(current->mm);
 280                         goto retry;
 281                 }
 282         }
 283         mmap_read_unlock(current->mm);
 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         mmap_read_lock(current->mm);
 320         rc = reset_guest_reference_bit(current->mm, vmaddr);
 321         if (rc < 0) {
 322                 rc = fixup_user_fault(current->mm, vmaddr,
 323                                       FAULT_FLAG_WRITE, &unlocked);
 324                 if (!rc) {
 325                         mmap_read_unlock(current->mm);
 326                         goto retry;
 327                 }
 328         }
 329         mmap_read_unlock(current->mm);
 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                 mmap_read_lock(current->mm);
 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->mm, vmaddr,
 394                                               FAULT_FLAG_WRITE, &unlocked);
 395                         rc = !rc ? -EAGAIN : rc;
 396                 }
 397                 mmap_read_unlock(current->mm);
 398                 if (rc == -EFAULT)
 399                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 400                 if (rc < 0)
 401                         return rc;
 402                 start += PAGE_SIZE;
 403         }
 404
 405         if (m3 & (SSKE_MC | SSKE_MR)) {
 406                 if (m3 & SSKE_MB) {
 407                         /* skey in reg1 is unpredictable */
 408                         kvm_s390_set_psw_cc(vcpu, 3);
 409                 } else {
 410                         kvm_s390_set_psw_cc(vcpu, rc);
 411                         vcpu->run->s.regs.gprs[reg1] &= ~0xff00UL;
 412                         vcpu->run->s.regs.gprs[reg1] |= (u64) oldkey << 8;
 413                 }
 414         }
 415         if (m3 & SSKE_MB) {
 416                 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT)
 417                         vcpu->run->s.regs.gprs[reg2] &= ~PAGE_MASK;
 418                 else
 419                         vcpu->run->s.regs.gprs[reg2] &= ~0xfffff000UL;
 420                 end = kvm_s390_logical_to_effective(vcpu, end);
 421                 vcpu->run->s.regs.gprs[reg2] |= end;
 422         }
 423         return 0;
 424 }
 425
 426 static int handle_ipte_interlock(struct kvm_vcpu *vcpu)
 427 {
 428         vcpu->stat.instruction_ipte_interlock++;
 429         if (psw_bits(vcpu->arch.sie_block->gpsw).pstate)
 430                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 431         wait_event(vcpu->kvm->arch.ipte_wq, !ipte_lock_held(vcpu));
 432         kvm_s390_retry_instr(vcpu);
 433         VCPU_EVENT(vcpu, 4, "%s", "retrying ipte interlock operation");
 434         return 0;
 435 }
 436
 437 static int handle_test_block(struct kvm_vcpu *vcpu)
 438 {
 439         gpa_t addr;
 440         int reg2;
 441
 442         vcpu->stat.instruction_tb++;
 443
 444         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 445                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 446
 447         kvm_s390_get_regs_rre(vcpu, NULL, &reg2);
 448         addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 449         addr = kvm_s390_logical_to_effective(vcpu, addr);
 450         if (kvm_s390_check_low_addr_prot_real(vcpu, addr))
 451                 return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
 452         addr = kvm_s390_real_to_abs(vcpu, addr);
 453
 454         if (kvm_is_error_gpa(vcpu->kvm, addr))
 455                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 456         /*
 457          * We don't expect errors on modern systems, and do not care
 458          * about storage keys (yet), so let's just clear the page.
 459          */
 460         if (kvm_clear_guest(vcpu->kvm, addr, PAGE_SIZE))
 461                 return -EFAULT;
 462         kvm_s390_set_psw_cc(vcpu, 0);
 463         vcpu->run->s.regs.gprs[0] = 0;
 464         return 0;
 465 }
 466
 467 static int handle_tpi(struct kvm_vcpu *vcpu)
 468 {
 469         struct kvm_s390_interrupt_info *inti;
 470         unsigned long len;
 471         u32 tpi_data[3];
 472         int rc;
 473         u64 addr;
 474         u8 ar;
 475
 476         vcpu->stat.instruction_tpi++;
 477
 478         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 479         if (addr & 3)
 480                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 481
 482         inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0);
 483         if (!inti) {
 484                 kvm_s390_set_psw_cc(vcpu, 0);
 485                 return 0;
 486         }
 487
 488         tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr;
 489         tpi_data[1] = inti->io.io_int_parm;
 490         tpi_data[2] = inti->io.io_int_word;
 491         if (addr) {
 492                 /*
 493                  * Store the two-word I/O interruption code into the
 494                  * provided area.
 495                  */
 496                 len = sizeof(tpi_data) - 4;
 497                 rc = write_guest(vcpu, addr, ar, &tpi_data, len);
 498                 if (rc) {
 499                         rc = kvm_s390_inject_prog_cond(vcpu, rc);
 500                         goto reinject_interrupt;
 501                 }
 502         } else {
 503                 /*
 504                  * Store the three-word I/O interruption code into
 505                  * the appropriate lowcore area.
 506                  */
 507                 len = sizeof(tpi_data);
 508                 if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len)) {
 509                         /* failed writes to the low core are not recoverable */
 510                         rc = -EFAULT;
 511                         goto reinject_interrupt;
 512                 }
 513         }
 514
 515         /* irq was successfully handed to the guest */
 516         kfree(inti);
 517         kvm_s390_set_psw_cc(vcpu, 1);
 518         return 0;
 519 reinject_interrupt:
 520         /*
 521          * If we encounter a problem storing the interruption code, the
 522          * instruction is suppressed from the guest's view: reinject the
 523          * interrupt.
 524          */
 525         if (kvm_s390_reinject_io_int(vcpu->kvm, inti)) {
 526                 kfree(inti);
 527                 rc = -EFAULT;
 528         }
 529         /* don't set the cc, a pgm irq was injected or we drop to user space */
 530         return rc ? -EFAULT : 0;
 531 }
 532
 533 static int handle_tsch(struct kvm_vcpu *vcpu)
 534 {
 535         struct kvm_s390_interrupt_info *inti = NULL;
 536         const u64 isc_mask = 0xffUL << 24; /* all iscs set */
 537
 538         vcpu->stat.instruction_tsch++;
 539
 540         /* a valid schid has at least one bit set */
 541         if (vcpu->run->s.regs.gprs[1])
 542                 inti = kvm_s390_get_io_int(vcpu->kvm, isc_mask,
 543                                            vcpu->run->s.regs.gprs[1]);
 544
 545         /*
 546          * Prepare exit to userspace.
 547          * We indicate whether we dequeued a pending I/O interrupt
 548          * so that userspace can re-inject it if the instruction gets
 549          * a program check. While this may re-order the pending I/O
 550          * interrupts, this is no problem since the priority is kept
 551          * intact.
 552          */
 553         vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
 554         vcpu->run->s390_tsch.dequeued = !!inti;
 555         if (inti) {
 556                 vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
 557                 vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
 558                 vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
 559                 vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
 560         }
 561         vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
 562         kfree(inti);
 563         return -EREMOTE;
 564 }
 565
 566 static int handle_io_inst(struct kvm_vcpu *vcpu)
 567 {
 568         VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
 569
 570         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 571                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 572
 573         if (vcpu->kvm->arch.css_support) {
 574                 /*
 575                  * Most I/O instructions will be handled by userspace.
 576                  * Exceptions are tpi and the interrupt portion of tsch.
 577                  */
 578                 if (vcpu->arch.sie_block->ipa == 0xb236)
 579                         return handle_tpi(vcpu);
 580                 if (vcpu->arch.sie_block->ipa == 0xb235)
 581                         return handle_tsch(vcpu);
 582                 /* Handle in userspace. */
 583                 vcpu->stat.instruction_io_other++;
 584                 return -EOPNOTSUPP;
 585         } else {
 586                 /*
 587                  * Set condition code 3 to stop the guest from issuing channel
 588                  * I/O instructions.
 589                  */
 590                 kvm_s390_set_psw_cc(vcpu, 3);
 591                 return 0;
 592         }
 593 }
 594
 595 /*
 596  * handle_pqap: Handling pqap interception
 597  * @vcpu: the vcpu having issue the pqap instruction
 598  *
 599  * We now support PQAP/AQIC instructions and we need to correctly
 600  * answer the guest even if no dedicated driver's hook is available.
 601  *
 602  * The intercepting code calls a dedicated callback for this instruction
 603  * if a driver did register one in the CRYPTO satellite of the
 604  * SIE block.
 605  *
 606  * If no callback is available, the queues are not available, return this
 607  * response code to the caller and set CC to 3.
 608  * Else return the response code returned by the callback.
 609  */
 610 static int handle_pqap(struct kvm_vcpu *vcpu)
 611 {
 612         struct ap_queue_status status = {};
 613         crypto_hook pqap_hook;
 614         unsigned long reg0;
 615         int ret;
 616         uint8_t fc;
 617
 618         /* Verify that the AP instruction are available */
 619         if (!ap_instructions_available())
 620                 return -EOPNOTSUPP;
 621         /* Verify that the guest is allowed to use AP instructions */
 622         if (!(vcpu->arch.sie_block->eca & ECA_APIE))
 623                 return -EOPNOTSUPP;
 624         /*
 625          * The only possibly intercepted functions when AP instructions are
 626          * available for the guest are AQIC and TAPQ with the t bit set
 627          * since we do not set IC.3 (FIII) we currently will only intercept
 628          * the AQIC function code.
 629          * Note: running nested under z/VM can result in intercepts for other
 630          * function codes, e.g. PQAP(QCI). We do not support this and bail out.
 631          */
 632         reg0 = vcpu->run->s.regs.gprs[0];
 633         fc = (reg0 >> 24) & 0xff;
 634         if (fc != 0x03)
 635                 return -EOPNOTSUPP;
 636
 637         /* PQAP instruction is allowed for guest kernel only */
 638         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 639                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 640
 641         /* Common PQAP instruction specification exceptions */
 642         /* bits 41-47 must all be zeros */
 643         if (reg0 & 0x007f0000UL)
 644                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 645         /* APFT not install and T bit set */
 646         if (!test_kvm_facility(vcpu->kvm, 15) && (reg0 & 0x00800000UL))
 647                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 648         /* APXA not installed and APID greater 64 or APQI greater 16 */
 649         if (!(vcpu->kvm->arch.crypto.crycbd & 0x02) && (reg0 & 0x0000c0f0UL))
 650                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 651
 652         /* AQIC function code specific exception */
 653         /* facility 65 not present for AQIC function code */
 654         if (!test_kvm_facility(vcpu->kvm, 65))
 655                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 656
 657         /*
 658          * If the hook callback is registered, there will be a pointer to the
 659          * hook function pointer in the kvm_s390_crypto structure. Lock the
 660          * owner, retrieve the hook function pointer and call the hook.
 661          */
 662         down_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
 663         if (vcpu->kvm->arch.crypto.pqap_hook) {
 664                 pqap_hook = *vcpu->kvm->arch.crypto.pqap_hook;
 665                 ret = pqap_hook(vcpu);
 666                 if (!ret && vcpu->run->s.regs.gprs[1] & 0x00ff0000)
 667                         kvm_s390_set_psw_cc(vcpu, 3);
 668                 up_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
 669                 return ret;
 670         }
 671         up_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
 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 (!kvm_s390_pv_cpu_is_protected(vcpu) && (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_ACCOUNT);
 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_ACCOUNT);
 895                 if (!mem)
 896                         goto out_no_data;
 897                 handle_stsi_3_2_2(vcpu, (void *) mem);
 898                 break;
 899         }
 900         if (kvm_s390_pv_cpu_is_protected(vcpu)) {
 901                 memcpy((void *)sida_origin(vcpu->arch.sie_block), (void *)mem,
 902                        PAGE_SIZE);
 903                 rc = 0;
 904         } else {
 905                 rc = write_guest(vcpu, operand2, ar, (void *)mem, PAGE_SIZE);
 906         }
 907         if (rc) {
 908                 rc = kvm_s390_inject_prog_cond(vcpu, rc);
 909                 goto out;
 910         }
 911         if (vcpu->kvm->arch.user_stsi) {
 912                 insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
 913                 rc = -EREMOTE;
 914         }
 915         trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
 916         free_page(mem);
 917         kvm_s390_set_psw_cc(vcpu, 0);
 918         vcpu->run->s.regs.gprs[0] = 0;
 919         return rc;
 920 out_no_data:
 921         kvm_s390_set_psw_cc(vcpu, 3);
 922 out:
 923         free_page(mem);
 924         return rc;
 925 }
 926
 927 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
 928 {
 929         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
 930         case 0x02:
 931                 return handle_stidp(vcpu);
 932         case 0x04:
 933                 return handle_set_clock(vcpu);
 934         case 0x10:
 935                 return handle_set_prefix(vcpu);
 936         case 0x11:
 937                 return handle_store_prefix(vcpu);
 938         case 0x12:
 939                 return handle_store_cpu_address(vcpu);
 940         case 0x14:
 941                 return kvm_s390_handle_vsie(vcpu);
 942         case 0x21:
 943         case 0x50:
 944                 return handle_ipte_interlock(vcpu);
 945         case 0x29:
 946                 return handle_iske(vcpu);
 947         case 0x2a:
 948                 return handle_rrbe(vcpu);
 949         case 0x2b:
 950                 return handle_sske(vcpu);
 951         case 0x2c:
 952                 return handle_test_block(vcpu);
 953         case 0x30:
 954         case 0x31:
 955         case 0x32:
 956         case 0x33:
 957         case 0x34:
 958         case 0x35:
 959         case 0x36:
 960         case 0x37:
 961         case 0x38:
 962         case 0x39:
 963         case 0x3a:
 964         case 0x3b:
 965         case 0x3c:
 966         case 0x5f:
 967         case 0x74:
 968         case 0x76:
 969                 return handle_io_inst(vcpu);
 970         case 0x56:
 971                 return handle_sthyi(vcpu);
 972         case 0x7d:
 973                 return handle_stsi(vcpu);
 974         case 0xaf:
 975                 return handle_pqap(vcpu);
 976         case 0xb1:
 977                 return handle_stfl(vcpu);
 978         case 0xb2:
 979                 return handle_lpswe(vcpu);
 980         default:
 981                 return -EOPNOTSUPP;
 982         }
 983 }
 984
 985 static int handle_epsw(struct kvm_vcpu *vcpu)
 986 {
 987         int reg1, reg2;
 988
 989         vcpu->stat.instruction_epsw++;
 990
 991         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 992
 993         /* This basically extracts the mask half of the psw. */
 994         vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
 995         vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
 996         if (reg2) {
 997                 vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
 998                 vcpu->run->s.regs.gprs[reg2] |=
 999                         vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
1000         }
1001         return 0;
1002 }
1003
1004 #define PFMF_RESERVED   0xfffc0101UL
1005 #define PFMF_SK         0x00020000UL
1006 #define PFMF_CF         0x00010000UL
1007 #define PFMF_UI         0x00008000UL
1008 #define PFMF_FSC        0x00007000UL
1009 #define PFMF_NQ         0x00000800UL
1010 #define PFMF_MR         0x00000400UL
1011 #define PFMF_MC         0x00000200UL
1012 #define PFMF_KEY        0x000000feUL
1013
1014 static int handle_pfmf(struct kvm_vcpu *vcpu)
1015 {
1016         bool mr = false, mc = false, nq;
1017         int reg1, reg2;
1018         unsigned long start, end;
1019         unsigned char key;
1020
1021         vcpu->stat.instruction_pfmf++;
1022
1023         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
1024
1025         if (!test_kvm_facility(vcpu->kvm, 8))
1026                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1027
1028         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1029                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1030
1031         if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
1032                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1033
1034         /* Only provide non-quiescing support if enabled for the guest */
1035         if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ &&
1036             !test_kvm_facility(vcpu->kvm, 14))
1037                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1038
1039         /* Only provide conditional-SSKE support if enabled for the guest */
1040         if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK &&
1041             test_kvm_facility(vcpu->kvm, 10)) {
1042                 mr = vcpu->run->s.regs.gprs[reg1] & PFMF_MR;
1043                 mc = vcpu->run->s.regs.gprs[reg1] & PFMF_MC;
1044         }
1045
1046         nq = vcpu->run->s.regs.gprs[reg1] & PFMF_NQ;
1047         key = vcpu->run->s.regs.gprs[reg1] & PFMF_KEY;
1048         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
1049         start = kvm_s390_logical_to_effective(vcpu, start);
1050
1051         if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1052                 if (kvm_s390_check_low_addr_prot_real(vcpu, start))
1053                         return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
1054         }
1055
1056         switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1057         case 0x00000000:
1058                 /* only 4k frames specify a real address */
1059                 start = kvm_s390_real_to_abs(vcpu, start);
1060                 end = (start + PAGE_SIZE) & ~(PAGE_SIZE - 1);
1061                 break;
1062         case 0x00001000:
1063                 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
1064                 break;
1065         case 0x00002000:
1066                 /* only support 2G frame size if EDAT2 is available and we are
1067                    not in 24-bit addressing mode */
1068                 if (!test_kvm_facility(vcpu->kvm, 78) ||
1069                     psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_24BIT)
1070                         return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1071                 end = (start + _REGION3_SIZE) & ~(_REGION3_SIZE - 1);
1072                 break;
1073         default:
1074                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1075         }
1076
1077         while (start != end) {
1078                 unsigned long vmaddr;
1079                 bool unlocked = false;
1080
1081                 /* Translate guest address to host address */
1082                 vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
1083                 if (kvm_is_error_hva(vmaddr))
1084                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1085
1086                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1087                         if (kvm_clear_guest(vcpu->kvm, start, PAGE_SIZE))
1088                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1089                 }
1090
1091                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
1092                         int rc = kvm_s390_skey_check_enable(vcpu);
1093
1094                         if (rc)
1095                                 return rc;
1096                         mmap_read_lock(current->mm);
1097                         rc = cond_set_guest_storage_key(current->mm, vmaddr,
1098                                                         key, NULL, nq, mr, mc);
1099                         if (rc < 0) {
1100                                 rc = fixup_user_fault(current->mm, vmaddr,
1101                                                       FAULT_FLAG_WRITE, &unlocked);
1102                                 rc = !rc ? -EAGAIN : rc;
1103                         }
1104                         mmap_read_unlock(current->mm);
1105                         if (rc == -EFAULT)
1106                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1107                         if (rc == -EAGAIN)
1108                                 continue;
1109                         if (rc < 0)
1110                                 return rc;
1111                 }
1112                 start += PAGE_SIZE;
1113         }
1114         if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1115                 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT) {
1116                         vcpu->run->s.regs.gprs[reg2] = end;
1117                 } else {
1118                         vcpu->run->s.regs.gprs[reg2] &= ~0xffffffffUL;
1119                         end = kvm_s390_logical_to_effective(vcpu, end);
1120                         vcpu->run->s.regs.gprs[reg2] |= end;
1121                 }
1122         }
1123         return 0;
1124 }
1125
1126 /*
1127  * Must be called with relevant read locks held (kvm->mm->mmap_lock, kvm->srcu)
1128  */
1129 static inline int __do_essa(struct kvm_vcpu *vcpu, const int orc)
1130 {
1131         int r1, r2, nappended, entries;
1132         unsigned long gfn, hva, res, pgstev, ptev;
1133         unsigned long *cbrlo;
1134
1135         /*
1136          * We don't need to set SD.FPF.SK to 1 here, because if we have a
1137          * machine check here we either handle it or crash
1138          */
1139
1140         kvm_s390_get_regs_rre(vcpu, &r1, &r2);
1141         gfn = vcpu->run->s.regs.gprs[r2] >> PAGE_SHIFT;
1142         hva = gfn_to_hva(vcpu->kvm, gfn);
1143         entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1144
1145         if (kvm_is_error_hva(hva))
1146                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1147
1148         nappended = pgste_perform_essa(vcpu->kvm->mm, hva, orc, &ptev, &pgstev);
1149         if (nappended < 0) {
1150                 res = orc ? 0x10 : 0;
1151                 vcpu->run->s.regs.gprs[r1] = res; /* Exception Indication */
1152                 return 0;
1153         }
1154         res = (pgstev & _PGSTE_GPS_USAGE_MASK) >> 22;
1155         /*
1156          * Set the block-content state part of the result. 0 means resident, so
1157          * nothing to do if the page is valid. 2 is for preserved pages
1158          * (non-present and non-zero), and 3 for zero pages (non-present and
1159          * zero).
1160          */
1161         if (ptev & _PAGE_INVALID) {
1162                 res |= 2;
1163                 if (pgstev & _PGSTE_GPS_ZERO)
1164                         res |= 1;
1165         }
1166         if (pgstev & _PGSTE_GPS_NODAT)
1167                 res |= 0x20;
1168         vcpu->run->s.regs.gprs[r1] = res;
1169         /*
1170          * It is possible that all the normal 511 slots were full, in which case
1171          * we will now write in the 512th slot, which is reserved for host use.
1172          * In both cases we let the normal essa handling code process all the
1173          * slots, including the reserved one, if needed.
1174          */
1175         if (nappended > 0) {
1176                 cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo & PAGE_MASK);
1177                 cbrlo[entries] = gfn << PAGE_SHIFT;
1178         }
1179
1180         if (orc) {
1181                 struct kvm_memory_slot *ms = gfn_to_memslot(vcpu->kvm, gfn);
1182
1183                 /* Increment only if we are really flipping the bit */
1184                 if (ms && !test_and_set_bit(gfn - ms->base_gfn, kvm_second_dirty_bitmap(ms)))
1185                         atomic64_inc(&vcpu->kvm->arch.cmma_dirty_pages);
1186         }
1187
1188         return nappended;
1189 }
1190
1191 static int handle_essa(struct kvm_vcpu *vcpu)
1192 {
1193         /* entries expected to be 1FF */
1194         int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1195         unsigned long *cbrlo;
1196         struct gmap *gmap;
1197         int i, orc;
1198
1199         VCPU_EVENT(vcpu, 4, "ESSA: release %d pages", entries);
1200         gmap = vcpu->arch.gmap;
1201         vcpu->stat.instruction_essa++;
1202         if (!vcpu->kvm->arch.use_cmma)
1203                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1204
1205         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1206                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1207         /* Check for invalid operation request code */
1208         orc = (vcpu->arch.sie_block->ipb & 0xf0000000) >> 28;
1209         /* ORCs 0-6 are always valid */
1210         if (orc > (test_kvm_facility(vcpu->kvm, 147) ? ESSA_SET_STABLE_NODAT
1211                                                 : ESSA_SET_STABLE_IF_RESIDENT))
1212                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1213
1214         if (!vcpu->kvm->arch.migration_mode) {
1215                 /*
1216                  * CMMA is enabled in the KVM settings, but is disabled in
1217                  * the SIE block and in the mm_context, and we are not doing
1218                  * a migration. Enable CMMA in the mm_context.
1219                  * Since we need to take a write lock to write to the context
1220                  * to avoid races with storage keys handling, we check if the
1221                  * value really needs to be written to; if the value is
1222                  * already correct, we do nothing and avoid the lock.
1223                  */
1224                 if (vcpu->kvm->mm->context.uses_cmm == 0) {
1225                         mmap_write_lock(vcpu->kvm->mm);
1226                         vcpu->kvm->mm->context.uses_cmm = 1;
1227                         mmap_write_unlock(vcpu->kvm->mm);
1228                 }
1229                 /*
1230                  * If we are here, we are supposed to have CMMA enabled in
1231                  * the SIE block. Enabling CMMA works on a per-CPU basis,
1232                  * while the context use_cmma flag is per process.
1233                  * It's possible that the context flag is enabled and the
1234                  * SIE flag is not, so we set the flag always; if it was
1235                  * already set, nothing changes, otherwise we enable it
1236                  * on this CPU too.
1237                  */
1238                 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
1239                 /* Retry the ESSA instruction */
1240                 kvm_s390_retry_instr(vcpu);
1241         } else {
1242                 int srcu_idx;
1243
1244                 mmap_read_lock(vcpu->kvm->mm);
1245                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1246                 i = __do_essa(vcpu, orc);
1247                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1248                 mmap_read_unlock(vcpu->kvm->mm);
1249                 if (i < 0)
1250                         return i;
1251                 /* Account for the possible extra cbrl entry */
1252                 entries += i;
1253         }
1254         vcpu->arch.sie_block->cbrlo &= PAGE_MASK;       /* reset nceo */
1255         cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
1256         mmap_read_lock(gmap->mm);
1257         for (i = 0; i < entries; ++i)
1258                 __gmap_zap(gmap, cbrlo[i]);
1259         mmap_read_unlock(gmap->mm);
1260         return 0;
1261 }
1262
1263 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
1264 {
1265         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1266         case 0x8a:
1267         case 0x8e:
1268         case 0x8f:
1269                 return handle_ipte_interlock(vcpu);
1270         case 0x8d:
1271                 return handle_epsw(vcpu);
1272         case 0xab:
1273                 return handle_essa(vcpu);
1274         case 0xaf:
1275                 return handle_pfmf(vcpu);
1276         default:
1277                 return -EOPNOTSUPP;
1278         }
1279 }
1280
1281 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
1282 {
1283         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1284         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1285         int reg, rc, nr_regs;
1286         u32 ctl_array[16];
1287         u64 ga;
1288         u8 ar;
1289
1290         vcpu->stat.instruction_lctl++;
1291
1292         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1293                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1294
1295         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1296
1297         if (ga & 3)
1298                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1299
1300         VCPU_EVENT(vcpu, 4, "LCTL: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1301         trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
1302
1303         nr_regs = ((reg3 - reg1) & 0xf) + 1;
1304         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1305         if (rc)
1306                 return kvm_s390_inject_prog_cond(vcpu, rc);
1307         reg = reg1;
1308         nr_regs = 0;
1309         do {
1310                 vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
1311                 vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
1312                 if (reg == reg3)
1313                         break;
1314                 reg = (reg + 1) % 16;
1315         } while (1);
1316         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1317         return 0;
1318 }
1319
1320 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
1321 {
1322         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1323         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1324         int reg, rc, nr_regs;
1325         u32 ctl_array[16];
1326         u64 ga;
1327         u8 ar;
1328
1329         vcpu->stat.instruction_stctl++;
1330
1331         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1332                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1333
1334         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1335
1336         if (ga & 3)
1337                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1338
1339         VCPU_EVENT(vcpu, 4, "STCTL r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1340         trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
1341
1342         reg = reg1;
1343         nr_regs = 0;
1344         do {
1345                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1346                 if (reg == reg3)
1347                         break;
1348                 reg = (reg + 1) % 16;
1349         } while (1);
1350         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1351         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1352 }
1353
1354 static int handle_lctlg(struct kvm_vcpu *vcpu)
1355 {
1356         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1357         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1358         int reg, rc, nr_regs;
1359         u64 ctl_array[16];
1360         u64 ga;
1361         u8 ar;
1362
1363         vcpu->stat.instruction_lctlg++;
1364
1365         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1366                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1367
1368         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1369
1370         if (ga & 7)
1371                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1372
1373         VCPU_EVENT(vcpu, 4, "LCTLG: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1374         trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
1375
1376         nr_regs = ((reg3 - reg1) & 0xf) + 1;
1377         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1378         if (rc)
1379                 return kvm_s390_inject_prog_cond(vcpu, rc);
1380         reg = reg1;
1381         nr_regs = 0;
1382         do {
1383                 vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
1384                 if (reg == reg3)
1385                         break;
1386                 reg = (reg + 1) % 16;
1387         } while (1);
1388         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1389         return 0;
1390 }
1391
1392 static int handle_stctg(struct kvm_vcpu *vcpu)
1393 {
1394         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1395         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1396         int reg, rc, nr_regs;
1397         u64 ctl_array[16];
1398         u64 ga;
1399         u8 ar;
1400
1401         vcpu->stat.instruction_stctg++;
1402
1403         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1404                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1405
1406         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1407
1408         if (ga & 7)
1409                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1410
1411         VCPU_EVENT(vcpu, 4, "STCTG r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1412         trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
1413
1414         reg = reg1;
1415         nr_regs = 0;
1416         do {
1417                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1418                 if (reg == reg3)
1419                         break;
1420                 reg = (reg + 1) % 16;
1421         } while (1);
1422         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1423         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1424 }
1425
1426 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
1427 {
1428         switch (vcpu->arch.sie_block->ipb & 0x000000ff) {
1429         case 0x25:
1430                 return handle_stctg(vcpu);
1431         case 0x2f:
1432                 return handle_lctlg(vcpu);
1433         case 0x60:
1434         case 0x61:
1435         case 0x62:
1436                 return handle_ri(vcpu);
1437         default:
1438                 return -EOPNOTSUPP;
1439         }
1440 }
1441
1442 static int handle_tprot(struct kvm_vcpu *vcpu)
1443 {
1444         u64 address1, address2;
1445         unsigned long hva, gpa;
1446         int ret = 0, cc = 0;
1447         bool writable;
1448         u8 ar;
1449
1450         vcpu->stat.instruction_tprot++;
1451
1452         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1453                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1454
1455         kvm_s390_get_base_disp_sse(vcpu, &address1, &address2, &ar, NULL);
1456
1457         /* we only handle the Linux memory detection case:
1458          * access key == 0
1459          * everything else goes to userspace. */
1460         if (address2 & 0xf0)
1461                 return -EOPNOTSUPP;
1462         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1463                 ipte_lock(vcpu);
1464         ret = guest_translate_address(vcpu, address1, ar, &gpa, GACC_STORE);
1465         if (ret == PGM_PROTECTION) {
1466                 /* Write protected? Try again with read-only... */
1467                 cc = 1;
1468                 ret = guest_translate_address(vcpu, address1, ar, &gpa,
1469                                               GACC_FETCH);
1470         }
1471         if (ret) {
1472                 if (ret == PGM_ADDRESSING || ret == PGM_TRANSLATION_SPEC) {
1473                         ret = kvm_s390_inject_program_int(vcpu, ret);
1474                 } else if (ret > 0) {
1475                         /* Translation not available */
1476                         kvm_s390_set_psw_cc(vcpu, 3);
1477                         ret = 0;
1478                 }
1479                 goto out_unlock;
1480         }
1481
1482         hva = gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
1483         if (kvm_is_error_hva(hva)) {
1484                 ret = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1485         } else {
1486                 if (!writable)
1487                         cc = 1;         /* Write not permitted ==> read-only */
1488                 kvm_s390_set_psw_cc(vcpu, cc);
1489                 /* Note: CC2 only occurs for storage keys (not supported yet) */
1490         }
1491 out_unlock:
1492         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1493                 ipte_unlock(vcpu);
1494         return ret;
1495 }
1496
1497 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
1498 {
1499         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1500         case 0x01:
1501                 return handle_tprot(vcpu);
1502         default:
1503                 return -EOPNOTSUPP;
1504         }
1505 }
1506
1507 static int handle_sckpf(struct kvm_vcpu *vcpu)
1508 {
1509         u32 value;
1510
1511         vcpu->stat.instruction_sckpf++;
1512
1513         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1514                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1515
1516         if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
1517                 return kvm_s390_inject_program_int(vcpu,
1518                                                    PGM_SPECIFICATION);
1519
1520         value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
1521         vcpu->arch.sie_block->todpr = value;
1522
1523         return 0;
1524 }
1525
1526 static int handle_ptff(struct kvm_vcpu *vcpu)
1527 {
1528         vcpu->stat.instruction_ptff++;
1529
1530         /* we don't emulate any control instructions yet */
1531         kvm_s390_set_psw_cc(vcpu, 3);
1532         return 0;
1533 }
1534
1535 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1536 {
1537         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1538         case 0x04:
1539                 return handle_ptff(vcpu);
1540         case 0x07:
1541                 return handle_sckpf(vcpu);
1542         default:
1543                 return -EOPNOTSUPP;
1544         }
1545 }