arch/s390/kvm/priv.c

   1 /*
   2  * handling privileged instructions
   3  *
   4  * Copyright IBM Corp. 2008, 2013
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License (version 2 only)
   8  * as published by the Free Software Foundation.
   9  *
  10  *    Author(s): Carsten Otte <cotte@de.ibm.com>
  11  *               Christian Borntraeger <borntraeger@de.ibm.com>
  12  */
  13
  14 #include <linux/kvm.h>
  15 #include <linux/gfp.h>
  16 #include <linux/errno.h>
  17 #include <linux/compat.h>
  18 #include <linux/mm_types.h>
  19
  20 #include <asm/asm-offsets.h>
  21 #include <asm/facility.h>
  22 #include <asm/current.h>
  23 #include <asm/debug.h>
  24 #include <asm/ebcdic.h>
  25 #include <asm/sysinfo.h>
  26 #include <asm/pgtable.h>
  27 #include <asm/page-states.h>
  28 #include <asm/pgalloc.h>
  29 #include <asm/gmap.h>
  30 #include <asm/io.h>
  31 #include <asm/ptrace.h>
  32 #include <asm/compat.h>
  33 #include <asm/sclp.h>
  34 #include "gaccess.h"
  35 #include "kvm-s390.h"
  36 #include "trace.h"
  37
  38 static int handle_ri(struct kvm_vcpu *vcpu)
  39 {
  40         if (test_kvm_facility(vcpu->kvm, 64)) {
  41                 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (lazy)");
  42                 vcpu->arch.sie_block->ecb3 |= ECB3_RI;
  43                 kvm_s390_retry_instr(vcpu);
  44                 return 0;
  45         } else
  46                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
  47 }
  48
  49 int kvm_s390_handle_aa(struct kvm_vcpu *vcpu)
  50 {
  51         if ((vcpu->arch.sie_block->ipa & 0xf) <= 4)
  52                 return handle_ri(vcpu);
  53         else
  54                 return -EOPNOTSUPP;
  55 }
  56
  57 static int handle_gs(struct kvm_vcpu *vcpu)
  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         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
  93                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
  94
  95         op2 = kvm_s390_get_base_disp_s(vcpu, &ar);
  96         if (op2 & 7)    /* Operand must be on a doubleword boundary */
  97                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
  98         rc = read_guest(vcpu, op2, ar, &gtod.tod, sizeof(gtod.tod));
  99         if (rc)
 100                 return kvm_s390_inject_prog_cond(vcpu, rc);
 101
 102         VCPU_EVENT(vcpu, 3, "SCK: setting guest TOD to 0x%llx", gtod.tod);
 103         kvm_s390_set_tod_clock(vcpu->kvm, &gtod);
 104
 105         kvm_s390_set_psw_cc(vcpu, 0);
 106         return 0;
 107 }
 108
 109 static int handle_set_prefix(struct kvm_vcpu *vcpu)
 110 {
 111         u64 operand2;
 112         u32 address;
 113         int rc;
 114         u8 ar;
 115
 116         vcpu->stat.instruction_spx++;
 117
 118         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 119                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 120
 121         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 122
 123         /* must be word boundary */
 124         if (operand2 & 3)
 125                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 126
 127         /* get the value */
 128         rc = read_guest(vcpu, operand2, ar, &address, sizeof(address));
 129         if (rc)
 130                 return kvm_s390_inject_prog_cond(vcpu, rc);
 131
 132         address &= 0x7fffe000u;
 133
 134         /*
 135          * Make sure the new value is valid memory. We only need to check the
 136          * first page, since address is 8k aligned and memory pieces are always
 137          * at least 1MB aligned and have at least a size of 1MB.
 138          */
 139         if (kvm_is_error_gpa(vcpu->kvm, address))
 140                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 141
 142         kvm_s390_set_prefix(vcpu, address);
 143         trace_kvm_s390_handle_prefix(vcpu, 1, address);
 144         return 0;
 145 }
 146
 147 static int handle_store_prefix(struct kvm_vcpu *vcpu)
 148 {
 149         u64 operand2;
 150         u32 address;
 151         int rc;
 152         u8 ar;
 153
 154         vcpu->stat.instruction_stpx++;
 155
 156         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 157                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 158
 159         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 160
 161         /* must be word boundary */
 162         if (operand2 & 3)
 163                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 164
 165         address = kvm_s390_get_prefix(vcpu);
 166
 167         /* get the value */
 168         rc = write_guest(vcpu, operand2, ar, &address, sizeof(address));
 169         if (rc)
 170                 return kvm_s390_inject_prog_cond(vcpu, rc);
 171
 172         VCPU_EVENT(vcpu, 3, "STPX: storing prefix 0x%x into 0x%llx", address, operand2);
 173         trace_kvm_s390_handle_prefix(vcpu, 0, address);
 174         return 0;
 175 }
 176
 177 static int handle_store_cpu_address(struct kvm_vcpu *vcpu)
 178 {
 179         u16 vcpu_id = vcpu->vcpu_id;
 180         u64 ga;
 181         int rc;
 182         u8 ar;
 183
 184         vcpu->stat.instruction_stap++;
 185
 186         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 187                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 188
 189         ga = kvm_s390_get_base_disp_s(vcpu, &ar);
 190
 191         if (ga & 1)
 192                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 193
 194         rc = write_guest(vcpu, ga, ar, &vcpu_id, sizeof(vcpu_id));
 195         if (rc)
 196                 return kvm_s390_inject_prog_cond(vcpu, rc);
 197
 198         VCPU_EVENT(vcpu, 3, "STAP: storing cpu address (%u) to 0x%llx", vcpu_id, ga);
 199         trace_kvm_s390_handle_stap(vcpu, ga);
 200         return 0;
 201 }
 202
 203 int kvm_s390_skey_check_enable(struct kvm_vcpu *vcpu)
 204 {
 205         int rc = 0;
 206         struct kvm_s390_sie_block *sie_block = vcpu->arch.sie_block;
 207
 208         trace_kvm_s390_skey_related_inst(vcpu);
 209         if (!(sie_block->ictl & (ICTL_ISKE | ICTL_SSKE | ICTL_RRBE)) &&
 210             !(atomic_read(&sie_block->cpuflags) & CPUSTAT_KSS))
 211                 return rc;
 212
 213         rc = s390_enable_skey();
 214         VCPU_EVENT(vcpu, 3, "enabling storage keys for guest: %d", rc);
 215         if (!rc) {
 216                 if (atomic_read(&sie_block->cpuflags) & CPUSTAT_KSS)
 217                         atomic_andnot(CPUSTAT_KSS, &sie_block->cpuflags);
 218                 else
 219                         sie_block->ictl &= ~(ICTL_ISKE | ICTL_SSKE |
 220                                              ICTL_RRBE);
 221         }
 222         return rc;
 223 }
 224
 225 static int try_handle_skey(struct kvm_vcpu *vcpu)
 226 {
 227         int rc;
 228
 229         vcpu->stat.instruction_storage_key++;
 230         rc = kvm_s390_skey_check_enable(vcpu);
 231         if (rc)
 232                 return rc;
 233         if (sclp.has_skey) {
 234                 /* with storage-key facility, SIE interprets it for us */
 235                 kvm_s390_retry_instr(vcpu);
 236                 VCPU_EVENT(vcpu, 4, "%s", "retrying storage key operation");
 237                 return -EAGAIN;
 238         }
 239         return 0;
 240 }
 241
 242 static int handle_iske(struct kvm_vcpu *vcpu)
 243 {
 244         unsigned long addr;
 245         unsigned char key;
 246         int reg1, reg2;
 247         int rc;
 248
 249         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 250                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 251
 252         rc = try_handle_skey(vcpu);
 253         if (rc)
 254                 return rc != -EAGAIN ? rc : 0;
 255
 256         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 257
 258         addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 259         addr = kvm_s390_logical_to_effective(vcpu, addr);
 260         addr = kvm_s390_real_to_abs(vcpu, addr);
 261         addr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(addr));
 262         if (kvm_is_error_hva(addr))
 263                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 264
 265         down_read(&current->mm->mmap_sem);
 266         rc = get_guest_storage_key(current->mm, addr, &key);
 267         up_read(&current->mm->mmap_sem);
 268         if (rc)
 269                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 270         vcpu->run->s.regs.gprs[reg1] &= ~0xff;
 271         vcpu->run->s.regs.gprs[reg1] |= key;
 272         return 0;
 273 }
 274
 275 static int handle_rrbe(struct kvm_vcpu *vcpu)
 276 {
 277         unsigned long addr;
 278         int reg1, reg2;
 279         int rc;
 280
 281         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 282                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 283
 284         rc = try_handle_skey(vcpu);
 285         if (rc)
 286                 return rc != -EAGAIN ? rc : 0;
 287
 288         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 289
 290         addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 291         addr = kvm_s390_logical_to_effective(vcpu, addr);
 292         addr = kvm_s390_real_to_abs(vcpu, addr);
 293         addr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(addr));
 294         if (kvm_is_error_hva(addr))
 295                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 296
 297         down_read(&current->mm->mmap_sem);
 298         rc = reset_guest_reference_bit(current->mm, addr);
 299         up_read(&current->mm->mmap_sem);
 300         if (rc < 0)
 301                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 302
 303         kvm_s390_set_psw_cc(vcpu, rc);
 304         return 0;
 305 }
 306
 307 #define SSKE_NQ 0x8
 308 #define SSKE_MR 0x4
 309 #define SSKE_MC 0x2
 310 #define SSKE_MB 0x1
 311 static int handle_sske(struct kvm_vcpu *vcpu)
 312 {
 313         unsigned char m3 = vcpu->arch.sie_block->ipb >> 28;
 314         unsigned long start, end;
 315         unsigned char key, oldkey;
 316         int reg1, reg2;
 317         int rc;
 318
 319         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 320                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 321
 322         rc = try_handle_skey(vcpu);
 323         if (rc)
 324                 return rc != -EAGAIN ? rc : 0;
 325
 326         if (!test_kvm_facility(vcpu->kvm, 8))
 327                 m3 &= ~SSKE_MB;
 328         if (!test_kvm_facility(vcpu->kvm, 10))
 329                 m3 &= ~(SSKE_MC | SSKE_MR);
 330         if (!test_kvm_facility(vcpu->kvm, 14))
 331                 m3 &= ~SSKE_NQ;
 332
 333         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 334
 335         key = vcpu->run->s.regs.gprs[reg1] & 0xfe;
 336         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 337         start = kvm_s390_logical_to_effective(vcpu, start);
 338         if (m3 & SSKE_MB) {
 339                 /* start already designates an absolute address */
 340                 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
 341         } else {
 342                 start = kvm_s390_real_to_abs(vcpu, start);
 343                 end = start + PAGE_SIZE;
 344         }
 345
 346         while (start != end) {
 347                 unsigned long addr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
 348
 349                 if (kvm_is_error_hva(addr))
 350                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 351
 352                 down_read(&current->mm->mmap_sem);
 353                 rc = cond_set_guest_storage_key(current->mm, addr, key, &oldkey,
 354                                                 m3 & SSKE_NQ, m3 & SSKE_MR,
 355                                                 m3 & SSKE_MC);
 356                 up_read(&current->mm->mmap_sem);
 357                 if (rc < 0)
 358                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 359                 start += PAGE_SIZE;
 360         }
 361
 362         if (m3 & (SSKE_MC | SSKE_MR)) {
 363                 if (m3 & SSKE_MB) {
 364                         /* skey in reg1 is unpredictable */
 365                         kvm_s390_set_psw_cc(vcpu, 3);
 366                 } else {
 367                         kvm_s390_set_psw_cc(vcpu, rc);
 368                         vcpu->run->s.regs.gprs[reg1] &= ~0xff00UL;
 369                         vcpu->run->s.regs.gprs[reg1] |= (u64) oldkey << 8;
 370                 }
 371         }
 372         if (m3 & SSKE_MB) {
 373                 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT)
 374                         vcpu->run->s.regs.gprs[reg2] &= ~PAGE_MASK;
 375                 else
 376                         vcpu->run->s.regs.gprs[reg2] &= ~0xfffff000UL;
 377                 end = kvm_s390_logical_to_effective(vcpu, end);
 378                 vcpu->run->s.regs.gprs[reg2] |= end;
 379         }
 380         return 0;
 381 }
 382
 383 static int handle_ipte_interlock(struct kvm_vcpu *vcpu)
 384 {
 385         vcpu->stat.instruction_ipte_interlock++;
 386         if (psw_bits(vcpu->arch.sie_block->gpsw).pstate)
 387                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 388         wait_event(vcpu->kvm->arch.ipte_wq, !ipte_lock_held(vcpu));
 389         kvm_s390_retry_instr(vcpu);
 390         VCPU_EVENT(vcpu, 4, "%s", "retrying ipte interlock operation");
 391         return 0;
 392 }
 393
 394 static int handle_test_block(struct kvm_vcpu *vcpu)
 395 {
 396         gpa_t addr;
 397         int reg2;
 398
 399         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 400                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 401
 402         kvm_s390_get_regs_rre(vcpu, NULL, &reg2);
 403         addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 404         addr = kvm_s390_logical_to_effective(vcpu, addr);
 405         if (kvm_s390_check_low_addr_prot_real(vcpu, addr))
 406                 return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
 407         addr = kvm_s390_real_to_abs(vcpu, addr);
 408
 409         if (kvm_is_error_gpa(vcpu->kvm, addr))
 410                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 411         /*
 412          * We don't expect errors on modern systems, and do not care
 413          * about storage keys (yet), so let's just clear the page.
 414          */
 415         if (kvm_clear_guest(vcpu->kvm, addr, PAGE_SIZE))
 416                 return -EFAULT;
 417         kvm_s390_set_psw_cc(vcpu, 0);
 418         vcpu->run->s.regs.gprs[0] = 0;
 419         return 0;
 420 }
 421
 422 static int handle_tpi(struct kvm_vcpu *vcpu)
 423 {
 424         struct kvm_s390_interrupt_info *inti;
 425         unsigned long len;
 426         u32 tpi_data[3];
 427         int rc;
 428         u64 addr;
 429         u8 ar;
 430
 431         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 432         if (addr & 3)
 433                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 434
 435         inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0);
 436         if (!inti) {
 437                 kvm_s390_set_psw_cc(vcpu, 0);
 438                 return 0;
 439         }
 440
 441         tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr;
 442         tpi_data[1] = inti->io.io_int_parm;
 443         tpi_data[2] = inti->io.io_int_word;
 444         if (addr) {
 445                 /*
 446                  * Store the two-word I/O interruption code into the
 447                  * provided area.
 448                  */
 449                 len = sizeof(tpi_data) - 4;
 450                 rc = write_guest(vcpu, addr, ar, &tpi_data, len);
 451                 if (rc) {
 452                         rc = kvm_s390_inject_prog_cond(vcpu, rc);
 453                         goto reinject_interrupt;
 454                 }
 455         } else {
 456                 /*
 457                  * Store the three-word I/O interruption code into
 458                  * the appropriate lowcore area.
 459                  */
 460                 len = sizeof(tpi_data);
 461                 if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len)) {
 462                         /* failed writes to the low core are not recoverable */
 463                         rc = -EFAULT;
 464                         goto reinject_interrupt;
 465                 }
 466         }
 467
 468         /* irq was successfully handed to the guest */
 469         kfree(inti);
 470         kvm_s390_set_psw_cc(vcpu, 1);
 471         return 0;
 472 reinject_interrupt:
 473         /*
 474          * If we encounter a problem storing the interruption code, the
 475          * instruction is suppressed from the guest's view: reinject the
 476          * interrupt.
 477          */
 478         if (kvm_s390_reinject_io_int(vcpu->kvm, inti)) {
 479                 kfree(inti);
 480                 rc = -EFAULT;
 481         }
 482         /* don't set the cc, a pgm irq was injected or we drop to user space */
 483         return rc ? -EFAULT : 0;
 484 }
 485
 486 static int handle_tsch(struct kvm_vcpu *vcpu)
 487 {
 488         struct kvm_s390_interrupt_info *inti = NULL;
 489         const u64 isc_mask = 0xffUL << 24; /* all iscs set */
 490
 491         /* a valid schid has at least one bit set */
 492         if (vcpu->run->s.regs.gprs[1])
 493                 inti = kvm_s390_get_io_int(vcpu->kvm, isc_mask,
 494                                            vcpu->run->s.regs.gprs[1]);
 495
 496         /*
 497          * Prepare exit to userspace.
 498          * We indicate whether we dequeued a pending I/O interrupt
 499          * so that userspace can re-inject it if the instruction gets
 500          * a program check. While this may re-order the pending I/O
 501          * interrupts, this is no problem since the priority is kept
 502          * intact.
 503          */
 504         vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
 505         vcpu->run->s390_tsch.dequeued = !!inti;
 506         if (inti) {
 507                 vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
 508                 vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
 509                 vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
 510                 vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
 511         }
 512         vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
 513         kfree(inti);
 514         return -EREMOTE;
 515 }
 516
 517 static int handle_io_inst(struct kvm_vcpu *vcpu)
 518 {
 519         VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
 520
 521         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 522                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 523
 524         if (vcpu->kvm->arch.css_support) {
 525                 /*
 526                  * Most I/O instructions will be handled by userspace.
 527                  * Exceptions are tpi and the interrupt portion of tsch.
 528                  */
 529                 if (vcpu->arch.sie_block->ipa == 0xb236)
 530                         return handle_tpi(vcpu);
 531                 if (vcpu->arch.sie_block->ipa == 0xb235)
 532                         return handle_tsch(vcpu);
 533                 /* Handle in userspace. */
 534                 return -EOPNOTSUPP;
 535         } else {
 536                 /*
 537                  * Set condition code 3 to stop the guest from issuing channel
 538                  * I/O instructions.
 539                  */
 540                 kvm_s390_set_psw_cc(vcpu, 3);
 541                 return 0;
 542         }
 543 }
 544
 545 static int handle_stfl(struct kvm_vcpu *vcpu)
 546 {
 547         int rc;
 548         unsigned int fac;
 549
 550         vcpu->stat.instruction_stfl++;
 551
 552         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 553                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 554
 555         /*
 556          * We need to shift the lower 32 facility bits (bit 0-31) from a u64
 557          * into a u32 memory representation. They will remain bits 0-31.
 558          */
 559         fac = *vcpu->kvm->arch.model.fac_list >> 32;
 560         rc = write_guest_lc(vcpu, offsetof(struct lowcore, stfl_fac_list),
 561                             &fac, sizeof(fac));
 562         if (rc)
 563                 return rc;
 564         VCPU_EVENT(vcpu, 3, "STFL: store facility list 0x%x", fac);
 565         trace_kvm_s390_handle_stfl(vcpu, fac);
 566         return 0;
 567 }
 568
 569 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
 570 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
 571 #define PSW_ADDR_24 0x0000000000ffffffUL
 572 #define PSW_ADDR_31 0x000000007fffffffUL
 573
 574 int is_valid_psw(psw_t *psw)
 575 {
 576         if (psw->mask & PSW_MASK_UNASSIGNED)
 577                 return 0;
 578         if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
 579                 if (psw->addr & ~PSW_ADDR_31)
 580                         return 0;
 581         }
 582         if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
 583                 return 0;
 584         if ((psw->mask & PSW_MASK_ADDR_MODE) ==  PSW_MASK_EA)
 585                 return 0;
 586         if (psw->addr & 1)
 587                 return 0;
 588         return 1;
 589 }
 590
 591 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
 592 {
 593         psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
 594         psw_compat_t new_psw;
 595         u64 addr;
 596         int rc;
 597         u8 ar;
 598
 599         if (gpsw->mask & PSW_MASK_PSTATE)
 600                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 601
 602         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 603         if (addr & 7)
 604                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 605
 606         rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
 607         if (rc)
 608                 return kvm_s390_inject_prog_cond(vcpu, rc);
 609         if (!(new_psw.mask & PSW32_MASK_BASE))
 610                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 611         gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
 612         gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
 613         gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
 614         if (!is_valid_psw(gpsw))
 615                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 616         return 0;
 617 }
 618
 619 static int handle_lpswe(struct kvm_vcpu *vcpu)
 620 {
 621         psw_t new_psw;
 622         u64 addr;
 623         int rc;
 624         u8 ar;
 625
 626         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 627                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 628
 629         addr = kvm_s390_get_base_disp_s(vcpu, &ar);
 630         if (addr & 7)
 631                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 632         rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
 633         if (rc)
 634                 return kvm_s390_inject_prog_cond(vcpu, rc);
 635         vcpu->arch.sie_block->gpsw = new_psw;
 636         if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
 637                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 638         return 0;
 639 }
 640
 641 static int handle_stidp(struct kvm_vcpu *vcpu)
 642 {
 643         u64 stidp_data = vcpu->kvm->arch.model.cpuid;
 644         u64 operand2;
 645         int rc;
 646         u8 ar;
 647
 648         vcpu->stat.instruction_stidp++;
 649
 650         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 651                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 652
 653         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 654
 655         if (operand2 & 7)
 656                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 657
 658         rc = write_guest(vcpu, operand2, ar, &stidp_data, sizeof(stidp_data));
 659         if (rc)
 660                 return kvm_s390_inject_prog_cond(vcpu, rc);
 661
 662         VCPU_EVENT(vcpu, 3, "STIDP: store cpu id 0x%llx", stidp_data);
 663         return 0;
 664 }
 665
 666 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
 667 {
 668         int cpus = 0;
 669         int n;
 670
 671         cpus = atomic_read(&vcpu->kvm->online_vcpus);
 672
 673         /* deal with other level 3 hypervisors */
 674         if (stsi(mem, 3, 2, 2))
 675                 mem->count = 0;
 676         if (mem->count < 8)
 677                 mem->count++;
 678         for (n = mem->count - 1; n > 0 ; n--)
 679                 memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
 680
 681         memset(&mem->vm[0], 0, sizeof(mem->vm[0]));
 682         mem->vm[0].cpus_total = cpus;
 683         mem->vm[0].cpus_configured = cpus;
 684         mem->vm[0].cpus_standby = 0;
 685         mem->vm[0].cpus_reserved = 0;
 686         mem->vm[0].caf = 1000;
 687         memcpy(mem->vm[0].name, "KVMguest", 8);
 688         ASCEBC(mem->vm[0].name, 8);
 689         memcpy(mem->vm[0].cpi, "KVM/Linux       ", 16);
 690         ASCEBC(mem->vm[0].cpi, 16);
 691 }
 692
 693 static void insert_stsi_usr_data(struct kvm_vcpu *vcpu, u64 addr, u8 ar,
 694                                  u8 fc, u8 sel1, u16 sel2)
 695 {
 696         vcpu->run->exit_reason = KVM_EXIT_S390_STSI;
 697         vcpu->run->s390_stsi.addr = addr;
 698         vcpu->run->s390_stsi.ar = ar;
 699         vcpu->run->s390_stsi.fc = fc;
 700         vcpu->run->s390_stsi.sel1 = sel1;
 701         vcpu->run->s390_stsi.sel2 = sel2;
 702 }
 703
 704 static int handle_stsi(struct kvm_vcpu *vcpu)
 705 {
 706         int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
 707         int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
 708         int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
 709         unsigned long mem = 0;
 710         u64 operand2;
 711         int rc = 0;
 712         u8 ar;
 713
 714         vcpu->stat.instruction_stsi++;
 715         VCPU_EVENT(vcpu, 3, "STSI: fc: %u sel1: %u sel2: %u", fc, sel1, sel2);
 716
 717         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 718                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 719
 720         if (fc > 3) {
 721                 kvm_s390_set_psw_cc(vcpu, 3);
 722                 return 0;
 723         }
 724
 725         if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
 726             || vcpu->run->s.regs.gprs[1] & 0xffff0000)
 727                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 728
 729         if (fc == 0) {
 730                 vcpu->run->s.regs.gprs[0] = 3 << 28;
 731                 kvm_s390_set_psw_cc(vcpu, 0);
 732                 return 0;
 733         }
 734
 735         operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
 736
 737         if (operand2 & 0xfff)
 738                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 739
 740         switch (fc) {
 741         case 1: /* same handling for 1 and 2 */
 742         case 2:
 743                 mem = get_zeroed_page(GFP_KERNEL);
 744                 if (!mem)
 745                         goto out_no_data;
 746                 if (stsi((void *) mem, fc, sel1, sel2))
 747                         goto out_no_data;
 748                 break;
 749         case 3:
 750                 if (sel1 != 2 || sel2 != 2)
 751                         goto out_no_data;
 752                 mem = get_zeroed_page(GFP_KERNEL);
 753                 if (!mem)
 754                         goto out_no_data;
 755                 handle_stsi_3_2_2(vcpu, (void *) mem);
 756                 break;
 757         }
 758
 759         rc = write_guest(vcpu, operand2, ar, (void *)mem, PAGE_SIZE);
 760         if (rc) {
 761                 rc = kvm_s390_inject_prog_cond(vcpu, rc);
 762                 goto out;
 763         }
 764         if (vcpu->kvm->arch.user_stsi) {
 765                 insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
 766                 rc = -EREMOTE;
 767         }
 768         trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
 769         free_page(mem);
 770         kvm_s390_set_psw_cc(vcpu, 0);
 771         vcpu->run->s.regs.gprs[0] = 0;
 772         return rc;
 773 out_no_data:
 774         kvm_s390_set_psw_cc(vcpu, 3);
 775 out:
 776         free_page(mem);
 777         return rc;
 778 }
 779
 780 static const intercept_handler_t b2_handlers[256] = {
 781         [0x02] = handle_stidp,
 782         [0x04] = handle_set_clock,
 783         [0x10] = handle_set_prefix,
 784         [0x11] = handle_store_prefix,
 785         [0x12] = handle_store_cpu_address,
 786         [0x14] = kvm_s390_handle_vsie,
 787         [0x21] = handle_ipte_interlock,
 788         [0x29] = handle_iske,
 789         [0x2a] = handle_rrbe,
 790         [0x2b] = handle_sske,
 791         [0x2c] = handle_test_block,
 792         [0x30] = handle_io_inst,
 793         [0x31] = handle_io_inst,
 794         [0x32] = handle_io_inst,
 795         [0x33] = handle_io_inst,
 796         [0x34] = handle_io_inst,
 797         [0x35] = handle_io_inst,
 798         [0x36] = handle_io_inst,
 799         [0x37] = handle_io_inst,
 800         [0x38] = handle_io_inst,
 801         [0x39] = handle_io_inst,
 802         [0x3a] = handle_io_inst,
 803         [0x3b] = handle_io_inst,
 804         [0x3c] = handle_io_inst,
 805         [0x50] = handle_ipte_interlock,
 806         [0x56] = handle_sthyi,
 807         [0x5f] = handle_io_inst,
 808         [0x74] = handle_io_inst,
 809         [0x76] = handle_io_inst,
 810         [0x7d] = handle_stsi,
 811         [0xb1] = handle_stfl,
 812         [0xb2] = handle_lpswe,
 813 };
 814
 815 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
 816 {
 817         intercept_handler_t handler;
 818
 819         /*
 820          * A lot of B2 instructions are priviledged. Here we check for
 821          * the privileged ones, that we can handle in the kernel.
 822          * Anything else goes to userspace.
 823          */
 824         handler = b2_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
 825         if (handler)
 826                 return handler(vcpu);
 827
 828         return -EOPNOTSUPP;
 829 }
 830
 831 static int handle_epsw(struct kvm_vcpu *vcpu)
 832 {
 833         int reg1, reg2;
 834
 835         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 836
 837         /* This basically extracts the mask half of the psw. */
 838         vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
 839         vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
 840         if (reg2) {
 841                 vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
 842                 vcpu->run->s.regs.gprs[reg2] |=
 843                         vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
 844         }
 845         return 0;
 846 }
 847
 848 #define PFMF_RESERVED   0xfffc0101UL
 849 #define PFMF_SK         0x00020000UL
 850 #define PFMF_CF         0x00010000UL
 851 #define PFMF_UI         0x00008000UL
 852 #define PFMF_FSC        0x00007000UL
 853 #define PFMF_NQ         0x00000800UL
 854 #define PFMF_MR         0x00000400UL
 855 #define PFMF_MC         0x00000200UL
 856 #define PFMF_KEY        0x000000feUL
 857
 858 static int handle_pfmf(struct kvm_vcpu *vcpu)
 859 {
 860         bool mr = false, mc = false, nq;
 861         int reg1, reg2;
 862         unsigned long start, end;
 863         unsigned char key;
 864
 865         vcpu->stat.instruction_pfmf++;
 866
 867         kvm_s390_get_regs_rre(vcpu, &reg1, &reg2);
 868
 869         if (!test_kvm_facility(vcpu->kvm, 8))
 870                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
 871
 872         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
 873                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
 874
 875         if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
 876                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 877
 878         /* Only provide non-quiescing support if enabled for the guest */
 879         if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ &&
 880             !test_kvm_facility(vcpu->kvm, 14))
 881                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 882
 883         /* Only provide conditional-SSKE support if enabled for the guest */
 884         if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK &&
 885             test_kvm_facility(vcpu->kvm, 10)) {
 886                 mr = vcpu->run->s.regs.gprs[reg1] & PFMF_MR;
 887                 mc = vcpu->run->s.regs.gprs[reg1] & PFMF_MC;
 888         }
 889
 890         nq = vcpu->run->s.regs.gprs[reg1] & PFMF_NQ;
 891         key = vcpu->run->s.regs.gprs[reg1] & PFMF_KEY;
 892         start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
 893         start = kvm_s390_logical_to_effective(vcpu, start);
 894
 895         if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
 896                 if (kvm_s390_check_low_addr_prot_real(vcpu, start))
 897                         return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
 898         }
 899
 900         switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
 901         case 0x00000000:
 902                 /* only 4k frames specify a real address */
 903                 start = kvm_s390_real_to_abs(vcpu, start);
 904                 end = (start + PAGE_SIZE) & ~(PAGE_SIZE - 1);
 905                 break;
 906         case 0x00001000:
 907                 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
 908                 break;
 909         case 0x00002000:
 910                 /* only support 2G frame size if EDAT2 is available and we are
 911                    not in 24-bit addressing mode */
 912                 if (!test_kvm_facility(vcpu->kvm, 78) ||
 913                     psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_24BIT)
 914                         return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 915                 end = (start + _REGION3_SIZE) & ~(_REGION3_SIZE - 1);
 916                 break;
 917         default:
 918                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
 919         }
 920
 921         while (start != end) {
 922                 unsigned long useraddr;
 923
 924                 /* Translate guest address to host address */
 925                 useraddr = gfn_to_hva(vcpu->kvm, gpa_to_gfn(start));
 926                 if (kvm_is_error_hva(useraddr))
 927                         return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 928
 929                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
 930                         if (clear_user((void __user *)useraddr, PAGE_SIZE))
 931                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 932                 }
 933
 934                 if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
 935                         int rc = kvm_s390_skey_check_enable(vcpu);
 936
 937                         if (rc)
 938                                 return rc;
 939                         down_read(&current->mm->mmap_sem);
 940                         rc = cond_set_guest_storage_key(current->mm, useraddr,
 941                                                         key, NULL, nq, mr, mc);
 942                         up_read(&current->mm->mmap_sem);
 943                         if (rc < 0)
 944                                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 945                 }
 946
 947                 start += PAGE_SIZE;
 948         }
 949         if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
 950                 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT) {
 951                         vcpu->run->s.regs.gprs[reg2] = end;
 952                 } else {
 953                         vcpu->run->s.regs.gprs[reg2] &= ~0xffffffffUL;
 954                         end = kvm_s390_logical_to_effective(vcpu, end);
 955                         vcpu->run->s.regs.gprs[reg2] |= end;
 956                 }
 957         }
 958         return 0;
 959 }
 960
 961 static inline int do_essa(struct kvm_vcpu *vcpu, const int orc)
 962 {
 963         struct kvm_s390_migration_state *ms = vcpu->kvm->arch.migration_state;
 964         int r1, r2, nappended, entries;
 965         unsigned long gfn, hva, res, pgstev, ptev;
 966         unsigned long *cbrlo;
 967
 968         /*
 969          * We don't need to set SD.FPF.SK to 1 here, because if we have a
 970          * machine check here we either handle it or crash
 971          */
 972
 973         kvm_s390_get_regs_rre(vcpu, &r1, &r2);
 974         gfn = vcpu->run->s.regs.gprs[r2] >> PAGE_SHIFT;
 975         hva = gfn_to_hva(vcpu->kvm, gfn);
 976         entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
 977
 978         if (kvm_is_error_hva(hva))
 979                 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 980
 981         nappended = pgste_perform_essa(vcpu->kvm->mm, hva, orc, &ptev, &pgstev);
 982         if (nappended < 0) {
 983                 res = orc ? 0x10 : 0;
 984                 vcpu->run->s.regs.gprs[r1] = res; /* Exception Indication */
 985                 return 0;
 986         }
 987         res = (pgstev & _PGSTE_GPS_USAGE_MASK) >> 22;
 988         /*
 989          * Set the block-content state part of the result. 0 means resident, so
 990          * nothing to do if the page is valid. 2 is for preserved pages
 991          * (non-present and non-zero), and 3 for zero pages (non-present and
 992          * zero).
 993          */
 994         if (ptev & _PAGE_INVALID) {
 995                 res |= 2;
 996                 if (pgstev & _PGSTE_GPS_ZERO)
 997                         res |= 1;
 998         }
 999         if (pgstev & _PGSTE_GPS_NODAT)
1000                 res |= 0x20;
1001         vcpu->run->s.regs.gprs[r1] = res;
1002         /*
1003          * It is possible that all the normal 511 slots were full, in which case
1004          * we will now write in the 512th slot, which is reserved for host use.
1005          * In both cases we let the normal essa handling code process all the
1006          * slots, including the reserved one, if needed.
1007          */
1008         if (nappended > 0) {
1009                 cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo & PAGE_MASK);
1010                 cbrlo[entries] = gfn << PAGE_SHIFT;
1011         }
1012
1013         if (orc && gfn < ms->bitmap_size) {
1014                 /* increment only if we are really flipping the bit to 1 */
1015                 if (!test_and_set_bit(gfn, ms->pgste_bitmap))
1016                         atomic64_inc(&ms->dirty_pages);
1017         }
1018
1019         return nappended;
1020 }
1021
1022 static int handle_essa(struct kvm_vcpu *vcpu)
1023 {
1024         /* entries expected to be 1FF */
1025         int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1026         unsigned long *cbrlo;
1027         struct gmap *gmap;
1028         int i, orc;
1029
1030         VCPU_EVENT(vcpu, 4, "ESSA: release %d pages", entries);
1031         gmap = vcpu->arch.gmap;
1032         vcpu->stat.instruction_essa++;
1033         if (!vcpu->kvm->arch.use_cmma)
1034                 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1035
1036         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1037                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1038         /* Check for invalid operation request code */
1039         orc = (vcpu->arch.sie_block->ipb & 0xf0000000) >> 28;
1040         /* ORCs 0-6 are always valid */
1041         if (orc > (test_kvm_facility(vcpu->kvm, 147) ? ESSA_SET_STABLE_NODAT
1042                                                 : ESSA_SET_STABLE_IF_RESIDENT))
1043                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1044
1045         if (likely(!vcpu->kvm->arch.migration_state)) {
1046                 /*
1047                  * CMMA is enabled in the KVM settings, but is disabled in
1048                  * the SIE block and in the mm_context, and we are not doing
1049                  * a migration. Enable CMMA in the mm_context.
1050                  * Since we need to take a write lock to write to the context
1051                  * to avoid races with storage keys handling, we check if the
1052                  * value really needs to be written to; if the value is
1053                  * already correct, we do nothing and avoid the lock.
1054                  */
1055                 if (vcpu->kvm->mm->context.use_cmma == 0) {
1056                         down_write(&vcpu->kvm->mm->mmap_sem);
1057                         vcpu->kvm->mm->context.use_cmma = 1;
1058                         up_write(&vcpu->kvm->mm->mmap_sem);
1059                 }
1060                 /*
1061                  * If we are here, we are supposed to have CMMA enabled in
1062                  * the SIE block. Enabling CMMA works on a per-CPU basis,
1063                  * while the context use_cmma flag is per process.
1064                  * It's possible that the context flag is enabled and the
1065                  * SIE flag is not, so we set the flag always; if it was
1066                  * already set, nothing changes, otherwise we enable it
1067                  * on this CPU too.
1068                  */
1069                 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
1070                 /* Retry the ESSA instruction */
1071                 kvm_s390_retry_instr(vcpu);
1072         } else {
1073                 /* Account for the possible extra cbrl entry */
1074                 i = do_essa(vcpu, orc);
1075                 if (i < 0)
1076                         return i;
1077                 entries += i;
1078         }
1079         vcpu->arch.sie_block->cbrlo &= PAGE_MASK;       /* reset nceo */
1080         cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
1081         down_read(&gmap->mm->mmap_sem);
1082         for (i = 0; i < entries; ++i)
1083                 __gmap_zap(gmap, cbrlo[i]);
1084         up_read(&gmap->mm->mmap_sem);
1085         return 0;
1086 }
1087
1088 static const intercept_handler_t b9_handlers[256] = {
1089         [0x8a] = handle_ipte_interlock,
1090         [0x8d] = handle_epsw,
1091         [0x8e] = handle_ipte_interlock,
1092         [0x8f] = handle_ipte_interlock,
1093         [0xab] = handle_essa,
1094         [0xaf] = handle_pfmf,
1095 };
1096
1097 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
1098 {
1099         intercept_handler_t handler;
1100
1101         /* This is handled just as for the B2 instructions. */
1102         handler = b9_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
1103         if (handler)
1104                 return handler(vcpu);
1105
1106         return -EOPNOTSUPP;
1107 }
1108
1109 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
1110 {
1111         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1112         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1113         int reg, rc, nr_regs;
1114         u32 ctl_array[16];
1115         u64 ga;
1116         u8 ar;
1117
1118         vcpu->stat.instruction_lctl++;
1119
1120         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1121                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1122
1123         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1124
1125         if (ga & 3)
1126                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1127
1128         VCPU_EVENT(vcpu, 4, "LCTL: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1129         trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
1130
1131         nr_regs = ((reg3 - reg1) & 0xf) + 1;
1132         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1133         if (rc)
1134                 return kvm_s390_inject_prog_cond(vcpu, rc);
1135         reg = reg1;
1136         nr_regs = 0;
1137         do {
1138                 vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
1139                 vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
1140                 if (reg == reg3)
1141                         break;
1142                 reg = (reg + 1) % 16;
1143         } while (1);
1144         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1145         return 0;
1146 }
1147
1148 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
1149 {
1150         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1151         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1152         int reg, rc, nr_regs;
1153         u32 ctl_array[16];
1154         u64 ga;
1155         u8 ar;
1156
1157         vcpu->stat.instruction_stctl++;
1158
1159         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1160                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1161
1162         ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1163
1164         if (ga & 3)
1165                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1166
1167         VCPU_EVENT(vcpu, 4, "STCTL r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1168         trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
1169
1170         reg = reg1;
1171         nr_regs = 0;
1172         do {
1173                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1174                 if (reg == reg3)
1175                         break;
1176                 reg = (reg + 1) % 16;
1177         } while (1);
1178         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1179         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1180 }
1181
1182 static int handle_lctlg(struct kvm_vcpu *vcpu)
1183 {
1184         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1185         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1186         int reg, rc, nr_regs;
1187         u64 ctl_array[16];
1188         u64 ga;
1189         u8 ar;
1190
1191         vcpu->stat.instruction_lctlg++;
1192
1193         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1194                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1195
1196         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1197
1198         if (ga & 7)
1199                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1200
1201         VCPU_EVENT(vcpu, 4, "LCTLG: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1202         trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
1203
1204         nr_regs = ((reg3 - reg1) & 0xf) + 1;
1205         rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1206         if (rc)
1207                 return kvm_s390_inject_prog_cond(vcpu, rc);
1208         reg = reg1;
1209         nr_regs = 0;
1210         do {
1211                 vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
1212                 if (reg == reg3)
1213                         break;
1214                 reg = (reg + 1) % 16;
1215         } while (1);
1216         kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1217         return 0;
1218 }
1219
1220 static int handle_stctg(struct kvm_vcpu *vcpu)
1221 {
1222         int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1223         int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1224         int reg, rc, nr_regs;
1225         u64 ctl_array[16];
1226         u64 ga;
1227         u8 ar;
1228
1229         vcpu->stat.instruction_stctg++;
1230
1231         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1232                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1233
1234         ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1235
1236         if (ga & 7)
1237                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1238
1239         VCPU_EVENT(vcpu, 4, "STCTG r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1240         trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
1241
1242         reg = reg1;
1243         nr_regs = 0;
1244         do {
1245                 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1246                 if (reg == reg3)
1247                         break;
1248                 reg = (reg + 1) % 16;
1249         } while (1);
1250         rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1251         return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1252 }
1253
1254 static const intercept_handler_t eb_handlers[256] = {
1255         [0x2f] = handle_lctlg,
1256         [0x25] = handle_stctg,
1257         [0x60] = handle_ri,
1258         [0x61] = handle_ri,
1259         [0x62] = handle_ri,
1260 };
1261
1262 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
1263 {
1264         intercept_handler_t handler;
1265
1266         handler = eb_handlers[vcpu->arch.sie_block->ipb & 0xff];
1267         if (handler)
1268                 return handler(vcpu);
1269         return -EOPNOTSUPP;
1270 }
1271
1272 static int handle_tprot(struct kvm_vcpu *vcpu)
1273 {
1274         u64 address1, address2;
1275         unsigned long hva, gpa;
1276         int ret = 0, cc = 0;
1277         bool writable;
1278         u8 ar;
1279
1280         vcpu->stat.instruction_tprot++;
1281
1282         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1283                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1284
1285         kvm_s390_get_base_disp_sse(vcpu, &address1, &address2, &ar, NULL);
1286
1287         /* we only handle the Linux memory detection case:
1288          * access key == 0
1289          * everything else goes to userspace. */
1290         if (address2 & 0xf0)
1291                 return -EOPNOTSUPP;
1292         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1293                 ipte_lock(vcpu);
1294         ret = guest_translate_address(vcpu, address1, ar, &gpa, GACC_STORE);
1295         if (ret == PGM_PROTECTION) {
1296                 /* Write protected? Try again with read-only... */
1297                 cc = 1;
1298                 ret = guest_translate_address(vcpu, address1, ar, &gpa,
1299                                               GACC_FETCH);
1300         }
1301         if (ret) {
1302                 if (ret == PGM_ADDRESSING || ret == PGM_TRANSLATION_SPEC) {
1303                         ret = kvm_s390_inject_program_int(vcpu, ret);
1304                 } else if (ret > 0) {
1305                         /* Translation not available */
1306                         kvm_s390_set_psw_cc(vcpu, 3);
1307                         ret = 0;
1308                 }
1309                 goto out_unlock;
1310         }
1311
1312         hva = gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
1313         if (kvm_is_error_hva(hva)) {
1314                 ret = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1315         } else {
1316                 if (!writable)
1317                         cc = 1;         /* Write not permitted ==> read-only */
1318                 kvm_s390_set_psw_cc(vcpu, cc);
1319                 /* Note: CC2 only occurs for storage keys (not supported yet) */
1320         }
1321 out_unlock:
1322         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1323                 ipte_unlock(vcpu);
1324         return ret;
1325 }
1326
1327 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
1328 {
1329         /* For e5xx... instructions we only handle TPROT */
1330         if ((vcpu->arch.sie_block->ipa & 0x00ff) == 0x01)
1331                 return handle_tprot(vcpu);
1332         return -EOPNOTSUPP;
1333 }
1334
1335 static int handle_sckpf(struct kvm_vcpu *vcpu)
1336 {
1337         u32 value;
1338
1339         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1340                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1341
1342         if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
1343                 return kvm_s390_inject_program_int(vcpu,
1344                                                    PGM_SPECIFICATION);
1345
1346         value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
1347         vcpu->arch.sie_block->todpr = value;
1348
1349         return 0;
1350 }
1351
1352 static int handle_ptff(struct kvm_vcpu *vcpu)
1353 {
1354         /* we don't emulate any control instructions yet */
1355         kvm_s390_set_psw_cc(vcpu, 3);
1356         return 0;
1357 }
1358
1359 static const intercept_handler_t x01_handlers[256] = {
1360         [0x04] = handle_ptff,
1361         [0x07] = handle_sckpf,
1362 };
1363
1364 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1365 {
1366         intercept_handler_t handler;
1367
1368         handler = x01_handlers[vcpu->arch.sie_block->ipa & 0x00ff];
1369         if (handler)
1370                 return handler(vcpu);
1371         return -EOPNOTSUPP;
1372 }