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 <dahi@linux.vnet.ibm.com>
   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         /* branch prediction */
 514         if (test_kvm_facility(vcpu->kvm, 82))
 515                 scb_s->fpf |= scb_o->fpf & FPF_BPBC;
 516         /* SIMD */
 517         if (test_kvm_facility(vcpu->kvm, 129)) {
 518                 scb_s->eca |= scb_o->eca & ECA_VX;
 519                 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
 520         }
 521         /* Run-time-Instrumentation */
 522         if (test_kvm_facility(vcpu->kvm, 64))
 523                 scb_s->ecb3 |= scb_o->ecb3 & ECB3_RI;
 524         /* Instruction Execution Prevention */
 525         if (test_kvm_facility(vcpu->kvm, 130))
 526                 scb_s->ecb2 |= scb_o->ecb2 & ECB2_IEP;
 527         /* Guarded Storage */
 528         if (test_kvm_facility(vcpu->kvm, 133)) {
 529                 scb_s->ecb |= scb_o->ecb & ECB_GS;
 530                 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
 531         }
 532         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIIF))
 533                 scb_s->eca |= scb_o->eca & ECA_SII;
 534         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IB))
 535                 scb_s->eca |= scb_o->eca & ECA_IB;
 536         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_CEI))
 537                 scb_s->eca |= scb_o->eca & ECA_CEI;
 538         /* Epoch Extension */
 539         if (test_kvm_facility(vcpu->kvm, 139))
 540                 scb_s->ecd |= scb_o->ecd & ECD_MEF;
 541
 542         /* etoken */
 543         if (test_kvm_facility(vcpu->kvm, 156))
 544                 scb_s->ecd |= scb_o->ecd & ECD_ETOKENF;
 545
 546         scb_s->hpid = HPID_VSIE;
 547         scb_s->cpnc = scb_o->cpnc;
 548
 549         prepare_ibc(vcpu, vsie_page);
 550         rc = shadow_crycb(vcpu, vsie_page);
 551 out:
 552         if (rc)
 553                 unshadow_scb(vcpu, vsie_page);
 554         return rc;
 555 }
 556
 557 void kvm_s390_vsie_gmap_notifier(struct gmap *gmap, unsigned long start,
 558                                  unsigned long end)
 559 {
 560         struct kvm *kvm = gmap->private;
 561         struct vsie_page *cur;
 562         unsigned long prefix;
 563         struct page *page;
 564         int i;
 565
 566         if (!gmap_is_shadow(gmap))
 567                 return;
 568         if (start >= 1UL << 31)
 569                 /* We are only interested in prefix pages */
 570                 return;
 571
 572         /*
 573          * Only new shadow blocks are added to the list during runtime,
 574          * therefore we can safely reference them all the time.
 575          */
 576         for (i = 0; i < kvm->arch.vsie.page_count; i++) {
 577                 page = READ_ONCE(kvm->arch.vsie.pages[i]);
 578                 if (!page)
 579                         continue;
 580                 cur = page_to_virt(page);
 581                 if (READ_ONCE(cur->gmap) != gmap)
 582                         continue;
 583                 prefix = cur->scb_s.prefix << GUEST_PREFIX_SHIFT;
 584                 /* with mso/msl, the prefix lies at an offset */
 585                 prefix += cur->scb_s.mso;
 586                 if (prefix <= end && start <= prefix + 2 * PAGE_SIZE - 1)
 587                         prefix_unmapped_sync(cur);
 588         }
 589 }
 590
 591 /*
 592  * Map the first prefix page and if tx is enabled also the second prefix page.
 593  *
 594  * The prefix will be protected, a gmap notifier will inform about unmaps.
 595  * The shadow scb must not be executed until the prefix is remapped, this is
 596  * guaranteed by properly handling PROG_REQUEST.
 597  *
 598  * Returns: - 0 on if successfully mapped or already mapped
 599  *          - > 0 if control has to be given to guest 2
 600  *          - -EAGAIN if the caller can retry immediately
 601  *          - -ENOMEM if out of memory
 602  */
 603 static int map_prefix(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 604 {
 605         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 606         u64 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
 607         int rc;
 608
 609         if (prefix_is_mapped(vsie_page))
 610                 return 0;
 611
 612         /* mark it as mapped so we can catch any concurrent unmappers */
 613         prefix_mapped(vsie_page);
 614
 615         /* with mso/msl, the prefix lies at offset *mso* */
 616         prefix += scb_s->mso;
 617
 618         rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, prefix, NULL);
 619         if (!rc && (scb_s->ecb & ECB_TE))
 620                 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
 621                                            prefix + PAGE_SIZE, NULL);
 622         /*
 623          * We don't have to mprotect, we will be called for all unshadows.
 624          * SIE will detect if protection applies and trigger a validity.
 625          */
 626         if (rc)
 627                 prefix_unmapped(vsie_page);
 628         if (rc > 0 || rc == -EFAULT)
 629                 rc = set_validity_icpt(scb_s, 0x0037U);
 630         return rc;
 631 }
 632
 633 /*
 634  * Pin the guest page given by gpa and set hpa to the pinned host address.
 635  * Will always be pinned writable.
 636  *
 637  * Returns: - 0 on success
 638  *          - -EINVAL if the gpa is not valid guest storage
 639  */
 640 static int pin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t *hpa)
 641 {
 642         struct page *page;
 643
 644         page = gfn_to_page(kvm, gpa_to_gfn(gpa));
 645         if (is_error_page(page))
 646                 return -EINVAL;
 647         *hpa = (hpa_t) page_to_virt(page) + (gpa & ~PAGE_MASK);
 648         return 0;
 649 }
 650
 651 /* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */
 652 static void unpin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t hpa)
 653 {
 654         kvm_release_pfn_dirty(hpa >> PAGE_SHIFT);
 655         /* mark the page always as dirty for migration */
 656         mark_page_dirty(kvm, gpa_to_gfn(gpa));
 657 }
 658
 659 /* unpin all blocks previously pinned by pin_blocks(), marking them dirty */
 660 static void unpin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 661 {
 662         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 663         hpa_t hpa;
 664
 665         hpa = (u64) scb_s->scaoh << 32 | scb_s->scaol;
 666         if (hpa) {
 667                 unpin_guest_page(vcpu->kvm, vsie_page->sca_gpa, hpa);
 668                 vsie_page->sca_gpa = 0;
 669                 scb_s->scaol = 0;
 670                 scb_s->scaoh = 0;
 671         }
 672
 673         hpa = scb_s->itdba;
 674         if (hpa) {
 675                 unpin_guest_page(vcpu->kvm, vsie_page->itdba_gpa, hpa);
 676                 vsie_page->itdba_gpa = 0;
 677                 scb_s->itdba = 0;
 678         }
 679
 680         hpa = scb_s->gvrd;
 681         if (hpa) {
 682                 unpin_guest_page(vcpu->kvm, vsie_page->gvrd_gpa, hpa);
 683                 vsie_page->gvrd_gpa = 0;
 684                 scb_s->gvrd = 0;
 685         }
 686
 687         hpa = scb_s->riccbd;
 688         if (hpa) {
 689                 unpin_guest_page(vcpu->kvm, vsie_page->riccbd_gpa, hpa);
 690                 vsie_page->riccbd_gpa = 0;
 691                 scb_s->riccbd = 0;
 692         }
 693
 694         hpa = scb_s->sdnxo;
 695         if (hpa) {
 696                 unpin_guest_page(vcpu->kvm, vsie_page->sdnx_gpa, hpa);
 697                 vsie_page->sdnx_gpa = 0;
 698                 scb_s->sdnxo = 0;
 699         }
 700 }
 701
 702 /*
 703  * Instead of shadowing some blocks, we can simply forward them because the
 704  * addresses in the scb are 64 bit long.
 705  *
 706  * This works as long as the data lies in one page. If blocks ever exceed one
 707  * page, we have to fall back to shadowing.
 708  *
 709  * As we reuse the sca, the vcpu pointers contained in it are invalid. We must
 710  * therefore not enable any facilities that access these pointers (e.g. SIGPIF).
 711  *
 712  * Returns: - 0 if all blocks were pinned.
 713  *          - > 0 if control has to be given to guest 2
 714  *          - -ENOMEM if out of memory
 715  */
 716 static int pin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 717 {
 718         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 719         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 720         hpa_t hpa;
 721         gpa_t gpa;
 722         int rc = 0;
 723
 724         gpa = READ_ONCE(scb_o->scaol) & ~0xfUL;
 725         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO))
 726                 gpa |= (u64) READ_ONCE(scb_o->scaoh) << 32;
 727         if (gpa) {
 728                 if (gpa < 2 * PAGE_SIZE)
 729                         rc = set_validity_icpt(scb_s, 0x0038U);
 730                 else if ((gpa & ~0x1fffUL) == kvm_s390_get_prefix(vcpu))
 731                         rc = set_validity_icpt(scb_s, 0x0011U);
 732                 else if ((gpa & PAGE_MASK) !=
 733                          ((gpa + sizeof(struct bsca_block) - 1) & PAGE_MASK))
 734                         rc = set_validity_icpt(scb_s, 0x003bU);
 735                 if (!rc) {
 736                         rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 737                         if (rc)
 738                                 rc = set_validity_icpt(scb_s, 0x0034U);
 739                 }
 740                 if (rc)
 741                         goto unpin;
 742                 vsie_page->sca_gpa = gpa;
 743                 scb_s->scaoh = (u32)((u64)hpa >> 32);
 744                 scb_s->scaol = (u32)(u64)hpa;
 745         }
 746
 747         gpa = READ_ONCE(scb_o->itdba) & ~0xffUL;
 748         if (gpa && (scb_s->ecb & ECB_TE)) {
 749                 if (gpa < 2 * PAGE_SIZE) {
 750                         rc = set_validity_icpt(scb_s, 0x0080U);
 751                         goto unpin;
 752                 }
 753                 /* 256 bytes cannot cross page boundaries */
 754                 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 755                 if (rc) {
 756                         rc = set_validity_icpt(scb_s, 0x0080U);
 757                         goto unpin;
 758                 }
 759                 vsie_page->itdba_gpa = gpa;
 760                 scb_s->itdba = hpa;
 761         }
 762
 763         gpa = READ_ONCE(scb_o->gvrd) & ~0x1ffUL;
 764         if (gpa && (scb_s->eca & ECA_VX) && !(scb_s->ecd & ECD_HOSTREGMGMT)) {
 765                 if (gpa < 2 * PAGE_SIZE) {
 766                         rc = set_validity_icpt(scb_s, 0x1310U);
 767                         goto unpin;
 768                 }
 769                 /*
 770                  * 512 bytes vector registers cannot cross page boundaries
 771                  * if this block gets bigger, we have to shadow it.
 772                  */
 773                 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 774                 if (rc) {
 775                         rc = set_validity_icpt(scb_s, 0x1310U);
 776                         goto unpin;
 777                 }
 778                 vsie_page->gvrd_gpa = gpa;
 779                 scb_s->gvrd = hpa;
 780         }
 781
 782         gpa = READ_ONCE(scb_o->riccbd) & ~0x3fUL;
 783         if (gpa && (scb_s->ecb3 & ECB3_RI)) {
 784                 if (gpa < 2 * PAGE_SIZE) {
 785                         rc = set_validity_icpt(scb_s, 0x0043U);
 786                         goto unpin;
 787                 }
 788                 /* 64 bytes cannot cross page boundaries */
 789                 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 790                 if (rc) {
 791                         rc = set_validity_icpt(scb_s, 0x0043U);
 792                         goto unpin;
 793                 }
 794                 /* Validity 0x0044 will be checked by SIE */
 795                 vsie_page->riccbd_gpa = gpa;
 796                 scb_s->riccbd = hpa;
 797         }
 798         if (((scb_s->ecb & ECB_GS) && !(scb_s->ecd & ECD_HOSTREGMGMT)) ||
 799             (scb_s->ecd & ECD_ETOKENF)) {
 800                 unsigned long sdnxc;
 801
 802                 gpa = READ_ONCE(scb_o->sdnxo) & ~0xfUL;
 803                 sdnxc = READ_ONCE(scb_o->sdnxo) & 0xfUL;
 804                 if (!gpa || gpa < 2 * PAGE_SIZE) {
 805                         rc = set_validity_icpt(scb_s, 0x10b0U);
 806                         goto unpin;
 807                 }
 808                 if (sdnxc < 6 || sdnxc > 12) {
 809                         rc = set_validity_icpt(scb_s, 0x10b1U);
 810                         goto unpin;
 811                 }
 812                 if (gpa & ((1 << sdnxc) - 1)) {
 813                         rc = set_validity_icpt(scb_s, 0x10b2U);
 814                         goto unpin;
 815                 }
 816                 /* Due to alignment rules (checked above) this cannot
 817                  * cross page boundaries
 818                  */
 819                 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 820                 if (rc) {
 821                         rc = set_validity_icpt(scb_s, 0x10b0U);
 822                         goto unpin;
 823                 }
 824                 vsie_page->sdnx_gpa = gpa;
 825                 scb_s->sdnxo = hpa | sdnxc;
 826         }
 827         return 0;
 828 unpin:
 829         unpin_blocks(vcpu, vsie_page);
 830         return rc;
 831 }
 832
 833 /* unpin the scb provided by guest 2, marking it as dirty */
 834 static void unpin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
 835                       gpa_t gpa)
 836 {
 837         hpa_t hpa = (hpa_t) vsie_page->scb_o;
 838
 839         if (hpa)
 840                 unpin_guest_page(vcpu->kvm, gpa, hpa);
 841         vsie_page->scb_o = NULL;
 842 }
 843
 844 /*
 845  * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o.
 846  *
 847  * Returns: - 0 if the scb was pinned.
 848  *          - > 0 if control has to be given to guest 2
 849  */
 850 static int pin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
 851                    gpa_t gpa)
 852 {
 853         hpa_t hpa;
 854         int rc;
 855
 856         rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 857         if (rc) {
 858                 rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 859                 WARN_ON_ONCE(rc);
 860                 return 1;
 861         }
 862         vsie_page->scb_o = (struct kvm_s390_sie_block *) hpa;
 863         return 0;
 864 }
 865
 866 /*
 867  * Inject a fault into guest 2.
 868  *
 869  * Returns: - > 0 if control has to be given to guest 2
 870  *            < 0 if an error occurred during injection.
 871  */
 872 static int inject_fault(struct kvm_vcpu *vcpu, __u16 code, __u64 vaddr,
 873                         bool write_flag)
 874 {
 875         struct kvm_s390_pgm_info pgm = {
 876                 .code = code,
 877                 .trans_exc_code =
 878                         /* 0-51: virtual address */
 879                         (vaddr & 0xfffffffffffff000UL) |
 880                         /* 52-53: store / fetch */
 881                         (((unsigned int) !write_flag) + 1) << 10,
 882                         /* 62-63: asce id (alway primary == 0) */
 883                 .exc_access_id = 0, /* always primary */
 884                 .op_access_id = 0, /* not MVPG */
 885         };
 886         int rc;
 887
 888         if (code == PGM_PROTECTION)
 889                 pgm.trans_exc_code |= 0x4UL;
 890
 891         rc = kvm_s390_inject_prog_irq(vcpu, &pgm);
 892         return rc ? rc : 1;
 893 }
 894
 895 /*
 896  * Handle a fault during vsie execution on a gmap shadow.
 897  *
 898  * Returns: - 0 if the fault was resolved
 899  *          - > 0 if control has to be given to guest 2
 900  *          - < 0 if an error occurred
 901  */
 902 static int handle_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 903 {
 904         int rc;
 905
 906         if (current->thread.gmap_int_code == PGM_PROTECTION)
 907                 /* we can directly forward all protection exceptions */
 908                 return inject_fault(vcpu, PGM_PROTECTION,
 909                                     current->thread.gmap_addr, 1);
 910
 911         rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
 912                                    current->thread.gmap_addr, NULL);
 913         if (rc > 0) {
 914                 rc = inject_fault(vcpu, rc,
 915                                   current->thread.gmap_addr,
 916                                   current->thread.gmap_write_flag);
 917                 if (rc >= 0)
 918                         vsie_page->fault_addr = current->thread.gmap_addr;
 919         }
 920         return rc;
 921 }
 922
 923 /*
 924  * Retry the previous fault that required guest 2 intervention. This avoids
 925  * one superfluous SIE re-entry and direct exit.
 926  *
 927  * Will ignore any errors. The next SIE fault will do proper fault handling.
 928  */
 929 static void handle_last_fault(struct kvm_vcpu *vcpu,
 930                               struct vsie_page *vsie_page)
 931 {
 932         if (vsie_page->fault_addr)
 933                 kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
 934                                       vsie_page->fault_addr, NULL);
 935         vsie_page->fault_addr = 0;
 936 }
 937
 938 static inline void clear_vsie_icpt(struct vsie_page *vsie_page)
 939 {
 940         vsie_page->scb_s.icptcode = 0;
 941 }
 942
 943 /* rewind the psw and clear the vsie icpt, so we can retry execution */
 944 static void retry_vsie_icpt(struct vsie_page *vsie_page)
 945 {
 946         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 947         int ilen = insn_length(scb_s->ipa >> 8);
 948
 949         /* take care of EXECUTE instructions */
 950         if (scb_s->icptstatus & 1) {
 951                 ilen = (scb_s->icptstatus >> 4) & 0x6;
 952                 if (!ilen)
 953                         ilen = 4;
 954         }
 955         scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, ilen);
 956         clear_vsie_icpt(vsie_page);
 957 }
 958
 959 /*
 960  * Try to shadow + enable the guest 2 provided facility list.
 961  * Retry instruction execution if enabled for and provided by guest 2.
 962  *
 963  * Returns: - 0 if handled (retry or guest 2 icpt)
 964  *          - > 0 if control has to be given to guest 2
 965  */
 966 static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 967 {
 968         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 969         __u32 fac = READ_ONCE(vsie_page->scb_o->fac) & 0x7ffffff8U;
 970
 971         if (fac && test_kvm_facility(vcpu->kvm, 7)) {
 972                 retry_vsie_icpt(vsie_page);
 973                 if (read_guest_real(vcpu, fac, &vsie_page->fac,
 974                                     sizeof(vsie_page->fac)))
 975                         return set_validity_icpt(scb_s, 0x1090U);
 976                 scb_s->fac = (__u32)(__u64) &vsie_page->fac;
 977         }
 978         return 0;
 979 }
 980
 981 /*
 982  * Get a register for a nested guest.
 983  * @vcpu the vcpu of the guest
 984  * @vsie_page the vsie_page for the nested guest
 985  * @reg the register number, the upper 4 bits are ignored.
 986  * returns: the value of the register.
 987  */
 988 static u64 vsie_get_register(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, u8 reg)
 989 {
 990         /* no need to validate the parameter and/or perform error handling */
 991         reg &= 0xf;
 992         switch (reg) {
 993         case 15:
 994                 return vsie_page->scb_s.gg15;
 995         case 14:
 996                 return vsie_page->scb_s.gg14;
 997         default:
 998                 return vcpu->run->s.regs.gprs[reg];
 999         }
1000 }
1001
1002 static int vsie_handle_mvpg(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1003 {
1004         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1005         unsigned long pei_dest, pei_src, src, dest, mask, prefix;
1006         u64 *pei_block = &vsie_page->scb_o->mcic;
1007         int edat, rc_dest, rc_src;
1008         union ctlreg0 cr0;
1009
1010         cr0.val = vcpu->arch.sie_block->gcr[0];
1011         edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
1012         mask = _kvm_s390_logical_to_effective(&scb_s->gpsw, PAGE_MASK);
1013         prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
1014
1015         dest = vsie_get_register(vcpu, vsie_page, scb_s->ipb >> 20) & mask;
1016         dest = _kvm_s390_real_to_abs(prefix, dest) + scb_s->mso;
1017         src = vsie_get_register(vcpu, vsie_page, scb_s->ipb >> 16) & mask;
1018         src = _kvm_s390_real_to_abs(prefix, src) + scb_s->mso;
1019
1020         rc_dest = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, dest, &pei_dest);
1021         rc_src = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, src, &pei_src);
1022         /*
1023          * Either everything went well, or something non-critical went wrong
1024          * e.g. because of a race. In either case, simply retry.
1025          */
1026         if (rc_dest == -EAGAIN || rc_src == -EAGAIN || (!rc_dest && !rc_src)) {
1027                 retry_vsie_icpt(vsie_page);
1028                 return -EAGAIN;
1029         }
1030         /* Something more serious went wrong, propagate the error */
1031         if (rc_dest < 0)
1032                 return rc_dest;
1033         if (rc_src < 0)
1034                 return rc_src;
1035
1036         /* The only possible suppressing exception: just deliver it */
1037         if (rc_dest == PGM_TRANSLATION_SPEC || rc_src == PGM_TRANSLATION_SPEC) {
1038                 clear_vsie_icpt(vsie_page);
1039                 rc_dest = kvm_s390_inject_program_int(vcpu, PGM_TRANSLATION_SPEC);
1040                 WARN_ON_ONCE(rc_dest);
1041                 return 1;
1042         }
1043
1044         /*
1045          * Forward the PEI intercept to the guest if it was a page fault, or
1046          * also for segment and region table faults if EDAT applies.
1047          */
1048         if (edat) {
1049                 rc_dest = rc_dest == PGM_ASCE_TYPE ? rc_dest : 0;
1050                 rc_src = rc_src == PGM_ASCE_TYPE ? rc_src : 0;
1051         } else {
1052                 rc_dest = rc_dest != PGM_PAGE_TRANSLATION ? rc_dest : 0;
1053                 rc_src = rc_src != PGM_PAGE_TRANSLATION ? rc_src : 0;
1054         }
1055         if (!rc_dest && !rc_src) {
1056                 pei_block[0] = pei_dest;
1057                 pei_block[1] = pei_src;
1058                 return 1;
1059         }
1060
1061         retry_vsie_icpt(vsie_page);
1062
1063         /*
1064          * The host has edat, and the guest does not, or it was an ASCE type
1065          * exception. The host needs to inject the appropriate DAT interrupts
1066          * into the guest.
1067          */
1068         if (rc_dest)
1069                 return inject_fault(vcpu, rc_dest, dest, 1);
1070         return inject_fault(vcpu, rc_src, src, 0);
1071 }
1072
1073 /*
1074  * Run the vsie on a shadow scb and a shadow gmap, without any further
1075  * sanity checks, handling SIE faults.
1076  *
1077  * Returns: - 0 everything went fine
1078  *          - > 0 if control has to be given to guest 2
1079  *          - < 0 if an error occurred
1080  */
1081 static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1082         __releases(vcpu->kvm->srcu)
1083         __acquires(vcpu->kvm->srcu)
1084 {
1085         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1086         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
1087         int guest_bp_isolation;
1088         int rc = 0;
1089
1090         handle_last_fault(vcpu, vsie_page);
1091
1092         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
1093
1094         /* save current guest state of bp isolation override */
1095         guest_bp_isolation = test_thread_flag(TIF_ISOLATE_BP_GUEST);
1096
1097         /*
1098          * The guest is running with BPBC, so we have to force it on for our
1099          * nested guest. This is done by enabling BPBC globally, so the BPBC
1100          * control in the SCB (which the nested guest can modify) is simply
1101          * ignored.
1102          */
1103         if (test_kvm_facility(vcpu->kvm, 82) &&
1104             vcpu->arch.sie_block->fpf & FPF_BPBC)
1105                 set_thread_flag(TIF_ISOLATE_BP_GUEST);
1106
1107         local_irq_disable();
1108         guest_enter_irqoff();
1109         local_irq_enable();
1110
1111         /*
1112          * Simulate a SIE entry of the VCPU (see sie64a), so VCPU blocking
1113          * and VCPU requests also hinder the vSIE from running and lead
1114          * to an immediate exit. kvm_s390_vsie_kick() has to be used to
1115          * also kick the vSIE.
1116          */
1117         vcpu->arch.sie_block->prog0c |= PROG_IN_SIE;
1118         barrier();
1119         if (test_cpu_flag(CIF_FPU))
1120                 load_fpu_regs();
1121         if (!kvm_s390_vcpu_sie_inhibited(vcpu))
1122                 rc = sie64a(scb_s, vcpu->run->s.regs.gprs);
1123         barrier();
1124         vcpu->arch.sie_block->prog0c &= ~PROG_IN_SIE;
1125
1126         local_irq_disable();
1127         guest_exit_irqoff();
1128         local_irq_enable();
1129
1130         /* restore guest state for bp isolation override */
1131         if (!guest_bp_isolation)
1132                 clear_thread_flag(TIF_ISOLATE_BP_GUEST);
1133
1134         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1135
1136         if (rc == -EINTR) {
1137                 VCPU_EVENT(vcpu, 3, "%s", "machine check");
1138                 kvm_s390_reinject_machine_check(vcpu, &vsie_page->mcck_info);
1139                 return 0;
1140         }
1141
1142         if (rc > 0)
1143                 rc = 0; /* we could still have an icpt */
1144         else if (rc == -EFAULT)
1145                 return handle_fault(vcpu, vsie_page);
1146
1147         switch (scb_s->icptcode) {
1148         case ICPT_INST:
1149                 if (scb_s->ipa == 0xb2b0)
1150                         rc = handle_stfle(vcpu, vsie_page);
1151                 break;
1152         case ICPT_STOP:
1153                 /* stop not requested by g2 - must have been a kick */
1154                 if (!(atomic_read(&scb_o->cpuflags) & CPUSTAT_STOP_INT))
1155                         clear_vsie_icpt(vsie_page);
1156                 break;
1157         case ICPT_VALIDITY:
1158                 if ((scb_s->ipa & 0xf000) != 0xf000)
1159                         scb_s->ipa += 0x1000;
1160                 break;
1161         case ICPT_PARTEXEC:
1162                 if (scb_s->ipa == 0xb254)
1163                         rc = vsie_handle_mvpg(vcpu, vsie_page);
1164                 break;
1165         }
1166         return rc;
1167 }
1168
1169 static void release_gmap_shadow(struct vsie_page *vsie_page)
1170 {
1171         if (vsie_page->gmap)
1172                 gmap_put(vsie_page->gmap);
1173         WRITE_ONCE(vsie_page->gmap, NULL);
1174         prefix_unmapped(vsie_page);
1175 }
1176
1177 static int acquire_gmap_shadow(struct kvm_vcpu *vcpu,
1178                                struct vsie_page *vsie_page)
1179 {
1180         unsigned long asce;
1181         union ctlreg0 cr0;
1182         struct gmap *gmap;
1183         int edat;
1184
1185         asce = vcpu->arch.sie_block->gcr[1];
1186         cr0.val = vcpu->arch.sie_block->gcr[0];
1187         edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
1188         edat += edat && test_kvm_facility(vcpu->kvm, 78);
1189
1190         /*
1191          * ASCE or EDAT could have changed since last icpt, or the gmap
1192          * we're holding has been unshadowed. If the gmap is still valid,
1193          * we can safely reuse it.
1194          */
1195         if (vsie_page->gmap && gmap_shadow_valid(vsie_page->gmap, asce, edat))
1196                 return 0;
1197
1198         /* release the old shadow - if any, and mark the prefix as unmapped */
1199         release_gmap_shadow(vsie_page);
1200         gmap = gmap_shadow(vcpu->arch.gmap, asce, edat);
1201         if (IS_ERR(gmap))
1202                 return PTR_ERR(gmap);
1203         gmap->private = vcpu->kvm;
1204         WRITE_ONCE(vsie_page->gmap, gmap);
1205         return 0;
1206 }
1207
1208 /*
1209  * Register the shadow scb at the VCPU, e.g. for kicking out of vsie.
1210  */
1211 static void register_shadow_scb(struct kvm_vcpu *vcpu,
1212                                 struct vsie_page *vsie_page)
1213 {
1214         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1215
1216         WRITE_ONCE(vcpu->arch.vsie_block, &vsie_page->scb_s);
1217         /*
1218          * External calls have to lead to a kick of the vcpu and
1219          * therefore the vsie -> Simulate Wait state.
1220          */
1221         kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
1222         /*
1223          * We have to adjust the g3 epoch by the g2 epoch. The epoch will
1224          * automatically be adjusted on tod clock changes via kvm_sync_clock.
1225          */
1226         preempt_disable();
1227         scb_s->epoch += vcpu->kvm->arch.epoch;
1228
1229         if (scb_s->ecd & ECD_MEF) {
1230                 scb_s->epdx += vcpu->kvm->arch.epdx;
1231                 if (scb_s->epoch < vcpu->kvm->arch.epoch)
1232                         scb_s->epdx += 1;
1233         }
1234
1235         preempt_enable();
1236 }
1237
1238 /*
1239  * Unregister a shadow scb from a VCPU.
1240  */
1241 static void unregister_shadow_scb(struct kvm_vcpu *vcpu)
1242 {
1243         kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
1244         WRITE_ONCE(vcpu->arch.vsie_block, NULL);
1245 }
1246
1247 /*
1248  * Run the vsie on a shadowed scb, managing the gmap shadow, handling
1249  * prefix pages and faults.
1250  *
1251  * Returns: - 0 if no errors occurred
1252  *          - > 0 if control has to be given to guest 2
1253  *          - -ENOMEM if out of memory
1254  */
1255 static int vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1256 {
1257         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1258         int rc = 0;
1259
1260         while (1) {
1261                 rc = acquire_gmap_shadow(vcpu, vsie_page);
1262                 if (!rc)
1263                         rc = map_prefix(vcpu, vsie_page);
1264                 if (!rc) {
1265                         gmap_enable(vsie_page->gmap);
1266                         update_intervention_requests(vsie_page);
1267                         rc = do_vsie_run(vcpu, vsie_page);
1268                         gmap_enable(vcpu->arch.gmap);
1269                 }
1270                 atomic_andnot(PROG_BLOCK_SIE, &scb_s->prog20);
1271
1272                 if (rc == -EAGAIN)
1273                         rc = 0;
1274                 if (rc || scb_s->icptcode || signal_pending(current) ||
1275                     kvm_s390_vcpu_has_irq(vcpu, 0) ||
1276                     kvm_s390_vcpu_sie_inhibited(vcpu))
1277                         break;
1278                 cond_resched();
1279         }
1280
1281         if (rc == -EFAULT) {
1282                 /*
1283                  * Addressing exceptions are always presentes as intercepts.
1284                  * As addressing exceptions are suppressing and our guest 3 PSW
1285                  * points at the responsible instruction, we have to
1286                  * forward the PSW and set the ilc. If we can't read guest 3
1287                  * instruction, we can use an arbitrary ilc. Let's always use
1288                  * ilen = 4 for now, so we can avoid reading in guest 3 virtual
1289                  * memory. (we could also fake the shadow so the hardware
1290                  * handles it).
1291                  */
1292                 scb_s->icptcode = ICPT_PROGI;
1293                 scb_s->iprcc = PGM_ADDRESSING;
1294                 scb_s->pgmilc = 4;
1295                 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, 4);
1296                 rc = 1;
1297         }
1298         return rc;
1299 }
1300
1301 /*
1302  * Get or create a vsie page for a scb address.
1303  *
1304  * Returns: - address of a vsie page (cached or new one)
1305  *          - NULL if the same scb address is already used by another VCPU
1306  *          - ERR_PTR(-ENOMEM) if out of memory
1307  */
1308 static struct vsie_page *get_vsie_page(struct kvm *kvm, unsigned long addr)
1309 {
1310         struct vsie_page *vsie_page;
1311         struct page *page;
1312         int nr_vcpus;
1313
1314         rcu_read_lock();
1315         page = radix_tree_lookup(&kvm->arch.vsie.addr_to_page, addr >> 9);
1316         rcu_read_unlock();
1317         if (page) {
1318                 if (page_ref_inc_return(page) == 2)
1319                         return page_to_virt(page);
1320                 page_ref_dec(page);
1321         }
1322
1323         /*
1324          * We want at least #online_vcpus shadows, so every VCPU can execute
1325          * the VSIE in parallel.
1326          */
1327         nr_vcpus = atomic_read(&kvm->online_vcpus);
1328
1329         mutex_lock(&kvm->arch.vsie.mutex);
1330         if (kvm->arch.vsie.page_count < nr_vcpus) {
1331                 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO | GFP_DMA);
1332                 if (!page) {
1333                         mutex_unlock(&kvm->arch.vsie.mutex);
1334                         return ERR_PTR(-ENOMEM);
1335                 }
1336                 page_ref_inc(page);
1337                 kvm->arch.vsie.pages[kvm->arch.vsie.page_count] = page;
1338                 kvm->arch.vsie.page_count++;
1339         } else {
1340                 /* reuse an existing entry that belongs to nobody */
1341                 while (true) {
1342                         page = kvm->arch.vsie.pages[kvm->arch.vsie.next];
1343                         if (page_ref_inc_return(page) == 2)
1344                                 break;
1345                         page_ref_dec(page);
1346                         kvm->arch.vsie.next++;
1347                         kvm->arch.vsie.next %= nr_vcpus;
1348                 }
1349                 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1350         }
1351         page->index = addr;
1352         /* double use of the same address */
1353         if (radix_tree_insert(&kvm->arch.vsie.addr_to_page, addr >> 9, page)) {
1354                 page_ref_dec(page);
1355                 mutex_unlock(&kvm->arch.vsie.mutex);
1356                 return NULL;
1357         }
1358         mutex_unlock(&kvm->arch.vsie.mutex);
1359
1360         vsie_page = page_to_virt(page);
1361         memset(&vsie_page->scb_s, 0, sizeof(struct kvm_s390_sie_block));
1362         release_gmap_shadow(vsie_page);
1363         vsie_page->fault_addr = 0;
1364         vsie_page->scb_s.ihcpu = 0xffffU;
1365         return vsie_page;
1366 }
1367
1368 /* put a vsie page acquired via get_vsie_page */
1369 static void put_vsie_page(struct kvm *kvm, struct vsie_page *vsie_page)
1370 {
1371         struct page *page = pfn_to_page(__pa(vsie_page) >> PAGE_SHIFT);
1372
1373         page_ref_dec(page);
1374 }
1375
1376 int kvm_s390_handle_vsie(struct kvm_vcpu *vcpu)
1377 {
1378         struct vsie_page *vsie_page;
1379         unsigned long scb_addr;
1380         int rc;
1381
1382         vcpu->stat.instruction_sie++;
1383         if (!test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIEF2))
1384                 return -EOPNOTSUPP;
1385         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1386                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1387
1388         BUILD_BUG_ON(sizeof(struct vsie_page) != PAGE_SIZE);
1389         scb_addr = kvm_s390_get_base_disp_s(vcpu, NULL);
1390
1391         /* 512 byte alignment */
1392         if (unlikely(scb_addr & 0x1ffUL))
1393                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1394
1395         if (signal_pending(current) || kvm_s390_vcpu_has_irq(vcpu, 0) ||
1396             kvm_s390_vcpu_sie_inhibited(vcpu))
1397                 return 0;
1398
1399         vsie_page = get_vsie_page(vcpu->kvm, scb_addr);
1400         if (IS_ERR(vsie_page))
1401                 return PTR_ERR(vsie_page);
1402         else if (!vsie_page)
1403                 /* double use of sie control block - simply do nothing */
1404                 return 0;
1405
1406         rc = pin_scb(vcpu, vsie_page, scb_addr);
1407         if (rc)
1408                 goto out_put;
1409         rc = shadow_scb(vcpu, vsie_page);
1410         if (rc)
1411                 goto out_unpin_scb;
1412         rc = pin_blocks(vcpu, vsie_page);
1413         if (rc)
1414                 goto out_unshadow;
1415         register_shadow_scb(vcpu, vsie_page);
1416         rc = vsie_run(vcpu, vsie_page);
1417         unregister_shadow_scb(vcpu);
1418         unpin_blocks(vcpu, vsie_page);
1419 out_unshadow:
1420         unshadow_scb(vcpu, vsie_page);
1421 out_unpin_scb:
1422         unpin_scb(vcpu, vsie_page, scb_addr);
1423 out_put:
1424         put_vsie_page(vcpu->kvm, vsie_page);
1425
1426         return rc < 0 ? rc : 0;
1427 }
1428
1429 /* Init the vsie data structures. To be called when a vm is initialized. */
1430 void kvm_s390_vsie_init(struct kvm *kvm)
1431 {
1432         mutex_init(&kvm->arch.vsie.mutex);
1433         INIT_RADIX_TREE(&kvm->arch.vsie.addr_to_page, GFP_KERNEL_ACCOUNT);
1434 }
1435
1436 /* Destroy the vsie data structures. To be called when a vm is destroyed. */
1437 void kvm_s390_vsie_destroy(struct kvm *kvm)
1438 {
1439         struct vsie_page *vsie_page;
1440         struct page *page;
1441         int i;
1442
1443         mutex_lock(&kvm->arch.vsie.mutex);
1444         for (i = 0; i < kvm->arch.vsie.page_count; i++) {
1445                 page = kvm->arch.vsie.pages[i];
1446                 kvm->arch.vsie.pages[i] = NULL;
1447                 vsie_page = page_to_virt(page);
1448                 release_gmap_shadow(vsie_page);
1449                 /* free the radix tree entry */
1450                 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1451                 __free_page(page);
1452         }
1453         kvm->arch.vsie.page_count = 0;
1454         mutex_unlock(&kvm->arch.vsie.mutex);
1455 }
1456
1457 void kvm_s390_vsie_kick(struct kvm_vcpu *vcpu)
1458 {
1459         struct kvm_s390_sie_block *scb = READ_ONCE(vcpu->arch.vsie_block);
1460
1461         /*
1462          * Even if the VCPU lets go of the shadow sie block reference, it is
1463          * still valid in the cache. So we can safely kick it.
1464          */
1465         if (scb) {
1466                 atomic_or(PROG_BLOCK_SIE, &scb->prog20);
1467                 if (scb->prog0c & PROG_IN_SIE)
1468                         atomic_or(CPUSTAT_STOP_INT, &scb->cpuflags);
1469         }
1470 }