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