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 "kvm-s390.h"
  22 #include "gaccess.h"
  23
  24 struct vsie_page {
  25         struct kvm_s390_sie_block scb_s;        /* 0x0000 */
  26         /*
  27          * the backup info for machine check. ensure it's at
  28          * the same offset as that in struct sie_page!
  29          */
  30         struct mcck_volatile_info mcck_info;    /* 0x0200 */
  31         /*
  32          * The pinned original scb. Be aware that other VCPUs can modify
  33          * it while we read from it. Values that are used for conditions or
  34          * are reused conditionally, should be accessed via READ_ONCE.
  35          */
  36         struct kvm_s390_sie_block *scb_o;       /* 0x0218 */
  37         /* the shadow gmap in use by the vsie_page */
  38         struct gmap *gmap;                      /* 0x0220 */
  39         /* address of the last reported fault to guest2 */
  40         unsigned long fault_addr;               /* 0x0228 */
  41         /* calculated guest addresses of satellite control blocks */
  42         gpa_t sca_gpa;                          /* 0x0230 */
  43         gpa_t itdba_gpa;                        /* 0x0238 */
  44         gpa_t gvrd_gpa;                         /* 0x0240 */
  45         gpa_t riccbd_gpa;                       /* 0x0248 */
  46         gpa_t sdnx_gpa;                         /* 0x0250 */
  47         __u8 reserved[0x0700 - 0x0258];         /* 0x0258 */
  48         struct kvm_s390_crypto_cb crycb;        /* 0x0700 */
  49         __u8 fac[S390_ARCH_FAC_LIST_SIZE_BYTE]; /* 0x0800 */
  50 };
  51
  52 /* trigger a validity icpt for the given scb */
  53 static int set_validity_icpt(struct kvm_s390_sie_block *scb,
  54                              __u16 reason_code)
  55 {
  56         scb->ipa = 0x1000;
  57         scb->ipb = ((__u32) reason_code) << 16;
  58         scb->icptcode = ICPT_VALIDITY;
  59         return 1;
  60 }
  61
  62 /* mark the prefix as unmapped, this will block the VSIE */
  63 static void prefix_unmapped(struct vsie_page *vsie_page)
  64 {
  65         atomic_or(PROG_REQUEST, &vsie_page->scb_s.prog20);
  66 }
  67
  68 /* mark the prefix as unmapped and wait until the VSIE has been left */
  69 static void prefix_unmapped_sync(struct vsie_page *vsie_page)
  70 {
  71         prefix_unmapped(vsie_page);
  72         if (vsie_page->scb_s.prog0c & PROG_IN_SIE)
  73                 atomic_or(CPUSTAT_STOP_INT, &vsie_page->scb_s.cpuflags);
  74         while (vsie_page->scb_s.prog0c & PROG_IN_SIE)
  75                 cpu_relax();
  76 }
  77
  78 /* mark the prefix as mapped, this will allow the VSIE to run */
  79 static void prefix_mapped(struct vsie_page *vsie_page)
  80 {
  81         atomic_andnot(PROG_REQUEST, &vsie_page->scb_s.prog20);
  82 }
  83
  84 /* test if the prefix is mapped into the gmap shadow */
  85 static int prefix_is_mapped(struct vsie_page *vsie_page)
  86 {
  87         return !(atomic_read(&vsie_page->scb_s.prog20) & PROG_REQUEST);
  88 }
  89
  90 /* copy the updated intervention request bits into the shadow scb */
  91 static void update_intervention_requests(struct vsie_page *vsie_page)
  92 {
  93         const int bits = CPUSTAT_STOP_INT | CPUSTAT_IO_INT | CPUSTAT_EXT_INT;
  94         int cpuflags;
  95
  96         cpuflags = atomic_read(&vsie_page->scb_o->cpuflags);
  97         atomic_andnot(bits, &vsie_page->scb_s.cpuflags);
  98         atomic_or(cpuflags & bits, &vsie_page->scb_s.cpuflags);
  99 }
 100
 101 /* shadow (filter and validate) the cpuflags  */
 102 static int prepare_cpuflags(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 103 {
 104         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 105         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 106         int newflags, cpuflags = atomic_read(&scb_o->cpuflags);
 107
 108         /* we don't allow ESA/390 guests */
 109         if (!(cpuflags & CPUSTAT_ZARCH))
 110                 return set_validity_icpt(scb_s, 0x0001U);
 111
 112         if (cpuflags & (CPUSTAT_RRF | CPUSTAT_MCDS))
 113                 return set_validity_icpt(scb_s, 0x0001U);
 114         else if (cpuflags & (CPUSTAT_SLSV | CPUSTAT_SLSR))
 115                 return set_validity_icpt(scb_s, 0x0007U);
 116
 117         /* intervention requests will be set later */
 118         newflags = CPUSTAT_ZARCH;
 119         if (cpuflags & CPUSTAT_GED && test_kvm_facility(vcpu->kvm, 8))
 120                 newflags |= CPUSTAT_GED;
 121         if (cpuflags & CPUSTAT_GED2 && test_kvm_facility(vcpu->kvm, 78)) {
 122                 if (cpuflags & CPUSTAT_GED)
 123                         return set_validity_icpt(scb_s, 0x0001U);
 124                 newflags |= CPUSTAT_GED2;
 125         }
 126         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GPERE))
 127                 newflags |= cpuflags & CPUSTAT_P;
 128         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GSLS))
 129                 newflags |= cpuflags & CPUSTAT_SM;
 130         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IBS))
 131                 newflags |= cpuflags & CPUSTAT_IBS;
 132         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_KSS))
 133                 newflags |= cpuflags & CPUSTAT_KSS;
 134
 135         atomic_set(&scb_s->cpuflags, newflags);
 136         return 0;
 137 }
 138 /* Copy to APCB FORMAT1 from APCB FORMAT0 */
 139 static int setup_apcb10(struct kvm_vcpu *vcpu, struct kvm_s390_apcb1 *apcb_s,
 140                         unsigned long apcb_o, struct kvm_s390_apcb1 *apcb_h)
 141 {
 142         struct kvm_s390_apcb0 tmp;
 143
 144         if (read_guest_real(vcpu, apcb_o, &tmp, sizeof(struct kvm_s390_apcb0)))
 145                 return -EFAULT;
 146
 147         apcb_s->apm[0] = apcb_h->apm[0] & tmp.apm[0];
 148         apcb_s->aqm[0] = apcb_h->aqm[0] & tmp.aqm[0] & 0xffff000000000000UL;
 149         apcb_s->adm[0] = apcb_h->adm[0] & tmp.adm[0] & 0xffff000000000000UL;
 150
 151         return 0;
 152
 153 }
 154
 155 /**
 156  * setup_apcb00 - Copy to APCB FORMAT0 from APCB FORMAT0
 157  * @vcpu: pointer to the virtual CPU
 158  * @apcb_s: pointer to start of apcb in the shadow crycb
 159  * @apcb_o: pointer to start of original apcb in the guest2
 160  * @apcb_h: pointer to start of apcb in the guest1
 161  *
 162  * Returns 0 and -EFAULT on error reading guest apcb
 163  */
 164 static int setup_apcb00(struct kvm_vcpu *vcpu, unsigned long *apcb_s,
 165                         unsigned long apcb_o, unsigned long *apcb_h)
 166 {
 167         if (read_guest_real(vcpu, apcb_o, apcb_s,
 168                             sizeof(struct kvm_s390_apcb0)))
 169                 return -EFAULT;
 170
 171         bitmap_and(apcb_s, apcb_s, apcb_h, sizeof(struct kvm_s390_apcb0));
 172
 173         return 0;
 174 }
 175
 176 /**
 177  * setup_apcb11 - Copy the FORMAT1 APCB from the guest to the shadow CRYCB
 178  * @vcpu: pointer to the virtual CPU
 179  * @apcb_s: pointer to start of apcb in the shadow crycb
 180  * @apcb_o: pointer to start of original guest apcb
 181  * @apcb_h: pointer to start of apcb in the host
 182  *
 183  * Returns 0 and -EFAULT on error reading guest apcb
 184  */
 185 static int setup_apcb11(struct kvm_vcpu *vcpu, unsigned long *apcb_s,
 186                         unsigned long apcb_o,
 187                         unsigned long *apcb_h)
 188 {
 189         if (read_guest_real(vcpu, apcb_o, apcb_s,
 190                             sizeof(struct kvm_s390_apcb1)))
 191                 return -EFAULT;
 192
 193         bitmap_and(apcb_s, apcb_s, apcb_h, sizeof(struct kvm_s390_apcb1));
 194
 195         return 0;
 196 }
 197
 198 /**
 199  * setup_apcb - Create a shadow copy of the apcb.
 200  * @vcpu: pointer to the virtual CPU
 201  * @crycb_s: pointer to shadow crycb
 202  * @crycb_o: pointer to original guest crycb
 203  * @crycb_h: pointer to the host crycb
 204  * @fmt_o: format of the original guest crycb.
 205  * @fmt_h: format of the host crycb.
 206  *
 207  * Checks the compatibility between the guest and host crycb and calls the
 208  * appropriate copy function.
 209  *
 210  * Return 0 or an error number if the guest and host crycb are incompatible.
 211  */
 212 static int setup_apcb(struct kvm_vcpu *vcpu, struct kvm_s390_crypto_cb *crycb_s,
 213                const u32 crycb_o,
 214                struct kvm_s390_crypto_cb *crycb_h,
 215                int fmt_o, int fmt_h)
 216 {
 217         struct kvm_s390_crypto_cb *crycb;
 218
 219         crycb = (struct kvm_s390_crypto_cb *) (unsigned long)crycb_o;
 220
 221         switch (fmt_o) {
 222         case CRYCB_FORMAT2:
 223                 if ((crycb_o & PAGE_MASK) != ((crycb_o + 256) & PAGE_MASK))
 224                         return -EACCES;
 225                 if (fmt_h != CRYCB_FORMAT2)
 226                         return -EINVAL;
 227                 return setup_apcb11(vcpu, (unsigned long *)&crycb_s->apcb1,
 228                                     (unsigned long) &crycb->apcb1,
 229                                     (unsigned long *)&crycb_h->apcb1);
 230         case CRYCB_FORMAT1:
 231                 switch (fmt_h) {
 232                 case CRYCB_FORMAT2:
 233                         return setup_apcb10(vcpu, &crycb_s->apcb1,
 234                                             (unsigned long) &crycb->apcb0,
 235                                             &crycb_h->apcb1);
 236                 case CRYCB_FORMAT1:
 237                         return setup_apcb00(vcpu,
 238                                             (unsigned long *) &crycb_s->apcb0,
 239                                             (unsigned long) &crycb->apcb0,
 240                                             (unsigned long *) &crycb_h->apcb0);
 241                 }
 242                 break;
 243         case CRYCB_FORMAT0:
 244                 if ((crycb_o & PAGE_MASK) != ((crycb_o + 32) & PAGE_MASK))
 245                         return -EACCES;
 246
 247                 switch (fmt_h) {
 248                 case CRYCB_FORMAT2:
 249                         return setup_apcb10(vcpu, &crycb_s->apcb1,
 250                                             (unsigned long) &crycb->apcb0,
 251                                             &crycb_h->apcb1);
 252                 case CRYCB_FORMAT1:
 253                 case CRYCB_FORMAT0:
 254                         return setup_apcb00(vcpu,
 255                                             (unsigned long *) &crycb_s->apcb0,
 256                                             (unsigned long) &crycb->apcb0,
 257                                             (unsigned long *) &crycb_h->apcb0);
 258                 }
 259         }
 260         return -EINVAL;
 261 }
 262
 263 /**
 264  * shadow_crycb - Create a shadow copy of the crycb block
 265  * @vcpu: a pointer to the virtual CPU
 266  * @vsie_page: a pointer to internal date used for the vSIE
 267  *
 268  * Create a shadow copy of the crycb block and setup key wrapping, if
 269  * requested for guest 3 and enabled for guest 2.
 270  *
 271  * We accept format-1 or format-2, but we convert format-1 into format-2
 272  * in the shadow CRYCB.
 273  * Using format-2 enables the firmware to choose the right format when
 274  * scheduling the SIE.
 275  * There is nothing to do for format-0.
 276  *
 277  * This function centralize the issuing of set_validity_icpt() for all
 278  * the subfunctions working on the crycb.
 279  *
 280  * Returns: - 0 if shadowed or nothing to do
 281  *          - > 0 if control has to be given to guest 2
 282  */
 283 static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 284 {
 285         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 286         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 287         const uint32_t crycbd_o = READ_ONCE(scb_o->crycbd);
 288         const u32 crycb_addr = crycbd_o & 0x7ffffff8U;
 289         unsigned long *b1, *b2;
 290         u8 ecb3_flags;
 291         u32 ecd_flags;
 292         int apie_h;
 293         int apie_s;
 294         int key_msk = test_kvm_facility(vcpu->kvm, 76);
 295         int fmt_o = crycbd_o & CRYCB_FORMAT_MASK;
 296         int fmt_h = vcpu->arch.sie_block->crycbd & CRYCB_FORMAT_MASK;
 297         int ret = 0;
 298
 299         scb_s->crycbd = 0;
 300
 301         apie_h = vcpu->arch.sie_block->eca & ECA_APIE;
 302         apie_s = apie_h & scb_o->eca;
 303         if (!apie_s && (!key_msk || (fmt_o == CRYCB_FORMAT0)))
 304                 return 0;
 305
 306         if (!crycb_addr)
 307                 return set_validity_icpt(scb_s, 0x0039U);
 308
 309         if (fmt_o == CRYCB_FORMAT1)
 310                 if ((crycb_addr & PAGE_MASK) !=
 311                     ((crycb_addr + 128) & PAGE_MASK))
 312                         return set_validity_icpt(scb_s, 0x003CU);
 313
 314         if (apie_s) {
 315                 ret = setup_apcb(vcpu, &vsie_page->crycb, crycb_addr,
 316                                  vcpu->kvm->arch.crypto.crycb,
 317                                  fmt_o, fmt_h);
 318                 if (ret)
 319                         goto end;
 320                 scb_s->eca |= scb_o->eca & ECA_APIE;
 321         }
 322
 323         /* we may only allow it if enabled for guest 2 */
 324         ecb3_flags = scb_o->ecb3 & vcpu->arch.sie_block->ecb3 &
 325                      (ECB3_AES | ECB3_DEA);
 326         ecd_flags = scb_o->ecd & vcpu->arch.sie_block->ecd & ECD_ECC;
 327         if (!ecb3_flags && !ecd_flags)
 328                 goto end;
 329
 330         /* copy only the wrapping keys */
 331         if (read_guest_real(vcpu, crycb_addr + 72,
 332                             vsie_page->crycb.dea_wrapping_key_mask, 56))
 333                 return set_validity_icpt(scb_s, 0x0035U);
 334
 335         scb_s->ecb3 |= ecb3_flags;
 336         scb_s->ecd |= ecd_flags;
 337
 338         /* xor both blocks in one run */
 339         b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask;
 340         b2 = (unsigned long *)
 341                             vcpu->kvm->arch.crypto.crycb->dea_wrapping_key_mask;
 342         /* as 56%8 == 0, bitmap_xor won't overwrite any data */
 343         bitmap_xor(b1, b1, b2, BITS_PER_BYTE * 56);
 344 end:
 345         switch (ret) {
 346         case -EINVAL:
 347                 return set_validity_icpt(scb_s, 0x0022U);
 348         case -EFAULT:
 349                 return set_validity_icpt(scb_s, 0x0035U);
 350         case -EACCES:
 351                 return set_validity_icpt(scb_s, 0x003CU);
 352         }
 353         scb_s->crycbd = ((__u32)(__u64) &vsie_page->crycb) | CRYCB_FORMAT2;
 354         return 0;
 355 }
 356
 357 /* shadow (round up/down) the ibc to avoid validity icpt */
 358 static void prepare_ibc(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 359 {
 360         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 361         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 362         /* READ_ONCE does not work on bitfields - use a temporary variable */
 363         const uint32_t __new_ibc = scb_o->ibc;
 364         const uint32_t new_ibc = READ_ONCE(__new_ibc) & 0x0fffU;
 365         __u64 min_ibc = (sclp.ibc >> 16) & 0x0fffU;
 366
 367         scb_s->ibc = 0;
 368         /* ibc installed in g2 and requested for g3 */
 369         if (vcpu->kvm->arch.model.ibc && new_ibc) {
 370                 scb_s->ibc = new_ibc;
 371                 /* takte care of the minimum ibc level of the machine */
 372                 if (scb_s->ibc < min_ibc)
 373                         scb_s->ibc = min_ibc;
 374                 /* take care of the maximum ibc level set for the guest */
 375                 if (scb_s->ibc > vcpu->kvm->arch.model.ibc)
 376                         scb_s->ibc = vcpu->kvm->arch.model.ibc;
 377         }
 378 }
 379
 380 /* unshadow the scb, copying parameters back to the real scb */
 381 static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 382 {
 383         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 384         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 385
 386         /* interception */
 387         scb_o->icptcode = scb_s->icptcode;
 388         scb_o->icptstatus = scb_s->icptstatus;
 389         scb_o->ipa = scb_s->ipa;
 390         scb_o->ipb = scb_s->ipb;
 391         scb_o->gbea = scb_s->gbea;
 392
 393         /* timer */
 394         scb_o->cputm = scb_s->cputm;
 395         scb_o->ckc = scb_s->ckc;
 396         scb_o->todpr = scb_s->todpr;
 397
 398         /* guest state */
 399         scb_o->gpsw = scb_s->gpsw;
 400         scb_o->gg14 = scb_s->gg14;
 401         scb_o->gg15 = scb_s->gg15;
 402         memcpy(scb_o->gcr, scb_s->gcr, 128);
 403         scb_o->pp = scb_s->pp;
 404
 405         /* branch prediction */
 406         if (test_kvm_facility(vcpu->kvm, 82)) {
 407                 scb_o->fpf &= ~FPF_BPBC;
 408                 scb_o->fpf |= scb_s->fpf & FPF_BPBC;
 409         }
 410
 411         /* interrupt intercept */
 412         switch (scb_s->icptcode) {
 413         case ICPT_PROGI:
 414         case ICPT_INSTPROGI:
 415         case ICPT_EXTINT:
 416                 memcpy((void *)((u64)scb_o + 0xc0),
 417                        (void *)((u64)scb_s + 0xc0), 0xf0 - 0xc0);
 418                 break;
 419         case ICPT_PARTEXEC:
 420                 /* MVPG only */
 421                 memcpy((void *)((u64)scb_o + 0xc0),
 422                        (void *)((u64)scb_s + 0xc0), 0xd0 - 0xc0);
 423                 break;
 424         }
 425
 426         if (scb_s->ihcpu != 0xffffU)
 427                 scb_o->ihcpu = scb_s->ihcpu;
 428 }
 429
 430 /*
 431  * Setup the shadow scb by copying and checking the relevant parts of the g2
 432  * provided scb.
 433  *
 434  * Returns: - 0 if the scb has been shadowed
 435  *          - > 0 if control has to be given to guest 2
 436  */
 437 static int shadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 438 {
 439         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 440         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 441         /* READ_ONCE does not work on bitfields - use a temporary variable */
 442         const uint32_t __new_prefix = scb_o->prefix;
 443         const uint32_t new_prefix = READ_ONCE(__new_prefix);
 444         const bool wants_tx = READ_ONCE(scb_o->ecb) & ECB_TE;
 445         bool had_tx = scb_s->ecb & ECB_TE;
 446         unsigned long new_mso = 0;
 447         int rc;
 448
 449         /* make sure we don't have any leftovers when reusing the scb */
 450         scb_s->icptcode = 0;
 451         scb_s->eca = 0;
 452         scb_s->ecb = 0;
 453         scb_s->ecb2 = 0;
 454         scb_s->ecb3 = 0;
 455         scb_s->ecd = 0;
 456         scb_s->fac = 0;
 457         scb_s->fpf = 0;
 458
 459         rc = prepare_cpuflags(vcpu, vsie_page);
 460         if (rc)
 461                 goto out;
 462
 463         /* timer */
 464         scb_s->cputm = scb_o->cputm;
 465         scb_s->ckc = scb_o->ckc;
 466         scb_s->todpr = scb_o->todpr;
 467         scb_s->epoch = scb_o->epoch;
 468
 469         /* guest state */
 470         scb_s->gpsw = scb_o->gpsw;
 471         scb_s->gg14 = scb_o->gg14;
 472         scb_s->gg15 = scb_o->gg15;
 473         memcpy(scb_s->gcr, scb_o->gcr, 128);
 474         scb_s->pp = scb_o->pp;
 475
 476         /* interception / execution handling */
 477         scb_s->gbea = scb_o->gbea;
 478         scb_s->lctl = scb_o->lctl;
 479         scb_s->svcc = scb_o->svcc;
 480         scb_s->ictl = scb_o->ictl;
 481         /*
 482          * SKEY handling functions can't deal with false setting of PTE invalid
 483          * bits. Therefore we cannot provide interpretation and would later
 484          * have to provide own emulation handlers.
 485          */
 486         if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_KSS))
 487                 scb_s->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
 488
 489         scb_s->icpua = scb_o->icpua;
 490
 491         if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_SM))
 492                 new_mso = READ_ONCE(scb_o->mso) & 0xfffffffffff00000UL;
 493         /* if the hva of the prefix changes, we have to remap the prefix */
 494         if (scb_s->mso != new_mso || scb_s->prefix != new_prefix)
 495                 prefix_unmapped(vsie_page);
 496          /* SIE will do mso/msl validity and exception checks for us */
 497         scb_s->msl = scb_o->msl & 0xfffffffffff00000UL;
 498         scb_s->mso = new_mso;
 499         scb_s->prefix = new_prefix;
 500
 501         /* We have to definetly flush the tlb if this scb never ran */
 502         if (scb_s->ihcpu != 0xffffU)
 503                 scb_s->ihcpu = scb_o->ihcpu;
 504
 505         /* MVPG and Protection Exception Interpretation are always available */
 506         scb_s->eca |= scb_o->eca & (ECA_MVPGI | ECA_PROTEXCI);
 507         /* Host-protection-interruption introduced with ESOP */
 508         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_ESOP))
 509                 scb_s->ecb |= scb_o->ecb & ECB_HOSTPROTINT;
 510         /* transactional execution */
 511         if (test_kvm_facility(vcpu->kvm, 73) && wants_tx) {
 512                 /* remap the prefix is tx is toggled on */
 513                 if (!had_tx)
 514                         prefix_unmapped(vsie_page);
 515                 scb_s->ecb |= ECB_TE;
 516         }
 517         /* branch prediction */
 518         if (test_kvm_facility(vcpu->kvm, 82))
 519                 scb_s->fpf |= scb_o->fpf & FPF_BPBC;
 520         /* SIMD */
 521         if (test_kvm_facility(vcpu->kvm, 129)) {
 522                 scb_s->eca |= scb_o->eca & ECA_VX;
 523                 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
 524         }
 525         /* Run-time-Instrumentation */
 526         if (test_kvm_facility(vcpu->kvm, 64))
 527                 scb_s->ecb3 |= scb_o->ecb3 & ECB3_RI;
 528         /* Instruction Execution Prevention */
 529         if (test_kvm_facility(vcpu->kvm, 130))
 530                 scb_s->ecb2 |= scb_o->ecb2 & ECB2_IEP;
 531         /* Guarded Storage */
 532         if (test_kvm_facility(vcpu->kvm, 133)) {
 533                 scb_s->ecb |= scb_o->ecb & ECB_GS;
 534                 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
 535         }
 536         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIIF))
 537                 scb_s->eca |= scb_o->eca & ECA_SII;
 538         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IB))
 539                 scb_s->eca |= scb_o->eca & ECA_IB;
 540         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_CEI))
 541                 scb_s->eca |= scb_o->eca & ECA_CEI;
 542         /* Epoch Extension */
 543         if (test_kvm_facility(vcpu->kvm, 139))
 544                 scb_s->ecd |= scb_o->ecd & ECD_MEF;
 545
 546         /* etoken */
 547         if (test_kvm_facility(vcpu->kvm, 156))
 548                 scb_s->ecd |= scb_o->ecd & ECD_ETOKENF;
 549
 550         scb_s->hpid = HPID_VSIE;
 551
 552         prepare_ibc(vcpu, vsie_page);
 553         rc = shadow_crycb(vcpu, vsie_page);
 554 out:
 555         if (rc)
 556                 unshadow_scb(vcpu, vsie_page);
 557         return rc;
 558 }
 559
 560 void kvm_s390_vsie_gmap_notifier(struct gmap *gmap, unsigned long start,
 561                                  unsigned long end)
 562 {
 563         struct kvm *kvm = gmap->private;
 564         struct vsie_page *cur;
 565         unsigned long prefix;
 566         struct page *page;
 567         int i;
 568
 569         if (!gmap_is_shadow(gmap))
 570                 return;
 571         if (start >= 1UL << 31)
 572                 /* We are only interested in prefix pages */
 573                 return;
 574
 575         /*
 576          * Only new shadow blocks are added to the list during runtime,
 577          * therefore we can safely reference them all the time.
 578          */
 579         for (i = 0; i < kvm->arch.vsie.page_count; i++) {
 580                 page = READ_ONCE(kvm->arch.vsie.pages[i]);
 581                 if (!page)
 582                         continue;
 583                 cur = page_to_virt(page);
 584                 if (READ_ONCE(cur->gmap) != gmap)
 585                         continue;
 586                 prefix = cur->scb_s.prefix << GUEST_PREFIX_SHIFT;
 587                 /* with mso/msl, the prefix lies at an offset */
 588                 prefix += cur->scb_s.mso;
 589                 if (prefix <= end && start <= prefix + 2 * PAGE_SIZE - 1)
 590                         prefix_unmapped_sync(cur);
 591         }
 592 }
 593
 594 /*
 595  * Map the first prefix page and if tx is enabled also the second prefix page.
 596  *
 597  * The prefix will be protected, a gmap notifier will inform about unmaps.
 598  * The shadow scb must not be executed until the prefix is remapped, this is
 599  * guaranteed by properly handling PROG_REQUEST.
 600  *
 601  * Returns: - 0 on if successfully mapped or already mapped
 602  *          - > 0 if control has to be given to guest 2
 603  *          - -EAGAIN if the caller can retry immediately
 604  *          - -ENOMEM if out of memory
 605  */
 606 static int map_prefix(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 607 {
 608         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 609         u64 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
 610         int rc;
 611
 612         if (prefix_is_mapped(vsie_page))
 613                 return 0;
 614
 615         /* mark it as mapped so we can catch any concurrent unmappers */
 616         prefix_mapped(vsie_page);
 617
 618         /* with mso/msl, the prefix lies at offset *mso* */
 619         prefix += scb_s->mso;
 620
 621         rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, prefix);
 622         if (!rc && (scb_s->ecb & ECB_TE))
 623                 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
 624                                            prefix + PAGE_SIZE);
 625         /*
 626          * We don't have to mprotect, we will be called for all unshadows.
 627          * SIE will detect if protection applies and trigger a validity.
 628          */
 629         if (rc)
 630                 prefix_unmapped(vsie_page);
 631         if (rc > 0 || rc == -EFAULT)
 632                 rc = set_validity_icpt(scb_s, 0x0037U);
 633         return rc;
 634 }
 635
 636 /*
 637  * Pin the guest page given by gpa and set hpa to the pinned host address.
 638  * Will always be pinned writable.
 639  *
 640  * Returns: - 0 on success
 641  *          - -EINVAL if the gpa is not valid guest storage
 642  */
 643 static int pin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t *hpa)
 644 {
 645         struct page *page;
 646
 647         page = gfn_to_page(kvm, gpa_to_gfn(gpa));
 648         if (is_error_page(page))
 649                 return -EINVAL;
 650         *hpa = (hpa_t) page_to_virt(page) + (gpa & ~PAGE_MASK);
 651         return 0;
 652 }
 653
 654 /* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */
 655 static void unpin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t hpa)
 656 {
 657         kvm_release_pfn_dirty(hpa >> PAGE_SHIFT);
 658         /* mark the page always as dirty for migration */
 659         mark_page_dirty(kvm, gpa_to_gfn(gpa));
 660 }
 661
 662 /* unpin all blocks previously pinned by pin_blocks(), marking them dirty */
 663 static void unpin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 664 {
 665         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 666         hpa_t hpa;
 667
 668         hpa = (u64) scb_s->scaoh << 32 | scb_s->scaol;
 669         if (hpa) {
 670                 unpin_guest_page(vcpu->kvm, vsie_page->sca_gpa, hpa);
 671                 vsie_page->sca_gpa = 0;
 672                 scb_s->scaol = 0;
 673                 scb_s->scaoh = 0;
 674         }
 675
 676         hpa = scb_s->itdba;
 677         if (hpa) {
 678                 unpin_guest_page(vcpu->kvm, vsie_page->itdba_gpa, hpa);
 679                 vsie_page->itdba_gpa = 0;
 680                 scb_s->itdba = 0;
 681         }
 682
 683         hpa = scb_s->gvrd;
 684         if (hpa) {
 685                 unpin_guest_page(vcpu->kvm, vsie_page->gvrd_gpa, hpa);
 686                 vsie_page->gvrd_gpa = 0;
 687                 scb_s->gvrd = 0;
 688         }
 689
 690         hpa = scb_s->riccbd;
 691         if (hpa) {
 692                 unpin_guest_page(vcpu->kvm, vsie_page->riccbd_gpa, hpa);
 693                 vsie_page->riccbd_gpa = 0;
 694                 scb_s->riccbd = 0;
 695         }
 696
 697         hpa = scb_s->sdnxo;
 698         if (hpa) {
 699                 unpin_guest_page(vcpu->kvm, vsie_page->sdnx_gpa, hpa);
 700                 vsie_page->sdnx_gpa = 0;
 701                 scb_s->sdnxo = 0;
 702         }
 703 }
 704
 705 /*
 706  * Instead of shadowing some blocks, we can simply forward them because the
 707  * addresses in the scb are 64 bit long.
 708  *
 709  * This works as long as the data lies in one page. If blocks ever exceed one
 710  * page, we have to fall back to shadowing.
 711  *
 712  * As we reuse the sca, the vcpu pointers contained in it are invalid. We must
 713  * therefore not enable any facilities that access these pointers (e.g. SIGPIF).
 714  *
 715  * Returns: - 0 if all blocks were pinned.
 716  *          - > 0 if control has to be given to guest 2
 717  *          - -ENOMEM if out of memory
 718  */
 719 static int pin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 720 {
 721         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 722         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 723         hpa_t hpa;
 724         gpa_t gpa;
 725         int rc = 0;
 726
 727         gpa = READ_ONCE(scb_o->scaol) & ~0xfUL;
 728         if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO))
 729                 gpa |= (u64) READ_ONCE(scb_o->scaoh) << 32;
 730         if (gpa) {
 731                 if (gpa < 2 * PAGE_SIZE)
 732                         rc = set_validity_icpt(scb_s, 0x0038U);
 733                 else if ((gpa & ~0x1fffUL) == kvm_s390_get_prefix(vcpu))
 734                         rc = set_validity_icpt(scb_s, 0x0011U);
 735                 else if ((gpa & PAGE_MASK) !=
 736                          ((gpa + sizeof(struct bsca_block) - 1) & PAGE_MASK))
 737                         rc = set_validity_icpt(scb_s, 0x003bU);
 738                 if (!rc) {
 739                         rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 740                         if (rc)
 741                                 rc = set_validity_icpt(scb_s, 0x0034U);
 742                 }
 743                 if (rc)
 744                         goto unpin;
 745                 vsie_page->sca_gpa = gpa;
 746                 scb_s->scaoh = (u32)((u64)hpa >> 32);
 747                 scb_s->scaol = (u32)(u64)hpa;
 748         }
 749
 750         gpa = READ_ONCE(scb_o->itdba) & ~0xffUL;
 751         if (gpa && (scb_s->ecb & ECB_TE)) {
 752                 if (gpa < 2 * PAGE_SIZE) {
 753                         rc = set_validity_icpt(scb_s, 0x0080U);
 754                         goto unpin;
 755                 }
 756                 /* 256 bytes cannot cross page boundaries */
 757                 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 758                 if (rc) {
 759                         rc = set_validity_icpt(scb_s, 0x0080U);
 760                         goto unpin;
 761                 }
 762                 vsie_page->itdba_gpa = gpa;
 763                 scb_s->itdba = hpa;
 764         }
 765
 766         gpa = READ_ONCE(scb_o->gvrd) & ~0x1ffUL;
 767         if (gpa && (scb_s->eca & ECA_VX) && !(scb_s->ecd & ECD_HOSTREGMGMT)) {
 768                 if (gpa < 2 * PAGE_SIZE) {
 769                         rc = set_validity_icpt(scb_s, 0x1310U);
 770                         goto unpin;
 771                 }
 772                 /*
 773                  * 512 bytes vector registers cannot cross page boundaries
 774                  * if this block gets bigger, we have to shadow it.
 775                  */
 776                 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 777                 if (rc) {
 778                         rc = set_validity_icpt(scb_s, 0x1310U);
 779                         goto unpin;
 780                 }
 781                 vsie_page->gvrd_gpa = gpa;
 782                 scb_s->gvrd = hpa;
 783         }
 784
 785         gpa = READ_ONCE(scb_o->riccbd) & ~0x3fUL;
 786         if (gpa && (scb_s->ecb3 & ECB3_RI)) {
 787                 if (gpa < 2 * PAGE_SIZE) {
 788                         rc = set_validity_icpt(scb_s, 0x0043U);
 789                         goto unpin;
 790                 }
 791                 /* 64 bytes cannot cross page boundaries */
 792                 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 793                 if (rc) {
 794                         rc = set_validity_icpt(scb_s, 0x0043U);
 795                         goto unpin;
 796                 }
 797                 /* Validity 0x0044 will be checked by SIE */
 798                 vsie_page->riccbd_gpa = gpa;
 799                 scb_s->riccbd = hpa;
 800         }
 801         if (((scb_s->ecb & ECB_GS) && !(scb_s->ecd & ECD_HOSTREGMGMT)) ||
 802             (scb_s->ecd & ECD_ETOKENF)) {
 803                 unsigned long sdnxc;
 804
 805                 gpa = READ_ONCE(scb_o->sdnxo) & ~0xfUL;
 806                 sdnxc = READ_ONCE(scb_o->sdnxo) & 0xfUL;
 807                 if (!gpa || gpa < 2 * PAGE_SIZE) {
 808                         rc = set_validity_icpt(scb_s, 0x10b0U);
 809                         goto unpin;
 810                 }
 811                 if (sdnxc < 6 || sdnxc > 12) {
 812                         rc = set_validity_icpt(scb_s, 0x10b1U);
 813                         goto unpin;
 814                 }
 815                 if (gpa & ((1 << sdnxc) - 1)) {
 816                         rc = set_validity_icpt(scb_s, 0x10b2U);
 817                         goto unpin;
 818                 }
 819                 /* Due to alignment rules (checked above) this cannot
 820                  * cross page boundaries
 821                  */
 822                 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 823                 if (rc) {
 824                         rc = set_validity_icpt(scb_s, 0x10b0U);
 825                         goto unpin;
 826                 }
 827                 vsie_page->sdnx_gpa = gpa;
 828                 scb_s->sdnxo = hpa | sdnxc;
 829         }
 830         return 0;
 831 unpin:
 832         unpin_blocks(vcpu, vsie_page);
 833         return rc;
 834 }
 835
 836 /* unpin the scb provided by guest 2, marking it as dirty */
 837 static void unpin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
 838                       gpa_t gpa)
 839 {
 840         hpa_t hpa = (hpa_t) vsie_page->scb_o;
 841
 842         if (hpa)
 843                 unpin_guest_page(vcpu->kvm, gpa, hpa);
 844         vsie_page->scb_o = NULL;
 845 }
 846
 847 /*
 848  * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o.
 849  *
 850  * Returns: - 0 if the scb was pinned.
 851  *          - > 0 if control has to be given to guest 2
 852  */
 853 static int pin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
 854                    gpa_t gpa)
 855 {
 856         hpa_t hpa;
 857         int rc;
 858
 859         rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
 860         if (rc) {
 861                 rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
 862                 WARN_ON_ONCE(rc);
 863                 return 1;
 864         }
 865         vsie_page->scb_o = (struct kvm_s390_sie_block *) hpa;
 866         return 0;
 867 }
 868
 869 /*
 870  * Inject a fault into guest 2.
 871  *
 872  * Returns: - > 0 if control has to be given to guest 2
 873  *            < 0 if an error occurred during injection.
 874  */
 875 static int inject_fault(struct kvm_vcpu *vcpu, __u16 code, __u64 vaddr,
 876                         bool write_flag)
 877 {
 878         struct kvm_s390_pgm_info pgm = {
 879                 .code = code,
 880                 .trans_exc_code =
 881                         /* 0-51: virtual address */
 882                         (vaddr & 0xfffffffffffff000UL) |
 883                         /* 52-53: store / fetch */
 884                         (((unsigned int) !write_flag) + 1) << 10,
 885                         /* 62-63: asce id (alway primary == 0) */
 886                 .exc_access_id = 0, /* always primary */
 887                 .op_access_id = 0, /* not MVPG */
 888         };
 889         int rc;
 890
 891         if (code == PGM_PROTECTION)
 892                 pgm.trans_exc_code |= 0x4UL;
 893
 894         rc = kvm_s390_inject_prog_irq(vcpu, &pgm);
 895         return rc ? rc : 1;
 896 }
 897
 898 /*
 899  * Handle a fault during vsie execution on a gmap shadow.
 900  *
 901  * Returns: - 0 if the fault was resolved
 902  *          - > 0 if control has to be given to guest 2
 903  *          - < 0 if an error occurred
 904  */
 905 static int handle_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 906 {
 907         int rc;
 908
 909         if (current->thread.gmap_int_code == PGM_PROTECTION)
 910                 /* we can directly forward all protection exceptions */
 911                 return inject_fault(vcpu, PGM_PROTECTION,
 912                                     current->thread.gmap_addr, 1);
 913
 914         rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
 915                                    current->thread.gmap_addr);
 916         if (rc > 0) {
 917                 rc = inject_fault(vcpu, rc,
 918                                   current->thread.gmap_addr,
 919                                   current->thread.gmap_write_flag);
 920                 if (rc >= 0)
 921                         vsie_page->fault_addr = current->thread.gmap_addr;
 922         }
 923         return rc;
 924 }
 925
 926 /*
 927  * Retry the previous fault that required guest 2 intervention. This avoids
 928  * one superfluous SIE re-entry and direct exit.
 929  *
 930  * Will ignore any errors. The next SIE fault will do proper fault handling.
 931  */
 932 static void handle_last_fault(struct kvm_vcpu *vcpu,
 933                               struct vsie_page *vsie_page)
 934 {
 935         if (vsie_page->fault_addr)
 936                 kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
 937                                       vsie_page->fault_addr);
 938         vsie_page->fault_addr = 0;
 939 }
 940
 941 static inline void clear_vsie_icpt(struct vsie_page *vsie_page)
 942 {
 943         vsie_page->scb_s.icptcode = 0;
 944 }
 945
 946 /* rewind the psw and clear the vsie icpt, so we can retry execution */
 947 static void retry_vsie_icpt(struct vsie_page *vsie_page)
 948 {
 949         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 950         int ilen = insn_length(scb_s->ipa >> 8);
 951
 952         /* take care of EXECUTE instructions */
 953         if (scb_s->icptstatus & 1) {
 954                 ilen = (scb_s->icptstatus >> 4) & 0x6;
 955                 if (!ilen)
 956                         ilen = 4;
 957         }
 958         scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, ilen);
 959         clear_vsie_icpt(vsie_page);
 960 }
 961
 962 /*
 963  * Try to shadow + enable the guest 2 provided facility list.
 964  * Retry instruction execution if enabled for and provided by guest 2.
 965  *
 966  * Returns: - 0 if handled (retry or guest 2 icpt)
 967  *          - > 0 if control has to be given to guest 2
 968  */
 969 static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 970 {
 971         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 972         __u32 fac = READ_ONCE(vsie_page->scb_o->fac) & 0x7ffffff8U;
 973
 974         if (fac && test_kvm_facility(vcpu->kvm, 7)) {
 975                 retry_vsie_icpt(vsie_page);
 976                 if (read_guest_real(vcpu, fac, &vsie_page->fac,
 977                                     sizeof(vsie_page->fac)))
 978                         return set_validity_icpt(scb_s, 0x1090U);
 979                 scb_s->fac = (__u32)(__u64) &vsie_page->fac;
 980         }
 981         return 0;
 982 }
 983
 984 /*
 985  * Run the vsie on a shadow scb and a shadow gmap, without any further
 986  * sanity checks, handling SIE faults.
 987  *
 988  * Returns: - 0 everything went fine
 989  *          - > 0 if control has to be given to guest 2
 990  *          - < 0 if an error occurred
 991  */
 992 static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
 993         __releases(vcpu->kvm->srcu)
 994         __acquires(vcpu->kvm->srcu)
 995 {
 996         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
 997         struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
 998         int guest_bp_isolation;
 999         int rc = 0;
1000
1001         handle_last_fault(vcpu, vsie_page);
1002
1003         if (need_resched())
1004                 schedule();
1005         if (test_cpu_flag(CIF_MCCK_PENDING))
1006                 s390_handle_mcck();
1007
1008         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
1009
1010         /* save current guest state of bp isolation override */
1011         guest_bp_isolation = test_thread_flag(TIF_ISOLATE_BP_GUEST);
1012
1013         /*
1014          * The guest is running with BPBC, so we have to force it on for our
1015          * nested guest. This is done by enabling BPBC globally, so the BPBC
1016          * control in the SCB (which the nested guest can modify) is simply
1017          * ignored.
1018          */
1019         if (test_kvm_facility(vcpu->kvm, 82) &&
1020             vcpu->arch.sie_block->fpf & FPF_BPBC)
1021                 set_thread_flag(TIF_ISOLATE_BP_GUEST);
1022
1023         local_irq_disable();
1024         guest_enter_irqoff();
1025         local_irq_enable();
1026
1027         /*
1028          * Simulate a SIE entry of the VCPU (see sie64a), so VCPU blocking
1029          * and VCPU requests also hinder the vSIE from running and lead
1030          * to an immediate exit. kvm_s390_vsie_kick() has to be used to
1031          * also kick the vSIE.
1032          */
1033         vcpu->arch.sie_block->prog0c |= PROG_IN_SIE;
1034         barrier();
1035         if (!kvm_s390_vcpu_sie_inhibited(vcpu))
1036                 rc = sie64a(scb_s, vcpu->run->s.regs.gprs);
1037         barrier();
1038         vcpu->arch.sie_block->prog0c &= ~PROG_IN_SIE;
1039
1040         local_irq_disable();
1041         guest_exit_irqoff();
1042         local_irq_enable();
1043
1044         /* restore guest state for bp isolation override */
1045         if (!guest_bp_isolation)
1046                 clear_thread_flag(TIF_ISOLATE_BP_GUEST);
1047
1048         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1049
1050         if (rc == -EINTR) {
1051                 VCPU_EVENT(vcpu, 3, "%s", "machine check");
1052                 kvm_s390_reinject_machine_check(vcpu, &vsie_page->mcck_info);
1053                 return 0;
1054         }
1055
1056         if (rc > 0)
1057                 rc = 0; /* we could still have an icpt */
1058         else if (rc == -EFAULT)
1059                 return handle_fault(vcpu, vsie_page);
1060
1061         switch (scb_s->icptcode) {
1062         case ICPT_INST:
1063                 if (scb_s->ipa == 0xb2b0)
1064                         rc = handle_stfle(vcpu, vsie_page);
1065                 break;
1066         case ICPT_STOP:
1067                 /* stop not requested by g2 - must have been a kick */
1068                 if (!(atomic_read(&scb_o->cpuflags) & CPUSTAT_STOP_INT))
1069                         clear_vsie_icpt(vsie_page);
1070                 break;
1071         case ICPT_VALIDITY:
1072                 if ((scb_s->ipa & 0xf000) != 0xf000)
1073                         scb_s->ipa += 0x1000;
1074                 break;
1075         }
1076         return rc;
1077 }
1078
1079 static void release_gmap_shadow(struct vsie_page *vsie_page)
1080 {
1081         if (vsie_page->gmap)
1082                 gmap_put(vsie_page->gmap);
1083         WRITE_ONCE(vsie_page->gmap, NULL);
1084         prefix_unmapped(vsie_page);
1085 }
1086
1087 static int acquire_gmap_shadow(struct kvm_vcpu *vcpu,
1088                                struct vsie_page *vsie_page)
1089 {
1090         unsigned long asce;
1091         union ctlreg0 cr0;
1092         struct gmap *gmap;
1093         int edat;
1094
1095         asce = vcpu->arch.sie_block->gcr[1];
1096         cr0.val = vcpu->arch.sie_block->gcr[0];
1097         edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
1098         edat += edat && test_kvm_facility(vcpu->kvm, 78);
1099
1100         /*
1101          * ASCE or EDAT could have changed since last icpt, or the gmap
1102          * we're holding has been unshadowed. If the gmap is still valid,
1103          * we can safely reuse it.
1104          */
1105         if (vsie_page->gmap && gmap_shadow_valid(vsie_page->gmap, asce, edat))
1106                 return 0;
1107
1108         /* release the old shadow - if any, and mark the prefix as unmapped */
1109         release_gmap_shadow(vsie_page);
1110         gmap = gmap_shadow(vcpu->arch.gmap, asce, edat);
1111         if (IS_ERR(gmap))
1112                 return PTR_ERR(gmap);
1113         gmap->private = vcpu->kvm;
1114         WRITE_ONCE(vsie_page->gmap, gmap);
1115         return 0;
1116 }
1117
1118 /*
1119  * Register the shadow scb at the VCPU, e.g. for kicking out of vsie.
1120  */
1121 static void register_shadow_scb(struct kvm_vcpu *vcpu,
1122                                 struct vsie_page *vsie_page)
1123 {
1124         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1125
1126         WRITE_ONCE(vcpu->arch.vsie_block, &vsie_page->scb_s);
1127         /*
1128          * External calls have to lead to a kick of the vcpu and
1129          * therefore the vsie -> Simulate Wait state.
1130          */
1131         kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
1132         /*
1133          * We have to adjust the g3 epoch by the g2 epoch. The epoch will
1134          * automatically be adjusted on tod clock changes via kvm_sync_clock.
1135          */
1136         preempt_disable();
1137         scb_s->epoch += vcpu->kvm->arch.epoch;
1138
1139         if (scb_s->ecd & ECD_MEF) {
1140                 scb_s->epdx += vcpu->kvm->arch.epdx;
1141                 if (scb_s->epoch < vcpu->kvm->arch.epoch)
1142                         scb_s->epdx += 1;
1143         }
1144
1145         preempt_enable();
1146 }
1147
1148 /*
1149  * Unregister a shadow scb from a VCPU.
1150  */
1151 static void unregister_shadow_scb(struct kvm_vcpu *vcpu)
1152 {
1153         kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
1154         WRITE_ONCE(vcpu->arch.vsie_block, NULL);
1155 }
1156
1157 /*
1158  * Run the vsie on a shadowed scb, managing the gmap shadow, handling
1159  * prefix pages and faults.
1160  *
1161  * Returns: - 0 if no errors occurred
1162  *          - > 0 if control has to be given to guest 2
1163  *          - -ENOMEM if out of memory
1164  */
1165 static int vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1166 {
1167         struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1168         int rc = 0;
1169
1170         while (1) {
1171                 rc = acquire_gmap_shadow(vcpu, vsie_page);
1172                 if (!rc)
1173                         rc = map_prefix(vcpu, vsie_page);
1174                 if (!rc) {
1175                         gmap_enable(vsie_page->gmap);
1176                         update_intervention_requests(vsie_page);
1177                         rc = do_vsie_run(vcpu, vsie_page);
1178                         gmap_enable(vcpu->arch.gmap);
1179                 }
1180                 atomic_andnot(PROG_BLOCK_SIE, &scb_s->prog20);
1181
1182                 if (rc == -EAGAIN)
1183                         rc = 0;
1184                 if (rc || scb_s->icptcode || signal_pending(current) ||
1185                     kvm_s390_vcpu_has_irq(vcpu, 0) ||
1186                     kvm_s390_vcpu_sie_inhibited(vcpu))
1187                         break;
1188         }
1189
1190         if (rc == -EFAULT) {
1191                 /*
1192                  * Addressing exceptions are always presentes as intercepts.
1193                  * As addressing exceptions are suppressing and our guest 3 PSW
1194                  * points at the responsible instruction, we have to
1195                  * forward the PSW and set the ilc. If we can't read guest 3
1196                  * instruction, we can use an arbitrary ilc. Let's always use
1197                  * ilen = 4 for now, so we can avoid reading in guest 3 virtual
1198                  * memory. (we could also fake the shadow so the hardware
1199                  * handles it).
1200                  */
1201                 scb_s->icptcode = ICPT_PROGI;
1202                 scb_s->iprcc = PGM_ADDRESSING;
1203                 scb_s->pgmilc = 4;
1204                 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, 4);
1205         }
1206         return rc;
1207 }
1208
1209 /*
1210  * Get or create a vsie page for a scb address.
1211  *
1212  * Returns: - address of a vsie page (cached or new one)
1213  *          - NULL if the same scb address is already used by another VCPU
1214  *          - ERR_PTR(-ENOMEM) if out of memory
1215  */
1216 static struct vsie_page *get_vsie_page(struct kvm *kvm, unsigned long addr)
1217 {
1218         struct vsie_page *vsie_page;
1219         struct page *page;
1220         int nr_vcpus;
1221
1222         rcu_read_lock();
1223         page = radix_tree_lookup(&kvm->arch.vsie.addr_to_page, addr >> 9);
1224         rcu_read_unlock();
1225         if (page) {
1226                 if (page_ref_inc_return(page) == 2)
1227                         return page_to_virt(page);
1228                 page_ref_dec(page);
1229         }
1230
1231         /*
1232          * We want at least #online_vcpus shadows, so every VCPU can execute
1233          * the VSIE in parallel.
1234          */
1235         nr_vcpus = atomic_read(&kvm->online_vcpus);
1236
1237         mutex_lock(&kvm->arch.vsie.mutex);
1238         if (kvm->arch.vsie.page_count < nr_vcpus) {
1239                 page = alloc_page(GFP_KERNEL | __GFP_ZERO | GFP_DMA);
1240                 if (!page) {
1241                         mutex_unlock(&kvm->arch.vsie.mutex);
1242                         return ERR_PTR(-ENOMEM);
1243                 }
1244                 page_ref_inc(page);
1245                 kvm->arch.vsie.pages[kvm->arch.vsie.page_count] = page;
1246                 kvm->arch.vsie.page_count++;
1247         } else {
1248                 /* reuse an existing entry that belongs to nobody */
1249                 while (true) {
1250                         page = kvm->arch.vsie.pages[kvm->arch.vsie.next];
1251                         if (page_ref_inc_return(page) == 2)
1252                                 break;
1253                         page_ref_dec(page);
1254                         kvm->arch.vsie.next++;
1255                         kvm->arch.vsie.next %= nr_vcpus;
1256                 }
1257                 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1258         }
1259         page->index = addr;
1260         /* double use of the same address */
1261         if (radix_tree_insert(&kvm->arch.vsie.addr_to_page, addr >> 9, page)) {
1262                 page_ref_dec(page);
1263                 mutex_unlock(&kvm->arch.vsie.mutex);
1264                 return NULL;
1265         }
1266         mutex_unlock(&kvm->arch.vsie.mutex);
1267
1268         vsie_page = page_to_virt(page);
1269         memset(&vsie_page->scb_s, 0, sizeof(struct kvm_s390_sie_block));
1270         release_gmap_shadow(vsie_page);
1271         vsie_page->fault_addr = 0;
1272         vsie_page->scb_s.ihcpu = 0xffffU;
1273         return vsie_page;
1274 }
1275
1276 /* put a vsie page acquired via get_vsie_page */
1277 static void put_vsie_page(struct kvm *kvm, struct vsie_page *vsie_page)
1278 {
1279         struct page *page = pfn_to_page(__pa(vsie_page) >> PAGE_SHIFT);
1280
1281         page_ref_dec(page);
1282 }
1283
1284 int kvm_s390_handle_vsie(struct kvm_vcpu *vcpu)
1285 {
1286         struct vsie_page *vsie_page;
1287         unsigned long scb_addr;
1288         int rc;
1289
1290         vcpu->stat.instruction_sie++;
1291         if (!test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIEF2))
1292                 return -EOPNOTSUPP;
1293         if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1294                 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1295
1296         BUILD_BUG_ON(sizeof(struct vsie_page) != PAGE_SIZE);
1297         scb_addr = kvm_s390_get_base_disp_s(vcpu, NULL);
1298
1299         /* 512 byte alignment */
1300         if (unlikely(scb_addr & 0x1ffUL))
1301                 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1302
1303         if (signal_pending(current) || kvm_s390_vcpu_has_irq(vcpu, 0) ||
1304             kvm_s390_vcpu_sie_inhibited(vcpu))
1305                 return 0;
1306
1307         vsie_page = get_vsie_page(vcpu->kvm, scb_addr);
1308         if (IS_ERR(vsie_page))
1309                 return PTR_ERR(vsie_page);
1310         else if (!vsie_page)
1311                 /* double use of sie control block - simply do nothing */
1312                 return 0;
1313
1314         rc = pin_scb(vcpu, vsie_page, scb_addr);
1315         if (rc)
1316                 goto out_put;
1317         rc = shadow_scb(vcpu, vsie_page);
1318         if (rc)
1319                 goto out_unpin_scb;
1320         rc = pin_blocks(vcpu, vsie_page);
1321         if (rc)
1322                 goto out_unshadow;
1323         register_shadow_scb(vcpu, vsie_page);
1324         rc = vsie_run(vcpu, vsie_page);
1325         unregister_shadow_scb(vcpu);
1326         unpin_blocks(vcpu, vsie_page);
1327 out_unshadow:
1328         unshadow_scb(vcpu, vsie_page);
1329 out_unpin_scb:
1330         unpin_scb(vcpu, vsie_page, scb_addr);
1331 out_put:
1332         put_vsie_page(vcpu->kvm, vsie_page);
1333
1334         return rc < 0 ? rc : 0;
1335 }
1336
1337 /* Init the vsie data structures. To be called when a vm is initialized. */
1338 void kvm_s390_vsie_init(struct kvm *kvm)
1339 {
1340         mutex_init(&kvm->arch.vsie.mutex);
1341         INIT_RADIX_TREE(&kvm->arch.vsie.addr_to_page, GFP_KERNEL);
1342 }
1343
1344 /* Destroy the vsie data structures. To be called when a vm is destroyed. */
1345 void kvm_s390_vsie_destroy(struct kvm *kvm)
1346 {
1347         struct vsie_page *vsie_page;
1348         struct page *page;
1349         int i;
1350
1351         mutex_lock(&kvm->arch.vsie.mutex);
1352         for (i = 0; i < kvm->arch.vsie.page_count; i++) {
1353                 page = kvm->arch.vsie.pages[i];
1354                 kvm->arch.vsie.pages[i] = NULL;
1355                 vsie_page = page_to_virt(page);
1356                 release_gmap_shadow(vsie_page);
1357                 /* free the radix tree entry */
1358                 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1359                 __free_page(page);
1360         }
1361         kvm->arch.vsie.page_count = 0;
1362         mutex_unlock(&kvm->arch.vsie.mutex);
1363 }
1364
1365 void kvm_s390_vsie_kick(struct kvm_vcpu *vcpu)
1366 {
1367         struct kvm_s390_sie_block *scb = READ_ONCE(vcpu->arch.vsie_block);
1368
1369         /*
1370          * Even if the VCPU lets go of the shadow sie block reference, it is
1371          * still valid in the cache. So we can safely kick it.
1372          */
1373         if (scb) {
1374                 atomic_or(PROG_BLOCK_SIE, &scb->prog20);
1375                 if (scb->prog0c & PROG_IN_SIE)
1376                         atomic_or(CPUSTAT_STOP_INT, &scb->cpuflags);
1377         }
1378 }