arch/s390/kvm/vsie.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * kvm nested virtualization support for s390x
   4  *
   5  * Copyright IBM Corp. 2016, 2018
   6  *
   7  *    Author(s): David Hildenbrand <[email protected]>
   8  */
   9 #include <linux/vmalloc.h>
  10 #include <linux/kvm_host.h>
  11 #include <linux/bug.h>
  12 #include <linux/list.h>
  13 #include <linux/bitmap.h>
  14 #include <linux/sched/signal.h>
  15 #include <linux/io.h>
  16
  17 #include <asm/gmap.h>
  18 #include <asm/mmu_context.h>
  19 #include <asm/sclp.h>
  20 #include <asm/nmi.h>
  21 #include <asm/dis.h>
  22 #include <asm/facility.h>
  23 #include "kvm-s390.h"
  24 #include "gaccess.h"
  25
  26 struct vsie_page {
  27         struct kvm_s390_sie_block scb_s;        /* 0x0000 */
  28         /*
  29          * the backup info for machine check. ensure it's at
  30          * the same offset as that in struct sie_page!
  31          */
  32         struct mcck_volatile_info mcck_info;    /* 0x0200 */
  33         /*
  34          * The pinned original scb. Be aware that other VCPUs can modify
  35          * it while we read from it. Values that are used for conditions or
  36          * are reused conditionally, should be accessed via READ_ONCE.
  37          */
  38         struct kvm_s390_sie_block *scb_o;       /* 0x0218 */
  39         /* the shadow gmap in use by the vsie_page */
  40         struct gmap *gmap;                      /* 0x0220 */
  41         /* address of the last reported fault to guest2 */
  42         unsigned long fault_addr;               /* 0x0228 */
  43         /* calculated guest addresses of satellite control blocks */
  44         gpa_t sca_gpa;                          /* 0x0230 */
  45         gpa_t itdba_gpa;                        /* 0x0238 */
  46         gpa_t gvrd_gpa;                         /* 0x0240 */
  47         gpa_t riccbd_gpa;                       /* 0x0248 */
  48         gpa_t sdnx_gpa;                         /* 0x0250 */
  49         __u8 reserved[0x0700 - 0x0258];         /* 0x0258 */
  50         struct kvm_s390_crypto_cb crycb;        /* 0x0700 */
  51         __u8 fac[S390_ARCH_FAC_LIST_SIZE_BYTE]; /* 0x0800 */
  52 };
  53
  54 /* trigger a validity icpt for the given scb */
  55 static int set_validity_icpt(struct kvm_s390_sie_block *scb,
  56                              __u16 reason_code)
  57 {
  58         scb->ipa = 0x1000;
  59         scb->ipb = ((__u32) reason_code) << 16;
  60         scb->icptcode = ICPT_VALIDITY;
  61         return 1;
  62 }
  63
  64 /* mark the prefix as unmapped, this will block the VSIE */
  65 static void prefix_unmapped(struct vsie_page *vsie_page)
  66 {
  67         atomic_or(PROG_REQUEST, &vsie_page->scb_s.prog20);
  68 }
  69
  70 /* mark the prefix as unmapped and wait until the VSIE has been left */
  71 static void prefix_unmapped_sync(struct vsie_page *vsie_page)
  72 {
  73         prefix_unmapped(vsie_page);
  74         if (vsie_page->scb_s.prog0c & PROG_IN_SIE)
  75                 atomic_or(CPUSTAT_STOP_INT, &vsie_page->scb_s.cpuflags);
  76         while (vsie_page->scb_s.prog0c & PROG_IN_SIE)
  77                 cpu_relax();
  78 }
  79
  80 /* mark the prefix as mapped, this will allow the VSIE to run */
  81 static void prefix_mapped(struct vsie_page *vsie_page)
  82 {
  83         atomic_andnot(PROG_REQUEST, &vsie_page->scb_s.prog20);
  84 }
  85
  86 /* test if the prefix is mapped into the gmap shadow */
  87 static int prefix_is_mapped(struct vsie_page *vsie_page)
  88 {
  89         return !(atomic_read(&vsie_page->scb_s.prog20) & PROG_REQUEST);
  90 }
  91
  92 /* copy the updated intervention request bits into the shadow scb */
  93 static void update_intervention_requests(struct vsie_page *vsie_page)
  94 {
  95         const int bits = CPUSTAT_STOP_INT | CPUSTAT_IO_INT | CPUSTAT_EXT_INT;
  96         int cpuflags;
  97
  98         cpuflags = atomic_read(&vsie_page->scb_o->cpuflags);
  99         atomic_andnot(bits, &vsie_page->scb_s.cpuflags);
 100         atomic_or(cpuflags & bits, &vsie_page->scb_s.cpuflags);
 101 }
 102
 103 /* shadow (filter and validate) the cpuflags  */
 104 static int prepare_cpuflags(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 105 {
 106         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 107         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 108         int newflags, cpuflags = atomic_read(&scb_o->cpuflags);
 109
 110         /* we don't allow ESA/390 guests */
 111         if (!(cpuflags & CPUSTAT_ZARCH))
 112                 return set_validity_icpt(scb_s, 0x0001U);
 113
 114         if (cpuflags & (CPUSTAT_RRF | CPUSTAT_MCDS))
 115                 return set_validity_icpt(scb_s, 0x0001U);
 116         else if (cpuflags & (CPUSTAT_SLSV | CPUSTAT_SLSR))
 117                 return set_validity_icpt(scb_s, 0x0007U);
 118
 119         /* intervention requests will be set later */
 120         newflags = CPUSTAT_ZARCH;
 121         if (cpuflags & CPUSTAT_GED && test_kvm_facility(vcpu->kvm, 8))
 122                 newflags |= CPUSTAT_GED;
 123         if (cpuflags & CPUSTAT_GED2 && test_kvm_facility(vcpu->kvm, 78)) {
 124                 if (cpuflags & CPUSTAT_GED)
 125                         return set_validity_icpt(scb_s, 0x0001U);
 126                 newflags |= CPUSTAT_GED2;
 127         }
 128         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GPERE))
 129                 newflags |= cpuflags & CPUSTAT_P;
 130         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GSLS))
 131                 newflags |= cpuflags & CPUSTAT_SM;
 132         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IBS))
 133                 newflags |= cpuflags & CPUSTAT_IBS;
 134         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_KSS))
 135                 newflags |= cpuflags & CPUSTAT_KSS;
 136
 137         atomic_set(&scb_s->cpuflags, newflags);
 138         return 0;
 139 }
 140 /* Copy to APCB FORMAT1 from APCB FORMAT0 */
 141 static int setup_apcb10(struct kvm_vcpu *vcpu, struct kvm_s390_apcb1 *apcb_s,
 142                         unsigned long crycb_gpa, struct kvm_s390_apcb1 *apcb_h)
 143 {
 144         struct kvm_s390_apcb0 tmp;
 145         unsigned long apcb_gpa;
 146
 147         apcb_gpa = crycb_gpa + offsetof(struct kvm_s390_crypto_cb, apcb0);
 148
 149         if (read_guest_real(vcpu, apcb_gpa, &tmp,
 150                             sizeof(struct kvm_s390_apcb0)))
 151                 return -EFAULT;
 152
 153         apcb_s->apm[0] = apcb_h->apm[0] & tmp.apm[0];
 154         apcb_s->aqm[0] = apcb_h->aqm[0] & tmp.aqm[0] & 0xffff000000000000UL;
 155         apcb_s->adm[0] = apcb_h->adm[0] & tmp.adm[0] & 0xffff000000000000UL;
 156
 157         return 0;
 158
 159 }
 160
 161 /**
 162  * setup_apcb00 - Copy to APCB FORMAT0 from APCB FORMAT0
 163  * @vcpu: pointer to the virtual CPU
 164  * @apcb_s: pointer to start of apcb in the shadow crycb
 165  * @crycb_gpa: guest physical address to start of original guest crycb
 166  * @apcb_h: pointer to start of apcb in the guest1
 167  *
 168  * Returns 0 and -EFAULT on error reading guest apcb
 169  */
 170 static int setup_apcb00(struct kvm_vcpu *vcpu, unsigned long *apcb_s,
 171                         unsigned long crycb_gpa, unsigned long *apcb_h)
 172 {
 173         unsigned long apcb_gpa;
 174
 175         apcb_gpa = crycb_gpa + offsetof(struct kvm_s390_crypto_cb, apcb0);
 176
 177         if (read_guest_real(vcpu, apcb_gpa, apcb_s,
 178                             sizeof(struct kvm_s390_apcb0)))
 179                 return -EFAULT;
 180
 181         bitmap_and(apcb_s, apcb_s, apcb_h,
 182                    BITS_PER_BYTE * sizeof(struct kvm_s390_apcb0));
 183
 184         return 0;
 185 }
 186
 187 /**
 188  * setup_apcb11 - Copy the FORMAT1 APCB from the guest to the shadow CRYCB
 189  * @vcpu: pointer to the virtual CPU
 190  * @apcb_s: pointer to start of apcb in the shadow crycb
 191  * @crycb_gpa: guest physical address to start of original guest crycb
 192  * @apcb_h: pointer to start of apcb in the host
 193  *
 194  * Returns 0 and -EFAULT on error reading guest apcb
 195  */
 196 static int setup_apcb11(struct kvm_vcpu *vcpu, unsigned long *apcb_s,
 197                         unsigned long crycb_gpa,
 198                         unsigned long *apcb_h)
 199 {
 200         unsigned long apcb_gpa;
 201
 202         apcb_gpa = crycb_gpa + offsetof(struct kvm_s390_crypto_cb, apcb1);
 203
 204         if (read_guest_real(vcpu, apcb_gpa, apcb_s,
 205                             sizeof(struct kvm_s390_apcb1)))
 206                 return -EFAULT;
 207
 208         bitmap_and(apcb_s, apcb_s, apcb_h,
 209                    BITS_PER_BYTE * sizeof(struct kvm_s390_apcb1));
 210
 211         return 0;
 212 }
 213
 214 /**
 215  * setup_apcb - Create a shadow copy of the apcb.
 216  * @vcpu: pointer to the virtual CPU
 217  * @crycb_s: pointer to shadow crycb
 218  * @crycb_gpa: guest physical address of original guest crycb
 219  * @crycb_h: pointer to the host crycb
 220  * @fmt_o: format of the original guest crycb.
 221  * @fmt_h: format of the host crycb.
 222  *
 223  * Checks the compatibility between the guest and host crycb and calls the
 224  * appropriate copy function.
 225  *
 226  * Return 0 or an error number if the guest and host crycb are incompatible.
 227  */
 228 static int setup_apcb(struct kvm_vcpu *vcpu, struct kvm_s390_crypto_cb *crycb_s,
 229                const u32 crycb_gpa,
 230                struct kvm_s390_crypto_cb *crycb_h,
 231                int fmt_o, int fmt_h)
 232 {
 233         switch (fmt_o) {
 234         case CRYCB_FORMAT2:
 235                 if ((crycb_gpa & PAGE_MASK) != ((crycb_gpa + 256) & PAGE_MASK))
 236                         return -EACCES;
 237                 if (fmt_h != CRYCB_FORMAT2)
 238                         return -EINVAL;
 239                 return setup_apcb11(vcpu, (unsigned long *)&crycb_s->apcb1,
 240                                     crycb_gpa,
 241                                     (unsigned long *)&crycb_h->apcb1);
 242         case CRYCB_FORMAT1:
 243                 switch (fmt_h) {
 244                 case CRYCB_FORMAT2:
 245                         return setup_apcb10(vcpu, &crycb_s->apcb1,
 246                                             crycb_gpa,
 247                                             &crycb_h->apcb1);
 248                 case CRYCB_FORMAT1:
 249                         return setup_apcb00(vcpu,
 250                                             (unsigned long *) &crycb_s->apcb0,
 251                                             crycb_gpa,
 252                                             (unsigned long *) &crycb_h->apcb0);
 253                 }
 254                 break;
 255         case CRYCB_FORMAT0:
 256                 if ((crycb_gpa & PAGE_MASK) != ((crycb_gpa + 32) & PAGE_MASK))
 257                         return -EACCES;
 258
 259                 switch (fmt_h) {
 260                 case CRYCB_FORMAT2:
 261                         return setup_apcb10(vcpu, &crycb_s->apcb1,
 262                                             crycb_gpa,
 263                                             &crycb_h->apcb1);
 264                 case CRYCB_FORMAT1:
 265                 case CRYCB_FORMAT0:
 266                         return setup_apcb00(vcpu,
 267                                             (unsigned long *) &crycb_s->apcb0,
 268                                             crycb_gpa,
 269                                             (unsigned long *) &crycb_h->apcb0);
 270                 }
 271         }
 272         return -EINVAL;
 273 }
 274
 275 /**
 276  * shadow_crycb - Create a shadow copy of the crycb block
 277  * @vcpu: a pointer to the virtual CPU
 278  * @vsie_page: a pointer to internal date used for the vSIE
 279  *
 280  * Create a shadow copy of the crycb block and setup key wrapping, if
 281  * requested for guest 3 and enabled for guest 2.
 282  *
 283  * We accept format-1 or format-2, but we convert format-1 into format-2
 284  * in the shadow CRYCB.
 285  * Using format-2 enables the firmware to choose the right format when
 286  * scheduling the SIE.
 287  * There is nothing to do for format-0.
 288  *
 289  * This function centralize the issuing of set_validity_icpt() for all
 290  * the subfunctions working on the crycb.
 291  *
 292  * Returns: - 0 if shadowed or nothing to do
 293  *          - > 0 if control has to be given to guest 2
 294  */
 295 static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 296 {
 297         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 298         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 299         const uint32_t crycbd_o = READ_ONCE(scb_o->crycbd);
 300         const u32 crycb_addr = crycbd_o & 0x7ffffff8U;
 301         unsigned long *b1, *b2;
 302         u8 ecb3_flags;
 303         u32 ecd_flags;
 304         int apie_h;
 305         int apie_s;
 306         int key_msk = test_kvm_facility(vcpu->kvm, 76);
 307         int fmt_o = crycbd_o & CRYCB_FORMAT_MASK;
 308         int fmt_h = vcpu->arch.sie_block->crycbd & CRYCB_FORMAT_MASK;
 309         int ret = 0;
 310
 311         scb_s->crycbd = 0;
 312
 313         apie_h = vcpu->arch.sie_block->eca & ECA_APIE;
 314         apie_s = apie_h & scb_o->eca;
 315         if (!apie_s && (!key_msk || (fmt_o == CRYCB_FORMAT0)))
 316                 return 0;
 317
 318         if (!crycb_addr)
 319                 return set_validity_icpt(scb_s, 0x0039U);
 320
 321         if (fmt_o == CRYCB_FORMAT1)
 322                 if ((crycb_addr & PAGE_MASK) !=
 323                     ((crycb_addr + 128) & PAGE_MASK))
 324                         return set_validity_icpt(scb_s, 0x003CU);
 325
 326         if (apie_s) {
 327                 ret = setup_apcb(vcpu, &vsie_page->crycb, crycb_addr,
 328                                  vcpu->kvm->arch.crypto.crycb,
 329                                  fmt_o, fmt_h);
 330                 if (ret)
 331                         goto end;
 332                 scb_s->eca |= scb_o->eca & ECA_APIE;
 333         }
 334
 335         /* we may only allow it if enabled for guest 2 */
 336         ecb3_flags = scb_o->ecb3 & vcpu->arch.sie_block->ecb3 &
 337                      (ECB3_AES | ECB3_DEA);
 338         ecd_flags = scb_o->ecd & vcpu->arch.sie_block->ecd &
 339                      (ECD_ECC | ECD_HMAC);
 340         if (!ecb3_flags && !ecd_flags)
 341                 goto end;
 342
 343         /* copy only the wrapping keys */
 344         if (read_guest_real(vcpu, crycb_addr + 72,
 345                             vsie_page->crycb.dea_wrapping_key_mask, 56))
 346                 return set_validity_icpt(scb_s, 0x0035U);
 347
 348         scb_s->ecb3 |= ecb3_flags;
 349         scb_s->ecd |= ecd_flags;
 350
 351         /* xor both blocks in one run */
 352         b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask;
 353         b2 = (unsigned long *)
 354                             vcpu->kvm->arch.crypto.crycb->dea_wrapping_key_mask;
 355         /* as 56%8 == 0, bitmap_xor won't overwrite any data */
 356         bitmap_xor(b1, b1, b2, BITS_PER_BYTE * 56);
 357 end:
 358         switch (ret) {
 359         case -EINVAL:
 360                 return set_validity_icpt(scb_s, 0x0022U);
 361         case -EFAULT:
 362                 return set_validity_icpt(scb_s, 0x0035U);
 363         case -EACCES:
 364                 return set_validity_icpt(scb_s, 0x003CU);
 365         }
 366         scb_s->crycbd = (u32)virt_to_phys(&vsie_page->crycb) | CRYCB_FORMAT2;
 367         return 0;
 368 }
 369
 370 /* shadow (round up/down) the ibc to avoid validity icpt */
 371 static void prepare_ibc(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 372 {
 373         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 374         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 375         /* READ_ONCE does not work on bitfields - use a temporary variable */
 376         const uint32_t __new_ibc = scb_o->ibc;
 377         const uint32_t new_ibc = READ_ONCE(__new_ibc) & 0x0fffU;
 378         __u64 min_ibc = (sclp.ibc >> 16) & 0x0fffU;
 379
 380         scb_s->ibc = 0;
 381         /* ibc installed in g2 and requested for g3 */
 382         if (vcpu->kvm->arch.model.ibc && new_ibc) {
 383                 scb_s->ibc = new_ibc;
 384                 /* takte care of the minimum ibc level of the machine */
 385                 if (scb_s->ibc < min_ibc)
 386                         scb_s->ibc = min_ibc;
 387                 /* take care of the maximum ibc level set for the guest */
 388                 if (scb_s->ibc > vcpu->kvm->arch.model.ibc)
 389                         scb_s->ibc = vcpu->kvm->arch.model.ibc;
 390         }
 391 }
 392
 393 /* unshadow the scb, copying parameters back to the real scb */
 394 static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 395 {
 396         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 397         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 398
 399         /* interception */
 400         scb_o->icptcode = scb_s->icptcode;
 401         scb_o->icptstatus = scb_s->icptstatus;
 402         scb_o->ipa = scb_s->ipa;
 403         scb_o->ipb = scb_s->ipb;
 404         scb_o->gbea = scb_s->gbea;
 405
 406         /* timer */
 407         scb_o->cputm = scb_s->cputm;
 408         scb_o->ckc = scb_s->ckc;
 409         scb_o->todpr = scb_s->todpr;
 410
 411         /* guest state */
 412         scb_o->gpsw = scb_s->gpsw;
 413         scb_o->gg14 = scb_s->gg14;
 414         scb_o->gg15 = scb_s->gg15;
 415         memcpy(scb_o->gcr, scb_s->gcr, 128);
 416         scb_o->pp = scb_s->pp;
 417
 418         /* branch prediction */
 419         if (test_kvm_facility(vcpu->kvm, 82)) {
 420                 scb_o->fpf &= ~FPF_BPBC;
 421                 scb_o->fpf |= scb_s->fpf & FPF_BPBC;
 422         }
 423
 424         /* interrupt intercept */
 425         switch (scb_s->icptcode) {
 426         case ICPT_PROGI:
 427         case ICPT_INSTPROGI:
 428         case ICPT_EXTINT:
 429                 memcpy((void *)((u64)scb_o + 0xc0),
 430                        (void *)((u64)scb_s + 0xc0), 0xf0 - 0xc0);
 431                 break;
 432         }
 433
 434         if (scb_s->ihcpu != 0xffffU)
 435                 scb_o->ihcpu = scb_s->ihcpu;
 436 }
 437
 438 /*
 439  * Setup the shadow scb by copying and checking the relevant parts of the g2
 440  * provided scb.
 441  *
 442  * Returns: - 0 if the scb has been shadowed
 443  *          - > 0 if control has to be given to guest 2
 444  */
 445 static int shadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 446 {
 447         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 448         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 449         /* READ_ONCE does not work on bitfields - use a temporary variable */
 450         const uint32_t __new_prefix = scb_o->prefix;
 451         const uint32_t new_prefix = READ_ONCE(__new_prefix);
 452         const bool wants_tx = READ_ONCE(scb_o->ecb) & ECB_TE;
 453         bool had_tx = scb_s->ecb & ECB_TE;
 454         unsigned long new_mso = 0;
 455         int rc;
 456
 457         /* make sure we don't have any leftovers when reusing the scb */
 458         scb_s->icptcode = 0;
 459         scb_s->eca = 0;
 460         scb_s->ecb = 0;
 461         scb_s->ecb2 = 0;
 462         scb_s->ecb3 = 0;
 463         scb_s->ecd = 0;
 464         scb_s->fac = 0;
 465         scb_s->fpf = 0;
 466
 467         rc = prepare_cpuflags(vcpu, vsie_page);
 468         if (rc)
 469                 goto out;
 470
 471         /* timer */
 472         scb_s->cputm = scb_o->cputm;
 473         scb_s->ckc = scb_o->ckc;
 474         scb_s->todpr = scb_o->todpr;
 475         scb_s->epoch = scb_o->epoch;
 476
 477         /* guest state */
 478         scb_s->gpsw = scb_o->gpsw;
 479         scb_s->gg14 = scb_o->gg14;
 480         scb_s->gg15 = scb_o->gg15;
 481         memcpy(scb_s->gcr, scb_o->gcr, 128);
 482         scb_s->pp = scb_o->pp;
 483
 484         /* interception / execution handling */
 485         scb_s->gbea = scb_o->gbea;
 486         scb_s->lctl = scb_o->lctl;
 487         scb_s->svcc = scb_o->svcc;
 488         scb_s->ictl = scb_o->ictl;
 489         /*
 490          * SKEY handling functions can't deal with false setting of PTE invalid
 491          * bits. Therefore we cannot provide interpretation and would later
 492          * have to provide own emulation handlers.
 493          */
 494         if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_KSS))
 495                 scb_s->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
 496
 497         scb_s->icpua = scb_o->icpua;
 498
 499         if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_SM))
 500                 new_mso = READ_ONCE(scb_o->mso) & 0xfffffffffff00000UL;
 501         /* if the hva of the prefix changes, we have to remap the prefix */
 502         if (scb_s->mso != new_mso || scb_s->prefix != new_prefix)
 503                 prefix_unmapped(vsie_page);
 504          /* SIE will do mso/msl validity and exception checks for us */
 505         scb_s->msl = scb_o->msl & 0xfffffffffff00000UL;
 506         scb_s->mso = new_mso;
 507         scb_s->prefix = new_prefix;
 508
 509         /* We have to definitely flush the tlb if this scb never ran */
 510         if (scb_s->ihcpu != 0xffffU)
 511                 scb_s->ihcpu = scb_o->ihcpu;
 512
 513         /* MVPG and Protection Exception Interpretation are always available */
 514         scb_s->eca |= scb_o->eca & (ECA_MVPGI | ECA_PROTEXCI);
 515         /* Host-protection-interruption introduced with ESOP */
 516         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_ESOP))
 517                 scb_s->ecb |= scb_o->ecb & ECB_HOSTPROTINT;
 518         /*
 519          * CPU Topology
 520          * This facility only uses the utility field of the SCA and none of
 521          * the cpu entries that are problematic with the other interpretation
 522          * facilities so we can pass it through
 523          */
 524         if (test_kvm_facility(vcpu->kvm, 11))
 525                 scb_s->ecb |= scb_o->ecb & ECB_PTF;
 526         /* transactional execution */
 527         if (test_kvm_facility(vcpu->kvm, 73) && wants_tx) {
 528                 /* remap the prefix is tx is toggled on */
 529                 if (!had_tx)
 530                         prefix_unmapped(vsie_page);
 531                 scb_s->ecb |= ECB_TE;
 532         }
 533         /* specification exception interpretation */
 534         scb_s->ecb |= scb_o->ecb & ECB_SPECI;
 535         /* branch prediction */
 536         if (test_kvm_facility(vcpu->kvm, 82))
 537                 scb_s->fpf |= scb_o->fpf & FPF_BPBC;
 538         /* SIMD */
 539         if (test_kvm_facility(vcpu->kvm, 129)) {
 540                 scb_s->eca |= scb_o->eca & ECA_VX;
 541                 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
 542         }
 543         /* Run-time-Instrumentation */
 544         if (test_kvm_facility(vcpu->kvm, 64))
 545                 scb_s->ecb3 |= scb_o->ecb3 & ECB3_RI;
 546         /* Instruction Execution Prevention */
 547         if (test_kvm_facility(vcpu->kvm, 130))
 548                 scb_s->ecb2 |= scb_o->ecb2 & ECB2_IEP;
 549         /* Guarded Storage */
 550         if (test_kvm_facility(vcpu->kvm, 133)) {
 551                 scb_s->ecb |= scb_o->ecb & ECB_GS;
 552                 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
 553         }
 554         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIIF))
 555                 scb_s->eca |= scb_o->eca & ECA_SII;
 556         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IB))
 557                 scb_s->eca |= scb_o->eca & ECA_IB;
 558         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_CEI))
 559                 scb_s->eca |= scb_o->eca & ECA_CEI;
 560         /* Epoch Extension */
 561         if (test_kvm_facility(vcpu->kvm, 139)) {
 562                 scb_s->ecd |= scb_o->ecd & ECD_MEF;
 563                 scb_s->epdx = scb_o->epdx;
 564         }
 565
 566         /* etoken */
 567         if (test_kvm_facility(vcpu->kvm, 156))
 568                 scb_s->ecd |= scb_o->ecd & ECD_ETOKENF;
 569
 570         scb_s->hpid = HPID_VSIE;
 571         scb_s->cpnc = scb_o->cpnc;
 572
 573         prepare_ibc(vcpu, vsie_page);
 574         rc = shadow_crycb(vcpu, vsie_page);
 575 out:
 576         if (rc)
 577                 unshadow_scb(vcpu, vsie_page);
 578         return rc;
 579 }
 580
 581 void kvm_s390_vsie_gmap_notifier(struct gmap *gmap, unsigned long start,
 582                                  unsigned long end)
 583 {
 584         struct kvm *kvm = gmap->private;
 585         struct vsie_page *cur;
 586         unsigned long prefix;
 587         struct page *page;
 588         int i;
 589
 590         if (!gmap_is_shadow(gmap))
 591                 return;
 592         /*
 593          * Only new shadow blocks are added to the list during runtime,
 594          * therefore we can safely reference them all the time.
 595          */
 596         for (i = 0; i < kvm->arch.vsie.page_count; i++) {
 597                 page = READ_ONCE(kvm->arch.vsie.pages[i]);
 598                 if (!page)
 599                         continue;
 600                 cur = page_to_virt(page);
 601                 if (READ_ONCE(cur->gmap) != gmap)
 602                         continue;
 603                 prefix = cur->scb_s.prefix << GUEST_PREFIX_SHIFT;
 604                 /* with mso/msl, the prefix lies at an offset */
 605                 prefix += cur->scb_s.mso;
 606                 if (prefix <= end && start <= prefix + 2 * PAGE_SIZE - 1)
 607                         prefix_unmapped_sync(cur);
 608         }
 609 }
 610
 611 /*
 612  * Map the first prefix page and if tx is enabled also the second prefix page.
 613  *
 614  * The prefix will be protected, a gmap notifier will inform about unmaps.
 615  * The shadow scb must not be executed until the prefix is remapped, this is
 616  * guaranteed by properly handling PROG_REQUEST.
 617  *
 618  * Returns: - 0 on if successfully mapped or already mapped
 619  *          - > 0 if control has to be given to guest 2
 620  *          - -EAGAIN if the caller can retry immediately
 621  *          - -ENOMEM if out of memory
 622  */
 623 static int map_prefix(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 624 {
 625         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 626         u64 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
 627         int rc;
 628
 629         if (prefix_is_mapped(vsie_page))
 630                 return 0;
 631
 632         /* mark it as mapped so we can catch any concurrent unmappers */
 633         prefix_mapped(vsie_page);
 634
 635         /* with mso/msl, the prefix lies at offset *mso* */
 636         prefix += scb_s->mso;
 637
 638         rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, prefix, NULL);
 639         if (!rc && (scb_s->ecb & ECB_TE))
 640                 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
 641                                            prefix + PAGE_SIZE, NULL);
 642         /*
 643          * We don't have to mprotect, we will be called for all unshadows.
 644          * SIE will detect if protection applies and trigger a validity.
 645          */
 646         if (rc)
 647                 prefix_unmapped(vsie_page);
 648         if (rc > 0 || rc == -EFAULT)
 649                 rc = set_validity_icpt(scb_s, 0x0037U);
 650         return rc;
 651 }
 652
 653 /*
 654  * Pin the guest page given by gpa and set hpa to the pinned host address.
 655  * Will always be pinned writable.
 656  *
 657  * Returns: - 0 on success
 658  *          - -EINVAL if the gpa is not valid guest storage
 659  */
 660 static int pin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t *hpa)
 661 {
 662         struct page *page;
 663
 664         page = gfn_to_page(kvm, gpa_to_gfn(gpa));
 665         if (!page)
 666                 return -EINVAL;
 667         *hpa = (hpa_t)page_to_phys(page) + (gpa & ~PAGE_MASK);
 668         return 0;
 669 }
 670
 671 /* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */
 672 static void unpin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t hpa)
 673 {
 674         kvm_release_page_dirty(pfn_to_page(hpa >> PAGE_SHIFT));
 675         /* mark the page always as dirty for migration */
 676         mark_page_dirty(kvm, gpa_to_gfn(gpa));
 677 }
 678
 679 /* unpin all blocks previously pinned by pin_blocks(), marking them dirty */
 680 static void unpin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 681 {
 682         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 683         hpa_t hpa;
 684
 685         hpa = (u64) scb_s->scaoh << 32 | scb_s->scaol;
 686         if (hpa) {
 687                 unpin_guest_page(vcpu->kvm, vsie_page->sca_gpa, hpa);
 688                 vsie_page->sca_gpa = 0;
 689                 scb_s->scaol = 0;
 690                 scb_s->scaoh = 0;
 691         }
 692
 693         hpa = scb_s->itdba;
 694         if (hpa) {
 695                 unpin_guest_page(vcpu->kvm, vsie_page->itdba_gpa, hpa);
 696                 vsie_page->itdba_gpa = 0;
 697                 scb_s->itdba = 0;
 698         }
 699
 700         hpa = scb_s->gvrd;
 701         if (hpa) {
 702                 unpin_guest_page(vcpu->kvm, vsie_page->gvrd_gpa, hpa);
 703                 vsie_page->gvrd_gpa = 0;
 704                 scb_s->gvrd = 0;
 705         }
 706
 707         hpa = scb_s->riccbd;
 708         if (hpa) {
 709                 unpin_guest_page(vcpu->kvm, vsie_page->riccbd_gpa, hpa);
 710                 vsie_page->riccbd_gpa = 0;
 711                 scb_s->riccbd = 0;
 712         }
 713
 714         hpa = scb_s->sdnxo;
 715         if (hpa) {
 716                 unpin_guest_page(vcpu->kvm, vsie_page->sdnx_gpa, hpa);
 717                 vsie_page->sdnx_gpa = 0;
 718                 scb_s->sdnxo = 0;
 719         }
 720 }
 721
 722 /*
 723  * Instead of shadowing some blocks, we can simply forward them because the
 724  * addresses in the scb are 64 bit long.
 725  *
 726  * This works as long as the data lies in one page. If blocks ever exceed one
 727  * page, we have to fall back to shadowing.
 728  *
 729  * As we reuse the sca, the vcpu pointers contained in it are invalid. We must
 730  * therefore not enable any facilities that access these pointers (e.g. SIGPIF).
 731  *
 732  * Returns: - 0 if all blocks were pinned.
 733  *          - > 0 if control has to be given to guest 2
 734  *          - -ENOMEM if out of memory
 735  */
 736 static int pin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 737 {
 738         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 739         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 740         hpa_t hpa;
 741         gpa_t gpa;
 742         int rc = 0;
 743
 744         gpa = READ_ONCE(scb_o->scaol) & ~0xfUL;
 745         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO))
 746                 gpa |= (u64) READ_ONCE(scb_o->scaoh) << 32;
 747         if (gpa) {
 748                 if (gpa < 2 * PAGE_SIZE)
 749                         rc = set_validity_icpt(scb_s, 0x0038U);
 750                 else if ((gpa & ~0x1fffUL) == kvm_s390_get_prefix(vcpu))
 751                         rc = set_validity_icpt(scb_s, 0x0011U);
 752                 else if ((gpa & PAGE_MASK) !=
 753                          ((gpa + sizeof(struct bsca_block) - 1) & PAGE_MASK))
 754                         rc = set_validity_icpt(scb_s, 0x003bU);
 755                 if (!rc) {
 756                         rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 757                         if (rc)
 758                                 rc = set_validity_icpt(scb_s, 0x0034U);
 759                 }
 760                 if (rc)
 761                         goto unpin;
 762                 vsie_page->sca_gpa = gpa;
 763                 scb_s->scaoh = (u32)((u64)hpa >> 32);
 764                 scb_s->scaol = (u32)(u64)hpa;
 765         }
 766
 767         gpa = READ_ONCE(scb_o->itdba) & ~0xffUL;
 768         if (gpa && (scb_s->ecb & ECB_TE)) {
 769                 if (gpa < 2 * PAGE_SIZE) {
 770                         rc = set_validity_icpt(scb_s, 0x0080U);
 771                         goto unpin;
 772                 }
 773                 /* 256 bytes cannot cross page boundaries */
 774                 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 775                 if (rc) {
 776                         rc = set_validity_icpt(scb_s, 0x0080U);
 777                         goto unpin;
 778                 }
 779                 vsie_page->itdba_gpa = gpa;
 780                 scb_s->itdba = hpa;
 781         }
 782
 783         gpa = READ_ONCE(scb_o->gvrd) & ~0x1ffUL;
 784         if (gpa && (scb_s->eca & ECA_VX) && !(scb_s->ecd & ECD_HOSTREGMGMT)) {
 785                 if (gpa < 2 * PAGE_SIZE) {
 786                         rc = set_validity_icpt(scb_s, 0x1310U);
 787                         goto unpin;
 788                 }
 789                 /*
 790                  * 512 bytes vector registers cannot cross page boundaries
 791                  * if this block gets bigger, we have to shadow it.
 792                  */
 793                 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 794                 if (rc) {
 795                         rc = set_validity_icpt(scb_s, 0x1310U);
 796                         goto unpin;
 797                 }
 798                 vsie_page->gvrd_gpa = gpa;
 799                 scb_s->gvrd = hpa;
 800         }
 801
 802         gpa = READ_ONCE(scb_o->riccbd) & ~0x3fUL;
 803         if (gpa && (scb_s->ecb3 & ECB3_RI)) {
 804                 if (gpa < 2 * PAGE_SIZE) {
 805                         rc = set_validity_icpt(scb_s, 0x0043U);
 806                         goto unpin;
 807                 }
 808                 /* 64 bytes cannot cross page boundaries */
 809                 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 810                 if (rc) {
 811                         rc = set_validity_icpt(scb_s, 0x0043U);
 812                         goto unpin;
 813                 }
 814                 /* Validity 0x0044 will be checked by SIE */
 815                 vsie_page->riccbd_gpa = gpa;
 816                 scb_s->riccbd = hpa;
 817         }
 818         if (((scb_s->ecb & ECB_GS) && !(scb_s->ecd & ECD_HOSTREGMGMT)) ||
 819             (scb_s->ecd & ECD_ETOKENF)) {
 820                 unsigned long sdnxc;
 821
 822                 gpa = READ_ONCE(scb_o->sdnxo) & ~0xfUL;
 823                 sdnxc = READ_ONCE(scb_o->sdnxo) & 0xfUL;
 824                 if (!gpa || gpa < 2 * PAGE_SIZE) {
 825                         rc = set_validity_icpt(scb_s, 0x10b0U);
 826                         goto unpin;
 827                 }
 828                 if (sdnxc < 6 || sdnxc > 12) {
 829                         rc = set_validity_icpt(scb_s, 0x10b1U);
 830                         goto unpin;
 831                 }
 832                 if (gpa & ((1 << sdnxc) - 1)) {
 833                         rc = set_validity_icpt(scb_s, 0x10b2U);
 834                         goto unpin;
 835                 }
 836                 /* Due to alignment rules (checked above) this cannot
 837                  * cross page boundaries
 838                  */
 839                 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 840                 if (rc) {
 841                         rc = set_validity_icpt(scb_s, 0x10b0U);
 842                         goto unpin;
 843                 }
 844                 vsie_page->sdnx_gpa = gpa;
 845                 scb_s->sdnxo = hpa | sdnxc;
 846         }
 847         return 0;
 848 unpin:
 849         unpin_blocks(vcpu, vsie_page);
 850         return rc;
 851 }
 852
 853 /* unpin the scb provided by guest 2, marking it as dirty */
 854 static void unpin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
 855                       gpa_t gpa)
 856 {
 857         hpa_t hpa = (hpa_t) vsie_page->scb_o;
 858
 859         if (hpa)
 860                 unpin_guest_page(vcpu->kvm, gpa, hpa);
 861         vsie_page->scb_o = NULL;
 862 }
 863
 864 /*
 865  * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o.
 866  *
 867  * Returns: - 0 if the scb was pinned.
 868  *          - > 0 if control has to be given to guest 2
 869  */
 870 static int pin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
 871                    gpa_t gpa)
 872 {
 873         hpa_t hpa;
 874         int rc;
 875
 876         rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 877         if (rc) {
 878                 rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 879                 WARN_ON_ONCE(rc);
 880                 return 1;
 881         }
 882         vsie_page->scb_o = phys_to_virt(hpa);
 883         return 0;
 884 }
 885
 886 /*
 887  * Inject a fault into guest 2.
 888  *
 889  * Returns: - > 0 if control has to be given to guest 2
 890  *            < 0 if an error occurred during injection.
 891  */
 892 static int inject_fault(struct kvm_vcpu *vcpu, __u16 code, __u64 vaddr,
 893                         bool write_flag)
 894 {
 895         struct kvm_s390_pgm_info pgm = {
 896                 .code = code,
 897                 .trans_exc_code =
 898                         /* 0-51: virtual address */
 899                         (vaddr & 0xfffffffffffff000UL) |
 900                         /* 52-53: store / fetch */
 901                         (((unsigned int) !write_flag) + 1) << 10,
 902                         /* 62-63: asce id (always primary == 0) */
 903                 .exc_access_id = 0, /* always primary */
 904                 .op_access_id = 0, /* not MVPG */
 905         };
 906         int rc;
 907
 908         if (code == PGM_PROTECTION)
 909                 pgm.trans_exc_code |= 0x4UL;
 910
 911         rc = kvm_s390_inject_prog_irq(vcpu, &pgm);
 912         return rc ? rc : 1;
 913 }
 914
 915 /*
 916  * Handle a fault during vsie execution on a gmap shadow.
 917  *
 918  * Returns: - 0 if the fault was resolved
 919  *          - > 0 if control has to be given to guest 2
 920  *          - < 0 if an error occurred
 921  */
 922 static int handle_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 923 {
 924         int rc;
 925
 926         if ((current->thread.gmap_int_code & PGM_INT_CODE_MASK) == PGM_PROTECTION)
 927                 /* we can directly forward all protection exceptions */
 928                 return inject_fault(vcpu, PGM_PROTECTION,
 929                                     current->thread.gmap_teid.addr * PAGE_SIZE, 1);
 930
 931         rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
 932                                    current->thread.gmap_teid.addr * PAGE_SIZE, NULL);
 933         if (rc > 0) {
 934                 rc = inject_fault(vcpu, rc,
 935                                   current->thread.gmap_teid.addr * PAGE_SIZE,
 936                                   kvm_s390_cur_gmap_fault_is_write());
 937                 if (rc >= 0)
 938                         vsie_page->fault_addr = current->thread.gmap_teid.addr * PAGE_SIZE;
 939         }
 940         return rc;
 941 }
 942
 943 /*
 944  * Retry the previous fault that required guest 2 intervention. This avoids
 945  * one superfluous SIE re-entry and direct exit.
 946  *
 947  * Will ignore any errors. The next SIE fault will do proper fault handling.
 948  */
 949 static void handle_last_fault(struct kvm_vcpu *vcpu,
 950                               struct vsie_page *vsie_page)
 951 {
 952         if (vsie_page->fault_addr)
 953                 kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
 954                                       vsie_page->fault_addr, NULL);
 955         vsie_page->fault_addr = 0;
 956 }
 957
 958 static inline void clear_vsie_icpt(struct vsie_page *vsie_page)
 959 {
 960         vsie_page->scb_s.icptcode = 0;
 961 }
 962
 963 /* rewind the psw and clear the vsie icpt, so we can retry execution */
 964 static void retry_vsie_icpt(struct vsie_page *vsie_page)
 965 {
 966         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 967         int ilen = insn_length(scb_s->ipa >> 8);
 968
 969         /* take care of EXECUTE instructions */
 970         if (scb_s->icptstatus & 1) {
 971                 ilen = (scb_s->icptstatus >> 4) & 0x6;
 972                 if (!ilen)
 973                         ilen = 4;
 974         }
 975         scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, ilen);
 976         clear_vsie_icpt(vsie_page);
 977 }
 978
 979 /*
 980  * Try to shadow + enable the guest 2 provided facility list.
 981  * Retry instruction execution if enabled for and provided by guest 2.
 982  *
 983  * Returns: - 0 if handled (retry or guest 2 icpt)
 984  *          - > 0 if control has to be given to guest 2
 985  */
 986 static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 987 {
 988         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 989         __u32 fac = READ_ONCE(vsie_page->scb_o->fac);
 990
 991         /*
 992          * Alternate-STFLE-Interpretive-Execution facilities are not supported
 993          * -> format-0 flcb
 994          */
 995         if (fac && test_kvm_facility(vcpu->kvm, 7)) {
 996                 retry_vsie_icpt(vsie_page);
 997                 /*
 998                  * The facility list origin (FLO) is in bits 1 - 28 of the FLD
 999                  * so we need to mask here before reading.
1000                  */
1001                 fac = fac & 0x7ffffff8U;
1002                 /*
1003                  * format-0 -> size of nested guest's facility list == guest's size
1004                  * guest's size == host's size, since STFLE is interpretatively executed
1005                  * using a format-0 for the guest, too.
1006                  */
1007                 if (read_guest_real(vcpu, fac, &vsie_page->fac,
1008                                     stfle_size() * sizeof(u64)))
1009                         return set_validity_icpt(scb_s, 0x1090U);
1010                 scb_s->fac = (u32)virt_to_phys(&vsie_page->fac);
1011         }
1012         return 0;
1013 }
1014
1015 /*
1016  * Get a register for a nested guest.
1017  * @vcpu the vcpu of the guest
1018  * @vsie_page the vsie_page for the nested guest
1019  * @reg the register number, the upper 4 bits are ignored.
1020  * returns: the value of the register.
1021  */
1022 static u64 vsie_get_register(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, u8 reg)
1023 {
1024         /* no need to validate the parameter and/or perform error handling */
1025         reg &= 0xf;
1026         switch (reg) {
1027         case 15:
1028                 return vsie_page->scb_s.gg15;
1029         case 14:
1030                 return vsie_page->scb_s.gg14;
1031         default:
1032                 return vcpu->run->s.regs.gprs[reg];
1033         }
1034 }
1035
1036 static int vsie_handle_mvpg(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1037 {
1038         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1039         unsigned long pei_dest, pei_src, src, dest, mask, prefix;
1040         u64 *pei_block = &vsie_page->scb_o->mcic;
1041         int edat, rc_dest, rc_src;
1042         union ctlreg0 cr0;
1043
1044         cr0.val = vcpu->arch.sie_block->gcr[0];
1045         edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
1046         mask = _kvm_s390_logical_to_effective(&scb_s->gpsw, PAGE_MASK);
1047         prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
1048
1049         dest = vsie_get_register(vcpu, vsie_page, scb_s->ipb >> 20) & mask;
1050         dest = _kvm_s390_real_to_abs(prefix, dest) + scb_s->mso;
1051         src = vsie_get_register(vcpu, vsie_page, scb_s->ipb >> 16) & mask;
1052         src = _kvm_s390_real_to_abs(prefix, src) + scb_s->mso;
1053
1054         rc_dest = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, dest, &pei_dest);
1055         rc_src = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, src, &pei_src);
1056         /*
1057          * Either everything went well, or something non-critical went wrong
1058          * e.g. because of a race. In either case, simply retry.
1059          */
1060         if (rc_dest == -EAGAIN || rc_src == -EAGAIN || (!rc_dest && !rc_src)) {
1061                 retry_vsie_icpt(vsie_page);
1062                 return -EAGAIN;
1063         }
1064         /* Something more serious went wrong, propagate the error */
1065         if (rc_dest < 0)
1066                 return rc_dest;
1067         if (rc_src < 0)
1068                 return rc_src;
1069
1070         /* The only possible suppressing exception: just deliver it */
1071         if (rc_dest == PGM_TRANSLATION_SPEC || rc_src == PGM_TRANSLATION_SPEC) {
1072                 clear_vsie_icpt(vsie_page);
1073                 rc_dest = kvm_s390_inject_program_int(vcpu, PGM_TRANSLATION_SPEC);
1074                 WARN_ON_ONCE(rc_dest);
1075                 return 1;
1076         }
1077
1078         /*
1079          * Forward the PEI intercept to the guest if it was a page fault, or
1080          * also for segment and region table faults if EDAT applies.
1081          */
1082         if (edat) {
1083                 rc_dest = rc_dest == PGM_ASCE_TYPE ? rc_dest : 0;
1084                 rc_src = rc_src == PGM_ASCE_TYPE ? rc_src : 0;
1085         } else {
1086                 rc_dest = rc_dest != PGM_PAGE_TRANSLATION ? rc_dest : 0;
1087                 rc_src = rc_src != PGM_PAGE_TRANSLATION ? rc_src : 0;
1088         }
1089         if (!rc_dest && !rc_src) {
1090                 pei_block[0] = pei_dest;
1091                 pei_block[1] = pei_src;
1092                 return 1;
1093         }
1094
1095         retry_vsie_icpt(vsie_page);
1096
1097         /*
1098          * The host has edat, and the guest does not, or it was an ASCE type
1099          * exception. The host needs to inject the appropriate DAT interrupts
1100          * into the guest.
1101          */
1102         if (rc_dest)
1103                 return inject_fault(vcpu, rc_dest, dest, 1);
1104         return inject_fault(vcpu, rc_src, src, 0);
1105 }
1106
1107 /*
1108  * Run the vsie on a shadow scb and a shadow gmap, without any further
1109  * sanity checks, handling SIE faults.
1110  *
1111  * Returns: - 0 everything went fine
1112  *          - > 0 if control has to be given to guest 2
1113  *          - < 0 if an error occurred
1114  */
1115 static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1116         __releases(vcpu->kvm->srcu)
1117         __acquires(vcpu->kvm->srcu)
1118 {
1119         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1120         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
1121         int guest_bp_isolation;
1122         int rc = 0;
1123
1124         handle_last_fault(vcpu, vsie_page);
1125
1126         kvm_vcpu_srcu_read_unlock(vcpu);
1127
1128         /* save current guest state of bp isolation override */
1129         guest_bp_isolation = test_thread_flag(TIF_ISOLATE_BP_GUEST);
1130
1131         /*
1132          * The guest is running with BPBC, so we have to force it on for our
1133          * nested guest. This is done by enabling BPBC globally, so the BPBC
1134          * control in the SCB (which the nested guest can modify) is simply
1135          * ignored.
1136          */
1137         if (test_kvm_facility(vcpu->kvm, 82) &&
1138             vcpu->arch.sie_block->fpf & FPF_BPBC)
1139                 set_thread_flag(TIF_ISOLATE_BP_GUEST);
1140
1141         local_irq_disable();
1142         guest_enter_irqoff();
1143         local_irq_enable();
1144
1145         /*
1146          * Simulate a SIE entry of the VCPU (see sie64a), so VCPU blocking
1147          * and VCPU requests also hinder the vSIE from running and lead
1148          * to an immediate exit. kvm_s390_vsie_kick() has to be used to
1149          * also kick the vSIE.
1150          */
1151         vcpu->arch.sie_block->prog0c |= PROG_IN_SIE;
1152         current->thread.gmap_int_code = 0;
1153         barrier();
1154         if (!kvm_s390_vcpu_sie_inhibited(vcpu))
1155                 rc = sie64a(scb_s, vcpu->run->s.regs.gprs, vsie_page->gmap->asce);
1156         barrier();
1157         vcpu->arch.sie_block->prog0c &= ~PROG_IN_SIE;
1158
1159         local_irq_disable();
1160         guest_exit_irqoff();
1161         local_irq_enable();
1162
1163         /* restore guest state for bp isolation override */
1164         if (!guest_bp_isolation)
1165                 clear_thread_flag(TIF_ISOLATE_BP_GUEST);
1166
1167         kvm_vcpu_srcu_read_lock(vcpu);
1168
1169         if (rc == -EINTR) {
1170                 VCPU_EVENT(vcpu, 3, "%s", "machine check");
1171                 kvm_s390_reinject_machine_check(vcpu, &vsie_page->mcck_info);
1172                 return 0;
1173         }
1174
1175         if (rc > 0)
1176                 rc = 0; /* we could still have an icpt */
1177         else if (current->thread.gmap_int_code)
1178                 return handle_fault(vcpu, vsie_page);
1179
1180         switch (scb_s->icptcode) {
1181         case ICPT_INST:
1182                 if (scb_s->ipa == 0xb2b0)
1183                         rc = handle_stfle(vcpu, vsie_page);
1184                 break;
1185         case ICPT_STOP:
1186                 /* stop not requested by g2 - must have been a kick */
1187                 if (!(atomic_read(&scb_o->cpuflags) & CPUSTAT_STOP_INT))
1188                         clear_vsie_icpt(vsie_page);
1189                 break;
1190         case ICPT_VALIDITY:
1191                 if ((scb_s->ipa & 0xf000) != 0xf000)
1192                         scb_s->ipa += 0x1000;
1193                 break;
1194         case ICPT_PARTEXEC:
1195                 if (scb_s->ipa == 0xb254)
1196                         rc = vsie_handle_mvpg(vcpu, vsie_page);
1197                 break;
1198         }
1199         return rc;
1200 }
1201
1202 static void release_gmap_shadow(struct vsie_page *vsie_page)
1203 {
1204         if (vsie_page->gmap)
1205                 gmap_put(vsie_page->gmap);
1206         WRITE_ONCE(vsie_page->gmap, NULL);
1207         prefix_unmapped(vsie_page);
1208 }
1209
1210 static int acquire_gmap_shadow(struct kvm_vcpu *vcpu,
1211                                struct vsie_page *vsie_page)
1212 {
1213         unsigned long asce;
1214         union ctlreg0 cr0;
1215         struct gmap *gmap;
1216         int edat;
1217
1218         asce = vcpu->arch.sie_block->gcr[1];
1219         cr0.val = vcpu->arch.sie_block->gcr[0];
1220         edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
1221         edat += edat && test_kvm_facility(vcpu->kvm, 78);
1222
1223         /*
1224          * ASCE or EDAT could have changed since last icpt, or the gmap
1225          * we're holding has been unshadowed. If the gmap is still valid,
1226          * we can safely reuse it.
1227          */
1228         if (vsie_page->gmap && gmap_shadow_valid(vsie_page->gmap, asce, edat)) {
1229                 vcpu->kvm->stat.gmap_shadow_reuse++;
1230                 return 0;
1231         }
1232
1233         /* release the old shadow - if any, and mark the prefix as unmapped */
1234         release_gmap_shadow(vsie_page);
1235         gmap = gmap_shadow(vcpu->arch.gmap, asce, edat);
1236         if (IS_ERR(gmap))
1237                 return PTR_ERR(gmap);
1238         vcpu->kvm->stat.gmap_shadow_create++;
1239         WRITE_ONCE(vsie_page->gmap, gmap);
1240         return 0;
1241 }
1242
1243 /*
1244  * Register the shadow scb at the VCPU, e.g. for kicking out of vsie.
1245  */
1246 static void register_shadow_scb(struct kvm_vcpu *vcpu,
1247                                 struct vsie_page *vsie_page)
1248 {
1249         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1250
1251         WRITE_ONCE(vcpu->arch.vsie_block, &vsie_page->scb_s);
1252         /*
1253          * External calls have to lead to a kick of the vcpu and
1254          * therefore the vsie -> Simulate Wait state.
1255          */
1256         kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
1257         /*
1258          * We have to adjust the g3 epoch by the g2 epoch. The epoch will
1259          * automatically be adjusted on tod clock changes via kvm_sync_clock.
1260          */
1261         preempt_disable();
1262         scb_s->epoch += vcpu->kvm->arch.epoch;
1263
1264         if (scb_s->ecd & ECD_MEF) {
1265                 scb_s->epdx += vcpu->kvm->arch.epdx;
1266                 if (scb_s->epoch < vcpu->kvm->arch.epoch)
1267                         scb_s->epdx += 1;
1268         }
1269
1270         preempt_enable();
1271 }
1272
1273 /*
1274  * Unregister a shadow scb from a VCPU.
1275  */
1276 static void unregister_shadow_scb(struct kvm_vcpu *vcpu)
1277 {
1278         kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
1279         WRITE_ONCE(vcpu->arch.vsie_block, NULL);
1280 }
1281
1282 /*
1283  * Run the vsie on a shadowed scb, managing the gmap shadow, handling
1284  * prefix pages and faults.
1285  *
1286  * Returns: - 0 if no errors occurred
1287  *          - > 0 if control has to be given to guest 2
1288  *          - -ENOMEM if out of memory
1289  */
1290 static int vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1291 {
1292         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1293         int rc = 0;
1294
1295         while (1) {
1296                 rc = acquire_gmap_shadow(vcpu, vsie_page);
1297                 if (!rc)
1298                         rc = map_prefix(vcpu, vsie_page);
1299                 if (!rc) {
1300                         update_intervention_requests(vsie_page);
1301                         rc = do_vsie_run(vcpu, vsie_page);
1302                 }
1303                 atomic_andnot(PROG_BLOCK_SIE, &scb_s->prog20);
1304
1305                 if (rc == -EAGAIN)
1306                         rc = 0;
1307
1308                 /*
1309                  * Exit the loop if the guest needs to process the intercept
1310                  */
1311                 if (rc || scb_s->icptcode)
1312                         break;
1313
1314                 /*
1315                  * Exit the loop if the host needs to process an intercept,
1316                  * but rewind the PSW to re-enter SIE once that's completed
1317                  * instead of passing a "no action" intercept to the guest.
1318                  */
1319                 if (signal_pending(current) ||
1320                     kvm_s390_vcpu_has_irq(vcpu, 0) ||
1321                     kvm_s390_vcpu_sie_inhibited(vcpu)) {
1322                         kvm_s390_rewind_psw(vcpu, 4);
1323                         break;
1324                 }
1325                 cond_resched();
1326         }
1327
1328         if (rc == -EFAULT) {
1329                 /*
1330                  * Addressing exceptions are always presentes as intercepts.
1331                  * As addressing exceptions are suppressing and our guest 3 PSW
1332                  * points at the responsible instruction, we have to
1333                  * forward the PSW and set the ilc. If we can't read guest 3
1334                  * instruction, we can use an arbitrary ilc. Let's always use
1335                  * ilen = 4 for now, so we can avoid reading in guest 3 virtual
1336                  * memory. (we could also fake the shadow so the hardware
1337                  * handles it).
1338                  */
1339                 scb_s->icptcode = ICPT_PROGI;
1340                 scb_s->iprcc = PGM_ADDRESSING;
1341                 scb_s->pgmilc = 4;
1342                 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, 4);
1343                 rc = 1;
1344         }
1345         return rc;
1346 }
1347
1348 /*
1349  * Get or create a vsie page for a scb address.
1350  *
1351  * Returns: - address of a vsie page (cached or new one)
1352  *          - NULL if the same scb address is already used by another VCPU
1353  *          - ERR_PTR(-ENOMEM) if out of memory
1354  */
1355 static struct vsie_page *get_vsie_page(struct kvm *kvm, unsigned long addr)
1356 {
1357         struct vsie_page *vsie_page;
1358         struct page *page;
1359         int nr_vcpus;
1360
1361         rcu_read_lock();
1362         page = radix_tree_lookup(&kvm->arch.vsie.addr_to_page, addr >> 9);
1363         rcu_read_unlock();
1364         if (page) {
1365                 if (page_ref_inc_return(page) == 2)
1366                         return page_to_virt(page);
1367                 page_ref_dec(page);
1368         }
1369
1370         /*
1371          * We want at least #online_vcpus shadows, so every VCPU can execute
1372          * the VSIE in parallel.
1373          */
1374         nr_vcpus = atomic_read(&kvm->online_vcpus);
1375
1376         mutex_lock(&kvm->arch.vsie.mutex);
1377         if (kvm->arch.vsie.page_count < nr_vcpus) {
1378                 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO | GFP_DMA);
1379                 if (!page) {
1380                         mutex_unlock(&kvm->arch.vsie.mutex);
1381                         return ERR_PTR(-ENOMEM);
1382                 }
1383                 page_ref_inc(page);
1384                 kvm->arch.vsie.pages[kvm->arch.vsie.page_count] = page;
1385                 kvm->arch.vsie.page_count++;
1386         } else {
1387                 /* reuse an existing entry that belongs to nobody */
1388                 while (true) {
1389                         page = kvm->arch.vsie.pages[kvm->arch.vsie.next];
1390                         if (page_ref_inc_return(page) == 2)
1391                                 break;
1392                         page_ref_dec(page);
1393                         kvm->arch.vsie.next++;
1394                         kvm->arch.vsie.next %= nr_vcpus;
1395                 }
1396                 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1397         }
1398         page->index = addr;
1399         /* double use of the same address */
1400         if (radix_tree_insert(&kvm->arch.vsie.addr_to_page, addr >> 9, page)) {
1401                 page_ref_dec(page);
1402                 mutex_unlock(&kvm->arch.vsie.mutex);
1403                 return NULL;
1404         }
1405         mutex_unlock(&kvm->arch.vsie.mutex);
1406
1407         vsie_page = page_to_virt(page);
1408         memset(&vsie_page->scb_s, 0, sizeof(struct kvm_s390_sie_block));
1409         release_gmap_shadow(vsie_page);
1410         vsie_page->fault_addr = 0;
1411         vsie_page->scb_s.ihcpu = 0xffffU;
1412         return vsie_page;
1413 }
1414
1415 /* put a vsie page acquired via get_vsie_page */
1416 static void put_vsie_page(struct kvm *kvm, struct vsie_page *vsie_page)
1417 {
1418         struct page *page = pfn_to_page(__pa(vsie_page) >> PAGE_SHIFT);
1419
1420         page_ref_dec(page);
1421 }
1422
1423 int kvm_s390_handle_vsie(struct kvm_vcpu *vcpu)
1424 {
1425         struct vsie_page *vsie_page;
1426         unsigned long scb_addr;
1427         int rc;
1428
1429         vcpu->stat.instruction_sie++;
1430         if (!test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIEF2))
1431                 return -EOPNOTSUPP;
1432         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1433                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1434
1435         BUILD_BUG_ON(sizeof(struct vsie_page) != PAGE_SIZE);
1436         scb_addr = kvm_s390_get_base_disp_s(vcpu, NULL);
1437
1438         /* 512 byte alignment */
1439         if (unlikely(scb_addr & 0x1ffUL))
1440                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1441
1442         if (signal_pending(current) || kvm_s390_vcpu_has_irq(vcpu, 0) ||
1443             kvm_s390_vcpu_sie_inhibited(vcpu)) {
1444                 kvm_s390_rewind_psw(vcpu, 4);
1445                 return 0;
1446         }
1447
1448         vsie_page = get_vsie_page(vcpu->kvm, scb_addr);
1449         if (IS_ERR(vsie_page))
1450                 return PTR_ERR(vsie_page);
1451         else if (!vsie_page)
1452                 /* double use of sie control block - simply do nothing */
1453                 return 0;
1454
1455         rc = pin_scb(vcpu, vsie_page, scb_addr);
1456         if (rc)
1457                 goto out_put;
1458         rc = shadow_scb(vcpu, vsie_page);
1459         if (rc)
1460                 goto out_unpin_scb;
1461         rc = pin_blocks(vcpu, vsie_page);
1462         if (rc)
1463                 goto out_unshadow;
1464         register_shadow_scb(vcpu, vsie_page);
1465         rc = vsie_run(vcpu, vsie_page);
1466         unregister_shadow_scb(vcpu);
1467         unpin_blocks(vcpu, vsie_page);
1468 out_unshadow:
1469         unshadow_scb(vcpu, vsie_page);
1470 out_unpin_scb:
1471         unpin_scb(vcpu, vsie_page, scb_addr);
1472 out_put:
1473         put_vsie_page(vcpu->kvm, vsie_page);
1474
1475         return rc < 0 ? rc : 0;
1476 }
1477
1478 /* Init the vsie data structures. To be called when a vm is initialized. */
1479 void kvm_s390_vsie_init(struct kvm *kvm)
1480 {
1481         mutex_init(&kvm->arch.vsie.mutex);
1482         INIT_RADIX_TREE(&kvm->arch.vsie.addr_to_page, GFP_KERNEL_ACCOUNT);
1483 }
1484
1485 /* Destroy the vsie data structures. To be called when a vm is destroyed. */
1486 void kvm_s390_vsie_destroy(struct kvm *kvm)
1487 {
1488         struct vsie_page *vsie_page;
1489         struct page *page;
1490         int i;
1491
1492         mutex_lock(&kvm->arch.vsie.mutex);
1493         for (i = 0; i < kvm->arch.vsie.page_count; i++) {
1494                 page = kvm->arch.vsie.pages[i];
1495                 kvm->arch.vsie.pages[i] = NULL;
1496                 vsie_page = page_to_virt(page);
1497                 release_gmap_shadow(vsie_page);
1498                 /* free the radix tree entry */
1499                 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1500                 __free_page(page);
1501         }
1502         kvm->arch.vsie.page_count = 0;
1503         mutex_unlock(&kvm->arch.vsie.mutex);
1504 }
1505
1506 void kvm_s390_vsie_kick(struct kvm_vcpu *vcpu)
1507 {
1508         struct kvm_s390_sie_block *scb = READ_ONCE(vcpu->arch.vsie_block);
1509
1510         /*
1511          * Even if the VCPU lets go of the shadow sie block reference, it is
1512          * still valid in the cache. So we can safely kick it.
1513          */
1514         if (scb) {
1515                 atomic_or(PROG_BLOCK_SIE, &scb->prog20);
1516                 if (scb->prog0c & PROG_IN_SIE)
1517                         atomic_or(CPUSTAT_STOP_INT, &scb->cpuflags);
1518         }
1519 }