arch/s390/kvm/priv.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * handling privileged instructions
   4  *
   5  * Copyright IBM Corp. 2008, 2018
   6  *
   7  *    Author(s): Carsten Otte <cotte@de.ibm.com>
   8  *               Christian Borntraeger <borntraeger@de.ibm.com>
   9  */
  10
  11 #include <linux/kvm.h>
  12 #include <linux/gfp.h>
  13 #include <linux/errno.h>
  14 #include <linux/compat.h>
  15 #include <linux/mm_types.h>
  16
  17 #include <asm/asm-offsets.h>
  18 #include <asm/facility.h>
  19 #include <asm/current.h>
  20 #include <asm/debug.h>
  21 #include <asm/ebcdic.h>
  22 #include <asm/sysinfo.h>
  23 #include <asm/pgtable.h>
  24 #include <asm/page-states.h>
  25 #include <asm/pgalloc.h>
  26 #include <asm/gmap.h>
  27 #include <asm/io.h>
  28 #include <asm/ptrace.h>
  29 #include <asm/sclp.h>
  30 #include "gaccess.h"
  31 #include "kvm-s390.h"
  32 #include "trace.h"
  33
  34 static int handle_ri(struct kvm_vcpu *vcpu)
  35 {
  36         vcpu->stat.instruction_ri++;
  37
  38         if (test_kvm_facility(vcpu->kvm, 64)) {
  39                 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (lazy)");
  40                 vcpu->arch.sie_block->ecb3 |= ECB3_RI;
  41                 kvm_s390_retry_instr(vcpu);
  42                 return 0;
  43         } else
  44                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
  45 }
  46
  47 int kvm_s390_handle_aa(struct kvm_vcpu *vcpu)
  48 {
  49         if ((vcpu->arch.sie_block->ipa & 0xf) <= 4)
  50                 return handle_ri(vcpu);
  51         else
  52                 return -EOPNOTSUPP;
  53 }
  54
  55 static int handle_gs(struct kvm_vcpu *vcpu)
  56 {
  57         vcpu->stat.instruction_gs++;
  58
  59         if (test_kvm_facility(vcpu->kvm, 133)) {
  60                 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (lazy)");
  61                 preempt_disable();
  62                 __ctl_set_bit(2, 4);
  63                 current->thread.gs_cb = (struct gs_cb *)&vcpu->run->s.regs.gscb;
  64                 restore_gs_cb(current->thread.gs_cb);
  65                 preempt_enable();
  66                 vcpu->arch.sie_block->ecb |= ECB_GS;
  67                 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
  68                 vcpu->arch.gs_enabled = 1;
  69                 kvm_s390_retry_instr(vcpu);
  70                 return 0;
  71         } else
  72                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
  73 }
  74
  75 int kvm_s390_handle_e3(struct kvm_vcpu *vcpu)
  76 {
  77         int code = vcpu->arch.sie_block->ipb & 0xff;
  78
  79         if (code == 0x49 || code == 0x4d)
  80                 return handle_gs(vcpu);
  81         else
  82                 return -EOPNOTSUPP;
  83 }
  84 /* Handle SCK (SET CLOCK) interception */
  85 static int handle_set_clock(struct kvm_vcpu *vcpu)
  86 {
  87         struct kvm_s390_vm_tod_clock gtod = { 0 };
  88         int rc;
  89         u8 ar;
  90         u64 op2;
  91
  92         vcpu->stat.instruction_sck++;
  93
  94         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
  95                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
  96
  97         op2 = kvm_s390_get_base_disp_s(vcpu, &ar);
  98         if (op2 & 7)    /* Operand must be on a doubleword boundary */
  99                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 100         rc = read_guest(vcpu, op2, ar, &gtod.tod, sizeof(gtod.tod));
 101         if (rc)
 102                 return kvm_s390_inject_prog_cond(vcpu, rc);
 103
 104         VCPU_EVENT(vcpu, 3, "SCK: setting guest TOD to 0x%llx", gtod.tod);
 105         kvm_s390_set_tod_clock(vcpu->kvm, &gtod);
 106
 107         kvm_s390_set_psw_cc(vcpu, 0);
 108         return 0;
 109 }
 110
 111 static int handle_set_prefix(struct kvm_vcpu *vcpu)
 112 {
 113         u64 operand2;
 114         u32 address;
 115         int rc;
 116         u8 ar;
 117
 118         vcpu->stat.instruction_spx++;
 119
 120         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 121                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 122
 123         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 124
 125         /* must be word boundary */
 126         if (operand2 & 3)
 127                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 128
 129         /* get the value */
 130         rc = read_guest(vcpu, operand2, ar, &address, sizeof(address));
 131         if (rc)
 132                 return kvm_s390_inject_prog_cond(vcpu, rc);
 133
 134         address &= 0x7fffe000u;
 135
 136         /*
 137          * Make sure the new value is valid memory. We only need to check the
 138          * first page, since address is 8k aligned and memory pieces are always
 139          * at least 1MB aligned and have at least a size of 1MB.
 140          */
 141         if (kvm_is_error_gpa(vcpu->kvm, address))
 142                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 143
 144         kvm_s390_set_prefix(vcpu, address);
 145         trace_kvm_s390_handle_prefix(vcpu, 1, address);
 146         return 0;
 147 }
 148
 149 static int handle_store_prefix(struct kvm_vcpu *vcpu)
 150 {
 151         u64 operand2;
 152         u32 address;
 153         int rc;
 154         u8 ar;
 155
 156         vcpu->stat.instruction_stpx++;
 157
 158         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 159                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 160
 161         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 162
 163         /* must be word boundary */
 164         if (operand2 & 3)
 165                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 166
 167         address = kvm_s390_get_prefix(vcpu);
 168
 169         /* get the value */
 170         rc = write_guest(vcpu, operand2, ar, &address, sizeof(address));
 171         if (rc)
 172                 return kvm_s390_inject_prog_cond(vcpu, rc);
 173
 174         VCPU_EVENT(vcpu, 3, "STPX: storing prefix 0x%x into 0x%llx", address, operand2);
 175         trace_kvm_s390_handle_prefix(vcpu, 0, address);
 176         return 0;
 177 }
 178
 179 static int handle_store_cpu_address(struct kvm_vcpu *vcpu)
 180 {
 181         u16 vcpu_id = vcpu->vcpu_id;
 182         u64 ga;
 183         int rc;
 184         u8 ar;
 185
 186         vcpu->stat.instruction_stap++;
 187
 188         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 189                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 190
 191         ga = kvm_s390_get_base_disp_s(vcpu, &ar);
 192
 193         if (ga & 1)
 194                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 195
 196         rc = write_guest(vcpu, ga, ar, &vcpu_id, sizeof(vcpu_id));
 197         if (rc)
 198                 return kvm_s390_inject_prog_cond(vcpu, rc);
 199
 200         VCPU_EVENT(vcpu, 3, "STAP: storing cpu address (%u) to 0x%llx", vcpu_id, ga);
 201         trace_kvm_s390_handle_stap(vcpu, ga);
 202         return 0;
 203 }
 204
 205 int kvm_s390_skey_check_enable(struct kvm_vcpu *vcpu)
 206 {
 207         int rc;
 208         struct kvm_s390_sie_block *sie_block = vcpu->arch.sie_block;
 209
 210         trace_kvm_s390_skey_related_inst(vcpu);
 211         /* Already enabled? */
 212         if (vcpu->kvm->arch.use_skf &&
 213             !(sie_block->ictl & (ICTL_ISKE | ICTL_SSKE | ICTL_RRBE)) &&
 214             !kvm_s390_test_cpuflags(vcpu, CPUSTAT_KSS))
 215                 return 0;
 216
 217         rc = s390_enable_skey();
 218         VCPU_EVENT(vcpu, 3, "enabling storage keys for guest: %d", rc);
 219         if (rc)
 220                 return rc;
 221
 222         if (kvm_s390_test_cpuflags(vcpu, CPUSTAT_KSS))
 223                 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_KSS);
 224         if (!vcpu->kvm->arch.use_skf)
 225                 sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
 226         else
 227                 sie_block->ictl &= ~(ICTL_ISKE | ICTL_SSKE | ICTL_RRBE);
 228         return 0;
 229 }
 230
 231 static int try_handle_skey(struct kvm_vcpu *vcpu)
 232 {
 233         int rc;
 234
 235         rc = kvm_s390_skey_check_enable(vcpu);
 236         if (rc)
 237                 return rc;
 238         if (vcpu->kvm->arch.use_skf) {
 239                 /* with storage-key facility, SIE interprets it for us */
 240                 kvm_s390_retry_instr(vcpu);
 241                 VCPU_EVENT(vcpu, 4, "%s", "retrying storage key operation");
 242                 return -EAGAIN;
 243         }
 244         return 0;
 245 }
 246
 247 static int handle_iske(struct kvm_vcpu *vcpu)
 248 {
 249         unsigned long gaddr, vmaddr;
 250         unsigned char key;
 251         int reg1, reg2;
 252         bool unlocked;
 253         int rc;
 254
 255         vcpu->stat.instruction_iske++;
 256
 257         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 258                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 259
 260         rc = try_handle_skey(vcpu);
 261         if (rc)
 262                 return rc != -EAGAIN ? rc : 0;
 263
 264         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 265
 266         gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 267         gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
 268         gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
 269         vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr));
 270         if (kvm_is_error_hva(vmaddr))
 271                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 272 retry:
 273         unlocked = false;
 274         down_read(&current->mm->mmap_sem);
 275         rc = get_guest_storage_key(current->mm, vmaddr, &key);
 276
 277         if (rc) {
 278                 rc = fixup_user_fault(current, current->mm, vmaddr,
 279                                       FAULT_FLAG_WRITE, &unlocked);
 280                 if (!rc) {
 281                         up_read(&current->mm->mmap_sem);
 282                         goto retry;
 283                 }
 284         }
 285         if (rc)
 286                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 287         up_read(&current->mm->mmap_sem);
 288         vcpu->run->s.regs.gprs[reg1] &= ~0xff;
 289         vcpu->run->s.regs.gprs[reg1] |= key;
 290         return 0;
 291 }
 292
 293 static int handle_rrbe(struct kvm_vcpu *vcpu)
 294 {
 295         unsigned long vmaddr, gaddr;
 296         int reg1, reg2;
 297         bool unlocked;
 298         int rc;
 299
 300         vcpu->stat.instruction_rrbe++;
 301
 302         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 303                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 304
 305         rc = try_handle_skey(vcpu);
 306         if (rc)
 307                 return rc != -EAGAIN ? rc : 0;
 308
 309         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 310
 311         gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 312         gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
 313         gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
 314         vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(gaddr));
 315         if (kvm_is_error_hva(vmaddr))
 316                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 317 retry:
 318         unlocked = false;
 319         down_read(&current->mm->mmap_sem);
 320         rc = reset_guest_reference_bit(current->mm, vmaddr);
 321         if (rc < 0) {
 322                 rc = fixup_user_fault(current, current->mm, vmaddr,
 323                                       FAULT_FLAG_WRITE, &unlocked);
 324                 if (!rc) {
 325                         up_read(&current->mm->mmap_sem);
 326                         goto retry;
 327                 }
 328         }
 329         if (rc < 0)
 330                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 331         up_read(&current->mm->mmap_sem);
 332         kvm_s390_set_psw_cc(vcpu, rc);
 333         return 0;
 334 }
 335
 336 #define SSKE_NQ 0x8
 337 #define SSKE_MR 0x4
 338 #define SSKE_MC 0x2
 339 #define SSKE_MB 0x1
 340 static int handle_sske(struct kvm_vcpu *vcpu)
 341 {
 342         unsigned char m3 = vcpu->arch.sie_block->ipb >> 28;
 343         unsigned long start, end;
 344         unsigned char key, oldkey;
 345         int reg1, reg2;
 346         bool unlocked;
 347         int rc;
 348
 349         vcpu->stat.instruction_sske++;
 350
 351         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 352                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 353
 354         rc = try_handle_skey(vcpu);
 355         if (rc)
 356                 return rc != -EAGAIN ? rc : 0;
 357
 358         if (!test_kvm_facility(vcpu->kvm, 8))
 359                 m3 &= ~SSKE_MB;
 360         if (!test_kvm_facility(vcpu->kvm, 10))
 361                 m3 &= ~(SSKE_MC | SSKE_MR);
 362         if (!test_kvm_facility(vcpu->kvm, 14))
 363                 m3 &= ~SSKE_NQ;
 364
 365         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 366
 367         key = vcpu->run->s.regs.gprs[reg1] & 0xfe;
 368         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 369         start = kvm_s390_logical_to_effective(vcpu, start);
 370         if (m3 & SSKE_MB) {
 371                 /* start already designates an absolute address */
 372                 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
 373         } else {
 374                 start = kvm_s390_real_to_abs(vcpu, start);
 375                 end = start + PAGE_SIZE;
 376         }
 377
 378         while (start != end) {
 379                 unsigned long vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
 380                 unlocked = false;
 381
 382                 if (kvm_is_error_hva(vmaddr))
 383                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 384
 385                 down_read(&current->mm->mmap_sem);
 386                 rc = cond_set_guest_storage_key(current->mm, vmaddr, key, &oldkey,
 387                                                 m3 & SSKE_NQ, m3 & SSKE_MR,
 388                                                 m3 & SSKE_MC);
 389
 390                 if (rc < 0) {
 391                         rc = fixup_user_fault(current, current->mm, vmaddr,
 392                                               FAULT_FLAG_WRITE, &unlocked);
 393                         rc = !rc ? -EAGAIN : rc;
 394                 }
 395                 if (rc == -EFAULT)
 396                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 397
 398                 up_read(&current->mm->mmap_sem);
 399                 if (rc >= 0)
 400                         start += PAGE_SIZE;
 401         }
 402
 403         if (m3 & (SSKE_MC | SSKE_MR)) {
 404                 if (m3 & SSKE_MB) {
 405                         /* skey in reg1 is unpredictable */
 406                         kvm_s390_set_psw_cc(vcpu, 3);
 407                 } else {
 408                         kvm_s390_set_psw_cc(vcpu, rc);
 409                         vcpu->run->s.regs.gprs[reg1] &= ~0xff00UL;
 410                         vcpu->run->s.regs.gprs[reg1] |= (u64) oldkey << 8;
 411                 }
 412         }
 413         if (m3 & SSKE_MB) {
 414                 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT)
 415                         vcpu->run->s.regs.gprs[reg2] &= ~PAGE_MASK;
 416                 else
 417                         vcpu->run->s.regs.gprs[reg2] &= ~0xfffff000UL;
 418                 end = kvm_s390_logical_to_effective(vcpu, end);
 419                 vcpu->run->s.regs.gprs[reg2] |= end;
 420         }
 421         return 0;
 422 }
 423
 424 static int handle_ipte_interlock(struct kvm_vcpu *vcpu)
 425 {
 426         vcpu->stat.instruction_ipte_interlock++;
 427         if (psw_bits(vcpu->arch.sie_block->gpsw).pstate)
 428                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 429         wait_event(vcpu->kvm->arch.ipte_wq, !ipte_lock_held(vcpu));
 430         kvm_s390_retry_instr(vcpu);
 431         VCPU_EVENT(vcpu, 4, "%s", "retrying ipte interlock operation");
 432         return 0;
 433 }
 434
 435 static int handle_test_block(struct kvm_vcpu *vcpu)
 436 {
 437         gpa_t addr;
 438         int reg2;
 439
 440         vcpu->stat.instruction_tb++;
 441
 442         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 443                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 444
 445         kvm_s390_get_regs_rre(vcpu, NULL, &reg2);
 446         addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 447         addr = kvm_s390_logical_to_effective(vcpu, addr);
 448         if (kvm_s390_check_low_addr_prot_real(vcpu, addr))
 449                 return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
 450         addr = kvm_s390_real_to_abs(vcpu, addr);
 451
 452         if (kvm_is_error_gpa(vcpu->kvm, addr))
 453                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 454         /*
 455          * We don't expect errors on modern systems, and do not care
 456          * about storage keys (yet), so let's just clear the page.
 457          */
 458         if (kvm_clear_guest(vcpu->kvm, addr, PAGE_SIZE))
 459                 return -EFAULT;
 460         kvm_s390_set_psw_cc(vcpu, 0);
 461         vcpu->run->s.regs.gprs[0] = 0;
 462         return 0;
 463 }
 464
 465 static int handle_tpi(struct kvm_vcpu *vcpu)
 466 {
 467         struct kvm_s390_interrupt_info *inti;
 468         unsigned long len;
 469         u32 tpi_data[3];
 470         int rc;
 471         u64 addr;
 472         u8 ar;
 473
 474         vcpu->stat.instruction_tpi++;
 475
 476         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 477         if (addr & 3)
 478                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 479
 480         inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0);
 481         if (!inti) {
 482                 kvm_s390_set_psw_cc(vcpu, 0);
 483                 return 0;
 484         }
 485
 486         tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr;
 487         tpi_data[1] = inti->io.io_int_parm;
 488         tpi_data[2] = inti->io.io_int_word;
 489         if (addr) {
 490                 /*
 491                  * Store the two-word I/O interruption code into the
 492                  * provided area.
 493                  */
 494                 len = sizeof(tpi_data) - 4;
 495                 rc = write_guest(vcpu, addr, ar, &tpi_data, len);
 496                 if (rc) {
 497                         rc = kvm_s390_inject_prog_cond(vcpu, rc);
 498                         goto reinject_interrupt;
 499                 }
 500         } else {
 501                 /*
 502                  * Store the three-word I/O interruption code into
 503                  * the appropriate lowcore area.
 504                  */
 505                 len = sizeof(tpi_data);
 506                 if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len)) {
 507                         /* failed writes to the low core are not recoverable */
 508                         rc = -EFAULT;
 509                         goto reinject_interrupt;
 510                 }
 511         }
 512
 513         /* irq was successfully handed to the guest */
 514         kfree(inti);
 515         kvm_s390_set_psw_cc(vcpu, 1);
 516         return 0;
 517 reinject_interrupt:
 518         /*
 519          * If we encounter a problem storing the interruption code, the
 520          * instruction is suppressed from the guest's view: reinject the
 521          * interrupt.
 522          */
 523         if (kvm_s390_reinject_io_int(vcpu->kvm, inti)) {
 524                 kfree(inti);
 525                 rc = -EFAULT;
 526         }
 527         /* don't set the cc, a pgm irq was injected or we drop to user space */
 528         return rc ? -EFAULT : 0;
 529 }
 530
 531 static int handle_tsch(struct kvm_vcpu *vcpu)
 532 {
 533         struct kvm_s390_interrupt_info *inti = NULL;
 534         const u64 isc_mask = 0xffUL << 24; /* all iscs set */
 535
 536         vcpu->stat.instruction_tsch++;
 537
 538         /* a valid schid has at least one bit set */
 539         if (vcpu->run->s.regs.gprs[1])
 540                 inti = kvm_s390_get_io_int(vcpu->kvm, isc_mask,
 541                                            vcpu->run->s.regs.gprs[1]);
 542
 543         /*
 544          * Prepare exit to userspace.
 545          * We indicate whether we dequeued a pending I/O interrupt
 546          * so that userspace can re-inject it if the instruction gets
 547          * a program check. While this may re-order the pending I/O
 548          * interrupts, this is no problem since the priority is kept
 549          * intact.
 550          */
 551         vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
 552         vcpu->run->s390_tsch.dequeued = !!inti;
 553         if (inti) {
 554                 vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
 555                 vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
 556                 vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
 557                 vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
 558         }
 559         vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
 560         kfree(inti);
 561         return -EREMOTE;
 562 }
 563
 564 static int handle_io_inst(struct kvm_vcpu *vcpu)
 565 {
 566         VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
 567
 568         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 569                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 570
 571         if (vcpu->kvm->arch.css_support) {
 572                 /*
 573                  * Most I/O instructions will be handled by userspace.
 574                  * Exceptions are tpi and the interrupt portion of tsch.
 575                  */
 576                 if (vcpu->arch.sie_block->ipa == 0xb236)
 577                         return handle_tpi(vcpu);
 578                 if (vcpu->arch.sie_block->ipa == 0xb235)
 579                         return handle_tsch(vcpu);
 580                 /* Handle in userspace. */
 581                 vcpu->stat.instruction_io_other++;
 582                 return -EOPNOTSUPP;
 583         } else {
 584                 /*
 585                  * Set condition code 3 to stop the guest from issuing channel
 586                  * I/O instructions.
 587                  */
 588                 kvm_s390_set_psw_cc(vcpu, 3);
 589                 return 0;
 590         }
 591 }
 592
 593 static int handle_stfl(struct kvm_vcpu *vcpu)
 594 {
 595         int rc;
 596         unsigned int fac;
 597
 598         vcpu->stat.instruction_stfl++;
 599
 600         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 601                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 602
 603         /*
 604          * We need to shift the lower 32 facility bits (bit 0-31) from a u64
 605          * into a u32 memory representation. They will remain bits 0-31.
 606          */
 607         fac = *vcpu->kvm->arch.model.fac_list >> 32;
 608         rc = write_guest_lc(vcpu, offsetof(struct lowcore, stfl_fac_list),
 609                             &fac, sizeof(fac));
 610         if (rc)
 611                 return rc;
 612         VCPU_EVENT(vcpu, 3, "STFL: store facility list 0x%x", fac);
 613         trace_kvm_s390_handle_stfl(vcpu, fac);
 614         return 0;
 615 }
 616
 617 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
 618 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
 619 #define PSW_ADDR_24 0x0000000000ffffffUL
 620 #define PSW_ADDR_31 0x000000007fffffffUL
 621
 622 int is_valid_psw(psw_t *psw)
 623 {
 624         if (psw->mask & PSW_MASK_UNASSIGNED)
 625                 return 0;
 626         if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
 627                 if (psw->addr & ~PSW_ADDR_31)
 628                         return 0;
 629         }
 630         if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
 631                 return 0;
 632         if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
 633                 return 0;
 634         if (psw->addr & 1)
 635                 return 0;
 636         return 1;
 637 }
 638
 639 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
 640 {
 641         psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
 642         psw_compat_t new_psw;
 643         u64 addr;
 644         int rc;
 645         u8 ar;
 646
 647         vcpu->stat.instruction_lpsw++;
 648
 649         if (gpsw->mask & PSW_MASK_PSTATE)
 650                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 651
 652         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 653         if (addr & 7)
 654                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 655
 656         rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
 657         if (rc)
 658                 return kvm_s390_inject_prog_cond(vcpu, rc);
 659         if (!(new_psw.mask & PSW32_MASK_BASE))
 660                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 661         gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
 662         gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
 663         gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
 664         if (!is_valid_psw(gpsw))
 665                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 666         return 0;
 667 }
 668
 669 static int handle_lpswe(struct kvm_vcpu *vcpu)
 670 {
 671         psw_t new_psw;
 672         u64 addr;
 673         int rc;
 674         u8 ar;
 675
 676         vcpu->stat.instruction_lpswe++;
 677
 678         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 679                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 680
 681         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 682         if (addr & 7)
 683                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 684         rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
 685         if (rc)
 686                 return kvm_s390_inject_prog_cond(vcpu, rc);
 687         vcpu->arch.sie_block->gpsw = new_psw;
 688         if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
 689                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 690         return 0;
 691 }
 692
 693 static int handle_stidp(struct kvm_vcpu *vcpu)
 694 {
 695         u64 stidp_data = vcpu->kvm->arch.model.cpuid;
 696         u64 operand2;
 697         int rc;
 698         u8 ar;
 699
 700         vcpu->stat.instruction_stidp++;
 701
 702         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 703                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 704
 705         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 706
 707         if (operand2 & 7)
 708                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 709
 710         rc = write_guest(vcpu, operand2, ar, &stidp_data, sizeof(stidp_data));
 711         if (rc)
 712                 return kvm_s390_inject_prog_cond(vcpu, rc);
 713
 714         VCPU_EVENT(vcpu, 3, "STIDP: store cpu id 0x%llx", stidp_data);
 715         return 0;
 716 }
 717
 718 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
 719 {
 720         int cpus = 0;
 721         int n;
 722
 723         cpus = atomic_read(&vcpu->kvm->online_vcpus);
 724
 725         /* deal with other level 3 hypervisors */
 726         if (stsi(mem, 3, 2, 2))
 727                 mem->count = 0;
 728         if (mem->count < 8)
 729                 mem->count++;
 730         for (n = mem->count - 1; n > 0 ; n--)
 731                 memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
 732
 733         memset(&mem->vm[0], 0, sizeof(mem->vm[0]));
 734         mem->vm[0].cpus_total = cpus;
 735         mem->vm[0].cpus_configured = cpus;
 736         mem->vm[0].cpus_standby = 0;
 737         mem->vm[0].cpus_reserved = 0;
 738         mem->vm[0].caf = 1000;
 739         memcpy(mem->vm[0].name, "KVMguest", 8);
 740         ASCEBC(mem->vm[0].name, 8);
 741         memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
 742         ASCEBC(mem->vm[0].cpi, 16);
 743 }
 744
 745 static void insert_stsi_usr_data(struct kvm_vcpu *vcpu, u64 addr, u8 ar,
 746                                  u8 fc, u8 sel1, u16 sel2)
 747 {
 748         vcpu->run->exit_reason = KVM_EXIT_S390_STSI;
 749         vcpu->run->s390_stsi.addr = addr;
 750         vcpu->run->s390_stsi.ar = ar;
 751         vcpu->run->s390_stsi.fc = fc;
 752         vcpu->run->s390_stsi.sel1 = sel1;
 753         vcpu->run->s390_stsi.sel2 = sel2;
 754 }
 755
 756 static int handle_stsi(struct kvm_vcpu *vcpu)
 757 {
 758         int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
 759         int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
 760         int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
 761         unsigned long mem = 0;
 762         u64 operand2;
 763         int rc = 0;
 764         u8 ar;
 765
 766         vcpu->stat.instruction_stsi++;
 767         VCPU_EVENT(vcpu, 3, "STSI: fc: %u sel1: %u sel2: %u", fc, sel1, sel2);
 768
 769         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 770                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 771
 772         if (fc > 3) {
 773                 kvm_s390_set_psw_cc(vcpu, 3);
 774                 return 0;
 775         }
 776
 777         if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
 778             || vcpu->run->s.regs.gprs[1] & 0xffff0000)
 779                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 780
 781         if (fc == 0) {
 782                 vcpu->run->s.regs.gprs[0] = 3 << 28;
 783                 kvm_s390_set_psw_cc(vcpu, 0);
 784                 return 0;
 785         }
 786
 787         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 788
 789         if (operand2 & 0xfff)
 790                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 791
 792         switch (fc) {
 793         case 1: /* same handling for 1 and 2 */
 794         case 2:
 795                 mem = get_zeroed_page(GFP_KERNEL);
 796                 if (!mem)
 797                         goto out_no_data;
 798                 if (stsi((void *) mem, fc, sel1, sel2))
 799                         goto out_no_data;
 800                 break;
 801         case 3:
 802                 if (sel1 != 2 || sel2 != 2)
 803                         goto out_no_data;
 804                 mem = get_zeroed_page(GFP_KERNEL);
 805                 if (!mem)
 806                         goto out_no_data;
 807                 handle_stsi_3_2_2(vcpu, (void *) mem);
 808                 break;
 809         }
 810
 811         rc = write_guest(vcpu, operand2, ar, (void *)mem, PAGE_SIZE);
 812         if (rc) {
 813                 rc = kvm_s390_inject_prog_cond(vcpu, rc);
 814                 goto out;
 815         }
 816         if (vcpu->kvm->arch.user_stsi) {
 817                 insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
 818                 rc = -EREMOTE;
 819         }
 820         trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
 821         free_page(mem);
 822         kvm_s390_set_psw_cc(vcpu, 0);
 823         vcpu->run->s.regs.gprs[0] = 0;
 824         return rc;
 825 out_no_data:
 826         kvm_s390_set_psw_cc(vcpu, 3);
 827 out:
 828         free_page(mem);
 829         return rc;
 830 }
 831
 832 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
 833 {
 834         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
 835         case 0x02:
 836                 return handle_stidp(vcpu);
 837         case 0x04:
 838                 return handle_set_clock(vcpu);
 839         case 0x10:
 840                 return handle_set_prefix(vcpu);
 841         case 0x11:
 842                 return handle_store_prefix(vcpu);
 843         case 0x12:
 844                 return handle_store_cpu_address(vcpu);
 845         case 0x14:
 846                 return kvm_s390_handle_vsie(vcpu);
 847         case 0x21:
 848         case 0x50:
 849                 return handle_ipte_interlock(vcpu);
 850         case 0x29:
 851                 return handle_iske(vcpu);
 852         case 0x2a:
 853                 return handle_rrbe(vcpu);
 854         case 0x2b:
 855                 return handle_sske(vcpu);
 856         case 0x2c:
 857                 return handle_test_block(vcpu);
 858         case 0x30:
 859         case 0x31:
 860         case 0x32:
 861         case 0x33:
 862         case 0x34:
 863         case 0x35:
 864         case 0x36:
 865         case 0x37:
 866         case 0x38:
 867         case 0x39:
 868         case 0x3a:
 869         case 0x3b:
 870         case 0x3c:
 871         case 0x5f:
 872         case 0x74:
 873         case 0x76:
 874                 return handle_io_inst(vcpu);
 875         case 0x56:
 876                 return handle_sthyi(vcpu);
 877         case 0x7d:
 878                 return handle_stsi(vcpu);
 879         case 0xb1:
 880                 return handle_stfl(vcpu);
 881         case 0xb2:
 882                 return handle_lpswe(vcpu);
 883         default:
 884                 return -EOPNOTSUPP;
 885         }
 886 }
 887
 888 static int handle_epsw(struct kvm_vcpu *vcpu)
 889 {
 890         int reg1, reg2;
 891
 892         vcpu->stat.instruction_epsw++;
 893
 894         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 895
 896         /* This basically extracts the mask half of the psw. */
 897         vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
 898         vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
 899         if (reg2) {
 900                 vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
 901                 vcpu->run->s.regs.gprs[reg2] |=
 902                         vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
 903         }
 904         return 0;
 905 }
 906
 907 #define PFMF_RESERVED   0xfffc0101UL
 908 #define PFMF_SK         0x00020000UL
 909 #define PFMF_CF         0x00010000UL
 910 #define PFMF_UI         0x00008000UL
 911 #define PFMF_FSC        0x00007000UL
 912 #define PFMF_NQ         0x00000800UL
 913 #define PFMF_MR         0x00000400UL
 914 #define PFMF_MC         0x00000200UL
 915 #define PFMF_KEY        0x000000feUL
 916
 917 static int handle_pfmf(struct kvm_vcpu *vcpu)
 918 {
 919         bool mr = false, mc = false, nq;
 920         int reg1, reg2;
 921         unsigned long start, end;
 922         unsigned char key;
 923
 924         vcpu->stat.instruction_pfmf++;
 925
 926         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 927
 928         if (!test_kvm_facility(vcpu->kvm, 8))
 929                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
 930
 931         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 932                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 933
 934         if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
 935                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 936
 937         /* Only provide non-quiescing support if enabled for the guest */
 938         if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ &&
 939             !test_kvm_facility(vcpu->kvm, 14))
 940                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 941
 942         /* Only provide conditional-SSKE support if enabled for the guest */
 943         if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK &&
 944             test_kvm_facility(vcpu->kvm, 10)) {
 945                 mr = vcpu->run->s.regs.gprs[reg1] & PFMF_MR;
 946                 mc = vcpu->run->s.regs.gprs[reg1] & PFMF_MC;
 947         }
 948
 949         nq = vcpu->run->s.regs.gprs[reg1] & PFMF_NQ;
 950         key = vcpu->run->s.regs.gprs[reg1] & PFMF_KEY;
 951         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 952         start = kvm_s390_logical_to_effective(vcpu, start);
 953
 954         if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
 955                 if (kvm_s390_check_low_addr_prot_real(vcpu, start))
 956                         return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
 957         }
 958
 959         switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
 960         case 0x00000000:
 961                 /* only 4k frames specify a real address */
 962                 start = kvm_s390_real_to_abs(vcpu, start);
 963                 end = (start + PAGE_SIZE) & ~(PAGE_SIZE - 1);
 964                 break;
 965         case 0x00001000:
 966                 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
 967                 break;
 968         case 0x00002000:
 969                 /* only support 2G frame size if EDAT2 is available and we are
 970                    not in 24-bit addressing mode */
 971                 if (!test_kvm_facility(vcpu->kvm, 78) ||
 972                     psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_24BIT)
 973                         return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 974                 end = (start + _REGION3_SIZE) & ~(_REGION3_SIZE - 1);
 975                 break;
 976         default:
 977                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 978         }
 979
 980         while (start != end) {
 981                 unsigned long vmaddr;
 982                 bool unlocked = false;
 983
 984                 /* Translate guest address to host address */
 985                 vmaddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
 986                 if (kvm_is_error_hva(vmaddr))
 987                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 988
 989                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
 990                         if (clear_user((void __user *)vmaddr, PAGE_SIZE))
 991                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 992                 }
 993
 994                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
 995                         int rc = kvm_s390_skey_check_enable(vcpu);
 996
 997                         if (rc)
 998                                 return rc;
 999                         down_read(&current->mm->mmap_sem);
1000                         rc = cond_set_guest_storage_key(current->mm, vmaddr,
1001                                                         key, NULL, nq, mr, mc);
1002                         if (rc < 0) {
1003                                 rc = fixup_user_fault(current, current->mm, vmaddr,
1004                                                       FAULT_FLAG_WRITE, &unlocked);
1005                                 rc = !rc ? -EAGAIN : rc;
1006                         }
1007                         if (rc == -EFAULT)
1008                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1009
1010                         up_read(&current->mm->mmap_sem);
1011                         if (rc >= 0)
1012                                 start += PAGE_SIZE;
1013                 }
1014         }
1015         if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1016                 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT) {
1017                         vcpu->run->s.regs.gprs[reg2] = end;
1018                 } else {
1019                         vcpu->run->s.regs.gprs[reg2] &= ~0xffffffffUL;
1020                         end = kvm_s390_logical_to_effective(vcpu, end);
1021                         vcpu->run->s.regs.gprs[reg2] |= end;
1022                 }
1023         }
1024         return 0;
1025 }
1026
1027 static inline int do_essa(struct kvm_vcpu *vcpu, const int orc)
1028 {
1029         struct kvm_s390_migration_state *ms = vcpu->kvm->arch.migration_state;
1030         int r1, r2, nappended, entries;
1031         unsigned long gfn, hva, res, pgstev, ptev;
1032         unsigned long *cbrlo;
1033
1034         /*
1035          * We don't need to set SD.FPF.SK to 1 here, because if we have a
1036          * machine check here we either handle it or crash
1037          */
1038
1039         kvm_s390_get_regs_rre(vcpu, &r1, &r2);
1040         gfn = vcpu->run->s.regs.gprs[r2] >> PAGE_SHIFT;
1041         hva = gfn_to_hva(vcpu->kvm, gfn);
1042         entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1043
1044         if (kvm_is_error_hva(hva))
1045                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1046
1047         nappended = pgste_perform_essa(vcpu->kvm->mm, hva, orc, &ptev, &pgstev);
1048         if (nappended < 0) {
1049                 res = orc ? 0x10 : 0;
1050                 vcpu->run->s.regs.gprs[r1] = res; /* Exception Indication */
1051                 return 0;
1052         }
1053         res = (pgstev & _PGSTE_GPS_USAGE_MASK) >> 22;
1054         /*
1055          * Set the block-content state part of the result. 0 means resident, so
1056          * nothing to do if the page is valid. 2 is for preserved pages
1057          * (non-present and non-zero), and 3 for zero pages (non-present and
1058          * zero).
1059          */
1060         if (ptev & _PAGE_INVALID) {
1061                 res |= 2;
1062                 if (pgstev & _PGSTE_GPS_ZERO)
1063                         res |= 1;
1064         }
1065         if (pgstev & _PGSTE_GPS_NODAT)
1066                 res |= 0x20;
1067         vcpu->run->s.regs.gprs[r1] = res;
1068         /*
1069          * It is possible that all the normal 511 slots were full, in which case
1070          * we will now write in the 512th slot, which is reserved for host use.
1071          * In both cases we let the normal essa handling code process all the
1072          * slots, including the reserved one, if needed.
1073          */
1074         if (nappended > 0) {
1075                 cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo & PAGE_MASK);
1076                 cbrlo[entries] = gfn << PAGE_SHIFT;
1077         }
1078
1079         if (orc && gfn < ms->bitmap_size) {
1080                 /* increment only if we are really flipping the bit to 1 */
1081                 if (!test_and_set_bit(gfn, ms->pgste_bitmap))
1082                         atomic64_inc(&ms->dirty_pages);
1083         }
1084
1085         return nappended;
1086 }
1087
1088 static int handle_essa(struct kvm_vcpu *vcpu)
1089 {
1090         /* entries expected to be 1FF */
1091         int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1092         unsigned long *cbrlo;
1093         struct gmap *gmap;
1094         int i, orc;
1095
1096         VCPU_EVENT(vcpu, 4, "ESSA: release %d pages", entries);
1097         gmap = vcpu->arch.gmap;
1098         vcpu->stat.instruction_essa++;
1099         if (!vcpu->kvm->arch.use_cmma)
1100                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1101
1102         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1103                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1104         /* Check for invalid operation request code */
1105         orc = (vcpu->arch.sie_block->ipb & 0xf0000000) >> 28;
1106         /* ORCs 0-6 are always valid */
1107         if (orc > (test_kvm_facility(vcpu->kvm, 147) ? ESSA_SET_STABLE_NODAT
1108                                                 : ESSA_SET_STABLE_IF_RESIDENT))
1109                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1110
1111         if (likely(!vcpu->kvm->arch.migration_state)) {
1112                 /*
1113                  * CMMA is enabled in the KVM settings, but is disabled in
1114                  * the SIE block and in the mm_context, and we are not doing
1115                  * a migration. Enable CMMA in the mm_context.
1116                  * Since we need to take a write lock to write to the context
1117                  * to avoid races with storage keys handling, we check if the
1118                  * value really needs to be written to; if the value is
1119                  * already correct, we do nothing and avoid the lock.
1120                  */
1121                 if (vcpu->kvm->mm->context.uses_cmm == 0) {
1122                         down_write(&vcpu->kvm->mm->mmap_sem);
1123                         vcpu->kvm->mm->context.uses_cmm = 1;
1124                         up_write(&vcpu->kvm->mm->mmap_sem);
1125                 }
1126                 /*
1127                  * If we are here, we are supposed to have CMMA enabled in
1128                  * the SIE block. Enabling CMMA works on a per-CPU basis,
1129                  * while the context use_cmma flag is per process.
1130                  * It's possible that the context flag is enabled and the
1131                  * SIE flag is not, so we set the flag always; if it was
1132                  * already set, nothing changes, otherwise we enable it
1133                  * on this CPU too.
1134                  */
1135                 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
1136                 /* Retry the ESSA instruction */
1137                 kvm_s390_retry_instr(vcpu);
1138         } else {
1139                 /* Account for the possible extra cbrl entry */
1140                 i = do_essa(vcpu, orc);
1141                 if (i < 0)
1142                         return i;
1143                 entries += i;
1144         }
1145         vcpu->arch.sie_block->cbrlo &= PAGE_MASK;       /* reset nceo */
1146         cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
1147         down_read(&gmap->mm->mmap_sem);
1148         for (i = 0; i < entries; ++i)
1149                 __gmap_zap(gmap, cbrlo[i]);
1150         up_read(&gmap->mm->mmap_sem);
1151         return 0;
1152 }
1153
1154 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
1155 {
1156         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1157         case 0x8a:
1158         case 0x8e:
1159         case 0x8f:
1160                 return handle_ipte_interlock(vcpu);
1161         case 0x8d:
1162                 return handle_epsw(vcpu);
1163         case 0xab:
1164                 return handle_essa(vcpu);
1165         case 0xaf:
1166                 return handle_pfmf(vcpu);
1167         default:
1168                 return -EOPNOTSUPP;
1169         }
1170 }
1171
1172 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
1173 {
1174         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1175         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1176         int reg, rc, nr_regs;
1177         u32 ctl_array[16];
1178         u64 ga;
1179         u8 ar;
1180
1181         vcpu->stat.instruction_lctl++;
1182
1183         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1184                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1185
1186         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1187
1188         if (ga & 3)
1189                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1190
1191         VCPU_EVENT(vcpu, 4, "LCTL: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1192         trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
1193
1194         nr_regs = ((reg3 - reg1) & 0xf) + 1;
1195         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1196         if (rc)
1197                 return kvm_s390_inject_prog_cond(vcpu, rc);
1198         reg = reg1;
1199         nr_regs = 0;
1200         do {
1201                 vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
1202                 vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
1203                 if (reg == reg3)
1204                         break;
1205                 reg = (reg + 1) % 16;
1206         } while (1);
1207         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1208         return 0;
1209 }
1210
1211 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
1212 {
1213         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1214         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1215         int reg, rc, nr_regs;
1216         u32 ctl_array[16];
1217         u64 ga;
1218         u8 ar;
1219
1220         vcpu->stat.instruction_stctl++;
1221
1222         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1223                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1224
1225         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1226
1227         if (ga & 3)
1228                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1229
1230         VCPU_EVENT(vcpu, 4, "STCTL r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1231         trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
1232
1233         reg = reg1;
1234         nr_regs = 0;
1235         do {
1236                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1237                 if (reg == reg3)
1238                         break;
1239                 reg = (reg + 1) % 16;
1240         } while (1);
1241         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1242         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1243 }
1244
1245 static int handle_lctlg(struct kvm_vcpu *vcpu)
1246 {
1247         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1248         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1249         int reg, rc, nr_regs;
1250         u64 ctl_array[16];
1251         u64 ga;
1252         u8 ar;
1253
1254         vcpu->stat.instruction_lctlg++;
1255
1256         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1257                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1258
1259         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1260
1261         if (ga & 7)
1262                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1263
1264         VCPU_EVENT(vcpu, 4, "LCTLG: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1265         trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
1266
1267         nr_regs = ((reg3 - reg1) & 0xf) + 1;
1268         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1269         if (rc)
1270                 return kvm_s390_inject_prog_cond(vcpu, rc);
1271         reg = reg1;
1272         nr_regs = 0;
1273         do {
1274                 vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
1275                 if (reg == reg3)
1276                         break;
1277                 reg = (reg + 1) % 16;
1278         } while (1);
1279         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1280         return 0;
1281 }
1282
1283 static int handle_stctg(struct kvm_vcpu *vcpu)
1284 {
1285         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1286         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1287         int reg, rc, nr_regs;
1288         u64 ctl_array[16];
1289         u64 ga;
1290         u8 ar;
1291
1292         vcpu->stat.instruction_stctg++;
1293
1294         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1295                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1296
1297         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1298
1299         if (ga & 7)
1300                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1301
1302         VCPU_EVENT(vcpu, 4, "STCTG r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1303         trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
1304
1305         reg = reg1;
1306         nr_regs = 0;
1307         do {
1308                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1309                 if (reg == reg3)
1310                         break;
1311                 reg = (reg + 1) % 16;
1312         } while (1);
1313         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1314         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1315 }
1316
1317 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
1318 {
1319         switch (vcpu->arch.sie_block->ipb & 0x000000ff) {
1320         case 0x25:
1321                 return handle_stctg(vcpu);
1322         case 0x2f:
1323                 return handle_lctlg(vcpu);
1324         case 0x60:
1325         case 0x61:
1326         case 0x62:
1327                 return handle_ri(vcpu);
1328         default:
1329                 return -EOPNOTSUPP;
1330         }
1331 }
1332
1333 static int handle_tprot(struct kvm_vcpu *vcpu)
1334 {
1335         u64 address1, address2;
1336         unsigned long hva, gpa;
1337         int ret = 0, cc = 0;
1338         bool writable;
1339         u8 ar;
1340
1341         vcpu->stat.instruction_tprot++;
1342
1343         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1344                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1345
1346         kvm_s390_get_base_disp_sse(vcpu, &address1, &address2, &ar, NULL);
1347
1348         /* we only handle the Linux memory detection case:
1349          * access key == 0
1350          * everything else goes to userspace. */
1351         if (address2 & 0xf0)
1352                 return -EOPNOTSUPP;
1353         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1354                 ipte_lock(vcpu);
1355         ret = guest_translate_address(vcpu, address1, ar, &gpa, GACC_STORE);
1356         if (ret == PGM_PROTECTION) {
1357                 /* Write protected? Try again with read-only... */
1358                 cc = 1;
1359                 ret = guest_translate_address(vcpu, address1, ar, &gpa,
1360                                               GACC_FETCH);
1361         }
1362         if (ret) {
1363                 if (ret == PGM_ADDRESSING || ret == PGM_TRANSLATION_SPEC) {
1364                         ret = kvm_s390_inject_program_int(vcpu, ret);
1365                 } else if (ret > 0) {
1366                         /* Translation not available */
1367                         kvm_s390_set_psw_cc(vcpu, 3);
1368                         ret = 0;
1369                 }
1370                 goto out_unlock;
1371         }
1372
1373         hva = gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
1374         if (kvm_is_error_hva(hva)) {
1375                 ret = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1376         } else {
1377                 if (!writable)
1378                         cc = 1;         /* Write not permitted ==> read-only */
1379                 kvm_s390_set_psw_cc(vcpu, cc);
1380                 /* Note: CC2 only occurs for storage keys (not supported yet) */
1381         }
1382 out_unlock:
1383         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1384                 ipte_unlock(vcpu);
1385         return ret;
1386 }
1387
1388 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
1389 {
1390         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1391         case 0x01:
1392                 return handle_tprot(vcpu);
1393         default:
1394                 return -EOPNOTSUPP;
1395         }
1396 }
1397
1398 static int handle_sckpf(struct kvm_vcpu *vcpu)
1399 {
1400         u32 value;
1401
1402         vcpu->stat.instruction_sckpf++;
1403
1404         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1405                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1406
1407         if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
1408                 return kvm_s390_inject_program_int(vcpu,
1409                                                    PGM_SPECIFICATION);
1410
1411         value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
1412         vcpu->arch.sie_block->todpr = value;
1413
1414         return 0;
1415 }
1416
1417 static int handle_ptff(struct kvm_vcpu *vcpu)
1418 {
1419         vcpu->stat.instruction_ptff++;
1420
1421         /* we don't emulate any control instructions yet */
1422         kvm_s390_set_psw_cc(vcpu, 3);
1423         return 0;
1424 }
1425
1426 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1427 {
1428         switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1429         case 0x04:
1430                 return handle_ptff(vcpu);
1431         case 0x07:
1432                 return handle_sckpf(vcpu);
1433         default:
1434                 return -EOPNOTSUPP;
1435         }
1436 }