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