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         unsigned long reg0;
 614         int ret;
 615         uint8_t fc;
 616
 617         /* Verify that the AP instruction are available */
 618         if (!ap_instructions_available())
 619                 return -EOPNOTSUPP;
 620         /* Verify that the guest is allowed to use AP instructions */
 621         if (!(vcpu->arch.sie_block->eca & ECA_APIE))
 622                 return -EOPNOTSUPP;
 623         /*
 624          * The only possibly intercepted functions when AP instructions are
 625          * available for the guest are AQIC and TAPQ with the t bit set
 626          * since we do not set IC.3 (FIII) we currently will only intercept
 627          * the AQIC function code.
 628          * Note: running nested under z/VM can result in intercepts for other
 629          * function codes, e.g. PQAP(QCI). We do not support this and bail out.
 630          */
 631         reg0 = vcpu->run->s.regs.gprs[0];
 632         fc = (reg0 >> 24) & 0xff;
 633         if (fc != 0x03)
 634                 return -EOPNOTSUPP;
 635
 636         /* PQAP instruction is allowed for guest kernel only */
 637         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 638                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 639
 640         /* Common PQAP instruction specification exceptions */
 641         /* bits 41-47 must all be zeros */
 642         if (reg0 & 0x007f0000UL)
 643                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 644         /* APFT not install and T bit set */
 645         if (!test_kvm_facility(vcpu->kvm, 15) && (reg0 & 0x00800000UL))
 646                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 647         /* APXA not installed and APID greater 64 or APQI greater 16 */
 648         if (!(vcpu->kvm->arch.crypto.crycbd & 0x02) && (reg0 & 0x0000c0f0UL))
 649                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 650
 651         /* AQIC function code specific exception */
 652         /* facility 65 not present for AQIC function code */
 653         if (!test_kvm_facility(vcpu->kvm, 65))
 654                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 655
 656         /*
 657          * Verify that the hook callback is registered, lock the owner
 658          * and call the hook.
 659          */
 660         if (vcpu->kvm->arch.crypto.pqap_hook) {
 661                 if (!try_module_get(vcpu->kvm->arch.crypto.pqap_hook->owner))
 662                         return -EOPNOTSUPP;
 663                 ret = vcpu->kvm->arch.crypto.pqap_hook->hook(vcpu);
 664                 module_put(vcpu->kvm->arch.crypto.pqap_hook->owner);
 665                 if (!ret && vcpu->run->s.regs.gprs[1] & 0x00ff0000)
 666                         kvm_s390_set_psw_cc(vcpu, 3);
 667                 return ret;
 668         }
 669         /*
 670          * A vfio_driver must register a hook.
 671          * No hook means no driver to enable the SIE CRYCB and no queues.
 672          * We send this response to the guest.
 673          */
 674         status.response_code = 0x01;
 675         memcpy(&vcpu->run->s.regs.gprs[1], &status, sizeof(status));
 676         kvm_s390_set_psw_cc(vcpu, 3);
 677         return 0;
 678 }
 679
 680 static int handle_stfl(struct kvm_vcpu *vcpu)
 681 {
 682         int rc;
 683         unsigned int fac;
 684
 685         vcpu->stat.instruction_stfl++;
 686
 687         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 688                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 689
 690         /*
 691          * We need to shift the lower 32 facility bits (bit 0-31) from a u64
 692          * into a u32 memory representation. They will remain bits 0-31.
 693          */
 694         fac = *vcpu->kvm->arch.model.fac_list >> 32;
 695         rc = write_guest_lc(vcpu, offsetof(struct lowcore, stfl_fac_list),
 696                             &fac, sizeof(fac));
 697         if (rc)
 698                 return rc;
 699         VCPU_EVENT(vcpu, 3, "STFL: store facility list 0x%x", fac);
 700         trace_kvm_s390_handle_stfl(vcpu, fac);
 701         return 0;
 702 }
 703
 704 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
 705 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
 706 #define PSW_ADDR_24 0x0000000000ffffffUL
 707 #define PSW_ADDR_31 0x000000007fffffffUL
 708
 709 int is_valid_psw(psw_t *psw)
 710 {
 711         if (psw->mask & PSW_MASK_UNASSIGNED)
 712                 return 0;
 713         if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
 714                 if (psw->addr & ~PSW_ADDR_31)
 715                         return 0;
 716         }
 717         if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
 718                 return 0;
 719         if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
 720                 return 0;
 721         if (psw->addr & 1)
 722                 return 0;
 723         return 1;
 724 }
 725
 726 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
 727 {
 728         psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
 729         psw_compat_t new_psw;
 730         u64 addr;
 731         int rc;
 732         u8 ar;
 733
 734         vcpu->stat.instruction_lpsw++;
 735
 736         if (gpsw->mask & PSW_MASK_PSTATE)
 737                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 738
 739         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 740         if (addr & 7)
 741                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 742
 743         rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
 744         if (rc)
 745                 return kvm_s390_inject_prog_cond(vcpu, rc);
 746         if (!(new_psw.mask & PSW32_MASK_BASE))
 747                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 748         gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
 749         gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
 750         gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
 751         if (!is_valid_psw(gpsw))
 752                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 753         return 0;
 754 }
 755
 756 static int handle_lpswe(struct kvm_vcpu *vcpu)
 757 {
 758         psw_t new_psw;
 759         u64 addr;
 760         int rc;
 761         u8 ar;
 762
 763         vcpu->stat.instruction_lpswe++;
 764
 765         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 766                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 767
 768         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 769         if (addr & 7)
 770                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 771         rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
 772         if (rc)
 773                 return kvm_s390_inject_prog_cond(vcpu, rc);
 774         vcpu->arch.sie_block->gpsw = new_psw;
 775         if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
 776                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 777         return 0;
 778 }
 779
 780 static int handle_stidp(struct kvm_vcpu *vcpu)
 781 {
 782         u64 stidp_data = vcpu->kvm->arch.model.cpuid;
 783         u64 operand2;
 784         int rc;
 785         u8 ar;
 786
 787         vcpu->stat.instruction_stidp++;
 788
 789         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 790                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 791
 792         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 793
 794         if (operand2 & 7)
 795                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 796
 797         rc = write_guest(vcpu, operand2, ar, &stidp_data, sizeof(stidp_data));
 798         if (rc)
 799                 return kvm_s390_inject_prog_cond(vcpu, rc);
 800
 801         VCPU_EVENT(vcpu, 3, "STIDP: store cpu id 0x%llx", stidp_data);
 802         return 0;
 803 }
 804
 805 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
 806 {
 807         int cpus = 0;
 808         int n;
 809
 810         cpus = atomic_read(&vcpu->kvm->online_vcpus);
 811
 812         /* deal with other level 3 hypervisors */
 813         if (stsi(mem, 3, 2, 2))
 814                 mem->count = 0;
 815         if (mem->count < 8)
 816                 mem->count++;
 817         for (n = mem->count - 1; n > 0 ; n--)
 818                 memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
 819
 820         memset(&mem->vm[0], 0, sizeof(mem->vm[0]));
 821         mem->vm[0].cpus_total = cpus;
 822         mem->vm[0].cpus_configured = cpus;
 823         mem->vm[0].cpus_standby = 0;
 824         mem->vm[0].cpus_reserved = 0;
 825         mem->vm[0].caf = 1000;
 826         memcpy(mem->vm[0].name, "KVMguest", 8);
 827         ASCEBC(mem->vm[0].name, 8);
 828         memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
 829         ASCEBC(mem->vm[0].cpi, 16);
 830 }
 831
 832 static void insert_stsi_usr_data(struct kvm_vcpu *vcpu, u64 addr, u8 ar,
 833                                  u8 fc, u8 sel1, u16 sel2)
 834 {
 835         vcpu->run->exit_reason = KVM_EXIT_S390_STSI;
 836         vcpu->run->s390_stsi.addr = addr;
 837         vcpu->run->s390_stsi.ar = ar;
 838         vcpu->run->s390_stsi.fc = fc;
 839         vcpu->run->s390_stsi.sel1 = sel1;
 840         vcpu->run->s390_stsi.sel2 = sel2;
 841 }
 842
 843 static int handle_stsi(struct kvm_vcpu *vcpu)
 844 {
 845         int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
 846         int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
 847         int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
 848         unsigned long mem = 0;
 849         u64 operand2;
 850         int rc = 0;
 851         u8 ar;
 852
 853         vcpu->stat.instruction_stsi++;
 854         VCPU_EVENT(vcpu, 3, "STSI: fc: %u sel1: %u sel2: %u", fc, sel1, sel2);
 855
 856         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 857                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 858
 859         if (fc > 3) {
 860                 kvm_s390_set_psw_cc(vcpu, 3);
 861                 return 0;
 862         }
 863
 864         if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
 865             || vcpu->run->s.regs.gprs[1] & 0xffff0000)
 866                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 867
 868         if (fc == 0) {
 869                 vcpu->run->s.regs.gprs[0] = 3 << 28;
 870                 kvm_s390_set_psw_cc(vcpu, 0);
 871                 return 0;
 872         }
 873
 874         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 875
 876         if (!kvm_s390_pv_cpu_is_protected(vcpu) && (operand2 & 0xfff))
 877                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 878
 879         switch (fc) {
 880         case 1: /* same handling for 1 and 2 */
 881         case 2:
 882                 mem = get_zeroed_page(GFP_KERNEL_ACCOUNT);
 883                 if (!mem)
 884                         goto out_no_data;
 885                 if (stsi((void *) mem, fc, sel1, sel2))
 886                         goto out_no_data;
 887                 break;
 888         case 3:
 889                 if (sel1 != 2 || sel2 != 2)
 890                         goto out_no_data;
 891                 mem = get_zeroed_page(GFP_KERNEL_ACCOUNT);
 892                 if (!mem)
 893                         goto out_no_data;
 894                 handle_stsi_3_2_2(vcpu, (void *) mem);
 895                 break;
 896         }
 897         if (kvm_s390_pv_cpu_is_protected(vcpu)) {
 898                 memcpy((void *)sida_origin(vcpu->arch.sie_block), (void *)mem,
 899                        PAGE_SIZE);
 900                 rc = 0;
 901         } else {
 902                 rc = write_guest(vcpu, operand2, ar, (void *)mem, PAGE_SIZE);
 903         }
 904         if (rc) {
 905                 rc = kvm_s390_inject_prog_cond(vcpu, rc);
 906                 goto out;
 907         }
 908         if (vcpu->kvm->arch.user_stsi) {
 909                 insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
 910                 rc = -EREMOTE;
 911         }
 912         trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
 913         free_page(mem);
 914         kvm_s390_set_psw_cc(vcpu, 0);
 915         vcpu->run->s.regs.gprs[0] = 0;
 916         return rc;
 917 out_no_data:
 918         kvm_s390_set_psw_cc(vcpu, 3);
 919 out:
 920         free_page(mem);
 921         return rc;
 922 }
 923
 924 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
 925 {
 926         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
 927         case 0x02:
 928                 return handle_stidp(vcpu);
 929         case 0x04:
 930                 return handle_set_clock(vcpu);
 931         case 0x10:
 932                 return handle_set_prefix(vcpu);
 933         case 0x11:
 934                 return handle_store_prefix(vcpu);
 935         case 0x12:
 936                 return handle_store_cpu_address(vcpu);
 937         case 0x14:
 938                 return kvm_s390_handle_vsie(vcpu);
 939         case 0x21:
 940         case 0x50:
 941                 return handle_ipte_interlock(vcpu);
 942         case 0x29:
 943                 return handle_iske(vcpu);
 944         case 0x2a:
 945                 return handle_rrbe(vcpu);
 946         case 0x2b:
 947                 return handle_sske(vcpu);
 948         case 0x2c:
 949                 return handle_test_block(vcpu);
 950         case 0x30:
 951         case 0x31:
 952         case 0x32:
 953         case 0x33:
 954         case 0x34:
 955         case 0x35:
 956         case 0x36:
 957         case 0x37:
 958         case 0x38:
 959         case 0x39:
 960         case 0x3a:
 961         case 0x3b:
 962         case 0x3c:
 963         case 0x5f:
 964         case 0x74:
 965         case 0x76:
 966                 return handle_io_inst(vcpu);
 967         case 0x56:
 968                 return handle_sthyi(vcpu);
 969         case 0x7d:
 970                 return handle_stsi(vcpu);
 971         case 0xaf:
 972                 return handle_pqap(vcpu);
 973         case 0xb1:
 974                 return handle_stfl(vcpu);
 975         case 0xb2:
 976                 return handle_lpswe(vcpu);
 977         default:
 978                 return -EOPNOTSUPP;
 979         }
 980 }
 981
 982 static int handle_epsw(struct kvm_vcpu *vcpu)
 983 {
 984         int reg1, reg2;
 985
 986         vcpu->stat.instruction_epsw++;
 987
 988         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 989
 990         /* This basically extracts the mask half of the psw. */
 991         vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
 992         vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
 993         if (reg2) {
 994                 vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
 995                 vcpu->run->s.regs.gprs[reg2] |=
 996                         vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
 997         }
 998         return 0;
 999 }
1000
1001 #define PFMF_RESERVED   0xfffc0101UL
1002 #define PFMF_SK         0x00020000UL
1003 #define PFMF_CF         0x00010000UL
1004 #define PFMF_UI         0x00008000UL
1005 #define PFMF_FSC        0x00007000UL
1006 #define PFMF_NQ         0x00000800UL
1007 #define PFMF_MR         0x00000400UL
1008 #define PFMF_MC         0x00000200UL
1009 #define PFMF_KEY        0x000000feUL
1010
1011 static int handle_pfmf(struct kvm_vcpu *vcpu)
1012 {
1013         bool mr = false, mc = false, nq;
1014         int reg1, reg2;
1015         unsigned long start, end;
1016         unsigned char key;
1017
1018         vcpu->stat.instruction_pfmf++;
1019
1020         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
1021
1022         if (!test_kvm_facility(vcpu->kvm, 8))
1023                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1024
1025         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1026                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1027
1028         if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
1029                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1030
1031         /* Only provide non-quiescing support if enabled for the guest */
1032         if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ &&
1033             !test_kvm_facility(vcpu->kvm, 14))
1034                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1035
1036         /* Only provide conditional-SSKE support if enabled for the guest */
1037         if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK &&
1038             test_kvm_facility(vcpu->kvm, 10)) {
1039                 mr = vcpu->run->s.regs.gprs[reg1] & PFMF_MR;
1040                 mc = vcpu->run->s.regs.gprs[reg1] & PFMF_MC;
1041         }
1042
1043         nq = vcpu->run->s.regs.gprs[reg1] & PFMF_NQ;
1044         key = vcpu->run->s.regs.gprs[reg1] & PFMF_KEY;
1045         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
1046         start = kvm_s390_logical_to_effective(vcpu, start);
1047
1048         if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1049                 if (kvm_s390_check_low_addr_prot_real(vcpu, start))
1050                         return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
1051         }
1052
1053         switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1054         case 0x00000000:
1055                 /* only 4k frames specify a real address */
1056                 start = kvm_s390_real_to_abs(vcpu, start);
1057                 end = (start + PAGE_SIZE) & ~(PAGE_SIZE - 1);
1058                 break;
1059         case 0x00001000:
1060                 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
1061                 break;
1062         case 0x00002000:
1063                 /* only support 2G frame size if EDAT2 is available and we are
1064                    not in 24-bit addressing mode */
1065                 if (!test_kvm_facility(vcpu->kvm, 78) ||
1066                     psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_24BIT)
1067                         return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1068                 end = (start + _REGION3_SIZE) & ~(_REGION3_SIZE - 1);
1069                 break;
1070         default:
1071                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1072         }
1073
1074         while (start != end) {
1075                 unsigned long vmaddr;
1076                 bool unlocked = false;
1077
1078                 /* Translate guest address to host address */
1079                 vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
1080                 if (kvm_is_error_hva(vmaddr))
1081                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1082
1083                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1084                         if (kvm_clear_guest(vcpu->kvm, start, PAGE_SIZE))
1085                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1086                 }
1087
1088                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
1089                         int rc = kvm_s390_skey_check_enable(vcpu);
1090
1091                         if (rc)
1092                                 return rc;
1093                         mmap_read_lock(current->mm);
1094                         rc = cond_set_guest_storage_key(current->mm, vmaddr,
1095                                                         key, NULL, nq, mr, mc);
1096                         if (rc < 0) {
1097                                 rc = fixup_user_fault(current->mm, vmaddr,
1098                                                       FAULT_FLAG_WRITE, &unlocked);
1099                                 rc = !rc ? -EAGAIN : rc;
1100                         }
1101                         mmap_read_unlock(current->mm);
1102                         if (rc == -EFAULT)
1103                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1104                         if (rc == -EAGAIN)
1105                                 continue;
1106                         if (rc < 0)
1107                                 return rc;
1108                 }
1109                 start += PAGE_SIZE;
1110         }
1111         if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1112                 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT) {
1113                         vcpu->run->s.regs.gprs[reg2] = end;
1114                 } else {
1115                         vcpu->run->s.regs.gprs[reg2] &= ~0xffffffffUL;
1116                         end = kvm_s390_logical_to_effective(vcpu, end);
1117                         vcpu->run->s.regs.gprs[reg2] |= end;
1118                 }
1119         }
1120         return 0;
1121 }
1122
1123 /*
1124  * Must be called with relevant read locks held (kvm->mm->mmap_lock, kvm->srcu)
1125  */
1126 static inline int __do_essa(struct kvm_vcpu *vcpu, const int orc)
1127 {
1128         int r1, r2, nappended, entries;
1129         unsigned long gfn, hva, res, pgstev, ptev;
1130         unsigned long *cbrlo;
1131
1132         /*
1133          * We don't need to set SD.FPF.SK to 1 here, because if we have a
1134          * machine check here we either handle it or crash
1135          */
1136
1137         kvm_s390_get_regs_rre(vcpu, &r1, &r2);
1138         gfn = vcpu->run->s.regs.gprs[r2] >> PAGE_SHIFT;
1139         hva = gfn_to_hva(vcpu->kvm, gfn);
1140         entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1141
1142         if (kvm_is_error_hva(hva))
1143                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1144
1145         nappended = pgste_perform_essa(vcpu->kvm->mm, hva, orc, &ptev, &pgstev);
1146         if (nappended < 0) {
1147                 res = orc ? 0x10 : 0;
1148                 vcpu->run->s.regs.gprs[r1] = res; /* Exception Indication */
1149                 return 0;
1150         }
1151         res = (pgstev & _PGSTE_GPS_USAGE_MASK) >> 22;
1152         /*
1153          * Set the block-content state part of the result. 0 means resident, so
1154          * nothing to do if the page is valid. 2 is for preserved pages
1155          * (non-present and non-zero), and 3 for zero pages (non-present and
1156          * zero).
1157          */
1158         if (ptev & _PAGE_INVALID) {
1159                 res |= 2;
1160                 if (pgstev & _PGSTE_GPS_ZERO)
1161                         res |= 1;
1162         }
1163         if (pgstev & _PGSTE_GPS_NODAT)
1164                 res |= 0x20;
1165         vcpu->run->s.regs.gprs[r1] = res;
1166         /*
1167          * It is possible that all the normal 511 slots were full, in which case
1168          * we will now write in the 512th slot, which is reserved for host use.
1169          * In both cases we let the normal essa handling code process all the
1170          * slots, including the reserved one, if needed.
1171          */
1172         if (nappended > 0) {
1173                 cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo & PAGE_MASK);
1174                 cbrlo[entries] = gfn << PAGE_SHIFT;
1175         }
1176
1177         if (orc) {
1178                 struct kvm_memory_slot *ms = gfn_to_memslot(vcpu->kvm, gfn);
1179
1180                 /* Increment only if we are really flipping the bit */
1181                 if (ms && !test_and_set_bit(gfn - ms->base_gfn, kvm_second_dirty_bitmap(ms)))
1182                         atomic64_inc(&vcpu->kvm->arch.cmma_dirty_pages);
1183         }
1184
1185         return nappended;
1186 }
1187
1188 static int handle_essa(struct kvm_vcpu *vcpu)
1189 {
1190         /* entries expected to be 1FF */
1191         int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1192         unsigned long *cbrlo;
1193         struct gmap *gmap;
1194         int i, orc;
1195
1196         VCPU_EVENT(vcpu, 4, "ESSA: release %d pages", entries);
1197         gmap = vcpu->arch.gmap;
1198         vcpu->stat.instruction_essa++;
1199         if (!vcpu->kvm->arch.use_cmma)
1200                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1201
1202         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1203                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1204         /* Check for invalid operation request code */
1205         orc = (vcpu->arch.sie_block->ipb & 0xf0000000) >> 28;
1206         /* ORCs 0-6 are always valid */
1207         if (orc > (test_kvm_facility(vcpu->kvm, 147) ? ESSA_SET_STABLE_NODAT
1208                                                 : ESSA_SET_STABLE_IF_RESIDENT))
1209                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1210
1211         if (!vcpu->kvm->arch.migration_mode) {
1212                 /*
1213                  * CMMA is enabled in the KVM settings, but is disabled in
1214                  * the SIE block and in the mm_context, and we are not doing
1215                  * a migration. Enable CMMA in the mm_context.
1216                  * Since we need to take a write lock to write to the context
1217                  * to avoid races with storage keys handling, we check if the
1218                  * value really needs to be written to; if the value is
1219                  * already correct, we do nothing and avoid the lock.
1220                  */
1221                 if (vcpu->kvm->mm->context.uses_cmm == 0) {
1222                         mmap_write_lock(vcpu->kvm->mm);
1223                         vcpu->kvm->mm->context.uses_cmm = 1;
1224                         mmap_write_unlock(vcpu->kvm->mm);
1225                 }
1226                 /*
1227                  * If we are here, we are supposed to have CMMA enabled in
1228                  * the SIE block. Enabling CMMA works on a per-CPU basis,
1229                  * while the context use_cmma flag is per process.
1230                  * It's possible that the context flag is enabled and the
1231                  * SIE flag is not, so we set the flag always; if it was
1232                  * already set, nothing changes, otherwise we enable it
1233                  * on this CPU too.
1234                  */
1235                 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
1236                 /* Retry the ESSA instruction */
1237                 kvm_s390_retry_instr(vcpu);
1238         } else {
1239                 int srcu_idx;
1240
1241                 mmap_read_lock(vcpu->kvm->mm);
1242                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1243                 i = __do_essa(vcpu, orc);
1244                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1245                 mmap_read_unlock(vcpu->kvm->mm);
1246                 if (i < 0)
1247                         return i;
1248                 /* Account for the possible extra cbrl entry */
1249                 entries += i;
1250         }
1251         vcpu->arch.sie_block->cbrlo &= PAGE_MASK;       /* reset nceo */
1252         cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
1253         mmap_read_lock(gmap->mm);
1254         for (i = 0; i < entries; ++i)
1255                 __gmap_zap(gmap, cbrlo[i]);
1256         mmap_read_unlock(gmap->mm);
1257         return 0;
1258 }
1259
1260 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
1261 {
1262         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1263         case 0x8a:
1264         case 0x8e:
1265         case 0x8f:
1266                 return handle_ipte_interlock(vcpu);
1267         case 0x8d:
1268                 return handle_epsw(vcpu);
1269         case 0xab:
1270                 return handle_essa(vcpu);
1271         case 0xaf:
1272                 return handle_pfmf(vcpu);
1273         default:
1274                 return -EOPNOTSUPP;
1275         }
1276 }
1277
1278 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
1279 {
1280         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1281         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1282         int reg, rc, nr_regs;
1283         u32 ctl_array[16];
1284         u64 ga;
1285         u8 ar;
1286
1287         vcpu->stat.instruction_lctl++;
1288
1289         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1290                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1291
1292         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1293
1294         if (ga & 3)
1295                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1296
1297         VCPU_EVENT(vcpu, 4, "LCTL: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1298         trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
1299
1300         nr_regs = ((reg3 - reg1) & 0xf) + 1;
1301         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1302         if (rc)
1303                 return kvm_s390_inject_prog_cond(vcpu, rc);
1304         reg = reg1;
1305         nr_regs = 0;
1306         do {
1307                 vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
1308                 vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
1309                 if (reg == reg3)
1310                         break;
1311                 reg = (reg + 1) % 16;
1312         } while (1);
1313         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1314         return 0;
1315 }
1316
1317 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
1318 {
1319         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1320         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1321         int reg, rc, nr_regs;
1322         u32 ctl_array[16];
1323         u64 ga;
1324         u8 ar;
1325
1326         vcpu->stat.instruction_stctl++;
1327
1328         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1329                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1330
1331         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1332
1333         if (ga & 3)
1334                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1335
1336         VCPU_EVENT(vcpu, 4, "STCTL r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1337         trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
1338
1339         reg = reg1;
1340         nr_regs = 0;
1341         do {
1342                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1343                 if (reg == reg3)
1344                         break;
1345                 reg = (reg + 1) % 16;
1346         } while (1);
1347         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1348         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1349 }
1350
1351 static int handle_lctlg(struct kvm_vcpu *vcpu)
1352 {
1353         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1354         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1355         int reg, rc, nr_regs;
1356         u64 ctl_array[16];
1357         u64 ga;
1358         u8 ar;
1359
1360         vcpu->stat.instruction_lctlg++;
1361
1362         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1363                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1364
1365         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1366
1367         if (ga & 7)
1368                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1369
1370         VCPU_EVENT(vcpu, 4, "LCTLG: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1371         trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
1372
1373         nr_regs = ((reg3 - reg1) & 0xf) + 1;
1374         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1375         if (rc)
1376                 return kvm_s390_inject_prog_cond(vcpu, rc);
1377         reg = reg1;
1378         nr_regs = 0;
1379         do {
1380                 vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
1381                 if (reg == reg3)
1382                         break;
1383                 reg = (reg + 1) % 16;
1384         } while (1);
1385         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1386         return 0;
1387 }
1388
1389 static int handle_stctg(struct kvm_vcpu *vcpu)
1390 {
1391         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1392         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1393         int reg, rc, nr_regs;
1394         u64 ctl_array[16];
1395         u64 ga;
1396         u8 ar;
1397
1398         vcpu->stat.instruction_stctg++;
1399
1400         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1401                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1402
1403         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1404
1405         if (ga & 7)
1406                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1407
1408         VCPU_EVENT(vcpu, 4, "STCTG r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1409         trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
1410
1411         reg = reg1;
1412         nr_regs = 0;
1413         do {
1414                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1415                 if (reg == reg3)
1416                         break;
1417                 reg = (reg + 1) % 16;
1418         } while (1);
1419         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1420         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1421 }
1422
1423 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
1424 {
1425         switch (vcpu->arch.sie_block->ipb & 0x000000ff) {
1426         case 0x25:
1427                 return handle_stctg(vcpu);
1428         case 0x2f:
1429                 return handle_lctlg(vcpu);
1430         case 0x60:
1431         case 0x61:
1432         case 0x62:
1433                 return handle_ri(vcpu);
1434         default:
1435                 return -EOPNOTSUPP;
1436         }
1437 }
1438
1439 static int handle_tprot(struct kvm_vcpu *vcpu)
1440 {
1441         u64 address1, address2;
1442         unsigned long hva, gpa;
1443         int ret = 0, cc = 0;
1444         bool writable;
1445         u8 ar;
1446
1447         vcpu->stat.instruction_tprot++;
1448
1449         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1450                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1451
1452         kvm_s390_get_base_disp_sse(vcpu, &address1, &address2, &ar, NULL);
1453
1454         /* we only handle the Linux memory detection case:
1455          * access key == 0
1456          * everything else goes to userspace. */
1457         if (address2 & 0xf0)
1458                 return -EOPNOTSUPP;
1459         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1460                 ipte_lock(vcpu);
1461         ret = guest_translate_address(vcpu, address1, ar, &gpa, GACC_STORE);
1462         if (ret == PGM_PROTECTION) {
1463                 /* Write protected? Try again with read-only... */
1464                 cc = 1;
1465                 ret = guest_translate_address(vcpu, address1, ar, &gpa,
1466                                               GACC_FETCH);
1467         }
1468         if (ret) {
1469                 if (ret == PGM_ADDRESSING || ret == PGM_TRANSLATION_SPEC) {
1470                         ret = kvm_s390_inject_program_int(vcpu, ret);
1471                 } else if (ret > 0) {
1472                         /* Translation not available */
1473                         kvm_s390_set_psw_cc(vcpu, 3);
1474                         ret = 0;
1475                 }
1476                 goto out_unlock;
1477         }
1478
1479         hva = gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
1480         if (kvm_is_error_hva(hva)) {
1481                 ret = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1482         } else {
1483                 if (!writable)
1484                         cc = 1;         /* Write not permitted ==> read-only */
1485                 kvm_s390_set_psw_cc(vcpu, cc);
1486                 /* Note: CC2 only occurs for storage keys (not supported yet) */
1487         }
1488 out_unlock:
1489         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1490                 ipte_unlock(vcpu);
1491         return ret;
1492 }
1493
1494 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
1495 {
1496         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1497         case 0x01:
1498                 return handle_tprot(vcpu);
1499         default:
1500                 return -EOPNOTSUPP;
1501         }
1502 }
1503
1504 static int handle_sckpf(struct kvm_vcpu *vcpu)
1505 {
1506         u32 value;
1507
1508         vcpu->stat.instruction_sckpf++;
1509
1510         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1511                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1512
1513         if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
1514                 return kvm_s390_inject_program_int(vcpu,
1515                                                    PGM_SPECIFICATION);
1516
1517         value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
1518         vcpu->arch.sie_block->todpr = value;
1519
1520         return 0;
1521 }
1522
1523 static int handle_ptff(struct kvm_vcpu *vcpu)
1524 {
1525         vcpu->stat.instruction_ptff++;
1526
1527         /* we don't emulate any control instructions yet */
1528         kvm_s390_set_psw_cc(vcpu, 3);
1529         return 0;
1530 }
1531
1532 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1533 {
1534         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1535         case 0x04:
1536                 return handle_ptff(vcpu);
1537         case 0x07:
1538                 return handle_sckpf(vcpu);
1539         default:
1540                 return -EOPNOTSUPP;
1541         }
1542 }