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